Gene-statin interaction study identifies 9 loci that moderate statin effectiveness in blood lipids

doi:10.21203/rs.3.rs-4720582/v1

Download PDF

Article

Gene-statin interaction study identifies 9 loci that moderate statin effectiveness in blood lipids

https://doi.org/10.21203/rs.3.rs-4720582/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Hyperlipidemia, and high low-density lipoprotein cholesterol (LDL-c) in particular, is a risk factor for atherosclerosis, myocardial infarction, and stroke. High LDL-c is primarily treated with HMG-CoA reductase inhibitors, commonly known as statins. If statins interfere with the genetic pathways that endogenously increase the risk for hyperlipidemia, gene-statin interactions may identify individuals that are more sensitive to these drugs. Accordingly, we performed genome-wide gene-statin interaction analyses for LDL-c and two related lipids: high-density lipoprotein cholesterol (HDL-c) and triglycerides (TG). Our results suggest statins selectively reduce LDL-c heritability. More specifically, we identified five genome-wide significant gene-statin interactions for LDL-c, two gene-interactions for HDL-c, and four gene-interactions for TG. Fifteen loci remained associated with LDL-c despite statin treatment, revealing treatment-resistant genotypes and suggesting additional genetic targets for drug development, enhancement, and repurposing. These results are an important step towards using targeted treatments for patients with hyperlipidemia.

Health sciences/Cardiology/Cardiovascular biology/Cardiovascular genetics

Health sciences/Diseases/Cardiovascular diseases/Dyslipidaemias

Hyperlipidemia, a condition marked with aberrant lipid concentrations in the blood, increases the risk for atherosclerosis, myocardial infarction, and stroke. High blood levels of low-density lipoprotein cholesterol (LDL-c) appear to play the largest causal role in this process¹, though high levels of triglycerides (TG) and low levels of high-density lipoprotein cholesterol (HDL-c) may also contribute to negative hyperlipidemia outcomes. In general, lipid levels have a strong genetic component, which has two implications for the current study. First, because lipid levels are, in part, a function of genetic processes, targeting the genetic mechanisms that endogenously affect lipid levels will interfere with these genetic pathways and reduce the association between genes and lipid profiles. Second, because everyone has a unique constellation of genes, pharmaceutical interventions may be more effective for certain people. Gene-drug interactions highlight the genomic mechanisms that regulate lipid metabolism and homeostasis, identify genetic markers for individuals who may benefit the most (or least) from statin treatment, and lay the foundation to develop proactive and personalized solutions to reduce cardiovascular disease morbidity and mortality^2,3.

The primary treatment for hyperlipidemia is statins, which reduce LDL-c levels in almost everyone ^4–8. Importantly, the benefits of statins are heterogeneous: they are generally effective for most people, but approximately 10% of patients experience side effects such as myopathy or inflammation of the liver and kidney⁹, potentially due to genetic variation^10,11. While people who experience adverse events from statins are likely to stop or change treatments, others may quietly or unknowingly benefit more than average. Identifying gene-statin interactions for hyperlipidemia will highlight genomic variants that increase or decrease treatment efficacy, in turn highlighting the genotypes that are more (or less) sensitive to statin use and paving the way for personalized prescription protocols.

Our premise is that genetic factors increase an individual’s endogenous risk for hyperlipidemia, and statins reduce this risk by interfering with these pathways. This implies three hypotheses. First, if statins interfere with the genetic pathways that culminate in hyperlipidemia, then the heritability, or the variation accounted for by genetic factors, of lipids should be lower in individuals taking statins. Second, identifying single nucleotide polymorphisms (SNPs) implicated in gene-drug interactions can highlight drug mechanisms and identify individuals who may benefit the most (or least) from statin use. Third, genetic associations for individuals who are already taking statins can highlight putative gene-targets for drug improvement, repurposing, and novel development. The preponderance of causal evidence linking lipids to CVD focuses on LDL-c ^1,12, but due to correlations between LDL-c, HDL-c, and TG it is instructive to explore the possibility of statin effects on these related outcomes. As causal evidence is stronger for LDL-c, the specific pharmacogenomic hypotheses apply more directly to LDL-c than to either TG or HDL-c. In this study, we test these hypotheses by conducting genome-wide gene-statin interaction analyses for LDL-c, HDL-c, and TG with data from the European ancestry subset of the UK Biobank (see Method Section). Covariates include age, sex, the first 10 genetic ancestry principle components (PCs), and interactions between statins and the PCs¹³ (see Methods section for analytical details and sample descriptions).

Phenotypic Correlations

The means and standard deviations of the three lipid blood levels, the proportion of people taking a statin, and other relevant demographic information is described in Supplementary Table 1. The drug codes used to define statin use is described in Supplementary Table 2. LDL-c, HDL-c, and TG are moderately correlated, necessitating a simultaneous investigation of the three lipids to avoid inadvertently misinterpreting pleiotropic genetic associations. In the UK Biobank European subsample (sample details in Methods), statin use appears to induce a small phenotypic correlation between LDL-c and HDL-c (r_{no statin} =0.01, r_statin = 0.16) and modestly reduce the phenotypic correlation between LDL-c and TG (r_{no statin} =0.29, r_statin = 0.24). Thus, statin use alters the pattern of phenotypic correlations between the lipids, potentially by interfering with the genetic mechanisms regulating LDL-c.

Heritability and Genetic Correlations: We examined whether statins reduce the heritability of each lipid. We used the summary statistics from the gene-statin interaction analyses to calculate genetic marginal effects for statin users and non-users for each lipid¹⁴ (See Methods Section). We used the marginal genetic effects to estimate the SNP-based heritability (h²_snp) of each lipid using linkage disequilibrium score regression (LDSC)¹⁵ depending on statin use. We found that statin use only reduced h²_snp for LDL-c; the h²_snp of HDL-c and TG remained unchanged (Fig. 1; see Supplementary Table 3 for more details). The decrease in h²_snp for LDL-c is consistent with clinical observations where statins reduce LDL-c levels and minimally affect TG and HDL-c. Further, SNP-based genetic correlations (r_Gsnp) between statin users and non-users for LDL-c was significantly less than one (r_G = 0.79, p_rG<1 = 0.016), implying different sets of genes contribute to the genetic architecture of LDL-c depending on statin use. The r_Gsnp between statin users and non-users for HDL-c and TG are not different from one (HDL-c: r_G = 0.95, p_rG<1 = 0.55; TG: r_G = 0.99, p_rG<1 = 0.89). In other words, statin use appears to disrupt some of the genetic pathways that increase LDL-c by selectively interfering with genetic associations in LDL-c relevant loci.

SNP-statin Interactions in LDL-c, HDL-c and TG. We identified five genome wide significant interactions between statin use for LDL-c (Apolipoprotein B – APOB-rs934197; Fatty acid desaturase 2 –FADS2-rs1535; low density lipoprotein receptor – LDLR-rs17242381; SURP and G-patch domain containing 1 – SUGP1-rs10401969; and Apolipoprotein E – APOE-rs72654473), two interactions for HDL-c (cholesteryl ester transfer protein – CETP-rs247616; and APOE-rs429358), and four interactions for TG (Lipoprotein A – LPA-rs55730499, FADS2-rs102275, apolipoprotein A4 – APOA4-rs7118999, and APOE-rs7412; see Table 1). While SNPs in the APOE region interacted with all three lipids (Fig. 2), the lead SNP for each lipid varied despite the lipids being modestly correlated and the SNPs being in high LD. Accordingly, while we cannot assume to identify causal variants, it is clear that this region has a strong, broad association with lipid regulation. Specifically, in the APOE region, statins reduce (but do not eliminate) the marginal genetic effects of the SNPs on LDL-c and on HDL-c, with the pattern of effects being stronger and more consistent for LDL-c. For TG, statins appear to increase the marginal genetic effects in the APOE region. This pattern of results is consistent with the observation that taking statins may induce a small, positive phenotypic correlation between LDL-c and HDL-c, and reduce the correlation between LDL-c and TG. The global pattern of reduced marginal genetic effects is consistent across the other genome-wide significant interaction loci (p < 5x10^− 8) for LDL-c and HDL-c, but variable for TG depending on the genomic loci (see Supplementary Figs. 4–14 for regional plots of additional loci). Notably, the pattern of SNP-statin interactions for the three lipids are relatively distinct: significant interactions for one lipid are not necessarily significant for the others despite significant phenotypic and genetic correlations.

Table 1

Results for the genome-wide significant gene-statin interaction index variants identified in the GxE GWAS analysis of LDL Cholesterol, HDL Cholesterol and Triglycerides in 389,440 individuals.
					Main Effect (β₁)				Interaction Effect (β₃)
Index Variant	Chr	BP	A1	A2	β₁	SE	Z	P	β₃	SE	Z	P	Nearest Gene
LDL Cholesterol
rs934197	2	21267461	A	G	3.67	0.08	46.19	0.00E + 00	-1.35	0.22	6.14	8.31E-10	APOB
rs1535	11	61597972	A	G	1.54	0.08	19.36	1.53E-83	-1.72	0.22	7.70	1.40E-14	FADS2
rs17242381	19	11206729	T	C	6.21	0.12	53.29	0.00E + 00	-3.17	0.34	9.42	4.73E-21	LDLR
rs10401969	19	19407718	T	C	4.42	0.14	31.24	3.35E-214	-2.53	0.41	6.19	6.19E-10	SUGP1
rs72654473	19	45414399	C	A	12.11	0.11	105.34	0.00E + 00	-5.75	0.34	16.84	1.20E-63	APOE
HDL Cholesterol
rs247616	16	56989590	C	T	3.52	3.52	97.31	0.00E + 00	-0.58	0.09	-6.18	6.36E-10	NLRC5
rs429358	19	45411941	C	T	-1.03	0.05	-20.57	4.67E-94	0.71	0.12	6.00	1.96E-09	APOE
Triglycerides
rs55730499	6	161005610	C	T	-2.78	0.43	6.41	1.48E-10	-5.81	0.93	6.21	5.23E-10	LPA
rs102275	11	61557803	T	C	4.10	0.24	17.32	3.18E-67	4.20	0.53	7.91	2.48E-15	FADS2
rs7118999	11	116645275	C	T	-11.91	0.25	47.27	0.00E + 00	-3.28	0.56	5.85	5.04E-09	APOA4
rs7412	19	45412079	C	T	9.38	0.39	24.19	2.86E-129	12.63	0.94	13.46	2.68E-41	APOE

Variant-level differences in the Genetic Marginal Effects for Statin Users versus Non-users: We present circular Manhattan plots of the marginal genetic effects in the No Statin and Statin groups in Figs. 3a and 3b. As lipids are phenotypically correlated, we highlight genomic regions with significant SNP-statin interactions across lipids (standard Manhattan and QQ plots of all the marginal genetic effects are presented in Supplementary Figs. 1–3). A complete list of the lead SNPs for the genome-wide significant marginal genetic effects for statin users and non-users in each lipid are presented in Supplementary Tables 4–6. Marginal genetic effects can be interpreted in the same way as standard genome-wide association study (GWAS) associations for a specific level of the moderator (i.e., genetic associations for Non-Statin or Statin Users). The red, blue, and yellow bands around the Manhattan plots in Fig. 3 denote genome wide significant associations for LDL-c, HDL-c, and TG, respectively. The robust polygenic architecture of unmedicated respondents for all three lipids is evident in the myriad genome-wide significant associations in Fig. 3a. The large number of genome-wide risk loci for non-statin users (LDL-c = 247, HDL-c = 234, TG = 139) reflects the central role genetic factors play in endogenous lipid regulation and metabolism¹⁶. The results for non-statin users reflect the findings from the literature¹⁷. Compared with the non-statin group, there is a muted genetic architecture for the statin users, as seen in Fig. 3b (number of genome-wide significant risk loci for statin users: LDL-c = 15, HDL-c = 38, TG = 82). The differences in the magnitude of the p-values may reflect two factors. First, there is a large asymmetry in the sample sizes between statin users and non-users (N_{no statins} = 325,922 vs N_statins = 63,518) which reduces the statistical power and significance for the analyses of statin users. Second, statins may directly reduce the magnitudes of the genetic associations. In the current analyses, statin use appears to reduce the magnitude of the genetic associations for LDL-c, while the difference in the magnitudes of the genetic associations for HDL-c and TG for most loci are negligible. Thus, the interpretation of the differences between the results for statin and non-statin users for LDL-c are consistent with SNP-statin interactions, while the interpretation of the results for HDL-c and TG are more consistent with differences in sample size.

Integrating Pharmaceutical and Genomic Results. We integrated data from the LDL-c genome-wide gene-statin interaction analysis with the pharmGKB.org database^18,19 (see Supplementary Table 7 for the list of statin gene-targets) to answer two distinct research questions. First, we examined whether genome-wide significant interactions were in pharmacologically predicted genetic loci from the pharmGKB.org database to affirm purported mechanisms of action for statins. Second, we collated the genome-wide significant marginal effects for the statin group to identify variants that remain associated with LDL-c levels despite treatment to suggest additional genomic targets for future pharmacological development.

The bottom triad of Fig. 4 lists gene targets and chromosome positions for the statins included in our analysis. Figure 4 illustrates that statins target numerous genetic pathways. These results provide the basis for existing expectations for empirically testing the influence of statins on LDL-c in the general population.

The top left triad of Fig. 4 integrates the significant interaction loci for LDL-c with previously identified drug-gene targets to highlight pharmacodynamic mechanisms of statins. Notably, as statins are HMG-CoA reductase inhibitors, the HMGCR (3-hydroxy-3-methylglutaryl-CoA reductase) gene has the strongest a priori expectation for a gene-statin interaction. The lack of a significant interaction between statins and HMGCR, therefore, is extremely intriguing. While the lack of a HMGCR -statin interaction finding may imply a superficial disconnect between our results and the existing literature, the significant gene-statin interactions we identified paint a consistent picture of LDL-c regulation lending credence to our findings. The five genome-wide significant interaction loci have been previously associated with LDL-c with well characterized regulatory descriptions. LDLR, APOB, and APOE are existing secondary targets for statins, and FADS2 and SUGP1 have previously been associated with LDL-c levels. Specifically, SUGP1 stimulates alternative splicing of HMGCR and decreases its transcript stability, subsequently reducing cholesterol synthesis. This, in conjunction with the effect of LDLR, increases LDL-c uptake in the liver^20,21. Independently, both APOB and APOE play pivotal roles in regulating lipid metabolism and homeostasis, notably in liver uptake and clearance via the LDL receptor, as coded for by LDLR^22,23. Further, FADS2 plays a role in the regulation of cholesterol accumulation related to polyunsaturated fatty acids as the liver preferentially converts polyunsaturated fatty acids into ketone bodies rather than into LDL-c. The gene-statin interactions at these loci imply that statins will be more effective for individuals who have protective alleles in these regions. Taken together, the results paint a nuanced picture for how statins dampen the genetic associations between these loci and LDL-c by interfering with multiple biological and genomic pathways that regulate cholesterol uptake, metabolism, and homeostasis.

The top right triad of Fig. 4 identifies 15 genetic associations with LDL-c that remain significant despite statin use. If statins interrupted all the genetic pathways that culminate in LDL-c, we would not observe any genetic associations with LDL-c for people who are taking statins. These residual genetic associations, therefore, suggest avenues for 1) drug improvement, 2) drug repurposing, and 3) novel drug development. The first, and most straightforward implication of this analysis is for drug improvement. Specifically, associations with HMGCR (rs118039278), LDLR (rs17242381), and APOE (rs72654473) remain significant after taking statins, suggesting there is room for drug improvement that would more effectively disrupt these genetic pathways that increase endogenous LDL-c. These results do not imply that statins are not effective, but instead that by more accurately targeting these variants we can further regulate LDL-c and increase the effectiveness of statin-like drugs. This set of genes is also instructive for drug repurposing. For example, drugs, such as fenofibrate, warfarin, and ACE inhibitors, are commonly prescribed to treat related CVDs, such as hypertriglyceridemia, blood clots, or hypertension. As high LDL-c is often comorbid with other CVDs, drugs that treat these related disorders may also influence the biological pathways that affect LDL-c. Caution is necessary when translating these results, as drug repurposing can increase the risk of adverse drug-drug interactions. Finally, the residual genetic associations may highlight gene targets for novel drug development. Finding a residual genetic association for PCSK9 (Proprotein convertase subtilisin/kexin type 9 – rs1159147) exemplifies the potential for novel drug discovery. PCSK9 is the target gene of an emerging class of medications (PSCK9 inhibitors) used to lower LDL-c that work on a different set of biological pathways than statins. Because PCSK9 inhibitors were rarely prescribed to patients in our dataset, they were thus excluded from the analysis. Accordingly, our results “post-dict” the effectiveness of targeting PCSK9, underscoring the validity of our method for identifying novel drugs that could treat hyperlipidemia in the future. Other genes in Fig. 4 that are in printed black (e.g. CELSR2) are not targets of existing medications and therefore may be ideal for similar novel drug development.

Statins lower cholesterol for almost everyone, but responsiveness to statins is a heterogeneous function of individual genetic differences. Despite phenotypic and genetic correlations among LDL-c, HDL-c, and TG, the current study demonstrates that statins disrupt the specific genetic factors that contribute to the heritability of LDL-c, thereby altering the genomic architecture of lipids. This finding is consistent with the observation that statins lower the heritability of LDL-c but leave the heritability of both HDL-c and TG unchanged. We identified five genome-wide significant gene-statin interactions that highlight genomic pathways that allow statins to lower LDL-c. The specific variants that enhance the efficacy of statins can inform the underlying biology of LDL-c as well as provide a starting point for examining personalized medicine approaches for hyperlipidemia. Furthermore, our analyses identified 15 loci that remained genome-wide significant despite statin treatment. Many of these associations are in genes with well-validated links to LDL-c implying that statins do not eliminate the genetic risk for hyperlipidemia. Both the gene-statin interactions and the residual genetic associations have important pharmacogenomic implications. Identifying specific variants that interact with statins provides insights into biological mechanisms of action demonstrating how statins lower LDL-c, as well as suggesting ways of determining who will benefit most from statin use. Conversely, genetic associations that remain significant in statin users highlight potential gene-targets for drug improvement, repurposing, and development.

Gene-statin interaction analyses provide pharmacogenomic insights from both disease- and patient centered perspectives. From a disease-centered perspective, knowing which genes interact with statin use informs the biology of hyperlipidemia by identifying specific genomic pathways that statins interrupt. Many pharmacogenomic results have focused on the pharmacokinetic aspects of medication use (e.g., absorption, bioavailability, distribution, metabolization, and excretion). The broad implication of these findings is that poor metabolizers will require lower doses while ultrarapid metabolizers require higher doses to achieve optimal effects. Such findings are predicated on the central role of cytochrome P450 genes (e.g. CYP2D6 or CYP3A4). Because we did not examine dosage or other factors characterized by the movement of drugs throughout the body, we did not identify any genes from this class. Instead, our results highlight central pharmacodynamic aspects of statin use (e.g., mechanisms of action with respect to genomic pathways), emphasizing a distinct perspective for personalized pharmacogenomic research. By further investigating how statins affect these pharmacodynamic genetic pathways and then integrating them with the existing pharmacokinetic results, it will become possible to identify individuals who are likely to benefit more (or less) from statin use based on their unique genotypes.

From a patient-centered perspective, our results highlight five genotypes that are particularly responsive to statin use, allowing clinicians to begin tailoring treatments to individual patients. While statins decrease LDL-c by 33 mg/dL on average, each protective allele from APOB, FADS2, LDLR, SUGP1, and APOE provides an additional reduction of 1.35, 1.72, 3.17, 2.53, and 5.75 mg/dL per allele, respectively. Thus, if a hypothetical individual had one protective allele at each of these five loci, compared with no protective alleles, we would predict a reduction in LDL-c of 33 mg/dL from the main effect plus an additional reduction of 14.5 mg/dL due to their personal genetic sensitivity to statins. Knowing who will benefit the most from a prescription medication, based on their genotype, has profound implications for prescribing medications, titrating dosage, and reducing the probability of adverse medication-induced side effects.

From a mechanistic perspective, existing pharmacological evidence demonstrates that the principal role of statins is to inhibit the conversion of HMG-CoA into mevalonate primarily by interfering with the function of HMGCR ²⁴, whereas our analyses implicate other relevant loci. Specifically, while we did not find direct evidence for HMGCR sensitivity to statins, the five loci we identified have known pharmacodynamic regulatory roles in statin metabolism that may indirectly affect HMGCR function. Nevertheless, our analyses suggest that more research needs to be done to refine our understanding of the complex biological relationships between HMGCR, statins, and LDL-c in very large population samples. These results suggest that statins may bind to, and inactivate, HMG-CoA in a way that does not preferentially benefit individuals with a particular HMGCR allele. Accordingly, we do not see an interaction between statins and HMGCR that lowers LDL-c. More research is needed to unpack this intriguing non-finding.

Fifteen loci remained genome-wide significant in the statin use group. These residual genetic associations highlight statin-relevant drug targets. Importantly, while some of the identified loci are current statin targets, increasing the effectiveness of statins to further interrupt these genetic pathways could improve patient outcomes. Specifically, we identified gene-statin interactions with variants in APOB, APOE and LDLR, and therefore it may be relatively straightforward to increase statin efficacy by more efficiently interrupting these genetic pathways. Residual genetic effects can also be instructive for drug repurposing. For example, our analyses identified several FDA-approved medications that are not used as cholesterol-lowering agents: some that treat frequently comorbid conditions (i.e., warfarin for blood clots), and others for unrelated conditions (i.e., hormonal contraceptives). Accordingly, drugs that target genes that remain associated with LDL-c after taking statins are of interest for two reasons. First, it is possible that they have the potential to further lower LDL-c, and thus could be used as an adjunctive treatment for individuals whose LDL-c is not sufficiently lowered by statins alone. Second, drugs that target the residual LDL-c associations may have a higher likelihood of driving adverse reactions to statins. Further research is required to disaggregate these two possibilities. Finally, we identified several genes that are not currently targeted by existing drugs. Existing evidence suggests novel drugs based on standard GWAS analyses are more likely to pass Stage 3 clinical trials than drugs developed using other methods^25–27. Accordingly, just as standard GWAS have been effective for identifying genetic pathways that can be leveraged when developing pharmacological interventions²⁸, the genomic pathways that remain significant for individuals who are already taking statins may be even more effective for drug development as they highlight treatment resistant genes.

Our analytical methods provide a novel approach for assessing the genomic mechanisms of medication use broadly, which we anticipate will become increasingly applicable as the availability of high-quality genomic data in medicated samples continues to grow. The pharmacogenomic hypothesis is that medication will disrupt targeted genomic pathways, lower the heritability of the phenotype, or reduce or eliminate the genetic associations with variants in the targeted genes. Currently however, medications are typically developed to target specific genetic pathways in drug naïve populations and tested with clinical trials. Because of this strategy, most current research excludes individuals who take statins, or other cholesterol-lowering medications. Confirming the efficacy of statins in a large general population sample of both users and non-users provides a unique opportunity to examine the pharmacogenomic effects of the observed and intended gene targets. Our analyses go further than the existing literature by highlighting both the genomic pathways that are altered by statin use, as well as those that remain significant after use to identify biological pathways that existing medications do not abolish. However, we refrain from making causal inferences based on these results, as respondents in population-based samples are not randomly assigned to statin use. Those who take statins likely have a higher genetic predisposition for elevated LDL-c. Reciprocally, people with relatively low LDL-c are less likely to have genotypes that result in elevated cholesterol, and thus have no need for statins (which they would not have been prescribed in the first place). As non-random statin treatment could potentially result in collider bias²⁹, we conducted a simulation study (Supplement C) that demonstrated the genetic marginal effects are unaffected by non-random statin use but the statin use coefficient ($\:{\widehat{\beta\:}}_{2}$ in Eq. 2) was strongly underestimated. As such, the genetic marginal effects we rely on in this study are robust to the fact that only individuals with high cholesterol took statins in our sample.

Current pharmacogenomic studies tend to utilize small, targeted, randomized controlled trials^30,31 that lack the statistical power to detect genome-wide targets that could examine individual drug sensitivity. Small sample sizes, however, preclude genome-wide gene-drug interaction analyses that require large sample sizes to overcome multiple testing corrections. Existing evidence supports the fact that our novel gene-drug interaction methods can further refine our understanding of the biological mechanisms that regulate lipid production and metabolism, thereby improving the effectiveness of treatment by focusing attention on the genetic signatures that are not minimized by statins.

The current study should be interpreted in light of a few limitations. First, to maximize the statistical power to detect SNP-statin interactions, we collapsed statins into a single, binary moderator, which has several implications. Because statin use was assessed via self-report in an unselected population sample, we assume that respondents answered accurately and patients were following the treatment regimens established with their physician, including dosage and compliance. However, because medication use was based on self-reported data, we were confident in the respondent’s accuracy regarding the type of medications they used, while acknowledging the specific dosages had low levels of reliability. Because dosage information is particularly relevant for identifying pharmacokinetic effects, our analyses were not well-positioned to detect cytochrome P450 genes. Nevertheless, we detected relevant pharmacodynamic processes related to the interaction signals. Further, and related to statistical power, by collapsing the five discrete statins into a single dichotomous variable, we obscured the inherent pharmacological differences between statins. For example, atorvastatin is associated with a larger reduction in LDL-c levels compared with pravastatin³². Because specific statins target different biological pathways, future research should explore the potential differential effect of each individual statin. Finally, the current study was restricted to individuals of European ancestry, limiting the generalizability of the current results to ensure that the results were not a function of population stratification. The extreme discrepancy in sizes of the ancestry groups in the UK Biobank, as well as other existing genomic data frames, limit the ability to conduct well-powered analyses in non-European ancestry groups. While genetic associations with cardiovascular phenotypes show a reasonable degree of consistency across ancestries^33,34, replication remains necessary.

In summary, we demonstrated that statin use interferes with lipid levels by selectively reducing the heritability of LDL-c. We identified several loci that have genome-wide significant gene-statin interactions that provide insights into the biological mechanisms of action and suggest a process of determining which patients will benefit most from statin use. Furthermore, our analyses identified 15 loci that remained genome-wide significant despite treatment. Many of these associations are in genes with well-validated links to LDL-c implying that statins do not entirely break the genetic risk for high cholesterol, or that some genomic pathways are resistant to pharmacological intervention. These results are an important step towards using personal genomic information to provide more targeted treatments to patients with hyperlipidemia.

Sample. Data for this project come from the UK Biobank (UKB Application 57923). The full UKB dataset has approximately 500K UK individuals^35–37. This sample is substantially larger and more representative of the UK general population than any other publicly available, genetically-informed dataset. For the current analyses, we restrict the sample to individuals of European ancestry for two reasons. First, as allele frequencies vary across ancestral populations, and if samples from multiple ancestral populations are analyzed jointly, population stratification could undermine the validity of the genetic associations^38,39. Including ancestry PCs reduces, but does not entirely remove, the impact of population stratification⁴⁰. Second, existing non-Europeans sample sizes do not provide sufficient power for genome-wide analyses, both in our sample and in other available datasets^41,42. The vast sample size provides ample power to detect genomic moderation.

Outcome Variables. Our analyses focus on LDL-c, HDL-c, and TG. All lipid assays were conducted on blood samples, consistent with internationally recognized testing and calibration procedures. Lipid were initially measured in mmol/L and converted to mg/dL to simplify interpretation for American audiences. Outcomes were treated as quantitative traits in the moderated GWAS analyses.

Moderator. Statins are a class of medications that were developed in the 1990s to lower cholesterol, specifically LDL-c. There are numerous statins that are commonly prescribed for hyperlipidemia. We focus on patients who self-reported taking atorvastatin, fluvastatin, pravastatin, rosuvastatin, or simvastatin (or the brand-name equivalents). Patients were coded as statin users if they took any statin regardless of whether they took any other cholesterol-lowering medication (e.g., PCSK9 inhibitors) or used vitamin/dietary supplements to reduce cholesterol (e.g., niacin). Patients were coded as non-statin users if they only used vitamin/dietary supplements. Patients were excluded from all analyses if they ONLY used PCSK9 inhibitors to reduce their cholesterol.

Genotypes. DNA was extracted from blood samples taken from all willing participants. The DNA preparation and extraction was performed at UKB, genotyping was conducted by Affymetrix, and the initial quality control of the data was conducted by the Wellcome Trust Centre for Human Genetics. Genotype imputation was conducted in IMPUTE2 using the UK10K as a reference panel. Imputed genotype data was used in all moderated GWAS analyses. Analyses were restricted to individuals of European ancestry to reduce the impact of population stratification on the results. For some individuals, individual SNPs did not pass quality control thresholds, and were excluded from the analyses for those SNPs. This resulted in a maximum sample size of 389,440.

Covariates. To minimize the impact of known predictors of lipid variability, all the moderated GWAS analyses controlled for biological sex, age, the first 10 ancestry principal components (PCs), as well as interactions between the PCs and statins. The statin-PC interactions further account for the possibility of population stratification specific to the SNP-statin interaction coefficient that is not shared with the genetic main effects.

Genome Wide Gene-Statin Interaction Analysis. All the moderated GWAS analyses of raw data were conducted in R⁴³ using the moderated GWAS functions in GW-SEM ^44,45. Moderated GWAS analyses are an extension of standard GWAS where users conduct a series of regression analyses where a phenotype is regressed on each SNP in a genomic assay, adjusting for a set of predefined covariates such as age, sex, and the first 10 ancestry principal components to account for unobserved population stratification⁴⁶. As such, the standard GWAS model for each variant is:

$$\:{Y}_{i}=\:{\widehat{\beta\:}}_{0}+\:{\widehat{\beta\:}}_{1}{SNP}_{ij}+\:\widehat{\gamma\:}{Covariates}_{i}+\:{ϵ}_{i}$$

Where, in the current case, $\:{Y}_{i}\:$is the level of LDL-c, HDL-c, or TG for the i^th person, SNP_ij is the j^th genetic variant for the i^th person, and Covariates_i are the covariates for the i^th person (e.g., sex, age, and PCs). The estimate of the genetic association is $\:{\widehat{\beta\:}}_{1}$, which is the most interesting parameter for most GWAS. The other estimates, $\:{\widehat{\beta\:}}_{0}$ and $\:\widehat{\gamma\:}$, denote the intercept and regression coefficients for the covariates.

Genome-wide moderation analyses expand the standard GWAS model by regressing a phenotype on three independent variables: a) each SNP in the genomic assay, b) each individual’s statin use (the moderator), and the interaction between each SNP and the moderator, in addition to the covariates. To further rule out the possibility that population stratification may bias the interaction coefficients for the individual variants, interactions between the moderator and each of the first 10 ancestry principal components are also included as covariates¹³. Accordingly, the model for the genome-wide moderation analyses for each variant is:

$$\:{Y}_{i}=\:{\widehat{\beta\:}}_{0}+\:{\widehat{\beta\:}}_{1}{SNP}_{ij}+\:{\widehat{\beta\:}}_{2}{Statin}_{i}+\:{\widehat{\beta\:}}_{3}{SNP}_{ij}{Statin}_{i}+\:\widehat{\gamma\:}{Covariates}_{i}+\:{ϵ}_{i}$$

In the moderated GWAS model, both main effects, $\:{\widehat{\beta\:}}_{1}$ and $\:{\widehat{\beta\:}}_{2}$, depend on the interaction effect, $\:{\widehat{\beta\:}}_{3}$. Importantly, $\:{\widehat{\beta\:}}_{3}$ provides a test of whether the effect of a SNP on the phenotype changes at different levels of the environment. This parameter is difficult to interpret directly. Therefore, it is advantageous to calculate genetic marginal effects to examine how genetic association vary across levels of an environment.⁴⁶

Finally, the interaction coefficient provides a formal test of the difference in the marginal genetic effects (for each locus) between the statin-takers and the non-statin-takers. Specifically, the interaction coefficient tells us whether the genetic associations differ in the statin vs non-statin groups. This coefficient must be interpreted with caution as interactions may amplify or diminish the main effect of the SNP on the phenotype (lipid) potentially leading to non-sensical conclusions. For example, if the interaction effect ($\:{\widehat{\beta\:}}_{3}$) is in the opposite direction of the main effect ($\:{\widehat{\beta\:}}_{1}$), it is possible to have a significant interaction coefficient despite not having significant genetic associations in either the statin or non-statin groups. Accordingly, we restrict our discussion of the significant interaction coefficients to those with meaningful substantive interpretations where the marginal effect is significant for at least the statin users or non-users.

Marginal Genetic Effects. Summary statistics from the moderated GWAS of each of the lipids was used to calculate the marginal genetic effects and standard errors for the statin users and non-users. All calculations were done automatically in GW-SEM⁴⁵. Genetic marginal effects are the association between a SNP and a phenotype at a specific level of an environment. In a standard GWAS model, the genetic marginal effect of the SNP is the regression coefficient ($\:{\widehat{\beta\:}}_{1}$). Effectively, the standard GWAS regression coefficient is the effect of the SNP on the phenotype at the mean level of the moderating environment. In a moderated GWAS, the genetic marginal effect is a function of both genetic and moderating factors ⁴⁶. To calculate genetic marginal effects ($\:{\widehat{\beta\:}}_{ME})$, we take the first derivative of the GxE GWAS model with respect to the SNP, leaving:

$$\:{\widehat{\beta\:}}_{ME}=\:{\widehat{\beta\:}}_{1}+\:{\widehat{\beta\:}}_{3}Statin$$

As respondents either took a statin (1) or did not (0), the marginal effect for the no statin group is simply $\:{\widehat{\beta\:}}_{1}$ and the marginal effect for the statin group is $\:{\widehat{\beta\:}}_{1}+\:{\widehat{\beta\:}}_{3}$. We can then calculate the standard errors of the genetic marginal effects (SE_ME) using parameters from the variance-covariance (vcov) matrix of the moderated GWAS model for each SNP by:

$$\:{SE}_{ME}=\sqrt{{\sigma\:}_{{\beta\:}_{1}}^{2}\:+{\sigma\:}_{{\beta\:}_{3}}^{2}{\text{S}\text{t}\text{a}\text{t}\text{i}\text{n}}^{2}+\:2{{\sigma\:}}_{({\beta\:}_{1},{\beta\:}_{3})}\text{S}\text{t}\text{a}\text{t}\text{i}\text{n}}$$

inserting the corresponding value of the statin group (0 or 1) that was used to calculate the genetic marginal effects. Note that for the no statin group, the SE reduces to $\:\sqrt{{\sigma\:}_{{\beta\:}_{1}}^{2}\:}$ which is just the standard error of $\:{\widehat{\beta\:}}_{1}$. The additional terms used to calculate the SE in the statin group incorporate any collinearity between the main effect and the interaction effect. After calculating the genetic marginal effect and the standard error, the z-statistic and p-value are easily calculated for use in subsequent analyses. This process is then repeated for each SNP that is analyzed. The process is automated in GW-SEM⁴⁵, which is the only software platform that currently stores the $\:{{\sigma\:}}_{({\beta\:}_{1},{\beta\:}_{3})}$ statistic necessary to calculate the standard error of the marginal effects.

Pharmacogenomic Analyses. The aim of the pharmacogenomic analyses was to collate our moderated GWAS results with pharmaceutical information regarding genetic targets for existing drugs using the pharmGKB.org database (the most extensive pharmacogenomic database available)¹⁹. We first extracted known gene targets for the five separate statins used as moderators in our analyses: atorvastatin, fluvastatin, pravastatin, rosuvastatin and Simvastatin. The bottom triad of Fig. 4 presents gene-targets for two or more of the statins. A full list of gene-targets is available in SI 7. Second, we queried the pharmGKB.org database for drugs that are known to target the five genes with significant interaction effects: APOB, FADS2, LDLR, SUGP1, and APOE. The corresponding gene-drug combinations for the genes that interacted with statin use are presented in the left triad of Fig. 4. Finally, we queried the pharmGKB.org database for drugs that target the 15 genes that remained significant for people who were taking statins. The corresponding gene-drug combinations for the residual genetic associations in the statin group are presented in the right triad of Fig. 4.

Competing Interests.

The authors have no competing interests to declare.

Author Contribution

B.V. conceptualized the project, developed the methods, oversaw the analyses, and wrote the manuscript. J.H., A.M.A., and S.E.B. conducted the analyses. C.W.T, A.J.C, and J.M.H. edited the manuscript and provided scientific guidance.

Data Availability

Data was accessed through the UK Biobank Account Management System (application number 57923). In accordance with UK Biobank data access policies, raw can be accessed through the UK Biobank AMS after completing a data access agreement, but the sharing of raw phenotypic or genetic data is strictly prohibited. The detailed results of the major genome-wide significant loci are listing in supplementary tables. A complete set of summary statistics will be made available following the publication of the manuscript.

Holmes MV, Asselbergs FW, Palmer TM, et al. Mendelian randomization of blood lipids for coronary heart disease. Eur Heart J. 2015;36(9):539–550. doi:10.1093/eurheartj/eht571
Yengo L, Sidorenko J, Kemper KE, et al. Meta-analysis of genome-wide association studies for height and body mass index in ∼700000 individuals of European ancestry. Hum Mol Genet. 2018;27(20):3641–3649. doi:10.1093/hmg/ddy271
Sandhu MS, Waterworth DM, Debenham SL, et al. LDL-cholesterol concentrations: a genome-wide association study. The Lancet. 2008;371(9611):483–491. doi:10.1016/S0140-6736(08)60208-1
Pharoah PD, Hollingworth W. Cost effectiveness of lowering cholesterol concentration with statins in patients with and without pre-existing coronary heart disease: life table method applied to health authority population. BMJ. 1996;312(7044):1443–1448. doi:10.1136/bmj.312.7044.1443
Scandinavian Simvastatin Survival Study Group. Randomised trial of cholesterol lowering in 4444 patients with coronary heart disease: the Scandinavian Simvastatin Survival Study (4S). Lancet Lond Engl. 1994;344(8934):1383–1389.
Shepherd J, Cobbe SM, Ford I, et al. Prevention of coronary heart disease with pravastatin in men with hypercholesterolemia. West of Scotland Coronary Prevention Study Group. N Engl J Med. 1995;333(20):1301–1307. doi:10.1056/NEJM199511163332001
Ference BA, Yoo W, Alesh I, et al. Effect of long-term exposure to lower low-density lipoprotein cholesterol beginning early in life on the risk of coronary heart disease: a Mendelian randomization analysis. J Am Coll Cardiol. 2012;60(25):2631–2639. doi:10.1016/j.jacc.2012.09.017
Cholesterol Treatment Trialists’ (CTT) Collaboration. Efficacy and safety of more intensive lowering of LDL cholesterol: a meta-analysis of data from 170 000 participants in 26 randomised trials. The Lancet. 2010;376(9753):1670–1681. doi:10.1016/S0140-6736(10)61350-5
Bytyçi I, Penson PE, Mikhailidis DP, et al. Prevalence of statin intolerance: a meta-analysis. Eur Heart J. 2022;43(34):3213–3223. doi:10.1093/eurheartj/ehac015
Keen HI, Krishnarajah J, Bates TR, Watts GF. Statin myopathy: the fly in the ointment for the prevention of cardiovascular disease in the 21st century? Expert Opin Drug Saf. 2014;13(9):1227–1239. doi:10.1517/14740338.2014.937422
Bitzur R, Cohen H, Kamari Y, Harats D. Intolerance to Statins: Mechanisms and Management. Diabetes Care. 2013;36(Supplement_2):S325-S330. doi:10.2337/dcS13-2038
Clarke R, Peden JF, Hopewell JC, et al. Genetic variants associated with Lp(a) lipoprotein level and coronary disease. N Engl J Med. 2009;361(26):2518–2528. doi:10.1056/NEJMoa0902604
Keller MC. Gene × environment interaction studies have not properly controlled for potential confounders: the problem and the (simple) solution. Biol Psychiatry. 2014;75(1):18–24. doi:10.1016/j.biopsych.2013.09.006
Verhulst Brad, Joshua N. Pritikin, James Clifford, Prom-Wormley EC. The importance of genetic marginal effects for the interpretation of gene-environment interactions in the genome wide association studies (GWAS). Behav Genet.
Bulik-Sullivan BK, Loh PR, Finucane HK, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet. 2015;47(3):291–295. doi:10.1038/ng.3211
Willer CJ, Schmidt EM, Sengupta S, et al. Discovery and refinement of loci associated with lipid levels. Nat Genet. 2013;45(11):1274–1283. doi:10.1038/ng.2797
Graham SE, Clarke SL, Wu KHH, et al. The power of genetic diversity in genome-wide association studies of lipids. Nature. 2021;600(7890):675–679. doi:10.1038/s41586-021-04064-3
Whirl-Carrillo M, McDonagh EM, Hebert JM, et al. Pharmacogenomics knowledge for personalized medicine. Clin Pharmacol Ther. 2012;92(4):414–417. doi:10.1038/clpt.2012.96
Whirl-Carrillo M, Huddart R, Gong L, et al. An Evidence-Based Framework for Evaluating Pharmacogenomics Knowledge for Personalized Medicine. Clin Pharmacol Ther. 2021;110(3):563–572. doi:10.1002/cpt.2350
Kim MJ, Yu CY, Theusch E, et al. SUGP1 is a novel regulator of cholesterol metabolism. Hum Mol Genet. Published online May 20, 2016:ddw151. doi:10.1093/hmg/ddw151
Trapani L. Regulation and deregulation of cholesterol homeostasis: The liver as a metabolic “power station.” World J Hepatol. 2012;4(6):184. doi:10.4254/wjh.v4.i6.184
Behbodikhah J, Ahmed S, Elyasi A, et al. Apolipoprotein B and Cardiovascular Disease: Biomarker and Potential Therapeutic Target. Metabolites. 2021;11(10):690. doi:10.3390/metabo11100690
Getz GS, Reardon CA. Apoprotein E as a lipid transport and signaling protein in the blood, liver, and artery wall. J Lipid Res. 2009;50:S156-S161. doi:10.1194/jlr.R800058-JLR200
Friesen JA, Rodwell VW. The 3-hydroxy-3-methylglutaryl coenzyme-A (HMG-CoA) reductases. Genome Biol. 2004;5(11):248. doi:10.1186/gb-2004-5-11-248
Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32(1):40–51. doi:10.1038/nbt.2786
Barrett JC, Dunham I, Birney E. Using human genetics to make new medicines. Nat Rev Genet. 2015;16(10):561–562. doi:10.1038/nrg3998
King EA, Davis JW, Degner JF. Are drug targets with genetic support twice as likely to be approved? Revised estimates of the impact of genetic support for drug mechanisms on the probability of drug approval. PLoS Genet. 2019;15(12):e1008489. doi:10.1371/journal.pgen.1008489
Reay WR, Cairns MJ. Advancing the use of genome-wide association studies for drug repurposing. Nat Rev Genet. 2021;22(10):658–671. doi:10.1038/s41576-021-00387-z
Holmberg MJ, Andersen LW. Collider Bias. JAMA. 2022;327(13):1282. doi:10.1001/jama.2022.1820
Santos PCJL, Pereira AC. Type of LDLR mutation and the pharmacogenetics of familial hypercholesterolemia treatment. Pharmacogenomics. 2015;16(15):1743–1750. doi:10.2217/pgs.15.113
Leusink M, Onland-Moret NC, de Bakker PIW, de Boer A, Maitland-van der Zee AH. Seventeen years of statin pharmacogenetics: a systematic review. Pharmacogenomics. 2016;17(2):163–180. doi:10.2217/pgs.15.158
Cannon CP, Braunwald E, McCabe CH, et al. Intensive versus Moderate Lipid Lowering with Statins after Acute Coronary Syndromes. N Engl J Med. 2004;350(15):1495–1504. doi:10.1056/NEJMoa040583
Kavousi M, Bos MM, Barnes HJ, et al. Multi-ancestry genome-wide study identifies effector genes and druggable pathways for coronary artery calcification. Nat Genet. 2023;55(10):1651–1664. doi:10.1038/s41588-023-01518-4
Kilpeläinen TO, Bentley AR, Noordam R, et al. Multi-ancestry study of blood lipid levels identifies four loci interacting with physical activity. Nat Commun. 2019;10(1):376. doi:10.1038/s41467-018-08008-w
Allen NE, Sudlow C, Peakman T, Collins R, UK Biobank. UK biobank data: come and get it. Sci Transl Med. 2014;6(224):224ed4. doi:10.1126/scitranslmed.3008601
Sudlow C, Gallacher J, Allen N, et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 2015;12(3):e1001779. doi:10.1371/journal.pmed.1001779
Fry A, Littlejohns TJ, Sudlow C, et al. Comparison of Sociodemographic and Health-Related Characteristics of UK Biobank Participants With Those of the General Population. Am J Epidemiol. 2017;186(9):1026–1034. doi:10.1093/aje/kwx246
Price AL, Butler J, Patterson N, et al. Discerning the ancestry of European Americans in genetic association studies. PLoS Genet. 2008;4(1):e236. doi:10.1371/journal.pgen.0030236
Price AL, Zaitlen NA, Reich D, Patterson N. New approaches to population stratification in genome-wide association studies. Nat Rev Genet. 2010;11(7):459–463. doi:10.1038/nrg2813
Patterson N, Price AL, Reich D. Population structure and eigenanalysis. PLoS Genet. 2006;2(12):e190. doi:10.1371/journal.pgen.0020190
Márquez-Luna C, Loh PR, South Asian Type 2 Diabetes (SAT2D) Consortium, SIGMA Type 2 Diabetes Consortium, Price AL. Multiethnic polygenic risk scores improve risk prediction in diverse populations. Genet Epidemiol. 2017;41(8):811–823. doi:10.1002/gepi.22083
Peterson RE, Kuchenbaecker K, Walters RK, et al. Genome-wide Association Studies in Ancestrally Diverse Populations: Opportunities, Methods, Pitfalls, and Recommendations. Cell. 2019;179(3):589–603. doi:10.1016/j.cell.2019.08.051
R Core Team. \proglangR: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2014. http://www.R-project.org
Verhulst B, Maes HH, Neale MC. GW-SEM: A Statistical Package to Conduct Genome-Wide Structural Equation Modeling. Behav Genet. 2017;47(3):345–359. doi:10.1007/s10519-017-9842-6
Pritikin JN, Neale MC, Prom-Wormley EC, Clark SL, Verhulst B. GW-SEM 2.0: Efficient, Flexible, and Accessible Multivariate GWAS. Behav Genet. Published online February 2021. doi:10.1007/s10519-021-10043-1
Verhulst B, Pritikin JN, Clifford J, Prom-Wormley E. Using Genetic Marginal Effects to Study Gene-Environment Interactions with GWAS Data. Behav Genet. 2021;51(3):358–373. doi:10.1007/s10519-021-10058-8

No competing interests reported.

SupplementaryInformationsub.docx

Download PDF

Reviews received at journal
15 Aug, 2024
Reviewers agreed at journal
07 Aug, 2024
Reviewers agreed at journal
31 Jul, 2024
Reviewers invited by journal
31 Jul, 2024
Editor assigned by journal
23 Jul, 2024
Submission checks completed at journal
22 Jul, 2024
First submitted to journal
10 Jul, 2024

You are reading this latest preprint version

Gene-statin interaction study identifies 9 loci that moderate statin effectiveness in blood lipids

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Online Methods

Declarations

Competing Interests.

Author Contribution

Data Availability

References

Additional Declarations

Supplementary Files

Status:

Version 1