Hypercholesterolaemia and hypertriacylglycerolaemia are well established as risk factors for atherosclerosis and CHD(1, Reference Sarwar, Danesh and Eiriksdottir2). Dietary fats are important determinants of serum cholesterol, lipoproteins and TAG concentrations(Reference Hegsted, Ausman and Johnson3, Reference Keys, Anderson and Grande4). Links between dietary lipids and changes in serum lipids have been extensively investigated in well-controlled studies. The wide inter-individual variation in the response of the serum lipids to dietary components has been proposed to be partly due to genetic variation(Reference Katan, Beynen and de Vries5). Identification of common variations in genes related to dietary responsiveness may allow individual diet prescription optimising the treatment of dyslipidaemia(Reference Lovegrove and Gitau6, Reference Ordovas7). Candidate genes have been related to the metabolism, synthesis and intestinal absorption of fatty acids, cholesterol and lipoproteins(Reference Lovegrove and Gitau6–Reference Loktionov8).
In previous studies, we have examined cross-sectional associations between diet, blood lipids and apoE (APOE) single nucleotide polymorphisms (SNP) and, apart from a small proportion (3 %) of individuals who are of the APOE e2/e4 genotype, there is little evidence that different APOE genotypes respond differently to differences in SFA or total fat intake(Reference Wu, Bowman and Welch9). Here, we examine gene–nutrient interactions in cross-sectional associations between diet, blood lipids and a SNP in the 3-hydroxy-3-methylglutaryl-coenzyme A (HMGCoA) reductase gene (HMGCR).
The HMGCoA reductase enzyme catalyses a limiting step in endogenous cholesterol synthesis converting HMGCoA to mevalonate, a key intermediate in the production of cholesterol and other sterols(Reference Brown and Goldstein10). Inhibitors of this enzyme, such as simvastatin, pravastatin and lovastatin (statins), lower serum total cholesterol and LDL-cholesterol and are highly effective for cardiovascular risk reduction(1, Reference Amarenco, Lavallee and Touboul11). However, the wide variation in inter-individual response to the therapy suggests that genetic differences may contribute to this variation(Reference Mangravite, Wilke and Zhang12).
Some polymorphisms were identified in the HMGCR locus(Reference Zuliani and Hobbs13–Reference Chasman, Posada and Subrahmanyan17) and have been studied for associations with lipid levels and CHD. A variable nucleotide tandem repeat at the end of an Alu sequence located 10 kb 3′ of exon 2 consisting of (TTA)n repeats did not show association with cholesterol levels in either children(Reference Hubacek, Pistulková and Valenta18) or adults(Reference Plat and Mensink19), although a trend for hypercholesterolaemia was observed in children carrying alleles more than ten repeats(Reference Hubacek, Pistulková and Valenta18). CHD patients homozygous for the A allele of the 8302AC polymorphism in intron 2 showed higher levels of VLDL and TAG than controls(Reference Tong, Zhang and Li20). The association of several SNP in the HMGCR gene and the response to statins treatment has been recently studied(Reference Chasman, Posada and Subrahmanyan17). Two tightly linked SNP were found to be significantly associated with a difference in the change in the serum lipid response to pravastatin. A significant reduction in the overall efficacy of pravastatin of 22·3 % for the SNP rs17238540 was observed(Reference Chasman, Posada and Subrahmanyan17).
Our objective was to investigate the influence of the T/G SNP in the HMGCR gene (rs17238540) in the relationship between serum lipids and dietary fat in an initially healthy free-living population of the European Prospective Investigation into Cancer and Nutrition in Norfolk (EPIC-Norfolk) cohort study.
Methods
Study protocol
EPIC-Norfolk is a prospective population study of men and women recruited at age 45–75 years from a general practice age–sex register in Norfolk, UK, from 1993 to 1997. Approximately 25 000 individuals participating in the baseline survey, who had filled in a detailed health and lifestyle questionnaire, attended a first health check when blood and urine samples, and data on height, weight and blood pressure were collected by trained nurses(Reference Khaw, Bingham and Welch21). BMI was estimated as weight in kg/(height in m)2. Medical history was ascertained with the question, ‘Has your doctor ever told you that you have any of the following?’, which was followed by a list of conditions including ‘high blood pressure (hypertension) requiring treatment with drugs’ and ‘high lipid levels requiring treatment with drugs’. Habitual physical activity was assessed, both in work and during leisure time, during the previous year, and individuals were assigned to one of four ordered categories(Reference Wareham, Jakes and Rennie22). Total cholesterol, HDL-cholesterol and TAG were analysed using non fasting blood samples on an RA-1000 (Bayer Diagnostics, Basingstoke, Hants, UK) and LDL-cholesterol was calculated using the Friedewald formula(Reference Friedewald, Levy and Fredrickson23).
Genotype determination
DNA for genotyping was extracted from blood samples collected in EDTA or from stored buffy coat samples with a phenol–chloroform procedure after digestion with proteinase K. The HMGCR SNP (rs17238540) genotype was assessed using pyrosequencing (Pyrosequencing AB, Uppsala, Sweden). Briefly, forward biotin-labelled (5′-biotinGCAAGCCTGT TTGCAGGTAT) and reverse (5′-TCAGCCTAATCCATTGTGTCC) primers were designed flanking the polymorphic region of the HMGCR gene(Reference Chasman, Posada and Subrahmanyan17). The PCR reaction tube (12·5 μl) contained 10 ng DNA, 1 × PCR buffer, MgCl2 (2 mol/l), 0·125 mol/l of each dNTP (deoxynucleotide triphosphate), 10 pmol of each primer, and two units of AmpliTaq Gold (Applied Biosystems, Inc., Branchburg, NJ, USA). The annealing temperature was set at 56°C at forty-four cycles on the Thermal Cycler (PTC-225; MJ Research, Inc., Watertown, MA, USA). Detailed pyrosequencing sample preparation and procedure have been described elsewhere(Reference Wu, Bowman and Welch9, Reference Ahmadian, Gharizadeh and Gustafsson24–Reference Ronaghi, Karamohamed and Pettersson26). The dispensation order of the nucleotides for the machine was: TAACACGAGTG. Repeated blind genetic analysis for 6 % of the sample was 99·9 % concordant.
Dietary data
A FFQ consisting of 131 items was sent to all participants before the first health check(Reference Bingham, Welch and McTaggart27). The questionnaires were coded and analysed for nutrient intake with a custom-made developed program(Reference Welch, Luben and Khaw28). Participants also completed 7 d food diaries, but the time for processing the diaries is much longer and at the time of the present analysis, data from diaries were available for approximately half of those who had responded to the FFQ, which were available for all genotyped subjects. In this way, to gain enough power for the analysis in the separate genotype groups, the data from FFQ were used in the present study. Dietary fibre intake was estimated as the NSP content of each food, measured by the Englyst method. The response rate for the FFQ in EPIC-Norfolk was 99 % (n 25 350). Nutritional analysis of the FFQ was done as previously described(Reference Welch, Luben and Khaw28). Briefly, individuals with more than ten missing lines were omitted from the dataset (n 247, ninety men, 157 women, 0·97 % of data). We also excluded outliers based on the ratio of energy intake:BMR (EI:BMR), BMR estimated using sex-specific equations including age and body weight. Cut points based on the top and bottom 0·5 % of EI:BMI were introduced, identifying another 250 individuals. In addition, nutrient values>3 sd from the mean of the upper (fifth) quintile for energy, fat, carbohydrate, protein and alcohol were also excluded for each nutrient. For the present analyses additional inclusion and exclusion criteria were used, including availability of results for genotyping.
Statistical analysis
After excluding subjects for whom the genotyping data were not available, the statistical analysis for genetic data was conducted on 23 011 participants while complete data on dietary and serum lipids were available for 21 700 and 20 881 participants, respectively.
Characteristics of individuals in the different categories were compared. Differences in means were tested using ANOVA. Differences in the frequency of the categorical variables as well as the difference between the observed and the expected genotype frequency distributions were examined using the χ2 test. We compared serum lipids (total cholesterol, LDL-cholesterol, HDL-cholesterol and TAG) across the quartiles of dietary fat (as percentage energy of total energy intake) and fibre intake, univariate and then adjusted by sex, age, total energy intake (kJ/d), carbohydrate intake (percentage energy intake) alcohol intake (percentage energy intake), exercise index, smoking status and use of lipid-lowering drugs for the whole population then stratified by sex and genotype. The Bonferroni correction for multiple tests(Reference Ottenbacher29) was used to demarcate significant differences for the further multivariate regression and the P value for significance was set as < 0·01. Regressions between dietary component variables and serum lipids were adjusted as described above and were done for the whole cohort then stratified by sex and genotype. Regression coefficients (β) and standard error were normalised to show the change of serum lipids for every approximate sd change in the dietary lipid intake. The results were expressed as the two-tailed test for significance (P) and the 95 % CI. We also compared the regression coefficients of serum lipid on dietary fat intake for the different genotype groups.
The EPIC-Norfolk Study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects/patients were approved by the Norfolk Health District Ethics Committee. Written informed consent was obtained from all subjects.
Results
Genotype frequencies and allele distributions for 23 011 participants were: TT 95·65 %, TG 4·29 % and GG 0·06 %; T 97·8 % and G 2·2 %, respectively (Table 1). The genotype frequencies were in Hardy–Weinberg equilibrium and did not differ between men and women (P = 0·77).
Table 1 3-Hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) SNP29 genotype frequencies in the European Prospective Investigation into Cancer and Nutrition in Norfolk (EPIC-Norfolk) cohort study
* P = 0·77 (Pearson χ2 test for differences in the genotype frequencies between men and women).
As there was no difference in the genotype distribution between men and women, the baseline clinical, biochemical and dietary intake variables separated by genotype groups are presented sex combined in Table 2. There were no differences between the genotype groups for these variables. As the GG group had very few individuals (n 12), it was pooled with the TG group for the following analyses, forming a G variant allele carriers group (TG+GG).
Table 2 Baseline distribution of variables according to 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) genotype in the European Prospective Investigation into Cancer and Nutrition in Norfolk (EPIC-Norfolk) cohort study
(Mean values and standard deviations or percentages)
MUFA, monounsaturated fat; PUFA, polyunsaturated fat; SFA, saturated fat.
* P value for one-way ANOVA tests between genotype groups.
† P value for Pearson χ2 tests between genotype groups.
The comparison between adjusted means of total cholesterol, HDL-cholesterol, LDL-cholesterol and TAG by quartile of lipid (percentage of energy) and fibre intake (g/d) in the two genotype groups is shown in Table 3. Significant differences were observed in the TT genotype group for total and LDL-cholesterol levels which were reduced in the higher quartiles of PUFA and fibre intake and increased in the higher quartiles of SFA intake. In the TT group the HDL-cholesterol increased in the higher quartiles of total fat, SFA and fibre intake and did not change with MUFA and PUFA intake. TAG serum levels in the TT group showed a low inverse association with PUFA and fibre intake (P = 0·04). Conversely, in the TG+GG group, the only weak relationship (P = 0·048) observed was an inverse association between LDL-cholesterol and PUFA intake. Similar results were observed in sex-stratified analyses; so, multivariate regression was conducted sex combined, adjusting for sex.
Table 3 Adjusted total cholesterol, LDL-cholesterol, HDL-cholesterol and TAG by quartiles of fat intake separated by 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) genotype in the European Prospective Investigation into Cancer and Nutrition in Norfolk (EPIC-Norfolk) cohort study*
(Mean values with their standard errors)
MUFA, monounsaturated fat; PUFA, polyunsaturated fat; SFA, saturated fat.
* Univariate ANOVA adjusted by sex, BMI, age, total energy intake, carbohydrate intake (% energy), alcohol intake (% energy), exercise index, smoking status and lipid-lowering drug use.
† P value for trend of the serum lipid values across the nutrient intake quartiles within each genotype group.
The results presented in Table 4 show the linear relationships between the serum lipids, dietary fat and fibre, stratified by genotype. The results for the whole cohort (not shown) were similar to the results for the TT group, which presented a significant and positive relationship between total cholesterol and the intake of total fat, MUFA and SFA, with positive regression coefficients of 0·066 (95 % CI 0·031, 0·10), 0·059 (95 % CI 0·031, 0·088) and 0·076 (95 % CI 0·054, 0·098) for increments of 1 sd in the intake of total fat, MUFA and SFA, respectively. The same trend was observed for LDL-cholesterol with coefficients of 0·059 (95 % CI 0·027, 0·09), 0·058 (95 % CI 0·031, 0·084) and 0·062 (95 % CI 0·042, 0·082) for increments of 1 sd in the intake of total fat, MUFA and SFA, respectively. PUFA intake was inversely associated with all blood lipid fractions as was fibre, except for a positive association between fibre intake and HDL. The present results showed no association between total fat and serum HDL or TAG, while SFA was positively associated with HDL but did not show association with TAG. Also, low MUFA intake was associated with high TAG and HDL levels.
Table 4 Linear regression between total cholesterol, LDL-cholesterol and HDL-cholesterol and dietary fat intake assessed by FFQ multivariate adjusted* and separated according to 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) genotype in the European Prospective Investigation into Cancer and Nutrition in Norfolk (EPIC-Norfolk) cohort study (n 20 881)†
(Standardised regression coefficients, standard errors and 95 % confidence intervals)
MUFA, monounsaturated fat; PUFA, polyunsaturated fat; SFA, saturated fat.
* Analysis adjusted by sex, BMI, age, total energy intake, carbohydrate intake (% energy), alcohol intake (% energy), exercise index, smoking status and lipid-lowering drug use.
† The nutrient intake was adjusted to show differences for increase of 1 sd in the intake of total fat (approximately 6 % of energy), MUFA (approximately 2 % energy), PUFA (approximately 2 % energy), SFA (approximately 3 % energy) and fibre (approximately 7 g/d).
‡ Z = 2·27 (P = 0·023) for testing differences in the regression coefficients.
§ Z = 2·19 (P = 0·029) for testing differences in the regression coefficients.
The TG+GG group showed overall associations in the same direction as the TT group and mostly were of similar magnitude. However, differences between the regression coefficients for serum TAG and SFA intake and fibre were different amongst the genotype groups; whilst the TT individuals showed no change in the TAG levels related to SFA intake ( − 0·0007 (95 % CI − 0·02, 0·02) mmol TAG/l; P = 0·99), the TG+GG group showed a significant positive relationship between TAG and the intake of SFA (+0·109 (95 % CI 0·02, 0·20) mmol TAG/l; P = 0·017) for each 3 % SFA energy increase. In the TG+GG group the inverse relationship between TAG and fibre intake was three times higher ( − 0·137 (95 % CI − 0·22, − 0·05) mmol TAG/l; P = 0·002) than in the TT group ( − 0·038 (95 % CI − 0·06; − 0·02) mmol TAG/l; P < 0·001). In both cases the respective coefficient regressions of serum TAG were different amongst the genotype groups (Z = 2·27, P = 0·023 for SFA intake; Z = 2·19, P = 0·029 for fibre intake).
Discussion
In the present paper we report the effect of the SNP rs17238540 at the HMGCR gene on serum lipids in response to dietary fat and fibre in a largely Caucasian population from the UK.
While HMG-CoA reductase is the main target of regulatory mechanisms for endogenous cholesterol synthesis, which is being exploited in pharmacotherapy, the SNP was not related to baseline lipid levels. This is in agreement with previous smaller studies(Reference Chasman, Posada and Subrahmanyan17, Reference Polisecki, Muallem and Maeda30, Reference Singer, Holdaas and Jardine31).
Serum lipids of TT individuals, according to quartiles of dietary fat and fibre intake, showed significant variation. This is in accordance with previous results of our group for the overall population(Reference Wu, Bowman and Welch9, Reference Bingham, Luben and Welch32). In contrast, serum lipids, according to quartiles of dietary fat intake, for carriers of the minor allele (G) did not show significant variation. The G allele carriers showed additional relationships that were opposite of those observed for TT individuals, such as the tendency to present lower levels of HDL with higher intake of SFA, suggesting a different serum lipid response to dietary components.
In the general population carbohydrate-rich diets are associated with hypertriacylglycerolaemia more consistently than dietary fat(1, Reference Grundy and Denke33). In the Framingham study(Reference Sonnenberg, Quatromoni and Gagnon34) a positive correlation was found between SFA intake and TAG levels, but the model was not adjusted for carbohydrate and alcohol intakes that are well known to raise TAG levels(1, Reference Grundy and Denke33). Adjusting our model for these dietary components (Table 4), we have found that compared with TT individuals, for whom there was no overall relationship between serum TAG and dietary SFA, the G allele carriers showed a higher (β = − 0·109, Z = 2·27) and significantly (P = 0·02) positive correlation between TAG and SFA, indicating greater sensitivity of these individuals to SFA intake. This observation may suggest that the G allele is somehow linked to a liver overproduction or lower blood clearance of VLDL-TAG-rich particles elicited by dietary SFA which is not observed in the overall population, which may in turn have implications for CHD risk.
Fibre intake has been previously shown, using both FFQ and 7 d food diaries, to be inversely related to TAG in this population after adjustment for alcohol and carbohydrate intake(Reference Wu, Bowman and Welch9, Reference Bingham, Luben and Welch32). Cross-sectional inverse associations between fibre intake and serum TAG have also been shown in the Framingham study(Reference Sonnenberg, Quatromoni and Gagnon34), and we have suggested that serum TAG could be used as a biomarker of fibre intake(Reference Bingham, Luben and Welch32). The results of the present study showed that individuals carrying the G allele appeared to be more responsive to dietary fibre, presenting lower serum TAG compared with TT individuals. So, adopting healthier dietary patterns such as eating more fibre and less saturated fat would be more beneficial to these individuals.
The mechanisms underlying our observations are speculative. HMGCoA reductase is an enzyme of cholesterol metabolism, and when inhibited by statins has a moderate TAG-lowering efficacy, in the range of 10–35 %, when TAG exceed 150 mg/dl (1500 mg/l)(Reference Stein, Lane and Laskarzewski35, Reference Ginsberg36). The polymorphism does not apparently alter the basal activity of the enzyme; however, although not confirmed in recently published studies(Reference Polisecki, Muallem and Maeda30, Reference Singer, Holdaas and Jardine31), it was previously described to lower the response to pravastatin(Reference Hubacek, Pistulková and Valenta18). It is also possible that it is linked to other genetic changes within functional parts of the gene and the observed effect in the present study may reflect this. In fact it is in linkage disequilibrium with another SNP in intron 5 and also with a third SNP in a 3′ untranslated exon of the HMGCR gene, which is retained in the mature mRNA(Reference Chasman, Posada and Subrahmanyan17), but the biological effects of these SNP are also unknown.
The genotype frequencies found are in concordance with the frequencies found in a cohort study of largely Caucasians in the USA (TT 93·23 %; TG 6·70 % and GG 0·07 %)(Reference Chasman, Posada and Subrahmanyan17). The frequency found for the minor allele in the present study (0·022) is similar to the frequency (0·019) reported for a study comprising participants from Scotland, Republic of Ireland and The Netherlands(Reference Polisecki, Muallem and Maeda30). Data about allele frequencies in other ethnicities are scarce. In a recent investigation about the effects of several SNP in the HMGCR gene, the frequency of the G allele was found to be 0·09 in African-Americans(Reference Krauss, Mangravite and Smith37). This finding might indicate that our observations can have more impact in populations with different genetic backgrounds in which the number of individuals carrying the G allele is higher.
The present study has several limitations. First, it is a cross-sectional study with consequent limitations concerning the effect of dietary components and changes on blood lipids over time. However, the positive association of SFA intake and the negative association of PUFA intake with total serum cholesterol and LDL-cholesterol are consistent with the directions predicted by carefully controlled intervention studies(Reference Hegsted, Ausman and Johnson3, Reference Keys, Anderson and Grande4, Reference Mensink and Katan38). Second, we measured serum lipids in the non-fasting state. Nevertheless, blood sampling would not be expected to have a major effect on our analysis as indicated by a meta-analysis that showed no differences in IHD risk between non-fasting and fasting participants for TAG levels(Reference Sarwar, Danesh and Eiriksdottir2). Finally, the phenotypic variability of serum lipids is probably related to several common genetic variants in different genes(Reference Katan, Beynen and de Vries5, Reference Ordovas7) which were not considered.
The strength of the present study is the large number of individuals for whom both data on genotype and dietary variables were available along with data concerning important covariates such as physical activity and alcohol intake that allowed us to detect the interactions. Also, a previous study(Reference Wu, Bowman and Welch9) with the same population had found highly significant associations between fatty acids and fibre, assessed by the FFQ, and serum lipid fractions in the same directions predicted from carefully controlled intervention studies on blood lipids(Reference Hegsted, Ausman and Johnson3, Reference Keys, Anderson and Grande4). Although the FFQ cannot be the best method for assessing these nutrients compared with more detailed methods(Reference Bingham, Luben and Welch39), it might rather attenuate the interactions.
The present results suggest that individuals carrying the G allele for the SNP may show a greater response to dietary fibre intake with lower TAG levels and higher TAG levels with increased SFA intake compared with those homozygous for the major allele. In this way the individuals carrying the minor allele may benefit more from dietary intervention to control serum TAG (for example, substituting SFA for PUFA and increasing fibre intake). However, even though the present study involved a large cohort, only a small number of individuals carried the minor allele. Thus, whether this conclusion would be substantiated by data on a larger population of minor allele carriers remains unknown. Nonetheless, the present study does provide unique information on diet, genotype and blood lipids.
Acknowledgements
The present study was supported by programme grants from the Medical Research Council, UK, Cancer Research UK and Conselho Nacional de Ciência e Tecnologia – CNPq, Brazil (fellowship to R. N. F).
R. N. F. performed the statistical analysis, analysed and interpreted the data, and drafted the manuscript; K.-T. K. and S. A. B. conceived of and designed the research, handled funding and supervision, analysed and interpreted the data, and made critical revision of the manuscript for important intellectual content; R. B., K. W. and H. J. acquired the data; R. L. and N. J. W. conceived of and designed the research and acquired the data; all authors reviewed the manuscript.
There are no conflicts of interest to declare.
