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 0
  • Views 364
  • 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.

References

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

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

  • 364
    Page views
  • 48
    Downloads
  • 0
    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)

Further reading

    1. Epidemiology and Global Health
    2. Medicine
    Thao Phuong Ho-Le et al.
    Research Article Updated

    This study sought to redefine the concept of fracture risk that includes refracture and mortality, and to transform the risk into "skeletal age". We analysed data obtained from 3521 women and men aged 60 years and older, whose fracture incidence, mortality, and bone mineral density (BMD) have been monitored since 1989. During the 20-year follow-up period, among 632 women and 184 men with a first incident fracture, the risk of sustaining a second fracture was higher in women (36%) than in men (22%), but mortality risk was higher in men (41%) than in women (25%). The increased risk of mortality was not only present with an initial fracture, but was accelerated with refractures. Key predictors of post-fracture mortality were male gender (hazard ratio [HR] 2.4; 95% CI, 1.79–3.21), advancing age (HR 1.67; 1.53–1.83), and lower femoral neck BMD (HR 1.16; 1.01–1.33). A 70-year-old man with a fracture is predicted to have a skeletal age of 75. These results were incorporated into a prediction model to aid patient-doctor discussion about fracture vulnerability and treatment decisions.

    1. Epidemiology and Global Health
    2. Genetics and Genomics
    William L Hamilton et al.
    Research Article

    COVID-19 poses a major challenge to care homes, as SARS-CoV-2 is readily transmitted and causes disproportionately severe disease in older people. Here, 1,167 residents from 337 care homes were identified from a dataset of 6,600 COVID-19 cases from the East of England. Older age and being a care home resident were associated with increased mortality. SARS-CoV-2 genomes were available for 700 residents from 292 care homes. By integrating genomic and temporal data, 409 viral clusters within the 292 homes were identified, indicating two different patterns - outbreaks among care home residents and independent introductions with limited onward transmission. Approximately 70% of residents in the genomic analysis were admitted to hospital during the study, providing extensive opportunities for transmission between care homes and hospitals. Limiting viral transmission within care homes should be a key target for infection control to reduce COVID-19 mortality in this population.