Genetic variants in CETP increase risk of intracerebral hemorrhage

Objective In observational epidemiologic studies, higher plasma high‐density lipoprotein cholesterol (HDL‐C) has been associated with increased risk of intracerebral hemorrhage (ICH). DNA sequence variants that decrease cholesteryl ester transfer protein (CETP) gene activity increase plasma HDL‐C; as such, medicines that inhibit CETP and raise HDL‐C are in clinical development. Here, we test the hypothesis that CETP DNA sequence variants associated with higher HDL‐C also increase risk for ICH. Methods We performed 2 candidate‐gene analyses of CETP. First, we tested individual CETP variants in a discovery cohort of 1,149 ICH cases and 1,238 controls from 3 studies, followed by replication in 1,625 cases and 1,845 controls from 5 studies. Second, we constructed a genetic risk score comprised of 7 independent variants at the CETP locus and tested this score for association with HDL‐C as well as ICH risk. Results Twelve variants within CETP demonstrated nominal association with ICH, with the strongest association at the rs173539 locus (odds ratio [OR] = 1.25, standard error [SE] = 0.06, p = 6.0 × 10−4) with no heterogeneity across studies (I 2 = 0%). This association was replicated in patients of European ancestry (p = 0.03). A genetic score of CETP variants found to increase HDL‐C by ∼2.85mg/dl in the Global Lipids Genetics Consortium was strongly associated with ICH risk (OR = 1.86, SE = 0.13, p = 1.39 × 10−6). Interpretation Genetic variants in CETP associated with increased HDL‐C raise the risk of ICH. Given ongoing therapeutic development in CETP inhibition and other HDL‐raising strategies, further exploration of potential adverse cerebrovascular outcomes may be warranted. Ann Neurol 2016;80:730–740

Objective: In observational epidemiologic studies, higher plasma high-density lipoprotein cholesterol (HDL-C) has been associated with increased risk of intracerebral hemorrhage (ICH). DNA sequence variants that decrease cholesteryl ester transfer protein (CETP) gene activity increase plasma HDL-C; as such, medicines that inhibit CETP and raise HDL-C are in clinical development. Here, we test the hypothesis that CETP DNA sequence variants associated with higher HDL-C also increase risk for ICH. Methods: We performed 2 candidate-gene analyses of CETP. First, we tested individual CETP variants in a discovery cohort of 1,149 ICH cases and 1,238 controls from 3 studies, followed by replication in 1,625 cases and 1,845 controls from 5 studies. Second, we constructed a genetic risk score comprised of 7 independent variants at the CETP locus and tested this score for association with HDL-C as well as ICH risk. Results: Twelve variants within CETP demonstrated nominal association with ICH, with the strongest association at the rs173539 locus (odds ratio [OR] 5 1.25, standard error [SE] 5 0.06, p 5 6.0 3 10 24 ) with no heterogeneity across studies (I 2 5 0%). This association was replicated in patients of European ancestry (p 5 0.03). A genetic score of CETP variants found to increase HDL-C by 2.85mg/dl in the Global Lipids Genetics Consortium was strongly associated with ICH risk (OR 5 1.86, SE 5 0.13, p 5 1.39 3 10 26 ). Interpretation: Genetic variants in CETP associated with increased HDL-C raise the risk of ICH. Given ongoing therapeutic development in CETP inhibition and other HDL-raising strategies, further exploration of potential adverse cerebrovascular outcomes may be warranted. ANN NEUROL 2016;80:730-740 S erum levels of high-density lipoprotein cholesterol (HDL-C) are strongly and inversely associated with coronary artery disease (CAD) risk. 1 Of the many single nucleotide polymorphisms (SNPs) associated with HDL-C levels, those within cholesteryl ester transfer protein (CETP) have the strongest effect. [2][3][4] Inhibitory variants within CETP associated with increased HDL-C correlate with reduced risk of multiple cardiac risk factors, including metabolic syndrome and myocardial infarction. [5][6][7][8] Inhibitors of the CETP gene product, designed to raise HDL-C by limiting CETP-mediated exchange of cholesteryl esters and triglycerides between HDL and lowdensity lipoprotein (LDL)/very low-density lipoprotein particles, are being investigated in ongoing phase III trials as treatments to reduce CAD risk. 9,10 In contrast, substantial data suggest that elevations in HDL-C may increase risk of spontaneous intracerebral hemorrhage (ICH). 11,12 Furthermore, clinical trial data suggest an increased risk of ICH on statins despite a lack of significant differences in lipid levels. 13,14 Because of small sample sizes and confounding by environmental or medical exposures, prior studies have not been able to resolve this potentially paradoxical role of elevated HDL-C in ICH. Although ICH comprises only 15 to 20% of all strokes, it accounts for 50% of all stroke-related mortality and 30% of total costs. 15,16 Blood pressure control remains the only available preventive strategy. 17 As HDL-C evolves as a cardiovascular treatment target and clinical trial data on therapeutic modifiers accrue, an improved mechanistic understanding of the pathways involved in hemorrhagic cerebrovascular disease could lead to alternative treatments and prevention strategies for ICH.
It is not known whether CETP inhibitors, which endeavor to produce a biological effect similar to known genetic variants in CETP, increase ICH risk. The objective of this study was to use genome-wide genotypes from individuals with and without ICH from the International Stroke Genetics Consortium (ISGC) to test genetic variants within CETP for association with ICH risk, under the hypothesis that the HDL-raising effects of inhibitory variants within CETP will result in increased ICH. CETP genetic variants that impact HDL-C are unconfounded by other exposures, remain constant throughout life, and may be more reflective of long-term levels than periodic lipid measurements. 18 Thus, examination of CETP genetic variation constitutes a valuable causal inference tool to help strengthen or disclaim prior observations of association between elevated HDL-C and ICH, and could provide additional clues about potential adverse effects of pharmacologic CETP inhibition.

Study Design
We performed a 2-stage (discovery and replication) case-control candidate-gene association study using both genome-wide data and direct genotyping. The discovery phase utilized data from 3 genome-wide association studies (GWASs) of ICH, sampling patients of European ancestry (Table 1). 19 Replication involved direct genotyping of variants of interest from individuals recruited through 5 case-control studies of ICH, with no overlap between individuals from the discovery phase (Table 2). Detailed description of discovery and replication case and control recruitment architectures can be found in Supplementary  Table 1.
All studies had approval from the local institutional review board or ethics committee at each participating institution. Informed consent was obtained from all patients or their legally authorized representatives, or was waived via protocolspecific allowance.

Cases
ICH was defined as a new and acute neurological deficit with compatible brain imaging. Enrolled patients were adult consenting primary acute ICH cases that presented to participating institutions with confirmation of primary ICH through computed tomography or magnetic resonance imaging. Exclusion criteria included trauma, brain tumor, hemorrhagic transformation of a  19 Replication cases were recruited from ISGC participating centers using similar criteria as discovery cases (Supplementary  Table 2). Briefly, the University Medical Center (UMC) Utrecht ICH study included additional screening for secondary ICH cases in follow-up. The Edinburgh Stroke Study recruited subjects aged >55 years only, and specifically excluded individuals with antecedent illicit drug use or presentation >1 week from onset of symptoms. The LINCHPIN study identified ICH cases aged >16 years with acute or chronic ICH from a prospective cohort of individuals living in the Lothian region of Scotland, United Kingdom.
NEUROIMAGING. Stroke neurologists and neuroradiologists at each participating site performed the neuroimaging assessment. Following known differences in underlying biology, ICH was classified as lobar or nonlobar according to location. 20 ICH originating in the corticosubcortical junction (with or without involvement of subcortical white matter) was defined as lobar, whereas ICH selectively involving the thalamus, internal capsule, basal ganglia, brainstem, or cerebellum was defined as nonlobar.

Controls
Controls were ICH-free individuals >18 years of age and were enrolled from the same populations that gave rise to the cases. Controls were confirmed to have no history of previous ICH by interview and/or medical record review. Control population age restrictions were identical to case population age restrictions for all included studies.
CONTROL POPULATIONS. ICH-free controls were recruited from the same populations that gave rise to the ICH cases, through inpatient recruitment, ambulatory centers in the local communities, blood donation centers serving the same population, and in the case of the Lothian Birth Cohort, a population cohort study (Supplementary Table 3). The Genetic and Environmental Risk Factors for Hemorrhagic Stroke (GERFHS) and Ethic/Racial Variations of Intracerebral Hemorrhage (ERICH) studies 19,21 used random digit dialing, the Lothian Birth Cohort individuals were matched to case samples by local investigators, 22 and UMC Utrecht identified controls from the local blood donor population. The remainder of the studies used random selection from ambulatory clinics or geographically matched populations where cases were being recruited.

Exposure: Common Genetic Variants within CETP
In the discovery phase, we ascertained variants within CETP by means of genome-wide genotyping followed by imputation using methods and quality control procedures previously described. 19 Briefly, DNA was isolated from fresh or frozen peripheral whole blood collected from cases and controls at each participating institution at the time of consent, quantified with a quantification kit (Qiagen, Valencia, CA), and normalized to a concentration of 30ng/ll. Cases and controls were plated together and genotyped on Illumina (San Diego, CA) 610 or Affymetrix (Santa Clara, CA) 6.0 platforms. Standard quality controls for genome-wide data were applied, and the  (10) 69 (11) 69 (14) 68 (13) 69 (13) 63 (14) 62 (13) 56 (11) 75 (9)  resulting set of individuals and SNPs were carried forward to imputation, which was completed using IMPUTE2 with 1000 Genomes-based reference panels (March 2012 version). 23 Postimputation exclusion filters were minor allele frequency (MAF) < 0.01 and information score < 0.5. SNPs were extracted from the CETP gene region according to the human genome reference GRCh38.p2 annotated location (http://www. ncbi.nlm.nih.gov), 6 50 kilobases.

Independent Replication
CETP variants exceeding Bonferroni-corrected significance and without significant heterogeneity (I 2 < 40%) for association with ICH in the discovery phase were selected for replication. 24 Replication SNPs were chosen based on proxy status with index SNPs. Because replication of CETP variants was carried out as part of an ongoing GWAS of ICH, a constraint for the selection of replication SNPs was predicted genotyping success using iPLEX (Sequenom, San Diego, CA) and Taqman (Applied Biosystems, Foster City, CA) methodologies, which were employed at the Massachusetts General Hospital and University of Miami genotyping centers, respectively (see Table 2). Ancestryinformative markers were also genotyped to facilitate adjustment for population admixture.

Data Analysis
We present discrete variables as counts (percentage) and continuous variables as mean (standard deviation [SD]) or median (interquartile range), as appropriate.
POPULATION STRUCTURE. Principal component analysis was implemented in both discovery and replication to account for population structure, using genome-wide data in discovery and prespecified ancestry-informative markers in replication. 25,26 Caucasian population outliers were identified and removed by visual inspection of plots generated with principal components 1 and 2, and these principal components were included as covariates in regression models fitted for association testing. In the GERFHS and ERICH samples, further refinement of population structure was achieved using the ADMIX-TURE software tool to remove outliers. 27 ASSOCIATION TESTING. Prior to discovery association testing, SNPs within CETP were clumped into loci sharing linkage disequilibrium (LD) r 2 > 0.5 using PLINK to allow discrimination of semi-independent loci across the gene. Association testing for SNPs within the CETP locus and ICH risk was completed separately for all ICH, as well as for lobar and nonlobar hemorrhages. Logistic regression models were fitted assuming independent additive genetic effects for dosage of the minor allele (1 degree of freedom additive trend test), and adjusted for age, gender, and principal components 1 and 2. A similar analytic approach was undertaken for analysis of replication data, using additive allele genotype data rather than dosage.
META-ANALYSIS. Fixed effects, inverse variance weighted meta-analysis was used to pool effect estimates across studies, assessing heterogeneity by computing Cochrane's Q (with corresponding p) and I 2 (percentage of effect size attributable to heterogeneity). Identical meta-analysis procedures were used for pooling of effects across studies in discovery and replication, and across all studies. 28

Genetic Risk Score Analysis
Variants within the CETP locus with established association with HDL-C levels in the most recent Global Lipids Genetics Consortium (GLGC) analysis (unpublished data) were extracted from the discovery data set and tested for association with ICH using an additive multi-SNP genetic risk score approach using the GTX package (http://CRAN.R-project.org/package5gtx) in R (version 3.0). Ten variants surpassing exome array-wide significance (p < 2.1 3 10 27 ) and demonstrating independence using a sequential forward selection model in the GLGC data set were identified, of which 7 were available in our ICH discovery data set. 29 These 7 variants, on average, were associated with a 0.19-SD increase in HDL-C (2.85mg/dl) in the GLGC population (p < 1 3 10 2200 ). This corresponds to a proportion of variance explained of 0.032. ICH risk was predicted from summary statistics, weighted according to the established HDL-C effect, and oriented to the HDL-C increasing allele.

Statistical Testing and Software
We used a conservative Bonferroni-corrected threshold for statistical significance of p < 0.004, adjusted for the number of semiindependent loci within CETP with r 2 < 0.5 (12 tests in this analysis). Quality control procedures, genetic association testing for single variants, and score calculations were performed in SNPTest and PLINK v1.07. 26,30 Imputation was completed using IMPUTE2. 23 All other statistical analyses were performed in SAS 9.2 (SAS Institute, Cary, NC).

Results
Following relevant exclusions during quality control and principal component analysis, 1,149 ICH cases and 1,238 controls from 3 case-control studies of ICH were included in the discovery phase, 43% of which were of the lobar ICH subtype (see Table 1).

CETP Genetic Variants
After imputation using 1000 Genomes reference panels and application of genome-wide quality control filters, a total of 390 common variants of MAF > 0.01 were extracted from the CETP gene and 50kb flanking regions (Supplementary Table 4). 31 These 390 variants were present either via array-based ascertainment or imputation in all 3 of the discovery data sets, and were used for association testing.

Single-SNP Association Testing
After testing all 390 SNPs within CETP clumped into regions sharing r 2 > 0.5, 12 loci demonstrating nominal association with ICH (p < 0.05) were identified (Supplementary Table 5). Three of these loci surpassed Bonferroni correction ( ) met prespecified criteria for replication due to its homogeneity across discovery data sets (I 2 5 0%). Of note, rs173539 was in high LD with rs3764261 (r 2 5 0.98), the strongest associated SNP with HDL-C in published GWASs of lipid levels (Fig 1). 32 Comparison of effects of the rs173539 locus on risk of lobar versus nonlobar hemorrhage revealed no significant differences by ICH subtype (Supplementary Table 6).

Replication and Meta-Analysis of the rs173539 Locus
A total of 1,625 ICH cases and 1,845 controls of Caucasian ancestry were available for replication. Following application of predictive algorithms for SNP genotype ascertainment success using both genotyping methodologies employed, 4 SNPs in LD with rs173539 locus were selected for replication genotyping according to the constraints described (Tables 4 and 5). Both rs173539 and rs3764261 were predicted to fail in one or both replication pools. All 4 selected SNPs were successfully genotyped in all replication data sets. All replication results showed minimal heterogeneity and consistent directions of effect, and 2 variants replicated at p < 0.05. In metaanalysis, all 4 SNPs within the rs173539 locus chosen for replication were strengthened by addition of the replication SNP data, with minimal heterogeneity in the final total sample size of 2,595 ICH cases and 3,030 controls (see Table 5).

Genetic Risk Score Analysis
An additive multi-SNP genetic risk score was constructed using independent HDL-association data. 29 Ten variants were selected, of which 7 were present in the ICH discovery data set (Table 6). Three variants were unavailable in the ICH data set due to differences in genotyping platforms (exome array vs GWAS array) between the two studies. The genetic risk score of these 7 variants demonstrated association with ICH (OR 5 1.86, SE 5 0.13, p 5 1.39 3 10 26 ).

Discussion
Our results demonstrate an association between CETP gene variants in the rs173539 locus and risk of ICH, opposite in direction from their effect on risk of CAD and metabolic syndrome. 5,7,8 Furthermore, an aggregated score of variants within CETP that raise HDL-C is  strongly associated with increased ICH risk. These results suggest that there may be substantial differences in the roles of lipids in the progression of cerebrovascular and cardiometabolic diseases. Novel therapies targeting CETP along with other approaches to increase HDL-C are currently under active investigation in an effort to reduce the risk of CAD. 33 Because the cerebral small vessel diseases that lead to ICH are common in the aging population and frequently coincide with risk factors for cardiometabolic disease, 34,35 our observations supporting opposing effects of HDL-C on ICH and CAD underscore the need for a better understanding of which patients could be at increased risk of ICH on therapies aimed at increasing HDL-C. Our findings support prior studies linking elevated HDL-C with increased risk of ICH. Unlike prior studies, however, our genetic approach limits confounding by dietary, environmental, or medication exposures. A recent meta-analysis of epidemiological studies examining associations between cholesterol levels and ICH found a dose-response relationship between HDL-C and ICH risk, with each 1mmol/l increase in HDL-C associated with a 17% increase in ICH risk. 11 This result was nullified when studies of subarachnoid hemorrhage patients were included, but strengthened by restriction to studies from the United States, highlighting the potential confounds of case misspecification and unmeasured environmental exposures in testing associations of this nature.
HDL-C appears to have a complex and contextdependent role in cerebrovascular disease. In contrast to ICH, elevated HDL-C is associated with reduced risk of ischemic stroke, particularly strokes caused by large artery atherosclerotic disease, consistent with the observed associations of HDL-C in CAD. 36 However, Mendelian randomization (MR) studies of genetic variants predisposing to elevated HDL-C have not demonstrated association with either ischemic stroke or CAD, suggesting the observed relationships may not be causal. 37,38 Unfortunately, the limited sample size of genetics efforts in ICH coupled with acute changes in lipid values around the Association results for rs173539 in association with ICH risk, as well as 4 additional SNPs in LD with rs173539 chosen for replication.  time of ICH currently preclude the use of this MR approach in our analyses. 39 No study, including the present, has yet established a direct causal relationship between HDL-C and ICH risk. Although associations between CETP genetic variants and ICH are almost certainly unidirectional due to the immutability of the genetic code, they still could impact an unseen risk factor that lies outside of the known HDL-C level determining effects of the gene. Even if causality can be ultimately established, the mechanism by which a CETP-mediated increase in HDL-C may worsen ICH risk remains unclear. Inhibition of CETP results in changes to HDL particle size and cholesterol efflux capacity in addition to the observed changes in HDL-C serum levels, and it may be through these accompanying changes in HDL function that ICH risk is conferred. 40 Furthermore, accumulating evidence suggests that HDL effects on endothelium are dynamic and modifiable, and can even become proinflammatory with the incorporation of serum amyloid A1, complement C3, and ceramides, resulting in altered immune regulation and reduced antioxidant effects. 41,42 It is therefore possible that elevated HDL-C provides a platform to further the vascular inflammatory processes that play a substantial role in the cerebral small vessel disease underlying ICH. 43 Further studies will be needed to dissect the pathways intersecting with HDL-C to clarify the foundational biology of its role in ICH.
Therapeutic development of small molecule and biologic compounds designed to raise HDL-C continue. 44 Although the first wave of phase III trials of CETP inhibitors were plagued by off-target effects and futility, 45 the REVEAL trial of anacetrapib was recently continued after unblinded interim review. Other HDL-raising strategies, including apolipoprotein-A1 (ApoA1)-rich reconstituted HDL particle infusions and ApoA1-mimetic peptides, continue to be evaluated in preclinical and early phase trials. 44 Given this pipeline of HDL-based therapeutic development, it is imperative that potential adverse clinical effects of such strategies be clarified. Early experiences with US Food and Drug Administrationapproved PCSK-9 inhibitors have led to predictions of widespread adoption of this new class of drugs, and it is reasonable to expect that HDL-C targeted treatments would be no different, resulting in a potentially large population of aging individuals with pharmacologically induced high HDL-C levels of uncertain long-term cerebrovascular risk. 46 The proportion of variance in HDL-C levels explained by our genetic risk score was 0.032. This is roughly commensurate with observed effects of statins, which in clinical trials raised HDL by 0.04 to 0.10mg/ dl. 47 With emerging HDL-C modifying strategies likely to exert more profound effects, the impact on ICH risk, if confirmed and verified to be causal, could be more substantial than indicated by our CETP genetic risk score.
As noted above, our study cannot determine whether the observed association between CETP and ICH risk is through HDL-C alone. Although they exhibit their largest effect on HDL-C levels, CETP variants are also associated with LDL, triglycerides, and total cholesterol levels. 3 Although we cannot perform formal MR, the association between our HDL-C increasing genetic risk score at CETP and risk of ICH provides support for an HDL-specific effect. Even with this suggestion of HDL-C specificity, the composition of HDL particles can vary with respect to ratios of esterified to unesterified cholesterol as well as apolipoprotein content. Genetic variation that determines circulating HDL-C does not necessarily capture these secondary characteristics, which could have a substantial impact on biological effects. An additional limitation of our study is the aggregation of case and control data across multiple sites, which could result in biases between cases and controls. We have attempted to control for study demographics and population structure in our regression analyses, and performed independent replication, but unmeasured confounding could still have impacted the observed associations. Related to this point, all analyses presented were in individuals of European ancestry due to small study populations, and therefore low statistical power, in individuals of other racial and ethnic backgrounds. As a result, our findings cannot be extended to minority populations at this time.
Although our study utilized genome-wide data for discovery and genetic risk score analyses, our approach was fundamentally a candidate gene study of CETP. Using GWAS data allowed for control of population stratification, which can be a major confounder in traditional candidate gene designs employing only direct genotyping. However, it was still based on an a priori hypothesis about CETP association with ICH. Therefore, the false discovery rate for association between variants at CETP and ICH risk, although stringently controlled using Bonferroni correction at the CETP locus, may still be elevated in comparison with a standard GWAS. Due to the hypothesis-driven nature of our study, we by definition cannot provide novel results about lipid-related genetic loci that lie outside of the tested gene region.
Finally, the CETP gene contains several independent loci which have been associated with lipid levels and clinical endpoints. 3,5,7,32 This resulted in a more complex replication phase than would have been needed if the genetic architecture of the locus were centered about a single region of association. Coupled with the limitations of variant selection in our replication phase, we cannot distinguish a culprit variant to the exclusion of others. Although all variants chosen for replication demonstrated refined effect size estimates and greater statistical significance in meta-analysis with discovery data, replication was strongest for variants in slightly lower LD than the lead variant from discovery, and with slightly higher between-study heterogeneity. Whether this observation represents true heterogeneity of effect at the replicated variants will depend on future validation and extension studies.
We have demonstrated an association between genetic variants in CETP and risk of ICH, and have shown that CETP's HDL-C raising effects could play a role in the pathogenesis of ICH. Further work will be needed to identify how the biological pathways impacted by HDL-C may impart increased risk of hemorrhage. These pathways may yield crucial novel targets for prevention of ICH and the cerebral small vessel diseases that lead to vessel rupture.