Subarachnoid hemorrhage: tests of association with apolipoprotein E and elastin genes

The methodology of the Genetic and Environmental Risk Factors of Hemorrhagic Stroke study have been previously reported [5, 11]. Cases of potential ICH or SAH in the Greater Cincinnati and Northern Kentucky are identified by surveillance of 16 hospital emergency and radiology departments and through monitoring of hospital discharge diagnoses. Eligible cases are ≥ 18 years, are without trauma or brain tumor as the cause of hemorrhage, and reside within a 50-mile radius of the University of Cincinnati. A subset of cases was invited to enroll in a direct interview and genetic sampling arm of the study. The response rate was reasonable with over 60% of the cases agreeing to participate. Two controls for each interviewed case, matched by age (± 5 years), race, and gender were recruited from the general population through random digit dialing. Controls were informed of their participation in a risk factor study. Institutional Review Boards at each hospital approved the study, and informed consents were obtained from the participants.

SAH was defined as non-traumatic abrupt onset of severe headache or altered level of consciousness associated with blood in the subarachnoid space on CT or at autopsy, or with a clinical history and examination consistent with SAH where xanthochromia and increased red blood cells are found in the cerebrospinal fluid.

A total of 309 Caucasians were included for analysis, of which 107 were SAH cases matched to 202 controls. There was no significant difference in the average age of cases and controls (51.48 ± 12.93 yrs vs. 50.68 ± 12.23 yrs; p = 0.521) or gender distribution (64.4% vs. 64.8% female; p = 0.992) between the cases and the controls. We also genotyped samples from a small group of African American subjects. However, the limited sample size (24 cases and 43 matched controls) lacked sufficient power to identify associations. Thus, our results refer only to Caucasian cases and controls.

Buccal swabs were collected from each participant at the time of interview, and DNA was extracted by standard methods. Genotyping for APOE ε2/ε3/ε4 alleles was performed by RFLP [17]. For analysis of the SNP markers, genomic DNA was preamplified by whole genome amplification (WGA) using improved-primer extension preamplification. The WGA kit (High-Fidelity Expand Template System) was obtained from Roche Pharmaceuticals. Five nanograms of DNA was subjected to WGA and then diluted 50-fold, of which 2 μl was used for SNP genotyping. The WGA protocols are validated for analysis of genetic markers in our laboratory [18].

The TaqMan™ (fluorogenic 5' nuclease) assay was used for SNP genotyping. The primers and probes were obtained from Applied Biosystems. PCR was conducted in ABI 9700 thermocyclers, and the end-point results scored using the ABI 7900HT Sequence Detection System. In each 384-well plate, two reference samples and two negative controls were included for quality control.

We analyzed 12 SNPs spanning a 16 kb fragment on the APOE gene (Table 1) of which five upstream markers are now assigned locations on the TOMM40 gene [19]. Physical distance between the most distal 3'SNP (rs10119) on TOMM40 and the most proximal 5'SNP (rs769446) on APOE is approximately 2 kb. These 12 SNPs were analyzed in our previous study on ICH, which showed haplotypic association with lobar ICH [11]. We selected 10 SNPs on the ELN gene (Table 2), which were reported in previous association studies involving aneurysmal SAH [14‐16].

Allele frequencies were estimated by gene counting. Conformity of genotype proportions to Hardy-Weinberg equilibrium (HWE) was tested by a goodness-of-fit χ² test. Haplotype frequencies and probabilities of haplotype pairs for each individual were estimated by PHASE version 2.1. PHASE implements Bayesian methods for estimating haplotypes from population data [20]. Allele and genotype frequency differences were tested using allelic and genotypic χ² tests, respectively. Haplotype associations were tested using χ² and haplotype trend regression (HTR) [21]. All p-values for allele, genotype and haplotype associations were empirically determined by Monte Carlo simulations as described by Becker and colleagues [22, 23]. Multiple testing was accounted for by testing a global hypothesis of no association for each of the single locus and haplotype test [22, 23].

We performed a genomic control analysis to adjust for population substructure [24]. This was done by typing 30 unlinked SNPs distributed throughout the genome as null markers in all samples (cases and controls) and conducting test statistics to estimate λ following the methods as described [25].

Genotype and allele frequencies among the cases and the controls are presented in Table 3. Genotypes at all 12 markers were in HWE. We did not observe significant association either at allelic or genotypic level (global p = 0.452 and 0.807, respectively). Marginal frequency differences between cases and controls at allelic and genotype levels were observed in SNP 1 and 4 (Table 3). We performed a separate analysis for the isoformic ε2/ε3/ε4 alleles, which also revealed no significant association (Table 4).

We conducted an analysis to test for association at the haplotype level based on data from all 12 SNPs (Table 5). Using PHASE, we observed a total of 66 haplotypes. In Table 5, we present data on 9 common haplotypes (frequency >0.02 in either cases or controls as inferred by PHASE. TGTCGATCCTTC (Hap1) was inferred as the most common haplotype, which accounted for 30% and 19% of all haplotypes in cases and controls, respectively. Global tests of association were significant for both χ² test and HTR (p = 0.019 and 0.036, respectively). Among all haplotypes, Hap1 showed significant difference between cases and controls by both methods (Table 5).

To explore the haplotype association at a finer level, we conducted a 'sliding haplotype' analysis to decompose Hap1, the most common and significantly associated haplotype. None of the other haplotypes showed significant association even with the sliding window analyses (data not shown), and they were consequently dropped from further analysis. A graphical representation of this analysis is presented in Figure 1. The uppermost circle in the figure, with the number '1–12', represents Hap1 inferred using all SNPs (SNPs 1 to12). Progressively smaller windows of this haplotype were formed at each level of the pyramid. These depict windows created with combinations of SNPs used in haplotype construction for that level (e.g., circle 1–11 represents the most common haplotype formed by SNPs 1 to 11, circle 2–12 represents the most common haplotype formed by SNPs 2 to 12, and so on). Haplotypes with significant associations are shown in yellow (p ~ 0.05–0.1), green (p ~ 0.01–0.05) and red (p < 0.01). A trend of ascending significance is observed at the 5' region. Marginal associations are also observed at the 3' region, where SNPs encoding the isoformic ε2/ε3/ε4 variants are located (SNP10 and SNP12 together form the ε2/ε3/ε4 variants – cytosine at both sites result in the ε4 isoform, thymine at the first site and cytosine at the second site form ε3, and thymine at both sites yield ε2).

Genotype and allele frequencies of ELN SNPs among the cases with aneurysmal SAH and controls are presented in Table 6. Genotypes were in HWE with the exception of SNP2 and SNP7. One marker rs13229379 (SNP6) was found to be monomorphic in our population and was thus dropped from subsequent analysis. None of the SNPs were associated with SAH.

Using PHASE, we inferred haplotypes based on all 9 SNPs and did not observe significant associations (global p = 0.229 and 0.198; Table 7). There was no association of haplotypes even after excluding the two markers (SNP2 and SNP7) that were not in HWE (data not shown). Further, the sliding window analysis showed no association between ELN and SAH.