Background
Coronary heart disease (CHD) remains the leading cause of death globally [
1,
2]. Recent decades have seen age-adjusted incidence and mortality rates decline substantially in higher-income countries [
3], but those declines are now slowing [
4,
5] and total numbers of cases are rising in most countries owing to population ageing and growth [
2,
6]. Age-adjusted CHD rates are higher among males than females [
7], and reasons for this are gaining clarity. For example, males are known to store more fat in visceral and ectopic compartments which drives insulin resistance [
8,
9] and results in higher type 2 diabetes rates among males [
10]. Males also have higher systolic blood pressure than females from adolescence to middle adulthood, although this difference narrows or even reverses in older age [
11‐
13].
Sex differences in circulating lipids are more contradictory. Adult females tend to have lower triglyceride levels compared with adult males, potentially due to hormonal mechanisms [
14,
15], yet adult females also tend to have higher cholesterol in low-density lipoprotein (LDL) particles [
12,
16]. Such comparisons have been based mostly on circulating traits measured by conventional clinical assays. More detailed measures from targeted metabolomic platforms now exist [
17] which have helped characterise the cardiometabolic profile of pregnancy and menopause [
18,
19]. Knowledge of sex differences in these more detailed traits at multiple life stages may help reveal more specific circulating pathways that underpin sex differences in age-adjusted rates of CHD, but no such investigation has yet been conducted.
We aimed in this study to better characterise sex differences in CHD-relevant metabolites at multiple life stages, to help identify circulating traits that may underpin known sex differences in CHD burden. Using data from a multi-generational pregnancy cohort study, we estimated the total effect of biological sex on over 200 systemic metabolites quantified using targeted metabolomics on EDTA-plasma or serum samples, including lipoprotein subclass-specific cholesterol and triglycerides, amino acids, and inflammatory glycoprotein acetyls, at four life stages. These metabolites were measured once in childhood (mean age 8 years), twice in adolescence (16 years and 18 years), and once in early adulthood (25 years) on the same male and female offspring (Generation-1 (G1)), as well as on their parents (Generation-0 (G0)) in middle adulthood (50 years). We also examined the extent to which adiposity may mediate any total effect of sex on metabolites by adjusting for body mass index (BMI) at each life stage.
Discussion
In this study, we examined sex differences in systemic metabolites at multiple life stages, from childhood to middle adulthood, to help identify circulating traits that may underpin higher age-adjusted CHD rates commonly seen among males. Our results suggest that, from adolescence onwards, lipids (particularly triglycerides) in VLDL are higher among males while levels of other CHD-related traits including LDL cholesterol, apolipoprotein-B, and inflammatory glycoprotein acetyls, are higher among females. These patterns of effect were highly consistent when adjusting for BMI measured on each occasion. Causal analyses of these traits in relation to clinical endpoints are needed to understand whether they differentially affect CHD risk among males and females.
A causal role for LDL cholesterol in CHD aetiology is strongly supported by human genetic [
35,
36] and pharmacological intervention [
37,
38] studies. Recent genetic evidence also supports a role of higher apolipoprotein-B in CHD independent of LDL cholesterol concentration [
39‐
42]. Despite this, our results suggest that differences in absolute levels of LDL cholesterol or apolipoprotein-B are unlikely to underpin the higher risk of CHD experienced among males, since levels were lower (more favourable) among males in adolescence and young adulthood among G1s, as well as in middle adulthood among G0s. However, this does not exclude the possibility of LDL cholesterol, apolipoprotein-B, or other metabolites which did not differ by sex in these analyses having a differential effect on CHD among males and females; sex-stratified causal analyses of these in relation to CHD itself are needed to determine this. The present study aimed to estimate absolute effects of sex on metabolites, not relative effects of metabolites on CHD by sex.
In contrast, our results suggest that triglyceride content—particularly in VLDL—are higher (more adverse) among males in adolescence and that this sex difference is larger in young adulthood and larger still in middle adulthood. The tendency for males to have higher circulating triglycerides has been observed previously [
14,
15], but how differences relative to females progress across multiple life stages has been unclear due to lack of repeated measures. Higher triglycerides among males at multiple life stages observed here, together with previous genetic evidence of a causal role of triglycerides for CHD [
40,
43], support triglycerides as a key target for CHD prevention, particularly among males. Whether triglycerides increase CHD risk more greatly among males requires sex-stratified analyses.
Equally strong tendencies were found for lower HDL cholesterol among males than females at later life stages. However, despite robust observational associations of lower HDL cholesterol with higher CHD risk [
44], genetic [
35,
45] and pharmacological intervention studies [
46] do not support a causal effect of HDL cholesterol on CHD. Nevertheless, HDL cholesterol is increasingly regarded as a useful non-causal marker for insulin resistance and other traits that are causal, namely circulating triglycerides, with which it tracks strongly in opposing directions [
44,
47]. The utility of HDL cholesterol as a marker of pre-glycaemic changes is further supported by results here showing progressively lower HDL cholesterol levels to coincide with progressively higher branched chain amino acid levels, which are likely components of early-stage insulin resistance [
48].
The earlier measurement occasions used in this study (8 years and 16 years) spanned puberty for most participants—an important period of growth and development. The mean age at puberty onset here was estimated at 13.6 years for males and 11.7 years for females based on growth-curve modelling of repeated height measures. This transition profoundly influences physical and sexual maturity [
49], and among females results in the release of oestrogen and other sex hormones which are thought to result in less adverse lipid profiles [
15]; whether testosterone release among males itself results in more adverse lipid profiles is less clear [
15,
50]. Potential benefits of oestrogen or harms of testosterone on lipid profiles are supported by present results which suggest that total lipids and triglycerides in VLDL are lower among males before puberty, but then switch direction after puberty onset to become higher among males in adolescence and young adulthood, a difference which is even greater in middle adulthood. This was also true of cholesterol in HDL but in the reverse direction (higher levels among males before puberty, then lower levels after puberty). Together, this supports puberty as a pivotal time for the emergence of life-long sex differences in triglycerides and risk-marking HDL cholesterol.
How these sex differences extend beyond middle age, when the menopausal transition is complete, is less clear. Females in middle adulthood were measured here at mean age 47.9 years, when natural menopause has either not yet begun or is typically in early stages. Menopause status was examined here using the rigorous STRAW criteria [
18,
28], and sex differences in metabolites were re-examined when including only those females who were pre-menopause; these indicated similar results as seen when all females were included. Sex differences were then re-examined when including only those females who were post-menopause (also excluding those who were peri-menopause), and sex differences in VLDL lipids appeared narrower than were seen when all females were included. In contrast, sex differences in IDL, LDL, and HDL lipids appeared wider. Differences also appeared narrower in branched chain amino acids, but wider in most other metabolites including fatty acids and glucose. This suggests that proposed cardio-protective effects of female oestrogen release [
14,
15], which would be reduced post-menopause [
18,
28], are VLDL-specific, but further studies in females with natural and surgical menopause are needed to confirm this.
Limitations
The limitations of this study include modest sample sizes among G1s, particularly for complete case analyses, and greater loss to follow-up among males. Unequal numbers of males and females may result in biased estimates of sex differences if loss to follow-up is related to both sex and outcomes; however, most characteristics were comparable between included and excluded males and females, suggesting that such bias is unlikely or small. Estimates were based on serial cross-sectional associations from linear regression models which did not account for correlation between repeated measures of metabolites over time. In the current sample, the moderate correlations between repeated measures of select metabolites tended to weaken with increasing follow-up time, e.g. the Pearson correlation between VLDL triglycerides measured at 8 years and 16 years, 8 years and 18 years, and 8 years and 25 years was 0.34, 0.33, and 0.25, respectively. For LDL cholesterol, the correlations for these same time periods were 0.61, 0.57, and 0.46; for apolipoprotein-B, these were 0.57, 0.51, and 0.43; and for glycoprotein acetyls, these were 0.31, 0.26, and 0.22. Mixed modelling was not presently feasible given the volume of traits examined, the sparsity of repeated measures for reliable non-linear modelling, and substantial occasion-level variability in several metabolite ratios. The development of a uniform approach for modelling trajectories under such conditions will improve estimates in future. Data were of a unique multi-generational nature, comprising offspring and their parents; this opens the possibility of additional sources of bias from cohort and period effects. Selection bias could also be differential between G1s and G0s. This is indicated by higher proportions of maternal smoking among included G1s than among included G0s who subsequently participated in clinic assessments several years after pregnancy, indicating that mothers who attended clinic assessments were a relatively healthy subset of those initially recruited.
We examined metabolites from a targeted NMR platform which is comprised of several traits considered a priori to be etiologically important for CHD, such as cholesterol and triglyceride content in various non-HDL particle types. However, this platform does not capture other potentially important factors like insulin, sex hormones, regulatory proteins, or different inflammatory traits. Examining additional metabolites measured using untargeted mass-spectrometry (MS) could reveal more detailed and novel sex differences. Examining circulating proteins from novel proteomic platforms could also be advantageous and help identify traits which are potentially more readily targetable via drugs. Such data are not currently available at scale or with repeated measures at different life stages.
Biological sex is an essentially randomised exposure within a causal inference framework with no expected influence of common causes (confounders), although such factors could influence study participation. Numerous explanations exist for sex differences in metabolites described here; these would largely be considered mediators. Estimates generated here therefore pertain to total effects of sex (male vs female), rather than direct effects of sex which are independent of potential mediating factors. In additional analyses, we adjusted for BMI measured at the time of metabolite assessment because biological sex is known to influence BMI [
7,
12] and BMI is supported by several observational and MR studies as influencing these same metabolites [
29‐
32]. This adjustment did not appear to substantially attenuate effect estimates for sex, suggesting that adiposity may not mediate/underpin the sex differences in metabolites observed here. Numerous potential lifestyle-related factors could still have mediating roles, such as adverse dietary patterns, smoking behaviour, or alcohol consumption. Future studies could examine these pathways using formal mediation analyses; although such lifestyle-related factors carry the added challenge of high measurement error and variability with time, with changing prevalence and potentially changing metabolic impacts across the life course.
Acknowledgments
We are extremely grateful to the families who participated in this study, the midwives for their help in recruiting them, and the whole ALSPAC team which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.