Association of LPA Variants With Risk of Coronary Disease and the Implications for Lipoprotein(a)-Lowering Therapies: A Mendelian Randomization Analysis

Importance Human genetic studies have indicated that plasma lipoprotein(a) (Lp[a]) is causally associated with the risk of coronary heart disease (CHD), but randomized trials of several therapies that reduce Lp(a) levels by 25% to 35% have not provided any evidence that lowering Lp(a) level reduces CHD risk. Objective To estimate the magnitude of the change in plasma Lp(a) levels needed to have the same evidence of an association with CHD risk as a 38.67-mg/dL (ie, 1-mmol/L) change in low-density lipoprotein cholesterol (LDL-C) level, a change that has been shown to produce a clinically meaningful reduction in the risk of CHD. Design, Setting, and Participants A mendelian randomization analysis was conducted using individual participant data from 5 studies and with external validation using summarized data from 48 studies. Population-based prospective cohort and case-control studies featured 20 793 individuals with CHD and 27 540 controls with individual participant data, whereas summarized data included 62 240 patients with CHD and 127 299 controls. Data were analyzed from November 2016 to March 2018. Exposures Genetic LPA score and plasma Lp(a) mass concentration. Main Outcomes and Measures Coronary heart disease. Results Of the included study participants, 53% were men, all were of white European ancestry, and the mean age was 57.5 years. The association of genetically predicted Lp(a) with CHD risk was linearly proportional to the absolute change in Lp(a) concentration. A 10-mg/dL lower genetically predicted Lp(a) concentration was associated with a 5.8% lower CHD risk (odds ratio [OR], 0.942; 95% CI, 0.933-0.951; P = 3 × 10−37), whereas a 10-mg/dL lower genetically predicted LDL-C level estimated using an LDL-C genetic score was associated with a 14.5% lower CHD risk (OR, 0.855; 95% CI, 0.818-0.893; P = 2 × 10−12). Thus, a 101.5-mg/dL change (95% CI, 71.0-137.0) in Lp(a) concentration had the same association with CHD risk as a 38.67-mg/dL change in LDL-C level. The association of genetically predicted Lp(a) concentration with CHD risk appeared to be independent of changes in LDL-C level owing to genetic variants that mimic the relationship of statins, PCSK9 inhibitors, and ezetimibe with CHD risk. Conclusions and Relevance The clinical benefit of lowering Lp(a) is likely to be proportional to the absolute reduction in Lp(a) concentration. Large absolute reductions in Lp(a) of approximately 100 mg/dL may be required to produce a clinically meaningful reduction in the risk of CHD similar in magnitude to what can be achieved by lowering LDL-C level by 38.67 mg/dL (ie, 1 mmol/L).

lowering effect of these therapies.6-11 Although these trials were not specifically designed to assess the Lp(a)-lowering effect of these agents, these trials raise the question of how much Lp(a) concentration must be lowered to produce a clinically meaningful reduction in cardiovascular events. Therapies that more specifically and potently lower Lp(a) concentrations by up to 90% by inhibiting apolipoprotein(a) synthesis are in development. 12 Whether lowering Lp(a) concentrations with these new therapies will reduce the risk of cardiovascular events is unknown.
Owing to the skewed distribution of plasma Lp(a) concentration, prior studies have reported the association between log-transformed concentrations of Lp(a) and CHD risk. 2,[13][14][15] Changes in log-transformed Lp(a) concentrations represent proportional changes in Lp(a) concentrations. However, proportional reduction is not a useful metric for assessing the potential clinical benefit of lowering Lp(a) level because concentrations can vary by as much as 1000-fold among members of the same population, and therefore, the same proportional change in Lp(a) concentration can result in markedly different absolute changes, depending on the initial Lp(a) concentration. 16 Importantly, statins and other therapies that reduce LDL particle concentrations are associated with a dose-dependent reduction in the risk of cardiovascular events that is determined by the absolute (rather than the proportional) change in LDL cholesterol (LDL-C) level. [17][18][19] Because Lp(a) contains an LDL particle, we hypothesized that there would be evidence to support a clinical association of Lp(a) with the risk of CHD that may also be proportional to the absolute change in circulating Lp(a) mass concentration. To test this hypothesis, we created a genetic score to estimate the magnitude and shape of the relationship of Lp(a) with the risk of CHD. We then estimated the absolute change in plasma Lp(a) concentration required to achieve the equivalent change in CHD risk as a 38.67-mg/dL (ie, 1-mmol/L) change in LDL-C level (to convert to millimoles per liter, multiply by 0.0259), a change in LDL-C that has been demonstrated to produce a clinically meaningful 20% to 25% reduction in the risk of cardiovascular events in short-term trials. [17][18][19] Our objective was to make inferences about how much Lp(a) concentration must be reduced pharmacologically to produce a clinically meaningful reduction in CHD risk and thereby determine who is most likely to benefit from treatment with Lp(a)-lowering therapy to inform clinical guidelines and the design of randomized trials evaluating Lp(a)-lowering therapies.

Study Population and Outcomes
We studied 48 333 participants of European descent (including 20 793 with CHD) from 5 studies for whom individual participant-level data were available as part of the CHD Exome + Consortium. Descriptions of the included studies are provided in eMethods 1 of the Supplement. The primary outcome was CHD, defined as CHD death, nonfatal myocardial infarction, or (for 3 of the studies) other coronary events with International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes I20-25.
Participants provided written informed consent for genetic studies. As this was an analysis of anonymized data that had already been collected, ethical approval was not sought for this particular investigation.

LPA Genetic Score
All CHD Exome+ Consortium participants were genotyped using a customized version of the Illumina Exome Beadchip array, which included ultrafine mapping of the LPA gene region involving 2426 variants genotyped within a 660-kb window (eFigure 1 in the Supplement). To select variants for inclusion in the genetic score, we identified variants in the LPA gene region that were conditionally associated with Lp(a) concentrations at a genome-wide level of significance (P < 5 × 10 −8 ) using forward stepwise regression among participants free from CHD at baseline in each study. We adjusted for study, age, sex, and 5 principal components of ancestry. Genetic variants correlated with a selected variant at r 2 greater than 0.4 were excluded from further steps of the procedure (eFigure 2 in the Supplement). For each participant, we calculated a weighted genetic score by summing the number of Lp(a)-raising alleles inherited at each variant included in the score, weighted by each variant's association with absolute change in Lp(a) mass concentration (measured in milligrams per deciliter). In sensitivity analyses, we repeated the primary analyses using different choices of variants in the genetic scores, as described in eMethods 2 in the Supplement.

Study Design
To assess the dose-response shape of the association between genetically predicted Lp(a) and CHD risk, we divided participants into deciles of the genetic score and measured the association between each decile of genetically predicted Lp(a) concentration and the risk of CHD using the first decile as the reference group. Informed by the shape of the association, we estimated the association between the LPA score and the risk of CHD for absolute changes in Lp(a) concentration.
To estimate the absolute reduction in Lp(a) concentration required to have the same change in CHD risk as a 38.67-mg/dL decrease in LDL-C level, we used the following protocol (eFigure 3 in the Supplement). First, we measured the association between the LPA score and the risk of CHD per 10-mg/dL decrease in genetically predicted Lp(a) concentration. Next, we measured the association between a genetic score consisting of variants in or near genes that encode the targets of currently available LDL-C-lowering therapies and CHD risk per 10-mg/dL decrease in genetically predicted LDL-C (eTable 1 in the Supplement). 20 We then calculated the ratio between these 2 estimates to obtain the change in Lp(a) concentration that has an equivalent association with CHD risk as a 1-mg/dL change in LDL-C level. To estimate the amount Lp(a) concentration must be reduced to have the same association with CHD risk as a 38.67-mg/dL reduction in LDL-C level, we multiplied this ratio by 38.67. Finally, we estimated the predicted short-term change associated with different magnitudes of pharma-cological lowering of Lp(a) concentration by converting the change in Lp(a) concentration into a change in LDL-C level having an equivalent predicted effect on CHD risk and using the estimated change associated with statin treatment per 38.67-mg/dL reduction in LDL-C level, as reported by the Cholesterol Treatment Trialists' Collaboration.17

Statistical Analyses
We estimated the association of each variant with Lp(a) or LDL-C concentration using linear regression and with CHD risk using logistic regression. All regression analyses were performed separately in each of the studies, adjusting for age, sex, and the first 5 principal components of ancestry; these estimates were combined across studies in a fixed-effects inverse variance-weighted meta-analysis. Heterogeneity was assessed using the I 2 statistic.
Mendelian randomization estimates were then obtained from these variant-specific estimates using a previously reported method that accounts for correlation between variants.21 Nonlinearity in the mendelian randomization estimates of the shape of the association of Lp(a) change with the risk of CHD was assessed using fractional polynomials, as described elsewhere.22 For external replication in an independent sample, we performed the same analyses using summarized genetic associations with CHD risk from the Coronary Artery Disease Genome Wide Replication and Meta-analysis (CARDIOGRAM) plus The Coronary Artery Disease (C4D) Genetics (CARDIOGRAMplusC4D) consortium in up to 62 240 patients and 127 299 controls.23 All analyses were performed using the statistical software platform R version 3.4.1 (R Programming). A detailed description of the methods is provided in eMethods 2 of the Supplement.

Participant Characteristics
The baseline characteristics of participants are presented in Table 1. Across the 5 studies contributing to the initial sample, the median Lp(a) concentration varied from 13.6 mg/dL to 43.3 mg/dL (eTable 2 and eFigure 4 in the Supplement).

LPA Genetic Score
The stepwise selection procedure identified 43 genetic variants conditionally associated with Lp(a) (eTable 3 in the Supplement). The genetic score comprising these variants explained 51% to 63% of the variance in Lp(a) concentration in each study (eFigure 5 in the Supplement). This explanatory ability is lower than observed previously24 because our genetic score was constructed conservatively to minimize bias owing to overfitting. Associations of each variant with Lp(a) concentration and CHD risk are displayed in Figure  1.

Association of LPA Genetic Score With CHD
In analyses dividing the population into deciles of genetically predicted absolute Lp(a) mass concentration, the exposure-outcome association for log-transformed CHD risk was approximately linear, ie, fixed changes in absolute Lp(a) concentrations led to equal odds ratios (ORs) for CHD regardless of the starting Lp(a) concentration ( Figure 2A). By contrast, the exposure-outcome association for deciles of log-transformed Lp(a) concentration was curvilinear ( Figure 2B), with fixed proportional changes in Lp(a) concentrations leading to greater log-ORs for individuals with higher baseline Lp(a) concentrations (and hence, increasingly greater absolute changes in Lp[a] concentrations).
These findings are consistent and support the hypothesis that the risk of CHD is log-linearly proportional to absolute changes in Lp(a) concentration.
Overall, each 10-mg/dL lower genetically predicted Lp(a) level was associated with a 5.8% lower risk of CHD (OR, 0.942; 95% CI, 0.933-0.951; P = 3 × 10 −37 ). There was no evidence of heterogeneity with similar genetic association estimates obtained across all studies independent of the type of Lp(a) assay used (eFigures 6 and 7 in the Supplement). Estimates were also similar in sensitivity analyses that varied the number of genetic variants included in the LPA score (eTable 4 in the Supplement). In external replication analyses involving participants from CARDIOGRAMplusC4D, a 10-mg/dL lower genetically predicted Lp(a) level was associated with a 5.2% lower risk of CHD (OR, 0.948; 95% CI, 0.941-0.955; P = 1 × 10 −47 ).

Expected Clinical Benefit of Lowering Lp(a) Concentration
Using the LDL-C genetic score, a 10-mg/dL genetically predicted lower LDL-C level was associated with a 14.5% lower risk of CHD (OR, 0.855; 95% CI, 0.818-0.893; P = 2 × 10 −12 ) (eFigure 8 in the Supplement). This finding suggests that a 1-mg/dL difference in LDL-C level has the same association with CHD risk as a 2.63-mg/dL difference in Lp(a) concentration (ie, log[0.855] / log[0.942] = 2.63), and therefore, a 38.67-mg/dL difference in LDL-C level has the same association as a 101.5-mg/dL (95% CI, 71.0-137.0) difference in Lp(a) concentration. In external replication analyses using data from CARDIOGRAMplusC4D, a 10-mg/dL lower LDL-C level was associated with a 14.0% lower CHD risk (OR, 0.860; 95% CI, 0.841-0.879; P = 3 × 10 −40 ), suggesting that a 109.1mg/dL (95% CI, 89.0-133.1) difference in Lp(a) concentration has the same association with CHD risk as a 38.67-mg/dL difference in LDL-C level.
Changes in genetically predicted Lp(a) and LDL-C concentrations represent lifelong exposure to these lipoproteins. Hence, to estimate the effect of lowering Lp(a) concentration in a short-term trial, we assumed that if lifelong exposure to 101.5-mg/dL lower Lp(a) concentration has the same association with CHD risk as lifelong exposure to 38.67-mg/dL lower LDL-C level, then short-term exposure to 101.5-mg/dL lower Lp(a) concentration should have the same association with CHD risk as short-term exposure to 38.67-mg/dL lower LDL-C levels observed in randomized trials. This assumption is valid only if changes in Lp(a) concentration and LDL-C level have similar cumulative associations with CHD over time. It is further supported by the observation that the ratio of the association of lifelong exposure to Lp(a) with CHD risk estimated from mendelian randomization to the association of intermediate-term exposure to Lp(a), estimated from observational studies in the Emerging Risk Factors Consortium,2 is very similar to the ratio of the association of lifelong exposure to LDL-C with CHD risk estimated from mendelian randomization to the association of intermediate-term exposure to LDL-C in the Emerging Risk Factors Consortium (Figure 3; eFigure 9 in the Supplement).25 Therefore, Lp(a) and LDL-C appear to have similar cumulative associations with the risk of CHD over time. Table 2 shows the expected clinical benefit in CHD risk from both lifelong and short-term exposure to absolute differences in Lp(a) concentration. Lifelong estimates are conventional mendelian randomization estimates, while short-term estimates are calculated using the difference in Lp(a) concentration needed to achieve the same change for a given reduction in LDL-C level over a median of 5 years of treatment with a statin, as reported by the Cholesterol Treatment Trialists' Collaboration.18

Independent Association of Lp(a) and LDL-C-Lowering Therapies
To assess whether the association of lowering Lp(a) concentration with the risk of CHD is likely to be independent of lowering LDL-C level with statins, we divided the population into 3 groups based on the number of LDL-C-lowering alleles each participant inherited at a common variant (rs12916) in the 3-hydroxy-3-methyl-glutaryl-CoA reductase (HMGCR) gene, which encodes the target of statins.26,27 The LPA score had nearly identical associations per 10-mg/dL lower Lp(a) concentration in each of these 3 groups (CC genotype group: OR, 0.945; 95% CI, 0.927-0.964; CT genotype group: OR, 0.939; 95% CI, 0.927-0.952; TT genotype group: OR, 0.945; 95% CI, 0.932-0.957; P = .79 for difference) (eFigure 10 in the Supplement), suggesting that the relative risk reduction of lowering Lp(a) concentration is likely to be independent of lowering LDL-C level with statins. Similar findings were observed for genetic variants in the PCSK9 and NPC1L1 gene regions that mimic the changes associated with PCSK9 inhibitors and ezetimibe, respectively.

Discussion
We found that the association of genetically predicted plasma Lp(a) with the risk of CHD was linearly proportional to the absolute difference in Lp(a) concentration. Absolute differences in Lp(a) concentration of approximately 100 mg/dL had an equivalent association with CHD risk as a 38.67-mg/dL difference in LDL-C level. The results of this study may have important implications for informing clinical practice guidelines on the use of Lp(a)-lowering therapies, for designing randomized trials to evaluate Lp(a)-lowering therapies currently in development, and for designing screening programs to reduce the global burden of CHD.
Because a 100-mg/dL difference in Lp(a) concentration had the same association with CHD risk as a 38.67-mg/dL difference in LDL-C level, the results of this study suggest that pharmacologically lowering Lp(a) concentration by approximately 100 mg/dL should reduce the risk of CHD (CHD death or nonfatal myocardial infarction) by approximately 22% to 25% in a 3-to 5-year randomized trial, similar to the association that has been observed for a 38.67-mg/dL reduction in LDL-C level during treatment with a statin.17-19 Therefore, it follows that lowering Lp(a) concentration by 80 mg/dL might be expected to reduce the risk of CHD events by approximately 18% to 20%, while lowering Lp(a) concentration by 50 mg/dL might reduce CHD events by 10% to 12% (Table 2), assuming that there are no unrecognized competing risks associated with lowering Lp(a) concentration. Therefore, only persons with very high Lp(a) concentrations are likely to benefit substantially from therapies that reduce Lp(a) concentration.
This finding likely explains why therapies that reduce Lp(a) concentration by 20% to 35% have failed to provide clear evidence that lowering Lp(a) concentration reduces the risk of cardiovascular events in previous randomized trials even though Lp(a) is a genetically supported target. The median Lp(a) concentration among participants enrolled in these trials Burgess

Europe PMC Funders Author Manuscripts
was approximately 12 to 20 mg/dL.7-11 Therefore, a 30% reduction in Lp(a) concentration would translate into only a 3-to 6-mg/dL absolute reduction in circulating plasma Lp(a) concentration, a small absolute reduction that was likely far too modest to reduce the risk of cardiovascular events appreciably in a short-term randomized trial.
The results of this study suggest that randomized trials evaluating new, more potent Lp(a)lowering therapies in development should be designed to enroll individuals with very high baseline Lp(a) concentrations of 90 to 100 mg/dL or more. Reducing Lp(a) concentration by 80% to 90% in such individuals should translate into large absolute reductions in Lp(a) concentrations of 70 to 90 mg/dL, which should in turn translate into approximately a 15% to 20% proportional reduction in the risk of CHD events. Enrolling patients at high risk of CHD owing to markedly elevated Lp(a) concentration in the initial proof-of-concept clinical trials is similar to the strategy used by the Scandinavian Simvastatin Survival Study trial,28 which enrolled high-risk patients with markedly elevated LDL-C concentrations and was the first trial to demonstrate that treatment with statins led to large, clinically meaningful reductions in the risk of cardiovascular events.
The magnitude of the pharmacologic reduction in Lp(a) mass that is likely needed to produce clinically meaningful reductions in CHD risk estimated in this study is larger than estimated in a 2018 study evaluating changes in Lp(a) during treatment with niacin. 29 However, whereas that previous study involved informal estimates of the reversible CHD risk by lowering Lp(a) concentrations in a short-term trial,29 we used a more systematic approach. In particular, our study estimated the differences in genetically predicted Lp(a) and LDL-C concentrations needed to have the same change in lifetime CHD risk and incorporated an assessment of the differential cumulative associations of Lp(a) and LDL-C with CHD risk over time to estimate how much Lp(a) concentration must be lowered pharmacologically to produce the same change as lowering LDL-C level by 38.67 mg/dL (ie, 1 mmol/L) with a statin. This approach has been successfully used to accurately anticipate the results of several recent trials.26,27,30 Similar analyses to those used in the current study are needed before it would be possible to accurately anticipate the potential effect of pharmacologically lowering Lp(a) on the risk of stroke, peripheral vascular disease, aortic stenosis, or composite end points that include these outcomes.31 Finally, it should be noted that plasma Lp(a) concentration is largely heritable. Therefore, if the linear relationship with CHD risk continues at very high absolute Lp(a) concentrations (as occurs for LDL-C), then Lp(a) concentrations in excess of 200 mg/dL may be associated with a 3-to 4-fold increased lifetime risk of CHD (OR, 3.30; 95% CI, 2.75-3.96) and thus may represent an inherited lipoprotein disorder that is associated with a similar lifetime risk of CHD as heterozygous familial hypercholesterolemia but with a prevalence that may be 2fold higher than that of heterozygous familial hypercholesterolemia.32,33 Therefore, screening for individuals with extremely elevated Lp(a) concentrations and treating them with one of the new Lp(a)-lowering therapies in development could potentially have the same effect on reducing the global burden of CHD as current screening programs designed to detect and treat individuals with familial hypercholesterolemia.

Limitations
Our study has limitations. Multiple different assays were used to measure Lp(a) concentrations in the included studies. However, we focused only on absolute differences in Lp(a) associated with genetic variants, which were very similar across all included studies, regardless of assay used or baseline Lp(a) concentrations. In addition, our estimate of the effect of lowering Lp(a) is agnostic to the mechanism of action, and hence our use of plasma Lp(a) mass concentration to estimate the dose-response relationship does not imply that our estimates are solely via changes in plasma Lp(a) mass concentration. If pharmacologic Lp(a) lowering has associations not adequately captured by the genetic variants (eg, antithrombotic associations), then smaller absolute reductions in Lp(a) than estimated here may produce clinically meaningful reductions in CHD risk.34

Conclusions
The association of genetically predicted Lp(a) with CHD risk was linearly proportional to the absolute change in Lp(a) mass concentration. Large absolute reductions in Lp(a) concentration of approximately100mg/dLare likely necessary to achieve clinically meaningful reductions in the risk of CHD similar in magnitude to what can be achieved by lowering LDL-C level by 38.67mg/dL (ie, 1 mmol/L) with a statin.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Funding/Support
The study's coordinating centre has been underpinned by grants G0800270 and MR/L003120/ Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Key Points
Question How much does plasma lipoprotein(a) need to be lowered to produce a clinically meaningful reduction in the risk of coronary heart disease?
Findings In a mendelian randomization analysis involving more than 80 000 patients and more than 150 000 controls, coronary heart disease risk was proportionally associated with the absolute change in plasma lipoprotein(a) mass concentration; a 101.5-mg/dL change in lipoprotein(a) concentration was associated with the same coronary heart disease risk as a 38.67-mg/dL (ie, 1-mmol/L) change in low-density lipoprotein cholesterol level.
Meaning Lipoprotein(a) concentration must be lowered by approximately 100 mg/dL to achieve the same reduction in coronary heart disease risk as can be achieved by lowering low-density lipoprotein cholesterol level by 38.67 mg/dL.      Abbreviations: CHD, coronary heart disease; LDL-C, low-density lipoprotein cholesterol; Lp(a), lipoprotein (a). SI conversion factor: To convert LDL-C to millimoles per liter, multiply by 0.0259.

Figure 2. Shape of Association Between Genetically Predicted Lipoprotein(a) (Lp[a]) and Coronary Heart Disease (CHD) Risk
a Each mg/dL lower Lp(a) has an association with CHD risk that is equivalent to a 0.38-mg/dL reduction in LDL-C based on the ratio of the associations of the genetic scores with CHD risk.
b Effect size (95% confidence interval) for Lp(a) reduction obtained from mendelian randomization approach.
c Effect size (95% confidence interval) for Lp(a) reduction obtained by considering equivalent lowering of LDL-C and in comparison with estimate from randomized trials of statins on major coronary events (risk ratio, 0.76; 95% CI, 0.73-0.78).