 |
|
Association of Apolipoprotein E Genotypes With Lipid Levels and Coronary Risk
Anna M. Bennet, PhD;
Emanuele Di Angelantonio, MD, MSc;
Zheng Ye, PhD;
Frances Wensley, MSc;
Anette Dahlin, BSc;
Anders Ahlbom, PhD;
Bernard Keavney, MD, FRCP;
Rory Collins, FRCP, FMedSci;
Björn Wiman, MD, PhD;
Ulf de Faire, MD, PhD;
John Danesh, MSc, DPhil, FRCP
JAMA. 2007;298:1300-1311.
ABSTRACT
| |
Context Previous reviews of associations of apolipoprotein E (apoE) genotype and coronary disease have been dominated by smaller studies that are liable to biases.
Objective To reassess associations of apoE genotypes with circulating lipid levels and with coronary risk.
Data Sources We conducted an updated meta-analysis including both published and previously unreported studies, using MEDLINE, EMBASE, BIOSIS, Science Citation Index, and the Chinese National Knowledge Infrastructure Database published between January 1970 and January 2007, reference lists of articles retrieved, and a registry of relevant studies.
Study Selection Eighty-two studies of lipid levels (86 067 healthy participants) and 121 studies of coronary outcomes (37 850 cases and 82 727 controls) were identified, with prespecified principal focus on studies with at least 1000 healthy participants for lipids and those with at least 500 coronary outcomes.
Data Extraction Information on genotype frequencies, lipid levels, coronary outcomes, and laboratory and population characteristics were recorded independently by 2 investigators and/or supplied by study investigators.
Results In the most extreme comparison, people with the  2/2 genotype had 1.14 mmol/L (95% confidence interval [CI], 0.87-1.40 mmol/L [44.0 mg/dL; 95% CI; 33.6-51.1 mg/dL]) or about 31% (95% CI, 23%-38%) lower mean low-density lipoprotein cholesterol (LDL-C) values than those with the 4/4 genotype. There were approximately linear relationships of apoE genotypes (when ordered 2/2, 2/3, 2/4, 3/3, 3/4, 4/4) with LDL-C and with coronary risk. The relationship with high-density lipoprotein cholesterol was inverse and shallow and that with triglycerides was nonlinear and largely confined to the 2/2 genotype. Compared with 3/3, the odds ratio for coronary disease was 0.80 (95% CI, 0.70-0.90) in 2 carriers and was 1.06 (95% CI, 0.99-1.13) in 4 carriers.
Conclusions There are approximately linear relationships of apoE genotypes with both LDL-C levels and coronary risk. Compared with individuals with the 3/3 genotype, 2 carriers have a 20% lower risk of coronary heart disease and 4 carriers have a slightly higher risk.
INTRODUCTION
Apolipoprotein E (apoE) is a multifunctional protein that plays a key role in the metabolism of cholesterol and triglycerides by binding to receptors on the liver to help mediate clearance of chylomicrons and very low-density lipoproteins from the bloodstream.1-3 Although individuals carrying the 4 allele have higher and those carrying the 2 allele have lower total cholesterol levels than people with the commonest 3/3 genotype, studies of lipid markers have typically involved too few participants to characterize relationships with different lipid subfractions across the 6 common genotypes.4 A previous review of 48 published studies among a total of 15 492 disease cases reported that, compared with 3/3 individuals, 4 carriers have a much greater risk of coronary disease and that 2 carriers have a neutral risk.5 But about half of those data were from studies with fewer than 500 coronary cases, which may be more liable to publication biases.6-9
Our reassessment of associations of apoE genotypes with circulating lipid levels and with coronary risk uses the following approach to maximize power and minimize bias: (1) we report updated meta-analyses of studies of apoE genotypes with total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), or triglycerides (involving data on up to 86 067 participants in 82 studies) and with coronary outcomes (involving data on up to 37 850 cases and 82 727 controls in 121 studies), with tabular data sought from investigators to supplement and update published data; (2) we contacted principal investigators listed in a registry of coronary genetic studies to seek unreported data; and (3) we prespecified that principal analyses would be based on studies of lipid fractions with at least 1000 healthy participants and on studies of coronary disease with at least 500 cases, involving only studies that had adequately assessed apoE status and lipid levels and/or coronary outcomes.
METHODS
We sought studies published between January 1970 and January 2007 on apoE genotype associations with concentrations of total cholesterol, LDL-C, HDL-C, or triglycerides or with risk of myocardial infarction (defined by World Health Organization Multinational Monitoring of Trends and Determinants in Cardiovascular Disease [MONICA] criteria10) or angiographic coronary stenosis (generally defined as at least 50% stenosis of  1 major coronary arteries). For lipid fractions, data were used from only apparently healthy controls (ie, people without known coronary or other diseases or clinical lipid abnormalities) who had information on all relevant genotypes. Electronic searches, not limited to the English language, were performed using MEDLINE, EMBASE, BIOSIS, Science Citation Index, and the Chinese National Knowledge Infrastructure Database by scanning the reference lists of articles identified for all relevant studies and review articles (including meta-analyses), hand searching of relevant journals, and by correspondence with authors of included studies. The computer-based searches combined search terms related to the relevant gene (eg, Apolipoprotein E, ApoE genotypes), lipid phenotypes (eg, total cholesterol, LDL, HDL, and triglycerides), and coronary disease (eg, myocardial infarction, atherosclerosis, coronary heart disease, and coronary stenosis) without language restriction (Figure 1).
|
|
|
|
Figure 1. Study Flow Diagram
|
|
|
The following data were extracted independently by 2 investigators, using a prepiloted data extraction form: genotype frequencies by categorical disease outcome; means and standard deviations of lipid fractions by genotype; mean age of cases; proportions of men and ethnic subgroups (defined as people of white European continental ancestry, East Asian, or other); fasting status; genotyping and lipid assay methods; and use of blinding of laboratory workers. Discrepancies were resolved by discussion and by adjudication of a third reviewer. We used the most up-to-date information in cases of multiple publications. We supplemented published data by a tabular data request to authors of published reports and to investigators of 62 potentially relevant unreported studies listed in published meta-analyses11-14 who had published on variants other than apoE.
Statistical Analysis
Analyses involved only within-study comparisons to avoid possible biases, with principal analyses of larger studies that had used accepted assessments of apoE genotype status (eg, polymerase chain reaction, isoelectric phenotyping), lipid markers (eg, enzymatic and precipitation methods), and coronary outcomes (as described above). Individuals with the 3/3 genotype were defined as the reference group. Separate analyses were conducted for each genotype (in the following prespecified order: 2/2, 2/3, 3/3, 3/4, and 4/4, with the position of 2/4 genotype inserted after data exploration) and for 2 and 4 carrier status (this particular analysis excluded, of course, the 2/4 genotype).
Summary odds ratios (ORs) for coronary disease and mean plasma levels of total cholesterol, LDL-C, HDL-C, and triglycerides (and differences in mean plasma levels between each genotype and the reference group) were calculated for each genotype using a random effects model that included between-study heterogeneity. We avoided any double counting by analyzing different coronary cases separately before combining them into a single coronary disease group for the few studies that included a single control group and nonoverlapping coronary stenosis cases and nonfatal myocardial infarction cases.
Consistency of findings across studies was assessed using the I2 statistic.15 Publication bias was assessed using funnel plots, Egger test16 and the trim-and-fill17 method. Heterogeneity was assessed using the Q statistic18 and by examining prespecified groupings of studies characteristics. All analyses were performed using Stata Statistical Software, Release 9 (StataCorp LP, College Station, Texas).
RESULTS
ApoE Genotypes and Lipid Outcomes
Eighty-two studies19-57 58-101 (44 previously published [19 in MEDLINE journals, 25 in non-MEDLINE journals], 6 expanded and/or updated, and 32 previously unreported in relation to lipid markers) were identified with data on apoE genotypes and lipid outcomes from a total of 86 067 disease-free participants (details of study characteristics available from the authors upon request). The principal analyses in the current review are based on data from the 22 studies that each involved at least 1000 participants (Table 1), collectively comprising about 84% of the total available data (ie, information was available from 72 150 individuals for total cholesterol; 61 463 for LDL-C; 69 142 for HDL-C, and 67 852 for triglycerides). Of these 22 studies (9 of which were previously published36, 48, 54, 69, 71, 81, 84, 86-87 and 13 previously unreported24-25,34-35,37, 44, 51, 59, 62-63,73, 77, 82), 12 involved European populations,25, 37, 44, 48, 51, 59, 62, 69, 73, 81-82,84 6 were based in North America,24, 34-35,54, 63, 87 and 4 in East Asia.36, 71, 77, 86 Nine of these studies were based in prospective cohorts24-25,34-36,62-63,73, 84 (typically recruiting participants from population registers, such as general practitioners lists or electoral rolls), 13 were either cross-sectional surveys or case-control studies37, 44, 48, 51, 54, 59, 69, 71, 77, 81-82,86-87 (with controls sampled from general populations in 4 of the case-control studies37, 44, 48, 59 and from blood donors in 1 such study51). Sixteen of the larger studies24, 35-36,44, 48, 54, 59, 63, 69, 71, 73, 77, 81-82,86-87 involved fasted individuals, and 1 did not report fasting status.84
|
|
|
|
Table 1. Summary of Data Available in the Current Analyses on Apolipoprotein E Genotypes and Circulating Lipid Levels or Coronary Risk
|
|
|
All of the studies used enzymatic methods to measure total cholesterol and triglycerides, and all used precipitation methods to assess HDL-C; LDL-C was directly measured in 4 studies44, 81, 86-87 and calculated in the remainder. All but 6 studies35, 44, 48, 51, 69, 81 used polymerase chain reaction–based methods to establish apoE genotypes. The overall allele frequencies among people without coronary disease were 0.07 for 2, 0.82 for 3, and 0.11 for 4; the overall genotype frequencies were 0.007 for 2/2, 0.116 for 2/3, 0.022 for 2/4, 0.623 for 3/3, 0.213 for 3/4, and 0.019 for 4/4. These frequencies were broadly similar in men and women and in adults older or younger than 55 years (although in East African populations, the frequencies of 2 and 4 were 0.08 and 0.09, respectively).26
Associations of apoE genotypes with levels of total cholesterol or LDL-C were strongly positive and approximately linear when ordered as described above (Figure 2). Comparison of people with 2/3 vs those with 3/4 (which are, apart from 3/3, the most common genotypes) yielded differences in total cholesterol of –0.43 mmol/L (95% confidence interval [CI], –0.36 to –0.51 mmol/L [–16.6 mg/dL; 95% CI, –13.9 to –19.7 mg/dL] or about –8%; 95% CI, –6% to –9%) and in LDL-C of 0.52 mmol/L (95% CI, –0.44 to –0.61 mmol/L [–20.1 mg/dL; 95% CI, –17.0 to –23.6 mg/dL] or about –14%; 95% CI, –12% to –17%). Comparison of people with 2/2 vs those with 4/4 (ie, the 2 most extreme but rarest, genotypes) yielded differences in total cholesterol of –0.81 mmol/L (95% CI, –0.61 to –1.02 mmol/L [–31.3, mg/dL; 95% CI, –23.6 to –39.4 mg/dL] or about –14%, 95% CI, –11% to –18%) and in LDL-C of –1.14 mmol/L (–0.87 to –1.40 mmol/L [–44.0 mg/dL; 95% CI, –33.6 to –54.1 mg/dL] or about –31%; 95% CI, –23% to –38%).
Associations of apoE genotypes with HDL-C levels were weakly inverse, with a difference of 0.07 mmol/L (95% CI, 0.06 to 0.09 mmol/L [2.7 mg/dL (95% CI, 2.3 to 3.5 mg/dL] or about 5%; 95% CI, 4% to 7%) in people with 2/3 vs those with 3/4, and a difference of 0.07 mmol/L (95% CI, 0.02 to 0.11 mmol/L [2.7 mg/dL; 95% CI, 0.8 to 4.3 mg/dL], or about 5%; 95% CI, 2% to 8%) in people with 2/2 vs those with 4/4. The association of apoE genotypes with triglycerides was nonlinear, with the highest levels in people with the comparatively rare 2/2 genotype and the lowest levels in the common 3/3 reference group, corresponding to a difference between these groups of 0.34 mmol/L (95% CI, 0.18 to 0.50 mmol/L [30.1 mg/dL; 95% CI, 15.9 to 44.2 mg/dL] or about 21%; 95% CI, 11% to 32%). Associations of apoE genotypes with lipid fractions generally did not vary importantly when studies were grouped by potentially relevant characteristics (details available from the authors upon request).
ApoE Genotypes and Coronary Risk
One hundred twenty-one studies19-64,65-68,88, 90-94,96, 100-135,136-164 (96 previously published [57 in MEDLINE journals, 39 in non-MEDLINE journals], 7 expanded and/or updated, and 18 previously unreported) were identified with data on apoE genotypes and coronary outcomes from a total of 37 850 cases and 82 727 controls (details of study characteristics available from the authors upon request). The principal prespecified analyses are based on data from 17 of these studies that each involved at least 500 cases (Table 1), collectively comprising about 21 331 cases and 47 467 controls (or about 56% of the total available data). Of the 17 larger studies (10 of which were published in journals indexed by MEDLINE25, 37, 39-40,44, 47-48,51, 62, 131, 148 and 7 previously unreported27, 34-35,45, 59, 63, 92, 111), 13 involved European populations,25, 27, 37, 39-40,44-45,47-48,51, 59, 62, 92, 148 3 were based in North America,34-35,63 and 1 was from Australia.131 Six of these were prospective cohort studies,25, 34-35,62-63,92 and 11 were case-control studies27, 37, 39-40,44-45,47-48,51, 59, 131, 148; there were no case-cohort studies. Studies involved patients either with confirmed myocardial infarction (generally defined by World Health Organization criteria) or with coronary stenosis (defined as 50% or 70% stenosis of 1 major coronary arteries). All but 5 studies35, 44, 48, 51, 148 used polymerase chain reaction–based genotyping methods, and none reported genotyping call rates.
Figure 3 shows that the combined ORs for coronary disease in the studies with at least 500 cases were 0.80 (95% CI, 0.70-0.90) in 2 carriers and 1.06 (95% CI, 0.99-1.13) in 4 carriers. With the 3/3 genotype as the reference group, Figure 4 shows that the ORs increased progressively between 2/2 (0.83; 95% CI, 0.55-1.25), 2/3 (0.82; 95% CI, 0.72-0.92;), 2/4 (0.93; 95% CI, 0.81-1.07), 3/4 (1.05; 95% CI, 0.99-1.12;), and 4/4 genotypes (1.22; 95% CI, 1.08-1.38;). Recorded features of the populations studied did not explain much of the moderately high degree of heterogeneity among the studies noted in Figure 3 . When based on the studies with at least 500 cases, the risk associations were broadly similar in men and women, people older or younger than 55 years, and in studies grouped by various characteristics (P value for interaction >.05 for each characteristic recorded, except data source [P=.003], Figure 5).
|
|
|
|
Figure 5. Odds Ratios for Coronary Disease With Apolipoprotein E Genotypes in Studies With 500 or More Cases
CHD indicates coronary heart disease; CI confidence interval; MI, myocardial infarction; PCR, polymerase chain reaction; phenotype, use of isoelectric methods to classify apolipoprotein E genotype; and PI, principal investigator of study. Exploration of potential sources of heterogeneity yielded P >.05 for location, publication status, and genotyping method, P =.03 for study design, and P=.003 for data source in 2 carriers. All corresponding P values were >.05 in 4 carriers. Size of the data markers is proportional to the inverse of the variance of the odds ratios.
aTotal number for exposed and reference groups.
bIncludes 1 Australian study.
cRefers to status of the source study at the time of current analysis.
|
|
|
Findings in the case-control studies were broadly similar to those in cohort studies, arguing against major survival bias (Figure 5). By contrast, in the meta-analysis based on studies with fewer than 500 cases, the ORs for coronary disease were 1.00 (95% CI, 0.91-1.11) in 2 carriers and 1.66 (95% CI, 1.50-1.84) in 4 carriers. There was a high degree of heterogeneity among findings in the smaller studies, mainly related to differences in geographical location, study design, and type of publication (Figure 6). These findings were not materially altered by using fixed effect meta-analysis (which does not incorporate heterogeneity between studies) or exclusion of the few studies departing from Hardy-Weinberg equilibrium (details available from the authors upon request).
|
|
|
|
Figure 6. Odds Ratios for Coronary Disease With Apolipoprotein E Genotypes in Studies With Fewer Than 500 Cases
CHD indicates coronary heart disease; CI, confidence interval; MI, myocardial infarction; PCR, polymerase chain reaction; phenotype, use of isoelectric methods to classify apolipoprotein E genotype; and PI, principal investigator of study. Several characteristics explained a considerable part of the heterogeneity, including study location (P<.001), design (P<.001), publication status (P=.004), data source (P<.001), and type of journal (P<.001). Size of data markers is proportional to the inverse of the variance of the odds ratios.
aTotal number for exposed and reference groups.
bRefers to status of the source study by January 2007.
cGenotype-specific data was not available from 3 studies.
d The weighted average of these strata-specific odds ratios (and the numbers of participants contributing to them) does not equal the overall odds ratio because only partial data were available on these characteristics since they were provided as tabular data by only a subset of relevant studies.
|
|
|
Evidence of Publication Bias
Figure 5 and Figure 6 display different ORs in the prespecified comparison of results for studies with at least 500 cases vs those for smaller studies (combined ORs of 0.80 (95% CI, 0.70-0.90) vs 1.00 (95% CI, 0.91-1.11), respectively, comparing 2 carriers with 3/3; or 1.06 (95% CI, 0.99-1.13) vs 1.66 (95% CI, 1.50-1.84), respectively, comparing 4 carriers with 3/3). Table 2 shows a similar pattern of findings when cut-off levels for numbers of cases in studies were varied. Funnel plots show a clear excess of extreme findings in studies with fewer than 500 coronary outcomes (Egger test, P < .001), and trim-and-fill analyses imply that 15 studies of 2 and 35 studies of 4 are required to make the funnel plots symmetrical. A cumulative meta-analysis, subdivided by study sample size, showed that this divergence in ORs by study size was evident by about the year 2000 (details available from the authors upon request).
|
|
|
|
Table 2. Odds Ratios for Coronary Disease According to Different Cut-off Levels of Study Size Used in Meta-analyses
|
|
|
COMMENT
Because previous reviews of apoE genotypes have been dominated by many smaller reports that are liable to biases,4-5,165 we conducted a more detailed analysis focusing on larger studies, both published and previously unreported, which fulfilled quality criteria in relation to assessment of apoE status, lipid levels, and coronary outcomes. We have demonstrated approximately linear relationships of apoE genotypes (when ordered 2/2, 2/3, 2/4, 3/3, 3/4, 4/4) with LDL-C levels and with coronary risk. The LDL-C levels were approximately 30% lower in people 2/2 than with 4/4 genotypes, a difference comparble with that produced by "statin" medication.166 The relationship of apoE genotypes with HDL-C was shallow and inverse and that with triglycerides was nonlinear and largely confined to the 2/2 genotype, with the latter about 2 times weaker than previously reported4 (Table 3). We found that, in comparison with the commonest 3/3 genotype, 2 carriers had a 20% reduced coronary risk, in contrast with previous estimates that 2 carriage is neutral for coronary risk.5 We noted strong evidence of selective publication in previous estimates based on smaller studies. This is a serious concern given that apoE genotypes and coronary risk had hitherto been considered among the few quantitatively secure associations in cardiovascular disease genetics. Our findings may have several implications, as described below.
|
|
|
|
Table 3. Comparison of Findings of the Current Analyses With Those Reported in the Most Recent Previous Meta-analyses of Apolipoprotein E Genotypes
|
|
|
The precise mechanisms by which 2 carriage (and, hence, apo E2 isoforms) might confer advantageous lipid profiles (or other possible cardioprotective effects) are only partially understood.167 They may relate to comparatively more efficient binding of apo E2 isoforms with heparin (which could enhance remnant lipoprotein metabolism through heparan sulfate proteoglycans receptors on the liver) and with small, phospholipid-enriched HDL (which could enhance reverse cholesterol transport).168-169 Although apo E2 isoforms bind to LDL receptors much more weakly than do apo E3 or apo E4 isoforms, most 2 carriers have, as demonstrated by the current data, advantageous lipid profiles and reduced coronary risk, perhaps due to compensatory up-regulation of LDL receptors. (By contrast, about 5% of 2/2 homozygotes develop type III hyperlipoproteinemia, a disorder characterized by increased levels of cholesterol and triglycerides and premature cardiovascular disease.170) The differing effects of different apoE genotypes on coronary risk might also be explained by influences on additional lipid-related phenotypes (eg, on levels of apoE,171 apolipoproteins A-I or apolipoprotein B,172-173 or very low-density lipoprotein174) and/or on markers of inflammation,173, 175 immunity,176 or oxidative status.177 Our findings should stimulate further investigation into possible mechanisms.
Given that the prevalence of the 2 allele is only about 7% in Western populations, even if the 20% lower coronary risk associated with it were to be entirely causal, it would still explain only a few percent of coronary disease cases in Western populations. Although the magnitude of this relative risk is insufficiently strong to justify population-wide screening for apoE genotypes,1 it has been proposed that the effects of apoE genotypes may be particularly strong in certain subgroups, such as in women.5 The current data, however, do not support the existence of such interactions in relation to sex and several other characteristics. Individual participant data would, however, be needed to assess any interactions with other potentially relevant characteristics not recorded in the present study (such as obesity,178 diet,179-180 medication use,181 smoking,147, 182 and glycemic status178). More detailed work is needed to help understand reasons for the comparatively modest amount of heterogeneity observed among the larger studies of apoE and coronary disease, such as factors related to assessment of apoE status, coronary outcomes, and study populations.
Our approach to identify previously unreported data yielded information on an extra 8028 cases of coronary disease from 7 studies with at least 500 cases and on an extra 50 907 participants from 13 studies of lipid outcomes with at least 1000 healthy participants. This experience reinforces the rationale for registry-based initiatives such as the Human Genome Epidemiology Network (HuGENet).183 Our cumulative meta-analysis showed that, in retrospect, the divergence in findings between smaller and larger studies was apparent by the year 2000. This observation underscores the potential value of regularly updated reviews for certain rapidly evolving hypotheses, both to enhance understanding and to optimize the use of resources. The observation that previous analyses both underestimated and overestimated effects of particular apoE genotypes on coronary risk suggests that selective publication could work in surprisingly complex ways. Smaller studies may have preferentially reported striking findings in relation to 4 and coronary risk but underreported the unexpected inverse association between the uncommon 2 allele and coronary risk (perhaps because these differences would have been more difficult to detect). This finding encourages further study of the impact of selective publication in different contexts.6-9
The strengths and limitations of the current study merit consideration. Our analyses involved 5 times more data than in any previous relevant analysis, including tabular data from a considerable number of larger studies (both published and previously unreported). Even though we cannot entirely exclude publication bias in our estimates, any effect should be minor compared with that in previous estimates because of the comprehensive nature of the current review and its focus on larger studies. Our inference that the large discrepancy between ORs in smaller and larger studies was mainly due to selective publication is based on evidence from statistical tests (showing, for example, an excess of extreme findings in the smaller studies of 4) and on lack of any other plausible explanations for the observed differences (eg, genotyping procedures used and departure from Hardy-Weinberg equilibrium did not differ much between smaller and larger studies, nor among published and unreported studies; unfortunately, studies were not able to provide genotyping call rates). Because we did not have access to individual data, we could not control for population stratification nor conduct "mendelian randomization" analyses,37 nor could we adjust for variables in possible intermediate pathways.
CONCLUSIONS
There are approximately linear relationships of apoE genotypes with both LDL-C levels and coronary risk. Compared with 3/3 individuals, 2 carriers have a 20% reduced risk of coronary disease whereas 4 carriers have only a slightly increased risk.
AUTHOR INFORMATION
|
