Burden of Risk Alleles for Hypertension Increases Risk of Intracerebral Hemorrhage
Background and Purpose—Genetic variation influences risk of intracerebral hemorrhage (ICH). Hypertension (HTN) is a potent risk factor for ICH and several common genetic variants (single nucleotide polymorphisms [SNPs]) associated with blood pressure levels have been identified. We sought to determine whether the cumulative burden of blood pressure-related SNPs is associated with risk of ICH and pre-ICH diagnosis of HTN.
Methods—We conducted a prospective multicenter case–control study in 2272 subjects of European ancestry (1025 cases and 1247 control subjects). Thirty-nine SNPs reported to be associated with blood pressure levels were identified from the National Human Genome Research Institute genomewide association study catalog. Single-SNP association analyses were performed for the outcomes ICH and pre-ICH HTN. Subsequently, weighted and unweighted genetic risk scores were constructed using these SNPs and entered as the independent variable in logistic regression models with ICH and pre-ICH HTN as the dependent variables.
Results—No single SNP was associated with either ICH or pre-ICH HTN. The blood pressure-based unweighted genetic risk score was associated with risk of ICH (OR, 1.11; 95% CI, 1.02–1.21; P=0.01) and the subset of ICH in deep regions (OR, 1.18; 95% CI, 1.07–1.30; P=0.001), but not with the subset of lobar ICH. The score was associated with a history of HTN among control subjects (OR, 1.17; 95% CI, 1.04–1.31; P=0.009) and ICH cases (OR, 1.15; 95% CI, 1.01–1.31; P=0.04). Similar results were obtained when using a weighted score.
Conclusion—Increasing numbers of high blood pressure-related alleles are associated with increased risk of deep ICH as well as with clinically identified HTN.
Wrldwide stroke is the second leading cause of death and the leading cause of acquired disability.1 Intracerebral hemorrhage (ICH), the severest form of stroke, accounts for 15% of acute strokes in the United States. Despite advances in neurocritical care, >75% of patients will die or become severely disabled as a result of their ICH.2 Effective preventive and acute treatments are therefore urgently needed.
Hypertension (HTN) is a potent risk factor for ICH.3 This effect is strongest for ICH in deep hemispheric locations.4 HTN has also been associated with increased ICH volumes and worse clinical outcome.5,6 In recent years, genomewide association studies have identified several common genetic variants (or single nucleotide polymorphisms [SNPs]) associated with blood pressure (BP) levels.7–9 Each of these common genetics variants, however, exerts only a small effect on BP. Consequently, estimating the combined effect that all these SNPs produced may be the only way to determine whether these variants influence risk of ICH. Genetic risk scores (GRSs) can be implemented to obtain an aggregate measure of the burden of risk alleles related to high BP carried by each individual.10,11 This approach has already shown that larger burdens of risk alleles for high BP levels are associated with increased risk of stroke.9
Within the International Stroke Genetics Consortium’s (ISGC) ongoing genomewide association study of ICH, we investigated the role of BP-associated SNPs on both ICH and pre-ICH diagnosis of HTN. We hypothesized that individuals with larger burdens of BP alleles will have an increased risk of ICH, specifically in deep locations of the brain. We also postulated that subjects with ICH carrying greater numbers of risk alleles for high BP should have an increased risk of HTN.
Materials and Methods
Study Design and Patients
We used a multicenter case–control design for the outcomes ICH and pre-ICH HTN in subjects of self-reported European ancestry from the following ISGC studies: Hospital del Mar Intracerebral Hemorrhage study12 in Barcelona, Spain; the Jagiellonian University Hemorrhagic Stroke Study13 in Krakow, Poland; the Lund Stroke Register14 in Lund, Sweden; the Vall d’Hebron Hospital ICH Study15 in Barcelona, Spain; the Medical University of Graz Intracerebral Hemorrhage study16 in Graz, Austria; the Genetic and Environmental Risk Factors for Hemorrhagic Stroke4 at the University of Cincinnati in Cincinnati, Cincinnati, OH; and the Genetics of Cerebral Hemorrhage on Anticoagulation study17 in the United States (participating sites included Massachusetts General Hospital, Beth Israel Deaconess Medical Center, Mayo Clinic Jacksonville, and the Universities of Michigan, Virginia, Florida at Jacksonville, Washington, and Utah).
All studies were approved by the Institutional Review Board or ethics committee of participating institutions. All participants provided informed consent; when subjects were not able to communicate, written consent was obtained from their legal proxies.
Cases were enrolled according to methods previously described.18 ICH was defined as a new and acute (<24 hours) neurological deficit with compatible brain imaging showing the presence of intraparenchymal bleeding. Enrolled subjects were primary acute ICH cases that presented to the emergency department of participating institutions (all accredited stroke centers) who provided written consent, were >18 years of age, and had confirmation of primary ICH through neuroimaging (either CT or MRI). Exclusion criteria included: anticoagulation, trauma, brain tumor, hemorrhagic transformation of a cerebral infarction, vascular malformation, or any other cause of secondary ICH. Additional recorded clinical characteristics included pre-ICH exposure to antiplatelet drugs or statins, history of ICH in a first-degree relative, and alcohol or tobacco use.
Ascertainment of ICH cases and assignment of hemorrhage location were performed by stroke neurologists at each ISGC site. ICH located in the cortex (with or without involvement of subcortical white matter) was defined as lobar, whereas ICH selectively involving the thalamus, internal capsule, basal ganglia, or brain stem was defined as deep. Cerebellar hemorrhages were excluded from the study.
Control subjects were >18 years of age and were enrolled from the same population that gave rise to the cases at each participating institution. Control subjects came from the same geographical area as the cases and 2 different sampling techniques were used for enrollment. For the Genetic and Environmental Risk Factors for Hemorrhagic Stroke study, random digit dialing was implemented. For the remainder of the studies, control subjects were enrolled through ambulatory clinics. This last control sampling strategy can sometimes introduce selection bias. To assess this possibility, the distribution of the BP-based GRS was compared between Genetic and Environmental Risk Factors for Hemorrhagic Stroke (that used random digit dialing) and the rest of studies by means of analysis of variance. Control subjects were confirmed to have no history of previous ICH by means of interview and review of medical records. Recorded clinical characteristics were identical to ICH cases.
Cases and control subjects were considered to have HTN when they (or their proxies) reported a medical history of HTN or when receiving anti-HTN medications at the time of admission with ICH. Several validation studies have shown that this approach has acceptable accuracy as compared with direct ascertainment of HTN through BP measurement.19–22
Peripheral whole blood was collected from cases and control subjects at each participating institution at the time of consent. Blood samples were subsequently shipped to the Massachusetts General Hospital, the coordinating center, and genotyping was carried out at the Broad Institute. DNA was isolated from fresh or frozen blood, quantified with a quantification kit (Qiagen, Valencia, CA), and normalized to a concentration of 30 ng/μL. Genotyping was performed using Affymetrix 6.0 (Santa Clara, CA) in Genetic and Environmental Risk Factors for Hemorrhagic Stroke and Illumina 610k (San Diego, CA) in the rest of the studies. Quality control procedures were implemented as described in online-only Data Supplement Figure I. MACH software23 was used to impute unobserved SNPs based on reference panels from HapMap24 and the 1000-genomes project.25
Selection of SNPs Associated With BP
SNPs associated with BP levels at P<1 × 10E−7 were selected from the National Human Genome Research Institute genomewide association study catalog.26 To ensure that the results of this study reflect independent effects, SNPs in each chromosome were pruned to avoid including variants in linkage disequilibrium (r2 >0.5).
Assessment of the relation between each BP-associated SNP and risk of ICH or HTN was carried out after principal components analysis was implemented to account for population stratification.27 Principal components were initially applied to identify and remove population outliers and subsequently entered as covariates in the regression models that were fit to test each hypothesis.
Genetic Association Analysis for Individual Variants
Single-SNP genetic association testing was completed within each sample using logistic regression, assuming additive effects for each risk allele present, and including age, sex, and principal components 1 and 2 in the model. Results for individual samples were combined in meta-analysis using inverse varianceweighted, fixed-effects meta-analysis.
The main exposure of interest in the present study is the burden of risk alleles for increased BP, as expressed by a GRS. Both weighted GRS (wGRS) and unweighted GRS (uGRS) were calculated. A wGRS is the sum of the products of the risk allele count (0, 1, or 2) at each locus multiplied by the reported effect of that risk allele on BP. An uGRS is simply the sum of the risk alleles for BP across the selected loci. In both instances, for SNPs reported to have minor alleles that reduce BP, the risk allele was set to be the other (major) allele.
Association Analysis for GRS
Multivariate logistic regression was used to model the risk of HTN or ICH using age and sex as covariates. These covariates were included in the model for efficiency. In all models, the GRS was converted to the standard normal distribution and entered as a continuous predictor. In this context, the β for the GRS can be interpreted as the increase in risk of the outcome per 1-SD increase of the GRS.
Two additional association analyses involving GRSs were carried out, one excluding brain stem hemorrhages and the other adding principal components 1 and 2 as covariates in the model. To ascertain if the effect of the GRS on ICH was mediated through clinically observed HTN, the multivariate model described for ICH was rerun entering HTN as a covariate. Finally, the same models were also implemented after stratifying by HTN status.
Statistical significance was considered to be Bonferroni-corrected P<0.001 and P<0.017 for single-SNP association analyses (39 tests) and GRS analyses (3 tests: all, deep, and lobar ICH), respectively, all tests being 2-sided. Genetic association testing for single variants as well as score calculations were performed in PLINK (http://pngu.mgh.harvard.edu/purcell/plink/).28 All other statistical analyses were performed in SAS 9.2 (SAS Institute, Cary, NC). Post hoc power analysis showed that the study would achieve 90% power to detect a risk increase of 10% per additional SD of the GRS.
A total of 2272 subjects were included in the study: 1025 ICH cases and 1247 ICH-free control subjects (mean [SD] age 71 [±12] years, 48% female). Among cases, 521 (53%) had deep and 462 (47%) had lobar hemorrhages (Table 1). Fifty-two SNPs associated with BP levels were identified from the National Human Genome Research Institute genomewide association study catalog. After pruning to remove genetic variants in high linkage disequilibrium, 38 SNPs remained to be used in the GRS (online-only Data Supplement Table I). Tested independently in meta-analysis, no single SNP was associated with either pre-ICH HTN among controls (online-only Data Supplement Table II) or ICH (online-only Data Supplement Tables III, IV, and V).
The GRS was associated with a diagnosis of pre-ICH HTN among control subjects and among subjects with lobar hemorrhages (Table 2). Within control subjects, each additional SD of the GRS produced an increase in risk of HTN of 22% (OR, 1.22; 95% CI, 1.08–1.37; P=0.001) and 17% (OR, 1.17; 95% CI, 1.10–1.31; P=0.009) for wGRS and uGRS, respectively (Table 2). Within cases with lobar ICH, each additional SD of the GRS produced an increase in risk of HTN of 32% (OR, 1.32; 95% CI, 1.08–1.61; P=0.006) and 25% (OR, 1.25; 95% CI, 1.03–1.51; P=0.02; Table 2) for wGRS and uGRS, respectively.
The GRS was associated with risk of all (deep and lobar) and deep ICH but not with lobar ICH (Table 3). When considering all (deep and lobar) ICH cases, each additional SD of the GRS produced an increase in risk of ICH of 10% (OR, 1.10; 95% CI, 1.01–1.19; P=0.03) and 11% (OR, 1.11; 95% CI, 1.02–1.21; P=0.01; Table 3) for wGRS and uGRS, respectively. When including only deep hemorrhages in the analysis, each additional SD of the GRS produced an increase in risk of ICH of 15% (OR, 1.15; 95% CI, 1.04–1.27; P=0.008) and 18% (OR, 1.18; 95% CI, 1.07–1.30; P=0.001; Table 3) for wGRS and uGRS, respectively. These results remained unchanged when excluding brain stem hemorrhages and when adding principal components 1 and 2 to the model.
The association between the GRS and ICH appears to be stronger in nonhypertensives (Table 4; Figure). When stratifying by HTN status, the effect of the GRS remained present within deep ICH in nonhypertensives (per increase in 1 SD of the uGRS OR, 1.25; 95% CI, 1.06–1.50; P=0.007) but not in hypertensives (per increase in 1 SD of the uGRS OR, 1.1; 95% CI, 0.97–1.25; P=0.14). No effect was observed within HTN strata for all and lobar ICH. When incorporating HTN into the model, the strength of the association was not substantially modified (online-only Data Supplement Table VI).
The present study demonstrates that the burden of risk alleles for BP, as measured by the GRS, is associated with risk of ICH. We also show that, as expected, the GRS is associated with clinically identified pre-ICH HTN. In line with previous findings suggesting differences in underlying biology between deep and lobar ICH,29 the association between the GRS and risk of ICH appears to be driven by ICH in deep regions of the brain. This association remained significant even when HTN was incorporated into the model, a finding that raises the possibility of misclassification of HTN status given that self-report and medication intake, and not actual BP levels, were used to ascertain this status.
This is the first demonstration that genetic variants for BP also influence risk of ICH. These findings build on previous reports demonstrating the feasibility of applying GRSs to stroke. A mitochondrial genomewide association study of ischemic stroke described a relation between a GRS generated with mitochondrial variants and risk of ischemic stroke.30 Our results also complement the conclusions of a recent report that specifically looked into the role of the genetics of hypertension in stroke. This report was a subanalysis of a large meta-analysis of genomewide association studies of BP9 and found an association between the aggregate burden of BP variants and stroke. In that study, however, the effect of the BP-associated GRS was not assessed specifically for ICH or for ICH subtypes.
Deep ICH has been primarily attributed to the effects on the cerebral vessels of long-standing hypertension, whereas a substantial proportion of lobar ICH appears to arise in the setting of amyloid angiopathy. In the present study, the association between the BP-based GRS and ICH was restricted to deep ICH cases. Furthermore, the association for deep ICH was predominantly observed in subjects who had not been labeled as hypertensives. One explanation could be that the GRS captured increases in risk of ICH produced by BP levels that are below those currently used to establish a diagnosis of HTN. Given the limitations of the approach implemented in the study to ascertain HTN, a second possibility is that subjects labeled as nonhypertensives in this population were misclassified.
These data have important implications for risk prediction of ICH. Given the limited impact of acute treatments for this condition, identifying subjects at highest risk of sustaining an ICH is of paramount importance, because it would open the possibility of implementing aggressive preventive strategies in high-risk individuals. Genetic data can aid in this goal, as they are available from birth, long before hypertension is diagnosed, are constant over time, and are not subject to misclassification and can be collected quickly, inexpensively, and painlessly. Importantly, this same approach could be applied to other risk factors and intermediates, and combined genetic data on common variants for these intermediates could be used to build increasingly precise risk prediction models.
With regard to future research directions, our results demonstrate that, considered in isolation, no single variant related to HTN is associated with ICH. Indeed, it is the aggregate burden of these variants that, in the end, increase the risk of sustaining an ICH. Future investigations can leverage this finding and test if genetic variation affecting entire biological pathways known to influence BP influence risk of ICH. Furthermore, this same approach may be applied to other biological processes and risk factors known to play a role in ICH, like hypercholesterolemia, alcohol abuse, smoking, and obesity.
The present study was undertaken within the largest sample assembled of ICH cases with available genomewide data. In addition, the collection of clinical data and biological samples was done following standardized, prespecified guidelines at every participating site, and stroke neurologists and neuroradiologists ascertained the cases and described important phenotypic characteristics, including ICH location. These last 2 features combined greatly decrease the possibility of outcome misclassification. This is particularly important in the field of stroke, where misclassification of stroke subtypes is usually a limitation.
A number of limitations in the study should be addressed. First, the fact that some of the controls were selected in ambulatory clinics introduces the possibility of selection bias. This would be important because the proportion of hypertensive controls could be higher using this sampling scheme, with a consequent increase in the presence of SNPs associated with hypertension among controls. It should be noted, however, that this situation, if anything, would bias the results toward the null. Additionally, the distribution of the GRS among controls enrolled by Genetic and Environmental Risk Factors for Hemorrhagic Stroke, a study that implemented random digit dialing, was similar to that observed in controls enrolled by studies that applied a sampling scheme based on ambulatory clinics (data not shown). Second, selection bias could also be present in the form of survival bias. Patients picked up in the setting of a case–control design would be those who survived the onset of an ICH, thus reaching the hospital and allowing for their enrollment. As has been shown recently, however, simulation results suggest that the effect on risk estimates introduced in this setting would be relatively small.31
In conclusion, we show that an association exists between the burden of risk alleles for elevated BP and the risk of deep ICH. This association is stronger for those individuals labeled as nonhypertensives. Further research is needed to evaluate the clinical value of this genetic risk score.
Sources of Funding
This study is funded by National Institutes of Health–National Institute of Neurological Disorders and Stroke (NIH-NINDS) grant 5U54NS057405 (Differences in the Imaging of Primary Hemorrhage Based on Ethnicity or Race [DECIPHER]), Genetic and Environmental Risk Factors for Hemorrhagic Stroke. This study was supported by National Institutes of Health grants NS36695 (Genetic and Environmental Risk Factors for Hemorrhagic Stroke) and NS30678 (Hemorrhagic and Ischemic Stroke Among Blacks and Whites) and by the Greater Cincinnati Foundation Grant (Cincinnati Control Cohort), Genetics of Cerebral Hemorrhage on Anticoagulation. This study was funded by NIH-NINDS grants R01NS059727, the Keane Stroke Genetics Research Fund, the Edward and Maybeth Sonn Research Fund, by the University of Michigan General Clinical Research Center (M01 RR000042), and by a grant from the National Center for Research Resources. Drs Biffi and Anderson were supported in part by the American Heart Association/Bugher Foundation Centers for Stroke Prevention Research (0775010N), Hospital del Mar ICH study: Ministerio de Sanidad y Consumo de España, Instituto de Salud Carlos III with the grants: “Registro BASICMAR” Funding for Research in Health (PI051737); “GWALA project (PI10/02064)”; Grant from Spanish Research Networks “Red HERACLES” (RD06/0009). Jagiellonian University Hemorrhagic Stroke Study: This study is supported by a grant funded by the Polish Ministry of Education (N N402 083934). Lund Stroke Register: Lund University, Region Skåne, the Swedish Research Council (K2010-61X-20378-04-3), the Freemasons Lodge of Instruction EOS in Lund, King Gustav V and Queen Victoria’s Foundation, Lund University, the Department of Neurology Lund, and the Swedish Stroke Association. Biobank services and genotyping were performed at Region Skåne Competence Centre (RSKC Malmö), Skåne University Hospital, Malmö, Sweden. Medical University of Graz ICH Study: control subjects of the MUG-ICH study are from the Austrian Stroke Prevention Study (ASPS), a population-based study funded by the Austrian Science Fond (FWF) grant numbers P20545-P05 and P13180. The Medical University of Graz supports the databank of the ASPS. All funding entities had no involvement in study design, data collection, analysis, and interpretation, writing of the report and in the decision to submit the paper for publication.
Christoph Kleinschnitz, MD, was the guest editor for this article.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.112.659755/-/DC1.
- Received April 4, 2012.
- Revision received June 15, 2012.
- Accepted June 20, 2012.
- © 2012 American Heart Association, Inc.
- Brott T,
- Thalinger K,
- Hertzberg V
- Woo D,
- Sauerbeck LR,
- Kissela BM,
- Khoury JC,
- Szaflarski JP,
- Gebel J,
- et al
- Qureshi AI
- Morrison AC,
- Bare LA,
- Chambless LE,
- Ellis SG,
- Malloy M,
- Kane JP,
- et al
- Hivert M-F,
- Jablonski KA,
- Perreault L,
- Saxena R,
- McAteer JB,
- Franks PW,
- et al
- Gomis M,
- Ois A,
- Rodríguez-Campello A,
- Cuadrado-Godia E,
- Jiménez-Conde J,
- Subirana I,
- et al
- Hallström B,
- Jönsson A-C,
- Nerbrand C,
- Norrving B,
- Lindgren A
- 17.↵Genes for Cerebral Hemorrhage on Anticoagulation (GOCHA) Collaborative Group. Exploiting common genetic variation to make anticoagulation safer. Stroke. 2009;40:S64–S66.
- Colditz GA,
- Martin P,
- Stampfer MJ,
- Willett WC,
- Sampson L,
- Rosner B,
- et al
- Hindorff LA,
- MacArthur J,
- Wise A,
- Junkins HA,
- Hall PN,
- Klemm AK,
- et al
- Anderson CD,
- Nalls MA,
- Biffi A,
- Rost NS,
- Greenberg SM,
- Singleton AB,
- et al