Data from 1668 men (316 cardiovascular disease events) from the Framingham Offspring Study was reanalysed, specifically examining APOE:smoking interactions. Overall hazard ratio (HR) for smoking was 1.95 (1.52, 2.50) compared to non-smokers. Using ɛ3/3 as a referent group, in non-smokers HRs for ɛ2 carriers (ɛ2+; 1.04 (0.61, 1.76) and ɛ4 carriers (ɛ4+; 1.04 (0.70, 1.54) showed no major risk increase. In smokers, HRs were 1.96 (1.26, 2.78) in ɛ3ɛ3 men, 3.46 (2.14, 5.60; p = 0.09 for interaction) in ɛ2+ and 3.81 (2.49, 5.84; p = 0.01 for interaction), with a significant interaction between daily cigarette consumption and APOE genotype on risk (p = 0.03). The potential mechanism for this APOEɛ4:smoking interaction was examined in a second study of 728 Caucasian patients with diabetes, where markers of reactive oxygen species were available. APOE genotype was not associated with plasma OX-LDL or total antioxidant status (TAOS) in non-smokers. However, in smokers ɛ4+ had 26.7% higher plasma OX-LDL than other genotypes (APOE:smoking interaction p = 0.04), while ɛ2+ had 28.4% higher plasma TAOS than ɛ3ɛ3 and ɛ4+ combined (APOE:smoking interaction p = 0.026). Although direct extrapolation needs to be considered with caution, these results identify that the cardiovascular disease risk-raising effect of ɛ4+ is confined to smokers, and a feasible mechanism is presented by the reduced antioxidant capacity/increased OX-LDL of apoE4.
Almost all chronic disease results from gene: environment interactions, and acknowledging and understanding this is the challenge that faces health scientists, and cardiovascular disease (CVD) is no exception (Sing et al. 2003). It is one of the accepted tenets of cardiovascular genetics that common variants of the apolipoprotein E gene (APOE), the ɛ2, ɛ3 and ɛ4 alleles, are associated with differences in coronary heart disease (CHD) and CVD risk. The 1996 meta-analysis estimated a summary unadjusted odds ratio (OR) for ɛ4, compared to ɛ3 homozygotes, as 1.38 (1.22, 1.57) in men and 1.82 (1.30,2.54) in women, whereas the ɛ2 allele was not significantly associated with CHD risk in men, OR 0.94 (0.79, 1.11), or in women, 1.07 (0.66, 1.76) (Wilson et al. 1996). Thus while the ɛ4 allele was risk-enhancing in both men and women, the effect of the ɛ2 allele was not (Wilson et al. 1996; Ioannidis et al. 2001). A recent meta-analysis of studies published since 1996 concurred that in men and women ɛ4 carriers (ɛ4+) had 42% higher CHD risk than ɛ3/3 individuals, while the risk-association for ɛ2 carriers was not consistent, but overall showed risk effects that were no different from ɛ3 homozygotes (Song et al. 2004).
Lahoz et al. (2001) examined the association of APOE genotype and CVD risk in 3413 men and women participating in the Framingham Offspring Study (FOS). After adjustment for non-lipid risk factors including smoking, in the men alone both the ɛ2 [OR 1.79 (1.15, 2.77)], and ɛ4 [(OR 1.63 (1.13, 2.34)] alleles, were associated with an increased risk of CVD compared to ɛ3 homozygotes (Lahoz et al. 2001). In FOS, this atypical effect of the ɛ2 allele on risk was suggested to be due to the increased postprandial lipemia associated with the ɛ2 allele and elevated fasting insulin levels in ɛ2 carriers (Lahoz et al. 2001).
The modification of the CHD risk-associated effects of smoking, by APOE genotype, was examined in the Northwick Park Heart Study II (NPHSII), a prospective study of CHD in over 3000 middle-aged UK men. In never-smokers the CHD risk was similar in all APOE genotypes. Compared to this reference group, ɛ3/3 current smokers had a hazard ratio (HR) of 1.68 (95%CI 1.01, 2.83), which is within the range of the well-established doubling of risk linking cigarette smoking and CHD (Doll et al. 1994; Kannel et al. 2003). Carriers of the ɛ4 allele, however, showed significant interaction of genotype and smoking on risk (p < 0.007), with an HR of 3.17 (95%CI 1.82, 5.50). Thus the association of the ɛ4 allele with CHD risk was confined to current smokers alone, and was independent of other classical CHD risk factors, including plasma lipids (Humphries et al. 2001).
We have reanalysed the FOS APOE data (Lahoz et al. 2001), examining the potential interaction between APOE and smoking on CVD risk. The study did not have the power to examine the APOE: smoking interaction in women, where CVD incidence was low; therefore the men alone were considered in this analysis. To explore the possible mechanisms for the interaction of smoking with APOE, and since smoking is a generator of reactive oxygen species (ROS) (Churg, 2003), we examined the relationship between APOE genotype and markers of oxidative stress, namely oxidised LDL (OX-LDL) and total antioxidant status (TAOS). Oxidative stress results from the excess of ROS overwhelming any endogenous antioxidant protection, and is well-documented in CVD (Harrison et al. 2003). ROS production promotes the formation of foam cells due to scavenger receptor-mediated uptake of OX-LDL by macrophages (Azumi et al. 2002). Since neither OX-LDL nor TAOS measures were available in FOS, we examined their relationship with APOE in patients with diabetes (University College London Diabetes and Cardiovascular Disease Study (Dhamrait et al. 2004)) who typically display increased oxidative stress (Brownlee, 2001), to determine whether there was a relationship between these markers of ROS and APOE genotype, which would reflect the differing antioxidant capacity of the apoE isoforms.
The details of the Framingham Offspring Study (FOS) and recruitment have been published previously (Kannel et al. 1979). Complete data was available on 1697 men. Between 1971–1994, there were 316 recorded CVD events in these men during this 23 year follow-up. (One hundred and twenty-two women had had an event, but because of the lack of power only the men were considered for further analysis). CVD included CHD (MI, angina pectoris, coronary insufficiency and coronary death), stoke, peripheral vascular disease and congestive heart failure, based on clinical information from physician records and hospitalisation. The study had ethical approval from institutional review committees and the subjects gave full informed consent.
Details of the University College Diabetes and Cardiovascular Study (UDACS) have been described elsewhere (Stephens et al. 2004). Briefly, this comprises 1011 consecutive men and women (of whom 439 men and 284 women were Caucasian), recruited from the diabetes clinic at University College London Hospitals NHS Trust (UCLH) between the years 2001-2. All patients had diabetes according to WHO criteria (Alberti et al. 1998). Subjects were categorised by the presence/absence of clinically manifest CVD. Ethical approval was obtained from the UCL/UCLH ethics committee.
These measurements were performed on citrated plasma collected at the time of recruitment. All samples were immediately placed on ice and underwent centrifuge within one hour of collection. All samples were stored at −80°C in aliquots and only defrosted for these measurements, and were analysed within 12 months of collection. OX-LDL was measure by ELISA (Mercodia AB, Uppsala Sweden). Plasma TAOS was measured by a photometric microassay previously described by Sampson et al. (2002) and is inversely related to oxidative stress (the higher the oxidative stress, the lower the TAOS). The inter-assay coefficient of variation (CV) was 14.1%, and the intra assay CV was 4.3%. Combined APOE genotype and OX-LDL, and APOE genotype and TAOS levels were available for 282 males and 178 females (63.6%) and 424 men and 277 women (96.9%), respectively. There was no difference in the APOE genotype distribution or the baseline characteristics of these two groups, despite the difference in participant number, and thus does not represent selection bias (Supplementary Table 1).
Table 1. Baseline characteristic at exam 1 by CVD event status at exam 6 in the Framingham Offspring Study (men)
No CVD event (1352)
CVD event (316)
*29 ɛ2/4 not included
Smokers (number (%))
Systolic BP (mm Hg)
In the FOS skewed variables were log transformed to normalise their distribution before analysis, while geometric means and approximate SD are presented. Differences in the baseline characteristics between those men who had a CVD event and those who were event-free were tested using the t-test for continuous variables and χ2 for categorical variables. Deviation from Hardy Weinberg equilibrium using the χ2 was assessed on the whole sample and included the ɛ2ɛ4 individuals. Baseline characteristics and exam 6 CVD prevalence data are presented for descriptive purposes only. We used Cox's regression models with time-dependent covariates, with survival time over the 23 year period (time to CVD or censoring) as a dependent variable; APOE genotypes and smoking as primary independent variables; and age, alcohol consumption, diabetes, hypertension, BMI, total cholesterol, HDL-C, and cholesterol treatment as covariates. Interactions between genotype and smoking were considered using a likelihood ratio test to compare models with and without the interaction term. Data on smoking and covariates were measured at exams 1 through to 6 and value change over time. The models were specified such that the hazard at a given exam depended on the values of smoking and covariates prior to that exam. We used lagged covariates to allow the proper time sequence of risk factors (exposure) and outcome. These analyses were performed using the PHREG procedure in SAS.
In the UDACS, ANOVA was used to assess the association between genotype and plasma TAOS/OX-LDL. The distribution of OX-LDL was normalised for analysis by using a square root transformation. The relationships between baseline parameters and plasma OX-LDL and TAOS were tested by Pearson's correlation coefficients. ANCOVA was performed to test the association between genotype and plasma OX-LDL and TAOS, after adjustment for potential confounders using multiple regression analysis to obtain a residual. The interaction between genotype and smoking status in determining plasma TAOS or OX-LDL (and adjusted measures) was performed using linear regression modelling, including an interaction term for genotype and smoking status. For these analyses, since there was no heterogeneity of effect, to test the interaction between APOE genotype and smoking on OX-LDL, ɛ3/3 and ɛ2+ were pooled and contrasted with ɛ4+, while to test the interaction of genotype and smoking on TAOS, ɛ3/3 and ɛ4+ genotypes were pooled and contrasted with ɛ2+. In all cases a p-value of less than 0.05 was considered statistically significant.
Framingham Offspring Study
Lipid profile, DNA and APOE genotypes were available on 1668 men after exclusion of 29 ɛ2ɛ4 men. Genotype distribution was in Hardy-Weinberg equilibrium and the allele frequencies were 0.80 for ɛ3, 0.12 for ɛ4 and 0.079 for ɛ2. The baseline characteristics of the cohort are shown in Table 1. Compared to men free of CVD, those men who had CVD were significantly older (p = 0.0032), more likely to be a current smoker (p = 0.0001), had a greater BMI (p = 0.03), and higher systolic BP (p = 0.0001), cholesterol (p = 0.005) and LDL (p = 0.004).
The Effects of Smoking on CHD Risk According to APOE Genotype
Forty-two percent of men smoked, and overall the effect of smoking was to increase CVD risk almost two fold [crude HR 1.93 (1.50, 2.49)]. The smoking prevalence according to APOE genotype is presented in Table 2. In a covariate-adjusted Cox's regression model, smoking was associated with an increase in CVD risk in ɛ3 homozygotes, HR1.96 (1.34, 2.77); in ɛ2+ smokers compared to ɛ2+ non-smokers the HR was 3.45 (1.73, 6.45) and in ɛ4+ smokers compared to ɛ4+ non-smokers HR was 3.68 (2.22, 6.11). Thus in ɛ4+ and ɛ2+, smoking was having an effect of a similar magnitude on CVD risk, which was about 70–80% higher than in ɛ3/3 men who smoked, suggesting a potential gene: environment/CVD risk interaction.
Table 2. Prevalence of smoking according to APOE genotype in the FOS men
in whole group*
Combined Effects of Smoking and APOE Genotype on CHD Risk
Compared to ɛ3/3 non-smokers, risk in ɛ2+ and ɛ4+ non-smokers was not statistically different, HRs 1.04 (0.61, 1.76) and 1.04 (0.70, 1.54), respectively. For the ɛ2+ men who smoked the HR was 3.46 (2.14, 5.60; p = 0.09 for interaction), and for ɛ4+ HR 3.81 (2.49, 5.84; p = 0.01 for interaction) compared to ɛ3/3 non-smokers (Figure 1).
To assess the quantitative effect of smoking on the genotypic effect on CVD risk, the interaction between smoking and APOE genotypes was evaluated using smoking as a continuous measure (number of cigarettes smoked per day). These results indicate a clear gradient of effect of smoking and genotype on risk (Figure 2), with an overall significant APOE: smoking interaction (p = 0.03).
University College Diabetes and Cardiovascular Study (UDACS)
The baseline characteristics of the UDACS Caucasian men and women by CVD status are presented in Table 3. In total 723 (100%) Caucasian men and women were successfully genotyped for the APOE genotype. Genotype distribution was in Hardy-Weinberg equilibrium and the allele frequencies were 0.81 for ɛ3, 0.12 for ɛ4 and 0.07 for ɛ2. Mean age was significantly higher for those with CVD compared to those free of CVD. Total cholesterol and LDL cholesterol were significantly lower in those with CVD and this could be explained by the significantly higher proportion of patients with CVD taking lipid lowering medications; ACE inhibitors (ACEI), aspirin, insulin and statins. HDL was lower and TGs higher in those with CVD compared to the CVD free individuals.
Table 3. Baseline characteristics of 723 Caucasian subjects by CVD status in the UDACS
No CVD (n = 523)
CVD (n = 200)
* ɛ2/4 excluded from the start
Smokers (number %)
Systolic BP (mmHg)
Body mass index (kg/m2)
Association of APOE Genotype and Smoking on TAOS and OX-LDL in UDACS
To test whether the genotype: smoking interaction could be explained by measures of oxidative stress, which were not available in FOS, the association between APOE and plasma OX-LDL and TAOS was examined in UDACS. Only Caucasian men and women were considered, and since there was no significant evidence for heterogeneity of effect between men and women for these parameters (not shown) the analysis was carried out in the group as a whole. Plasma TAOS was correlated with glucose, triglycerides and HDL (r =−0.13 −0.14, 0.11, respectively; all p < 0.05) and OX-LDL with age, LDL and HDL (r = 0.31, −0.13, respectively; all p < 0.05) and these variables were adjusted for in the analysis. No difference was observed in plasma TAOS and OX-LDL by pharmacological treatments with ACE inhibitors, statins, aspirin or insulin. Overall there was no significant association of genotype with either plasma TAOS or OX-LDL and similarly, after stratification by smoking status, in non-smokers there was no association of genotype with either adjusted TAOS or OX-LDL (Figure 3a, b). However, in current smokers ɛ2+ had 28.4% higher plasma TAOS than ɛ3/3 and ɛ4+ combined (p = 0.026, Figure 3a), with a significant APOE: smoking interaction on plasma TAOS (p = 0.026). Considering OX-LDL in smokers, ɛ4+ had 26.7% higher OX-LDL than ɛ3/3 and ɛ2+ combined (p = 0.04, Figure 3b), displaying a significant APOE: smoking interaction on OX-LDL (p = 0.04).
Many studies, when analysing genotypic effects on CVD risk, correct for all CVD risk factors, including smoking, to take into account any confounding chance differences in their prevalence by genotype. In doing so the possibility of analysing for potential gene: environment interactions are removed. The studies included in the 1996 meta-analysis (Wilson et al. 1996) corrected for confounders, including smoking, and thus gave validity to the concept that APOEɛ4 carriers were at increased risk of CHD compared to ɛ3/3 individuals, while ɛ2 carriers were protected from risk. This concept of ɛ4-related risk became an accepted tenet of the field. The data from this present study are compelling, since they highlight two essential points. Firstly, reanalysis of the effect of APOE genotype on CVD risk in the Framingham Offspring Study, specifically testing for gene: smoking interactions, clearly demonstrates that when stratified on the basis of smoking, the APOEɛ4 and ɛ2 carriers who do not smoke were at no greater CVD risk than ɛ3/3 non-smokers. This is particularly powerful since a direct comparison can be made with the same FOS data when smoking status was not considered, and where both ɛ4 and ɛ2 carriers showed increased CHD risk (Lahoz et al. 2001). Secondly, compared to ɛ3/3 men, ɛ4 men who smoked showed a significant interaction of ɛ4 and smoking on risk (p = 0.01). Both ɛ4+ and ɛ2+ showed a clear cigarette/dose response, which has not been previously reported.
Significant effects of APOE genotype on risk remained even after adjustment for lipid variables in the present study (in smokers) and in several other studies (Lahoz et al. 2001; Stengard et al. 1999; Humphries et al. 2001). Taken together with the fact that the ɛ4 non-smokers show no increased CVD risk, despite their higher lipid levels, this suggests that the APOE risk-associations are independent of lipids.
APOE Genotype and Smoking Interaction on CVD Risk
In the FOS men, smoking per se was associated with a 95% increase in risk. Compared to ɛ3/3 non-smokers, the effect of smoking was to increase risk by 96% in ɛ3/3men, 246% in ɛ2+ and 281% in ɛ4+, with evidence of a significant interaction of ɛ4+ and smoking on risk. These results support the previous study, which suggested that the accepted observation of the universal risk-association of ɛ4 needed re-assessment (Humphries et al. 2001).
Two studies assessing the APOE: smoking interaction reported failure to confirm these findings. The International Studies of Infarct Survival (ISIS), a large case control study, demonstrated an incremental OR increase of 1.17 for ɛ2ɛ3 to ɛ3ɛ3 to ɛ3ɛ4 subjects, but no ɛ4:smoking interaction was detected (Keavney et al. 2003). ISIS used ɛ3ɛ2 subjects as the reference group, rather than ɛ3ɛ3, and excluded ɛ2ɛ2 (n = 78) and high-risk ɛ4ɛ4 individuals (n = 160). In re-analysis (Humphries et al. 2003) a significantly greater than additive effect of genotype and smoking on risk was seen, with a relative excess of risk interaction (RERI of 1.62 (0.4, 2.97), thus confirming APOE:smoking interaction on CHD risk in the ISIS. In the Physicians Health Study, a nested case: control cohort matched for smoking, neither an APOE genotype effect on risk nor an ɛ4:smoking interaction were identified (Liu et al. 2003). As an indicator that this was a low-risk cohort, the expected association of APOE genotype with plasma lipid levels was not seen. Several studies using carotid artery atherosclerosis (CAA) as an endpoint have identified an ɛ4:smoking interaction (Djousse et al. 2002; Karvonen et al. 2002; Pezzini et al. 2004). Overall, these data support an ɛ4: smoking interaction on risk, with the exception of those studies where there is insufficient power, low smoking prevalence, or inappropriate analysis.
APOE ɛ2 Genotype and Smoking Interaction on CVD Risk
Lahoz et al. in reporting their initial findings of the association of ɛ2 and ɛ4 alleles with increased risk in FOS, acknowledged that the ɛ2 findings were unusual and contrary to the results from the meta-analysis (Lahoz et al. 2001). Thus the FOS is atypical since ɛ2 carriers usually show CVD risk protection compared to ɛ3/3 men. In the present analysis, both ɛ2+ and ɛ4+ non-smokers showed hazard ratios which were not significantly different from ɛ3/3 non-smokers. However, in the smokers bothɛ2 and ɛ4 carriers showed increased risk. While the interaction with smoking was significant for ɛ4+ carriers it was not for ɛ2 carriers. In the FOS it is possible that this lack of interaction between ɛ2 and smoking on risk merely reflects the smaller number of ɛ2 carriers.
Two possible mechanisms for the increased risk-association of the ɛ2 allele, in the FOS men but not the women were suggested by Lahoz et al. (2001); increased postprandial lipemia associated with the ɛ2 allele, and elevated fasting insulin levels compared to ɛ3/3 and ɛ3/4 men (Orchard et al. 1994).
Biological Basis for APOE: Smoking Interactions
In order to consolidate the statistically-established ɛ4:smoking interaction with a biological mechanism, we examined two measures of oxidant stress in a second study (UDACS), taking APOE genotype and smoking into account in a study of patients with diabetes, a group where increased ROS is typical. In the UDACS, ɛ4+ smokers had significant higher levels of OX-LDL (26.7%, p = 0.04) than the other genotypes combined, with significant APOE: smoking interaction on OX-LDL (p = 0.04). One mechanism causing this could be the higher levels of ROS in ɛ4+ subjects, since apoE4 is acting as a poorer antioxidant. This is supported by the fact that recombinant apoE protection against oxidation in vitro is greater in ɛ2 > ɛ3> ɛ4 (Smith et al. 1998; Jolivalt et al. 2000), and could be explained by the fact that ɛ2 has two free SH-groups, ɛ3 has one and ɛ4 none. Alternatively, subjects with the ɛ4 allele who have amongst the lowest plasma levels of apoE (Davignon et al. 1988) have been reported to have a preponderance of small dense LDL (Haffner et al. 1996). Since small dense LDL is more susceptible to oxidation (Chait et al. 1993; Sevanian et al. 1996), this would result in higher OX-LDL in ɛ4 carriers, particularly in those exposed to oxidative stress, namely smokers.
In the smokers, but not the non-smokers, carriers of the ɛ2 allele had 28% higher TAOS levels compared to both ɛ3/3 and ɛ4+ subjects (p = 0.026), with a significant genotype: smoking interaction on TAOS. High plasma TAOS implies reduced ROS and less oxidative stress, and goes some way to explain the protective nature of the ɛ2 allele in smokers, as seen in the NPHSII. In the FOS, although the ɛ2+ men had higher CVD risk this showed no interaction with smoking, and this supports the idea that increased risk in ɛ2+ in FOS is independent of the higher antioxidant status of apoE2. Even though plasma TAOS is not a highly specific measure of plasma oxidative stress, for a large number of samples such as in the UDACS it is a practical and inexpensive assay. By comparison, assays of other plasma markers (e.g. plasma F2-isoprostanes) are time consuming and expensive. To test whether TAOS was an adequate measure of oxidative stress, we previously observed a significant correlation (r =−0.65; p = 0.003) between plasma TAOS and esterified F2-isoprostane, supporting the use of plasma TAOS as an appropriate marker of reactive oxygen species (Dhamrait et al. 2004; Stephens et al. 2004).
Although direct extrapolation from one study to another needs to be considered with caution, these data provide a plausible explanation for the ɛ4:smoking interaction seen in the FOS and the lack of ɛ2:smoking interaction reported elsewhere (Humphries et al. 2001). These data provide no additional insight into the ɛ2 effect reported in the FOS (Lahoz et al. 2001).
These results provide strong confirmation that in men, the ɛ4 association with CVD risk is essentially confined to smokers. Since ɛ4 non-smokers show no major CVD risk, i.e. when the environmental insult is not present, this has important public health implications. No studies to date have examined this ɛ4: smoking effect in women, so it is not clear whether these effects are gender specific. The multifactorial nature of CHD/CVD implies not only independent effects of genes and environmental factors but also their interaction on risk. Thus our study of ɛ4: smoking interaction and association with measures of ROS confirms and extends our understanding of the multifactorial basis of CHD/CVD risk.
PJT, EH, and SEH are supported by the British Heart Foundation (RG2000/015). JWS is supported by Diabetes UK (BDA: RD01/0001357). This work was supported by grants HL 54779 and HL35243 and NIH/NILBI contract NO1-38038 and contract 53-K06-5-10 from the US Department of Agriculture Research Services.