Coronary artery disease data are from 60 801 cases with coronary artery disease and 123 504 controls from the Coronary ARtery DIsease Genome wide Replication and Meta-analysis (CARDIoGRAM) plus the Coronary Artery Disease (C4D) Genetics (CARDIoGRAMplusC4D) Consortium.19 Type 2 diabetes data are from 50 775 cases of type 2 diabetes and 270 269 controls from European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct study,13 the UK Biobank study,14 and the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM).15 In addition to the EPIC-InterAct study,13 the UK Biobank study,14 and DIAGRAM,15 type 2 diabetes association analyses of rs12916 at HMGCR included 11 studies (4496 cases and 50 677 controls) previously reported by Swerdlow et al.5 Therefore, the sample size of HMGCR genetic variants association with type 2 diabetes was 55 271 cases of type 2 diabetes and 320 946 controls. All results are scaled to represent the odds ratio per 1-mmol/L (38.7-mg/dL) genetically predicted reduction in LDL-C.
The EPIC-InterAct Consortium
eTable 1. Participating studies
eTable 2. Genetic variants included in the main analysis
eTable 3. Sensitivity analyses at the NPC1L1 and PCSK9 loci
eTable 4. Correlation between genetic variants
eTable 5. Burden of rare alleles in exome sequencing studies
eFigure 1. Meta-analysis results
eFigure 2. Conditional analysis at the NPC1L1 locus
eFigure 3. Conditional analysis at the PCSK9 locus
eFigure 4. Associations of LDL-lowering alleles with continuous cardiometabolic traits
eFigure 5. Stratified associations of NPC1L1 variants
eFigure 6. Associations with continuous cardiometabolic traits
Customize your JAMA Network experience by selecting one or more topics from the list below.
Lotta LA, Sharp SJ, Burgess S, et al. Association Between Low-Density Lipoprotein Cholesterol–Lowering Genetic Variants and Risk of Type 2 DiabetesA Meta-analysis. JAMA. 2016;316(13):1383–1391. doi:10.1001/jama.2016.14568
Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Low-density lipoprotein cholesterol (LDL-C)–lowering alleles in or near NPC1L1 or HMGCR, encoding the respective molecular targets of ezetimibe and statins, have previously been used as proxies to study the efficacy of these lipid-lowering drugs. Alleles near HMGCR are associated with a higher risk of type 2 diabetes, similar to the increased incidence of new-onset diabetes associated with statin treatment in randomized clinical trials. It is unknown whether alleles near NPC1L1 are associated with the risk of type 2 diabetes.
To investigate whether LDL-C-lowering alleles in or near NPC1L1 and other genes encoding current or prospective molecular targets of lipid-lowering therapy (ie, HMGCR, PCSK9, ABCG5/G8, LDLR) are associated with the risk of type 2 diabetes.
Design, Setting, and Participants
The associations with type 2 diabetes and coronary artery disease of LDL-C-lowering genetic variants were investigated in meta-analyses of genetic association studies. Meta-analyses included 50 775 individuals with type 2 diabetes and 270 269 controls and 60 801 individuals with coronary artery disease and 123 504 controls. Data collection took place in Europe and the United States between 1991 and 2016.
Low-density lipoprotein cholesterol–lowering alleles in or near NPC1L1, HMGCR, PCSK9, ABCG5/G8, and LDLR.
Main Outcomes and Measures
Odds ratios (ORs) for type 2 diabetes and coronary artery disease.
Low-density lipoprotein cholesterol–lowering genetic variants at NPC1L1 were inversely associated with coronary artery disease (OR for a genetically predicted 1-mmol/L [38.7-mg/dL] reduction in LDL-C of 0.61 [95% CI, 0.42-0.88]; P = .008) and directly associated with type 2 diabetes (OR for a genetically predicted 1-mmol/L reduction in LDL-C of 2.42 [95% CI, 1.70-3.43]; P < .001). For PCSK9 genetic variants, the OR for type 2 diabetes per 1-mmol/L genetically predicted reduction in LDL-C was 1.19 (95% CI, 1.02-1.38; P = .03). For a given reduction in LDL-C, genetic variants were associated with a similar reduction in coronary artery disease risk (I2 = 0% for heterogeneity in genetic associations; P = .93). However, associations with type 2 diabetes were heterogeneous (I2 = 77.2%; P = .002), indicating gene-specific associations with metabolic risk of LDL-C-lowering alleles.
Conclusions and Relevance
In this meta-analysis, exposure to LDL-C-lowering genetic variants in or near NPC1L1 and other genes was associated with a higher risk of type 2 diabetes. These data provide insights into potential adverse effects of LDL-C-lowering therapy.
Treatment with statins, the pharmacological agents of choice for low-density lipoprotein cholesterol (LDL-C)–lowering therapy in cardiovascular prevention,1,2 is associated with weight gain and a higher incidence of new-onset type 2 diabetes.3-5 Ezetimibe, an inhibitor of the LDL-C transporter Niemann-Pick C1-Like 1 (NPC1L1),6,7 has been approved as a lipid-lowering agent, but it is unclear whether its use will be associated with an adverse metabolic risk profile.
There is considerable interest in predicting the efficacy and safety of therapeutic targets early in the drug development process. Drug targets with supporting human genetic evidence have been shown to have lower attrition rates during drug development.8 Variation in genes encoding drug targets has been used to predict both the efficacy and safety of pharmacological perturbation of those targets.9,10 In particular, LDL-C-lowering alleles in HMGCR,5,11 which encodes the molecular target of statins, have been successfully used as genetic proxies to study the effects of these drugs.5,11 Furthermore, LDL-C-lowering alleles in HMGCR are associated with higher risk of type 2 diabetes and higher body mass index (BMI) in genetic studies,5 similar to the safety profile of statins in meta-analyses of randomized clinical trials (RCTs).3-5
The efficacy of adding ezetimibe to simvastatin in secondary cardiovascular prevention was supported by the Improved Reduction of Outcomes: Vytorin Efficacy International Trial (IMPROVE-IT).6,7 Immediately before and after the publication of the trial results, studies were reported describing the use of genetic variants at NPC1L1 to predict the efficacy of NPC1L1 inhibition in the prevention of coronary events.11,12 The purpose of this study was to use naturally occurring LDL-C-lowering alleles at NPC1L1 to investigate the potential associations between NPC1L1 inhibition and the risk of type 2 diabetes. Alleles that lower LDL-C in or near genes encoding other current or prospective molecular targets of LDL-C-lowering therapy were studied.
Question Are low-density lipoprotein cholesterol (LDL-C)–lowering alleles at NPC1L1 or other genes associated with the risk of type 2 diabetes?
Findings In a meta-analysis of genetic association studies including 50 775 individuals with type 2 diabetes and 270 269 controls, LDL-C-lowering polymorphisms at NPC1L1 were associated with a statistically significant odds ratio of 2.42 for type 2 diabetes per genetically predicted reduction of 1 mmol/L (38.7 mg/dL) in LDL-C. Low-density lipoprotein cholesterol–lowering polymorphisms at HMGCR and PCSK9 were associated with a higher risk of diabetes.
Meaning These data provide insights into potential adverse effects of LDL-C-lowering therapy.
The association of LDL-C-lowering polymorphisms near NPC1L1 with the risk of type 2 diabetes was investigated in meta-analyses of genetic association studies. The associations of LDL-C-lowering alleles in or near genes encoding other current or prospective molecular targets of LDL-C-lowering therapy11 (ie, HMGCR, PCSK9, ABCG5/G8, LDLR) with type 2 diabetes, coronary artery disease, and continuous cardiometabolic traits were studied. A summary of the studies included in each analysis appears in eTable 1 in the Supplement.
The association of LDL-C-lowering alleles with type 2 diabetes was estimated in a meta-analysis of 50 775 individuals with type 2 diabetes and 270 269 controls from the European Prospective Investigation into Cancer and Nutrition (EPIC)–InterAct study (a case-cohort study nested within the EPIC study, which was a cohort study of 500 000 European participants followed-up for an average of 8 years),13 the UK Biobank study,14 and the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM).15 An additional 11 studies (4496 cases and 50 677 controls) previously reported by Swerdlow et al5 were included in the analyses of the association with type 2 diabetes of rs12916 in HMGCR (eFigure 1 in the Supplement). The combined association of NPC1L1 genetic variants in subgroups by age, sex, and BMI was analyzed in 14 657 unrelated cases of type 2 diabetes and 118 854 controls from the EPIC-InterAct study and the UK Biobank study with available individual-level genotyping data.
Eight of the 10 constituent EPIC cohorts agreed to take part in the EPIC-InterAct study, leaving 455 680 participants for screening. Individuals were excluded from EPIC-InterAct if they did not have stored blood (n = 109 625) or information on diabetes status (n = 5821; 1.3% of participants screened for inclusion). From the remaining 340 234 participants, 12 403 individuals who developed type 2 diabetes during follow-up constituted the incident case group of EPIC-InterAct and a random group of 16 154 individuals free of diabetes at baseline constituted the subcohort group of EPIC-InterAct.13 Subcohort participants were previously shown to be representative of eligible EPIC participants within each country.13 Data on 20 831 participants with available genotyping (with no overlap with DIAGRAM15) were included in the main analysis. Data on the 22 494 participants (including participants overlapping with DIAGRAM) with available genotyping were included in subgroup analyses. Type 2 diabetes status was available for all participants. Individuals without genotype data were excluded from the study. Data collection took place between 1991 and 2016. Participant characteristics and genotyping methods have been previously reported in detail13 and are summarized in Table 1 and eTable 2 in the Supplement.
The UK Biobank study is a population-based cohort of 500 000 people aged 40 to 69 years who were recruited from 2006 through 2010 from several centers across the United Kingdom.14 The association of genetic variants with prevalent type 2 diabetes was estimated in 6627 cases and 143 765 controls of the UK Biobank study data set who had available genotype data. Genotyping was attempted in 152 770 individuals and failed in only 480 instances (0.3%). Among a total of 152 290 participants with available genotype data, type 2 diabetes status was adjudicated in 150 392 (98.8%) participants. Type 2 diabetes was defined on the basis of self-reported physician diagnosis at nurse interview or digital questionnaire, age at diagnosis older than 36 years, and use of oral medications for diabetes. Data collection took place between 2006 and 2016. Participant characteristics and genotyping information appear in Table 1 and eTable 2 in the Supplement.
DIAGRAM is a research consortium that published the largest meta-analysis of genome-wide association studies for type 2 diabetes in individuals of European descent.15 Type 2 diabetes association results were made publicly available for up to 34 840 cases and 114 981 controls from 38 genetic association studies with a case-control or cohort design.15 Fifty percent of the participants were women and the average age was 55 years.15 Imputation was performed using the HapMap reference panel.15 Participant exclusion criteria encompassed duplicate samples, relatedness, mismatch between self-reported and genotype-determined sex, outlying heterozygosity and non-European descent. Type 2 diabetes status was available in all participants. Data collection took place between 2002 and 2012. Participant characteristics are reported in Table 1 and further characteristics of the studies included in the DIAGRAM meta-analysis were reported previously.15
The likelihood of bias for studies participating in this meta-analysis was deemed low on the basis of (1) the low proportion of participants with missing data on exposure or outcome, (2) the high-quality genotyping or imputation of genetic variants included in the study (eTable 2 in the Supplement), (3) the low likelihood of bias by case status in genotyping errors or genotype misclassification, (4) the consideration that if any nondifferential misclassification of exposure or outcome occurred, it would result in a bias toward the null, and (5) the consideration that genetic variants are less likely to be affected by confounding or reverse causality.16,17 On this basis, studies were deemed suitable for pooling by meta-analysis.
For the genetic variants included in these analyses, LDL-C association estimates were obtained from genetic association results in up to 188 577 participants of the Global Lipids Genetics Consortium.18 In addition to type 2 diabetes, the association of these LDL-C-lowering alleles with coronary artery disease and continuous cardiometabolic traits was estimated in large meta-analyses of genome-wide association studies. For coronary artery disease, data were from the Coronary ARtery DIsease Genome wide Replication and Meta-analysis (CARDIoGRAM) plus the Coronary Artery Disease (C4D) Genetics (CARDIoGRAMplusC4D) Consortium meta-analysis (60 801 cases and 123 504 controls).19 For glycemic traits, including fasting glucose20,21 (n = 133 010), glucose level 2 hours after an oral glucose challenge20,22 (n = 42 854) and fasting insulin levels20,21 (natural-logarithm transformed; n = 108 557), data were from the Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC).20-22 For anthropometric traits, including BMI (n = 333 495) and waist-to-hip ratio (n = 224 047), data were from the Genetic Investigation of ANthropometric Traits (GIANT) Consortium.23,24 Details appear in eTable 1 in the Supplement.
In exploratory analyses, the burden of protein-truncating and probably deleterious missense variants in NPC1L1, HMGCR, PCSK9, ABCG5, ABCG8 and LDLR was estimated from exome sequencing studies of 8373 cases with type 2 diabetes and 8466 controls (Type 2 Diabetes Multi-Ethnic Sequencing [T2D-GENES] Consortium, Genetics of Type 2 Diabetes [GoT2D] Consortium, DIAGRAM Consortium. May 26, 2016; http://www.type2diabetesgenetics.org/).
The combined association of 2 LDL-C-lowering genetic variants near NPC1L1 with type 2 diabetes constituted the primary analysis of the study (Table 2). These variants were identified as having distinct effects on LDL-C levels in approximate conditional analyses using the GCTA software25,26 (description appears below and in eFigure 2 in the Supplement). In sensitivity analyses, the combined association of 5 LDL-C-lowering alleles near NPC1L1 (previously used to predict the efficacy of ezetimibe11) was investigated (eTable 3 in the Supplement).
For comparison with NPC1L1, other LDL-C-lowering alleles in or near genes encoding other current or prospective molecular targets of LDL-C-lowering therapy (ie, HMGCR, PCSK9, ABCG5/G8, LDLR) were studied.11 Three LDL-C-lowering polymorphisms in or near HMGCR, previously demonstrated to mimic the efficacy and metabolic effects of statins,5,11 were analyzed (Table 2). At the ABCG5/G8 and LDLR loci, polymorphisms previously used to investigate genetic relationships between LDL-C and coronary artery disease11 were studied (Table 2). At the PCSK9 locus, in addition to the rs11591147 (p.R46L) variant (Table 2), the combined association of up to an additional 8 likely independent LDL-C-lowering polymorphisms was investigated (eFigure 3 in the Supplement). Genetic variants included in the analyses were strongly and specifically associated with LDL-C (eFigure 4 in the Supplement).
Approximate conditional analyses on large-scale LDL-C association data from the Global Lipids Genetics Consortium18 using the GCTA software25,26 were performed to identify distinct association signals for LDL-C at the NPC1L1 and PCSK9 loci. This approach uses genetic association results in addition to the linkage disequilibrium pattern in a reference population to estimate the association of genetic variants in a region after accounting for 1 or more index genetic variants. In doing so, the software allows for the identification of likely independent association signals in a given region using result-level data. At the PCSK9 locus, in a smaller sample of individuals with individual-level genotypes, formal conditional analyses of the association with LDL-C of polymorphisms after adjusting for rs11591147 genotype status were conducted (eFigure 3 in the Supplement).
The HUGO Gene Nomenclature Committee27 (http://www.genenames.org) gene names of the investigated genes were HGNC:7898 (NPC1L1), HGNC:5006 (HMGCR), HGNC:20001 (PCSK9), HGNC:13886 (ABCG5), HGNC:13887 (ABCG8), and HGNC:6547 (LDLR). Genomic coordinates reported in this article represent the chromosome and physical position of genetic variants according to the Human Reference Genome Build 37 (http://www.ncbi.nlm.nih.gov/projects/genome/assembly/grc/). Polymorphism names reported in the manuscript represent rsID entries from dbSNP release 147 (http://www.ncbi.nlm.nih.gov/SNP/).
Genetic association data for the meta-analyses were either generated or gathered from available sources at the MRC Epidemiology Unit, University of Cambridge. For each genetic variant and outcome, inverse variance–weighted meta-analysis using fixed-effect models was used to obtain pooled estimates. The I2 statistic was used to quantify heterogeneity. For each gene, associations of LDL-C-lowering genetic variants and outcomes were estimated using Mendelian randomization statistical methods.17 Estimates of genetic variant to LDL-C and genetic variant to outcome associations were used to calculate estimates for the LDL-C reduction to outcome association at each gene.17 When multiple genetic variants at a given gene were included in the model, estimates were pooled with a weighted generalized linear regression method that accounts for the correlation between genetic variants.17 The correlation values were obtained from the SNAP software28 or from the 1000 Genomes Project phase 3 data on individuals of European ancestry (http://browser.1000genomes.org/; eTable 4 in the Supplement). Results were scaled to represent the odds ratio (OR) per 1-mmol/L (38.7-mg/dL) for the genetically predicted reduction in LDL-C. Absolute risk differences were estimated assuming that the incidence rate of type 2 diabetes in the EPIC-InterAct study subcohort would be the baseline incidence rate in unexposed individuals (ie, 3.76 incident cases per 1000 person-years of follow-up).13 This baseline rate was then multiplied by the OR estimated from genetic analyses to obtain the at-risk incidence rate. The absolute risk difference estimate was the at-risk incidence rate minus the baseline incidence rate. Absolute risk differences were expressed in incident events per 1000 person-years for a 1-mmol/L genetically predicted reduction in LDL-C. Statistical analyses were conducted using Stata version 14.1 (StataCorp), R version 3.2.2 (R Foundation for Statistical Computing), and METAL version 2011-03-25.29 A 2-tailed P < .05 was considered statistically significant.
Alleles that lower LDL-C at the NPC1L1 locus were inversely associated with coronary artery disease (OR for a genetically predicted 1-mmol/L [38.7-mg/dL] reduction in LDL-C of 0.61 [95% CI, 0.42-0.88]; P = .008) and directly associated with type 2 diabetes both individually (Table 2) and collectively (OR for a genetically predicted 1-mmol/L reduction in LDL-C of 2.42 [95% CI, 1.70-3.43], P < .001; estimated absolute risk difference, 5.3 incident cases per 1000 person-years for a 1-mmol/L genetically predicted reduction in LDL-C; Figure). For both polymorphisms, estimates of the association with type 2 diabetes were consistent across the studies included in the meta-analysis (eFigure 1 in the Supplement). In the periphery of the NPC1L1 locus, approximately 400 kilobases from the lead rs2073547 polymorphism, there was a known association signal for type 2 diabetes and glycemic traits near the GCK gene.15,20,21 After accounting for variation in GCK, the association with type 2 diabetes at NPC1L1 did not change (eTable 3 in the Supplement). Association estimates remained unchanged when modeling the association of 5 polymorphisms previously used by Ference et al11 as a proxy for NPC1L1 inhibition (eTable 3). In 14 657 cases of type 2 diabetes and 118 854 controls for whom we had access to individual-level genotyping data, there was no evidence of heterogeneity in the association between NPC1L1 alleles and type 2 diabetes in analyses stratified by age, sex, or BMI (eFigure 5 in the Supplement). In exome-sequencing association results, there was no evidence of enrichment with NPC1L1 protein truncating alleles in cases with type 2 diabetes compared with controls (OR of 1.12 [95% CI, 0.88-1.43] for type 2 diabetes among individuals carrying a truncating allele [P = .34]), but missense variants in NPC1L1 predicted to be probably deleterious were overrepresented in individuals with type 2 diabetes compared with controls (OR, 1.26 [95% CI, 1.07-1.47]; P = .005).
As previously reported,5,11 LDL-C lowering alleles at HMGCR were associated with type 2 diabetes and coronary artery disease in opposite directions (Table 2 and Figure). An association of the loss-of-function p.R46L (rs11591147) variant in PCSK9 with higher risk of type 2 diabetes was found (OR of 1.19 [95% CI, 1.02-1.38] for type 2 diabetes per 1-mmol/L genetically predicted LDL-C reduction [P = .03]; estimated absolute risk difference, 0.7 incident cases per 1000 person-years for a 1-mmol/L genetically predicted reduction in LDL-C). At PCSK9, analyses of the LDL-C association data suggested the presence of distinct association signals. In formal conditional analyses, there was evidence of at least 2 distinct association signals (rs11591147 and rs471705; eFigure 3). Using the GCTA software,25,26 approximate conditional analyses suggested the presence of 9 distinct association signals (rs11591147 plus 8 additional genetic variants; eFigure 3 in the Supplement). Inclusion of these additional signals gave similar associations with type 2 diabetes as the p.R46L variant alone (OR of 1.21 [95% CI, 1.04-1.41] for type 2 diabetes per 1-mmol/L genetically predicted reduction in LDL-C using rs11591147 plus rs471705 [P = .01] and OR of 1.16 [95% CI, 1.03-1.31] using rs11591147 plus the 8 additional polymorphisms [P = .02]; eTable 3 in the Supplement). The association with type 2 diabetes of LDL-C-lowering alleles at the ABCG5/G8 and LDLR loci did not reach statistical significance. There was no evidence of association with type 2 diabetes for missense variants predicted to be probably deleterious or protein truncating alleles in the HMGCR, PCSK9, ABCG5, ABCG8 and LDLR genes (eTable 5 in the Supplement), but the 95% CIs around the risk estimates were generally wide, reflecting the low prevalence of these genetic variants and the relatively small sample size of this analysis.
In analyses of the association with disease risk of a given genetically predicted reduction in LDL-C, there was a similar reduction in coronary artery disease risk across genes (I2 for heterogeneity in genetic associations = 0%, P = .93). However, for the same reduction in LDL-C, the association with type 2 diabetes risk differed by gene (I2 = 77.2%, P = .002). The different magnitudes and directions of association of LDL-C-lowering alleles with continuous glycemic and anthropometric traits suggested gene-specific mechanisms underlying the altered risk of type 2 diabetes (eFigure 6 in the Supplement). For example, at the HMGCR locus there were associations with BMI and waist-to-hip ratio, whereas at the PCSK9 locus there were associations with higher fasting glucose levels and 2-hour glucose levels (eFigure 6).
In this meta-analysis, exposure to LDL-C-lowering genetic variants in or near the NPC1L1 gene was associated with a higher risk of type 2 diabetes. This finding is consistent with the results of a small-scale open-label RCT showing increased glycated hemoglobin in association with ezetimibe treatment.30 Blazing et al31 reported that the addition of ezetimibe to simvastatin for secondary cardiovascular prevention in IMPROVE-IT resulted in a small and not statistically significant increase in risk of new-onset diabetes (9% relative risk increase per 1-mmol/L [38.7-mg/dL] reduction in LDL-C). However, the IMPROVE-IT results may not be sufficient to rule out an effect of inhibiting NPC1L1 on diabetes risk because (1) some of the effects of NPC1L1 inhibition may be apparent only after several years of treatment; (2) the risk of type 2 diabetes in individuals with a history of acute coronary syndrome yet free from type 2 diabetes in IMPROVE-IT may not reflect that of the general population on which this genetic analysis is based; (3) limited compliance to drug treatment, as observed in IMPROVE-IT,7 may dilute etiological effect estimates. By analogy, the association of statin treatment with higher risk of diabetes was only demonstrable in a meta-analysis of several RCTs including more than 90 000 individuals.3 Therefore, these results warrant the continued monitoring of the glycemic effects of ezetimibe in RCTs and clinical practice, particularly in a primary prevention setting.
The results of this study show that multiple LDL-C-lowering mechanisms, including those mediated by the molecular targets of available LDL-C-lowering drugs (ie, statins, ezetimibe, and proprotein convertase subtilisin/kexin type 9 [PCSK9] inhibitors), are associated with adverse metabolic consequences and increased type 2 diabetes risk. These findings are consistent with other studies of the association with type 2 diabetes of genetic scores aggregating multiple polymorphisms affecting LDL-C and other lipid fractions.32 They are consistent with the observation that patients with familial hypercholesterolemia are less likely to have type 2 diabetes.33 The genes that were associated both with lower LDL-C levels and higher risk of type 2 diabetes have an effect on LDL-C level by distinct pathways including cholesterol absorption (NPC1L1),34 endogenous cholesterol synthesis (HMGCR),35 and internalization of cholesterol-rich particles into the cell (PCSK9).36,37 For a similar reduction in LDL-C, the association with type 2 diabetes differed by gene, which would be consistent with the mediation of their associations by different mechanisms. Besseling et al33 have proposed that an increased internalization of cholesterol into pancreatic beta cells may result in impaired secretion of insulin, a hypothesis supported by murine experimental models.38 Alleles that lower LDL-C at HMGCR are associated with higher levels of fasting insulin and BMI, suggesting an insulin resistance-related mechanism.5 In contrast with early evidence showing metabolic benefits of NPC1L1 knockout in mice,39 recent studies suggest that its overexpression in the liver may suppress gluconeogenesis and, therefore, that its inhibition could perhaps enhance glucose production.40 Overall, these results indicate complex relationships between the mechanisms that lead to lower LDL-C and metabolic risk.
Contrary to previous, smaller-scale investigations,41 there were associations of the p.R46L variant in PCSK9 (rs11591147) with a higher risk of type 2 diabetes, and higher fasting and 2-hour glucose. These associations have to be interpreted with caution, given the level of statistical significance for the association and the context of multiple comparisons presented in this study. This finding suggests that an effect of LDL-C-lowering drugs on increased risk of diabetes might extend to the newly developed PCSK9 inhibitors, encouraging further genetic and clinical trial investigations.
In general, unlike the association of LDL-C-lowering alleles with cardiovascular risk, the association of these alleles with metabolic risk appears to be specific to particular genes, which in turn might suggest that the adverse consequences of lipid-lowering agents on diabetes risk could be specific to a particular drug target. This may have clinical implications for the future of lipid-lowering therapy in the context of the increasing number of approved drugs acting on different molecular targets. The overall safety profile of these drugs, including the magnitude of risk of new-onset type 2 diabetes, may be relevant to the choice of specific agent for subsets of the patient population (eg, those at high risk for type 2 diabetes who are candidates for lipid-lowering therapy).
A number of assumptions and potential limitations of the genetic approach used in this study should be considered. Mendelian randomization generally assumes that genetic variants are associated with the end point exclusively via the risk factor of interest.16,17 The strong and specific association with LDL-C, the well-known role of target genes in LDL-C metabolism and the use of conditionally distinct genetic variants at given loci strengthen the validity of the genetic models used in this study. Similar to previous examples,5,11,42 the aim of this study was to use genetic variants that mimic the action of pharmacological therapy and therefore pleiotropy (ie, the association with variables other than LDL-C) may be more informative than concerning. For instance, HMGCR genetic variants are associated with higher BMI, consistent with the effects on body weight observed in RCTs of statins.5 However, the consequences of modest reductions in LDL-C associated with LDL-C-lowering alleles over several decades, as assessed in this study, may differ from the short-term pharmacological inhibition of a molecular target in RCTs or clinical practice. In addition, several of the included studies were population-based and therefore association estimates from these studies may not be applicable to patient groups in whom a particular therapy is indicated.
In this meta-analysis, exposure to LDL-C-lowering genetic variants in or near NPC1L1 and other genes was associated with a higher risk of type 2 diabetes. These data provide insights into potential adverse effects of LDL-C-lowering therapy.
Corresponding Author: Nicholas J. Wareham, MBBS, PhD, MRC Epidemiology Unit, University of Cambridge, Cambridge CB20QQ, United Kingdom (firstname.lastname@example.org).
Author Contributions: Dr Lotta had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Scott and Wareham contributed equally to this article.
Concept and design: Lotta, Burgess, Perry, Ardanaz, Franks, Overvad, Riboli, Tjonneland, Tumino, van der A, van der Schouw, Langenberg, Scott, Wareham.
Acquisition, analysis, or interpretation of data: Lotta, Sharp, Perry, Stewart, Willems, Luan, Ardanaz, Arriola, Balkau, Boeing, Deloukas, Forouhi, Franks, Grioni, Kaaks, Key, Navarro, Nilsson, Overvad, Palli, Panico, Quirós, Riboli, Rolandsson, Sacerdote, Salamanca-Fernandez, Slimani, Spijkerman, McCarthy, Barroso, O'Rahilly, Savage, Sattar, Langenberg, Scott, Wareham.
Drafting of the manuscript: Lotta, Scott, Wareham.
Critical revision of the manuscript for important intellectual content: Lotta, Sharp, Burgess, Perry, Stewart, Willems, Luan, Ardanaz, Arriola, Balkau, Boeing, Deloukas, Forouhi, Franks, Grioni, Kaaks, Key, Navarro, Nilsson, Overvad, Palli, Panico, Quirós, Riboli, Rolandsson, Sacerdote, Salamanca-Fernandez, Slimani, Spijkerman, Tjonneland, Tumino, van der A, van der Schouw, McCarthy, Barroso, O'Rahilly, Savage, Sattar, Langenberg, Scott, Wareham.
Statistical analysis: Lotta, Sharp, Burgess, Perry, Stewart, Willems, Luan, Langenberg, Scott.
Obtained funding: Ardanaz, Tjonneland.
Administrative, technical, or material support: Ardanaz, Arriola, Deloukas, Forouhi, Franks, Grioni, Kaaks, Overvad, Quirós, Rolandsson, Sacerdote, Slimani, Spijkerman, van der Schouw, Barroso.
Study supervision: Ardanaz, Deloukas, Forouhi, Riboli, Salamanca-Fernandez, Tumino, van der Schouw, Sattar, Langenberg, Scott, Wareham.
Conflict of Interest Disclosures: The authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr McCarthy reported receiving grants from Eli Lilly, Roche, AstraZeneca, Merck, Janssen, Servier, Novo Nordisk, Sanofi-Aventis, Boehringer Ingelheim, Pfizer, and Takeda; and honoraria from Novo Nordisk and Pfzier. Dr Barroso reported receiving grants from Wellcome Trust; and owning stock in GlaxoSmithKline and Incyte. Dr O'Rahilly reported receiving personal fees from Pfizer, AstraZeneca, iMed, and ERX Pharmaceuticals for serving on advisory boards and scientific panels. Dr Sattar reported receiving personal fees from Amgen and Sanofi for serving on advisory boards and from Merck for giving a speech. No other disclosures were reported.
Funding/Support: Funding for the MRC Epidemiology Unit was provided by the United Kingdom’s Medical Research Council through grants MC_UU_12015/1, MC_PC_13046, MC_PC_13048 and MR/L00002/1. We acknowledge support from the National Institute of Health Research Biomedical Research Centre. Funding for the EPIC-InterAct Study was provided by the EU FP6 program grant LSHM_CT_2006_037197. Dr Burgess is supported by a postdoctoral fellowship 100114 from the Wellcome Trust. Dr McCarthy is a Wellcome Trust senior investigator and was supported by grants 090532 and 098381 from the Wellcome Trust. Dr Barroso was supported by grant WT098051 from the Wellcome Trust. Dr Savage was supported by the Wellcome Trust grant 107064.
Role of the Funder/Sponsor: The funders/sponsors had no role in the design or conduct of the study; collection, management, analysis or interpretation of the data; preparation, review or approval of the manuscript or the decision to submit the manuscript for publication.
Additional Contributions: A list of the members of the EPIC-InterAct Consortium appears in the Supplement. We gratefully acknowledge the help of the MRC Epidemiology Unit Support Teams, including the Field Teams, the Laboratory Team, and the Data Management Team. This research was conducted using the UK Biobank resource.