Lipoprotein(a): resurrected by genetics

Authors

  • F. Kronenberg,

    1. Division of Genetic Epidemiology, Department of Medical Genetics, Molecular and Clinical Pharmacology, Innsbruck Medical University, Innsbruck, Austria
    Search for more papers by this author
  • G. Utermann

    Corresponding author
    1. Division of Human Genetics, Department of Medical Genetics, Molecular and Clinical Pharmacology, Innsbruck Medical University, Innsbruck, Austria
    • Correspondence: Gerd Utermann MD, Division of Human Genetics, Department of Medical Genetics, Molecular and Clinical Pharmacology, Innsbruck Medical University, Schöpfstr. 41, A-6020 Innsbruck, Austria. (fax: +43 512 9003 73510; e-mail: gerd.utermann@i-med.ac.at)

    Search for more papers by this author

Abstract

Plasma lipoprotein(a) [Lp(a)] is a quantitative genetic trait with a very broad and skewed distribution, which is largely controlled by genetic variants at the LPA locus on chromosome 6q27. Based on genetic evidence provided by studies conducted over the last two decades, Lp(a) is currently considered to be the strongest genetic risk factor for coronary heart disease (CHD). The copy number variation of kringle IV in the LPA gene has been strongly associated with both Lp(a) levels in plasma and risk of CHD, thereby fulfilling the main criterion for causality in a Mendelian randomization approach. Alleles with a low kringle IV copy number that together have a population frequency of 25–35% are associated with a doubling of the relative risk for outcomes, which is exceptional in the field of complex genetic phenotypes. The recently identified binding of oxidized phospholipids to Lp(a) is considered as one of the possible mechanisms that may explain the pathogenicity of Lp(a). Drugs that have been shown to lower Lp(a) have pleiotropic effects on other CHD risk factors, and an improvement of cardiovascular endpoints is up to now lacking. However, it has been established in a proof of principle study that lowering of very high Lp(a) by apheresis in high-risk patients with already maximally reduced low-density lipoprotein cholesterol levels can dramatically reduce major coronary events.

Introduction

Lipoprotein(a) [Lp(a)] [1] is a complex particle present in human plasma, which is composed of a low-density lipoprotein (LDL) molecule and a high molecular weight glycoprotein, apolipoprotein(a) [apo(a)], homologous to plasminogen (PLG). The characteristics of Lp(a) (see also Table 1) are a more than 1000-fold range of concentrations amongst healthy individuals from <0.1 mg dL−1 to more than 200 mg dL−1, strict genetic control of the concentration which represents a quantitative genetic trait and the presence of a size polymorphism of apo(a) caused by a variable number of kringle (K) IV type 2 repeats (KIV-2) in the LPA gene [2].

Table 1. Summary of characteristics of lipoprotein(a) [Lp(a)] and apolipoprotein(a) [apo(a)]
An Lp(a) particle consists of one low-density lipoprotein (LDL) particle and the glycoprotein apo(a), which is linked to the apoB of LDL by a single disulphide bond
The LPA gene that codes for apo(a) evolved from plasminogen in Old World monkeys: high homology between apo(a) and plasminogen
A size polymorphism of the LPA gene on chromosome 6q26-q27 and of the apo(a) protein (>30 alleles and isoforms) results from CNV of KIV-2 in LPA
Convergent evolution of Lp(a); in the hedgehog, the KIII and not the KIV is the repetitive element
Lp(a) is synthesized in the liver
Site and mechanism of catabolism are unclear; no receptor specific for Lp(a)/apo(a) has been described but several observations point to a role of the kidney in Lp(a) removal
Lp(a) plasma concentrations: 1000-fold intra-population range (from 0 to >200 mg dL−1) and fourfold inter-population range. Skewed distribution in most populations (majority of Europeans have concentrations below 10 mg dL−1)
LPA is the major gene locus for Lp(a) concentrations in all populations
Inverse correlation between apo(a) isoform size (i.e. KIV repeats) and Lp(a) concentration
KIV size polymorphism of apo(a) explains about 20–80% of the variability of Lp(a) concentrations depending on ethnicity. This polymorphism and other sequence variations of the apo(a) gene explain 70–90% of the variability of Lp(a) concentration
Lp(a) concentration is relatively independent of age and gender
High Lp(a) concentration is associated with an increased risk of coronary heart disease
Small apo(a) alleles are associated with an increased risk of coronary heart disease; this is the strongest common genetic variation known to be associated with coronary heart disease
Lp(a) concentration is elevated in kidney disease, FH, familial defective apolipoprotein B-100, hypothyroidism and pregnancy, postmenopause and during growth hormone therapy
Lp(a) concentration is decreased in abetalipoproteinaemia, lecithin-cholesterol acyltransferase (LCAT) deficiency, lipoprotein lipase (LPL) deficiency, severe hypertriglyceridaemia, hepatic disorders associated with decreased liver function, alcoholism and hyperthyroidism, and during treatment with sex hormones or anabolic steroids

The finding in case–control studies that high Lp(a) plasma concentrations were associated with cardiovascular disease (CVD) generated considerable interest in Lp(a), which faded as this association could not be reproduced in the first prospective studies [3, 4]. However, a considerable number of subsequent prospective studies including meta-analyses [5] have provided sufficient evidence to support Lp(a) as a risk factor. Furthermore, following recent large genetic studies using the Mendelian randomization approach [6-9], Lp(a) is now considered important and consensus statements have been produced for the management of patients with high levels of Lp(a) [10].

LPA gene and the structure of Lp(a)

The LPA gene evolved from the PLG gene during primate evolution about 40 million years ago and is only present in Old World monkeys and primates (including humans) [11, 12]. PLG contains five types of K domains (KI–KV) and a protease domain (Fig. 1). The PLG gene has been duplicated and extensively remodelled during the evolution of the human LPA gene, and KI, KII and KIII have been lost. KV has been retained in single copy. The protease domain in LPA has been changed by mutations in critical residues in the human and rhesus LPA gene, predicting that it has lost plasmin activity [11, 13]. KIV has expanded and diversified by mutation into 10 different types (KIV types 1–10). One of these, the KIV-2 domain exists in multiple copies. The number of copies is variable ranging from two to >40 in one allele (Fig. 1). Therefore, few individuals have two alleles of identical copy number in their genomes. The resulting polymorphism represents a rare type of copy number variation (CNV), and heterozygosity is >95% in most populations; thus, this CNV is very informative in terms of genetic heritability. Different methods have been used to determine the number of KIV-2 copies. These include pulsed-field gel electrophoresis (PFGE)/Southern blotting of genomic DNA [14, 15] and quantitative polymerase chain reaction (qPCR) [7]. Whereas the first of these methods allows determination of the number of KIV-2 copies in separated alleles, the latter can only be used to estimate the total number (sum) of the KIV-2 copies of the two alleles of the investigated genome. The most precise method is fibre-fluorescence in situ hybridization (FISH), which allows the number of repeats to be counted under fluorescence microscopy [16] (Fig. 1). Fibre-FISH is not applicable for large sample sizes but has been used for standardization.

Figure 1.

Schematic illustration of the structural homology between plasminogen and apolipoprotein(a) [apo(a)]. Plasminogen contains five different kringle structures (I–V) and a protease domain. Apo(a) is missing KI–KIII but has a variable number of kringle (K) IV (KIV) copies. The minimum number is 11 (KIV type 1–10 with two copies of KIV type 2). An individual can have more than 40 KIV type 2 copies, which is the variable part of apo(a). The lower part of the illustration shows a two-colour fibre-FISH image of the LPA KIV type 2 domain from a 20-repeat allele using 4-kb (red) and 1.2-kb (green) intron probes, which enable the KIV type-2 repeat number to be counted (for details see [16]).

The KIV-2 CNV is translated into mRNA and transcribed into protein [apo(a) isoforms]. The resulting size polymorphism of apo(a) can be determined by immunoblotting from plasma. This not only allows an indirect estimate of the number of KIV-2 repeats but also provides information on the amount of apo(a) contributed by each allele.

During assembly of Lp(a), the apo(a) isoform binds covalently to apoB of LDL in a stoichiometric manner forming the Lp(a) complex [2]. This complex combines structural features with two different biochemical functions, that is, the fibrinolytic and the cholesterol transport systems in one particle. A surprising feature of the LPA/Lp(a) system is that it has developed twice by convergent evolution [12]. The insectivorous hedgehog has an LPA gene composed entirely of KIII domains, lacking not only all other K types but also the protease domain. Similar to its primate counterpart, hedgehog apo(a) forms a covalent complex with LDL, that is, the Lp(a) particle [12].

Regulation of LPA gene expression

The LPA gene is highly expressed in the liver but not in other organs [11]. The KIV-2 CNV can be demonstrated by mRNA analysis/Northern blotting [17]. Differences in LPA gene expression have not been observed between humans in promoter studies [18] but it has been reported that LPA gene expression is much higher in chimpanzees (Pan troglodytes) than in humans [19]. This may explain the several-fold higher Lp(a) levels in chimpanzees, compared with humans [20]. Three base variations in the LPA promoter between humans and chimpanzees, responsible for the expression difference, have been identified [19]. Binding sites for several transcription factors including hepatocyte nuclear factor HNF1α, HNF4α, sex hormones and acute phase inducers, which modulate LPA expression in baboons, have been identified in the 5′-region of the LPA gene [21]. A retinoid response element is present in the LPA promoter, and an enhancer residing in a long interspersed nuclear element (LINE) has been found in the intragenic region between LPA and PLG [21]. An Ets-binding element in the human LPA promoter functions as an ELK-1-binding site that mediates repression of LPA transcription by fibroblast growth factor 19 (FGF19) [22]. LPA gene expression is downregulated by bile acids. In line with this, patients with biliary obstruction have very low plasma Lp(a) levels, which are normalized following therapy. The Lp(a)-lowering effect of bile acids is mediated by the farnesoid-X receptor (FXR), which represses hepatic LPA gene expression in humans by interfering with HNF4α. This is explained by the observation that the LPA promoter contains a direct repeat element-1 (DR-1) between nucleotides 826 and 814 to which HNF4α binds, promoting LPA transcription [23]. FXR competes thereby with HNF4α for binding to the DR-1 element. Modulation of FXR has therefore been proposed as a potential target for Lp(a)-lowering therapy. However, the expected pleiotropic effects cast doubt on the usefulness of such a strategy.

Metabolism of Lp(a)

Apolipoprotein(a) is synthesized exclusively by the liver [24]. The site of assembly with apoB in LDL is still debated [25, 26] but may include the secretory pathway of the hepatocyte, the hepatocyte plasma membrane [27], the space of Disse or the plasma compartment. Enzymatic activity, catalysing disulphide bond formation between apoB in LDL and apo(a) has been identified [28]. Apo(a) synthesis and Lp(a) biogenesis have been studied in detail in primary baboon hepatocytes, hepatocytes from Lp(a)-transgenic mice and transfected HepG2 cells (reviewed in [25, 29, 30]). No intracellular assembly of Lp(a) could be observed in these direct experiments [27, 31-33]. Assembly of Lp(a) most likely occurs at the hepatocyte surface or in plasma and proceeds in two steps. First, apo(a) docks to LDL, and then a disulphide bond is formed between KIV-9 of apo(a) and apoB of LDL [26]. A puzzling finding is that an apoB synthesis inhibitor that lowers LDL in plasma also reduces plasma Lp(a) levels [34]. By contrast, the inhibition of cholesterol synthesis by HMG-CoA reductase inhibitors (statins) does not lower Lp(a) [35], and the inhibition of intracellular lipoprotein assembly by microsomal triglyceride transfer protein (MTP) inhibitors has only a moderate effect [36]. This may point to a critical role of newly synthesized apoB for Lp(a) formation. It is interesting that patients with abetalipoproteinaemia, who cannot assemble apoB-containing lipoproteins due to mutations in the MTP gene, have a lipid-poor apo(a)/apoB complex in plasma [37].

Turnover studies in humans have demonstrated that the residence time of Lp(a) in plasma is longer than that of LDL; there was no indication of a VLDL-like precursor in these experiments [38]. Differences in Lp(a) plasma levels are caused by differences in the synthetic rate for isoforms of different size, which is in line with the findings of cell culture experiments (reviewed in [25]). Variation in Lp(a) concentrations amongst individuals with isoforms of identical size is also determined by the rate of Lp(a) production rather than by differences in the catalytic rate [39]. Together with the genetic data, this suggests that KIV-2 repeat number, as well as the sequence variation, affects the synthetic rate.

In contrast to the site of synthesis, it is unclear how and where Lp(a) is removed from plasma. Although binding to several members of the LDL receptor (LDLR) family including VLDLR/megalin [40] and the LDLR itself [41] has been demonstrated, it is unknown whether or to which extend these LDLRs function as Lp(a) receptors in vivo. Binding to the LDLR was not confirmed in several studies, but a ‘hitchhiking-like process’ whereby Lp(a) attached to LDL is removed by the LDLR pathway has been proposed [42]. The observations that Lp(a) level is elevated in patients with familial hypercholesterolaemia (FH) caused by LDLR mutations, and is decreased in transgenic mice overexpressing the LDLR, provide indirect evidence that the LDLR may be involved in vivo [43]. On the other hand, statins, which cause an upregulation of LDLR, markedly decrease levels of LDL cholesterol (LDL-C) but not Lp(a) (33). The findings of an arteriovenous difference in Lp(a) concentrations in the renal circulation [44], of apo(a) fragments in urine [45, 46] and of disturbed Lp(a) metabolism in kidney disease [47, 48] have suggested a major role of the kidney in Lp(a) catabolism (see also section below: ‘Kidney disease and kidney function’).

Physiological function and pathophysiology of Lp(a)

The physiological function of Lp(a) remains unknown. Studies of the physiological function and pathogenic mechanism of Lp(a) have been notoriously amalgamated: almost all proposed functions also serve as an explanation for the pathogenicity of Lp(a). Studies have been greatly influenced by the finding of a high degree of homology between the apo(a) glycoprotein and PLG [11]. This suggested that Lp(a) may (i) provide a link between the cholesterol transport system in plasma and the fibrinolytic system and (ii) act as a modulator of the delicate balance between blood clotting and fibrinolysis. Numerous studies have tested the latter hypothesis and demonstrated, at least in vitro, that Lp(a) indeed interferes with the blood clotting/fibrinolytic cascades at several steps [49]. The reported functions of Lp(a) include inhibition of streptokinase- and urokinase-mediated activation of PLG by the tissue-type PLG activator (t-PA), inhibition of t-PA in solution, fibrin and fibrinogen binding, competition with PLG and t-PA binding for soluble fibrinogen, competition with PLG for binding to cellular receptors and enhancement of activity of the PLG activator inhibitor PAI-1 (reviewed in [26]). Because of these multiple interactions, it has been suggested that Lp(a) is an ‘interloper’ in the fibrinolytic system [50]. In view of the more than 1000-fold differences in plasma Lp(a) levels between individuals, major effects on the balance between clotting and fibrinolysis might be expected in vivo but have not convincingly been described. In this context, it is interesting that a clear association between Lp(a) concentrations or certain apo(a) isoforms and atherosclerotic stenosis but not venous thrombosis was recently detected in more than 40 000 subjects using a Mendelian randomization approach [51]. Similar observations were made in a large series of case–control samples including 4607 cases with venous thromboembolism [52].

Lipoprotein(a)/apo(a) also interacts with several components of the extracellular matrix, including fibrin, fibronectin, tetranectin, proteoglycans and DANCE [53], and binds to β2-glycoprotein [54]. Apo(a) lysine-binding sites in KIV-8 and KIV-10 mediate binding to fibrin/fibrinogen [26]. Binding of Lp(a) to fibrin has been proposed as a mechanism to deliver cholesterol to sites of injury and wound healing [55]. A negative side effect of this beneficial property might be deposition of cholesterol in growing atherosclerotic plaques by Lp(a) as well as inhibition of fibrinolysis at the plaque surface. Immunochemical studies have demonstrated apo(a) immunoreactivity in human atherosclerotic plaques and coronary artery bypass vein grafts colocalizing with apoB [56, 57], and intact Lp(a) has been isolated from plaques [58]. Furthermore, high Lp(a) levels impair activation of transforming growth factor-β by downregulation of plasmin generation, thereby contributing to smooth muscle cell proliferation [59]. Clear evidence that Lp(a) can interfere with many key reactions of clotting/fibrinolysis in vitro and is deposited in atherosclerotic plaques has been provided [60, 61]. Mechanisms independent of antifibrinolytic activity have also been described, including effects on monocytes/macrophages resulting in foam cell formation [62, 63]. Lp(a) induces chemoattractant activity of monocytes and macrophage expression of interleukin-8 (IL-8) [64].

In addition to in vitro studies, in vivo experiments have been conducted in mice and rabbits transgenic for the human LPA gene alone or both LPA and APOB to allow the formation of Lp(a) under physiological conditions [65-67]. Antifibrinolytic activities of the LPA transgene were demonstrated [68]. There is, however, controversy regarding whether these transgenic animals are prone to develop atherosclerotic plaques [69-71]. Hence, Lp(a) could not be unequivocally identified as a CVD risk factor from these studies.

How and whether the proposed functions of Lp(a) translate into in vivo functions and pathology in humans are presently unclear. First, in vivo functions have been demonstrated mainly in transgenic animals. Introducing apo(a)/Lp(a) into species that lack an endogenous LPA gene is a highly artificial situation. Species lacking Lp(a) may also lack co-evolved structures such as specific receptors. Second, almost all described functional properties of Lp(a) are also demonstrated by apo(a) alone and hence do not explain the existence of the complex Lp(a) particle. An exception is the ‘cholesterol delivery to wound healing’ hypothesis [55], which requires both components of Lp(a) but is not supported by experimental evidence.

An unexpected and intriguing observation is the binding of oxidized phospholipids (OxPls) to Lp(a) [72]. The binding site for OxPls has been identified in the protein moiety of Lp(a), specifically in the KV domain of apo(a) [73]. The covalent modification of apo(a) by oxidized phosphatidylcholine seems to be responsible for the effect of Lp(a) on IL-8 expression by macrophages [26]. Levels of Lp(a) and OxPls in human plasma are highly correlated, suggesting that individuals with high Lp(a) levels have a higher binding capacity for OxPls and higher OxPl plasma levels. Not unexpectedly, this association also results in an association between OxPl levels and CVD [74, 75]. OxPls on apoB-containing lipoproteins, which reflect Lp(a), have therefore been suggested as biomarkers to predict CVD [76]. The well-known atherogenic potential of oxidized LDL could therefore emerge as a property of Lp(a).

Although Lp(a) has been proposed to function as a ‘sink’ for OxPls, other binding sites must exist because humans lacking Lp(a) and species without an LPA gene apparently have the capacity to remove OxPls. Plasminogen has already been identified as a binding site for OxPls [77] suggesting that apo(a) has ‘inherited’ this property from its evolutionary ancestor PLG. The quantitative contribution of these two binding sites for the removal of OxPls is not known.

It also remains unclear whether apo(a) is an active protease. The sequences of the human and rhesus LPA gene predict that the protease domain has lost its plasmin-like activity, or cannot be activated, respectively. It has, however, been suggested that apo(a) has a proteolytic activity [78, 79].

There is no doubt that high Lp(a) levels increase the risk of CVD, although the mechanism is unclear. Whether Lp(a) is atherogenic, thrombogenic or both remains to be seen.

Genetics of Lp(a)

Explained genetic variance and KIV-2 repeat polymorphism

Lipoprotein(a) levels in human plasma undoubtedly have the strictest genetic control of all lipoproteins; more than 90% of the variance of Lp(a) concentrations is explained by genetics. The genetic determinants of Lp(a) levels have been identified and reside in the LPA gene itself [15, 80]. In fact, Lp(a) was discovered during a study to find new genetic variants of lipoproteins [1]. In this classical experiment 50 years ago, Berg injected rabbits with LDL (beta lipoproteins) and tested the generated antisera against a panel of different human serum samples: 30% of samples reacted positively and were designated Lp+ [1]. With the refinement of assay technologies, it became apparent that Lp(a) is a quantitative genetic trait [81-84] and that very few people completely lack Lp(a) in plasma [85]. The distribution of Lp(a) level is very broad in all populations and is highly skewed towards low levels in most ethnic groups (Fig. 2a); exceptions to this are populations in sub-Saharan Africa (see below) [86, 87].

Figure 2.

(a) Distribution of lipoprotein(a) [Lp(a)] concentration in a typical general population from Central Europe. (b) Mean Lp(a) concentration in various groups of subjects stratified by the number of kringle (K) IV (KIV) repeats; 11–22 KIV repeats are considered as low molecular weight (LMW) or small isoforms and those with >22 KIV repeats are considered as high molecular weight (HMW) or large apolipoprotein(a) [apo(a)] isoforms. There is a clear inverse correlation between the number of KIV repeats and Lp(a) concentration.

Early family studies established the genetic nature of the trait, and twin studies demonstrated that the heritability of Lp(a) is very high, exceeding 90% in populations of European and African descent [88, 89]. The cloning and sequencing of the LPA gene [11] and the discovery of the size polymorphism of apo(a) [81] and KIV-2 CNV in the gene [2, 14, 15, 90] enabled LPA to be identified as the major determinant of Lp(a) levels by association and sib-pair linkage studies [15, 80, 91-93]. As described, the LPA gene is characterized by an extensive size polymorphism caused by a variable number of KIV-2 repeats, which are transcribed and translated into protein isoforms of different sizes (Fig. 2b). Association studies using the size polymorphism of apo(a) [apo(a) isoforms] or the number of KIV-2 repeats in the gene demonstrated an inverse correlation between isoform size/KIV-2 repeat number and Lp(a) levels [81, 87, 94]. Depending on the population, between about 30% and 70% of the variance in Lp(a) concentrations is explained by the apo(a) size polymorphism [87].

The relation between KIV-2 repeat number and Lp(a) levels has been confirmed in cell culture experiments thus establishing causality [25]. The amount of apo(a) secreted by transfected HepG2 cells or primary baboon hepatocytes depends on the number of KIV-2 repeats in the construct [31] or endogenous LPA gene [28]. Together, these studies demonstrated that about 70–90% of the variance in Lp(a) levels can be explained by the LPA locus, suggesting a co-dominant model of inheritance. Heterozygotes for KIV-2 repeats have two different Lp(a) particles in plasma, which can be physically separated [95].

Single-nucleotide polymorphisms in LPA

The discrepancy between the fraction of the variance in Lp(a) concentration explained by association with the KIV-2 CNV and by linkage to the LPA locus leaves the variation and molecular mechanism responsible for the remaining 30–70% of the variance unexplained. This is in line with the observation that isoforms of the same size differ widely in concentration suggesting that KIV-2 variation in the LPA gene is not the only important contributor to Lp(a) level variation. Indeed, it has been reported that a pentanucleotide repeat polymorphism (PNRP) in the promoter region [18, 96] and several single-nucleotide polymorphisms (SNPs) in LPA are associated with Lp(a) levels [97-101]. However, not all variants are causal. A SNP may be associated with Lp(a) levels for the following three reasons: (i) The SNP is functional and causal. Clear examples of this are null alleles caused by truncating mutations in LPA [85], (ii) The SNP is not functional by itself but is in linkage disequilibrium (LD) with the KIV-2 CNV. SNPs rs3798220 and rs10455872, which have been associated with coronary heart disease (CHD) risk in Europeans, seem to fall into this category [9]. Together these SNPs have been reported to explain 36% of the variability of Lp(a) levels when statistically analysed [9], and (iii) The SNP is not causal but in LD with unknown causal variation in LPA. Hence the demonstration of causality of most SNPs requires stratification for the effect of the KIV-2 CNV and/or functional studies if the mutation type is not clearly predictive. The association between the promoter PNRP and Lp(a) levels does not reflect a causal mechanism [18] but might be due to unknown causal variation. It is important that the high estimates (70% to >90%) of the fraction of the variability in Lp(a) explained by linkage to the locus should not be confused with the estimates of ‘explained’ variability obtained from measured variation in LPA in association studies. Current methods do not capture the full spectrum of variation in LPA in individuals included in association studies. Therefore, estimates based on the contribution of the KIV-2 CNV and SNPs alone or in combination will necessarily be lower than those from linkage to the locus.

Lp(a) in genome-wide linkage and association studies

Many quantitative traits are influenced not only by the main coding gene. This was impressively demonstrated for the lipid traits total, LDL and HDL cholesterol as well as triglycerides, which are currently known to be influenced by 95 or more genetic loci [102-104]. It is therefore expected that Lp(a) will also be influenced by other modifying genes in addition to the LPA gene itself. LPA on chromosome 6q27 has also been identified as a major determinant of Lp(a) levels by genome-wide linkage [105-107] and association (GWA) studies [108-112]. In the genome-wide linkage studies, further loci were identified on chromosomes 1, 2, 11 and 13. These loci, however, were not confirmed by GWA studies. On the other hand, GWA studies performed to date have been limited by sample size, focus on certain subgroups (e.g. patients with diabetes mellitus, population isolates) or by the use of a specialized candidate gene chip. Due to these limitations, such studies were only able to identify the well-known region on chromosome 6q27 harbouring LPA, PLG and SLC22A3 [108-112]. It will require a large number of samples to identify further genes contributing to Lp(a) levels in addition to the LPA locus, if such loci exist. A conditional analysis adjusted for the effects of the LPA locus and especially apo(a) isoform size will greatly increase the power of such GWA studies.

In summary, Lp(a) concentrations are currently known to be mainly determined by two types of functional variation within the LPA gene, the KIV-2 CNV and SNPs [87]. Only a few established functional SNPs are known, which together explain only a minor fraction of the variance. At present, it therefore seems unlikely that a few SNPs with major effects are the cause of the non-KIV-2-mediated variability. Rather, multiple SNPs with different mostly small effect sizes probably contribute to the variance of the trait [87]. This may have practical consequences; for example, it might be difficult to design a test using a small number of SNPs to cover the genetic variation in LPA as a surrogate for the measurement of Lp(a) levels or the analysis of the KIV-2 CNV.

Lp(a)/LPA in different ethnic groups and species

Median and mean Lp(a) concentrations differ by up to fourfold between ethnic groups. Populations of sub-Saharan African descent have the highest reported levels whereas concentrations in Europeans are much lower [86, 87]. It has been suggested that there are differences in the types and frequencies of LPA gene variants between ethnic groups because the LPA gene is the major determinant of Lp(a) level variation in both Africans and Europeans [91-93]. It has been claimed that frequency differences in the KIV-2 alleles explain the differences in Lp(a) concentrations between ethnic groups. Such differences indeed exist [99, 113-115] but probably explain only a small fraction of the inter-ethnic variability. It was found in two separate studies that three SNPs in LPA can explain the higher Lp(a) levels in African Americans compared with North Americans of European descent [101, 116] but the SNPs identified in these two studies were not the same. In fact, Lp(a) levels associated with short KIV-2 repeat alleles are much higher in Europeans than in Africans, and null alleles exist at a minor allele frequency (MAF) of 2% in Europeans but have not been detected in sub-Saharan Africans. It is inconceivable that such a simple explanation explains the inter-ethnic variation. This variation is more likely to be associated with the complex genetic architecture of the trait such as the types, frequencies and LDs of CNVs and SNPs in LPA.

A nongenetic explanation for the differences in median Lp(a) levels between Africans and Europeans has also been suggested. Anuurad et al. [117] observed increased allele-specific Lp(a) levels for medium-sized apo(a) alleles in association with increased proinflammatory markers in African Americans, but not in white Americans. They suggested that inflammation-associated events may contribute to the inter-ethnic differences in Lp(a) concentrations.

Most species except Old World monkeys and primates lack an LPA gene and hence Lp(a) in plasma. In baboons, Lp(a) levels differ between subspecies, but in general are similar to levels in humans [118]. By contrast, the level of Lp(a) is 5-fold higher in chimpanzees than in humans [20].

Nongenetic factors influencing Lp(a) concentration

Kidney disease and kidney function

Lipoprotein(a) plasma levels are markedly influenced by the presence of chronic kidney disease (CKD) [48] and the glomerular filtration rate (GFR) [119] (Fig. 3). Lp(a) levels begin to increase in the earliest stages of renal impairment before GFR starts to decrease [120]. It is interesting that this increase can only be observed in patients with large but not small apo(a) isoforms, compared with isoform-matched controls; however, the reason for this is still unknown. This isoform-specific increase in Lp(a) was observed in several studies in patients with non-nephrotic CKD as well as end-stage renal disease (ESRD) undergoing haemodialysis [120-124]. By contrast, considerable increases in plasma Lp(a) levels of all apo(a) isoform groups occur in patients with nephrotic syndrome [125, 126] and in ESRD undergoing continuous peritoneal dialysis (CAPD) [122]. This might be a consequence of the pronounced protein loss in these patients and subsequently increased production in the liver [127]. In line with the findings in ESRD patients, a decrease in Lp(a) following a successful kidney transplantation was observed in haemodialysis patients with large apo(a) isoforms [128, 129] and in CAPD patients with all apo(a) isoforms [130]. Thus, the elevation of Lp(a) concentration in CKD is an acquired abnormality, mostly influenced by the impaired GFR and the degree of proteinuria. Furthermore, it has been suggested that the often observed malnutrition/inflammation syndrome in these patients additionally increases Lp(a) levels [123, 124]. The high concentrations of Lp(a) especially when of small apo(a) isoform size significantly increase the risk for CVD in these high-risk group of patients with CKD [131, 132].

Figure 3.

Lipoprotein(a) [Lp(a)] concentration at various stages of kidney disease compared with the control group. Data are provided for patients with various stages of kidney impairment not yet requiring renal-replacement therapy, patients with nephrotic syndrome and those requiring renal-replacement therapy such as haemodialysis (HD), continuous peritoneal dialysis (CAPD) and renal transplantation (RTX). GFR, glomerular filtration rate. Data from [120-122, 126, 128].

In vivo turnover studies using stable-isotope techniques pointed to pivotal differences in the metabolism of Lp(a) between haemodialysis patients [47] and those with nephrotic syndrome [127]. Patients with nephrotic syndrome showed no changes in the fractional catabolic rate of Lp(a) but an increased rate of Lp(a) synthesis when compared with control subjects [127]. This is in line with the generally increased synthesis of lipoproteins in nephrotic patients. By contrast, the production rates were normal in haemodialysis patients, compared with control subjects with similar plasma Lp(a) concentrations [47]. However, the fractional catabolic rate was significantly reduced resulting in a twofold longer residence time in plasma of almost 9 days in haemodialysis patients compared with only 4.4 days in controls. This decreased clearance of Lp(a) results in increased Lp(a) plasma concentrations [47]. A role of the kidney in the catabolism of Lp(a) is additionally supported by arteriovenous differences in Lp(a) concentrations between the arterial and renal vein, with lower concentrations in the vein [44], as well as by the presence of apo(a) fragments in urine [46, 133]. The urinary apo(a) excretion decreases with impaired kidney function [134].

Diabetes mellitus

Although type 2 diabetes mellitus (T2DM) is a common disease, the association between Lp(a) and T2DM or insulin resistance has not been extensively investigated. The results of several small studies published in the 1990s were inconclusive or pointed to a secondary effect of micro- and macroalbuminuria with Lp(a)-increasing effects in T2DM patients (reviewed in [135]). It was therefore interesting that Mora and colleagues recently reported an inverse association between low Lp(a) concentrations and incident as well as prevalent T2DM [136]. During 13 years of follow-up of almost 27 000 initially healthy women in the Women's Health Study (WHS), 1670 subjects developed T2DM. Women with Lp(a) levels in quintiles 2–5 had approximately a 20% lower risk of incident T2DM when compared with those in the lowest quintile of Lp(a). These surprising results were confirmed by two additional case–control studies included in the same publication [136]. First, 797 WHS participants who already had T2DM at baseline (and who were therefore excluded from the primary analysis) were compared with 25 076 women who remained free of T2DM during follow-up. Second, the results were externally validated amongst 419 prevalent T2DM cases and 9233 controls from the Copenhagen City Heart Study. Both these analyses confirmed an increased risk of T2DM in subjects with low Lp(a) concentrations (Fig. 4). These findings [136] were considered to be in marked contrast to the well-known association between high Lp(a) concentrations and CVD outcomes considering that T2DM is a major risk factor for CVD. Whether other large prospective and cross-sectional studies can confirm this interesting inverse association between low Lp(a) levels and an increased T2DM risk will be important to determine. The causality of this relationship would be further supported if subjects who have small apo(a) isoforms and therefore high Lp(a) concentrations are shown to be protected from T2DM.

Figure 4.

Risk of type 2 diabetes mellitus (T2DM) for individuals with lipoprotein(a) [Lp(a)] concentrations below versus above 10 mg dL−1. Data are from the Women's Health Study (WHS) and Copenhagen City Heart Study (CCHS) [104]. Numbers in parenthesis are patients with/without T2DM.

In line with these findings by Mora et al. [136], Rainwater and Haffner observed an inverse association between Lp(a) concentrations and both fasting and 2-h insulin as well as 2-h glucose concentrations in Mexican Americans who have a high average prevalence of T2DM. A similar but positive association was observed for apo(a) isoform size and insulin and glucose levels, which suggests a causal rather than a reverse association [137]. Furthermore, experimental studies demonstrated suppression of apo(a) by insulin in hepatocytes at the post-transcriptional level [138]; therefore, a secondary effect of insulin on Lp(a) and not the reverse is likely.

Hormonal influences

Sex hormones

There are no major differences in Lp(a) levels between men and women. However, Lp(a) levels are higher in postmenopausal compared with premenopausal women [139]. In line with this, meta-analyses of women using hormone-replacement therapy with various agents demonstrated an average decrease in Lp(a) of about 25% with no major difference between oral and transdermal agents [140, 141]. In an earlier meta-analysis that investigated not only various hormonal monotherapies but also various combination therapies, strong effects on Lp(a) level of conjugated equine oestrogens alone were observed (−24%), with a less pronounced reduction with oestradiol 17-beta (−13%) and an even smaller reduction with transdermal oestradiol 17-beta (−6%). The reduction was most pronounced for combinations with progestogens (up to −34%) or tibolone (−39%) [142]. In this context, a recent study indicated the complexity of the regulatory cycles involved in Lp(a) metabolism. Persson and colleagues studied healthy women undergoing in vitro fertilization during the phases of low and high endogenous oestrogens and did not observe any changes in Lp(a) levels [143]. It was suggested that the Lp(a)-decreasing effect of oestrogens is counterbalanced by growth hormone treatment, which increases the secretion of insulin-like growth factor-1 and Lp(a) resulting in a zero net effect on Lp(a) levels [143-145]. However, it should be kept in mind that interventional treatment for in vitro fertilization with 6 days of high oestrogen levels might be different from effects caused by long-term administration of oestrogens.

Similarly, androgens including testosterone have an Lp(a)-decreasing effect. A recent systematic review of the many studies performed in this field noted that the particular setting, the dose and the route of administration result in pronounced differences in the effect (for details see [146]). One explanation for this heterogeneity is that orally administered androgens do not form oestrogens whereas intramuscular testosterone is aromatized to 17-beta oestradiols, which might subsequently cause the Lp(a) decrease [147, 148]. This is in line with orchidectomy or pharmacological castration, which reduces testosterone and oestradiol and increases Lp(a) concentrations [146].

Thyroid hormones

Several studies have investigated Lp(a) concentrations in patients with either hypothyroidism or hyperthyroidism. The findings of these studies suggest that Lp(a) is elevated in the overt hypothyroid state but probably not in subclinical hypothyroidism [149]. Therapy with thyroid hormones shows an Lp(a)-lowering effect in overt but not in subclinical hypothyroidism [150, 151]. Patients with hyperthyroidism generally showed decreased Lp(a) levels in the hyperthyroid state and an increase in Lp(a) following therapeutic intervention with thyreostatic medication or radioactive iodine therapy [152-154]. Therefore, thyroid hormones have a strong modulatory effect on Lp(a) levels; however, the mechanism of this effect is unknown.

Interest in thyroid hormones increased with the development of thyroid hormone analogues that specifically activate thyroid hormone receptor-β, which is predominantly expressed in the liver. These analogues mainly affect lipoprotein metabolism without any thyroid hormone-related effects as they have no influence on the hypothalamic–pituitary–thyroid axis [155]. These thyromimetics indeed have a consistent effect not only on lowering LDL-C but also on Lp(a) level [156]. Reduced atherosclerosis has already been demonstrated in animal studies [157, 158] but needs to be confirmed in the clinical setting.

Growth hormone

A large number of studies have been conducted to investigate the effect of growth hormone substitution on Lp(a) levels in patients with congenital or acquired growth hormone deficiency [159-161]. A strong Lp(a)-increasing effect was demonstrated; however, this negative ‘side effect’ was accompanied by a significant reduction in LDL-C caused by increased LDLR activity and by lower triglyceride levels. These effects might counterbalance the negative effects of an increase in Lp(a) concentration. Of note, patients with acromegaly have markedly elevated Lp(a) levels [162].

Lp(a) and LDL-C measurement

Apo(a) isoform dependency of Lp(a) measurements

The CNV within LPA with its multiple copies of KIV translates into a protein with the corresponding size polymorphism of Lp(a). If an antibody used for the measurement of Lp(a) is directed against the repetitive KIV-2 structure, it is expected that Lp(a) concentration will be overestimated especially in those with a high number of KIV repeats. Therefore, an antibody directed against a unique structure such as the KV or the protease domain would be most desirable. Alternatively, an antibody against apo(a) to capture Lp(a) and a second antibody against apoB-100 can be used to measure Lp(a). This latter procedure captures Lp(a) particles and, by the use of apoB for detection, each Lp(a) particle is measured only once. This method, however, ignores ‘free’ LDL-unbound apo(a), the average amount of which is considered to be about 5% but can vary within a wide range [163]. Unfortunately, most of the antibodies used for measurement have not been fully characterized, and it is not known which structures they are directed against. Many might even be directed against the repetitive KIV-2 repeat. Therefore, in the mid-1990s, Marcovina and colleagues systematically investigated various assays [164]. Briefly, when they compared an assay using an antibody against KV with one using an antibody against KIV-2, they observed that Lp(a) was underestimated by about 10% in samples with 18 KIV repeats in the latter assay. Lp(a) was overestimated by about 35% in samples with 30–32 KIV repeats. However, this overestimation is not as high as may appear; individuals with 17–19 KIV repeats have an average Lp(a) level of 50 mg dL−1, which would be underestimated by only 5 mg dL−1. If measured in individuals with large isoforms with around 30 KIV repeats, Lp(a) levels would be overestimated by approximately a third and the true values in these subjects would decrease from about 14 to 9 mg dL−1. Neither situation would have an influence on clinical decisions in the majority of subjects. False-positive and false-negative rates of 5.1% and 11.0%, respectively, were reported [164]. These rates, however, depend very much on the assays used and the apo(a) isoform-specific bias of the particular assay [165]. In the most recent meta-analysis of the effects of Lp(a) on CHD, the accuracy of Lp(a) measurement was assessed in more than 100 000 participants. However, no difference in the estimates of Lp(a) level was observed when the analysis was stratified for the apo(a) isoform sensitivity of the assay used in the various studies [5].

Overestimation of LDL-C due to high Lp(a) concentrations

As Lp(a) is based on an LDL particle with an additional glycoprotein apo(a), the measurement of total or LDL cholesterol also includes for most methods the measurement of Lp(a) cholesterol. This might result in a pronounced overestimation of LDL-C in the case of high Lp(a) concentrations. Because Lp(a) contains roughly 45% cholesterol, the effect of this overestimation can be either low (i.e. 4.5 mg dL−1 in case of an Lp(a) concentration of 10 mg dL−1) or relatively high (i.e. 45 mg dL−1 in case of an Lp(a) level of 100 mg dL−1). This peculiarity can explain for the sometimes observed nonresponse or low response to statin treatment. Statins are known to have a pronounced effect on LDL-C but not on Lp(a) [35]. If a patient with a very high risk of coronary artery disease (CAD) is prescribed a statin to lower LDL-C to a target of below 70 mg dL−1 and if that patient already has LDL-C levels around 100 mg dL−1 and an Lp(a) concentration of 100 mg dL−1, the ‘statin-accessible’ concentration of LDL-C is only 55 mg dL−1: 100 mg dL−1 measured LDL-C minus 45% of measured Lp(a), that is, 45% of 100 mg dL−1 Lp(a). In this case, it might not be surprising if there is only a very small change in LDL-C after administration of a statin. This scenario is particularly likely in individuals who have very high concentrations of Lp(a) as is the case for patients with nephrotic syndrome [126] or undergoing peritoneal dialysis [122] in whom Lp(a) concentrations can reach more than 300 mg dL−1. For example, patients with nephrotic syndrome had on average 27 mg dL−1 higher LDL-C concentrations if not corrected for cholesterol derived from Lp(a) [166]. This idea was systematically investigated in a study in hypercholesterolaemic patients with concomitantly elevated Lp(a) concentrations above 60 mg dL−1 (mean 130 mg dL−1) who were treated with statins. A decrease in ‘true’ LDL-C proportional to the pretreatment ratio of true LDL-C/Lp(a)-derived cholesterol was observed [167]. Similarly, Miltiadous et al. [168] reported that the percentage decrease in LDL-C was inversely correlated with Lp(a) concentration in FH patients treated with statins. Finally, a recent GWA study investigating the LDL-C response to statin treatment identified besides the APOE gene region also the LPA gene to have a genome-wide significant effect on the LDL-C response [169]. Whereas the APOE association is a clear pharmacogenetic effect, the association with the LPA gene has to be considered as a ‘pseudo-pharmacogenetic’ effect that results from an incorrect measurement of LDL-C as we discussed recently [166].

Genetic evidence supports Lp(a) as an independent genetic risk factor for CHD

At present, genetic variation in the LPA gene is arguably the strongest single common genetic risk factor known for CVD. Given the large GWA studies that have been performed to search for gene variants associated with lipid levels as well as CHD [102, 170], it is unlikely that a stronger common variant will be identified. Therefore, it seems surprising that the question of whether Lp(a) is a risk factor for CVD remained unanswered for a long time. The reason for this is that Lp(a) is a quantitative trait and that reverse causality could not be rigorously excluded. It was argued that the elevated Lp(a) levels in patients with CVD might be the consequence rather than the cause of disease (Fig. 5); however, this possibility cannot be completely excluded even by prospective studies. With the exception of a few early studies, most prospective studies and subsequent meta-analyses [5] provided strong evidence for a causal role of Lp(a) in CHD although relative risk (RR) values were low. Only genetic studies following the Mendelian randomization approach could finally exclude reverse causality as the reason for elevated Lp(a) in CHD (Fig. 5). The rationale for such studies was defined at a European Lipoprotein Club Meeting [171]. Applied to Lp(a) it is as follows: given that high Lp(a) concentrations are associated with CHD and that variants in the LPA gene are associated with high Lp(a) concentrations, these variants must also be associated with CHD if causality underlies the association. An ideal candidate to test this hypothesis was the size polymorphism of apo(a)/KIV-2 CNV, which is strongly associated with Lp(a) levels [81]. Early studies of apo(a) isoforms in FH patients [172] and Chinese CHD patients and controls [173], which first applied the Mendelian randomization approach in practice, followed by a multicentre, multi-ethnic study including >1000 CHD cases [174] demonstrated that the KIV-2 CNV in the LPA gene is indeed associated with CHD. In the latter study, the odds ratio (OR) in the pooled sample was 1.78 for small apo(a) isoforms (Table 2). Several follow-up studies, including one in which KIV-2 repeats, were determined by PFGE/Southern blotting [175], and a meta-analysis of these studies confirmed the association [6]. The meta-analysis included 40 studies with 11 396 cases and 46 938 controls of which 30 studies with 7382 cases and 8514 controls applied broadly comparable phenotyping and analytical methods. Smaller apo(a) isoforms were associated with a twofold increased risk of CHD compared with large isoforms [RR = 2.08, 95% confidence interval (CI) 1.76–2.58)] [6]. Six studies in this meta-analysis focused on ischaemic stroke as the endpoint (718 cases and 1637 controls), and a combined RR for ischaemic stroke of 2.14 (95% CI 1.85 to −2.97) was observed [6] (Table 2). These risk estimates are much greater than the effects of common variants identified in recent GWA studies [170] and are of clinical relevance considering that about 25–35% of the population carry small apo(a) isoforms.

Figure 5.

Mendelian randomization approach to demonstrate a causal association between lipoprotein(a) [Lp(a)] concentration and coronary heart disease (CHD). Because a low number of KIV copies (11–22 copies) are associated with high Lp(a) levels and high Lp(a) levels are associated with CHD, it follows that a low number of KIV copies will be associated with CHD if the association between Lp(a) and CHD is causal. As the latter is indeed the case, reverse causation (i.e. that CHD is secondarily causing an increase in Lp(a) levels) can be excluded.

Table 2. Risk estimates for genetic variants in the LPA gene region: data from key studies over the last two decades
StudyStudy description and endpointSample sizeEstimateNotes
  1. CCHS, Copenhagen City Heart Study; CGPS, Copenhagen General Population Study; CIHDS, Copenhagen Ischemic Heart Disease Study; MAF, minor allele frequency (i.e. frequency of risk allele); f, frequency.

  2. a

    Significant when compared with the reference group (ref) of the upper quartile of the sum of KIV-2 repeats.

Sandholzer et al. [174]CAD patients and controls from six ethnic groups= 1013 CAD cases= 1570 controlsPooled OR for small isoforms B, S1 and S2 = 1.78First multicentre multiethnic study
Kraft et al. [175]Case–control study of CHD patients undergoing coronary angiography= 69 cases= 69 controls

17–19 KIV repeats: OR 4.63 (95% CI 1.24–17.23)

20–22 KIV repeats: OR 2.08 (95% CI 0.93–4.66)

23–25 KIV repeats: OR 1.09 (95% CI 0.49–2.42)

>25 KIV repeats: 0.31 (95% CI 0.16–0.64)

First study investigating the number of KIV repeats by pulsed-field gel electrophoresis
Kamstrup et al. [7]

Prospective CCHS: MI

Cross-sectional CGPS: MI

Case–control CIHDS: MI

= 8637 (599 MI)

n = 29 388 (994 MI)

n = 2461 (1231 MI)

HR in four quartiles of sum of KIV: 1.5a/1.3a/1.1/ref

OR in four quartiles of sum of KIV: 1.3a/1.1/0.9/ref

OR in four quartiles of sum of KIV: 1.4a/1.2a/1.3a/ref

First study using the sum of KIV-2 repeats on both alleles determined by real-time PCR
Erqou et al. [6]Meta-analysis of 40 studies including 11 396 cases and 46 938 controls= 7382 CHD, 8514 controls= 718 strokes, 1637 controls

RR for small versus large isoforms: CHD: RR 2.08 (95% CI 1.76–2.58)

Ischaemic stroke: 2.14 (95% CI 1.85–2.97)

First large meta-analysis of apo(a) isoforms. Risk estimates are derived from studies that used comparable methods
Clarke et al. [9]Meta-analysis of coronary disease cases and controls= 7991 cases= 7946 controls

rs10455872 (MAF = 0.07): OR 1.47 (95% CI 1.35–1.60)

rs3798220 (MAF = 0.02): OR 1.68 (95% CI 1.43–1.98)

First large study using two SNPs instead of KIV repeats
Trégouët et al. [8]Genome-wide haplotype association study using four SNPs from the SLC22A3-LPAL2-LPA gene cluster= 8999 CAD, 10 263 controls

Haplotype CCTC (f ≈ 2%): OR 1.82 (95% CI 1.57–2.12)

Haplotype CTTG (f ≈ 14%): OR 1.20 (95% CI 1.13–1.28)

First study using a genome-wide haplotype approach
Schunkert et al. [170]Meta-analysis of 14 GWA studies of CAD= 32 584rs3798220 (MAF = 0.02): OR 1.51 (95% CI 1.33–1.70)Large meta-analysis of GWA study data

Recently, four large studies, including two GWA studies [8, 170], a candidate gene approach using SNPs [9] and a Mendelian randomization study applying qPCR to quantify genomic KIV-2 repeat number [7], further confirmed and extended understanding of the associations between LPA gene variants and Lp(a) levels and myocardial infarction (MI) (Table 2). In the candidate gene study with almost 8000 CHD cases and 8000 controls, several SNPs in the LPA region were found to be associated with MI, the strongest being rs10455872 and rs3798220 [9]. rs10455872 is an intronic SNP but rs3798220 results in an amino acid substitution (Ile4399Met) in the protease domain of LPA. Both SNPs are associated with short KIV-2 repeats and high Lp(a) levels. These SNPs were associated with a 1.47- and 1.68-fold increased risk of CHD compared with noncarriers, respectively. A genotype score involving both SNPs revealed an OR for coronary disease of 1.51 for carriers of one variant and 2.57 for carriers of two or more variant alleles [9]. These findings were extended in a recent meta-analysis showing that the minor alleles of rs10455872 and rs3798220 increased the risk of CHD in carriers by 42% and 57%, respectively [176]. However, it should be noted that these two SNPs have only a minor allele frequency of 7% and 2%, respectively, and can therefore only explain a fraction of all small apo(a) isoforms. Using these two SNPs probably underestimates the risk but clearly underscores the importance of LPA as a risk gene for CAD, especially as the other genes for CAD detected by GWA studies were associated with OR values between 1.06 and 1.29 [170]. However, when the causative or the most appropriate variants are found, the risk estimates may well increase for some of these loci. The example of LPA very clearly demonstrates that SNPs can provide evidence of an association but underestimate the association if the SNPs only tag the underlying more complex genetic variants such as CNVs or other repeat structures. Similar observations were made for bilirubin where a TA-repeat polymorphism in the promoter of the UGT1A1 gene is tagged by nearby SNPs, which underestimate the association between the causative variants and bilirubin levels [177]. However, the accuracy of the estimation may improve with better imputations based on the 1000-Genomes Project.

The MI GWA study by Trégouët et al. [8] was the first to use haplotype information instead of single SNPs to search for associations. A haplotype composed of four SNPs, two in LPA (rs7767084 and rs10755578) and two in the neighbouring genes LPAL2 (rs3127599) and SLC22A3 (rs2048327), gave a strong association signal. The rare haplotype CCTC with a frequency of approximately 2% was associated with an 82% higher risk (OR 1.82, 95%CI 1.57–2.12) and the more common haplotype CTTG with a frequency of about 16% had a 20% higher risk (OR 1.20, 95%CI 1.13–1.28) when compared with the most frequent TCTC haplotype [8] (Table 2). The meta-analysis of 14 CAD GWA studies reported a 51% higher OR for the rare compared with the common allele of SNP rs3798220 (MAF = 0.02) [170]. It is interesting that no association was found between SNPs in LPA and early atherosclerosis in a study of 2000 Finnish young adults [111]. The reason for this inconsistent finding is unclear.

Using the Mendelian randomization approach with qPCR, Kamstrup and colleagues quantified the copy number of KIV-2 repeats in >40 000 subjects in the Copenhagen City Heart Study. Individuals with a low total sum of KIV-2 repeats from both alleles in their genome (first quartile) had an adjusted hazard ratio (HR) for MI of 1.50 compared with those with a high sum or repeats (fourth quartile) (Table 2). The KIV-2 CNV explained only about 25% of the variability in Lp(a) levels [7], which is lower than in other studies of European populations. However, the risk and explained variability are probably underestimated due to the characteristics of the applied method of qPCR; in contrast to studies using apo(a) isoforms or determination of KIV-2 repeats in alleles separated by PFGE, qPCR measures the total number of KIV-2 repeats of the two apo(a) alleles. Consequently, individuals with one very short KIV-2 repeat allele and one very large allele will be included in the same category as individuals with two intermediate copy number alleles. These two situations are, however, associated with very distinct Lp(a) concentrations. In the same study, Lp(a) levels for the 90th–95th percentile and above the 95th percentile were associated with multifactorially adjusted HRs for MI of 1.9 and 2.6, respectively, compared with levels below the 22nd percentile.

An important question is whether the associations with genetic variants in LPA reflect a causal effect of the SNPs, or KIV-2 repeats, or whether the variants are associated with and are only surrogate markers for high Lp(a), which is the pathogenic agent. rs3798220 is a coding SNP (I4399M), which therefore might be causally related to CHD risk. Luke et al. [178] have suggested that the I4399M variant in the protease domain of apo(a) is a risk factor independent from Lp(a) levels. Furthermore, a post hoc analysis including more the 25 000 Caucasian women from the WHS revealed a 56% relative risk reduction from aspirin therapy in carriers of the 4399M variant, whereas noncarriers of that allele did not benefit from aspirin therapy [179]. Similar observations were made in the Atherosclerosis Risk in Communities (ARIC) Study: amongst nonusers of aspirin, carriers of the 4399M variant were at an increased risk whereas the risk was similar amongst carriers and noncarriers of this variant if aspirin was used [180]. This is an interesting example of pharmacogenetics demonstrating that carriers of the high-risk variant 4399M might be able to compensate for the genetically elevated risk of CVD by the use of aspirin.

In the large studies by Trégouët et al. [8] and Clarke et al. [9], the effect of the SNPs on risk disappeared after adjustment for Lp(a) levels. Moreover, Clarke and co-workers [9] found that rs3798220 and rs10455872 are in LD with short KIV-2 repeats and high Lp(a) levels. Hence these data provide strong evidence that genetically determined high Lp(a) and not a structural variant is associated with CHD risk. As outlined, most but not all cases of short KIV-2 alleles are associated with high Lp(a) concentrations. It will be interesting to determine which fraction of short repeat/high Lp(a) alleles is tagged by SNPs rs3798220 and rs10455872.

Together the overwhelming genetic evidence has established Lp(a) as an emerging genetic risk factor for CVD. Of note, it is independent of other classical risk factors including lipids.

Lp(a) as a risk factor for CHD in different ethnic groups

In case–control studies, Lp(a) has been associated with CHD risk in European and Asian populations, but not unequivocally in African Americans [181, 182]. However, recent prospective data from the sufficiently powered ARIC Study indicate that the increased risk of CHD was at least as strong in African Americans as in white North Americans [183]. The situation is more complex for the association between the KIV-2 CNV and CHD. The KIV-2 CNV has been shown to be associated with CHD in different European populations, white North Americans and some (Chinese and Japanese) [6] but not other (Asian Indian) [184] Asian populations. One study of African Americans demonstrated an association between small apo(a) isoforms and high Lp(a) concentration in men but not women [181], whereas no association with isoform size was found in another study in this ethnic group [182]. In the Dallas Heart study, no association between Lp(a) levels or apo(a) isoforms and coronary calcification was found in African Americans [185]. A study amongst Asians from South and North India, using PFGE/Southern blotting to type the CNV in LPA, was particularly revealing [184]. It demonstrated the limitations of the Mendelian randomization approach, if markers selected in one population were used in studies of other ethnic groups. Lp(a) concentrations were highly significantly associated with MI risk in both Indian samples; however, KIV-2 repeats were not [184]. The authors suggested that this may be due to a difference in the genetic architecture of the Lp(a)/LPA trait between Indians and Europeans. In Europeans, short KIV-2 repeats are associated with very high Lp(a) levels whereas intermediate and high copy number alleles are associated with low or very low concentrations. By contrast, the inverse correlation between Lp(a) levels and KIV-2 repeats, although present and significant, is weaker in Asian Indians, and high Lp(a) levels are present over the whole range of KIV-2 repeat strata. Hence one of the prerequisites of a Mendelian randomization study, namely a strong association between the genetic variant and the putative risk factor, was not sufficiently fulfilled. Together with findings in other populations, the results of this study in Asian Indians support the notion that Lp(a) level and not KIV-2 number is the cause of increased CHD risk.

Lp(a) in FH

Classical FH is caused by mutations in the LDLR gene, which result in increased levels of LDL-C in plasma and a high risk of early CHD [186]. The presence of LDL as one component of the Lp(a) particle suggested that not only LDL but also Lp(a) may be catabolized by the LDLR pathway. This hypothesis has been tested by different approaches, which yielded conflicting results. The clinical phenotype of FH is very variable. Therefore, a further clinically important question is whether high Lp(a) confers additional CHD risk to FH patients over and above the risk posed by their elevated LDL-C. In the majority of larger and well-controlled studies, elevated Lp(a) levels were observed in FH patients compared with controls, and it is now generally accepted that Lp(a) is elevated 2- to 3-fold in FH patients compared with control subjects matched for the KIV-2 CNV (usually determined by apo(a) isoform size). Because of the very large inter-individual variation of Lp(a) levels and their distinct genetic control, studies intended to detect differences between groups (e.g. FH vs. non-FH) are easily biased if they do not control for the major effect of LPA gene variation or if they do not include a large number of participants. Probably the most convincing evidence for an in vivo effect of LDLR mutations on Lp(a) levels was provided by a large sib-pair study [187] and a study of South African FH families [188]. These studies included molecularly defined homozygous and heterozygous FH patients, in whom KIV-2 repeat genotypes and apo(a) isoforms by immunoblotting were determined. A clear dose effect of defective LDLR alleles on Lp(a) levels was observed [187, 188]. Moreover, this dose effect was present over the whole range of KIV-2 repeats (Fig. 6). Hence it appears that the FH status is associated with high Lp(a) levels, and elevated Lp(a) is considered to be part of the clinical syndrome of FH.

Figure 6.

Average lipoprotein(a) [Lp(a)] concentrations associated with five groups of binned apolipoprotein(a) [apo(a)] alleles (according to the number of KIV repeats). In each group, Lp(a) level increases with the number of low-density lipoprotein (LDL) receptor mutations demonstrating a positive gene–dosage effect. P values indicate statistical significance calculated for the three genotypic groups by KruskalWallis test and adjusted for multiple comparisons by the Bonferroni method. Figure adapted and reprinted with permission from [188].

What is not clear is the mechanism behind the elevated Lp(a) in FH. The most obvious means, that is, that Lp(a) is removed from the circulation by binding to the LDL receptor, has not been supported by most in vitro cell culture studies [see section above: Metabolism of Lp(a)] and is also inconsistent with other evidence. In vivo turnover studies by Rader et al. [189] determined the fractional catabolic rates of radioactively labelled Lp(a) and LDL in heterozygous and homozygous FH individuals and controls. No significant difference was found between the groups thus excluding the LDLR pathway as a major route of Lp(a) catabolism. However, this conclusion has been questioned by a re-analysis of the kinetic data [188]. Overall, the LDLR binding studies and turnover data suggest that Lp(a) is a ligand for the LDLR although with lower affinity than LDL. However, further investigation, for example, with turnover studies using stable isotopes, is required for verification.

An interesting observation is that the variability of Lp(a) levels is larger in sib-pairs with FH [187] and familial defective apoB [190] carrying KIV-2 repeat alleles identical by descent than in control sib-pairs. This indicates a non-LDL receptor-dependent mechanism. In line with this, results from segregation analysis in a large family study including 220 individuals from 14 families led Friedlander and Leitersdorf [191] to conclude that the elevation of Lp(a) levels in FH family members is due to the major effect of a nontransmitted environmental factor rather than the LDLR defect.

Whether high Lp(a) is a risk factor for CHD in FH patients has been addressed by several studies. Two early studies reported that Lp(a) is a CHD risk factor in FH patients [172, 192]. In one of these studies, which included 109 FH patients, the Mendelian randomization approach was applied to show that apo(a) isoforms predicted risk of CHD but not independent from Lp(a) levels [172]. Furthermore, the high Lp(a) in FH patients with CHD was shown to be the result of a genetic influence of the LPA locus [172]. Results from further studies with >200 patients each, but not controlled for the KIV-2 CNV, were controversial as discussed in [193]. However, two large studies [193, 194], one of which included >1600 FH patients [194], confirmed the initial findings and demonstrated that high Lp(a) is a CHD risk factor in FH patients. In one of these studies, Lp(a) concentrations >58 mg dL−1 were associated with a 2.59-fold increased risk [193]; in the other study, an RR of 1.50 for Lp(a) levels >30 mg dL−1 was calculated [194]. The true risk conferred by high Lp(a) level in FH patients is, however, unclear. All these investigations were retrospective case–control studies, whereas a large prospective study should be conducted in FH patients with measurement of baseline Lp(a) levels and possibly determination of LPA gene variation.

Application in clinical practice

The above-mentioned meta-analysis of the effect of Lp(a) levels on CHD reported an RR of 27% higher in the highest tertile compared with the lowest tertile [5], which is of similar magnitude to the risk associated with LDL-C tertiles [195]. At present, there are no proven dietary or therapeutic interventions that effectively lower Lp(a) and at the same time reduce CHD outcomes except apheresis procedures (Table 3). Statins, which dramatically lower total cholesterol and LDL-C, do not decrease Lp(a) concentrations significantly [35]. Therefore, any therapeutic regimens targeting patients with high Lp(a) must be indirect. The European Atherosclerosis Society (EAS) recently published a consensus statement [10] that recommended screening for high Lp(a) levels in subjects with intermediate or high risk of CVD or CHD, that is, those with premature CVD, FH, family history of premature CVD or high Lp(a), recurrent CVD despite statin therapy, ≥3% 10-year risk of fatal CVD according to European guidelines or ≥10% 10-year risk of fatal and/or nonfatal CHD according to US guidelines. If these recommendations are indeed followed, especially screening for those with ≥3% 10-year risk of fatal CVD or ≥10% 10-year risk of fatal and/or nonfatal CHD, a large number of individuals particularly above 65 years of age would require Lp(a) measurement. Although the above-mentioned meta-analysis demonstrated that risk increases linearly and without a threshold over the full range of Lp(a) concentrations [5], an arbitrary cut-off value of 50 mg dL−1 was recommended in the consensus statement [10]. This is higher than the 30 mg dL−1 that for a long time had defined the upper limit of the ‘normal’ range.

Table 3. Lipid-lowering treatment and the effect of this treatment on changes in lipoproteins including lipoprotein(a) and clinical endpoints
Treatment protocolPatient descriptionDosageObservation periodRelative change from baseline (%)Clinical endpoints
Lp(a)LDL-CHDL-CTGapoB
  1. CAD, coronary artery disease; CVD, cardiovascular disease; MTP, microsomal triglyceride transfer protein; CETP, cholesteryl ester transfer protein; n.a., not applicable; n.d., not determined; s.c. subcutaneously.

  2. a

    See text and Fig. 7 for discussion.

  3. b

    Composite endpoint consists of death from coronary heart disease (CHD), nonfatal MI, ischaemic stroke, hospitalization for an acute coronary syndrome, or symptom-driven coronary or cerebral revascularization.

Lipid apheresis combined with lipid-lowering drugs [201]120 patients with CAD and Lp(a) >95th percentileLipid apheresis every week, 2 weeks or 10 days5 years−73−65−15n.d.n.d.86% decrease of major coronary eventsa
Extended-release niacin [199]Patients with atherosclerotic CVD and LDL-C < 70 mg dL−1 treated with simvastatin plus ezetimibe (1718 niacin and 1696 placebo)1500–2000 mg3 years−25−12+25−29−14No effect on composite endpointb
Anacetrapib (CETP inhibitor) [200]Patients with CHD or at high risk of CHD (762 anacetrapib and 768 placebo)100 mg24 weeks (lipid changes) and 76 weeks (CVD endpoints)−36−40+138−7−21CVD events and death: 2.0% vs. 2.6%, = 0.40
Mipomersen (antisense oligonucleotide targeting apoB mRNA) [34]Homozygous FH with maximum tolerated dose of lipid-lowering medication (34 mipomersen and 17 placebo)200 mg per week s.c.26 weeks−31−25+15−17−27n.a.
Eprotirome (thyroid hormone analogue) [156]137 hypercholesterolaemia patients treated with statins and 47 receiving placebo25 μg50 μg100 μg12 weeks −27−32−43 −22−28−32n.d. −16−16−33 −20−24−33n.a.
Lomitapide (MTP inhibitor) [36]84 patients with hypercholesterolaemia10 mg ezetimibe (= 29), 10 mg lomitapide (= 28), both (= 28)12 weeks +3−17−16 −20−30−46 +6−6−9 −4−10−15 −15−24−37n.a.

Furthermore, the EAS consensus panel recommended treatment with niacin to lower Lp(a) below 50 mg dL−1 as a valid option in those patients with premature CVD, FH, family history of premature CVD or high Lp(a) and those with recurrent CVD already receiving aggressive LDL-C-lowering treatment with statins [10]. However, this recommendation may be premature as direct proof is lacking that the beneficial effect of niacin is due to lowering Lp(a) and because this recommendation would target a large number of individuals. Niacin lowers Lp(a) by about 25–30% and has been reported to prevent CHD events [196]. Whether the prevention of such events is caused by the lowering of Lp(a), however, is unclear. Niacin increases retention of apo(a) at the hepatocyte surface thereby inhibiting its secretion and lowering Lp(a) [197]. In addition, it was recently shown that niacin reduces the expression of apo(a) in transgenic animals as well as cell culture studies [198]. Niacin also lowers triglycerides, raises HDL-C and affects blood coagulation. Because of the extremely skewed distribution of Lp(a) levels, the finding that niacin lowers Lp(a) by 30% may also be misleading. Indeed, lowering Lp(a) from 15 mg dL−1 (which is about the average level in most European populations) to 10 mg dL−1 is certainly not the same as lowering the concentration from 100 to 70 mg mL−1. Of concern is the fact that a very recent large trial of the effect of niacin had to be stopped after an average follow-up period of 3 years due to lack of efficacy; despite significant improvements in triglycerides, HDL-C and Lp(a), the rate of composite endpoints did not differ significantly between the niacin and placebo groups [199].

Koch's postulate for Lp(a): is lowering Lp(a) beneficial?

If Koch's postulate is extrapolated to the situation of Lp(a), it would underscore causality of Lp(a) for CVD if (i) the presence of high concentrations of Lp(a) is associated with CVD (which was clearly demonstrated in numerous studies), and (ii) if removal of high concentrations is protective against CVD. Therefore, a key question is whether patients with high Lp(a) will benefit from lowering Lp(a) plasma concentrations. Furthermore, whether Lp(a) should be lowered in primary prevention, secondary prevention or only in highly selected risk groups is unclear as there are no simple ways to lower Lp(a) effectively and without changing other risk factors at the same time (Table 3). It will be difficult, if not impossible, to separate effects on LDL-C and Lp(a) lowering as well as on other changes in lipoproteins. At present, there are six therapeutic options for lowering Lp(a), including eprotirome (a thyroid analogue) [156], lomitapide (an MTP inhibitor) [36], anacetrapib (a cholesteryl ester transfer protein inhibitor) [200] and mipomersen (an antisense oligonucleotide against apoB mRNA) [34]. Most of these potential therapies are in Phase III clinical trials to determine efficacy in terms of ability to reduce plasma Lp(a). Furthermore, treatment with extended-release niacin [196] decreased Lp(a) but was not very effective in terms of reducing clinical endpoints as demonstrated recently [199]. LDL apheresis appears to have the same disadvantages as it removes Lp(a) and LDL-C simultaneously and can therefore not disentangle effects on LDL-C and Lp(a) lowering. However, Jaeger et al. [201] have recently overcome this problem. They selected a group of patients with excessive Lp(a) levels who continued to experience a high rate of major adverse coronary events (MACE) despite aggressive drug treatment and successful lowering of LDL-C in the first stage of the study. These patients then underwent LDL apheresis (Fig. 7). Lp(a) was lowered in this second stage on average by 73%, and the rate of MACE decreased dramatically by 86% from preapheresis (stage 1) to the 5-year follow-up (stage 2). However, this decrease in MACE might have been due to further lowering of LDL-C levels. The authors therefore analysed a subgroup of patients with apparently low LDL-C before the start of apheresis (≤100 mg dL−1). In fact, LDL-C was virtually absent in these patients because the measured ‘LDL-C’ in their plasma was mainly cholesterol due to their very high Lp(a) concentrations, and the true LDL-C was on average only 23 mg dL−1. Hence apheresis during subsequent years in stage 2 selectively and dramatically lowered Lp(a) in these patients, and the true LDL-C dropped only from 23 to 18 mg dL−1. Of most interest, the effect on MACE was of the same magnitude in the two subgroups with LDL-C concentrations above and below 100 mg dL−1 (−85% vs. −89%, respectively). In other words, the very small reduction in the true LDL-C in the low LDL-C group by 5 mg dL−1 is too small to explain this dramatic reduction in MACE by 89% [201]. Although this study is limited by its retrospective and observational nature, it is nevertheless the first to provide evidence that lowering Lp(a) is beneficial, at least in secondary prevention in some high-risk patients [202].

Figure 7.

Design and summary of the results of the study by Jaeger et al. [201] to investigate the effect of lipoprotein(a) [Lp(a)] lowering by lipid apheresis in very high-risk patients. ‘True LDL-C’ is considered to be LDL-C lacking the cholesterol from Lp(a) particles. The Lp(a)-derived cholesterol is not accessible for therapeutic intervention with statins and comprises about 45% of the Lp(a) mass (in mg dL−1). See text for further explanation.

Concluding remarks

Fifty years after its discovery, Lp(a) is experiencing a renaissance and is becoming again the focus of clinical interest for two main reasons. First, major clinical studies using SNPs have provided access to large population-based and clinical data with the availability of DNA and Lp(a) measurements as well as clinical outcomes. This enabled high-throughput studies to be performed and the Mendelian randomization approach to be applied, which clearly confirmed the causal association between high Lp(a) concentrations and CVD outcomes. Second, genetic studies and particularly GWA studies demonstrated that the expected risks associated with common genetic variants in other gene loci are very low, in most cases with RR values between 1.10 and 1.30. Thus, the RR associated with small apo(a) isoforms with a doubling of risk for outcomes is remarkable.

Some surprising results in recent years [e.g. an increased risk of T2DM in subjects with low Lp(a) concentrations] have demonstrated that much still remains unknown about this mysterious lipoprotein particle. We believe that understanding of the physiology and pathophysiology of Lp(a) and the associated risks will change over the next years as systems biology approaches and large studies attempt to confirm and extend earlier studies.

Conflict of interest statement

No conflicts of interest to declare.

Acknowledgements

The authors' work discussed in this review was supported by the Austrian Science Fund, the Austrian National Bank, the Austrian Genome Research Programme GEN-AU for the project Genomics of Lipid-associated Disorders (GOLD) and the Standortagentur Tirol. The authors have no relevant financial considerations to disclosure.

Ancillary