Human Genetics and the Causal Role of Lipoprotein(a) for Various Diseases

The interindividual range of Lp(a) concentrations is very wide from less than 0.1 mg/dL to more than 300 mg/dL. Few people lack Lp(a) in their plasma [19, 20]. The broad range in the Lp(a) distribution is known for all populations and is highly skewed towards low levels in most ethnic groups. Figure 1a demonstrates in a population-based study from Southern Germany that roughly 50 % of the individuals have concentrations below 10 mg/dL. Approximately 25 % of this typical population have concentrations above 30 mg/dL and 14 % have concentrations above 50 mg/dL. Both thresholds have been considered to be associated with an increased risk for cardiovascular disease [1, 21]. In contrast to European and Asian populations sub-Saharian Africans show markedly higher Lp(a) concentrations with a distribution that is closer to a Gaussian distribution.

These substantial differences between individuals are to a large extent genetically determined. Early family studies established the genetic nature of the trait and twin studies found that the heritability of Lp(a) is very high, exceeding 90 % in populations of European and African descent [22, 23]. Lp(a) is therefore the lipoprotein with the strongest genetic control. The discovery of the size polymorphism of apo(a) in serum [24] and KIV-2 CNV in the LPA gene [6, 25‐27] resulted in the identification of the LPA gene as the major gene for Lp(a) levels. Association and sib-pair linkage studies [27‐31] revealed very soon that the KIV repeat size polymorphism explains a major part of this variance ranging from 30 to 70 % depending on the ethnic population [1]. Individuals expressing a low number of K-IV repeats resulting in so-called small apo(a) isoforms (up to 22 KIV repeats) have on average markedly higher Lp(a) concentrations than individuals carrying only large apo(a) isoforms (more than 22 KIV repeats) (Fig. 1b).

Despite this pronounced inverse correlation between the number of KIV repeats and the Lp(a) concentrations, there is still a wide variability in Lp(a) concentrations within each KIV repeat group. That means isoforms of the same size differ widely in concentrations. This suggests that besides non-genetic factors other genetic variants than the KIV repeats are major contributors to Lp(a) concentration variation and that there are pronounced differences between ethnicities. And indeed several dozens of genetic variants within the LPA gene region have been described in the meanwhile and for most of them, the functional consequences are not yet known. This has to be seen in the context that a genetic variant can be in strong linkage disequilibrium with the real causal variant and without functional studies it will be hard to determine which variant is functionally responsible for Lp(a) concentration variability. One of the first additional polymorphisms besides the KIV repeat polymorphism described to be associated with Lp(a) concentrations was a pentanucleotide repeat polymorphism in the promoter region of the LPA gene [32, 33]. This polymorphism explains up to 14 % of the Lp(a) variation in Europeans but shows no association in Black Africans [32]. However, functional promoter studies suggested that causal variants in linkage disequilibrium with the pentanucleotide repeat polymorphism rather than pentanucleotide repeat polymorphism itself are contributing to Lp(a) concentrations [34, 35].

There are two SNPs known which are functionally characterized and which result in non-expressed apo(a) alleles. A +1 donor splice site mutation (rs41272114) with a frequency of about 6 % in Europeans results in alternate splicing leading to a truncated apo(a) isoform with congenital deficiency of Lp(a) [19, 36]. Another SNP in the first exon of KIV type-2 introduces a stop codon resulting in a truncated apo(a) protein. This nonsense mutation causing a null allele is observed in a frequency of 2 % in Europeans [20].

Besides these well characterized variants several dozens of SNPs in the wider LPA region have recently been brought to the attention of the scientific community [37]. Many of them show pronounced associations with Lp(a) concentrations (e.g., rs10455872 and rs3798220) but the functional significance is not yet clear. Furthermore, since sequencing of the highly repetitive and large KIV type-2 repeat is until now hard to accomplish, we might expect some surprises on genetic variants in those regions which could have a pronounced effect on Lp(a) concentrations.

The search for Lp(a)-modifying genes using genome-wide association studies (GWAS) is in progress. GWAS performed up to now were strongly limited by sample size, focus on certain subgroups (e.g., diabetes mellitus, population isolates) or by the use of a specialized candidate gene-chip. Due to these limitations they were only able to identify the well-known region on chromosome 6q27 harboring LPA, PLG, SLC22A3 [38‐42] and very recently the APOE gene locus in African Americans [43]. This finding will require confirmation in different populations and functional elucidation. It will require a large number of samples to identify further genes contributing to Lp(a) levels, should such loci exist. A conditional analysis adjusted for the effects of the LPA locus and especially the apo(a) isoform size will tremendously increase the power of such GWAS.

Two- to three-fold elevated Lp(a) levels were observed in patients with familial hypercholesterolemia vs. controls in the majority of larger and well controlled studies that were matched for apo(a) isoform size. These and other observations suggested that not only LDL but also Lp(a) may be catabolized by the LDLR pathway. A large sib-pair study [44] and a study of South African families with familial hypercholesterolemia [45] included molecularly defined homozygous and heterozygous patients, in which KIV-2 repeat genotypes and apo(a) isoforms were determined by Westernblots. They observed a clear dose effect of defective LDLR alleles on Lp(a) levels: when they binned apo(a) alleles by size they found in each isoform group Lp(a) levels to increase with the number of LDL receptor mutations demonstrating a positive gene dosage effect.

The mechanism behind the elevated Lp(a) in familial hypercholesterolemia is not clear. In vitro cell culture studies and in vivo turnover studies [46] excluded the LDLR pathway as a main route of Lp(a) catabolism. This is also in line with the observation that statins that result in an overexpression of LDL receptors do not lower Lp(a) concentrations, despite a pronounced effect on LDL-cholesterol (LDL-C) concentrations [47‐49]. Therefore, it is actually not clear which gene really influences Lp(a) concentrations in the case of familial hypercholesterolemia.

There was a major discussion in the mid-1990s whether Lp(a) is indeed a risk factor for CVD. In the meantime there is very strong evidence that increasing Lp(a) concentrations are associated with an increasing risk for CVD. Numerous studies have been published over the last three decades. Not all were positive and this resulted in major speculations on the reasons why some were negative. These speculations included the selection of appropriate controls, selection of patients, non-standardization of assays [50, 51] and sample storage effects [52], to mention only a few. One of the largest studies up to now, the Copenhagen City Heart Study, observed a 1.6-fold increased risk for an incident myocardial infarction for concentrations between 30 and 76 mg/dL (67th – 90th percentile) compared to individuals with Lp(a) concentrations below 5 mg/dL (<22nd percentile). This risk increased to 1.90 for individuals with Lp(a) concentrations between 77 and 117 mg/dL (90th - 95th percentile) and to 2.60 for individuals with Lp(a) concentrations above 117 mg/dL (>95th percentile) [53] (Fig. 3, panel a).

This pronounced association, however, is at first glance not proof that Lp(a) is causally related to CVD even if the results come from several prospective studies. As already mentioned, a major discussion dominated the field in the 1990s since reverse causation could not be convincingly excluded. Reverse causation means that elevated Lp(a) in patients with CVD might be the consequence rather than the cause of the disease. Only genetic studies following the principle of the Mendelian randomization approach finally excluded reverse causation as a reason for elevated Lp(a) in CVD. This approach became quite popular during the last two decades and was applied for the first time ever with Lp(a). It is based on a triangle of observations as illustrated in Fig. 3. The first leg starts with the finding that a biomarker (in our case high Lp(a) concentrations) is associated with CVD (Fig. 3, panel a). This association could be either causal or the result of reverse causation in the sense that the outcome CVD is changing the biomarker secondarily. Furthermore, the association can be influenced by several confounders. The second leg of the triangle is that certain alleles of genes (in our case this was the number of KIV repeats determining apo(a) isoform size) have an influence on Lp(a) concentrations (Fig. 3, panel b). Figure 4 is a simplified illustration how this random assortment of alleles from parents to offsprings works. If one of the parents carries one small apo(a) isoform (usually associated with high Lp(a) concentrations) and one large apo(a) isoform (associated with low concentrations), it is randomly determined at the time of conception which of the two alleles will be transmitted to the offspring. This holds also true for the alleles of the other parents. It will thereby be determined in this early stage whether someone is exposed all their life to lower or higher Lp(a) concentrations. If Lp(a) concentrations indeed influence the risk for CVD, the triangle will be finalized by the third leg (Fig. 3, panel c) as one would expect to see a preponderance of small apo(a) isoforms in persons who develop CVD. And this was indeed the case when the first studies were published in patients with familial hypercholesterolemia with and without CHD [54] and Chinese CHD patients and controls [55] and by a multicenter-multiethnic study including >1000 CHD cases [56]. In the latter study the OR in the pooled sample was 1.78 for small apo(a) isoforms. These were actually the first studies applying the Mendelian randomization approach in practice, although they did not use this term which was only introduced in 2003 [57]. Several follow-up studies, including one in which KIV-2 repeats were determined by pulsed-field gel electrophoresis [58] and a meta-analysis of these studies confirmed the association [59]. The meta-analysis included 40 studies with 11396 cases and 46938 controls and 30 studies with 7382 cases and 8514 controls that applied broadly comparable phenotyping and analytic methods. Smaller apo(a) isoforms were associated with a twofold increased risk for CHD compared to large isoforms (RR = 2.08, 95%CI 1.76-2.58) [59]. Similar relative risks were observed for ischemic stroke (RR = 2.14; 95 % CI: 1.85-2.97) [59]. These pronounced effect sizes are probably the strongest which will ever be identified for common variants and CVD, keeping in mind that approximately 25–35 % of the population carries small apo(a) isoforms (Fig. 3). This high prevalence of small apo(a) isoforms points to a high public health relevance.

The finding of the pronounced association between the number of KIV repeats and CVD is supported by further genetic studies using SNPs within the LPA gene region. A study of cases with myocardial infarction and controls investigated several SNPs in that gene region. Besides several other SNPs, rs10455872 and rs3798220 showed the strongest associations with myocardial infarction [37]. rs10455872 is an intronic SNP and rs3798220 results in an amino acid substitution in the protease domain of LPA. Both SNPs were described to be associated with small apo(a) isoforms and high Lp(a) levels. rs10455872 was associated with a 1.47-fold and rs3798220 with a 1.68-fold increased risk for CHD over non-carriers. Individuals who carried at least one risk allele of the two SNPs had a 1.51-fold elevated risk for myocardial infarction and subjects carrying two and more risk alleles of these two SNPs had a 2.57-fold increased risk [37]. These findings have been confirmed by a recent meta-analysis [60] and clearly underscore the LPA gene as a risk gene for CVD. This becomes even more accented in comparison to other genes for CAD detected by GWAS that were associated with odds ratios between 1.06 and 1.29 [61].

These impressive findings were a turning point which brought Lp(a) back to the stage of interest. Commercial laboratories started to offer the genotyping of the two SNPs rs10455872 and rs3798220 for diagnostic purpose which is easily possible in a high-throughput manner and with a simple interpretation. They were claimed to tag carriers of small apo(a) isoforms. However, the minor allele frequencies of these two SNPs are only 7 and 2 %, respectively. It was therefore quite interesting to see the data from almost 3000 individuals of a general Caucasian population that was genotyped for these two SNPs and phenotyped for the apo(a) isoforms [62]. The results clearly showed that 47 % of all subjects expressing a small apo(a) isoform were not tagged by one of these two high-risk SNPs. Furthermore, about 11 % of all subjects carrying at least one minor allele of these two SNPs did actually not have a small apo(a) isoform. That means that these two SNPs are far from being a good surrogate for risk evaluation instead of measurement of small apo(a) isoforms since roughly half of the small apo(a) isoforms will remain undetected when only these two SNPs are genotyped. As stated figuratively, from a population perspective we might see the tip of the iceberg but we might underestimate the size [62].

In addition, the situation might be different for various ethnicities. For example, a study investigating rs3798220 and K-IV repeat number in several ethnicities revealed that this variant was not found in Africans. Allele frequencies in East and Southeast Asians ranged from 2.9 to 11.6 %, and were very low (0.15 %) in CAD cases and controls from India. The variant was neither associated with small KIV CNV alleles nor elevated Lp(a) concentrations in Asians. The study concluded that it is unlikely that this SNP confers atherogenic potential on its own and that this SNP does not explain Lp(a)-attributed risk for CAD in Asian Indians [63]. This might also be the reason why a recent GWAS for CAD Japanese patients did not find the LPA locus to be associated with CAD [64] although case–control studies consistently found an association of the KIV repeat polymorphism with CAD in e.g., Chinese CAD patients and controls [55, 65].

To increase the number of SNPs by creating haplotypes instead of single SNPs for risk prediction was a further idea which was followed by a GWAS for myocardial infarction [66]. The authors used haplotypes built from four SNPs, two in LPA (rs7767084 and rs10755578) and two in the neighboring genes LPAL2 (rs3127599) and SLC22A3 (rs2048327) gene. A rare haplotype with a frequency of roughly 2 % was associated with an 82 % higher risk (OR 1.82, 95%CI 1.57-2.12) and a more common haplotype with a frequency of about 16 % had a 20 % higher risk (OR 1.20, 95%CI 1.13-1.28) for myocardial infarction [66]. Once again, this only supports the LPA gene as a risk gene for CVD but does not improve risk prediction when compared to apo(a) isoforms since either only for a small group of 2 % a tremendously increased risk or for a larger group of 16 % a slightly increased risk is predicted.

A further impressive genetic approach to support Lp(a) as a risk factor for CVD came from Kamstrup and colleagues who used qPCR to quantify the copy number of KIV type-2 repeats in >40,000 subjects. This method sums up the number of both apo(a) alleles and cannot distinguish each of the two alleles. They observed in the Copenhagen City Heart Study that individuals in the lower quartile of the sum of copy number in their genome had an adjusted hazard ratio for myocardial infarction of 1.50 compared to those in the highest quartile of copy number. The KIV type-2 CNV explained roughly 25 % of the variability in Lp(a) levels [53], which is lower than in other studies of European populations using methods such as apo(a) isoforms by Westernblot or separated alleles by pulsed-field gel electrophoresis. However, these estimates in risk and unexplained variability determined by qPCR are most likely an underestimation due to the special characteristics of qPCR (see below).

All these approaches and data together provide pronounced genetic evidence that Lp(a) is an emerging genetic risk factor for cardiovascular disease that is independent of other classical risk factors including lipids.

A recent GWAS identified rs10455872 of the LPA gene to be significantly associated with the presence of aortic-valve calcification with an odds ratio per allele of 2.05 (95%CI 1.63–2.57). In prospective analyses, the LPA genotype was associated with incident aortic stenosis and aortic-valve replacement [92]. In large studies from the general population, elevated Lp(a) levels and corresponding LPA risk genotypes predicted increased risk of incident aortic valve stenosis and the risk estimates were very similar to the observations for myocardial infarction [93, 94]. Furthermore, Lp(a) levels were also associated with aortic valve calcification in asymptomatic patients with familial hypercholesterolemia who were free of symptomatic CVD or symptoms suggestive of ischemic heart disease at the time of recruitment [95]. A prospective study in 220 patients with mild to moderate aortic stenosis described that progression of aortic stenosis was faster in the top tertiles of Lp(a) concentrations as well as oxidized phospholipids [96].

Lp(a) transports oxidized phospholipids with a high content in lysophosphatidylcholine. It has recently been demonstrated that autotaxin transforms lysophosphatidylcholine into lysophosphatidic acid that promotes deposition of hydroxyapatite of calcium in aortic valve. Autotaxin is transported in the aortic valve by Lp(a) and is also secreted by valve interstitial cells. This promotes inflammation and mineralization of the aortic valve [97].

The most recent findings published were on an association between Lp(a) concentrations and heart failure in more than 98000 individuals from the general population of Denmark [98]. In 4122 of them heart failure had been diagnosed during the observation period. Lp(a) concentrations in the upper tertile were significantly associated with heart failure compared to the lowest tertile and this association was graded within the upper tertile: HR = 1.24 (95%CI: 1.08-1.42) for the 67th to 90th percentiles, HR = 1.57 (95%CI: 1.32-1.87) for the 91st to 99th percentiles, and HR = 1.79 (95%CI: 1.18 to 2.73) for levels >99th percentile compared to the lowest tertile corresponding to a population-attributable risk of 9 %. LPA risk genotypes that are associated with high Lp(a) concentrations were in line with this association. When participants with a myocardial infarction or aortic valve stenosis were excluded, the risk estimates were attenuated and mediation analysis revealed that 63 % of heart failure risk was mediated via myocardial infarction or aortic valve stenosis [98].

The high homology of apo(a) and plasminogen suggested a link to the fibrinolytic system with blood clotting and fibrinolysis. It was therefore not clear whether the culprit is the development of atherosclerosis or thrombosis. Again a Mendelian randomization study in more than 41000 individuals has shed some light into the dark. Neither Lp(a) tertiles nor the sum of K-IV repeats were associated with the risk of venous thrombosis in general population studies [99]. A further study including 4607 cases with venous thromboembolism investigated the two LPA variants rs10455872 and rs3798220 and did also not find an association [100]. However, both studies found associations with CVD [99, 100].

However, a meta-analysis of eight studies including 589 children with venous thromboembolism and 1441 controls described elevated Lp(a) levels to be associated with an odds ratio of 4.50 (3.19–6.35) for first onset venous thromboembolism. The association with 135 recurrent cases of venous thromboembolisms was not statistically significant [101].

When Mora et al. described an association between very low Lp(a) concentrations and type 2 diabetes mellitus (T2DM) a few years ago, it was a major surprise [102]. The highest risk was found for Lp(a) <1 mg/dL (lowest 2.6 % of observations) with an odds ratio of 1.57 when compared to higher values. When the risk group was changed to Lp(a) concentrations below 4 mg/dL (equals the lowest quintile), the size of the risk was markedly attenuated to 1.18 [102]. These findings were contrary to expectations and the study could not clarify the causality between low Lp(a) levels and T2DM and could not exclude “reverse causation”. Kamstrup and Nordestgaard could not only replicate the association between low Lp(a) concentrations and T2DM in a large Danish study but also demonstrated that the highest quintile of the sum of the KIV-2 repeats from the two apo(a) alleles is associated with T2DM [103]. This quintile combines large isoforms and is associated with low and medium Lp(a) concentrations and therefore supports causality and excludes that the association between low Lp(a) concentrations and T2DM is merely due to reverse causation or confounding. Two other studies followed and confirmed the association between low Lp(a) concentrations and T2DM [104, 105]. An investigation in more than 10000 Chinese individuals extended the findings and found an association not only between low Lp(a) concentrations and prevalence of T2DM but also prevalent prediabetes, insulin resistance and hyperinsulinemia [105].

Mendelian randomization studies that investigated an association between the SNP rs10455872 and T2DM caused more confusion than support for causality [103, 104]. The carrier-status of this SNP is already well known for its association with CVD since it is associated with high Lp(a) concentrations [37]. The authors of these two studies [103, 104] used the non-carrier status of rs10455872 as a surrogate for genetically determined low Lp(a) concentrations to investigate the causality for the association between low Lp(a) concentrations and T2DM. Since they did not find an association between the non-carrier status of rs10455872 and T2DM, a causal role of Lp(a) concentrations has been called into question [103, 104]. As we discussed recently [106], the non-carrier status of rs10455872 includes already 86 % of the population with a very wide range in Lp(a) concentrations and with the inclusion of large and small apo(a) isoforms [103]. We have to bear in mind that the association of Lp(a) and T2DM is especially seen in subjects with ultra-low Lp(a) concentrations (e.g., <1 mg/dL) and was already weaker, when the low class was extended to the lowest quintile [103]. With other words, the non-carrier status of the SNP explains the wrong range of the Lp(a) values which might “dilute” the association of very low Lp(a) levels with T2DM. Therefore, rs10455872 is obviously an imprecise instrument to support or exclude causality of low Lp(a) levels for T2DM [106].

One might ask the question whether it would be counterproductive to lower Lp(a) to avoid CVD events but simultaneously increase the risk for T2DM? This is probably not the case since the risk for T2DM is especially elevated for persons with very low Lp(a) concentrations and an Lp(a)-lowering therapy will not bring Lp(a) in these concentration ranges.