Polymorphisms and Noncardioembolic Stroke in Three Case-Control StudiesLuke M.M.a · Berger K.e · Rowland C.M.a · Catanese J.J.a · Tong C.H.a · Ross D.A.a · Garcia V.a · Kuhlenbaeumer G.f · Ringelstein E.B.f · Pullinger C.R.b · Malloy M.J.b · Deedwania P.c · Ellis S.G.d · Kane J.P.b · Devlin J.J.a · Lalouschek W.g · Mannhalter C.g
aCelera, Alameda, Calif., bCardiovascular Research Institute, UCSF, San Francisco, Calif., cDepartment of Medicine, UCSF Fresno, Fresno, Calif., and dThe Cleveland Clinic, Department of Cardiovascular Medicine, Cleveland, Ohio, USA; eInstitute of Epidemiology and Social Medicine and fDepartment of Neurology, University of Muenster, Germany; gMedical University Vienna, Vienna, Austria Corresponding Author
Background: Gene variants associated with disease could reveal novel mechanisms. We searched for single nucleotide polymorphisms (SNPs) associated with noncardioembolic stroke (nonCES). Methods: We tested 24,926 SNPs in or near genes for association with nonCES in the Vienna Study (551 cases, 815 controls) and then evaluated the associated SNPs in the UCSF-CC Study (570 cases, 1,604 controls) first in pooled DNA samples and then in individual DNA samples. We then asked whether the risk alleles of the SNPs associated with increased risk in both studies were also associated with increased risk of nonCES in the German Study (728 cases, 1,041 controls). Results: Six of the 46 SNPs that were associated with nonCES in both the Vienna and the UCSF-CC Studies were also associated with nonCES in the German Study: rs362277 in HTT (OR 1.39, 90% CI 1.12–1.71), rs2924914 near CSMD1 (OR 1.22, 90% CI 1.04–1.43), rs1264352 near DDR1 (OR 1.20, 90% CI 1.02–1.41), rs544115 in NEU3 (OR 1.63, 90% CI 1.02–2.62), rs12481805 in UMODL1 (OR 1.31, 90% CI 1.01–1.81), and rs2857595 near NCR3 (OR 1.15, 90% CI 1.00–1.32). Accounting for multiple testing of 46 SNPs, these 6 SNPs had a false discovery rate of 0.69. Conclusions: Some of the 6 SNPs may be associated with nonCES but most may be false positives. These 6 SNPs merit investigation in additional nonCES study populations.
© 2011 S. Karger AG, Basel
Genetics association studies have been able to identify gene variants associated with ischemic stroke (IS). However, IS includes cardioembolic stroke (CES) and noncardioembolic stroke (nonCES) subtypes with distinct etiology and requiring different treatment strategies. It has been argued that genetic studies of heterogeneous phenotypes such as IS might not be valid subsequently  and that studies of specific stroke subtypes should be more reproducible . We previously reported a candidate-gene study for nonCES ; this study aimed to identify gene variants associated with nonCES using a non-candidate-gene approach in three case-control study populations.
Subjects and Methods
We tested 24,926 single nucleotide polymorphisms (SNPs) for association with nonCES among three case-control study populations using a previously described, staged study design . Briefly, we identified SNPs that had a p value of <0.05 when tested for association with nonCES in both of the first two study populations, first by pooled DNA studies and then by genotyping individuals. We then tested these SNPs in a third study population by genotyping.
All subjects were unrelated Caucasian men and women enrolled under Institutional Review Board- or Ethics Committee-approved protocols at the respective institutions. All participants gave written informed consent.
Cases and controls in the Vienna Study were as described  except that only 551 of the 562 cases had sufficient DNA and were included for the current analysis. The nonCES cases were IS cases other than the CES cases and included large vessel disease, small vessel disease, other determined etiology, and undetermined IS cases. There were 815 controls.
The UCSF-CC Study included 416 cases and 977 controls drawn from Genomic Resource at the University of California San Francisco (UCSF) and 154 cases and 627 controls drawn from the GeneBank of Cleveland Clinic (CC). Detailed information on the recruitment of the UCSF and CC participants is presented in the online supplementary material (for all online suppl. material, see www.karger.com/doi/10.1159/000333444). Cases in the UCSF-CC Study reported a history of IS in the questionnaire but information on the IS subtype was not available. To limit potential CES cases, cases with known risk factors for CES were excluded. Thus, cases from UCSF had no history of abnormal heart rhythm, heart valve disease or surgery. Cases and controls from CC were patients who had had coronary angiography. Cases had no history of atrial fibrillation, heart valve disease or surgery. Controls had no history of stroke or cardiovascular disease (including myocardial infarction, coronary stenosis >50%, peripheral vascular disease, or revascularization).
Cases in the German Study were recruited into the Westphalian Stroke Registry  and included 728 nonCES (consisting of 495 with atherothrombotic stroke, 230 with lacunar stroke, and 3 with stroke of undetermined cause) and 462 CES according to the TOAST criteria. Information on birthplace was not collected for the cases. The 1,041 controls in the German Study were recruited from the same region in Germany for the population-based Dortmund Health Study  and included 153 that were born outside of Germany. The German Study had at least 80% power to detect an association (2-sided test, type I error = 0.05) with an additive odds ratio of at least 1.36, 1.27 or 1.22 assuming minor allele frequencies (mAF) of 0.10, 0.20 and 0.40, respectively, before adjusting for multiple testing. Power calculations were performed with QUANTO.
The 24,926 SNPs tested (online suppl. table 1) were selected for being in or near 12,424 genes and either modify the encoded proteins (43%) or potentially the regulatory sequences (29% in introns, 7% each in untranslated regions and 3′or 5′ intergenic regions; the remaining 14% include SNPs in transcription factor or microRNA-binding sites, or encoding silent codon changes). Allele frequencies in pooled DNA samples were determined by real-time kinetic PCR  and genotypes of individual DNA samples were determined either by real-time kinetic PCR or by an oligonucleotide ligation assay  as detailed in the online supplementary materials.
The SNPs tested in the German Study were selected in the Vienna and the UCSF-CC Studies. The allele frequencies of 24,926 SNPs were determined in the Vienna Study in 8 pools of DNA from subjects categorized by case-control status, median age, and whether having either a history of diabetes or a family history of vascular disease in first-degree relatives. SNPs with allele frequencies associated with nonCES (2-sided p < 0.05) and mAF >0.01 were further tested in the UCSF-CC Study in 12 pools of DNA from subjects categorized by enrolling center, case-control status, and sex. SNPs that were associated with nonCES (1-sided Mantel-Haenszel p < 0.05 for UCSF and CC samples combined, and which had the same risk allele as in the Vienna Study), had consistent mAFs in UCSF and CC pooled DNA samples (Breslow-Day p > 0.05) and mAF >0.01 were confirmed by genotyping individual subjects of the Vienna and the UCSF-CC studies. SNPs meeting the following criteria in both the Vienna and the UCSF-CC studies were genotyped in the German Study: mAF >0.02, not deviated from Hardy-Weinberg equilibrium (HWE, p > 0.01), and associated with nonCES (1-sided p < 0.05, same risk allele) in additive (ADD), recessive (REC), or genotypic (where the odds ratios for risk allele heterozygotes (HET) and homozygotes (HOM), versus the non-risk homozygotes, were estimated in the same regression model) mode.
Differences between cases and controls were assessed by the Wilcoxon rank sum test (continuous variables) or χ2 test (discrete variables). Differences between case and control allele frequencies determined from DNA pools were assessed by the χ2 test. Association between SNPs and stroke was assessed by logistic regression models and, in the German Study, was adjusted for age, sex, diabetes, and hypertension (systolic blood pressure >140 mm Hg, diastolic blood pressure >90 mm Hg, a physician’s diagnosis of hypertension, or the use of anti-hypertensive medications). We considered associations with nonCES in the German Study to be significant only if the risk allele was the same as in the Vienna and the UCSF-CC studies and thus used 1-sided p values and 90% confidence intervals to assess the significance of the genetic associations. In the German Study, 8 SNPs were excluded from analysis: 4 had failed genotyping, and 4 had genotype frequencies that were out of Hardy-Weinberg equilibrium (p < 0.01). For 46 SNPs that were successfully genotyped for the nonCES analysis in the German Study, logistic regression models for three genetic models [ADD, dominant (DOM), and REC] were performed for each SNP and the minimum p value was recorded. A permutation test was performed for each SNP to calculate a corrected minimum p value. This is done by randomly permuting the genotypes among all subjects before recording for each SNP the permuted minimum p value from three correlated logistic regression models (ADD, DOM, REC) for 10,000 iterations and determining the proportion of permutations in which the permuted minimum p value was less than or equal to the observed minimum p value. To account for testing multiple SNPs, the false discovery rate q values were estimated by the method of Benjamini and Hochberg  using the corrected minimum p values obtained by permutation. The q value of a given SNP represents the expected proportion of false positives among the set of SNPs with equal or lower q values. For 6 SNPs that were genotyped for the CES analysis in the German Study, logistic regression models and a permutation test were similarly performed for each SNP to calculate a corrected minimum p value.
To detect the existence of population stratification, we used the Structure software  and the genotypes of the 46 SNPs for the German Study to evaluate models that assumed one, two, three, or four distinct ancestral populations among the German Study population. The proportions of ancestral populations attributed to each individual, as estimated by the Structure software, were recorded from the model having the highest likelihood. The association of the proportions of individual ancestry with case status was assessed using logistic regression. To adjust the odds ratios for ancestry, the proportions of the ancestral populations of each individual were included as an additional covariate in the multivariable logistic regression models.
The demographic and clinical characteristics of the case and control subjects are summarized in table 1.
|Table 1. Clinical characteristics of study populations|
Of the 24,926 SNPs tested in the Vienna Study, 3,280 SNPs had p values of <0.05 when tested for association with nonCES and were evaluated in the UCSF-CC Study. Of these, 54 had p values of <0.05 when tested for association with nonCES and had the same risk alleles in both studies (online suppl. table 2). Forty six of these SNPs were successfully genotyped and were analyzed for association with nonCES in the German Study (online suppl. table 3), and 6 SNPs in 6 loci had p values of <0.05 and had the same risk alleles as in the first two studies. After accounting for multiple testing in the German Study, the 6 SNPs associated with nonCES had a false discovery rate of 0.69. Adjusting for traditional risk factors (age, sex, hypertension, and diabetes) changed the odds ratios and the p values remained <0.05 for 3 SNPs (table 2).
|Table 2. Association of 6 SNPs with nonCES in the German Study|
We explored whether population stratification may have confounded the association between the 6 SNPs and nonCES in the German Study. We found that a model that assumed three ancestral populations was most consistent with the genotypes obtained from the German Study population and that population structure did not significantly differ between the cases and controls in the German Study (online suppl. figure). Adjusting for population structure did not appreciably change the odds ratios for the HTT, CSMD1, NEU3, and UMODL1 SNPs (the largest change in odds ratios was 0.03), but decreased the odds ratios of the DDR1 SNP (from 1.20 to 1.13) and the NCR3 SNP (from 1.15 to 1.08).
We then investigated the association of these 6 SNPs with CES in the German Study and found that the 3 SNPs in HTT, NEU3, and CSMD1 had p values of <0.05. Adjusting for traditional risk factors reduced the odds ratio of the CSMD1 SNP for CES (table 3). We also investigated the association of these 6 SNPs with two of the subtypes of nonCES, namely atherothrombotic stroke and lacunar stroke. We found that the SNP in HTT and the SNP in CSMD1 had p values of <0.05 for association with atherothrombotic stroke, while the SNPs in DDR1, UMODL, and NCR3 had p values of <0.05 for association with lacunar stroke (online suppl. table 4).
|Table 3. Association of 3 SNPs with CES in the German Study|
We investigated 24,926 SNPs and found that 6 SNPs in or near 6 genes (HTT, CSMD1, DDR1, NEU3, UMODL1, and NCR3) had p values of <0.05 when tested for association with nonCES in three case-control studies in which traditional risk factors of nonCES, such as age, male sex, hypertension, and diabetes were associated with the case status. Given multiple testing, there is insufficient evidence to consider these 6 SNPs as true associations, hence testing these 6 SNPs in additional nonCES studies will be important. For the same reason, the associations of these 6 SNPs with CES, or with the nonCES subtypes atherothrombotic stroke or lacunar stroke, should be further investigated. In addition, these SNPs could tag causative polymorphisms which may be revealed by testing other SNPs within the linkage disequilibrium blocks .
Some of these genes have plausible roles in vascular disease or neuronal function. HTT encodes huntingtin, and a variant form of the huntingtin protein is linked to Huntington’s Disease, a neurodegenerative disease . CSMD1 encodes CUB and Sushi multiple domains 1. The rat CSMD1 protein was reported to function in complement activation and inflammation, and is highly expressed in the leading edge of the nerve growth cone . DDR1 encodes discoidin domain receptor tyrosine kinase 1 which is highly expressed in oligodendrocytes during the neurodevelopmental myelination process  and deletion of the DDR1 gene decreases atherosclerosis in low-density lipoprotein receptor-deficient mice . NEU3 encodes membrane sialidase 3 which regulates insulin sensitivity and glucose tolerance in mice  and the activity of the NEU3 protein is associated with severity in acute stroke patients .
This report is limited by the case-control study design, which is subject to survival and treatment biases that could confound genetic associations. All CC subjects had had a coronary catheterization which may have introduced a selection bias. Furthermore, since IS subtype information was not recorded for the UCSF-CC cases, their nonCES status was inferred from the absence of risk factors of CES and not from physicians’ diagnosis. Even when IS subtype information was recorded, the diagnosis of IS subtype is still an evolving challenge [17,18]. In addition, although nonCES was a more homogeneous phenotype than IS, it was still heterogeneous and included large vessel, small vessel, and cryptogenic strokes. Allele frequencies determined in pooled DNA samples have greater measurement error compared with those determined by genotyping, and may have increased false-negative results during the pooled DNA stages of this study. This study investigated approximately 25,000 SNPs in 12,000 genes and did not cover all the linkage disequilibrium or haplotype blocks in the human genome. Finally, these three studies included only Caucasian subjects; analysis of current findings in large non-Caucasian stroke studies [10,19] is needed to determine whether these results may be applicable to other ethnicities.
In conclusion, we found that some of the 6 SNPs among the 24,926 SNPs tested may be associated with nonCES and these 6 SNPs should be further tested. Future genetic association studies of stroke could benefit from more homogeneous case definitions, larger numbers of subjects, and inclusion of non-Caucasian ethnicities.
Sources of Funding
The Vienna Stroke Registry and the Vienna Study were supported by grants of the Medizinisch-Wissenschaftlicher Fonds des Bürgermeisters der Bundeshauptstadt Wien (project numbers 1540, 1829, 1970), the Jubiläumsfonds der Oesterreichischen Nationalbank (project numbers 6866, 7115, 8281,9344), the Austrian Science Foundation (P13902-MED), the Wiener Krankenanstaltenverbund, and in part by Jubiläumsfonds der Österreichischen Nationalbank No. 9344. The UCSF subjects were drawn from Genomic Resource of Cardiovascular Research Institute funded in part by the Leducq Foundation, the Joseph Drown Foundation, and the Dhanem Foundation. The CC subjects were drawn from GeneBank of Cleveland Clinic funded by the Cleveland Clinic Foundation. Case ascertainment in the Westphalian Stroke Register was part of the German Competence Net Stroke, supported by the German Federal Ministry of Education and Research (01GI9909/3). Blood collection in the Dortmund Health Study (German controls) was done through funds from the Institute of Epidemiology and Social Medicine University of Muenster, while the collection of sociodemographic and clinical variables was supported by the German Migraine & Headache Society (DMKG) and by unrestricted grants of equal share from Astra Zeneca, Berlin Chemie, Boots Healthcare, GlaxoSmithKline, McNeil Pharma, MSD Sharp & Dohme and Pfizer to the University of Muenster. Genotyping and statistical analyses were conducted by the Core Laboratory and the Statistical Genetics Group, respectively, at Celera.
M.M.L., C.M.R., J.J.C., C.H.T., D.A.R., V.G., and J.J.D. are employees with ownership interest in Celera and have contributed to the study design, generation of data, interpretation of data, and writing of the manuscript. M.J.M., S.G.E., and J.P.K. had been paid consultants for Celera. The other authors have nothing to disclose.of the manuscript. M.J.M., S.G.E., and J.P.K. had been paid consultants for Celera. The other authors have nothing to disclose.
May M. Luke
Celera, 1401 Harbor Bay Parkway
Alameda, CA 94502 (USA)
Tel. +1 510 749 6267
Received: March 7, 2011
Accepted: September 19, 2011
Published online: December 1, 2011
Number of Print Pages : 6
Number of Figures : 0, Number of Tables : 3, Number of References : 19
Additional supplementary material is available online - Number of Parts : 3
Vol. 33, No. 1, Year 2012 (Cover Date: January 2012)
Journal Editor: Hennerici M.G. (Mannheim)
ISSN: 1015-9770 (Print), eISSN: 1421-9786 (Online)
For additional information: http://www.karger.com/CED