Large-Scale Phenome-Wide Association Study of PCSK9 Variants Demonstrates Protection Against Ischemic Stroke
Background: PCSK9 inhibition is a potent new therapy for hypercholesterolemia and cardiovascular disease. Although short-term clinical trial results have not demonstrated major adverse effects, long-term data will not be available for some time. Genetic studies in large biobanks offer a unique opportunity to predict drug effects and provide context for the evaluation of future clinical trial outcomes.
Methods: We tested the association of the PCSK9 missense variant rs11591147 with predefined phenotypes and phenome-wide, in 337 536 individuals of British ancestry in the UK Biobank, with independent discovery and replication. Using a Bayesian statistical method, we leveraged phenotype correlations to evaluate the phenome-wide impact of PCSK9 inhibition with higher power at a finer resolution.
Results: The T allele of rs11591147 showed a protective effect on hyperlipidemia (odds ratio, 0.63±0.04; P=2.32×10−38), coronary heart disease (odds ratio, 0.73±0.09; P=1.05×10−6), and ischemic stroke (odds ratio, 0.61±0.18; P=2.40×10−3) and was associated with increased type 2 diabetes mellitus risk adjusted for lipid-lowering medication status (odds ratio, 1.24±0.10; P=1.98×10−7). We did not observe associations with cataracts, heart failure, atrial fibrillation, and cognitive dysfunction. Leveraging phenotype correlations, we observed evidence of a protective association with cerebral infarction and vascular occlusion. These results explore the effects of direct PCSK9 inhibition; off-target effects cannot be predicted using this approach.
Conclusions: This result represents the first genetic evidence in a large cohort for the protective effect of PCSK9 inhibition on ischemic stroke and corroborates exploratory evidence from clinical trials. PCSK9 inhibition was not associated with variables other than those related to LDL (low-density lipoprotein) cholesterol, atherosclerosis, and type 2 diabetes mellitus, suggesting that other effects are either small or absent.
See Editorial by Sullivan et al
Cardiovascular disease (CVD) accounts for over a quarter of annual deaths in the United States and is the most common cause of mortality globally.1,2 Hypercholesterolemia has long been established as one of the most important risk factors for the development of CVD3; statins are widely prescribed to reduce low-density lipoprotein cholesterol (LDL-C), thereby decreasing CVD and all-cause mortality.4,5 Statins achieve their LDL-C–lowering effects by inhibiting HMGCR6 and have become among the most widely prescribed drugs across all drug classes. However, many patients on medium- to high-intensity statin therapy are unable to achieve a clinically recommended reduction in LDL-C, leading to the consideration of other LDL-C–lowering therapies.7,8
PCSK9 inhibitors are an important new class of LDL-C–lowering drugs that were discovered based on human genetic studies.9,10 The gene product of PCSK9 binds to the LDLR and promotes its degradation.11 PCSK9 binds to the EGF-A domain of LDLR; additional sites of interaction might be created in the acidic environment of endosomes, where the affinity of PCSK9 for LDLR increases. Increased LDLR degradation inhibits the LDLR-mediated endocytosis of circulating LDL-C, thereby increasing LDL-C levels.11,12 Genetic studies have established the association of variants in the PCSK9 locus with hypercholesterolemia and atherosclerotic CVD,9,13,14 and fine-mapping efforts have revealed that multiple variants are independently associated with these phenotypes.15,16 PCSK9 variants can independently influence lipid levels10,17 and also act to modify therapeutic response to statins.18 Early evidence of this association generated interest in PCSK9 as a potential drug target,19 and 2 recent clinical trials have confirmed the protective effects of the monoclonal antibodies evolocumab and alirocumab on composite cardiovascular end points.20,21 By convention, we use the term PCSK9 inhibitors to refer to these antibodies. PCSK9 inhibitors are currently approved for patients with atherosclerotic CVD or familial hypercholesterolemia who are unable to achieve optimal LDL-C levels using statins alone.22 Although existing clinical studies show that they are generally well tolerated in the short term,20,23 questions remain about their potential adverse effects, cost-effectiveness, and long-term effects on type 2 diabetes mellitus (T2D) and cognitive dysfunction, among other phenotypes.24–26
A phenome-wide association study of PCSK9 loss-of-function (LoF) variants has the potential to predict adverse effects of PCSK9 inhibition and generate hypotheses about unknown beneficial effects. Comparing carriers of such LoF variants with noncarriers can simulate the effects of blocking the corresponding protein. Hence, this approach can offer insights into the beneficial and adverse drug effects that are likely to result from direct inhibition of PCSK9—the intended drug target; off-target effects and other effects, such as long-term immune responses, cannot be predicted using this approach. In the present study, we combined a hypothesis-driven analysis of 11 phenotypes defined a priori based on concurrent literature with a phenome-wide association study of PCSK9 LoF variants in 337 536 individuals of British ancestry from the UK Biobank (UKB). We leveraged the increase in power offered by focused hypothesis-driven testing and applied newly developed methods that use the correlation structure of phenotypes to reduce the multiple testing burden in our phenome-wide screen.
The data used in this study come from the UKB and are available to other researchers through a simple application process. The UKB was approved by the North West Multi-Center Research Ethics Committee, and all participants provided written informed consent to participate. Detailed methods are available in the Data Supplement.
We performed a phenome-wide association study of the PCSK9 missense variant rs11591147 (R46L, G/T) in 337 536 individuals of British ancestry in the UKB, beginning with association tests in a 11-phenotype hypothesis-driven set and subsequently testing phenome-wide association across 278 manually curated phenotypes. In the hypothesis-driven set, the rs11591147 T allele showed a protective effect on cholesterol medication status, coronary heart disease, and ischemic stroke but was not associated with T2D initially (Figure 1). After independent discovery and replication, we retested all phenotypes in the full data set of 337 536 individuals to obtain better effect size point estimates and screen for suggestive but nonsignificant associations; power estimates for all phenotypes are included in Table I in the Data Supplement. We observed a nominally significant association with all stroke in the full data set (odds ratio [OR], 0.82±0.11; P=0.01). To explore whether a potential relationship between rs11591147 and T2D was masked by lipid medication status, we analyzed this association in the full data set after adjusting for lipid medication and separately after stratifying by lipid medication status. rs11591147 was significantly associated with T2D after adjustment (OR, 1.24±0.10; P=1.98×10−7). We observed this association in the subset of individuals on lipid-lowering medications (OR, 1.34±0.15; P=2.30×10−7; n=60 548; ncases=14 214) but not among individuals not on lipid-lowering medications (OR, 1.11±0.12; P=0.07; n=276 988; ncases=6733). ORs illustrating the dose-response on the use of lipid-lowering medication in individuals heterozygous and homozygous for the rs11591147 (T) allele are shown in Figure III in the Data Supplement. In the hypothesis-generating phenome-wide analysis, the variant was significantly associated with a reduction in metabolic disorders, ischemic heart disease, coronary artery bypass graft operations, percutaneous coronary interventions, history of angina, and cholesterol medication status (Figure 2). The association with history of myocardial infarction was significant in the discovery subset but did not formally replicate using our a priori defined thresholds (Data Supplement). We observed nominally significant associations (P<0.05) with 19 other variables in the hypothesis-generating set (Table III in the Data Supplement).
We then ran a phenome-wide association screen for rs11591147 using Tree-Wide Association Study27 with a Posterior Probability of Association significance threshold of 0.75. We observed that the association with metabolic diseases was driven by disorders of lipoprotein metabolism (E78) and the association with ischemic heart disease was driven by all individual ICD-10 codes comprising that phenotype (I20-I25). Cerebrovascular phenotypes that showed association in the single-variant analysis were primarily related to infarction (I63) and occlusion (I65). We did not observe significant or suggestive associations across 15 other disease categories. In total, we observed 11 active (significantly associated) nodes for the association of rs11591147 across the phenome (Figure 3). Tree-Wide Association Study using a 6-variant genetic score (Table) weighted based on minor allele frequencies (Figure IV in the Data Supplement) and LDL effect sizes (Figure V in the Data Supplement) showed similar results.
In our phenome-wide analysis of 337 536 participants of the UKB, we studied associations of the PCSK9 LoF variant rs11591147 with 11 predefined phenotypes (based on prior literature), as well as across the whole phenome to discover previously unknown associations indicating opportunities for drug repositioning and risks of adverse drug effects. We made several important observations.
First, we observed robust and replicable associations of the PCSK9 missense variant rs11591147 with cholesterol medication, coronary heart disease, and ischemic stroke; the last has not been observed in prior large genetic studies. We also observed an association with T2D after adjusting for lipid-lowering medication.
Second, we did not observe associations of this variant with cataracts, epilepsy, and cognitive dysfunction, all potential side effects of PCSK9 inhibitors that have been suggested in prior literature; we did not see evidence suggesting these associations even after reanalyzing the full data set (statistical power in Table I in the Data Supplement).
Third, using 2 alternative methods to study associations of PCSK9 LoF variants across the phenome, we did not observe associations with phenotypes other than those associated with LDL-C, atherosclerosis, and T2D, indicating that if there are additional side effects of PCSK9 inhibition, they are likely to be of small effect sizes or rare. We observed nominal associations with 19 variables across the phenome that were not significant using our discovery threshold (Table III in the Data Supplement). PCSK9 inhibitors can also have other monoclonal antibody-related side effects that are impossible to test using this approach.
Phenome-wide association studies of gene-disrupting variant that influence the activity of drug target genes represent a promising strategy to predict drug effects, which can include adverse effects or potential drug repositioning opportunities. Such studies depend on the availability of relevant coding variants that can mimic pharmaceutical interventions, although other variants associated with gene activity, such as expression quantitative trait loci, can also be used as proxies. Although randomized controlled trials are the gold standard for evidence of primary drug effects, they are costly and time-consuming; genetic studies can provide preliminary evidence predicting the outcome of such studies. Further, side effects reported in clinical trials are sometimes equally likely in the placebo and treatment arms, can be influenced by factors other than primary drug action,28 and are often too rare to discover in phase 3 trials. Studying gene-disrupting variant effects in large genomic cohorts is, therefore, a complimentary method to predict both beneficial and adverse side effects associated with primary drug action. There are limitations to this approach as well; studying the phenotypic consequences of LoF of a protein does not fully capture the range of effects that can occur because of therapeutic inhibition or neutralization of the activity of the gene product. For instance, this approach is unable to capture the downstream effects of an immune response to the therapeutic agent or explore the interaction of therapy with rare or extreme physiological stresses.
The development and success of PCSK9 inhibitors is one of the best examples of the value of genetics in drug discovery.9,20,22,29 Like studies on statin effects, several clinical trials and genetic studies have explored the effects of PCSK9 inhibitors on lipid levels, coronary heart disease, mortality, stroke, T2D, and cognitive effects; these studies are often hampered by small sample sizes and offer conflicting evidence that can impact our understanding of unintended drug effects and safety considerations.11,30,31 Hence, we developed a list of 11 hypothesis-driven phenotypes with careful definitions to evaluate the evidence of association of PCSK9 variants with phenotypes of interest in concurrent literature.20,23,25,26
Lipid Levels and Coronary Heart Disease
Consistent with many prior studies,14,17,20 we observed a protective effect of the PCSK9 LoF variant rs11591147 on cholesterol medication and coronary heart disease. Because lipid measurements have not been completed in the UKB, self-reported cholesterol medication served as a proxy response variable for high lipid levels. Using measured lipid levels is likely to increase power, so our effect estimate for this phenotype is likely to be biased toward the null. Effects of PCSK9 inhibition (genetic and pharmacological) on LDL-C levels and coronary heart disease are well established.24 Data from multiple clinical studies have shown a protective effect of PCSK9 inhibition on cardiovascular outcomes.29,32 Genetic studies have also confirmed this protective effect on varied individual definitions and composite end points of adverse cardiovascular outcomes.33
There are conflicting results about the effects of PCSK9 inhibitors on stroke, and only exploratory clinical evidence of this association is available. However, prior knowledge of the pathogenesis of stroke suggests that there might be an effect. Atherosclerotic plaques have a predisposition to thrombosis; ischemic heart disease can occur as the downstream consequence of calcification, ulceration, and thrombosis of coronary plaques. The pathogenesis of ischemic stroke can involve different mechanisms but often is a downstream consequence of an embolic thrombus released from the plaques in the carotid artery or another location. Ulceration of carotid plaques is often observed before thrombus formation and subsequent cerebrovascular ischemic events.34 Based on the protective effect of PCSK9 inhibition on atherosclerotic coronary heart disease, we explored whether its effect extends to cerebrovascular ischemic events; if associated, PCSK9 variants are likely to act primarily through lowering LDL-C levels. Given the large sample size in the UKB, we observed a significant protective association between rs11591147 and ischemic stroke (OR, 0.60±0.17; P=0.002) that replicated using our thresholds. In addition, we observed a nominally significant association with all stroke in the full data set (OR, 0.82±0.11; P=0.01), which was not significant in the discovery subset using our predefined criteria. We did not observe an association with hemorrhagic stroke (P=0.81). Further, Tree-Wide Association Study single-variant analysis showed an association with cerebral infarction and occlusion. Our results suggest that PCSK9 inhibitors are likely to be protective against ischemic stroke, which is consistent with our knowledge about associations of LDL-C, atherosclerosis, and ischemic stroke.
Prior clinical studies have included ischemic stroke as part of composite end points or only addressed it in exploratory analyses. Indeed, nonfatal ischemic stroke was included in the primary end point of the recent ODYSSEY Outcomes: Evaluation of Cardiovascular Outcomes After an Acute Coronary Syndrome During Treatment With Alirocumab clinical trial. However, clinical studies of anti-PCSK9 antibodies have involved shorter time spans and possibly individuals with more advanced atherosclerosis than in the current study and would be less likely to report a significant association between PCSK9 inhibition and stroke. A handful of small genetic analyses have shown suggestive evidence for the association between PCSK9 variants and ischemic stroke,11,35 but larger studies have offered conflicting evidence. Two recent meta-analyses using genetic data36 and randomized clinical data37 showed no association between PCSK9 inhibition and ischemic stroke. The association between clinical conditions resulting in altered lipid levels, such as familial hypercholesterolemia, and stroke is unclear; prospective studies have shown no relationship between familial hypercholesterolemia and stroke mortality but show a relationship between familial hypercholesterolemia and peripheral artery disease.38 A recent meta-analysis of clinical data suggested that all cholesterol-lowering interventions should reduce the risk of stroke.39 Our results using a Mendelian randomization approach in the UKB support this hypothesis. The genetic study of PCSK9 variants by Hopewell et al,36 which showed no association between rs11591147 and ischemic stroke in data from the METASTROKE consortium, was a meta-analysis of 12 case-control studies with a total of 10 307 cases and 19 326 controls; our study uses the UKB—a genetic cohort with 2622 cases and 334 914 controls (in the study sample). Ischemic stroke was defined similarly using clinical diagnoses in both studies, but cases in some studies used by Hopewell et al were also confirmed using brain imaging. The difference in the observed effect on ischemic stroke in the 2 studies could be because of these slight differences in case definitions, variability among individual datasets included in the meta-analysis, difference in statistical power, or chance.
Type 2 Diabetes Mellitus
Although we did not observe an association of rs11591147 with T2D in the hypothesis-driven screen, we observed this association after adjusting for lipid-lowering medications (OR, 0.24±0.10; P=1.98×10−7); lipid medication status likely serves as a proxy for high lipid levels in this analysis. Some genetic studies have suggested that LDL-C levels are causally linked with an increase in T2D risk independent of lipid-lowering medication.25,40 Meta-analyses of unintended effects of statin therapy have shown an increase in T2D risk,28 and genetic studies of variants in HMGCR and NPC1L1 (encoding the molecular targets of statins and ezetimibe, respectively) have shown increased risk of T2D associated with LDL-C–lowering variants.41 One study of PCSK9 variants showed a 6% (OR, 1.06) increase in T2D risk in individuals with alleles protective against high LDL-C levels; the detected risk increased to 12% (OR, 1.12) after adjustment for LDL-C levels.31 Another Mendelian randomization analysis showed a similar increase in T2D risk among carriers of PCSK9 variants.24 However, there is also conflicting evidence of this association. A pooled study of 10 phase 3 trials showed no increase in new-onset diabetes mellitus between alirocumab treatment and control groups during a follow-up period of 6 to 18 months.30 An analysis of the FOURIER trial (Further Cardiovascular Outcomes Research with PCSK9 Inhibition in Subjects with Elevated Risk) data showed that treatment with evolocumab did not worsen glycemia or increase the risk of new-onset diabetes mellitus.20 Our result confirms the evidence of association between LDL-C reduction and increased T2D risk seen across treatment modalities. Further, in a stratified analysis, rs11591147 was associated with T2D in individuals on lipid-lowering medication (OR, 1.34±0.15; P=2.30×10−7) but was not associated in individuals not on lipid-lowering medications (OR, 1.11±0.12; P=0.07). There was a large difference in the ratio of T2D cases to controls in these strata; there were 14 214 cases among individuals on medication (total n, ≈60K) and 6733 cases among individuals not on medication (total n, ≈277K). For additional insight, we explored stratification by type of lipid medication. In exploratory analyses, we observed an association of rs11591147 with T2D both in individuals on statins (OR, 1.35±0.15; P=1.97×10−7; n≈55K) and in those on other lipid-lowering medications (OR, 1.48±0.47; P=0.02; n≈5K). Cataracts are known to be associated with T2D,42 and we did not observe an association between PCSK9 inhibition and cataracts irrespective of adjustment for lipid medication status.
Cognition and Neurological Phenotypes
Prior studies have indicated that lowering LDL-C can alter the risk of epilepsy43 and cognitive dysfunction,26 but the evidence is weak and inconsistent. Similar concerns have been raised for PCSK9 inhibitors; the EBBINGHAUS clinical trial (Evaluating PCSK9 Binding Antibody Influence on Cognitive Health in High Cardiovascular Risk Subjects) found no evidence that the addition of evolocumab to statin therapy impacts incident changes in cognition.26,44 We found no association between rs11591147 and epilepsy or cognitive function measured using a trail making test. We also did not observe significant associations among cognitive function variables included in the phenome-wide analysis, including dementia and Parkinson disease diagnoses, but observed a nominal association with mood affective disorders (Table III in the Data Supplement).
We did not include peripheral artery disease as an independent phenotype in this analysis. However, TreeWAS analyses showed an association with a combined phenotype comprising diseases of the circulatory system. Although we did not find conclusive evidence for adverse effects in our phenome-wide search, we observed nominal associations with 19 variables in the discovery analysis (Table III in the Data Supplement). They were not statistically significant using our a priori defined thresholds; they need independent formal replication and should be considered as hypothesis generating. We observed that certain nominal associations had a direction of effect consistent with prior literature. A genetic study of LDL-lowering HMGCR variants showed an association with increased bodyweight and waist circumference.45 We observed a nominal increase in waist circumference and waist-hip ratio, along with a nominal decrease in maximum workload during a fitness test. There have been several studies showing a positive correlation between eosinophil counts and asthma. Eosinophilia—an increase in eosinophil counts—is a diagnostic feature of asthma,46 and sequence variants associated with eosinophil counts also show an association with asthma.47 We observed a nominal increase in eosinophil counts, self-reported asthma diagnosis, and self-reported history of wheezing. We also observed a nominal increase in heel bone mineral density, which has been discussed before in the context of circulating lipids and adiposity.48
Strengths and Limitations
The strengths of our study include a large study sample with consistent genotyping and collection of phenotypes, high statistical power, and a broad range of phenotypes across the whole phenome. Our study also has limitations. First, although our study sample was large, we acknowledge that we were not able to perform replication in an independent study sample. Most of our findings are consistent with prior literature and biological expectations, but some of the observations would benefit from replication in future studies. Second, our study was performed in middle-aged to elderly British individuals of European ancestry. We considered performing analyses also in individuals of black ancestry and South Asian origin from UKB, but unfortunately, there were <30 individuals in each of these populations who were carriers of the minor allele of rs11591147, making such analyses impossible because of low statistical power. Hence, the generalizability of our findings to other racial and ethnic groups is unknown.
PCSK9 inhibitors comprise a promising new category of lipid-lowering therapy for CVD. We provide genetic evidence that, in addition to coronary heart disease, reduced PCSK9 activity is likely to be protective against ischemic stroke, defined here using infarction and occlusion of relevant blood vessels. Confirmation of this protective effect in clinical trials is required before changing clinical practice and indications of PCSK9 antibody therapy. We did not find evidence of cognitive side effects of PCSK9 variants in the UKB; this result adds genetic support for the outcomes of the EBBINGHAUS trial. Finally, we observed an association of PCSK9 LoF variation with T2D among individuals on lipid-lowering medication.
We thank Stefan Gustafsson for assistance with data management. This research was conducted using the UK Biobank Resource under application number 16097.
Sources of Funding
This work was supported by a National Heart, Lung, and Blood Institute grant R01 HL135313-01 to Dr Ingelsson. Dr Knowles has received funding from the Doris Duke Foundation and the National Institutes of Health ClinGen study. Drs Ingelsson and Knowles have received funding from the Stanford Diabetes Research Center (award P30DK116074).
Dr Lindholm reports institutional research grants from AstraZeneca and GlaxoSmithKline and consultant fees to the institution from AstraZeneca (unrelated to the present work). The other authors report no conflicts.
Guest Editor for this article was Christopher Semsarian, MBBS, PhD, MPH.
The Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGEN.118.002162/-/DC1.
- Received January 9, 2018.
- Accepted May 21, 2018.
- © 2018 American Heart Association, Inc.
- Baigent C,
- et al
- Stone NJ
- Slimani A,
- et al
- Feng Q,
- et al
- Giugliano RP,
- et al
- Schmidt AF,
- et al
- Ingelsson E,
- et al
- Giugliano RP,
- et al
- Cortes A,
- et al
- Fisher M,
- et al
- Hopewell JC,
- et al
- Castilla-Guerra L,
- et al
- Fall T,
- et al
- Benn M,
- et al
- Gan W,
- et al