Associations of Dietary Cholesterol Consumption With Incident Diabetes and Cardiovascular Disease: The Role of Genetic Variability in Cholesterol Absorption and Disease Predisposition Free

Dietary cholesterol is naturally present in our diet, and it comes from animal products. Associations between dietary cholesterol consumption and incident diabetes or cardiovascular disease (CVD) have been studied for decades, but no consistent conclusion has been reached (1,2). Heterogeneities in the study design, confounding adjustments, and the genetic predisposition across different populations, among other reasons, may affect the dietary cholesterol-disease associations (1–3).

Genetic susceptibility to both disease and responsiveness of serum cholesterol to dietary cholesterol intake may contribute to the controversial associations between dietary cholesterol consumption and cardiometabolic diseases (4,5). However, except for a Chinese study, no other studies have investigated whether genetic susceptibility to disease modifies the association between dietary cholesterol consumption and cardiometabolic diseases (5). Genetic variations in absorption of dietary cholesterol play an important role in responsiveness to dietary cholesterol consumption. The existence of hyporesponse and hyperresponse to dietary cholesterol is well known (6). The Niemann-Pick C1-Like 1 (NPC1L1) protein is responsible for the uptake of sterols into enterocytes, and ATP Binding Cassettes G5 and G8 (ABCG5/8) transporters are responsible for exporting excess dietary sterols to the intestine (7). Although genetic variability in the serum cholesterol responsiveness to dietary cholesterol consumption is important, it was often overlooked by previous studies.

The objective of this study was to evaluate the associations of dietary cholesterol consumption with incident diabetes and CVD in people with different genetic susceptibility to diabetes, CVD, and dietary cholesterol absorption in the UK Biobank.

The UK Biobank recruited half a million participants aged 40 to 69 years between 2006 and 2010 from the general population at 22 assessment centers in the U.K. Details of the study design have been reported elsewhere (8). Data about demographic information, lifestyle, and other health-related aspects were collected through touchscreen questionnaires, interviews, and physical examination.

Participants with genetic data were excluded if they 1) were outliers for heterozygosity or missing rate, 2) had a missing rate >0.02 on autosomes, 3) were sex discordant, or 4) were not in a maximal set of unrelated individuals up to the third degree. Then, participants were also excluded if they had diabetes or CVD at baseline, did not have ≥2 dietary assessments, had extreme total energy intake (women: <500 or >3,500 kcal/day; men: <800 or >4,200 kcal/day], or had missing data for any covariates. The UK Biobank study was approved by the North West Multi-Center Research Ethics Committee in Manchester (REC reference 21/NW/0157). Electronic signed consent was obtained from all participants.

Dietary data were collected using the Oxford WebQ, a validated 24-h dietary assessment tool (9). Dietary data were collected at five time points. Participants were recruited from April 2009 to September 2010 for the first time and completed the Oxford WebQ at the assessment center. The four following time points were from February 2011 to June 2012. To estimate usual intake, mean dietary cholesterol and total energy were calculated for participants completing at least two dietary recalls. A principal component analysis was conducted to derive dietary patterns based on 43 food groups (10).

The outcomes for this study were incident diabetes and CVD (including stroke and coronary artery disease [CAD]). Diseases were defined based on the International Statistical Classification of Diseases and Related Health Problems 9th (ICD9) and 10th (ICD10) revisions and self-reported diagnosis. Surgical procedures according to the Office of Population, Censuses and Surveys: Classification of Interventions and Procedures, version 4 coding (OPCS4) were also used to define incident CVD. Diabetes was defined using ICD9 codes 25000, 25001, and 25002; ICD10 codes E10–E14; self-reported diagnosis; or medication use (11). CVD was defined using ICD9 codes 410, 412, 414, and 4273; ICD10 codes I20, I24–I25, I48, and I60–I64; OPCS4 codes K40–K46, K49–K50, K62, and K75; self-reported diagnosis; or CVD mortality (12).

Genome-wide genotyping was performed for 487,409 participants using the UK Biobank Axiom Array. Imputation was performed by the Wellcome Trust Centre for Human Genetics using the Haplotype Reference Consortium and the UK10 K haplotype resources (13). We constructed two types of polygenic risk scores (PRSs) for LDL cholesterol (LDL-C) and non-HDL cholesterol (non-HDL-C): one used all single nucleotide polymorphisms (SNPs) significantly associated with LDL-C and non-HDL-C, and the other used SNPs related to dietary cholesterol absorption only. SNPs related to dietary cholesterol absorption are in the ABCG5/8 and NPC1L1 genes. PRSs for diabetes, stroke, and CAD were also constructed. Information (including corresponding betas, position on the chromosome, effect allele, and non–effect allele) for creating PRSs was collected from previous genome-wide association studies (Supplementary Tables 1–7) (4,14,– 16). Each SNP was recoded as 0, 1, or 2, based on the number of risk alleles, and each SNP was weighted by its β-coefficient. SNPs were excluded if they were unavailable or had imputation information scores <0.4. Three PRS groups were defined as low (quintile 1), intermediate (quintiles 2–4), and high (quintile 5) for each PRS.

The mean (SD) or median (interquartile range [IQR]) for continuous variables and the count (proportion) for categorical variables were used to describe baseline characteristics of participants according to the quintiles of dietary cholesterol consumption. The duration of follow-up was calculated from the last time of dietary data collection to the diagnosis of diabetes or CVD, the time of death, or 31 October 2022, whichever occurred first.

Cox proportional hazard models were used to estimate the associations of dietary cholesterol consumption with incident diabetes and CVD in the overall sample. Models were sequentially adjusted for age, sex, ethnic background (White, non-White), education level (less than high school, high school graduate, some college, college graduate or higher, none of above), and Townsend Deprivation Index (quartiles 1–4) (model 1); plus smoking status (never, former, and current), drinking status (never, former, and current), alcohol intake (grams per day), physical activity (metabolic equivalent task minutes per week), total energy (kilocalories per day), saturated fat (grams per day), and four dietary patterns derived from principal component analysis (10) (model 2); plus statin use (yes/no) and family history of diabetes or CVD (yes/no) (model 3); plus BMI (BMI, kilograms per square meter) (model 4). To avoid reverse causation, events documented during the first 2 years of follow-up were excluded. Hazard ratios (HRs) with 95% CIs were reported. The proportional hazards assumption was verified before analysis using the Kolmogorov-type supremum test (17). The restricted cubic spline with four knots at the 5th, 35th, 65th, and 95th percentiles was used to assess dose-response relationships between dietary cholesterol consumption and incident diabetes and CVD. If the association was linear, each additional 300 mg of dietary cholesterol consumption per day was used as the interpretation unit to facilitate the comparison with previous studies (18).

Both model 3 and model 4 were used to evaluate the associations of dietary cholesterol consumption with incident diabetes and CVD according to three levels of genetic susceptibility to the diseases (i.e., diabetes, CAD, and stroke) and cholesterol absorption. The interactions between dietary cholesterol consumption and PRSs were examined using the likelihood ratio test. A post hoc analysis was conducted to assess whether genetic susceptibility to cholesterol absorption modifies the associations between dietary cholesterol and serum cholesterol levels. Linear regression models were used to evaluate the associations of dietary cholesterol consumption with LDL-C and non-HDL-C levels stratified by three genetic risk groups for cholesterol absorption.

A two-sided P < 0.05 was used to determine statistical significance. PLINK 2.0 was used to perform quality control and calculate PRSs. Other analyses were performed using SAS for Windows version 9.4 and R 4.2.2.

All data are available from the UK Biobank on request (https://www.ukbiobank.ac.uk/).

This study included 96,826 participants for diabetes analysis and 94,536 participants for CVD analysis (Supplementary Fig. 1). Over a median follow-up of 10.5 years (IQR, 10.4–10.9), 2,441 participants developed diabetes and 5,284 participants developed CVD. Participants with higher dietary cholesterol consumption were more likely to be current smokers and had higher total energy consumption compared with those with lower dietary cholesterol consumption (Table 1). The incidence rate of diabetes (3.0 per 1,000 person-years) and of CVD (6.3 per 1,000 person-years) was higher with greater dietary cholesterol consumption.

The relationships between dietary cholesterol consumption and incident diabetes and CVD were linear (P for nonlinear >0.05) (Supplementary Fig. 2). Each additional 300 mg of dietary cholesterol consumed per day was significantly associated with an elevated risk of diabetes (HR, 1.17 [95% CI, 1.06–1.28]) and CVD (HR, 1.10 [95% CI, 1.03–1.17]) based on model 3, while these associations were not significant after adjusting for BMI (HR for diabetes, 0.99 [95% CI, 0.90–1.09]; HR for CVD, 1.04 [95% CI, 0.98–1.12]) (Table 2). These associations did not differ by genetic susceptibility to diabetes (P for interaction = 0.44), stroke (P for interaction = 0.88), and CAD (P for interaction = 0.06) in both model 3 and model 4 (Fig. 1 and Supplementary Fig. 3).

No significant interactions were found between dietary cholesterol consumption and PRSs, for both those created from all SNPs and SNPs associated with cholesterol absorption only in relation to incident diabetes (P for interaction > 0.1). A significant interaction was observed between dietary cholesterol consumption and non-HDL-C absorption-related PRS in relation to incident CVD, regardless of whether BMI was adjusted for (P for interaction <0.05) (Fig. 2 and Supplementary Fig. 4). Compared with the association with incident CVD in the low genetic risk group (HR, 0.97 [95% CI, 0.82–1.15]), the association appeared to be stronger among people in the intermediate-risk group (HR, 1.07 [95% CI, 0.97–1.17]) and high-risk group (HR, 1.11 [95% CI, 0.96–1.28]). Similarly, the associations of dietary cholesterol consumption with serum LDL-C and non-HDL-C levels in the high-risk group based on non-HDL-C absorption-related PRS appeared to be stronger than those observed in the low- and intermediate-risk groups, although the subgroup association estimates were not statistically significant (Supplementary Table 8).

This study found that higher dietary cholesterol consumption was associated with a higher risk of incident diabetes and CVD among U.K. adults, but these associations were attenuated to the null after further adjusting for BMI. These associations did not differ by genetic susceptibility to the diseases. Only the effect of non–HDL-C variants on cholesterol absorption was found to interact with dietary cholesterol consumption and the association between dietary cholesterol consumption and incident CVD appeared to be stronger in people with greater genetic susceptibility to cholesterol absorption, although the stratum-specific association estimates were not statistically significant.

Many well-designed and well-executed prospective cohort studies have investigated the associations of dietary cholesterol consumption with incident diabetes and CVD, but both no association and positive association have been frequently reported, and a few studies even reported an inverse association (3,18–23). Previously published studies vary substantially in sample size, covariates included for adjustment, dietary assessment methods and their validity, ranges of dietary cholesterol consumption (mean intake: from <200 to >500 mg/day), outcome definition, follow-up time, and sociodemographic, dietary, and clinical characteristics of study samples, which have likely contributed to the inconsistent findings across studies (3,18–23). Even among the studies reporting positive associations of dietary cholesterol consumption with incident diabetes and CVD, the magnitude of the associations, comparing two groups with either extreme intakes (e.g., quintile 5 versus quintile 1) or a different intake of 300 mg/day, was modest, and the associations were commonly attenuated considerably with sequential adjustments. Proper confounding adjustment is key to obtain unbiased estimates. Our study implemented a most comprehensive adjustment to minimize confounding through taking into account a wide spectrum of demographic, socioeconomic, lifestyle, and dietary confounders including total energy, dietary patterns, saturated fat, and BMI. Other than the well-investigated nongenetic heterogeneities, the role of genetic variability in cholesterol absorption and disease predisposition in contributing to the inconsistent findings has been understudied (24).

Different genetic susceptibility to disease across populations is considered an important factor in explaining the inconsistent associations reported (5,25). Although the importance of gene-dietary cholesterol consumption interaction on the development of cardiometabolic diseases has been repeatedly discussed, only one study so far has reported that genetic susceptibility to CAD modified the association between egg consumption, a major dietary source of cholesterol, and incident CAD in a Chinese population (5). Our study did not find that the associations differed by genetic susceptibility to diabetes and CVD. The discrepancy between ours and the Chinese study may be attributable to the differences in dietary assessment approaches (food frequency questionnaires versus 24-h recall), dietary exposure (eggs versus dietary cholesterol), confounding adjustments, and cardiometabolic and genetic risk profiles of the study samples, among others (24). Further studies are needed to confirm these results among different study populations.

Our study is one of the first investigations to examine the influence of genetic variations in cholesterol metabolism on cardiometabolic consequences of cholesterol consumption. “High cholesterol absorbers” appeared to have a higher risk of CVD than “low cholesterol absorbers,” when using non-HDL-C-related PRS for cholesterol absorption in our study. A previous study also reported that ABCG5/8 gene variants influenced non-HDL-C levels, affecting the risk of CAD (4). When analyzing genetic variations in cholesterol absorption based on LDL-C, our study did not find a significant interaction. A possible explanation is that non-HDL-C is more closely related to dietary cholesterol than LDL-C because it includes chylomicron remnants (the intermediate metabolites of dietary cholesterol). Furthermore, studies have reported non-HDL-C as a stronger predictor of cardiometabolic risks than LDL-C (26). It is within our expectation that there was no significant interaction when the PRS was created based on all SNPs instead of absorption-related SNPs alone, because all SNPs reflect genetic variations in the totality of exogenous dietary cholesterol absorption and endogenous cholesterol synthesis. Therefore, it is challenging to isolate pure cardiometabolic effects of exogenous cholesterol on diseases. Also, our study did not investigate genetic variability in other metabolic steps in cholesterol metabolism, because the relevant genes involved in dietary cholesterol esterification, assembly, and transport also implicate in the endogenous cholesterol metabolism (24). That a significant interaction between dietary cholesterol consumption and genetic susceptibility to cholesterol absorption in relation to incident diabetes was not found may be attributed to the weaker association between serum cholesterol levels and diabetes compared with CVD (27). Regardless of whether the interaction was significant, we did not find statistically significant associations in any subgroups, even among those with high genetic susceptibility to cholesterol absorption. Identifying a significant association may rely on studies of larger size and discovery of additional novel SNPs implicated in cholesterol absorption through greater sequencing depth in the future.

In our study, the positive associations of dietary cholesterol consumption with incident diabetes and CVD survived the comprehensive adjustment until performing additional BMI adjustment, which nullified the associations. In line with most previous studies, BMI was treated as a confounder (1,23), because BMI level may affect both dietary intake and cardiometabolic risks. However, it cannot be completely ruled out that BMI may also be in the causal pathway connecting dietary intake and cardiometabolic diseases. Some studies did not control for BMI as a confounder or treated BMI as a mediator (18,28). Although certain foods containing dietary cholesterol have been associated with weight gain, limited evidence supports that dietary cholesterol consumption is directly associated with obesity (29). Therefore, to accurately determine whether BMI is a confounder, mediator, or both for the association between dietary cholesterol and cardiometabolic diseases, it is essential to use longitudinal data with repeated BMI measurements. This will ensure a clear temporal sequence between dietary cholesterol intake, weight change, and outcomes. Whether BMI is adjusted for is a critical consideration when comparing findings across studies.

The findings of our study do not support that higher dietary cholesterol consumption was associated with a higher risk of diabetes and CVD. Also, our study does not provide convincing data for supporting personalized recommendations for dietary cholesterol according to genetic susceptibility to diabetes and CVD or to genetic variability in cholesterol absorption, although “high cholesterol absorbers” appeared to be more sensitive to dietary cholesterol consumption compared with “low cholesterol absorbers.” Considering that foods high in cholesterol are often rich in saturated fat and animal protein, the latest Dietary Guidelines for Americans still recommends “dietary cholesterol consumption to be as low as possible without compromising the nutritional adequacy of the diet” (30).

Our study has several limitations. First, dietary cholesterol consumption is self-reported, which is prone to recall bias. Also, relying on two or more 24-h dietary recalls may not collect usual intake of dietary cholesterol accurately. Second, measurement errors for self-reported data including dietary intake are inevitable. Third, residual confounding was likely due to measurement errors and unmeasured confounding. Fourth, results of our study may not be generalized to non-U.K. populations, as dietary behaviors and culture, genetic backgrounds, and risk factor profiles of cardiometabolic diseases differ across different populations. Fifth, this is an observational study, and, thus, causal inferences cannot be made.

In conclusion, our study did not find significant associations of dietary cholesterol consumption with incident diabetes and CVD among U.K. adults, after adjusting for BMI as a confounder. These associations were also not significant in subgroups stratified by genetic predisposition to the diseases and cholesterol absorption, although the dietary cholesterol-CVD association appeared to be stronger among high versus low cholesterol absorbers, assessed using non-HDL-C related PRS. Our findings require replication, but do not support personalized recommendations for dietary cholesterol according to genetic susceptibility to cardiometabolic diseases or to genetic variability in cholesterol absorption.

Acknowledgments. This research has been conducted using the UK Biobank Resource under Application Number 101169. All authors thank the participants of the UK Biobank. The graphical abstract was created with BioRender.com.

Funding. This study was supported by the National Key Research and Development Program of China (2022YFC2705203 and 2023YFC2506700), the National Natural Science Foundation of China (82373551), the Shanghai Key Discipline of Public Health Grants Award (GWVI-11.1-20), and the Innovative Research Team of High-Level Local University in Shanghai.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.S. and V.W.Z. conceived and designed the study. S.S., Y.D., and S.W. did the data cleaning, analysis, and interpretation. S.S. and V.W.Z. wrote the manuscript. X.D., N.F., L.X., and V.W.Z. provided statistical expertise and assistance. All authors contributed to the interpretation of the data and critical revision of the manuscript for important intellectual content and approved the final draft. V.W.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Steven E. Kahn and Jonathan E. Shaw.