Association of genetically predicted blood sucrose with coronary heart disease and its risk factors in Mendelian randomization

We assessed the associations of genetically instrumented blood sucrose with risk of coronary heart disease (CHD) and its risk factors (i.e., type 2 diabetes, adiposity, blood pressure, lipids, and glycaemic traits), using two-sample Mendelian randomization. We used blood fructose as a validation exposure. Dental caries was a positive control outcome. We selected genetic variants strongly (P < 5 × 10–6) associated with blood sucrose or fructose as instrumental variables and applied them to summary statistics from the largest available genome-wide association studies of the outcomes. Inverse-variance weighting was used as main analysis. Sensitivity analyses included weighted median, MR-Egger and MR-PRESSO. Genetically higher blood sucrose was positively associated with the control outcome, dental caries (odds ratio [OR] 1.04 per log10 transformed effect size [median-normalized standard deviation] increase, 95% confidence interval [CI] 1.002–1.08, P = 0.04), but this association did not withstand allowing for multiple testing. The estimate for blood fructose was in the same direction. Genetically instrumented blood sucrose was not clearly associated with CHD (OR 1.01, 95% CI 0.997–1.02, P = 0.14), nor with its risk factors. Findings were similar for blood fructose. Our study found some evidence of the expected detrimental effect of sucrose on dental caries but no effect on CHD. Given a small effect on CHD cannot be excluded, further investigation with stronger genetic predictors is required.


Results
We obtained 9 and 2 uncorrelated (r 2 < 0.001) single nucleotide polymorphisms (SNP) strongly (P < 5 × 10 -6 ) predicting blood sucrose and fructose, respectively (Supplementary Table 1). Of the 9 SNPs for blood sucrose, only 2, 3 and 8 were available for lipids in the Global Lipids Genetics Consortium (GLGC), glycaemic traits, and blood pressure, respectively, while all were available for other outcomes. We used rs3821975 as a proxy for rs4693441 (r 2 = 1.0) for lipids in GLGC and glycaemic traits, but the other SNPs had no proxy, giving 3, 4, and 8 SNPs for lipids in GLGC, glycaemic traits, and blood pressure, and 9 SNPs for all other outcomes. Both of the 2 SNPs for blood fructose were available for all outcomes. The 9 and 2 SNPs explained 6.2% and 0.6% of the variance in blood sucrose and fructose, respectively. The mean F-statistic was 15.7 and 23.6, respectively. None of the SNPs was related to key confounders in the UK Biobank or had known pleiotropic effects according to Ensembl or PhenoScanner.
Genetically higher blood sucrose was positively associated with dental caries, but this association did not remain after accounting for multiple testing (Table 1). Genetically higher blood sucrose was suggestively positively associated with decayed and filled tooth surfaces divided by tooth surfaces (DFSS) but not with decayed, missing and filled tooth surfaces (DMFS). A directionally similar estimate with a much wider confidence interval (CI) was found for blood fructose with dental caries (Supplementary Table 2). Genetically instrumented blood sucrose was not significantly associated with CHD but the direction was positive, as it was for fructose (Table 1  and Supplementary Table 2). Genetically instrumented blood sucrose or fructose were not associated with T2D (Table 1 and Supplementary Table 2), adiposity, blood pressure, lipids, or glycaemic traits (Table 2 and Supplementary Table 3) using any method, with estimates in an inverse direction for adiposity, blood pressure and lipids.

Discussion
Our Mendelian randomization study suggests a positive association of genetically higher blood sucrose with dental caries, but this association did not withstand allowing for multiple testing. We found no significant association of genetically predicted sucrose with CHD, but a small effect of sucrose on CHD cannot be excluded. www.nature.com/scientificreports/ Genetically predicted blood sucrose was not associated with risk of T2D, adiposity, blood pressure, lipids, or glycaemic traits. No associations were found for blood fructose. Our finding of no significant association of blood sucrose with risk of CHD is inconsistent with some metaanalyses of observational studies suggesting a positive association of SSB with CHD 8 or myocardial infarction 27 and an observational study suggesting a positive association of sucrose intake with coronary events 28 . However, the observed associations in Western populations may be confounded by socioeconomic position, diet and lifestyle marked by SSB consumption 8,27 . In contrast, a Japanese cohort of 39,786 participants followed up for 18 years suggested no association of soft drink consumption with CHD 29 . Nevertheless, our finding of a nonsignificant positive association might be owing to inadequate power. Further Mendelian Randomization studies with stronger instruments for sugar intake or biomarkers would be very helpful about the specific role of sugar. More broadly genetic evidence seems to be emerging suggesting that protein rather than fat or sugar is the key dietary element affecting health 30 .
No association was found of genetically predicted blood sucrose or fructose with T2D, which is consistent with a meta-analysis of observational studies of sugar or fructose intake 31 but not with others suggesting positive associations of SSB with T2D 9,32 . However, the studies of SSB also found positive associations of artificially sweetened beverages with T2D 9,32 , implying that the associations might be due to confounding rather than to sugar itself. The null association with adiposity is concordant with our previous finding of SSB with adiposity in Hong Kong Chinese children 16 , but inconsistent with positive associations of sugar or SSB intake with weight gain in observational studies in Western populations 10,11 . The finding is more consistent with meta-analyses of RCTs showing no effect of substituting sugar for other carbohydrates on weight 10,17,20 . Sucrose or fructose intake, particularly in the form of SSB, may cause adiposity by adding to extra energy intake but inducing less satiety 33 thus promoting positive energy balance. However, an experimental study in mice found no difference in adiposity by sucrose intake but did find differences by fat intake 34 , suggesting that fat rather than sucrose drives over-consumption. Sensitivity analyses suggested a possible inverse association of blood sucrose with BMI, which might be chance or because sucrose might be at the expense of a lower fat intake. The lack of associations with blood pressure or lipids is consistent with meta-analyses of trials substituting sugar for complex carbohydrates on systolic blood pressure (SBP) and diastolic blood pressure (DBP) 20 and of trials substituting fructose for glucose on triglyceride 35 , but inconsistent with another showing positive effects of sugar intake on blood pressure and all lipids with or without energy control 18 . However, our MR findings differ from observed positive associations of SSB with hypertension 12 , triglycerides and LDL-C 13 . The lack of associations with glycaemic traits is consistent www.nature.com/scientificreports/ with a meta-analysis of RCTs showing no effect of sugar intake on fasting insulin when substituting sugar for other macronutrients or subtracting sugar from the diet 22 . Nevertheless, most of the trials had small sample sizes and short duration 18,20,22,35 , making the findings less certain. Sugar may cause CVD through several mechanisms. First, sugar may promote CVD through its effect on weight gain. Observational studies have suggested a weakened association of SSB with CHD after adjusting for body mass index (BMI) and energy intake 8 . However, while sugar at high doses in hypercaloric trials did increase weight 17 , sugar at usual intake especially as part of an isocaloric diet showed no effect on weight gain 10,17 . Second, sugar may cause CVD through metabolic effects of fructose. Unlike glucose, fructose can bypass phosphofructokinase and be metabolised to substrates for lipid synthesis in unregulated glycolysis 36 , thus leading to increased de novo lipogenesis, dyslipidaemia, and insulin resistance after excess intake 36 . Chronic fructose intake may activate gene expression of lipogenic enzymes to further enhance lipogenesis 36 . Unregulated fructose metabolism can also raise uric acid concentration, which has been linked to endothelial dysfunction, hypertension, and increased risk of CVD 33 . It is also possible that sugar intake might affect glucagon or insulin which may cause CHD 37,38 . However, we did not find any evidence of these pathways, which need validation using stronger instruments predicting sugar.
Our study has some limitations. First, we used genetic variants predicting sucrose and fructose from metabolomics studies using mass spectrometry 24,26 . Blood sucrose and fructose may not be perfect biomarkers of sugar intake. However, blood sucrose is mainly of dietary origin and metabolized very slowly 25 , so blood sucrose reflects sucrose intake. High fructose corn syrup (42% or 55% fructose, glucose and water) is another commonly used sweetener other than sucrose, which may be captured by blood fructose. Many other factors, including intestinal permeability and liver metabolism 39 , may also influence blood levels of sucrose and fructose, but are unlikely to confound the genetic associations. Second, we used the genetic variants associated with blood sucrose and fructose at less than genome-wide significance (P < 5 × 10 -6 ) and the variance of sucrose and fructose explained by the genetic variants were small (6.2% and 0.6%). Weak instrument bias could attenuate the causal estimates towards the null 23 . However, the SNPs had a mean F-statistic > 10. Weak instrument bias is expected to be small if the p-value is below 1 × 10 -540 . Third, confounding by population stratification is possible. However, the genetic associations concern people of European ancestry generated with genomic control. The GWAS of blood sucrose suggested no inflation or deflation due to population stratification 24 . The genetic variants were not related to key confounders in the UK Biobank. Fourth, genetic predictors of sucrose and fructose may have pleiotropic effects. However, we found no unknown pleiotropy from MR-Egger and MR-PRESSO and no known pleiotropy in Ensembl or PhennoScanner. All methods gave similar results. Fifth, our study was mainly limited to people of European ancestry and might not apply to other populations. However, causes would be expected to be consistent 41 , although might not always be relevant 42 . Sixth, the instruments predicting blood sucrose were obtained from primarily women, which may not be applicable to men. However, there is no evidence that genetic predisposition to blood sucrose or fructose differs by sex or has different effects by sex. The estimates for blood sucrose or fructose with dental caries, BMI and blood pressure in women were similar to those in both sexes. Seventh, we could not check whether the associations varied by age, or for non-linear effects. Eighth, canalization, i.e., compensatory mechanisms driving the association of genetic variants with sucrose or fructose, may result in different associations in Mendelian randomization than would be observed from interventions. However, whether that exists is unknown. Finally, although the estimates were close to the null, we cannot rule out a small effect for CHD, which should be investigated further. We only have 80% power to detect an OR of about 1.05 for sucrose on CHD. However, the estimates for adiposity, blood pressure and lipids were in the opposite direction to those expected.

Conclusions
Our study found some evidence of a detrimental effect of blood sucrose on dental caries. However, our study suggests no effect of blood sucrose on T2D, adiposity, blood pressure, lipids, or glycaemic traits, whilst it remains possible that there could be a small effect of sucrose on CHD risk.

Methods
Genetic instruments for blood sucrose and fructose. Genetic variants, i.e., SNP, strongly (P < 5 × 10 -6 ) predicting serum sucrose (log 10 transformed median-normalized values) were extracted from an existing GWAS of serum metabolites adjusted for sex and age, obtained from non-targeted metabolomics using four www.nature.com/scientificreports/ independent ultra-high-performance liquid chromatography-tandem mass spectrometry 24 , in 1439 individuals (median age 58 years, 96.6% women) of European ancestry 24 . These associations were not adjusted for principal components because there was no evidence of inflation or deflation due to population stratification 24 . SNPs strongly (P < 5 × 10 -6 ) predicting blood fructose (log 10 transformed normalized) were extracted from an existing GWAS of blood metabolites adjusted for age, sex, batch and genomic control 26 , obtained using liquid-phase chromatography, gas chromatography separation and tandem mass spectrometry 26 in 7824 individuals (mean age 55 years, 83.5% women) of European ancestry 26 . SNPs were discarded if they were non-biallelic or had a minor allele frequency less than 0.01.
Genetic associations with dental caries, CHD, and T2D. Genetic associations with dental caries (n case = 2110, n control = 359,084) were extracted from genetic summary statistics from the UK Biobank in ~ 361,194 white British individuals (mean age 57.5 years) 43 provided by Neale Lab (http://www.neale lab.is/uk-bioba nk/). Dental caries was defined as hospital inpatient admission due to International Classification of Diseases (ICD) 10 K02. These genetic associations were obtained from linear regresstion adjusted for age, age 2 , sex, interactions of sex with age and age 2 , and 20 principal components. The linear regression estimates were transformed to log odds ratios (OR) using an approximation 44 . Genetic associations with DFSS (inverse normal rank transformed) and DMFS (standard deviation of residuals) were taken from summary statistics from the Gene-Lifestyle Interactions in Dental Endpoints (GLIDE) consortium in up to 26,792 individuals of European ancestry adjusted for age, sex, and study-specific covariates 45 . Genetic associations with CHD were taken from summary statistics of the Coronary Artery DIsease Genome wide Replication and Meta-analysis plus Coronary Artery Disease (CAR-DIoGRAMplusC4D) consortium from a meta-analysis (n case = ~ 76,014, n control = ~ 264,785) of CARDIoGRAM-plusC4D 1000 Genomes, the Myocardial Infarction Genetics and CARDIoGRAM Exome, the UK Biobank SOFT CAD GWAS, and two small case-control studies in individuals of mainly European ancestry (77% in CARDIoGRAMplusC4D 1000 Genomes and 94% in the UK Biobank SOFT CAD GWAS) 46 with genetic control. CHD cases included fatal or nonfatal myocardial infarction, percutaneous transluminal coronary angioplasty, coronary artery bypass grafting, chronic ischemic heart disease and angina determined from clinical diagnosis, medical records and self-report of medication usage. Controls were those who were not cases. Genetic associations with T2D were obtained from summary statistics from the DIAbetes Genetics Replication And Metaanalysis (DIAGRAM) consortium from a meta-analysis of case (n = 74,124)-control (n = 824,006) studies 47 in individuals of European ancestry adjusted for study-specific covariates and population structure but not BMI.
Genetic associations with adiposity, blood pressure, lipids, and glycaemic traits. Genetic associatons with inverse-normally transformed BMI were obtained from summary statitsics from a meta-analysis of the Genetic Investigation of ANthropometric Traits (GIANT) Consortium and the UK Biobank in up to 806,834 individuals of European ancestry adjusted for age, age 2 , assessment centre, and sex 48 . Genetic associations with inverse-rank normalized waist circumference (WC), body fat percentage (BFP), high-density lipoprotein cholesterol (HDL-C), LDL-C, and triglyceride were obtained from UK Biobank summary statitsics 43 provided by Neale Lab (http://www.neale lab.is/uk-bioba nk/) in ~ 361,194 white British individuals adjusted for age, age 2 , and 20 principal components, sex and interactions of sex with age and age 2 . Genetic associations with inverse-normally transformed HDL-C, LDL-C, and triglyceride were also taken from GLGC in ~ 188,578 individuals of European ancestry adjusted for age, age 2 , and sex 49 . Genetic associations with SBP and DBP were obtained from a metaanalysis of the UK Biobank and the International Consortium of Blood Pressure-Genome Wide Association Studies (ICBP) in ~ 757,601 individuals of European ancestry adjusted for sex, age, age 2 , BMI, and genotyping chips 50 . Genetic associations with glycaemic traits, including indices of β-cell function (HOMA-B) and insulin resistance (HOMA-IR), and fasting insulin (all naturally log transformed), were taken from a meta-analysis of GWAS including the UK Biobank in up to 46,186 individuals of European ancestry adjusted for age, sex and study specific covariates with genomic control 51 provided by the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC).

Statistical analysis.
We obtained uncorrelated SNPs (r 2 ≤ 0.001) using the "ld_clump" function of "ieugwasr" R package. For SNPs not available for an outcome, we used proxies (r 2 ≥ 0.8) identified from LDlink (https ://analy sisto ols.nci.nih.gov/LDlin k/). We aligned genetic variants for exposure and outcome on effect allele as none were palindromic. We estimated the F-statistic for each SNP using an approximation 52 , where a F-statistic of 10 or above indicated a lower risk of weak instrument bias. We checked whether any of the SNPs were associated with key confounders (i.e., Townsend deprivation index, education, smoking, alcohol drinking, and moderate physical activity) in the UK Biobank at Bonferroni corrected significance (0.05 divided by the number of SNPs times the number of confounders) (http://www.neale lab.is/uk-bioba nk). We checked for known pleiotropic effects using two curated genotype-to-phenotype cross-references, Ensembl 99 (http://www.ensem bl.org/index .html, accessed April 15, 2020), and PhenoScanner (http://www.pheno scann er.medsc hl.cam.ac.uk, accessed April 15, 2020). We obtained Mendelian randomization estimates by meta-analysing SNP-specific Wald estimates (ratio of SNP on outcome to SNP on exposure) 23 using inverse variance weighting (IVW) with multiplicative random effects which allows for heterogeneity but assumes balanced pleiotropy. We also used a weighted median which is valid as long as < 50% of the weight are contributed by invalid SNPs 53 , and MR-Egger which is valid as long as the instrument strength independent of direct effect assumption (i.e., the size of the pleiotropic effects is independent of their strength as instruments) holds 52 . MR-Egger provides a check on the validity of the IVW estimate, with a significant p-value (< 0.05) for the intercept indicating that the IVW estimate may be invalid 52 . We also Scientific Reports | (2020) 10:21588 | https://doi.org/10.1038/s41598-020-78685-5 www.nature.com/scientificreports/ used the Mendelian randomization pleiotropy residual sum and outlier (MR-PRESSO) test which can correct for potential horizontal pleiotropy (i.e., the variant has an effect on disease outside of its effect on the exposure) 54 . We estimated the expected effect size of the association of blood sucrose with CHD for 80% power. At an R 2 (the proportion of variance in blood sucrose explained by the genetic variants) of 6.2%, we have 80% power to detect an OR of ∼1.05 for sucrose on CHD. Here, we are testing a hypothesis, the effect on CHD with related secondary outcomes (CHD risk factors) and a control outcome (dental caries), for comprehensiveness we considered a p-value of 0.05 and a Bonferroni-corrected p-value of 0.05/7 = 0.007 (given 7 outcomes including CHD, 5 CHD risk factors and dental caries) as significant. All statistical analyses were performed using R version 3.6.3 (The R Foundation for Statistical Computing, Vienna, Austria) 55 and the "MendelianRandomization" and "MRPRESSO" R packages. This study requires no ethical approval because only publicly available summary data was used.

Data availability
The data are available in the consortia.