|Home | About | Journals | Submit | Contact Us | Français|
Age-related macular degeneration (AMD) is the most common cause of incurable visual impairment in high-income countries. Previous studies report inconsistent associations between AMD and apolipoprotein E (APOE), a lipid transport protein involved in low-density cholesterol modulation. Potential interaction between APOE and sex, and smoking status, has been reported. We present a pooled analysis (n=21,160) demonstrating associations between late AMD and APOε4 (OR=0.72 per haplotype; CI: 0.65–0.74; P=4.41×10−11) and APOε2 (OR=1.83 for homozygote carriers; CI: 1.04–3.23; P=0.04), following adjustment for age-group and sex within each study and smoking status. No evidence of interaction between APOE and sex or smoking was found. Ever smokers had significant increased risk relative to never smokers for both neovascular (OR=1.54; CI: 1.38–1.72; P=2.8×10−15) and atrophic (OR=1.38; CI: 1.18–1.61; P=3.37×10−5) AMD but not early AMD (OR=0.94; CI: 0.86–1.03; P=0.16), implicating smoking as a major contributing factor to disease progression from early signs to the visually disabling late forms. Extended haplotype analysis incorporating rs405509 did not identify additional risks beyondε2 and ε4 haplotypes. Our expanded analysis substantially improves our understanding of the association between the APOE locus and AMD. It further provides evidence supporting the role of cholesterol modulation, and low-density cholesterol specifically, in AMD disease etiology.
Age-related macular degeneration (AMD; MIM# 603075) is in its late form, the leading cause of incurable visual impairment among individuals of European descent over the age of 50 [Centers for Disease Control and Prevention, 2004], accounting for more than half of all new cases of registered blindness [Evans and Wormald, 1996]. The socioeconomic burden associated with AMD is increasing in our aging societies with almost 30% of those aged 75 years and above showing early signs of disease [Klein et al., 1992; Vingerling et al., 1995]. AMD is a common disorder of complex etiology with multiple genetic, environmental and lifestyle factors contributing to the phenotype, although the specifics of the etiology remain largely unresolved [Swaroop et al., 2007]. AMD by definition affects the macular region of the retina which is associated with detailed central vision. AMD is commonly divided into early (eAMD) and late AMD, and the vision impairing late AMD is subdivided into geographic atrophic (GA) and neovascular AMD (NV) or mixed GA and NV together (GANV). eAMD is characterized by drusen formation and pigmentary changes at the level of the retinal pigment epithelium (RPE), and can progress through atrophy of the pigment epithelium to the visually disabling late atrophic and/or neovascular phenotypes.
Advances in our understanding of the genetic basis of the disease have identified risk and protective variants in several genes associated with the complement pathway and chronic inflammation such as factor H (CFH) [Edwards et al., 2005, Hageman et al., 2005; Haines et al., 2005; Hughes et al., 2006; Klein et al., 2005], component 2 (CC2)/factor B (CFB) region [Gold et al., 2006; McKay et al., 2009], component 3 (C3) [Maller et al., 2007; McKay et al., 2010; Park et al., 2009; Yates et al., 2007] and complement factor I (CFI) [Fagerness et al., 2009]. Beyond the complement pathway, chromosome 10q26 and specifically the Age Related Maculopathy Susceptibility 2 (ARMS2) locus has been implicated as a second major genetic contributor to the AMD disease process [Dewan et al., 2006; Jakobsdottir et al., 2005; Riveria et al., 2005; Yang et al., 2006]. The high linkage disequilibrium that exists between ARMS2 and the serine protease HTRA1 makes it difficult to determine the source of the genetic effect at this locus, although data have been reported supporting mitochondrial involvement through interaction with translocase of outer mitochondrial membrane proteins and co-localization to the mitochondrial-rich ellipsoid region of the photoreceptors [Fritsche et al., 2008; Kanda et al., 2007]. Mitochondrial dysfunction and oxidative stress have been implicated in AMD disease etiology with reports of increased mitochondrial damage in the neural retina and RPE with ageing. Reports of mitochondrial genomic variation and associated increased AMD risk [Canter et al., 2008; San Giovanni et al., 2009] provide an excellent rationale for the involvement of oxidative stress in the disease pathway. More recently, TIMP3, a metalloproteinase involved in degradation of the extracellular matrix, hepatic lipase C (LIPC) and cholesterylester transfer protein (CETP), key genes involved in the metabolism of triglycerides and high density lipoproteins (HDL), were implicated in AMD pathogenesis through genome-wide association studies (GWAS) [Chen et al., 2010; Neale et al., 2010].
One of the first reported genetic associations with AMD was the protective effect exerted by the ε4 haplotype of Apolipoprotein E (APOE; MIM# 107741), a lipid transport protein that acts as a ligand for the low density lipoprotein receptor which is involved in the maintenance and repair of neuronal cell membranes [Klaver et al., 1998; Souied et al., 1998]. Association of APOE with AMD pathogenesis is supportive of mechanisms such as immunoregulation and cell signalling [Zarbin, 2004]. Variation at two single nucleotide polymorphisms (SNPs) within the coding sequence of the APOE gene, rs429358 and rs7412, results in different isoforms reported to attenuate binding affinity to the low density lipoprotein (LDL)-receptor. For example, ε2 has a much reduced binding affinity leading to lower total cholesterol levels with respect to ε3 and ε4, which reveal a higher binding affinity with higher total cholesterol levels [Siest et al., 1995]. Three allelic variants derived from these SNPs commonly referred to as ε2, ε3 and ε4 are differentiated on the basis of cysteine (Cys) and arginine (Arg) residue interchanges at positions 112 (rs429358) and 158 (rs7412) in the amino acid sequence and give rise to 6 diplotypes (ε3/ε3,ε3/ε4, ε2/ε3, ε4/ε4, ε2/ε4 andε2/ε2, ranked from most to least common amongst European populations [Corbo and Scacchi, 1999]). The ε2 haplotype has a Cys residue at positions 112 and 158 in the receptor-binding region of APOE. The ε3 haplotype has residues Cys-112 and Arg-158 and the ε4 haplotype has Arg residues at both positions, with these amino acid substitutions having a strong physiological consequence on protein function. ε3 is by far the most common haplotype in all human populations, but considerable variation in the frequency distribution of all haplotypes have been reported across different ethnic groups [Corbo and Scacchi, 1999] and with human longevity [McKay et al., 2011].
Many independent replications of AMD and APOE have been undertaken and while the protective effect associated withε4 was confirmed in some studies, the significance of the increased risk associated withε2 has not been consistent [Baird et al., 2004, 2006; Bergeron-Sawitzke et al., 2009; Bojanowski et al., 2006; Conley et al., 2005; Francis et al., 2009; Fritsche et al., 2009; Schmidt et al., 2002, 2005; Thakkinstian et al., 2006; van Leeuwen et al., 2004; Zareparsi et al., 2004]. Some studies have reported increased risk of AMD associated withε2 [Baird et al., 2004; Schmidt et al., 2002; van Leeuwen et al., 2004; Zareparsi et al., 2004] while others have suggested that effect modification by sex [Baird et al., 2004; Schmidt et al., 2002] or smoking status may occur [Schmidt et al., 2005]. Previously, pooled data and meta analyses in much smaller sample sizes indicated a significant protective effect associated with ε4 and a non-significant increased risk associated with ε2 [Bojanowski et al., 2006; Thakkinstian et al., 2006]. In our investigation, we invited contributing studies to pool individual sample data to assess APOE genotype in AMD with respect to accompanying clinical phenotype data and to test for interaction with sex, age and smoking status.
Initially 18 research groups likely to have APOE genetic data in a well-phenotyped dataset were invited to participate. The analysis originated from 15 of these studies in 40 centers based in 11 countries which had data on APOE in AMD patients and suitable controls: nine European countries (United Kingdom, Germany, the Netherlands, Norway, Estonia, Italy, France, Greece and Spain), six centers in the United States and one in Australia [Augood et al., 2004; Baird et al., 2006; Bergeron-Sawitzke et al., 2009; Conley et al., 2005; Dandekar et al., 2006; Ennis et al., 2008; Francis et al., 2009; Fritsche et al., 2009; Haan et al., 2006; Hadley et al., 2010; McKay et al., 2009; Vingerling et al., 1995; Yates et al., 2007; Zareparsi et al., 2004]. Study demographics are summarized (n=21,160; Table 1). Three studies, the Rotterdam Study [Vingerling et al., 1995], the Women’s Health Initiative (WHI) [Haan et al., 2006] and the EUREYE study [Augood et al., 2004], provided data derived from samples acquired through population based surveys (n=10,253). The remainder of the data originated from case-control (association) studies. Individual participant data for age at examination, sex, smoking status (ever smoker or never smoker), AMD phenotype and APOE haplotype (SNPs rs429358, rs7412) were necessary from participating studies to facilitate a pooled data analysis in order to be eligible for this study. APOE genotypes were determined using several methods: TaqMan genotyping technology (Applied Biosystems, Foster City, CA): Portland, AREDS, Philadelphia, Edinburgh, Women’s Health Initiative, Cambridge, London; APOE-specific polymerase chain reaction followed by restriction digest: Michigan, Rotterdam, Melbourne, Los Angeles, Cambridge; KASPar genotyping (KBioscience, Hertfordshire, UK): EUREYE, Southampton; Sequenom (San Diego, CA): Regensburg; BigDye Terminator DNA Cycle Sequencing analysis (Applied Biosystems): Belfast. Where available, genotypes were also requested for rs405509, a SNP located in the promoter region of APOE (n=7,568). Extended haplotype frequencies were estimated using UNPHASED, where data were available for rs405509 [Dudbridge, 2008].
Control subjects were classified on the basis of direct clinical examination or by retinal imaging and were graded as having either no signs of AMD in either eye, or had fewer than five hard drusen of diameter ≤ 63um and no focal pigmentary irregularities such as hyperpigmentation or hypopigmentation (i.e., grades 0a and 0b using the definitions of the Wisconsin Age Related Maculopathy Grading System; n=10,623). Cases were classified according to AMD diagnosis in the worst eye. Participants with drusen > 63um and/or focal pigmentary irregularities but had not progressed to late AMD, were classified as early AMD (eAMD; i.e. grades 1a–3; n=4,143). Late AMD was defined as those individuals with geographic atrophy (grade 4a) and/or exudative AMD (grade 4b) in at least one eye. Samples with late AMD were sub-categorized to those with NV without GA (n=3,935); GA without NV (n=1,370) and both NV and GA in the same or fellow eye (GANV; n=1,089). Cases of macular disease due to other primary causes that mimic NV AMD, such as myopic maculopathy, adult vitelliform, any retinal scarring and idiopathic macular telangiectasia, were excluded.
All participants provided prior written informed consent, and the protocols were reviewed and approved by local institutional review boards. Recruitment procedures and detailed AMD grading methods for each study have been described previously [Augood et al., 2004; Baird et al., 2006; Bergeron-Sawitzke et al., 2009; Conley et al., 2005; Dandekar et al., 2006; Ennis et al., 2008; Francis et al., 2009; Fritsche et al., 2009; Haan et al., 2006; Hadley et al., 2010; McKay et al., 2009; Vingerling et al., 1995; Yates et al., 2007; Zareparsi et al., 2004].
Analysis was restricted to samples derived from participants of self-reported European descent and who had been assessed for AMD through retinal photography or clinical examination. Call rates for all SNPs were verified and minor allele frequencies assessed; departure from Hardy-Weinberg Equilibrium (HWE) was determined separately in cases and controls, using the χ2 goodness-of-fit test (P<0.01). Associations between haplotype and AMD and possible effect modification by smoking, age and sex were assessed using likelihood ratio χ2 tests in the logistic regression model. Sex and age were considered to be potential confounding variables, particularly in population surveys where cases of AMD were older and more often female than those who were AMD free. To allow for this and also for the different control group selection criteria employed in the various case-control studies, sex and age (categorized in five year groups) were included in all logistic regression models as were interactions between sex and study and between age-group and study. Similar findings were obtained with age treated as a continuous variable rather than catergorized.
To assess the most appropriate genetic model two diplotype analyses were explored separately with ε3ε3 assigned as the reference group in each: (i) ε2ε2, ε2ε3, and ε3ε3 and (ii) ε4ε4, ε3ε4, and ε3ε3. The diplotype effects were estimated using the model-free approach [Minelli et al., 2005], which does not assume that the underlying genetic model is known in advance and makes use of the information available on all diplotypes. Odds ratios (ORs), OR1 (comparing ε2ε2 withε3ε3), OR2 (ε2ε3 vs. ε3ε3), OR3 (ε4ε4 vs. ε3ε3), and OR4 (ε3ε4 vs. ε3ε3) were estimated, adjusting for study, age-group, sex, and smoking status, using methods previously described [Minelli et al., 2005; Thakkinstian et al., 2006]. These ORs were modeled on a logarithmic scale accounting for both between and within study variation. The ratios λ1=log OR2/log OR1 and λ2=log OR4/log OR3 were estimated. These parameters capture information about the genetic model, as follows; the model is recessive if λ = 0, dominant if λ = 1 and additive if λ = 0.5. A homozygous or heterosis model is appropriate for λ > 1 or λ < 0. Additionally, logistic regression was used to fit various genetic models and the Akaike Information Criteria was calculated to choose the most appropriate genetic model. ORs with 95% confidence intervals (CIs) were then estimated after first checking that there was no evidence of heterogeneity in genetic effects between studies. This was done using likelihood-ratio tests to assess the significance of interactions in the logistic model.
Multinomial logistic regression was used to test if the contribution of APOE and smoking to AMD varied between the 3 late AMD sub-phenotypes (GA, NV and GANV) and between eAMD and late AMD. In these analyses the various disease subgroups were compared with a common control group. The multinomial logistic regression was fitted using the mlogit command in STATA (StataCorp, College Station, Tex) while the remainder of the analysis was performed in SPSS (SPSS Inc, Chicago, Ill).
Of the 10,537 classified cases, 4,143 (39%) were categorized as eAMD and 6,394 (61%) as late AMD (Table 1). Late AMD cases were further categorized by sub-phenotype: GA: n=1370 (21%); NV: n=3935 (62%) and GANV: n=1089 (17%). APOE haplotype frequencies in controls varied between studies ranging from 6.7–10.8% for ε2; 70.8–81.8% for ε3 and 9.5–20.8% for ε4 (Supp. Table S1). APOE haplotype frequencies also differed by phenotype with less variation inε2 (7.9–10.2%) and ε3 (78.0–80.9%) than inε4 (9.1–14.1%) (Supp. Table S2). No evidence for departure from HWE was detected in participating studies for any SNP and genotyping quality control metrics are provided (Supp. Table S3).
The most appropriate genetic models for the APOE diplotypes were assessed from parameters λ1 andλ2 representing the ratio of risks in specific diplotypes [Minelli et al., 2005]. The parameter, λ2, estimated as 0.49 (CI: 0.29–0.89) was strongly indicative of an additive model for ε4 (λ2 =0.5). The parameter λ1 was estimated at 0.14 (CI: 0.01–0.89) suggesting a recessive model forε2 was likely (λ1=0) although an additive model (λ1 =0.5) could not be excluded (Table 2). This choice of model was further supported by the Akaike Information Criteria which attained its minimum value for the recessive model (12807.37) compared to the values obtained for the general (12809.37), additive (12811.34) and dominant models (12811.84). This model is most in keeping with the diplotype comparisons shown in Figure 1 (ε2ε3 andε2ε4 versusε3ε3) which clearly indicate minimal increase in risk associated with a single copy of ε2.
Initial comparison of diplotype frequencies between late AMD and controls used the ε3ε3 diplotype as a reference for calculating ORs, CI and P values. ORs were estimated after adjustment for age-group and sex within each study and for smoking status (ever versus never). The ε2ε2 diplotype showed a significantly increased risk for late AMD after adjustment for age-group and sex within each study and for smoking status (OR=1.83; CI: 1.04–3.23; P=0.04), with non-significant differences in risk associated with the two heterozygous ε2 diplotypes (ε2ε3: OR=0.97; CI: 0.85–1.10; P=0.62; ε2ε4: OR=0.88; CI: 0.64–1.20; P=0.41) relative to the referenceε3ε3 diplotype. Both the heterozygous ε3ε4 diplotype (OR=0.71; CI: 0.64–0.80; P=5.54 × 10−9) and the homozygous ε4ε4 diplotype (OR=0.45; CI: 0.29–0.69; P=2.97 × 10−4) showed significantly decreased risks compared to the reference ε3ε3 diplotype for late AMD (Figure 1).
The estimate of the risk per ε4 haplotype with late AMD was OR=0.72; CI: 0.65–0.79; P=4.41 × 10−11. Addition of interaction terms to the logistic model suggested no evidence of heterogeneity of the ε4 haplotype risk between studies (χ2=15.0, df=14, P=0.38; Figure 2). The risk estimate for the ε2ε2 diplotype relative to the ε3ε3diplotype was OR=1.83; CI: 1.04–3.23; P=0.036, with little evidence of heterogeneity between studies (χ2=19.8, df=14, P=0.14).
Tests for heterogeneity of the OR for ε4 between studies were not significant for any of the four AMD sub-phenotypes (NV: χ2=12.0, df=14, P=0.61; GA: χ2=9.90, df=14, P=0.77; GANV:χ2=15.2, df=14, P=0.37; eAMD: χ2=19.4, df=14, P=0.15). These results indicate that risk estimates per ε4 haplotype were homogeneous across studies and so can be validly pooled. Corresponding tests for the ε2ε2 diplotype were not considered robust because of the very low frequency of this diplotype and consequently are not presented.
Assessment of ε2ε2 risk by AMD sub-phenotype (Table 2) resulted in similar non-significant increases in risk after adjustment for age-group, sex, smoking status and ε4 (NV: OR=1.86; CI: 1.00–3.48; GA: OR=1.59; CI: 0.68–3.69; GANV: OR=1.82; CI: 0.68–4.85; eAMD: OR=1.59; CI: 0.96–2.65). Multinomial logistic regression analysis showed no significant difference in the ε2ε2 effect between the three late AMD sub-phenotypes (χ2=0.35, df=2, P=0.84) or between all late AMD combined and eAMD (χ2=0.01, df=1, P=0.91) (Figure 3a).
There was a significant decrease in risk associated with each copy of ε4 (Table 2) in all AMD sub-phenotypes (NV: OR=0.74; CI: 0.66–0.83; GA: OR=0.65; CI: 0.55–0.77; GANV: OR=0.71; CI: 0.59–0.85; eAMD: OR=0.84; CI: 0.77–0.92) after adjustment for age-group, sex, smoking status and ε2. The multinomial logistic model tests for differences in the ε4 effects between the three late AMD sub-phenotypes (χ2=4.33, df=2, P=0.11) and between all late AMD and eAMD (χ2=2.52, df=1, P=0.11) were not significant (Figure 3b).
Due to variability in recruitment procedures, including the use of some spouse controls in several of the case-control studies, sex was treated primarily as a confounder rather than an independent risk factor in the analysis. However, tests for differences between sexes were estimated in the late AMD risk associated with APOE by fitting interaction terms within the logistic regression model. These indicated no effect modification by sex for either ε2 (χ2=0.56, df=1, P=0.46) or ε4 (χ2=0.47, df=1, P=0.49). Analysis of APOE diplotype and late AMD in males and females separately is presented in Supp. Figure S1.
Smoking status (ever smoker versus never smoker) was associated with a highly significant increased risk for late AMD after adjustment for age-group and sex (OR: 1.50; CI: 1.36–1.65: P=7.92 × 10−17; Table 2). Tests for differences between ever smokers and never smokers in the late AMD risk associated with APOE were obtained by fitting interaction terms. These indicated no effect modification by smoking for either ε2 (χ2=0.46, df=1, P=0.50) or ε4 (χ2=2.27, df=1, P=0.13).
Although smoking status was not significantly associated with increased risk in eAMD (OR=0.94; CI: 0.86–1.03; P=0.16), it was significantly associated with each of the late AMD sub-phenotypes (NV: OR=1.54; CI: 1.38–1.72; GA: OR=1.38; CI: 1.18–1.61; GANV: OR=1.51; CI: 1.27–1.79) after adjustment for age-group, sex and APOE diplotype (Figure 4).
The test for differences in the effect of smoking between each late AMD sub-phenotype was not significant (χ2=3.4, df=2, P=0.18), but a difference in the effect of smoking was detected between late AMD and eAMD (χ2=66.5, df=1, P<0.00001; Figure 4). Significant heterogeneity in the effect of smoking was also detected between studies (Figure 4) (eAMD: χ2=34.0, df=14, P=0.002; NV: χ2=48.9, df=14, P<0.001; Late AMD:χ2=52.4, df=14, P<0.001). Moderate but non-significant heterogeneity was detected for the effect of smoking in GA (χ2=21.9, df=14, P=0.08) but none in GANV (χ2=11.7, df=14, P=0.63).
Test for differences between ever smokers and never smokers in late AMD risk associated with APOE were obtained by fitting interaction terms within the logistic regression model. These indicated no significant effect modification by smoking status for eitherε2 (χ2=0.08, df=1, P=0.78) or ε4 (χ2=1.91, df=1, P=0.17). Analysis of APOE diplotype and late AMD in never smokers and ever smokers separately is presented in Supp. Figure S2.
Genotype data to estimate extended haplotype frequencies (rs405509, rs429358 and rs7412) with adjustment for age, sex and smoking were available for 1,739 late AMD cases and 4,725 controls. Haplotype frequency estimates using UNPHASED [Dudbridge, 2008] are presented in Table 3 and linkage disequilibrium values in Supp. Figure S3. Comparison of ORs between the different haplotypes based on the genotype at rs405509 (i.e. G-ε2 v T-ε2, G-ε3 v T-ε3, G-ε4 v Tε4) showed no significant difference in effect between G-ε2 and T-ε2 (OR=1.58; CI: 0.85–2.95; P=0.15), between G-ε3 and T-ε3 (OR=0.99; CI: 0.88–1.11; P=0.85) or between G-ε4 and T-ε4 (OR=1.07; CI: 0.71–1.62; P=0.75) following adjustment for age, sex and smoking status. Comparison of extended haplotypes against a reference category G-ε3/G-ε3 did not identify significant variation in risk beyond G-ε3/G-ε4 (OR=0.61; CI: 0.38–0.99; P=0.04) and T-ε4/T-ε4 (OR=0.27; CI: 0.08–0.85; P=0.03).
The results generated for ε4 from this pooled data analysis (OR=0.72 per haplotype; CI: 0.65–0.79; P=4.41 × 10−11) provide support for previous studies which identified a protective role with late AMD [Bojanowski et al., 2006; Klaver et al., 1998; Souied et al., 1998]. Previous studies of APOE were dominated by smaller reports involving multiple comparisons which tend to be more liable to publication biases. By conducting a pooled analysis in a large dataset of both published and previously unreported studies, we attempted to clarify the relationship between AMD risk and APOE. Variation in geographic distribution of APOE haplotype frequencies as measured in this large study also highlights the limitations of small studies whose power to detect associations diminishes with decreased haplotype frequency; for example, the frequency of the ε4 haplotype in the control samples from the Edinburgh study (9.5%) was less than half that observed in the Melbourne study (20.8%) and a study in Edinburgh would thus have lower power than a study in Melbourne for any given sample size.
While the significantly increased risk associated with ε2 for AMD has been reported previously [Fritsche et al., 2009], most studies have found non-significant increases in risk. This study has demonstrated a significant increased risk associated with ε2ε2 for late AMD only (OR=1.83; CI: 1.04–3.23; P=0.04), with a corresponding low population attributable risk. The wide confidence intervals and imprecise estimates associated with the rare ε2ε2 diplotype meant that, although our results were most consistent with a recessive genetic model, we could not entirely exclude an additive genetic model. Little heterogeneity was detected between studies for the genetic effect exerted by either ε2 orε4 and, indeed, the genetic effects were not significantly different across all sub-phenotypes of AMD.
Previous reports have suggested that APOE effects may be stronger in women [Baird et al., 2004; Schmidt et al., 2002]. Our analyses did not find evidence of interaction between APOE haplotype risk and sex. Adjustment for both age and sex within studies was made using a logistic regression model, to adjust for possible confounding, but the differing designs of the studies included in this analysis were such that we could not use the data to assess the effect of age and sex on AMD risk. This would be best addressed in a population-based longitudinal study. An earlier meta-analysis examining the effect of APOE haplotype on coronary risk found no effect modification by sex in more than 100,000 participants [Bennet et al., 2007]. Similarly, we did not find evidence to support APOE effect modification by sex in AMD.
Smoking status (ever smoker versus never smoker) showed an increased risk associated with late AMD (OR=1.50; CI: 1.36–1.65) after adjustment for age-group, sex and APOE diplotype, and showed a non-significant difference in effect size between late AMD sub-phenotypes. Our finding of no significant association between smoking status and eAMD suggests that smoking exacerbates early symptoms leading to the progression from eAMD to late AMD, supporting previous epidemiology-based findings [Clemons et al., 2005; Klein et al., 2008; Smith et al. 1996; Vingerling et al., 1996; Xu et al., 2006]. Cigarette smoke may influence macular pigment concentrations [Hammond et al., 1996; Stryker et al., 1988], increase oxidative stress [Beatty et al., 2000] and impair choroidal microcirculation [Suner et al., 2004], all processes hypothesized to be involved in the pathogenesis of AMD. Significant heterogeneity in smoking effect was detected across studies for late AMD, which may be attributable in part to variation in smoking criteria definitions. Standardization of definitions could not be imposed retrospectively beyond the level of ever smoker versus never smoker. As such, this study is limited in its ability to estimate the effect of smoking on disease outcome. Previous reports of effect modification of APOE haplotype by smoking status were not supported in this study suggesting that APOE haplotype and smoking status are independent risk factors.
Previously, rs405509 has been implicated in attenuating APOE promoter activity and subsequent expression [Artiga et al., 1998; Campillos et al., 2003; Ramos et al., 2005] with possible consequences in attenuating AMD risk [Fritsche et al., 2009]. Results reported in this analysis, however, do not support an influence of rs405509, with OR comparisons showing no significant difference in effect between G-ε2 and T-ε2, G-ε3 and T-ε3 or G-ε4 and T-ε4 following adjustment for age group, sex and smoking status. However, several slight differences between this study and that of Fritsche and colleagues are worthy of consideration. Firstly, Fritsche and colleagues compared haplotype frequencies between all AMD cases and controls while this study compared frequencies between late AMD only and controls. Secondly, the original study compared the frequency of each haplotype combination against all others combined while this study compared the frequency of each haplotype against a single reference haplotype (G-ε3/G-ε3). Thirdly, while rs405509 should sufficiently define the extended APOE haplotype reported previously, additional SNPs were genotyped by Fritsche and colleagues which were not replicated in this study.
The retina hosts the body’s second highest level of APOE production after the liver and is likely to play an important role in the maintenance of normal retinal function [Anderson et al., 2001]. Several possible mechanisms of APOE effect on AMD have been proposed such as variability in isoform dimerization potential associated with lipid cholesterol transport or variation in receptor binding affinity [Klaver et al., 1998]. In addition, the positively charged ε4 haplotype has been proposed to improve permeability of Bruch’s membrane, for lipid transport and reducing debris accumulation associated with drusen formation [Crabb et al., 2002; Souied et al., 1998]. In older eyes, evidence supporting reduced lipoprotein transportation across Bruch’s membrane was reported as a consequence of ageing, leading to drusen deposition and RPE insult [Curcio et al., 2009]. More recently evidence to supportε4 as a potential lipoprotein transporter of the macular pigments lutein and zeaxanthin, has been suggested [Loane et al., 2010] and reduced dietary intake of these carotenoids has been associated with increased risk of AMD [Seddon et al., 1994]. As such, variation in genes that modulate retinal cholesterol levels may influence AMD risk [Connor et al., 2007].
The strengths and limitations of the current study should be considered. Our analysis is almost five times larger than published meta-analyses [Bojanowski et al., 2006; Swaroop et al., 2007; Thakkinstian et al., 2006] and, although we cannot completely exclude publication bias in our estimates, this should be limited by our inclusion of both published and unpublished studies. Access to individual level data enabled appropriate adjustment to limit potential confounding by sex, age and smoking and assessment of potential interaction between covariates. In addition, phenotypic sub-classification facilitated stratification by disease sub-phenotype identifying smoking as a major contributing factor to late AMD but not as a risk factor for the early form of the disease. However, variation in recruitment procedures between studies and indeed study type, i.e., case control versus population, limited the ability to measure the full influence of age, sex and smoking on disease risk. Nevertheless, comparison of effect sizes between population and case control studies, showed no significant differences in the odds ratios for ε2ε2,ε4 and smoking status between the two different study designs.
Recent GWAS identified novel variants in the LIPC and CETP genes associated with AMD and cholesterol level modulation, suggesting some alleles may influence cholesterol levels in the macula and in the blood in opposite directions [Chen et al., 2010; Neale et al., 2010]. Improved understanding of APOE effect on AMD disease etiology will improve accuracy of AMD risk prediction models, eventually offering some therapeutic benefit. While the complexity surrounding APOE and cholesterol modulation within the retina remains, potential benefit derived from statin therapy for the treatment of cardiovascular disease (CVD) warrants further investigation in AMD, given the overlap in risk factors for both conditions [Snow and Seddon, 1999]. While ε4 significantly increases risk associated with CVD, atherosclerosis, Alzheimer’s disease (AD) and other dementias [Ang et al., 2008], it clearly plays a protective role with respect to AMD. The mechanisms by which APOE and cholesterol levels modulate AMD risk, especially with respect to the opposing effects reported for other complex diseases such as AD and CVD, are worthy of further investigation.
The authors thank the patients, their families, and the control subjects who participated in the study. This work was supported in part by grants from The Guide Dogs for the Blind Association UK (2008-5a (GS) and OR2006-02d (ATM, JRWY)); Medical Research Council (JRWY, ATM G0000067); Research and Development Office, Northern Ireland Health Personal Social Services (RRG 4.5 - GS); EVI-GENORET, an integrated project funded through the European Union Research Project (FP6 - UC); The Deutsche Forschungsgemeinschaft (WE1259/18-1 and WE1259/19-1 to BHFW); the Alcon Research Institute and the Ruth and Milton Steinbach Foundation New York (BHFW). Russo Grant from Tufts University School of Medicine (JMS); Macular Degeneration Research Fund-Tufts Medical Center (JMS); Massachusetts Lions Eye Research Fund (JMS); Research to Prevent Blindness USA (MLK, PJF); Foundation Fighting Blindness (PJF); financial support from the Department of Health through the award made by the National Institute for Health Research to Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology for a Specialist Biomedical Research Centre for Ophthalmology (JRWY, ATM); the National Health & Medical Research Council of Australia Centre for Clinical Research Excellence #529923 – Translational Clinical Research in Major Eye Diseases and a Practitioner Fellowship to RHG. The Centre for Eye Research Australia receives operational infrastructure support from the Victorian Government (PNB, RHG); Macular Disease Society (GMK and AJL); T.F.C. Frost Charity (AJL) and British Council for the Prevention of Blindness (AJL). Genetic analysis in the EUREYE study was supported by an MRC Biomarker Award G0601354 (AH and AF). Target funding (SF0940026s07) by the Estonian Ministry of Education and Science (MR).
Supporting Information for this preprint is available from the Human Mutation editorial office request (humu/at/wiley.com)