1. Epidemiology and Global Health
  2. Genetics and Genomics
Download icon

Improved lipidomic profile mediates the effects of adherence to healthy lifestyles on coronary heart disease

  1. Jiahui Si
  2. Jiachen Li
  3. Canqing Yu
  4. Yu Guo
  5. Zheng Bian
  6. Iona Millwood
  7. Ling Yang
  8. Robin Walters
  9. Yiping Chen
  10. Huaidong Du
  11. Li Yin
  12. Jianwei Chen
  13. Junshi Chen
  14. Zhengming Chen
  15. Liming Li  Is a corresponding author
  16. Liming Liang  Is a corresponding author
  17. Jun Lv  Is a corresponding author
  1. Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, China
  2. Departments of Epidemiology and Biostatistics, Harvard T.H. Chan School of Public Health, United States
  3. Peking University Institute of Public Health & Emergency Preparedness, China
  4. Chinese Academy of Medical Sciences, China
  5. Medical Research Council Population Health Research Unit at the University of Oxford, United Kingdom
  6. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, United Kingdom
  7. NCDs Prevention and Control Department, Hunan Center for Disease Control & Prevention, China
  8. Liuyang Center for Disease Control & Prevention, Liuyang, China
  9. China National Center for Food Safety Risk Assessment, China
  10. Key Laboratory of Molecular Cardiovascular Sciences (Peking University), Ministry of Education, China
Research Article
  • Cited 1
  • Views 1,084
  • Annotations
Cite this article as: eLife 2021;10:e60999 doi: 10.7554/eLife.60999

Abstract

Adherence to healthy lifestyles is associated with reduced risk of coronary heart disease (CHD), but uncertainty persists about the underlying lipid pathway. In a case–control study of 4681 participants nested in the prospective China Kadoorie Biobank, 61 lipidomic markers in baseline plasma were measured by targeted nuclear magnetic resonance spectroscopy. Baseline lifestyles included smoking, alcohol consumption, dietary habit, physical activity, and adiposity levels. Genetic instrument was used to mimic the lipid-lowering effect of statins. We found that 35 lipid metabolites showed statistically significant mediation effects in the pathway from healthy lifestyles to CHD reduction, including very low-density lipoprotein (VLDL) particles and their cholesterol, large-sized high-density lipoprotein (HDL) particle and its cholesterol, and triglyceride in almost all lipoprotein subfractions. The statins genetic score was associated with reduced intermediate- and low-density lipoprotein, but weak or no association with VLDL and HDL. Lifestyle interventions and statins may improve different components of the lipid profile.

Introduction

Coronary heart disease (CHD) has become one of the leading causes of death worldwide (Roth et al., 2018). In China, the mortality rate from CHD increased almost four times from 2002 to 2014 (Chen et al., 2017). It is widely acknowledged that unhealthy lifestyles, such as smoking, excess alcohol consumption, inadequate physical activity, unhealthy diet, and adiposity, are major risk factors for CHD (Dariush et al., 2008). Studies on the impacts of adherence to a combination of healthy lifestyle factors (HLFs) on mortality (Li et al., 2018; Zhu et al., 2019), healthy life expectancy (Li et al., 2020), and risk of type 2 diabetes (Lv et al., 2017a) and cardiovascular diseases in the Chinese population (Lv et al., 2017b) have provided important information on the maximum public health benefit that lifestyle intervention could achieve.

Atherogenic dyslipidemia is one of the well-documented risk factors for CHD (Shaima et al., 2016; Peters et al., 2016). Conventional lipid markers fail to distinguish between the size, density, concentration, or composition of lipoprotein particles, which may have contrasting effects on CHD risk (Holmes et al., 2018; Würtz et al., 2015). Lipidomics provides a detailed snapshot of the systemic lipid profile beyond routine lipid markers. Only a few studies have examined the association between lipidomic profile and individual HLFs separately (Kujala et al., 2013; Würtz et al., 2016a; Würtz et al., 2014; Lankinen et al., 2014), which, however, typically correlates with one another. It is mostly unknown how much of the effects of combined HLFs on reduced CHD risk are mediated through an improved lipid profile and what the differences are between components of the lipid profile in their mediating effects.

Statins are HMG-CoA reductase (HMGCR) inhibitors, which reduce the low-density lipoprotein cholesterol (LDL-C) by interfering with the cholesterol-biosynthetic pathway and have become one of the first-line therapy options for dyslipidemia (Sirtori, 2014). Mendelian randomization studies constructed in the European population observed lipid-lowering effect of statins beyond the anticipated decrease in LDL particles (Ference et al., 2019; Würtz et al., 2016b). However, such genetic effects have not been examined in Asian populations. No study compares the effects of healthy lifestyle and genetically inferred lipid-lowering medications on lipidomic profile in the same set of study participants.

The primary aims of the present study were to examine the combined effect of HLFs on components of a comprehensive lipidomic profile measured by nuclear magnetic resonance (NMR) spectroscopy (Soininen et al., 2015), and further quantify how much of the combined effects of HLFs on CHD reduction are mediated through lipid metabolites. We also estimated the clinical effect of statins and bempedoic acid, a novel therapeutic approach by inhibiting ATP citrate lyase (ACLY) (Pinkosky et al., 2016) on lipidomic profile by creating a Chinese specific genetic score for HMGCR and ACLY functions. Finally, we examined the joint effects of HLFs and lipid-lowering medications on lipidomic profile. We did a nested case–control study comprising incident CHD cases, stroke cases, and controls identified from the 10-year follow-up of the China Kadoorie Biobank (CKB). We included all eligible participants to examine the impact of HLFs and genetic scores on lipid metabolites, and only included CHD cases and controls in the further mediation analysis.

Results

The mean age of 4681 participants was 46.7 ± 8.0 years. Five HLFs included never smoking, moderate alcohol consumption, having a healthy dietary score ≥4, being physically active, and healthy adiposity levels. Of the 4681 participants, 0.2%, 11.1%, and 47.4% had at least 5, 4, and 3 HLFs, respectively. The overall mean (SD) LDL-C concentration measured by the clinical chemistry assay was 88.8 (27.0) mg/dl. Younger, female, and more educated participants were more likely to adopt a healthy lifestyle (Table 1). Compared with control participants, the CHD cases were older, were less likely to be women, and had a higher prevalence of hypertension and diabetes at baseline (Supplementary file 2A).

Table 1
Age-, sex-, and study area-adjusted baseline characteristics of 4681 participants according to the number of healthy lifestyle factors (HLFs).

The results are presented as adjusted means or percentages, with adjustment for age, sex, and study area, as appropriate. All baseline characteristics were associated with the number of HLFs, with p<0.05 for trend across categories, except for urban or rural residence (0.155), family history of heart attack (p=0.905), and consumption of fresh vegetables (0.065).

Baseline characteristics
0123≥4
No. of participants, n (%)118
(2.5)
688
(14.7)
1656
(35.4)
1698
(36.3)
521
(11.1)
Age, year49.649.047.545.843.8
Female, %5.122.646.462.967.5
Urban area, %42.633.527.825.136.6
Middle school and above, %51.553.653.857.958.2
Married, %91.392.994.695.395.2
Prevalent hypertension, %62.652.648.740.638.5
Prevalent diabetes, %13.812.07.14.13.8
Family history of heart attack, %2.85.04.74.24.8
Having HLFs*, %
Never smoking47.157.070.685.9
Moderate alcohol consumption3.58.615.928.3
Being physically active13.337.666.296.0
Healthy dietary pattern23.941.460.294.0
Vegetables 7 days/week90.893.292.593.097.6
Fruit 7 days/week2.67.610.216.324.8
Read meat <7 days/week52.965.773.776.784.8
Soybean product ≥4 days/week2.44.07.010.719.2
Fish ≥1 day/week18.323.930.338.550.3
Coarse grains ≥4 days/week7.921.122.723.725.0
Healthy adiposity level25.855.589.098.0
  1. *HLFs were defined as: never smoking; weekly but not daily drinking or daily drinking less than 30 g of pure alcohol; engaging in a sex-specific median or higher level of physical activity; engaging in more than or equal to 4 of total six healthy diet components; having a body mass index between 18.5 and 27.9 kg/m2 and having a waist circumference <90 cm in men and <85 cm in women.

Associations of combined HLFs with lipid metabolites

Adherence to combined HLFs was associated with 50 components of the lipid profile (false dicovery rate [FDR] < 0.05). Compared with participants who adopted at most one HLF, the differences in the lipid metabolites, especially VLDL- and HDL-related measures, increased with the number of HLFs adhered to (Figures 13). Participants with four to five HLFs had lower VLDL particle concentrations and smaller VLDL particle size, with adjusted SD difference (95% CI) ranging from −0.54 (−0.66,–0.43) for large VLDL to −0.27 (−0.38,–0.15) for very small VLDL, and −0.47 (−0.59,–0.36) for VLDL diameter (Figure 1, Figure 1—source data 1). For HDL, adherence to four to five HLFs was associated with higher HDL particle concentrations and larger HDL particle size, with maximum SD difference (95% CI) of 0.39 (0.28, 0.50) for large HDL, and 0.32 (0.21, 0.43) for HDL diameter. The corresponding SD difference (95% CI) for apolipoprotein B/apolipoprotein A1 was −0.45 (−0.56,–0.34), resulting from higher apolipoprotein A1 and lower apolipoprotein B.

Figure 1 with 13 supplements see all
Associations of size-specific lipoprotein particle concentrations, mean lipoprotein particle diameter, and apolipoprotein concentrations with combined healthy lifestyle and risk of coronary heart disease.

(a) SD difference and 95% CI are for comparison of participants who adopted two to three or four to five combined healthy lifestyles with participants who adopted zero to one. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and case/control status. (b) Odds ratio and 95% CI are for the associations of 1-SD metabolic markers increasing with CHD risk. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and smoking status. Horizontal lines represent 95% CIs. ApoA1 = apolipoprotein A1; ApoB = apolipoprotein B; CHD = coronary heart disease; HDL = high-density lipoprotein; IDL = intermediate-density lipoprotein; LDL = low-density lipoprotein; Sig. = significance ***p≤0.0001, **p≤0.01, *p≤0.05, – p>0.05 (false discovery rate [FDR]–adjusted p-values); VLDL = very low-density lipoprotein.

Figure 1—source data 1

Associations of size-specific lipoprotein particle concentrations, mean lipoprotein particle diameter, and apolipoprotein concentrations with combined healthy lifestyle and risk of coronary heart disease.

https://cdn.elifesciences.org/articles/60999/elife-60999-fig1-data1-v1.docx
Associations of cholesterol concentrations in lipoprotein subfractions with combined healthy lifestyle and risk of coronary heart disease.

(a) SD difference and 95% CI are for comparison of participants who adopted two to three or four to five combined healthy lifestyles with participants who adopted zero to one. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and case/control status. (b) Odds ratio and 95% CI are for the associations of 1-SD metabolic markers increasing with CHD risk. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and smoking status. Horizontal lines represent 95% CIs. CHD = coronary heart disease; HDL2 = larger HDL particles; HDL3 = smaller HDL particles; Sig. = significance ***p≤0.0001, **p≤0.01, *p≤0.05, – p>0.05 (false discovery rate [FDR]–adjusted p-values); other abbreviations as in Figure 1.

Figure 2—source data 1

Associations of cholesterol concentrations in lipoprotein subfractions with combined healthy lifestyle and risk of coronary heart disease.

https://cdn.elifesciences.org/articles/60999/elife-60999-fig2-data1-v1.docx
Associations of triglyceride concentrations in lipoprotein subfractions with combined healthy lifestyle and risk of coronary heart disease.

(a) SD difference and 95% CI are for comparison of participants who adopted two to three or four to five combined healthy lifestyles with participants who adopted zero to one. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and case/control status. (b) Odds ratio and 95% CI are for the associations of 1-SD metabolic markers increasing with CHD risk. Multivariable model was adjusted for: age, sex, fasting time, study areas, education level, and smoking status. Horizontal lines represent 95% CIs. CHD = coronary heart disease; Sig. = significance ***p≤0.0001, **p≤0.01, *p≤0.05, – p>0.05 (false discovery rate [FDR]–adjusted p-values); Abbreviations as in Figure 1.

Figure 3—source data 1

Associations of triglyceride concentrations in lipoprotein subfractions with combined healthy lifestyle and risk of coronary heart disease.

https://cdn.elifesciences.org/articles/60999/elife-60999-fig3-data1-v1.docx

The associations of combined HLFs with cholesterol concentrations in lipoprotein subfractions were very similar to the associations with the corresponding lipoprotein particle concentrations (Figure 2, Figure 2—source data 1). Combined HLFs were consistently associated with lower TG concentrations in all lipoprotein subfractions except large HDL particles. The adjusted SD difference (95% CI) for participants with four to five HLFs ranged from −0.55 (−0.66,–0.43) for small VLDL-TG to −0.29 (−0.41,–0.18) for medium LDL-TG (Figure 3, Figure 3—source data 1).

The linear associations between each one factor increase in HLFs and lipid profile were illustrated in Figure 1—figure supplement 1. In sensitivity analyses, we further adjusted for prevalent diabetes, restricted analyses to control participants (Supplementary file 2B), did not adjust for fasting time, used more strict body mass index (BMI) and waist circumference (WC) cut-off points to define healthy adiposity, or excluded moderate alcohol consumption from the HLF definition; the associations between HLFs and metabolites were not substantially altered (Figure 1—figure supplements 24).

Associations of individual HLFs with lipid metabolites

Of the five individual HLFs analyzed, moderate alcohol consumption (Figure 1—figure supplement 5), being physically active (Figure 1—figure supplement 6), and having healthy adiposity levels (Figure 1—figure supplement 7), had the most significant influence on lipid metabolites. Participants who were physically active or had healthy adiposity levels had a cardioprotective lipid profile, with lower concentrations of VLDL-related measures, apolipoprotein B, and higher concentrations of larger HDL particles. The maximum SD differences (95% CI) related to physical activity (Figure 1—figure supplement 6) and healthy adiposity level (Figure 1—figure supplement 7) were −0.12 (−0.18,–0.06) for medium VLDL-TG and −0.54 (−0.60,–0.48) for total TG, respectively. Sensitivity analysis using more strict BMI and WC cut-off points (BMI in the range of 18.5–24.9 kg/m2 and WC <90 cm in men and <80 cm in women) observed similar and generally stronger associations between healthy adiposity level and lipidomic profile (Figure 1—figure supplement 8).

Moderate alcohol consumption was associated with higher concentrations of VLDL- and HDL-related measures and apolipoprotein A1. The maximum SD difference (95% CI) was 0.29 (0.19, 0.38) for small HDL particle concentration (Figure 1—figure supplement 5). We further divided participants into three groups according to their alcohol consumption at baseline: non-regular, moderate, and heavy use. Compared with non-regular use group, both heavy (with ≥30 g of pure alcohol per day) and moderate alcohol use (<30 g per day) had a similar pattern of effects on lipid metabolites, with the most significant changes observed in participants with heavy alcohol use (Figure 1—figure supplements 911).

Smoking (Figure 1—figure supplement 12) and dietary habit (Figure 1—figure supplement 13) had a relatively small impact on lipid metabolites.

Mediation effects of lipid metabolites in the association between HLFs and CHD risk

We restricted the following analyses in 927 incident CHD cases and 1513 controls. Incident CHD cases were those who developed fatal ischemic heart disease and nonfatal myocardial infarction during follow-up. The associations between lipid metabolites and CHD risk generally mirrored the associations between combined HLFs and lipid metabolites (Figures 13). None of the lipid metabolites showed interactions with the HLFs in their effect on CHD risk (all pinteration > 0.05). A total of 35 lipid metabolites showed statistically significant mediation effects from combined HLFs to CHD reduction (FDR ranging from <0.001 to 0.042). The proportions of reduced CHD risk associated with combined HLFs mediated by VLDL particle concentration ranged from 4.77% for very small VLDL to 10.15% for small VLDL (Figure 1, Figure 1—source data 1). Other strong mediators included large HDL (8.02%), apolipoprotein B (8.36%), and apolipoprotein B/apolipoprotein A1 (11.32%). For cholesterol, compared to LDL-C, VLDL- and HDL-C were relatively strong mediators (Figure 2, Figure 2—source data 1). TG carried within all lipoproteins (except for large-sized HDL) showed statistically significant mediating effects, with the maximum proportion of 10.47% for small VLDL-TG (Figure 3, Figure 3—source data 1). The top five principal components of all lipid metabolites mediated 14.05% of the reduced CHD risk associated with combined HLFs.

HMGCR and ACLY scores, HLFs, and lipid metabolites

The HMGCR and ACLY scores had a similar pattern of effects on lipid metabolites, with higher scores mainly associated with decreased concentrations of intermediate-density lipoprotein (IDL)- and LDL-related measures and apolipoprotein B (Supplementary file 2C and D). The sum of HMGCR and ACLY scores was associated with stronger changes in the above lipid metabolites (Supplementary file 2E). Use of genetic scores based on the European population (Ference et al., 2019) for HMGCR and ACLY observed similar but weaker associations (Supplementary file 2C–F).

In the joint association analysis of HLFs and HMGCR score with lipid metabolites, compared with participants who had higher genetic risk (median cutoffs) and adhered to zero to two HLFs, those with lower genetic risk and three to five HLFs had the most cardioprotective lipidomic profile, including 0.36 SD decrease in VLDL-C, 0.13 SD decrease in LDL-C, and 0.21 SD increase in HDL-C (Figure 4 and Figure 4—figure supplement 1 for ACLY score). We further compared the effect patterns of each one factor increase in HLFs with a 2-SD increase in HMGCR or ACLY score on lipid metabolites (Figure 4—figure supplements 2 and 3). The combined HLFs, as opposed to the effect by HMGCR and ACLY scores, were associated with lower VLDL-related measures, apolipoprotein B, and TG in almost all lipoprotein subfractions, and with higher HDL and HDL-C concentrations.

Figure 4 with 3 supplements see all
Joint association of combined healthy lifestyle and HMGCR scores based on Chinese population with lipid metabolites.

Participants who had higher genetic risk regarding HMGCR (3-hydroxy-3-methylglutaryl–coenzyme A reductase) and adhered to zero to two healthy lifestyle factors (HLFs) were reference group. SD difference and 95% CI of log-transformed lipid metabolites for participants with lower genetic risk and 0-2 HLFs, higher genetic risk and 3-5 HLFs, and lower genetic risk and 3-5 HLFs were shown in red, blue, and green, respectively. Abbreviations as in Figure 1.

When we stratified participants according to the score of HMGCR, ACLY, or their sum score, the associations between each one factor increase in HLFs and lipid metabolites were generally similar between high- and low- genetic risk stratum (all pinteraction >0.05) (Supplementary file 2C–E).

Discussion

In this prospective study of middle-aged Chinese, participants who adhered to healthy lifestyles tended to have a more cardioprotective lipidomic profile, which jointly mediated 14% of the protective effect of combined HLFs on CHD reduction. Similar to results in the European population, genetic scores for the targets of statins and ACLY inhibitors showed similar effects on reducing concentrations of IDL- and LDL-related measures, while the underlying mechanisms of lifestyle intervention were more strongly related to VLDL- and HDL-related measures, apolipoprotein B, and TG in almost all lipoprotein subfractions.

Our findings on the associations of lipid metabolites with individual lifestyle-related characteristics like physical activity, adiposity, and alcohol consumption were generally consistent with previous studies (Kujala et al., 2013; Würtz et al., 2016a; Würtz et al., 2014). One of the studies used the Mendelian randomization to indicate causal adverse effects of increased adiposity on lipoprotein subclass profiles within the non-obese weight range among young Finland adults (Würtz et al., 2014). Another study similarly showed a mixture of favorable and adverse effects of alcohol consumption on the lipid profile in relation to cardiovascular disease (Würtz et al., 2016a). Also, the lipoprotein lipid profile observed cross-sectionally was highly consistent with the pattern of their changes accompanying a change in alcohol consumption at 6-year follow-up. Numerous studies have found increased HDL level with higher alcohol consumption (Gepner et al., 2015; Brien et al., 2011). However, the association between alcohol consumption and apolipoprotein B-carrying lipoprotein is less clear (Brien et al., 2011; Roerecke and Rehm, 2014). Our results noted that alcohol consumption showed divergent relationships with different sized apolipoprotein B-carrying particles, for example, higher concentration of large-sized VLDL and lower concentration of IDL and LDL. Although the explanation for the complex association between alcohol consumption and lipid profile remains inconclusive, our detailed investigation of lipoprotein subclasses provides improved understanding of the diverse molecular process related to alcohol consumption. Regarding diet, a previous randomized trial found that diet in rich of whole grain, bilberries, and fatty fish caused changes in HDL particles (Lankinen et al., 2014). The present study characterized a healthy diet differently and found significant differences in VLDL-related measures but not HDL.

This is the first study, to our knowledge, to assess the combined effect of HLFs on a comprehensive lipidomic profile. The results indicated that participants with healthy lifestyles were characterized by an antiatherogenic lipidomic profile, which has been related to lower CHD risk previously (Varbo et al., 2013; Holmes et al., 2015; Peter et al., 2015; Natarajan et al., 2010; Inouye et al., 2010). The positive associations of combined HLFs with HDL and HDL-C were limited to large and medium subclasses, in line with previous studies which suggested that small HDL particles did not have protective effects on CHD (Peter et al., 2015; Natarajan et al., 2010; Inouye et al., 2010). Notably, the TG levels within all apolipoprotein B and most HDL particles were lower in participants who adopted healthy lifestyles. It is plausible that healthy lifestyles have opposing relationships with HDL-TG and HDL-C.

Limited prospective studies have investigated the mediating effects of individual HLFs on CHD through total cholesterol, suggesting that it mediates 13%, 8%, and 18% of the excess CHD risk related to inadequate physical activity (Mora et al., 2007), obesity, and overweight (Global Burden of Metabolic Risk Factors for Chronic Diseases Collaboration (BMI Mediated Effects), 2014), respectively. Our results showed all NMR-measured lipid metabolites jointly explained 14% of the protective effect of combined HLFs on CHD risk. We further highlighted the differences between various lipid metabolites in their mediating effects. The apolipoprotein B/A1 ratio was among the most influential mediators and has been previously reported to be a better predictor of CHD risk than any of the cholesterol ratios (McQueen et al., 2008). A lower apolipoprotein B/A1 ratio, together with manifestation of other metabolites, suggested that adherence to HLFs can reduce the risk of CHD through both lower proatherogenic and higher antiatherogenic lipoproteins.

Both statins and ACLY inhibitors have been associated with lowering LDL- and IDL-related measures and further with a reduction in CVD risk in western populations (Ference et al., 2019; Würtz et al., 2016b). In the present study, we used genetic scores to mimic the effects of these two LDL-C lowering targets and observed similar effects on the lipidomic profile. For LDL and IDL particle concentrations, adherence to two or more HLFs could achieve a similar beneficial effect as a 2-SD change in ACLY/HMGCR genetic scores. In other words, two or more HLFs could compensate for the deleterious effect on lipid metabolites due to inheriting risk alleles in these genes. More importantly, we found that adherence to combined HLFs had a much stronger effect on other components of the lipidomic profile than LDL- and IDL-related measures, including VLDL- and HDL-related components.

To our knowledge, this is the first study to reveal the potential underlying lipid pathways that may mediate the effects of adherence to combined HLFs on lower CHD risk. The strengths of the study include the prospective outcome ascertainment, a comprehensive assessment of lifestyle factors, and a population free of lipid-modifying therapy at the time of blood collection. The measurements of multiple lipids and lipoprotein particles provide a detailed snapshot of the systemic lipid profile. The concordance of measurements by both NMR spectroscopy and clinical chemistry assays, and by duplicate samples of NMR metabolomics provided evidence to support internal validity. The availability of genotyping data allowed us to use a Mendelian randomization approach to estimate the effects of statins and ACLY inhibition on lipid metabolites while avoiding potential confounding bias by indication.

Our study has limitations. First, the lifestyle behaviors were self-reported once at baseline. Second, lifestyle behaviors and lipid metabolites were measured at the same time. However, previous evidence supports the causative effects of individual HLFs on lipid metabolites and the resemblance between the cross-sectional and longitudinal association patterns (Würtz et al., 2016a; Würtz et al., 2014). Third, the lipid metabolites quantified by the NMR spectroscopy assay did not include some important measures such as lipoprotein(a), apolipoprotein CIII, and HDL functionality. There are also strong correlations between the lipid metabolites, with multiple measures representing the same underlying lipid fractions. As a result, the mediating role of lipid metabolites in the present study cannot extrapolate to that of the complete lipidomic profile and also cannot differentiate the individual mediating role of each lipid metabolites. Nevertheless, this does not detract from the value of the study in identifying potential pathways underlying the HLFs and CHD risk. Fourth, the effect of therapeutic agents mimicked by genetic scores is the effect of lifelong exposure to a biomarker on an outcome that is difficult to be translated into the expected effect of short-term pharmacologic changes (Ference et al., 2019). However, genetic scores served mainly for comparison of the underlying mechanisms of lifestyle interventions and lipid-lowering medications, rather than the effect size. Lastly, a more sophisticated HLFs score with appropriated weight might show stronger association. However, a more straightforward definition would be easier to understand and adapted by the public. Also, mediation analysis might be biased when the continuous exposure variable was dichotomized. Our simulation showed that this requires a particularly strong association between exposure and mediator, which was far from the realistic association between HLFs and metabolite in our study.

The present study of Chinese adults elucidated that the effects of adherence to a combination of HLFs on lower CHD risk were partly mediated by an improved lipid profile. Lifestyle interventions and lipid-lowering medication therapies may affect different components of the lipid profile, suggesting that they are not redundant strategies but could be combined for better benefits.

Materials and methods

Study population

Request a detailed protocol

The CKB is a prospective cohort of 512,715 adults (aged 30–79 years) from 10 geographically diverse areas across China (five urban sites and five rural sites) during 2004–2008. Details of the study design, survey methods, and long-term follow-up have been given elsewhere (Chen et al., 2005; Chen et al., 2011). Briefly, all participants had baseline data collected by questionnaire, including sociodemographic, lifestyle factors, and medical and medication history, and physical measurements. Participants also provided a 10 ml random blood sample for long-term storage, with the time since last meal recorded. Mortality and morbidity during follow-up were identified through linkage with local death and disease registries, with the national health insurance system, and by active follow-up if necessary (i.e., visiting local communities or directly contacting participants). Since 2014, 97% of the participants have been linked to the health insurance databases. By December 31, 2015, of all the cohort participants, only 4875 (<1%) were lost to follow-up. The mean follow-up duration of the cohort since baseline was 9.2 (1.4) years.

The study protocol was approved by the Ethics Review Committee of the Chinese Center for Disease Control and Prevention (005/2004, Beijing, China) and the Oxford Tropical Research Ethics Committee, University of Oxford (025–04, UK). All participants provided written informed consent.

Design of the present study

Request a detailed protocol

A subset of 4681 CKB participants was selected for metabolomics measurements in a nested case–control study of incident CHD and stroke occurring before the censoring date of January 1, 2015 (Holmes et al., 2018). Cases were those who had a newly developed fatal or nonfatal disease during follow-up: (1) CHD: fatal ischemic heart disease coded as ICD-10 I20-I25 and nonfatal myocardial infarction coded as I21-I23 (n = 927); (2) ischemic stroke: ICD-10 I63 or I69.3 (n = 1114); (3) intracerebral hemorrhage: ICD-10 I61 or I69.1 (n = 1127). Case status was defined as the disease first occurred in each participant. Common controls were selected by frequency matching to combined cases by age, sex, and study area (n = 1513). The diagnosis adjudication has finished for 34,000 reported cases of ischemic heart disease by a review of hospital medical records. Overall, 88% of the diagnoses were confirmed. All case and control participants did not report doctor-diagnosed CHD, stroke, transient ischemic attack, or cancer, and were not using statins and other lipid-lowering medications at baseline. Of the 4681 participants, 4592 had genotyping information, which was generated using a customized Affymetrix Axiom array including ~800,000 SNPs and further imputed to the 1000 Genomes reference panel (Phase 3) using IMPUTE v2.

Measurement of lipid metabolites

Request a detailed protocol

A high-throughput targeted NMR metabolomics platform (Soininen et al., 2015) was used for quantification of circulating lipid metabolites in baseline plasma samples (Brainshake Laboratory at Kuopio, Finland). All metabolites were assayed simultaneously. Cases and controls were measured in random order, with laboratory staff blinded to case/control status. Of the 4681 participants, 137 had duplicated measurements. The median coefficient of variation for duplicates was 5.0% (interquartile range: 2.7–6.7%) (Holmes et al., 2018). Six traits covered by NMR spectroscopy were also measured using standard clinical chemistry assays including total cholesterol, LDL-C, high-density lipoprotein cholesterol (HDL-C), triglyceride (TG), apolipoprotein B, and apolipoprotein A1 (Wolfson Laboratory at University of Oxford, UK). There were high correlations between NMR and clinical chemistry measured traits, with the correlation ranging from 0.80 to 0.90 (Holmes et al., 2018).

Definition and assessment of HLFs

Request a detailed protocol

We included five baseline lifestyle-related characteristics: smoking, alcohol consumption, dietary habit, physical activity, and body weight and fat to assess energy balance (Lloyd-Jones et al., 2010). In the baseline questionnaire, for smoking, we asked frequency, type, and amount of tobacco smoked per day for ever smokers, and years since quitting and reason to quit for former smokers. For alcohol consumption, we asked drinking frequency on a week, type of alcoholic beverage, and volume of alcohol consumed on a typical drinking day. For physical activity, we asked the usual type and duration of activities. The daily level of physical activity was calculated by multiplying the metabolic equivalent tasks (METs) value for a particular type of physical activity by hours spent on that activity per day and summing the MET hours for all activities. For dietary habit, we used a short qualitative food frequency questionnaire to assess habitual intakes of 12 conventional food groups (Supplementary file 2G). For adiposity level, trained staff measured weight, height, and WC with calibrated instruments. BMI was calculated as weight in kilograms divided by the square of the height in meters.

The HLFs that may be related to lower CHD risk were defined as follows: (1) never smoking; (2) moderate alcohol consumption: weekly but not daily drinking, or daily drinking less than 30 g of pure alcohol; (3) having ≥4 of the total six healthy dietary habits that are particularly addressed in the Chinese dietary guidelines (2016) (Yang et al., 2018): consuming fresh vegetables every day, fresh fruits every day, red meat <7 days/week, soybean products ≥4 days/week, fish ≥1 day/week, and coarse grains ≥4 days/week; (4) being physically active, i.e. having a sex-specific median or higher level of physical activity; (5) healthy adiposity levels: BMI in the range of 18.5–27.9 kg/m2 (normal or overweight according to the standard classification specific for Chinese) and WC <90 cm in men and <85 cm in women (Chen et al., 2018; Jia et al., 2010).

Genetic scores for HMGCR and ACLY

Request a detailed protocol

We constructed the genetic scores in the Chinese population with a previously adopted method (Ference et al., 2019). This approach has been used to accurately anticipate the results of several randomized trials that have evaluated lipid-lowering therapies (Ference, 2018). First, we tested the association of each variant within a 500 KB window on either side of the HMGCR gene in a linear regression model, with plasma LDL-C as dependent variable and age, sex, and the top 10 ancestry-informative principal components as covariates in 13,060 participants from the CKB cohort without overlapping with the lipidomic data set. All 13,060 participants were not using statin and other lipid-lowering medications at baseline. Second, we pruned the variant by keeping the top variants with most significant p-value and removed other variants that were correlated with the selected variant (r2 > 0.3). Next, we tested the association between each remaining variant and LDL-C, conditional on previously selected variants and covariates, and selected the variant with the smallest p-value. We iteratively repeated this step until all variants were selected, removed due to linkage disequilibrium with a selected variant, or were not associated with LDL-C (p>0.05). The exposure allele for each selected variant was defined as the allele associated with lower plasma LDL-C. The weight for each variant was the conditional effect of that variant on LDL-C level in mmol/l adjusted for all other variants included in the score among the 13,060 participants.

We multiplied the number of exposure alleles that a participant inherited at each variant by their weights. We then summed these values to construct a weighted HMGCR genetic score for each participant in the present analysis. We used the same protocol to construct a weighted ACLY genetic score. We called this as genetic score based on Chinese population.

We also used previously constructed HMGCR and ACLY genetic scores (Ference et al., 2019) for comparison with studies in the European population. Three of the total nine variants included in the ACLY genetic score (rs113201466, rs145940140, and rs117981684) were monomorphism in the eastern Asian population. Only the other six variants were used to construct the genetic score for ACLY. Linear regression was used to estimate the effect of each variant on LDL-C level, with adjustment for age, sex, and the top 10 ancestry-informative principal components. We called this as genetic score based on European population. We also used the conditional effect reported from the previous study (Ference et al., 2019) to construct alternative weighted genetic scores for sensitivity analyses.

Variants included in the genetic scores and their association with LDL-C were provided in Supplementary file 2H.

Statistical analysis

Request a detailed protocol

We classified participants into three groups according to the number of HLFs they adopted: zero to one, two to three, and four to five. All lipid metabolites were inverse normal transformed (SD = 1), which is useful for comparing variables expressed in different units. The associations between combined HLFs and lipid metabolites were assessed using linear regression adjusted for age (years), sex (male or female), fasting time (<8 or ≥8 hr), 10 study areas, education level (no formal or primary school, middle or high school, technical school or college or higher), and case/control status, with participants who adopted zero to one HLF as the reference group. Logistic regression was used to estimate odds ratios (ORs) of CHD per 1-SD higher lipid metabolite levels, adjusted for age, sex, fasting time, study areas, education level, and smoking status. The additional adjustment was also made for other HLFs (alcohol consumption, dietary habit, physical activity, and BMI) and prevalent diabetes, with results largely unchanged (data not shown).

We used the paramed package (Emsley and Liu, 2013) to perform causal mediation analysis (Valeri and VanderWeele, 2013) using parametric regression models. For each lipid metabolite, two models were estimated: (1) a model for the mediator (the lipid metabolite itself) conditional on exposure (the number of HLFs as a continuous variable) and covariates (age, sex, fasting time, study areas, and education level) in the control participants only (n = 1513); (2) a model for the risk of CHD conditional on exposure, the mediator, and covariates (age, sex, fasting time, study areas, and education level) in both CHD case and control participants (n = 2440). We also allowed for the presence of exposure-mediator interactions in the outcome regression model. We aimed to access how much of the total effect (TE) is due to neither mediation nor interaction; how much is due to interaction but not mediation; how much is due to both mediation and interaction; and how much of the effect is due to mediation but not interaction (natural indirect effect [NIE]). We used the delta method to estimate standard errors and confidence intervals. If exposure-mediator interactions did not exist, the proportion attributable to the NIE was calculated by dividing the NIE by TE on log odds scale, with 0 indicating no mediation effect. We also used the top five principal components of all lipid metabolites that explained ≥95% of the total variation to estimate the joint mediating effect of all lipid metabolites.

The HMGCR and ACLY genetic scores were also inverse normal transformed to facilitate comparisons. We estimated the association of a 2-SD increase in the HMGCR or ACLY score with lipid metabolites using linear regression, with adjustment for age, sex, study areas, and the top 10 genotype principal components. We further examined the joint association of HLFs and genetic scores with lipid metabolites, by classifying participants into four groups according to their genetic score (median cutoffs) and number of HLFs (zeo to two or three to five). We also examined whether the association between HLFs and lipid metabolites differed by scores of HMGCR, ACLY, or their sum score, which were dichotomized according to the median cutoffs of the genetic scores. The tests for interaction were performed using likelihood ratio tests comparing models with and without the cross-product term.

All analyses were performed with Stata version 14.2 (StataCorp) and R software version 3.5.2 (R Foundation for Statistical Computing). p-values are presented as unadjusted for multiple testing unless otherwise indicated. For testing of multiple lipid metabolites, we used the FDR (Benjamini and Hochberg, 1995) <0.05.

Data availability

According to the Regulation of the People's Republic of China on the Administration of Human Genetic Resources, we are not allowed to provide Chinese human clinical and genetic data abroad without an official approval. We are providing our syntax of statistical analysis and the output from the analysis.

References

    1. Peter W
    2. Havulinna Aki S
    3. Pasi S
    (2015)
    Metabolite profiling and cardiovascular event risk
    Circulation 131:774–785.
    1. Roth GA
    2. Abate D
    3. Abate KH
    4. Abay SM
    5. Abbafati C
    6. Abbasi N
    7. Abbastabar H
    8. Abd-Allah F
    9. Abdela J
    10. Abdelalim A
    11. Abdollahpour I
    12. Abdulkader RS
    13. Abebe HT
    14. Abebe M
    15. Abebe Z
    16. Abejie AN
    17. Abera SF
    18. Abil OZ
    19. Abraha HN
    20. Abrham AR
    21. Abu-Raddad LJ
    22. Accrombessi MMK
    23. Acharya D
    24. Adamu AA
    25. Adebayo OM
    26. Adedoyin RA
    27. Adekanmbi V
    28. Adetokunboh OO
    29. Adhena BM
    30. Adib MG
    31. Admasie A
    32. Afshin A
    33. Agarwal G
    34. Agesa KM
    35. Agrawal A
    36. Agrawal S
    37. Ahmadi A
    38. Ahmadi M
    39. Ahmed MB
    40. Ahmed S
    41. Aichour AN
    42. Aichour I
    43. Aichour MTE
    44. Akbari ME
    45. Akinyemi RO
    46. Akseer N
    47. Al-Aly Z
    48. Al-Eyadhy A
    49. Al-Raddadi RM
    50. Alahdab F
    51. Alam K
    52. Alam T
    53. Alebel A
    54. Alene KA
    55. Alijanzadeh M
    56. Alizadeh-Navaei R
    57. Aljunid SM
    58. Alkerwi A
    59. Alla F
    60. Allebeck P
    61. Alonso J
    62. Altirkawi K
    63. Alvis-Guzman N
    64. Amare AT
    65. Aminde LN
    66. Amini E
    67. Ammar W
    68. Amoako YA
    69. Anber NH
    70. Andrei CL
    71. Androudi S
    72. Animut MD
    73. Anjomshoa M
    74. Ansari H
    75. Ansha MG
    76. Antonio CAT
    77. Anwari P
    78. Aremu O
    79. Ärnlöv J
    80. Arora A
    81. Arora M
    82. Artaman A
    83. Aryal KK
    84. Asayesh H
    85. Asfaw ET
    86. Ataro Z
    87. Atique S
    88. Atre SR
    89. Ausloos M
    90. Avokpaho EFGA
    91. Awasthi A
    92. Quintanilla BPA
    93. Ayele Y
    94. Ayer R
    95. Azzopardi PS
    96. Babazadeh A
    97. Bacha U
    98. Badali H
    99. Badawi A
    100. Bali AG
    101. Ballesteros KE
    102. Banach M
    103. Banerjee K
    104. Bannick MS
    105. Banoub JAM
    106. Barboza MA
    107. Barker-Collo SL
    108. Bärnighausen TW
    109. Barquera S
    110. Barrero LH
    111. Bassat Q
    112. Basu S
    113. Baune BT
    114. Baynes HW
    115. Bazargan-Hejazi S
    116. Bedi N
    117. Beghi E
    118. Behzadifar M
    119. Behzadifar M
    120. Béjot Y
    121. Bekele BB
    122. Belachew AB
    123. Belay E
    124. Belay YA
    125. Bell ML
    126. Bello AK
    127. Bennett DA
    128. Bensenor IM
    129. Berman AE
    130. Bernabe E
    131. Bernstein RS
    132. Bertolacci GJ
    133. Beuran M
    134. Beyranvand T
    135. Bhalla A
    136. Bhattarai S
    137. Bhaumik S
    138. Bhutta ZA
    139. Biadgo B
    140. Biehl MH
    141. Bijani A
    142. Bikbov B
    143. Bilano V
    144. Bililign N
    145. Bin Sayeed MS
    146. Bisanzio D
    147. Biswas T
    148. Blacker BF
    149. Basara BB
    150. Borschmann R
    151. Bosetti C
    152. Bozorgmehr K
    153. Brady OJ
    154. Brant LC
    155. Brayne C
    156. Brazinova A
    157. Breitborde NJK
    158. Brenner H
    159. Briant PS
    160. Britton G
    161. Brugha T
    162. Busse R
    163. Butt ZA
    164. Callender CSKH
    165. Campos-Nonato IR
    166. Campuzano Rincon JC
    167. Cano J
    168. Car M
    169. Cárdenas R
    170. Carreras G
    171. Carrero JJ
    172. Carter A
    173. Carvalho F
    174. Castañeda-Orjuela CA
    175. Castillo Rivas J
    176. Castle CD
    177. Castro C
    178. Castro F
    179. Catalá-López F
    180. Cerin E
    181. Chaiah Y
    182. Chang J-C
    183. Charlson FJ
    184. Chaturvedi P
    185. Chiang PP-C
    186. Chimed-Ochir O
    187. Chisumpa VH
    188. Chitheer A
    189. Chowdhury R
    190. Christensen H
    191. Christopher DJ
    192. Chung S-C
    193. Cicuttini FM
    194. Ciobanu LG
    195. Cirillo M
    196. Cohen AJ
    197. Cooper LT
    198. Cortesi PA
    199. Cortinovis M
    200. Cousin E
    201. Cowie BC
    202. Criqui MH
    203. Cromwell EA
    204. Crowe CS
    205. Crump JA
    206. Cunningham M
    207. Daba AK
    208. Dadi AF
    209. Dandona L
    210. Dandona R
    211. Dang AK
    212. Dargan PI
    213. Daryani A
    214. Das SK
    215. Gupta RD
    216. Neves JD
    217. Dasa TT
    218. Dash AP
    219. Davis AC
    220. Davis Weaver N
    221. Davitoiu DV
    222. Davletov K
    223. De La Hoz FP
    224. De Neve J-W
    225. Degefa MG
    226. Degenhardt L
    227. Degfie TT
    228. Deiparine S
    229. Demoz GT
    230. Demtsu BB
    231. Denova-Gutiérrez E
    232. Deribe K
    233. Dervenis N
    234. Des Jarlais DC
    235. Dessie GA
    236. Dey S
    237. Dharmaratne SD
    238. Dicker D
    239. Dinberu MT
    240. Ding EL
    241. Dirac MA
    242. Djalalinia S
    243. Dokova K
    244. Doku DT
    245. Donnelly CA
    246. Dorsey ER
    247. Doshi PP
    248. Douwes-Schultz D
    249. Doyle KE
    250. Driscoll TR
    251. Dubey M
    252. Dubljanin E
    253. Duken EE
    254. Duncan BB
    255. Duraes AR
    256. Ebrahimi H
    257. Ebrahimpour S
    258. Edessa D
    259. Edvardsson D
    260. Eggen AE
    261. El Bcheraoui C
    262. El Sayed Zaki M
    263. El-Khatib Z
    264. Elkout H
    265. Ellingsen CL
    266. Endres M
    267. Endries AY
    268. Er B
    269. Erskine HE
    270. Eshrati B
    271. Eskandarieh S
    272. Esmaeili R
    273. Esteghamati A
    274. Fakhar M
    275. Fakhim H
    276. Faramarzi M
    277. Fareed M
    278. Farhadi F
    279. Farinha CSE
    280. Faro A
    281. Farvid MS
    282. Farzadfar F
    283. Farzaei MH
    284. Feigin VL
    285. Feigl AB
    286. Fentahun N
    287. Fereshtehnejad S-M
    288. Fernandes E
    289. Fernandes JC
    290. Ferrari AJ
    291. Feyissa GT
    292. Filip I
    293. Finegold S
    294. Fischer F
    295. Fitzmaurice C
    296. Foigt NA
    297. Foreman KJ
    298. Fornari C
    299. Frank TD
    300. Fukumoto T
    301. Fuller JE
    302. Fullman N
    303. Fürst T
    304. Furtado JM
    305. Futran ND
    306. Gallus S
    307. Garcia-Basteiro AL
    308. Garcia-Gordillo MA
    309. Gardner WM
    310. Gebre AK
    311. Gebrehiwot TT
    312. Gebremedhin AT
    313. Gebremichael B
    314. Gebremichael TG
    315. Gelano TF
    316. Geleijnse JM
    317. Genova-Maleras R
    318. Geramo YCD
    319. Gething PW
    320. Gezae KE
    321. Ghadami MR
    322. Ghadimi R
    323. Ghasemi Falavarjani K
    324. Ghasemi-Kasman M
    325. Ghimire M
    326. Gibney KB
    327. Gill PS
    328. Gill TK
    329. Gillum RF
    330. Ginawi IA
    331. Giroud M
    332. Giussani G
    333. Goenka S
    334. Goldberg EM
    335. Goli S
    336. Gómez-Dantés H
    337. Gona PN
    338. Gopalani SV
    339. Gorman TM
    340. Goto A
    341. Goulart AC
    342. Gnedovskaya EV
    343. Grada A
    344. Grosso G
    345. Gugnani HC
    346. Guimaraes ALS
    347. Guo Y
    348. Gupta PC
    349. Gupta R
    350. Gupta R
    351. Gupta T
    352. Gutiérrez RA
    353. Gyawali B
    354. Haagsma JA
    355. Hafezi-Nejad N
    356. Hagos TB
    357. Hailegiyorgis TT
    358. Hailu GB
    359. Haj-Mirzaian A
    360. Haj-Mirzaian A
    361. Hamadeh RR
    362. Hamidi S
    363. Handal AJ
    364. Hankey GJ
    365. Harb HL
    366. Harikrishnan S
    367. Haro JM
    368. Hasan M
    369. Hassankhani H
    370. Hassen HY
    371. Havmoeller R
    372. Hay RJ
    373. Hay SI
    374. He Y
    375. Hedayatizadeh-Omran A
    376. Hegazy MI
    377. Heibati B
    378. Heidari M
    379. Hendrie D
    380. Henok A
    381. Henry NJ
    382. Herteliu C
    383. Heydarpour F
    384. Heydarpour P
    385. Heydarpour S
    386. Hibstu DT
    387. Hoek HW
    388. Hole MK
    389. Homaie Rad E
    390. Hoogar P
    391. Hosgood HD
    392. Hosseini SM
    393. Hosseinzadeh M
    394. Hostiuc M
    395. Hostiuc S
    396. Hotez PJ
    397. Hoy DG
    398. Hsiao T
    399. Hu G
    400. Huang JJ
    401. Husseini A
    402. Hussen MM
    403. Hutfless S
    404. Idrisov B
    405. Ilesanmi OS
    406. Iqbal U
    407. Irvani SSN
    408. Irvine CMS
    409. Islam N
    410. Islam SMS
    411. Islami F
    412. Jacobsen KH
    413. Jahangiry L
    414. Jahanmehr N
    415. Jain SK
    416. Jakovljevic M
    417. Jalu MT
    418. James SL
    419. Javanbakht M
    420. Jayatilleke AU
    421. Jeemon P
    422. Jenkins KJ
    423. Jha RP
    424. Jha V
    425. Johnson CO
    426. Johnson SC
    427. Jonas JB
    428. Joshi A
    429. Jozwiak JJ
    430. Jungari SB
    431. Jürisson M
    432. Kabir Z
    433. Kadel R
    434. Kahsay A
    435. Kalani R
    436. Karami M
    437. Karami Matin B
    438. Karch A
    439. Karema C
    440. Karimi-Sari H
    441. Kasaeian A
    442. Kassa DH
    443. Kassa GM
    444. Kassa TD
    445. Kassebaum NJ
    446. Katikireddi SV
    447. Kaul A
    448. Kazemi Z
    449. Karyani AK
    450. Kazi DS
    451. Kefale AT
    452. Keiyoro PN
    453. Kemp GR
    454. Kengne AP
    455. Keren A
    456. Kesavachandran CN
    457. Khader YS
    458. Khafaei B
    459. Khafaie MA
    460. Khajavi A
    461. Khalid N
    462. Khalil IA
    463. Khan EA
    464. Khan MS
    465. Khan MA
    466. Khang Y-H
    467. Khater MM
    468. Khoja AT
    469. Khosravi A
    470. Khosravi MH
    471. Khubchandani J
    472. Kiadaliri AA
    473. Kibret GD
    474. Kidanemariam ZT
    475. Kiirithio DN
    476. Kim D
    477. Kim Y-E
    478. Kim YJ
    479. Kimokoti RW
    480. Kinfu Y
    481. Kisa A
    482. Kissimova-Skarbek K
    483. Kivimäki M
    484. Knudsen AKS
    485. Kocarnik JM
    486. Kochhar S
    487. Kokubo Y
    488. Kolola T
    489. Kopec JA
    490. Koul PA
    491. Koyanagi A
    492. Kravchenko MA
    493. Krishan K
    494. Kuate Defo B
    495. Kucuk Bicer B
    496. Kumar GA
    497. Kumar M
    498. Kumar P
    499. Kutz MJ
    500. Kuzin I
    501. Kyu HH
    502. Lad DP
    503. Lad SD
    504. Lafranconi A
    505. Lal DK
    506. Lalloo R
    507. Lallukka T
    508. Lam JO
    509. Lami FH
    510. Lansingh VC
    511. Lansky S
    512. Larson HJ
    513. Latifi A
    514. Lau KM-M
    515. Lazarus JV
    516. Lebedev G
    517. Lee PH
    518. Leigh J
    519. Leili M
    520. Leshargie CT
    521. Li S
    522. Li Y
    523. Liang J
    524. Lim L-L
    525. Lim SS
    526. Limenih MA
    527. Linn S
    528. Liu S
    529. Liu Y
    530. Lodha R
    531. Lonsdale C
    532. Lopez AD
    533. Lorkowski S
    534. Lotufo PA
    535. Lozano R
    536. Lunevicius R
    537. Ma S
    538. Macarayan ERK
    539. Mackay MT
    540. MacLachlan JH
    541. Maddison ER
    542. Madotto F
    543. Magdy Abd El Razek H
    544. Magdy Abd El Razek M
    545. Maghavani DP
    546. Majdan M
    547. Majdzadeh R
    548. Majeed A
    549. Malekzadeh R
    550. Malta DC
    551. Manda A-L
    552. Mandarano-Filho LG
    553. Manguerra H
    554. Mansournia MA
    555. Mapoma CC
    556. Marami D
    557. Maravilla JC
    558. Marcenes W
    559. Marczak L
    560. Marks A
    561. Marks GB
    562. Martinez G
    563. Martins-Melo FR
    564. Martopullo I
    565. März W
    566. Marzan MB
    567. Masci JR
    568. Massenburg BB
    569. Mathur MR
    570. Mathur P
    571. Matzopoulos R
    572. Maulik PK
    573. Mazidi M
    574. McAlinden C
    575. McGrath JJ
    576. McKee M
    577. McMahon BJ
    578. Mehata S
    579. Mehndiratta MM
    580. Mehrotra R
    581. Mehta KM
    582. Mehta V
    583. Mekonnen TC
    584. Melese A
    585. Melku M
    586. Memiah PTN
    587. Memish ZA
    588. Mendoza W
    589. Mengistu DT
    590. Mengistu G
    591. Mensah GA
    592. Mereta ST
    593. Meretoja A
    594. Meretoja TJ
    595. Mestrovic T
    596. Mezgebe HB
    597. Miazgowski B
    598. Miazgowski T
    599. Millear AI
    600. Miller TR
    601. Miller-Petrie MK
    602. Mini GK
    603. Mirabi P
    604. Mirarefin M
    605. Mirica A
    606. Mirrakhimov EM
    607. Misganaw AT
    608. Mitiku H
    609. Moazen B
    610. Mohammad KA
    611. Mohammadi M
    612. Mohammadifard N
    613. Mohammed MA
    614. Mohammed S
    615. Mohan V
    616. Mokdad AH
    617. Molokhia M
    618. Monasta L
    619. Moradi G
    620. Moradi-Lakeh M
    621. Moradinazar M
    622. Moraga P
    623. Morawska L
    624. Moreno Velásquez I
    625. Morgado-Da-Costa J
    626. Morrison SD
    627. Moschos MM
    628. Mouodi S
    629. Mousavi SM
    630. Muchie KF
    631. Mueller UO
    632. Mukhopadhyay S
    633. Muller K
    634. Mumford JE
    635. Musa J
    636. Musa KI
    637. Mustafa G
    638. Muthupandian S
    639. Nachega JB
    640. Nagel G
    641. Naheed A
    642. Nahvijou A
    643. Naik G
    644. Nair S
    645. Najafi F
    646. Naldi L
    647. Nam HS
    648. Nangia V
    649. Nansseu JR
    650. Nascimento BR
    651. Natarajan G
    652. Neamati N
    653. Negoi I
    654. Negoi RI
    655. Neupane S
    656. Newton CRJ
    657. Ngalesoni FN
    658. Ngunjiri JW
    659. Nguyen AQ
    660. Nguyen G
    661. Nguyen HT
    662. Nguyen HT
    663. Nguyen LH
    664. Nguyen M
    665. Nguyen TH
    666. Nichols E
    667. Ningrum DNA
    668. Nirayo YL
    669. Nixon MR
    670. Nolutshungu N
    671. Nomura S
    672. Norheim OF
    673. Noroozi M
    674. Norrving B
    675. Noubiap JJ
    676. Nouri HR
    677. Nourollahpour Shiadeh M
    678. Nowroozi MR
    679. Nyasulu PS
    680. Odell CM
    681. Ofori-Asenso R
    682. Ogbo FA
    683. Oh I-H
    684. Oladimeji O
    685. Olagunju AT
    686. Olivares PR
    687. Olsen HE
    688. Olusanya BO
    689. Olusanya JO
    690. Ong KL
    691. Ong SKS
    692. Oren E
    693. Orpana HM
    694. Ortiz A
    695. Ortiz JR
    696. Otstavnov SS
    697. Øverland S
    698. Owolabi MO
    699. Özdemir R
    700. P A M
    701. Pacella R
    702. Pakhale S
    703. Pakhare AP
    704. Pakpour AH
    705. Pana A
    706. Panda-Jonas S
    707. Pandian JD
    708. Parisi A
    709. Park E-K
    710. Parry CDH
    711. Parsian H
    712. Patel S
    713. Pati S
    714. Patton GC
    715. Paturi VR
    716. Paulson KR
    717. Pereira A
    718. Pereira DM
    719. Perico N
    720. Pesudovs K
    721. Petzold M
    722. Phillips MR
    723. Piel FB
    724. Pigott DM
    725. Pillay JD
    726. Pirsaheb M
    727. Pishgar F
    728. Polinder S
    729. Postma MJ
    730. Pourshams A
    731. Poustchi H
    732. Pujar A
    733. Prakash S
    734. Prasad N
    735. Purcell CA
    736. Qorbani M
    737. Quintana H
    738. Quistberg DA
    739. Rade KW
    740. Radfar A
    741. Rafay A
    742. Rafiei A
    743. Rahim F
    744. Rahimi K
    745. Rahimi-Movaghar A
    746. Rahman M
    747. Rahman MHU
    748. Rahman MA
    749. Rai RK
    750. Rajsic S
    751. Ram U
    752. Ranabhat CL
    753. Ranjan P
    754. Rao PC
    755. Rawaf DL
    756. Rawaf S
    757. Razo-García C
    758. Reddy KS
    759. Reiner RC
    760. Reitsma MB
    761. Remuzzi G
    762. Renzaho AMN
    763. Resnikoff S
    764. Rezaei S
    765. Rezaeian S
    766. Rezai MS
    767. Riahi SM
    768. Ribeiro ALP
    769. Rios-Blancas MJ
    770. Roba KT
    771. Roberts NLS
    772. Robinson SR
    773. Roever L
    774. Ronfani L
    775. Roshandel G
    776. Rostami A
    777. Rothenbacher D
    778. Roy A
    779. Rubagotti E
    780. Sachdev PS
    781. Saddik B
    782. Sadeghi E
    783. Safari H
    784. Safdarian M
    785. Safi S
    786. Safiri S
    787. Sagar R
    788. Sahebkar A
    789. Sahraian MA
    790. Salam N
    791. Salama JS
    792. Salamati P
    793. Saldanha RDF
    794. Saleem Z
    795. Salimi Y
    796. Salvi SS
    797. Salz I
    798. Sambala EZ
    799. Samy AM
    800. Sanabria J
    801. Sanchez-Niño MD
    802. Santomauro DF
    803. Santos IS
    804. Santos JV
    805. Milicevic MMS
    806. Sao Jose BP
    807. Sarker AR
    808. Sarmiento-Suárez R
    809. Sarrafzadegan N
    810. Sartorius B
    811. Sarvi S
    812. Sathian B
    813. Satpathy M
    814. Sawant AR
    815. Sawhney M
    816. Saxena S
    817. Sayyah M
    818. Schaeffner E
    819. Schmidt MI
    820. Schneider IJC
    821. Schöttker B
    822. Schutte AE
    823. Schwebel DC
    824. Schwendicke F
    825. Scott JG
    826. Sekerija M
    827. Sepanlou SG
    828. Serván-Mori E
    829. Seyedmousavi S
    830. Shabaninejad H
    831. Shackelford KA
    832. Shafieesabet A
    833. Shahbazi M
    834. Shaheen AA
    835. Shaikh MA
    836. Shams-Beyranvand M
    837. Shamsi M
    838. Shamsizadeh M
    839. Sharafi K
    840. Sharif M
    841. Sharif-Alhoseini M
    842. Sharma R
    843. She J
    844. Sheikh A
    845. Shi P
    846. Shiferaw MS
    847. Shigematsu M
    848. Shiri R
    849. Shirkoohi R
    850. Shiue I
    851. Shokraneh F
    852. Shrime MG
    853. Si S
    854. Siabani S
    855. Siddiqi TJ
    856. Sigfusdottir ID
    857. Sigurvinsdottir R
    858. Silberberg DH
    859. Silva DAS
    860. Silva JP
    861. Silva NTD
    862. Silveira DGA
    863. Singh JA
    864. Singh NP
    865. Singh PK
    866. Singh V
    867. Sinha DN
    868. Sliwa K
    869. Smith M
    870. Sobaih BH
    871. Sobhani S
    872. Sobngwi E
    873. Soneji SS
    874. Soofi M
    875. Sorensen RJD
    876. Soriano JB
    877. Soyiri IN
    878. Sposato LA
    879. Sreeramareddy CT
    880. Srinivasan V
    881. Stanaway JD
    882. Starodubov VI
    883. Stathopoulou V
    884. Stein DJ
    885. Steiner C
    886. Stewart LG
    887. Stokes MA
    888. Subart ML
    889. Sudaryanto A
    890. Sufiyan MB
    891. Sur PJ
    892. Sutradhar I
    893. Sykes BL
    894. Sylaja PN
    895. Sylte DO
    896. Szoeke CEI
    897. Tabarés-Seisdedos R
    898. Tabuchi T
    899. Tadakamadla SK
    900. Takahashi K
    901. Tandon N
    902. Tassew SG
    903. Taveira N
    904. Tehrani-Banihashemi A
    905. Tekalign TG
    906. Tekle MG
    907. Temsah M-H
    908. Temsah O
    909. Terkawi AS
    910. Teshale MY
    911. Tessema B
    912. Tessema GA
    913. Thankappan KR
    914. Thirunavukkarasu S
    915. Thomas N
    916. Thrift AG
    917. Thurston GD
    918. Tilahun B
    919. To QG
    920. Tobe-Gai R
    921. Tonelli M
    922. Topor-Madry R
    923. Torre AE
    924. Tortajada-Girbés M
    925. Touvier M
    926. Tovani-Palone MR
    927. Tran BX
    928. Tran KB
    929. Tripathi S
    930. Troeger CE
    931. Truelsen TC
    932. Truong NT
    933. Tsadik AG
    934. Tsoi D
    935. Tudor Car L
    936. Tuzcu EM
    937. Tyrovolas S
    938. Ukwaja KN
    939. Ullah I
    940. Undurraga EA
    941. Updike RL
    942. Usman MS
    943. Uthman OA
    944. Uzun SB
    945. Vaduganathan M
    946. Vaezi A
    947. Vaidya G
    948. Valdez PR
    949. Varavikova E
    950. Vasankari TJ
    951. Venketasubramanian N
    952. Villafaina S
    953. Violante FS
    954. Vladimirov SK
    955. Vlassov V
    956. Vollset SE
    957. Vos T
    958. Wagner GR
    959. Wagnew FS
    960. Waheed Y
    961. Wallin MT
    962. Walson JL
    963. Wang Y
    964. Wang Y-P
    965. Wassie MM
    966. Weiderpass E
    967. Weintraub RG
    968. Weldegebreal F
    969. Weldegwergs KG
    970. Werdecker A
    971. Werkneh AA
    972. West TE
    973. Westerman R
    974. Whiteford HA
    975. Widecka J
    976. Wilner LB
    977. Wilson S
    978. Winkler AS
    979. Wiysonge CS
    980. Wolfe CDA
    981. Wu S
    982. Wu Y-C
    983. Wyper GMA
    984. Xavier D
    985. Xu G
    986. Yadgir S
    987. Yadollahpour A
    988. Yahyazadeh Jabbari SH
    989. Yakob B
    990. Yan LL
    991. Yano Y
    992. Yaseri M
    993. Yasin YJ
    994. Yentür GK
    995. Yeshaneh A
    996. Yimer EM
    997. Yip P
    998. Yirsaw BD
    999. Yisma E
    1000. Yonemoto N
    1001. Yonga G
    1002. Yoon S-J
    1003. Yotebieng M
    1004. Younis MZ
    1005. Yousefifard M
    1006. Yu C
    1007. Zadnik V
    1008. Zaidi Z
    1009. Zaman SB
    1010. Zamani M
    1011. Zare Z
    1012. Zeleke AJ
    1013. Zenebe ZM
    1014. Zhang AL
    1015. Zhang K
    1016. Zhou M
    1017. Zodpey S
    1018. Zuhlke LJ
    1019. Naghavi M
    1020. Murray CJL
    (2018) Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017
    The Lancet 392:1736–1788.
    https://doi.org/10.1016/S0140-6736(18)32203-7

Decision letter

  1. Edward D Janus
    Reviewing Editor; University of Melbourne, Australia
  2. Matthias Barton
    Senior Editor; University of Zurich, Switzerland
  3. Edward D Janus
    Reviewer; University of Melbourne, Australia
  4. Corey Giles
    Reviewer

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

The novel approach is a detailed lipidomic profile of 61 plasma metabolites by NMR at baseline in a large number of Chinese subjects associated with five individual lifestyle related factors ( dietary habit, exercise, non-smoking, adiposity and alcohol consumption) either singly or in combination and the impact of these on vascular disease after 20 years of follow up. Adherence as expected improved both the profile and outcomes.

Decision letter after peer review:

Thank you for submitting your article "Improved lipidomic profile mediates the effects of adherence to healthy lifestyles on coronary heart disease" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Edward D Janus as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Matthias Barton as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Corey Giles (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, when editors judge that a submitted work as a whole belongs in eLife but that some conclusions require a modest amount of additional new data, as they do with your paper, we are asking that the manuscript be revised to either limit claims to those supported by data in hand, or to explicitly state that the relevant conclusions require additional supporting data.

Our expectation is that the authors will eventually carry out the additional experiments and report on how they affect the relevant conclusions either in a preprint on bioRxiv or medRxiv, or if appropriate, as a Research Advance in eLife, either of which would be linked to the original paper.

Summary:

The authors report, in 4681 subjects, the detailed baseline lipidomic profile (61 plasma metabolites by NMR) associated with five individual healthy lifestyle related factors singly or in combination (dietary habit, exercise, non-smoking, adiposity and exercise) and relate these to the presence of coronary or cerebrovascular disease after approximately 20 years of follow up vs the baseline parameters for a further 1513 non affected matched control subjects. All measurements were without lipid lowering drugs.

They used regression to assess associations between health life-style factors (HLF); mediation analysis to assess mediating effects of metabolites on coronary heart disease HLF; pharmacomimetic gene scores to estimate the genetic effects of HMGCR and ACLY.

The beneficial effects of the five lifestyle characteristics are already well known so the main new data is the detailed lipidomic profile in a Chinese population and its implications. Participants who adhered to a combined healthy lifestyle were more likely to have a cardio-protective lipidomic profile which jointly mediated 14% of the protective effect of combined healthy lifestyle on CHD risk.

The manuscript is well written overall and has important clinical interpretation although there is an overload of information in the supplementary material. There are some issues of clarity in the text and some major limitations in the analysis which need to be addressed.

Revisions for this paper:

1).While the Title is clear and acceptable the statement in the Abstract – Healthy lifestyles included baseline smoking, alcohol consumption, dietary habit, physical activity, and adiposity levels – is not appropriate as some of these are unhealthy lifestyles or as in the case of adiposity are not even lifestyles. Better to use "Baseline lifestyle related characteristics" throughout the manuscript.

Also in the Abstract – significant mediation effects in the pathway from healthy lifestyles to CHD – is not correct as this says that healthy lifestyles cause CHD. If you delete "healthy" it’s then correct but even better is to leave your wording but change it to "CHD reduction".

2) In the last paragraph of the Introduction say something in one or two sentences about the cohort, CHD and the controls. Without this the Results section is difficult to understand because it precedes the Materials and methods section.

3) Results

At beginning please list the five Healthy lifestyle factors.

4) Mediation effects of lipid metabolites in the association between HLFs and CHD risk. In the first line you need to make it clear what is meant by CHD risk that is state it is the presence of coronary or cerebrovascular disease in members of the cohort during the follow up period. Also CHD means coronary heart disease so cerebro-vascular disease is not CHD. You need to use Coronary heart disease AND cerebrovascular disease throughout the paper in title, text, tables, figures. You need to also make it very clear that you have included intracerebral haemorrhage because this is not atherosclerotic in origin. Ideally you would show that excluding haemorrhagic stroke while retaining ischemic stroke does not substantially change your overall findings/implications.

5) Please provide the mean and SD number of years of follow-up for the cohort and specify how reliable and complete follow up was. This is important in your analysis. If incomplete then it’s a limitation to be discussed.

6) A major limitation of this study is the dichotomization of health factors, summing the number of factors for each person, then stratifying participants into one of three groups. I appreciate the authors intention to create an aggregate score for “health behavior”, but used in this context, it is indeed a limitation.

The HLF used in this study have all been linked with CHD. Dichotomizing them results in substantial loss of information. Summing across these scores then leads to a loss in specificity for the underlying biology. In effect, you may produce a variable that associates with CHD, but does not provide any more biological insight than what we knew over 50 years ago.

As shown in the supplement, much of the HLF could be driven by adiposity alone. This is after dichotomization, so I would expect even stronger results if the authors used raw BMI or WC. Some of the individual HLF have vastly different association strengths with metabolites, some even in the opposite direction to each other. The aggregate HLF just ends up losing power in this situation and hides the underlying biology. Associations of individual HLF (without dichotomization) provides information that can help improve our understanding of the metabolic consequences of these HLF. The current manuscript is lacking this insight.

Importantly, mediation analysis is no longer valid when the exposure variable is dichotomized. Causal mediation analysis will suggest a mediation effect even when this is not true. There is no getting around this, unfortunately. Whenever information is removed from the exposure, the putative mediating variable will attempt to explain it.

7) Alcohol consumption appeared to have a complex effect on the lipoprotein profile. From Table 1, around 28.3% of the study population with >= 4 healthy lifestyle factors were considered as moderate alcohol consumers. What was the general distribution of alcohol intake among study population, e.g., non-regular, previous regular, or heavy intake? This information should be given to better understand the effect of alcohol consumption.

8) The authors appear to have attempted a two-factorial analysis of gene scores by the HLF aggregate. It is now known that using the derived scores and an interaction term provides much more power for identifying independence or importance.

I am not sure why this analysis was performed: "We also examined whether the association between HLFs and lipid metabolites differed by scores of HMGCR, ACLY, or their sum score". Each genetic variant often has small effect sizes (much smaller than a typical dose of statin). Even the combined score would be very small compared to this. Assuming there isn't some underlying confounding, the inheritance of the genetic variants should be randomly distributed among the population. I do not believe there is any rationale that the association of metabolites with HLF should differ under these conditions.

9) The authors use the 1000 Genomes reference panel (Phase 3) to impute variants. Is this database (relative dearth of Chinese) appropriate to use for imputing variants in a Chinese population? The authors constructed genetic sores to mimic the effect of statins and ACLY inhibitors based on the Chinese population and also the European population. Use of the genetic scores based on the European population observed similar but weaker associations, especially for some lipid metabolites (small VLDL-TG and LDL-TG) (Supplementary file 1E).

https://doi.org/10.7554/eLife.60999.sa1

Author response

Revisions for this paper:

1) While the Title is clear and acceptable the statement in the Abstract – Healthy lifestyles included baseline smoking, alcohol consumption, dietary habit, physical activity, and adiposity levels – is not appropriate as some of these are unhealthy lifestyles or as in the case of adiposity are not even lifestyles. Better to use "Baseline lifestyle related characteristics" throughout the manuscript.

Also in the Abstract – significant mediation effects in the pathway from healthy lifestyles to CHD – is not correct as this says that healthy lifestyles cause CHD. If you delete "healthy" it’s then correct but even better is to leave your wording but change it to "CHD reduction".

We thank the reviewer for pointing this out. We included adiposity measures as a lifestyle factor to assess energy balance, same as previous study (1). We have revised to use “baseline lifestyle related characteristics” or change the wording of each lifestyle throughout the manuscript to keep consistent. We also added “reduction” in the Abstract section.

2) In the last paragraph of the Introduction say something in one or two sentences about the cohort, CHD and the controls. Without this the Results section is difficult to understand because it precedes the Materials and methods section.

We thank the reviewer for pointing this out. We have added a brief description of the study population at the end of the Introduction section.

3) Results

At beginning please list the five Healthy lifestyle factors.

We have added to help better understanding.

4) Mediation effects of lipid metabolites in the association between HLFs and CHD risk. In the first line you need to make it clear what is meant by CHD risk that is state it is the presence of coronary or cerebrovascular disease in members of the cohort during the follow up period. Also CHD means coronary heart disease so cerebro-vascular disease is not CHD. You need to use Coronary heart disease AND cerebrovascular disease throughout the paper in title, text, tables, figures. You need to also make it very clear that you have included intracerebral haemorrhage because this is not atherosclerotic in origin. Ideally you would show that excluding haemorrhagic stroke while retaining ischemic stroke does not substantially change your overall findings/implications.

Of the total 4,681 participants, cases were those who had a newly developed fatal or nonfatal disease during follow-up: (1) CHD: fatal ischemic heart disease coded as ICD-10 I20-I25 and nonfatal myocardial infarction coded as I21-I23 (n=927); (2) ischaemic stroke: ICD-10 I63 or I69.3 (n=1,114); (3) intracerebral haemorrhage: ICD-10 I61 or I69.1 (n=1,127). Case status was defined as the disease first occurred in each participant. Common controls were selected by frequency matching to combined cases by age, sex, and study area (n=1,513).

In the analyses of HLFs and metabolites, we included all 4,681 participants and adjusted for case/control status to increase power. We also restricted analyses to control participants as a sensitivity analysis (Supplementary file 1B). In the analyses of metabolites and CHD, and corresponding mediation analyses, we excluded both ischaemic and haemorrhagic stroke. We apologize for the confusion because the Results section precedes the Materials and methods section. We have added a brief description of the study participants involved in this part of analyses in the “Result – Mediation effects” section and the last paragraph of the Introduction section.

5) Please provide the mean and SD number of years of follow-up for the cohort and specify how reliable and complete follow up was.This is important in your analysis. If incomplete then it’s a limitation to be discussed.

The mean follow-up duration of the cohort since baseline was 9.2 (1.4) years. We added the mean follow-up duration of the cohort in the Materials and methods – Study population section. The electronic linkage with the national health insurance (HI) claim databases started in 2011, which has become an important means of following up. Linkage to local HI databases has been achieved by 97% of the participants since 2014. By December 31 2015, of the all cohort participants, only 4875 (<1%) were lost to follow-up. The diagnosis adjudication has finished for 34,000 reported cases of ischemic heart disease by a review of hospital medical records. Overall, 88% of the diagnoses were confirmed. We have also added this information to the manuscript.

6) A major limitation of this study is the dichotomization of health factors, summing the number of factors for each person, then stratifying participants into one of three groups. I appreciate the authors intention to create an aggregate score for “health behavior”, but used in this context, it is indeed a limitation.

The HLF used in this study have all been linked with CHD. Dichotomizing them results in substantial loss of information. Summing across these scores then leads to a loss in specificity for the underlying biology. In effect, you may produce a variable that associates with CHD, but does not provide any more biological insight than what we knew over 50 years ago.

As shown in the supplement, much of the HLF could be driven by adiposity alone. This is after dichotomization, so I would expect even stronger results if the authors used raw BMI or WC. Some of the individual HLF have vastly different association strengths with metabolites, some even in the opposite direction to each other. The aggregate HLF just ends up losing power in this situation and hides the underlying biology. Associations of individual HLF (without dichotomization) provides information that can help improve our understanding of the metabolic consequences of these HLF. The current manuscript is lacking this insight.

Importantly, mediation analysis is no longer valid when the exposure variable is dichotomized. Causal mediation analysis will suggest a mediation effect even when this is not true. There is no getting around this, unfortunately. Whenever information is removed from the exposure, the putative mediating variable will attempt to explain it.

Previous studies have examined the association between individual lifestyle related characteristics (without dichotomization) and lipidomic profile separately, including physical activity (2), alcohol consumption (3), BMI (4), and consumption of whole grain, fish and bilberries (5). In our study, we also performed analyses for individual factors (Figure 1—figure supplement 2-7). Our findings on the associations of lipid metabolites with individual lifestyle related characteristics like physical activity, adiposity, and alcohol consumption were generally consistent with previous studies. We agree with the reviewer that individual HLF may lead to different or even opposite effect on metabolites. Lifestyle related characteristics are also typically correlated with one another. Previous studies suggested that a large proportion of premature death (6) and increased risk of cardiometabolic diseases (7, 8) are attributed to the combination of these unhealthy lifestyles (dichotomized), providing important information on the maximum public health benefit that lifestyle intervention could achieve. A comprehensive analysis of the impact of adopting HLFs on the lipidomic profile is still lacking. Studies are warranted to examine the combined effect on metabolites as the mediators to cardiometabolic diseases. Our study has this unique opportunity to address this important question. Therefore, our main aim was to align with these previously used definitions and provide evidence of the combined effects of lifestyle related characteristics on lipidomic profile.

We agree with the reviewer that dichotomizing a continuous variable may lose information. A more sophisticated score with appropriated weight might show stronger association. However, a more straightforward definition would be easier to understand and adapted by the public. We have addressed this issue in the Discussion section.

Regarding the mediation analysis, we agree that when the exposure and mediator are highly correlated and the exposure is dichotomized with losing information, the mediator might show a biased association with the outcome even when it is not. However, this requires a particularly strong association between exposure and mediator. To demonstrate this phenomenon, we picked the strongest exposure-mediator association observed in our study (i.e., BMI-ApoB/ApoA1 with R2=0.137). We simulated the data with same sample size and case-control numbers, the same BMI-disease association (effect size OR=1.023) and the same BMI-metabolite association as BMI-ApoB/ApoA1 but no association between metabolite and disease outcome. We designed a series of simulations using different R2 (0.10–0.99), each repeating 100 times. The simulated data showed as R2 increases the mediator will have larger bias in association with the outcome (Author response image 1), and subsequently larger bias in the corresponding mediation effect (Author response image 2).

Author response image 1
The p value frequency for the association between the mediator and the outcome among the 100 simulations.
Author response image 2
The p value frequency for the mediation effect among the 100 simulations.

Author response table 1 shows the Bonferroni-corrected minimum p value among the 100 simulations. The BMI-metabolite association needs to be increased to R2=0.80 to generate a falsely significant mediation effect to achieve Bonferroni-corrected p value < 0.05. This is far from the realistic association between HLFs and metabolite in our study (the strongest R2=0.137 for BMI – ApoB/ApoA1). To make a direct comparison, we tested the mediation effect for dichotomized BMI – ApoB/ApoA1 – CHD and the observed mediation p value was less than 2E-16 in our study, far from that would have been biased due to loss of information. We have discussed this issue in the Discussion section.

Author response table 1
The Bonferroni-corrected minimum p value in 100 simulations.
Bonferroni-corrected minimum p value for the mediator – outcome associationBonferroni-corrected minimum p value for the mediation effect
R2=0.100.2150.227
R2=0.1370.2580.216
R2=0.150.1240.117
R2=0.200.1330.189
R2=0.300.1140.154
R2=0.400.1530.198
R2=0.500.2900.234
R2=0.600.2170.247
R2=0.700.0570.096
R2=0.800.0200.047
R2=0.900.0140.018
R2=0.990.0110.005

7) Alcohol consumption appeared to have a complex effect on the lipoprotein profile. From Table 1, around 28.3% of the study population with >= 4 healthy lifestyle factors were considered as moderate alcohol consumers. What was the general distribution of alcohol intake among study population, e.g., non-regular, previous regular, or heavy intake? This information should be given to better understand the effect of alcohol consumption.

In the present study, around 77.5% of the participants reported drinking less than weekly at baseline. Please find the detailed distribution of alcohol consumption in Author response table 2.

Author response table 2
Detailed distributionNo. of participantsPercent
Non-regular alcohol consumptionNot regular (less than weekly) drinker3,62677.5
Former regular drinker2054.4
Moderate alcohol consumptionWeekly but not daily drinker4399.4
Daily drinker with <15g pure alcohol per day30.1
Daily drinker with 15-29g pure alcohol per day801.7
Heavy alcohol consumptionDaily drinker with 30-59g pure alcohol per day1202.6
Daily drinker with ≥60g pure alcohol per day2084.4

8) The authors appear to have attempted a two-factorial analysis of gene scores by the HLF aggregate. It is now known that using the derived scores and an interaction term provides much more power for identifying independence or importance.

I am not sure why this analysis was performed: "We also examined whether the association between HLFs and lipid metabolites differed by scores of HMGCR, ACLY, or their sum score". Each genetic variant often has small effect sizes (much smaller than a typical dose of statin). Even the combined score would be very small compared to this. Assuming there isn't some underlying confounding, the inheritance of the genetic variants should be randomly distributed among the population. I do not believe there is any rationale that the association of metabolites with HLF should differ under these conditions.

Because HLF and statin both have the potential impact on lowering cardiometabolic disease-associated lipids, we performed this analysis to explore the potential joint and interaction effect of HLFs with lipid-lowering drugs on lipid metabolites. Our results demonstrated the independent effect of both HLFs and lipid-lowering drugs on metabolites. Importantly we observed that when simultaneously adjusted in the model, they are targeting different components of the lipidomic profile. Similar to the intuition of the reviewer, we observed HLFs showed no interactions with neither genetic scores (to mimic the effect of HMGCR and ACLY) in its effect on lipid metabolites (Supplementary file 1C-1E). Our results suggest that the pathways affecting LDL and VLDL might be different. We prefer to keep these results to give a comprehensive understanding of the underlying pathway of HLFs and lipid-lowering drugs.

9) The authors use the 1000 Genomes reference panel (Phase 3) to impute variants. Is this database (relative dearth of Chinese) appropriate to use for imputing variants in a Chinese population? The authors constructed genetic sores to mimic the effect of statins and ACLY inhibitors based on the Chinese population and also the European population. Use of the genetic scores based on the European population observed similar but weaker associations, especially for some lipid metabolites (small VLDL-TG and LDL-TG) (Supplementary file 1E).

In our study, we used the 1000 Genomes Project reference panel (Phase 3, all populations) to impute ungenotyped SNPs in the CKB subjects (9, 10). The 1000 Genomes Project Consortium has demonstrated the imputation accuracy using all samples in 1000 Genomes Project as reference to impute other population data, the squared correlation between imputed and experimental genotypes was >95% for common variants in each population (including Chinese) (11). For both the HapMap and the 1000 Genomes Project, it has been a common recommendation that cosmopolitan reference panels be used to improve imputation accuracy for common and low-frequency variants (12, 13). We have provided details of imputation quality of the variants used to construct genetic scores (Supplementary file 1H). It is currently expected that genetic association effects learned from the European population sample are generally less applicable to other populations. Therefore we also developed the genetic score using effects learned from the CKB subjects.

References

1. Lloyd-Jones DM, Hong Y, Labarthe D, et al. Defining and setting national goals for cardiovascular health promotion and disease reduction: the American Heart Association’s strategic Impact Goal through 2020 and beyond. Circulation 2010;121:586–613.2. Kujala Urho M., Mäkinen Ville-Petteri, Heinonen Ilkka, et al. Long-term Leisure-time Physical Activity and Serum Metabolome. Circulation 2013;127:340–348.3. Würtz P, Cook S, Wang Q, et al. Metabolic profiling of alcohol consumption in 9778 young adults. Int. J. Epidemiol. 2016;45:1493–1506.4. Würtz P, Wang Q, Kangas AJ, et al. Metabolic Signatures of Adiposity in Young Adults: Mendelian Randomization Analysis and Effects of Weight Change. PLOS Med. 2014;11:e1001765.5. Lankinen M, Kolehmainen M, Jääskeläinen T, et al. Effects of Whole Grain, Fish and Bilberries on Serum Metabolic Profile and Lipid Transfer Protein Activities: A Randomized Trial (Sysdimet) Berthold HK, editor. PLoS ONE 2014;9:e90352.6. Li Y, Schoufour J, Wang DD, et al. Healthy lifestyle and life expectancy free of cancer, cardiovascular disease, and type 2 diabetes: prospective cohort study. BMJ 2020;368. Available at: https://www.bmj.com/content/368/bmj.l6669. Accessed February 18, 2020.7. Lv J, Yu C, Guo Y, et al. Adherence to a healthy lifestyle and the risk of type 2 diabetes in Chinese adults. Int. J. Epidemiol. 2017;46:1410–1420.8. Lv J, Yu C, Guo Y, et al. Adherence to Healthy Lifestyle and Cardiovascular Diseases in the Chinese Population. J. Am. Coll. Cardiol. 2017;69:1116–1125.9. Dai J, Lv J, Zhu M, et al. Identification of risk loci and a polygenic risk score for lung cancer: a large-scale prospective cohort study in Chinese populations. Lancet Respir. Med. 2019;7:881–891.10. Gan W, Bragg F, Walters RG, et al. Genetic Predisposition to Type 2 Diabetes and Risk of Subclinical Atherosclerosis and Cardiovascular Diseases Among 160,000 Chinese Adults. Diabetes 2019;68:2155–2164.11. Anon. A global reference for human genetic variation. Nature 2015;526:68–74.12. Howie B, Marchini J, Stephens M. Genotype Imputation with Thousands of Genomes. G3 Genes Genomes Genet. 2011;1:457–470.13. Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: using sequence and genotype data to estimate haplotypes and unobserved genotypes. Genet. Epidemiol. 2010;34:816–834.

https://doi.org/10.7554/eLife.60999.sa2

Article and author information

Author details

  1. Jiahui Si

    1. Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
    2. Departments of Epidemiology and Biostatistics, Harvard T.H. Chan School of Public Health, Boston, United States
    Contribution
    Formal analysis, Visualization, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
  2. Jiachen Li

    Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
    Contribution
    Formal analysis, Validation, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Canqing Yu

    1. Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
    2. Peking University Institute of Public Health & Emergency Preparedness, Beijing, China
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Yu Guo

    Chinese Academy of Medical Sciences, Beijing, China
    Contribution
    Data curation, Funding acquisition, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  5. Zheng Bian

    Chinese Academy of Medical Sciences, Beijing, China
    Contribution
    Data curation, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  6. Iona Millwood

    1. Medical Research Council Population Health Research Unit at the University of Oxford, Oxford, United Kingdom
    2. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  7. Ling Yang

    1. Medical Research Council Population Health Research Unit at the University of Oxford, Oxford, United Kingdom
    2. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  8. Robin Walters

    1. Medical Research Council Population Health Research Unit at the University of Oxford, Oxford, United Kingdom
    2. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  9. Yiping Chen

    1. Medical Research Council Population Health Research Unit at the University of Oxford, Oxford, United Kingdom
    2. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  10. Huaidong Du

    1. Medical Research Council Population Health Research Unit at the University of Oxford, Oxford, United Kingdom
    2. Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  11. Li Yin

    NCDs Prevention and Control Department, Hunan Center for Disease Control & Prevention, Changsha, China
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  12. Jianwei Chen

    Liuyang Center for Disease Control & Prevention, Liuyang, Hunan, China
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  13. Junshi Chen

    China National Center for Food Safety Risk Assessment, Beijing, China
    Contribution
    Data curation, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  14. Zhengming Chen

    Clinical Trial Service Unit & Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Supervision, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  15. Liming Li

    1. Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
    2. Peking University Institute of Public Health & Emergency Preparedness, Beijing, China
    Contribution
    Conceptualization, Resources, Data curation, Supervision, Funding acquisition, Investigation, Project administration, Writing - review and editing
    For correspondence
    lmleeph@vip.163.com
    Competing interests
    No competing interests declared
  16. Liming Liang

    Departments of Epidemiology and Biostatistics, Harvard T.H. Chan School of Public Health, Boston, United States
    Contribution
    Conceptualization, Supervision, Methodology, Writing - review and editing
    For correspondence
    lliang@hsph.harvard.edu
    Competing interests
    No competing interests declared
  17. Jun Lv

    1. Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
    2. Peking University Institute of Public Health & Emergency Preparedness, Beijing, China
    3. Key Laboratory of Molecular Cardiovascular Sciences (Peking University), Ministry of Education, Beijing, China
    Contribution
    Conceptualization, Data curation, Supervision, Funding acquisition, Investigation, Methodology, Project administration, Writing - review and editing
    For correspondence
    epi.lvjun@vip.163.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7916-3870

Funding

National Key R&D Program of China (2016YFC0900500)

  • Yu Guo

Wellcome Trust (202922/Z/16/Z)

  • Zhengming Chen

National Natural Science Foundation of China (81390540)

  • Liming Li

National Natural Science Foundation of China (81390544)

  • Jun Lv

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The most important acknowledgement is to the participants in the study and the members of the survey teams in each of the 10 regional centers, as well as to the project development and management teams based at Beijing, Oxford and the 10 regional centers.

Ethics

Human subjects: The study protocol was approved by the Ethics Review Committee of the Chinese Center for Disease Control and Prevention (005/2004, Beijing, China) and the Oxford Tropical Research Ethics Committee, University of Oxford (025-04, UK). All participants provided written informed consent.

Senior Editor

  1. Matthias Barton, University of Zurich, Switzerland

Reviewing Editor

  1. Edward D Janus, University of Melbourne, Australia

Reviewers

  1. Edward D Janus, University of Melbourne, Australia
  2. Corey Giles

Publication history

  1. Received: July 13, 2020
  2. Accepted: January 20, 2021
  3. Version of Record published: February 9, 2021 (version 1)

Copyright

© 2021, Si et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,084
    Page views
  • 106
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)