Association study of ABCA1 polymorphisms in singapore populations 6

The overall goal of the study was to establish the contributions of the selected ABCA1 SNPs to ethnic differences in CAD risk and lipid levels, especially HDL-C, among Singapore Chinese,

6 Association Study of ABCA1 Polymorphisms in Singapore Populations

6.1 Introduction

High density lipoprotein (HDL) levels are inversely correlated with the incidence of coronary artery disease (CAD; Wang and Briggs, 2004) A long-standing hypothesis to explain the atheroprotective effect of HDL is the process of reverse cholesterol transport (RCT; Glomset, 1968) In RCT, HDL or its apolipoproteins mediate the removal of excess free cholesterol from peripheral cells and, following a series of reactions in plasma, the cholesterol is delivered via either low density lipoprotein (LDL) or HDL to the liver for excretion into the bile Until recently, little was understood about what primes the initial step of RCT, namely, the removal of cholesterol from peripheral cells Moreover, evidence that definitively established the relationship between cholesterol efflux, HDL and atherosclerosis had been elusive

Tangier disease is a rare recessive disorder characterized by a near absence of plasma HDL-cholesterol (HDL-C) and a marked deposition of cholesterol esters in macrophage-rich tissues (Fredrickson et al., 1961) Biochemically, fibroblast cultures from Tangier disease patients are associated with a decrease in cholesterol (Walter et al., 1994)

as well as phospholipid efflux (von Eckardstein et al., 1998) Although first described in

1961, the molecular basis for Tangier disease has only been recently solved as homozygosity in ABCA1 gene mutations (Bodzioch et al., 1999; Brooks-Wilson et al., 1999; Remaley et al., 1999; Rust et al., 1999) Certain patients with the milder and more common form of familial hypoalphalipoproteinemia, also known as familial HDL deficiency, are characterized by heterozygosity in ABCA1 gene mutations (Brooks-Wilson et al., 1999; Marcil et al., 1999) The recognition of ABCA1 gene defects as the molecular basis for these two forms of inherited HDL deficiencies has shed much light on the role of the

regulated and energy-dependent transfer of excess cholesterol and phospholipids from peripheral cells including macrophages in the arterial wall to acceptor molecules in the plasma such as ApoAI, thereby generating nascent HDL particles and igniting the process

of RCT (Oram, 2003)

The discovery of ABCA1 as a crucial mediator in the first step of RCT, together with the well established inverse relationship between HDL-C levels and atherosclerosis, provide the scientific notion that ABCA1 is the much sought after direct link between HDL and atherosclerosis Direct evidence comes from observations that an impairment of cholesterol efflux is associated with increased risk of CAD and HDL-C seen in ABCA1 human heterozygotes (Clee et al., 2000; van Dam et al., 2003) as well as from transgenic mouse models (Singaraja et al., 2003)

More generally, there is interest that the more common genetic variations in ABCA1 may explain phenotypic variability in HDL-C levels and/or CAD risk among individuals of the general population To date, numerous association studies using single-nucleotide polymorphisms (SNPs) both in the coding and promoter regions of the ABCA1 gene have been reported in Caucasian, African-American and Japanese populations (Wang et al., 2000; Brousseau et al., 2001; Clee et al., 2001; Lutucuta et al., 2001; Zwarts

et al., 2002; Cenarro et al., 2003; Evan et al., 2003; Harada et al., 2003; Kakko et al., 2003; Srinivasan et al., 2003; Yamakawa-Kobayashi et al., 2003; Shioji et al., 2004; Cohen

et al., 2004; Frikke-Schmidt et al., 2004; Tregouet et al., 2004) A summary of the finding

of these studies are compiled in Table 6.1 The results, however, have not been entirely consistent To illustrate, for R219K, the most well-studied ABCA1 SNP to date, several studies show a protective association of the rarer K219 allele against CAD, which may or may not be accompanied by parallel increases in HDL-C and/or apolipoprotein AI (ApoAI) concentrations (Clee et al., 2001; Cenarro et al., 2003; Evans et al., 2003; Kakko et al., 2003;Tregouet et al., 2004; Yamakawa-Kobayashi et al., 2004) However, two recent large

sample studies (Cohen et al., 2004; Frikke-Schmidt et al., 2004) failed to replicate the protective association of the K219 allele Similarly, studies involving the M883I locus have also been variable, ranging from no evidence of association to entirely contradictory effects (Table 6.1)

Here, we have conducted an association study of seven ABCA1 SNPs spanning a region of ~145kb using population-based samples from Singapore The seven SNPs examined in the association study were: -14C>T, 237indelG, R219K, V825I, M883I, IVS44+18T>C and 8995A>G (listed in order according to their chromosomal location) A schematic of the locations of the SNPs with respect to the genomic sequence and mRNA (for coding and UTR SNPs only) is shown in Figure 6.1 These SNPs were selected for various reasons All are polymorphic in all three ethnic groups of Singapore with a minimum allele frequency of 5% With the exception of 8995A>G, we were also restricted

to evaluating SNPs which could be readily genotyped using simple restriction fragment length analysis without prior extensive assay development Associations with CAD susceptibility and/or HDL-C had been previously reported for the coding SNPs R219K, V825I and M883I (Table 6.1), and it was therefore of interest to determine the replicability

of those findings in the local context We also studied -14C>T, as well as two novel SNPs discovered in the course of the study, IVS44+18T>C and 8995A>G The promoter SNP -14C>T is less extensively investigated and nothing is known regarding the contributions of the intronic variant IVS44+18T>C and the 3’ untranslated region (UTR) variant 8995A>G

to the traits of interest

In Singapore, Indians show the highest incidence of CAD followed by Malays and Chinese, and this risk difference parallels ethnic differences in HDL-C (Heng et al., 2000) This observation, in the context of a fairly homogeneous physical environment, may suggest the presence of a strong genetic component in determining CAD risk and HDL-C

SNPs between male CAD patients (cases) and their ethnic-matched healthy controls In addition, a second analysis was carried out to explore the effect of these ABCA1 SNPs on quantitative variation in HDL-C, ApoAI and triglycerides (TG) The latter analysis was investigated only in control groups in order to exclude the effect of lipid-lowering treatment

in cases Both single- and multi-locus analyses were applied The overall goal of the study was to establish the contributions of the selected ABCA1 SNPs to ethnic differences in CAD risk and lipid levels, especially HDL-C, among Singapore Chinese, Malay and Indian males

Increased association with severity of atherosclerosis Shioji et al (2004) 3 large Japanese cohorts No association

Tregouet et al (2004) Two UK-based populations Marginally higher ApoAI for T allele

Ye et al (2005) UK white CAD patients Lower expression activity for T allele

Increased CAD in TT homozygotes especially in smokers -302C>T Promoter Shioji et al (2004) 3 large Japanese cohorts Association with higher HDL-C in general population but

not in hypertension group -278C>G Promoter Zwarts et al (2002) Dutch CAD patients, cohort design Increased CAD events, no effect on lipids and genotypes

Shioji et al (2004) 3 large Japanese cohorts Association with HDL-C in general population and hypertension group -99C>G Promoter Zwarts et al (2002) Dutch CAD patients, cohort design Fewer CAD events, no effect on lipids

Shioji et al (2004) 3 large Japanese cohorts No association -14C>T Promoter Zwarts et al (2002) Dutch CAD patients, cohort design Increased atherosclerosis among T allele carriers

Shioji et al (2004) 3 large Japanese cohorts No association 237indelG Promoter Zwarts et al (2002) Dutch CAD patients, cohort design Less severe atherosclerosis

R219K Coding Brousseau et al (2001) US case-control study Higher frequency in cases

No effect on lipids Clee et al (2001) Dutch CAD patients, cohort design Decreased CAD severity, progression

Decreased TG Increased cholesterol efflux and HDL-C in age-dependent manner Cenarro et al (2003) German CAD patients Higher frequency in patients without premature CAD

Interaction effect: Protection more among smokers than smokers

non-Evans et al (2003) German case-control study Lower K frequency in cases

Decreased TG in cases who had not received lipid-lowering treatment, especially those with ApoE3/3 genotypes.

Harada et al (2003) Japanese CAD patients No association with HDL-C and susceptibility to CAD

Association with TG in K carriers, but dissipiated after adjusting for M883I, gender, BMI, smoking, hypertension and diabetes Kakko et al (2003) Finnish population-based study Higher HDL-C in women KK individuals

Srinivasan et al (2003) US whites, population-based No significant marginal effects on HDL

Significant interaction of genotypes and age on HDL

K carriers associated with higher HDL with age.

Cohen et al (2004) US whites and blacks, general population No association with HDL-C Frikke-Schmidt et al (2004) Danish general population No association with HDL-C Shioji et al (2004) 3 large Japanese cohorts No association with HDL-C Tregouet et al (2004) Two UK-based populations Fewer myocardial infarction Yamakawa-Kobayashi et al (2004) Healthy Japanese young individuals Increased HDL-C and ApoAI in KK individuals V771M Coding Clee et al (2001) Dutch CAD patients, cohort design Decreased CAD severity in VM heterozygotes

Yamakawa-Kobayashi et al (2004) Healthy Japanese young individuals Higher HDL-C and ApoAI in M carriers.

Cohen et al (2004) US whites and blacks, general population No association with HDL-C Frikke-Schmidt et al (2004) Danish general population Higher HDL-C in VM heterozygotes in women Tregouet et al (2004) Two UK-based populations No association with ApoAI

V825I Coding Clee et al (2001) Dutch CAD patients, cohort design Increased events in I carriers during trial

Cohen et al (2004) US whites and blacks, general population Modest assocation seen in white men; direction of effect not specified Frikke-Schmidt et al (2004) Danish general population Higher HDL-C in I heterozygotes in women

Tregouet et al (2004) Two UK-based populations No association with ApoAI Yamakawa-Kobayashi et al (2004) Healthy Japanese young individuals No association with HDL-C, ApoAI, TG M883I Coding Wang et al (2000) Inuits, general population Higher HDL-C in MM homozygotes

Brousseau et al (2001) US case-control study Higher frequency of M allele in cases compared to controls

No association with CAD endpoints in cases.

M carriers have a 9% reduction in TG in cases, but no such association in controls.

Clee et al (2001) Dutch CAD patients, cohort design Increased CAD severity in MM compared to II homozygotes Harada et al (2003) Japanese CAD patients Higher HDL-C in M carriers (omitting patients on lipid-lowering

medication)

No association with CAD susceptibility Kakko et al (2003) Finnish population-based study Modest association with higher HDL-C in women Cohen et al (2004) US whites and blacks, general population Modest association with higher HDL-C in black and white men

homozygous for MM compared to II individuals.

Frikke-Schmidt et al (2004) Danish general population No association with HDL-C Shioji et al (2004) 3 large Japanese cohorts No association with HDL-C Tregouet et al (2004) Two UK-based populations No association with ApoAI Yamakawa-Kobayashi et al (2004) Healthy Japanese young individuals No association with HDL-C, ApoAI, TG R1587K Coding Clee et al (2001) Dutch CAD patients, cohort design Lower HDL-C in dose-dependent manner, independent of age, BMI,

smoking and TG

No effect on CAD Cohen et al (2004) US whites and blacks, general population No association with HDL-C Frikke-Schmidt et al (2004) Danish general population Stepwise decrease in HDL-C and ApoAI over time, especially in

to scale Promoter SNP (green), missense SNPs (red), UTR SNPs (blue), intronic SNP (black)

6.2 Results

6.2.1 Characteristics of Study Subjects

The demographic features and plasma lipid attributes of the association study subjects sampled from the Singapore Chinese, Malay and Indian populations are summarized in Table 6.2 Due to the small numbers of females recruited, only data from male subjects were analyzed

6.2.1.1 Age

Exploratory analyses indicated that age distributions for each sample followed a normal distribution, therefore parametric tests were used to examine age differences between groups Cases were significantly older than controls in individual analyses for each ethnic

population analysis (P<0.0005, Table 6.2), reflecting the difficulty in recruiting

age-matched controls A wider age disparity between cases and controls was observed for Malays compared to Chinese and Indians Malay cases (mean age 57.95 ± 9.21 years) averaged ~18 years older than their controls (mean age 39.70 ± 9.91 years) whereas Chinese (mean age 58.66 ± 8.57 years) and Indian cases (mean age 58.13 ± 9.78 years) averaged ~11 years younger than their respective controls (mean ages 47.15 ± 14.61 years and 46.89 ± 14.15 years respectively) Because the age profiles of cases were

comparable in all three ethnicities (ANCOVA, P=0.660), the wider age gap between Malay

cases and controls is attributed to the controls who averaged seven years younger than their Chinese and Indian counterparts

6.2.1.2 Body-Mass Index (BMI)

The BMI distributions in all samples did not show severe departures from normality Chinese and Indian cases possessed significantly lower mean BMI compared to their controls although the magnitude of the mean difference was slight (Chinese cases, mean

independent t-test, P<0.0005; Indians cases, mean BMI 25.03 ± 3.11 kg/m2, Indians

comparison, the BMI differences between cases and controls for the Malay and Indian populations were not longer evident after controlling for age

Comparing between cases of the various ethnicities, no difference in the mean BMI

was found (ANCOVA, P=0.166) However, a difference was detected between controls of the various races (ANCOVA, P<0.0005) Posthoc tests indicated that the mean BMI of

Chinese controls was significantly lower than their Malay (posthoc Fisher’s Least

Significant Difference (LSD) test, P<0.0005) and Indian counterparts (posthoc LSD test,

comparable (posthoc LSD test, P=0.338)

6.2.1.3 Smoking History

Smoking history of subjects was scored as a dichotomous variable in which non- and smokers were grouped into a single category whereas current smokers comprised the second The numbers of individuals in these two categories are expressed as percentage smokers in Table 6.2

ex-To estimate the CAD risk associated with smoking, logistic regression was performed using CAD status as the dependent variable and smoking status as the explanatory variable Smoking was more frequent among cases compared to controls across all ethnicities (Table 6.2) The risks of CAD among smokers were approximately

6.5- (OR 6.49, 95% CI 4.51-9.34, P<0.0005), 6- (OR 6.09, 95% CI 3.44-10.8, P<0.0005), and 17-fold (OR 16.7, 95% CI 7.10-39.2, P<0.0005) higher than non-smokers in Chinese,

Malays and Indians, respectively Since univariable analyses had indicated significant

differences in age and BMI between cases and controls, the logistic regression was repeated with age and BMI entered as additional covariates BMI was no longer a significant confounder in the multivariable model and was therefore excluded from the final fitted model Age-adjusted risks of CAD among smokers compared to non-smokers

remained significantly high with ORs of 4.29 (95% CI 2.86-6.42, P<0.0005), 2.89 (95% CI 1.29-6.49, P=0.010), and 13.25 (95% CI 5.46-32.14, P<0.0005) in Chinese, Malays and

Indians, respectively No interaction between age and smoking was detected, suggesting that age does not modify the association between smoking and CAD

To evaluate smoking habits between ethnic groups, we also applied logistic regression analysis using smoking as the dependent variable and race as the explanatory variable of interest for controls and cases separately Smoking habits appeared similar

between Chinese and Malay controls (OR 1.29, 95% CI 0.72-2.30, P=0.387) whereas

cigarette usage was most rarely observed for Indians (Indian vs Chinese controls, OR

0.26, 95% CI 0.11-0.62, P=0.003; Indian vs Malay controls, OR 0.20, 95% CI 0.08-0.51,

Malay cases, OR 1.21, 95% CI 0.85-1.73, P=0.293; Indian vs Chinese cases, OR 0.67, 95% CI 0.49-0.92, P=0.013; Indian vs Malay cases, OR 0.55, 95% CI 0.36-0.85, P=0.007)

All these observations remained valid even after excluding the potential confounding effect

of age

6.2.1.4 Plasma Lipid and Lipoprotein Levels

Distributions of TG and lipoprotein(a) (Lp(a)) levels were skewed and therefore a natural logarithmic transformation was performed prior to statistical analysis All other lipid and lipoprotein traits displayed normal distributions

Pairwise correlations between lipid variables among controls were examined graphically as well as summarized with Pearson correlation coefficients in Figures 6.2 to

B (ApoB) levels in every ethnic group, with 82-84% of the variation in TC attributed LDL-C (and vice versa), and 60-70% of the variation in ApoB attributed to LDL-C (and vice versa) These strong inter-relationships are related to the fact that the cholesterol content of LDL

is the primary contributor (60-70%) to TC measurements, and also LDL possesses ApoB

as its major structural protein component; additionally, ApoB synthesis is correlated with LDL-C levels (Kesaniemi and Grundy, 1982) A strong linear correlation was also observed between HDL-C and its major apolipoprotein, ApoAI, with 35-53% of the total variation attributed to either variable, depending on the ethnicity The next best trends that were observed consistently across all ethnic samples include the highly significant Pearson correlation coefficients between TG and HDL-C, TG and TC as well as LDL-C However, these correlations were non-linear as illustrated in the bivariate scatterplots (Figures 6.2-6.4)

Contrasts in crude (unadjusted) mean lipid levels between cases and controls are presented in Table 6.2 Contrary to expectations, cases displayed lower mean levels of

TC, TG, LDL-C and ApoB than their controls in each ethnic group Even after adjusting for potential confounders of age, BMI and smoking in multivariable linear regression analysis, these differences in mean lipid levels between cases and controls remained valid These lipid trends are secondary to the fact that many of the cases had already begun lipid-lowering therapy before enrollment into this study Statins represent the first-line drug therapy for treatment of dyslipidemia due to their efficacy in lowering LDL-C and good tolerability (Gaw, 2003) Furthermore, adoption of a healthier lifestyle such as diet modification, increased physical exercise and moderation in alcohol intake, might have also contributed to the lowering of TC, TG, LDL-C and ApoB levels in cases On the other hand, the levels of HDL-C and its major protein moiety, ApoAI, as well as Lp(a) were consistent with the expectations: HDL-C and ApoAI were significantly depressed and Lp(a)

was elevated among cases compared to controls (independent t-tests, P<0.0005) The

observed trends in HDL-C and ApoAI are concordant with the finding from the Framingham Offspring Study that showed 29% of established CAD patients had low HDL-

C without elevated LDL-C (Genest et al., 1992) Lp(a) is a known cardiovascular risk factor whose levels are regulated genetically (by apolipoprotein(a) synthetic rate) and affected to

a minor extent by age, sex and environmental factors (Kronenberg et al., 1999; Berglund and Ramakrishnan, 2004) Therefore the observed elevation in Lp(a) levels among cases relative to controls is not surprising

To explore the variability in mean levels of lipids among controls of various ethnicities, multivariable linear regression analysis was performed using the lipid trait as a dependent variable, and race, age, BMI and smoking as covariates (Table 6.3) Malay controls possessed significantly higher mean levels of TC, LDL-C and ApoB, as well as lower ApoAI compared to Chinese Compared to Chinese, Indians had a more atherogenic profile of higher LDL-C, ApoB, Lp(a) together with lower HDL-C and ApoAI Malay controls showed significantly higher LDL-C but lower Lp(a) compared to Indians Thus, ethnic variations in lipid profiles are evident, with Chinese male controls possessing an overall more favourable lipid profile compared to Malays and Indians It was not meaningful to conduct the same analysis for the cases since their lipid levels might not reflect baseline values for reasons stated previously

P values shown here are not adjusted for other covariates

d Comparisons of mean lipid levels between CAD and control groups are based on independent t-tests

e Comparisons are based on logistic regression with CAD status entered as the dependent variable and smoking as the covariate of interest

AGE BMI TC LDL ApoB TG HDL ApoAI Lp(a)

AGE BMI TC LDL ApoB TG HDL ApoAI Lp(a)

Figure 6.3 Bivariate correlations between quantitative variables in Malay controls Matrix

scatterplots between pairs of variables as well as the corresponding Pearson

correlations are shown

AGE BMI TC LDL ApoB TG HDL ApoAI Lp(a)

Figure 6.4 Bivariate correlations between quantitative variables in Indian controls

Matrix scatterplots between pairs of variables as well as the corresponding Pearson

Table 6.3 Comparison of mean lipid levels between controls of different ethnicities after adjusting for age, BMI and smoking using linear regression analysis

Predictor Variables Entered in Linear Regression Model Dependent

variable

Linear regression

Malays

vs Chinese

Indians

vs Chinese

6.2.2 Single-Locus Associations with CAD

In the following subsections, the associations of each SNP with CAD are described Logistic regression was used in which the binary trait CAD was specified as the dependent variable, and the SNP of interest as the covariate Analysis was conducted for two genetic risk models: (i) the additive model in which the SNP covariate is scored based on the number of allele of interest (0, 1 or 2) carried by an individual, and (ii) the general genotype model in which the individual effects of one of the homozygous and heterozygous genotypes are described relative to a pre-defined homozygous reference category These genotype-based methods have the advantage over conventional 2x2 allele-based tests in that they do not require the assumption of Hardy-Weinberg equilibrium (HWE; Sasieni, 1999) Because CAD is a late-onset disease and the controls were on average younger than the cases (Table 6.2), parallel analysis was also implemented using age as a non-genetic covariate

6.2.2.1 -14C>T

-14C>T causes a C to T transition change in the ABCA1 proximal promoter and is located

14 bases upstream of the transcriptional start site (Santamarina-Fojo et al., 2000) Although this SNP does not disrupt known or putative transcription factor binding sites, it resides in a phylogenetically conserved segment and may potentially effect allele-specific changes in gene expression A BsmA1 restriction site is created in the presence of the -14T allele and therefore genotypes at this locus could be readily discerned by simple restriction fragment length analysis (Figure 4.3)

Allele and genotype frequencies for -14C>T are summarized in Table 6.4 A total of

222 Chinese controls, 491 Chinese cases, 126 Malay controls, 110 Malay cases, 211 Indian controls and 159 Indian cases were analyzed Considering only control samples,

in Malays, with no significant differences among the three ethnic groups (P>0.05, Table

6.4) These -14C allele frequencies for the three diverse Singapore populations are within the estimates of 64-65% reported for two UK Caucasian samples by Tregouet et al (2004), but deviate significantly from the frequency of 86% estimated for a cohort of Dutch Caucasians with proven CAD (Zwarts et al., 2002)

Except Chinese controls and Indian cases, the remaining groups showed severe deviation from HWE in the direction of excess CT heterozygotes and a deficit of the less common TT homozygotes (Table 6.4) Even after a Bonferroni correction for multiple testing was applied, severe deviation from HWE remained for both Malay cases (exact

explanations for the departure from HWE were sought True deviations from HWE can due

to various genetic (natural selection, mutation, drift) or population evolutionary phenomenon (gene flow, inbreeding, population structure and admixture), sampling error and genotyping artifacts In the case of -14C>T, departure from HWE was evident even in certain control samples The large sample sizes ruled out the likelihood of sampling artifact To verify the RFLP genotypes, a smaller PCR fragment was resequenced for a random set of samples and we obtained complete concordance in genotype calls Allele dropout due to a polymorphism within the original PCR-RFLP primers was also an unlikely explanation because our earlier promoter SNP survey (Chapter 5) as well as many others

to date (Zwarts et al., 2001; Lutucuta et al., 2001; Probst et al., 2004; Shioji et al., 2004; Tregouet et al., 2004) have not identified any polymorphisms within the PCR primers used

in the -14C>T RFLP assay Also, allele dropout would be expected to create an excess of homozygotes and deficit of heterozygotes, opposite to what we had observed Finally, to determine whether the pair of PCR-RFLP primers was amplifying a homologous segment

of the human genome, a reverse electronic PCR was performed against the human consensus genome assembly (http://www.ncbi.nlm.nih.gov/projects/e-pcr/) The maximum

relaxed stringency of two gaps and two mismatches was specified No other hit besides the fragment of interest was returned, indicating specificity of the PCR

Based on the numbers genotyped and control allele frequencies, the estimated power to detect a crude allelic OR of 1.5 at a significance level of 5% was approximately 90%, 50% and 71% for the Chinese, Malay and Indian case-control groups respectively

To detect an association with a larger crude allelic OR of two, all case-control datasets had powers of at least 90% (Table 6.4)

For the Indian case-control dataset, we found evidence of a moderately elevated CAD risk associated with the rare -14T allele (Table 6.5) Evidence based on the general genotype model supported a recessive effect for the -14T allele: the heterozygous OR was

not significantly different from unity (OR 1.29, 95% CI 0.81-2.04, P=0.280; age-adjusted

OR 0.93, 95% CI 0.55-1.56, P=0.781) whereas the homozygous OR was significant, with

TT Indians possessing at least two-fold greater risk for CAD relative to CC individuals (OR

2.93, 95% CI 1.43-6.00, P=0.003; age-adjusted OR 2.70, 95% CI 1.22-5.95, P=0.007)

The additive genotype model was also significant before (OR 1.58, 95% CI 1.13-2.19,

interaction between age and the -14C>T locus was not found

No association of -14C>T with CAD was revealed for the Chinese population For the Malays, the age-adjusted general genotype model found a marginal recessive allele effect, with TT individuals at five-fold greater risk for CAD compared to the CC individuals

(age-adjusted OR 5.03, 95% CI 0.78-32.4, P=0.090) However, the wide confidence

interval of the recessive OR suggests that the association detected from the general genotype model might be unreliable Moreover, the additive model was not statistically significant

6.2.2.2 237indelG

237indelG is a one-base insertion/deletion (indel) polymorphism of a G nucleotide found within the untranslated exon 2 of the ABCA1 gene It was first described by Pullinger et al (2000) while mapping the transcriptional start site of the gene The two alleles of the SNP are denoted as 2g and 3g, with the latter indicating the insertion of the G nucleotide which creates a Bsl1 restriction enzyme (Figure 4.4)

237indel genotypes were successfully obtained for 248 Chinese controls, 517 Chinese cases, 176 Malay controls, 113 Malay cases, 231 Indian controls and 160 Indian cases (Table 6.4) Controls of Chinese and Malay ethnicities shared a similar 2g allele

frequency of 84% (P=0.924), which was significantly higher than the frequency of 70% in Indian controls (P<0.0005, Table 6.4) A higher 2g allele frequency of 91.5% had been

reported in Dutch CAD patients (Zwarts et al., 2002) Genotypes for all local samples

conformed with Hardy-Weinberg proportions (P>0.05, Table 6.4)

Given the observed control allele frequencies, the Chinese and Indian case-control datasets had 71.5% and 68% powers, respectively, to detect a crude allelic OR of 1.5 at a significance level of 5%, whereas the power was lower at 35% for Malays For a higher effect size of two, the sample sizes should have at least 73% power

The unadjusted general genotype test yielded a borderline protective effect of the

237indelG SNP in the Indian case-control dataset (P=0.091), with a crude heterozygous

OR of 0.69 (95% CI 0.44-1.06) relative to 2g2g genotype (Table 6.5) However, this result

was invalidated by an insignificant P value of 0.217 in the corresponding age-adjusted test

as well as a lack of detectable risk for the 3g3g genotype No further evidence of an association was found for either the crude and age-adjusted additive models The Chinese and Malay populations showed no evidence of an association with CAD at the 237indelG locus (Table 6.5)

6.2.2.3 R219K

R219K is a G to A transition in exon 7 that results in an arginine to lysine substitution at amino acid residue 219 in the ABCA1 protein Structurally the R219K polymorphism is located in the large extracellular loop between the first and second transmembrane spanning helices (Figure 5.12) The missense cSNP was first identified using a cDNA-based resequencing approach to examine genetic variation in the ABCA1 gene and was found to exist at a high frequency of ~50% in a white population sample (Wang et al., 2000) A RFLP assay using the EcoN1 restriction enzyme was used to distinguish between the two alleles at the R219K locus (Figure 4.5)

DNA samples from 136 Chinese controls, 152 Chinese cases, 120 Malay controls,

53 Malay cases, 102 Indian controls and 51 Indian cases were successfully genotyped at the R219K locus (Table 6.4) Allele frequencies were constant across all ethnic groups

(P>0.05, Table 6.4), with R219 being the common allele at a frequency of 62-66% In most

studies involving Caucasians, a comparable R219K allele frequency spectrum was detected (Clee et al., 2001; Cenarro et al., 2003; Evans et al., 2003; Srinivasan et al., 2003) albeit with a couple of exceptions (Kakko et al., 2003; Knoblauch et al., 2004) Somewhat different frequencies in Japanese (51%, Harada et al., 2003) and blacks (40%, Srinivasan et al., 2003) have been reported All samples observed in the present study

complied with the expected Hardy-Weinberg proportions (P>0.05, Table 6.4)

No statistical evidence of an association of R219K with CAD risk was detected in

any case-control datasets (P>0.05, Table 6.5) Retrospective calculations showed that for

an effect size of a crude allelic OR of 1.5 at the 5% significance level, the powers of the Chinese, Malay and Indian case-control datasets were 62.8%, 35.6% and 34.1%, respectively; and for a larger OR of two, the datasets had larger powers at 96.7% 75.8% and 69.5%, respectively

6.2.2.4 V825I

V825I encodes for G to A transition polymorphism in exon 17, resulting in a valine to leucine substitution at amino acid position 825 in the ABCA1 protein It lies in the sixth transmembrane helix location of the hypothetical topological model of the transporter protein (Figure 5.12) In the presence of the I825 allele, a DpnII restriction enzyme site is created and therefore genotypes at this locus can be readily discerned using a simple PCR-RFLP assay (Figure 4.6)

Allele and genotype distributions for V825I are summarized in Table 6.4 A total of

108 Chinese controls, 295 Chinese cases, 123 Malay controls, 77 Malay cases, 109 Indian controls and 83 Indian cases were analyzed for the V825I variant V825 allele distributions varied greatly between ethnic groups, with frequencies of 58%, 72% and 95% in the

Chinese, Malay and Indian control samples respectively (P<0.05, Table 6.4) Wang and

co-workers (2000) also encountered a wide range of frequencies across various ethnic populations, ranging from 49% in Chinese to 97% in African-Americans The V825 allele frequency in our Indian population is comparable to Caucasians (Clee et al., 2001; Tregouet et al., 2004)

Genotypes at the V825I showed no significant deviations from HWE (exact P>0.05,

Table 6.4) Power calculations showed that the Indian case-control dataset was extremely weakly powered (23.7%) to detect even a large allelic OR of two at the 5% significance level whereas the powers for the Chinese and Malay datasets were estimated at least 78% For detecting a smaller crude allelic OR of 1.5, the powers for the Chinese, Malay and Indian case-control datasets were 69.6%, 38.4% and 11.8%, respectively Varying sample sizes and allele frequencies account for heterogeneity in power

Examination of V825I genotype distributions between cases and controls found no

formal statistical support of an association with CAD in any ethnic group (P>0.05, Table

6.5)

6.2.2.5 M883I

The M883I polymorphism is a G to A transition in exon 18, causing a methionine to isoleucine substitution at amino acid position 883 in the ABCA1 protein The substitution is localized in the putative phosphorylation site of cAMP/cGMP-dependent protein kinase, approximately 50 amino acids upstream of the Walker A motif in the N-terminal ATP-binding cassette (Bodzioch et al., 1999; Figure 5.12) By introducing an artificial one-base mismatch in one of the PCR primers, M883I genotypes can be readily discerned using the Bsm1 restriction enzyme (Figure 4.7)

Genotypes were available from 250 Chinese controls, 364 Chinese cases, 167 Malay controls, 100 Malay cases, 223 Indian controls and 153 Indian cases (Table 6.4) Like the V825I polymorphism which lies just ~1.3 kb upstream (Figure 6.1), M883 allele frequencies varied in an ethnic-specific manner (all pairwise ethnic sample comparisons,

(66%) and lowest in Indian controls (9%) while Malay controls had an intermediate frequency (39%) A diverse range of M883 allele frequencies in different populations has been previously known (Wang et al., 2000; Clee et al., 2000; Harada et al., 2003; Kakko et al., 2003; Knoblauch et al., 2004; Tregouet et al., 2004) In the Chinese control samples, M883I genotypes did not follow the expected Hardy-Weinberg proportions (exact

departures from HWE (exact P>0.05, Table 6.4) It was estimated that the Chinese

case-control dataset had greater than 89% power to detect an allelic OR of at least 1.5 at a 5% significance level The Malay case-control dataset was estimated to have greater than 61% power to detect a minimum OR of 1.5 The number of Indian cases and controls genotyped had the weakest power (40.5%) to detect a modest crude allelic OR of 1.5, but

was otherwise reasonably powered (86.3%) for a larger OR of two, assuming a significance level of 5%

Association test results for CAD at the M883I locus are presented in Table 6.5 In Malays, a recessive effect was associated with the M883 homozygotes (crude OR 2.36,

95% CI 1.15-4.87, P=0.020, Table 6.5) However, this association might be spurious and

unrelated to a genetic effect because the corresponding age-adjusted genetic risk attributed to the MM genotype failed to reach statistical significance (age-adjusted OR

2.22, 95% CI 0.78-6.28, P=0.133) Similarly, the additive test detected a significant association with the M883 allele (crude OR 1.47, 95% CI 1.03-2.11, P=0.036) but this

result was rendered non-significant after adjustment for age (age adjusted OR 1.43, 95%

CI 0.86-2.40, P=0.168)

In Indian, a borderline recessive effect of the M883 allele emerged following correction for age (Table 6.5) The age-adjusted general genotype model showed that the heterozygous OR was not significantly different from unity (age-adjusted OR 1.01, 95% CI

0.52-1.96, P=0.970) whereas the relative risk of MM genotype was marginally larger adjusted OR 4.18, 95% CI 0.93-18.7, P=0.062), consistent with a recessive effect

(However, in view of the wide confidence intervals (95% CI 0.93-18.7) and marginal

age-adjusted P values (P=0.062), the genetic risk attributable to M883I in Indians might be

limited The additive genotype model did not generate a significant result

No association of the M883I locus with CAD was found in Chinese

6.2.2.6 IVS44+18T>C

IVS44+18T>C is a T to C transition polymorphism located within intron 44 of the ABCA1 gene, and was discovered during the DHPLC survey (Figure 5.13) The polymorphism is located 18 bases from the 5’ splice site of exon 44 and is not known to disrupt consensus splice signals Human-mouse comparison found a lack of sequence conservation here although this does not necessarily exclude possibility of a lineage-specific functional effect IVS44+18T>C genotypes were scored using the HaeIII restriction enzyme (Figure 4.8)

IVS44+18T>C genotypes were obtained for 144 Chinese controls, 336 Chinese cases, 109 Malay controls, 76 Malay cases, 100 Indian controls and 92 Indians (Table 6.4) Chinese and Malay controls shared a similar IVS44+18T allele frequency of 57%

(P=0.928, Table 6.4), but the frequency (78%) in Indian controls was quite different (Indian

vs Chinese controls, and Indian vs Malay controls, P<0.0005, Table 6.4) The observed genotype counts were consistent with Hardy-Weinberg proportions (P>0.05, Table 6.4) It

was estimated that for the Malay and Indian case-control datasets, the sample sizes possessed low powers of 46.3% and 34.1%, respectively, to detect an allelic OR of 1.5 at the 5% significance level whereas the Chinese case-control dataset had a power of 79% For a larger allelic OR of two, the power would exceed 71% for Indians and 85% for Chinese as well as Malays

A consistent association of the IVS44+18T>C locus with CAD was identified in the Indian population, both with or without adjusting for age (Table 6.5) Based on the general genotype model, the finding of a non-significant heterozygous relative risk (OR 1.48, 95%

CI 0.80-2.70, P=0.209; age-adjusted OR 1.78, 95% CI 0.82-3.86, P=0.143) but a significant homozygous relative risk (OR 3.35, 95% CI 1.10-10.2, P=0.034; age-adjusted

OR 5.68 95% CI 1.40-23.0, P=0.015) indicated that the IVS44+18C allele confers a

confounding effect of age was removed seems to suggest a strong genetic risk The additive tests also confirmed a significantly higher risk attributed to the IVS44+18C allele

did not result in a significant likelihood ratio test, suggesting that the effects of age and

genotype on CAD risk are independent

Suggestive evidence of an association of IVS44+18T>C with CAD in the Malay population emerged upon removing the effect of age (Table 6.5) A non-significant

heterozygous (age-adjusted OR 2.49, 95% CI 0.79-7.87, P=0.119) but an otherwise significant homozygous (age-adjusted OR 4.15, 95% CI 1.01-17.0, P=0.049) relative risks

from the age-adjusted general genotype model implicated a recessive risk for the IVS44+18C allele The age-adjusted additive genotype model also detected an increased

risk effect in the Malay population (age-adjusted OR 2.06, 95% CI 1.02-4.14, P=0.044,

Table 6.5) No statistical support of an interaction between age and genotype was evident

6.2.2.7 8995A>G

8995A>G is an A to G transition polymorphism located in the untranslated exon 50 of the

ABCA1 gene, and was initially discovered from an in silico analysis of ABCA1 expressed

sequence tags (ESTs) and mRNA (Figure 5.1) The two allelic variants are not expected to cause any major structural changes in the 3’ portion of the ABCA1 mRNA based on RNA folding analysis (Figure 5.4) 8995A>G genotypes were assessed using a PCR-SSCP assay (Figure 5.3)

272 Chinese controls, 516 Chinese cases, 179 Malay controls, 112 Malay cases,

233 Indian controls, and 166 Indian cases were successfully scored at the 8995A>G locus (Table 6.4) Like the -14C>T and R219K polymorphisms, 8995A>G allele frequencies

among controls of the three ethnic groups were uniform at ~84-85% (P>0.05, Table 6.4)

These frequencies are close to the 80% described in a Caucasian sample (Knoblauch et

al., 2004) HWE tests were non-significant (exact P>0.05, Table 6.4) Given the numbers

genotyped and the ability to detect a minimum allelic OR of 1.5 at 5% significance level, the Chinese case-control dataset possessed the highest power (71%) compared to the Malay (36%) and Indian (48%) datasets For a larger effect size of two, all three case-control datasets should have at least 74% power

The minor 8995G allele appeared to confer a modest recessive risk for CAD in the Chinese population (Table 6.5) By implementing the general genotype model, Chinese with the heterozygous AG genotype were not at enhanced risk for CAD relative to those

with the AA genotype (OR 1.14, 95% CI 0.82-1.58, P=0.433; age-adjusted OR 1.21, 95% 0.84-1.76, P=0.302); however, there was a greater risk for those with the GG genotype relative to the AA individuals (OR 2.95, 95% CI 1.00-8.73, P=0.050; age-adjusted OR 3.76, 95% CI 1.18-11.9, P=0.025) A modest difference in allele frequencies between Chinese

cases and controls was also obtained using the age-adjusted additive genotype tests

(age-adjusted OR=1.40, 95% CI 1.02-1.92, P=0.035) The effects of age and locus on CAD risk

were independent as the interaction term between these two variables did not result in a significant change in log likelihood when added to the model composed of main effects

No effect of the 8995A>G polymorphism on CAD risk was found in either Malay or Indian populations (Table 6.5)

6.2.2.8 Summary of Single-Locus Associations with CAD

models To allow one P value to be plotted for each test, the general genotype model

given in Table 6.5 is replaced by modelling specific dominant and recessive genotype

effects Only the smallest P values among the additive, dominant and recessive genotype

models are shown in Figure 6.5 Values above the dotted horizontal line indicate

associations that were significant at the P<0.05 level Because CAD is a late-onset

disease and the younger age of the controls in our study, we will limit interpretation to the age-adjusted association tests

Evidently, no one locus was consistently associated with CAD across all three ethnic groups The strongest associated SNP was -14C>T in Indians which showed a

significant dominant -14T allele effect (age-adjusted P=0.005, Figure 6.5C) However, this

finding should be treated cautiously in view of the severe Hardy-Weinberg disequilibrium at -14C>T (Table 6.4) Another SNP associated with CAD risk in Indians was IVS44+18T>C

for which the additive model gave the lowest P value (age-adjusted P=0.012, Figure 6.5C)

In Malays, an association of IVS44+18T>C was also found (additive model, age-adjusted

effect of the IVS44+18C allele) In Chinese, 8995A>G was associated with CAD (recessive

model, age-adjusted P=0.031; Figure 6.5) However, the modest P values and the multiple

hypothesis tests are conducted at the same significance level of α=5% may create a problem in interpreting these results because the probability of finding at least one spurious result among the tests will exceed α

SNP Sample Genotype 11 a

N Allele 1 Frequency HWE Test

Exact P

Chinese

vs Malays

Chinese

vs Indians

Malays

vs Indians

Genotype codes for the various SNPs:

-14C>T, 11=CC, 12=CT, 22=TT; 237indelG, 11=2g2g, 12=2g3g, 22=3g3g; R219K, 11=RR, 12=RK, 22=KK; V825I, 11=VV, 12=VI, 22=II; M883I, 11=MM, 12=MI, 22=II; IVS44+18T>C, 11=TT, 12=TC, 22=CC; 8995A>G, 11=AA, 12=AG, 22=GG.

b

Power calculations were computed fbased on the assumption of an additive genotype model and had not been unadjusted for covariates.

SNP Genotype risk model OR (95% CI) a P valueb -2LL c OR (95% CI) a P valueb -2LL c OR (95% CI) a P valueb -2LL c

Figure 6.5 Summary of single-locus association analyses with CAD (A) Chinese, (B)

against SNP positions (1=-14C>T, 2=237indelG, 3=R219K, 4=V825I, 5=M883I, 6=IVS44+18T>C, 7=8995A>G) For each SNP, genotype-based tests assuming additive,

A

B

C

6.2.3 Pairwise Linkage Disequilibrium (LD)

Figure 6.6 summarizes the pairwise LD patterns for cases and controls separately in each

two-locus haplotypes using the Expectation-Maximization (EM) algorithm (Excoffier and

Slatkin, 1995) P values for both LD metrics are also indicated in Figure 6.6 but statistical

significance should not be used to describe LD because its dependence on sample size

(Ardlie et al., 2002a) For instance, P values can be highly statistically significant even

when the magnitude of the LD metric is low

(Devlin and Risch, 1995), weak LD extended across the ABCA1 gene for all samples

LD existed between the V825I and M883I loci in four (Chinese controls, Malay cases and both Indian cases and controls) out of the six population samples Yet even in this case,

and M883I compared to other locus pairs is likely to be due to the physical proximity of the two markers which are separated by only 1.3 kb on the genomic map (Figure 6.1), although this reason alone fails to explain why a similar LD was not present in the Chinese cases and Malay controls; the latter observation might be due to sample size differences

appropriate since it is directly related to the amount of information encoded by one marker

the seven ABCA1 SNPs

across the seven ABCA1 SNPs both within and between populations (Figure 6.6A, upper

and are also less precise due to large variances (Ardlie et al., 2002a; Teare et al., 2002)

For ease of interpretation, if a highly significant and large |D’| of at least 0.8 was required

to indicate high LD (Ardlie et al., 2002a), the pattern of LD would tend to be weak across

6.2.4 Multi-Locus Associations with CAD

Besides assessing SNPs individually for association with CAD, we also analyzed SNPs simultaneously Although the pairwise LD analysis and single-locus case-control comparisons indicate that the effects of significant ABCA1 SNPs were largely independent and associations, if any, tended to be modest, a different risk effect may emerge when multiple loci are considered jointly Two types of multi-locus analysis were used: those encoded by unphased genotypes (also known as locus-based test) and haplotypes The

former does not consider cis effects of the alleles while the latter does The locus-based

test selects only important genetic risk factors in the model whereas haplotype-based methods can simultaneously test all genetic risk factors and in essence tests for simultaneous interaction between alleles On the other hand, locus-based methods do not require resolution of haplotype phase and can be carried out using standard statistical packages For multi-locus association analysis with CAD, only individuals with complete genotype data were analyzed The datasets consisted of: Chinese, 69 controls and 93 cases; Malays, 65 controls and 41 cases; and Indians, 73 controls and 29 cases Comparisons between these selected and unselected individuals (i.e those missing genotype data at one or more loci) showed no differences in age, smoking, ApoAI and ApoB levels for five out of the seven ABCA1 loci (237indelG, R219K, V825I, IVS44+18T>C and 8995A>G) Single-locus association analyses for these trimmed datasets were not conducted due to an inherent reduction in power associated with small samples

Chinese CAD Chinese Controls

Figure 6.6 Pairwise LD of ABCA1 SNPs genotyped in case-control association study

according to the colour key provided (B) Statistical significance of LD values is

diagonal SNPs 1, 2, 3, 4, 5, 6 and 7 denote -14C>T, 237indelG, R219K, V825I, M883I, IVS44+18T>C and 8995A>G respectively, ordered by their genomic locations

*P<0.05, **P<0.005, ***P<0.0005

A

6.2.4.1 Multi-Locus Associations with CAD Using Unphased Genotypes

Locus-based multi-locus associations were examined by standard logistic regression The relationship between the CAD dependent variable and loci takes the form of a generalized linear model in which alleles of each locus are assumed to act additively in the linear predictor Other potential non-genetic risk factors considered in the model included age, BMI and smoking A backward stepwise strategy was used to select variables in the

preliminary main effects model, with P values for entry and removal of variable set at 0.20

and 0.25 respectively Next, first-order interaction terms were added one at a time to the preliminary main effects model and their effects assessed by the log likelihood ratio test Multivariable models were fitted separately for each race

Table 6.6 provides the details of the multivariable logistic regression model fitted for the Chinese case-control dataset Hosmer and Lemeshow’s goodness of fit summary

statistic indicates a reasonable fit (P=0.307) and the classification table shows a specificity

of 74.6% and sensitivity of 85.1% Non-genetic confounders for CAD identified in the multivariable model included age, smoking and BMI Age and smoking showed positive associations with CAD, and BMI showed a significant albeit small negative association with CAD (Table 6.6), in line with the results from the univariable analyses (Table 6.2) Two loci towards the 3’ end of the gene were significantly related to CAD outcome The M883I

locus was positively associated with CAD (adjusted OR 1.86, 95% CI 1.02-3.39, P=0.044),

whereas IVS44+18T>C showed an inverse correlation (adjusted OR 0.54, 95% CI

0.30-0.95, P=0.034); the associated alleles were M883 in the positive (risk) direction and

IVS44+18C in the negative (protective) direction 8995A>G was also selected in the logistic model but its effect was not statistically significant (adjusted OR 1.77, 95% CI 0.81-

3.90, P=0.16) These genetic main effects fitted in the model coincide with the best three single-locus association test results (age-adjusted P<0.10) in Figure 6.5A No first-order

interactions between the main effects were detected, including those between genetic risk factors, consistent with the lack of LD

The parameters of the best fitted model for the Malay case-control dataset are given in Table 6.7 As expected, age was positively related to the CAD outcome (adjusted

IVS44+18C allele was associated with an enhanced CAD risk (adjusted OR 3.43, 95% CI

1.13-10.41, P=0.03) On the other hand, 237indelG (adjusted OR 0.27, 95% CI 0.06-1.20,

evidence of associations, with the 3g and 8995G alleles demonstrating a protective trend Single-locus association tests only found support of an effect at the IVS44+18T>C locus (Figure 6.5B) No first-order interactions were found therefore the effects of the covariates

in the multivariable model were independent of one another

In Indians, the variables in the best fitting model included four genetic risk factors and the three non-genetic covariates (Table 6.8) Age and smoking were positively associated with CAD outcome, whereas BMI showed an inverse association Of the four genetic risk factors in the model, only M883I and IVS44+18T>C showed significant

coefficients of P values less than 0.05 Both M883 (adjusted OR 46.08, 95% CI 1.94-1094,

strongly associated with increased CAD At -14C>T (adjusted OR 3.06, 95% CI

0.61-15.42, P=0.176) and 8995A>G (adjusted OR 0.15, 95% CI 0.01-2.78, P=0.20), there were

non-significant opposing trends towards positive and inverse associations with CAD respectively Previously, based on age-adjusted individual locus effects, the best statistical support was found for -14C>T, IVS44+18T>C and M883I, in order of decreasing statistical significance (Figure 6.5C) The lack of any significant first-order interaction term indicates that the effects of the variables were largely independent of one another

Table 6.6 Multi-locus association with CAD using unphased ABCA1 SNP

genotypes and other covariates in Chinese Variables are selected by a

backward stepwise strategy with P values for entry and removal set at 0.20 and

0.25 respectively An additive genotype model is assumed for the ABCA1

Table 6.7 Multi-locus association with CAD using unphased ABCA1 SNP

genotypes and other covariates in Malays Variables are selected by a

backward stepwise strategy with P values for entry and removal set at 0.20 and

0.25 respectively An additive genotype model is assumed for the ABCA1

Table 6.8 Multi-locus association with CAD using unphased ABCA1 SNP

genotypes and other covariates in Indians Variables are selected by a backward

stepwise strategy with P values for entry and removal set at 0.20 and 0.25

respectively An additive genotype model is assumed for the ABCA1 SNPs

6.2.4.2 Multi-Locus Associations with CAD Using Haplotypes

Analysis of haplotype association was carried out in the the HaploStats package HaploStats allows simultaneous adjustment for environmental covariates and accounts for the additional variance in the test statistic caused by haplotype uncertainty (Schaid et al., 2002; Lake et al., 2003) Haplotypes were inferred using the EM algorithm (Excoffier and Slatkin, 1995) Under the null hypothesis of no difference in haplotype frequencies for the trait of interest between cases and controls, data from both groups can be pooled to yield a better estimate of the haplotype frequencies than would be obtained by, say, using data from controls only Haplotype pairs were assigned to each subject and the trait was then regressed on the expected haplotype counts The expected counts are computed by the posterior probabilities of the haplotype pairs given the subject’s observed genotypes and assuming null association of the trait with haplotypes (Schaid et al., 2002)

In the first stage of haplotype association analysis, global score tests of association using six-locus haplotypes were performed using the haplo.score routine in HaploStats For this analysis, -14C>T was omitted because the observed excessive heterozygosity may increase the phase ambiguity during the EM estimation of haplotypes (Fallin and Schork, 2000) Increased phase uncertainty in turn can inflate Type I error rates with the prospective regression analysis (Lake et al., 2003)

The total number of haplotypes enumerated was 35, 32 and 27 for the Chinese, Malay and Indian populations (cases and controls combined) respectively (Figure 6.7A) The spectrum of haplotype distributions in all populations was skewed towards a large excess of rare haplotypes as reflected by low mean (2.9-3.7%) and median (1.2-1.7%) haplotype frequencies This observation is in line with the lack of LD (Figure 6.6) Relatively few common haplotypes (~5-6) with minimum frequencies of 5% were found for each ethnic population Together, these common haplotypes formed the bulk (Chinese, 52.4%; Malays, 51.9% Indians, 73.9%) of all chromosomes The most common haplotype