The early life nutritional environment and early life stress as potential pathways towards the metabolic syndrome in mid-life? A lifecourse analysis using the 1958 British Birth cohort

This study sets out a comprehensive lifecourse approach to examine the influence of both the early nutritional environment, and the psychosocial environment, on the subsequent risk of MetS. By taking a lifecourse approach, many mediating factors that may explain pathways between early nutrition and early psychosocial stress and subsequent MetS are examined. The main finding from our study is that both early nutritional and psychosocial environments are independently associated with subsequent risk of MetS. Mother’s pre-pregnancy BMI was associated with MetS in mid-life after taking into account a large range of confounders and three potential pathways (socioeconomic/materialist, psychological and behavioral). The association appeared to be mediated to a sizeable extent by offspring’s BMI at age 23, suggesting mechanisms via a mother to child transmission of BMI. However, participants’ BMI at age 23 did not fully explain the link between mother’s pre-pregnancy BMI and MetS suggesting that unidentified mechanisms may be operating other than those captured through the confounding and mediating variables used in this study, specifically health behaviors and socioeconomic pathways. The role of ACE was not clear. The positive association between ACE and MetS was indeed not significant after taking early life confounders into account, in particular childhood socioeconomic conditions and birth conditions that were independently associated with risk of MetS.

The main strength of this study is that data were prospectively measured and the large number of confounders and mediators we consider, which made that this study includes a large panel of confounders and mediators to explore MetS in a lifecourse perspective. Consequently, we were able to analyze both early nutritional and psychosocial exposures that are mainly studied separately in the literature and also to consider early birth conditions that are largely not taken into account in studies on MetS. There are also a number of limitations that need to be considered, however. The first is regarding the definition of MetS we used in the study. The definition of MetS has varied over the past decade. The prevalence of MetS may be lower when using definitions other than NCEP-ATPIII. However the risks of cardiovascular events, diabetes mellitus and hypertension are similar for NCEP-ATPIII and American Heart Association or International Diabetes Federation definitions [45]. Glycaemia was not collected in the biomedical survey used here therefore we used HbA1c with a cut-off above 6.5 % to define hyperglycaemia. HbA1c has been defined as a marker to identify diabetic status and used in other studies instead of glycaemia [46]. We were also unable to take into account whether people were treated for hypertension, diabetes or high cholesterol level because accurate data were not available. Our definition of MetS is thus a conservative one. Another weakness is the amount of missing data caused by attrition in the cohort. This weakness is partly addressed by the use of multiple imputations in our models. The standard application of multiple imputation assumes that data are MAR meaning that the probability of missing data depends on the observed data but not the missing data, which is an unverifiable assumption. Thus we cannot rule out that some data are Missing Not At Random. A way to render the MAR assumption more plausible and to limit the impact of missing not at random missingness is to include a large numbers of covariates in multiple imputations. In our analyses all important causes of missingness such as demographic characteristics, socio–economic position indicators and behavioral variables are included in the models. Any other missingness that is not accounted for by these variables is assumed to be completely random since all major systematic causes of missingness have been accounted for. We believe that this is a reasonable assumption since it has been shown that socio–economic position and age and are the main drivers of attrition in population surveys in the UK [47, 48]. Furthermore, analyses on complete cases gave a similar result which increases the robustness of our findings. Another limitation concerns the NCDS recruitment that was done in March 1958. So we do not take account for potential effects of birth-month variations that may influence the neuroendocrine regulation of metabolism and obesity [49]. Other potential limitations concern the analyses. To illustrate the influence of confounders and mediators on the associations between our two main exposures and MetS, we calculated B attenuation after inclusion of potential confounders and mediators. This approach is not a formal mediation analysis and may provide biased estimates, however when controlling for potential mediator-outcome confounders, as we did here by considering a large range of variables in analyses, the risk of bias is limited [41]. We also used a bootstrap method to improve the precision of the percent attenuation. Our use of health behaviors at 23 years as a proxy for behavioral patterns in early adulthood is also a limitation. Adult health behaviors were collected in the cohort at ages 23, 33, 41, 46. Constructing health behavior pathways using repeated measures, may explain a significantly greater part of the association between mother’s pre-pregnancy BMI and MetS compared to a baseline-only assessment of behaviors, as observed elsewhere [43]. However the inclusion of behaviors in the model did not change the association between maternal BMI before pregnancy and MetS except for BMI. So it is not likely that behaviors even by considering them in a longitudinal way, explain a significant part of the association between maternal BMI before pregnancy and MetS. Finally, it is possible that we omitted important confounders or mediators in our models. It would have been optimal to control for the mother’s gestational diabetes status, to control for an intrauterine exposure to hyperglycemia, however accurate information on this was not available. By including birthweight we should be able to capture any macrosomia effects of gestational diabetes on the cohort member. Moreover, some of the measures we used may be susceptible to measurement error, like self-reported data on maternal weight that may have resulted in some misclassification, even if self-reported weight is generally accurate in validity studies [50].

We found that mother’s pre-pregnancy BMI may be an important determinant of metabolic pathology in mid-life, mainly operating via a mother-child body-size transmission pathway. Previous works have shown that children exposed to maternal obesity in early life presented higher levels of some MetS components [17]. In our study this association was mainly attenuated after including offspring’s BMI at 23 years in the model, suggesting that this association could be explained in a significant part by a mother to child transmission of BMI. An intergenerational adiposity association is a widely observed phenomenon [16, 51‐53]. One of the main issues is to explain underlying pathways and notably the respective contributions of shared environments and genetic factors in this transmission. Some studies have observed a similar influence of mother and father’s BMI on the subsequent risk of obesity in offspring [52]. Such results may point towards an influence of shared family characteristics, such as diet and physical activity. However studies that included such mediators concluded that the mother to child transmission of BMI was attenuated but still persisted after taking adult health behaviors, including diet, or socioeconomic factors into account [16, 51]. In fact, the intrauterine environment may have an important role to play [9]. Offspring exposed to maternal hyperglycaemia during their intrauterine development are more at risk of metabolic disorders such as obesity in adulthood, and that changes in maternal nutrition, in particular during fetal development, may influence subsequent predisposition to the MetS [9, 19]. Such results are in accordance with the fetal overnutrition hypothesis or an early life nutritional programming of the metabolic syndrome, a phenomenon that has been shown in animals [18]. To investigate this hypothesis in humans the majority of studies use birthweight as a proxy of intrauterine exposures during pregnancy. In our study we consider birth and perinatal conditions, in particular birthweight, and the inclusion of such variables in models did not significantly change the association between mother’s pre-pregnancy BMI and MetS in offspring. Unfortunately, in our study, we had no information on mothers’ diet during pregnancy and after birth, no information on the way babies were fed except breastfeeding and no information on diet during childhood. The information collected on mother’s pre-pregnancy BMI, however, offers an opportunity for us to capture through a proxy variable something of cohort members’ early life nutritional environment of cohort members. Ideally, we would have liked to have information on weight gain during pregnancy or parents’ waist circumferences before pregnancy. However such data are not available in the NCDS cohort. The parent–child transmission of adiposity deserves further investigation using more relevant and detailed data to characterize both nutritional environment around pregnancy and early life, and mothers’ metabolic status. With this in mind, studies analyzing the effect of mother and father adiposity on offspring adiposity at different stages of life are needed.

We found no association between early ACE and MetS after adjusting for confounders such as early socioeconomic and birth conditions. Few previous studies have observed an influence of ACE by using retrospective measures for ACE [24, 54] or using both retrospective and prospective measures [30]. There is strong evidence for the role of stress in the long term etiology of cardiometabolic diseases [31]. This evidence leads to questions about the measures used to capture psychosocial stress and specifically our definition of adversity. Thomas et al. [30] have shown that the link between adversity and subsequent risk of obesity and type 2 diabetes in mid-life was different according to the definition used for adversity suggesting that we can be exposed to measurement error or that different measurements of ACE may be capturing different types of stress. In particular we did not use any variables on sexual or physical abuse that has been shown as associated with obesity [30], because these variables were retrospectively measured (at the same time as MetS was measured) as opposed to our definition which used only variables measured prospectively to avoid recall bias. Of course, childhood adversity may not be the only or best measure of stress in early life, both prenatal and postnatal periods could also a phase where early life stress may be picked-up. For example, maternal stress during pregnancy has been associated with central adiposity in children at 13 years [55]. In animals, prenatal stress has been associated with metabolic disturbances in adult offspring [56]. As we previously observed [38], emergency caesareans were associated with subsequent risk of MetS in the NCDS cohort. Recent reviews have shown that caesarean section was a risk factor of adult obesity [57, 58]. The main hypothesis refers to the nature of the offspring microbiome that could differ according the mode of delivery [59]. In our study, it was not caesarean sections per se that influenced the risk of MetS but only if they were carried-out in an emergency. Emergency caesareans may act as a proxy of perinatal stressful events that may play a role in subsequent risk of MetS. Other perinatal variables may also be proxies for early life stress [54, 60], like birthweight or smoking during pregnancy that were associated with risk of MetS in our study, in particular among women. It is noteworthy that the influence of these ‘perinatal stress’ proxies appear stronger in women than in men suggesting than women could be more sensitive to early life stress exposures than men in accordance with other studies. Early life exposures have indeed been found to be more strongly associated with autonomic nervous system reactivity in women [60]. Regarding risk of MetS, different biological and social pathways have been identified in men and women [13, 44].

Psychosocial stress is also associated with socioeconomic conditions [25] which are known to be associated with subsequent health and in particular with MetS [12, 14, 61]. In our study, psychosocial adversity was not significant after including socioeconomic variables such as mother’s level of education and father’s social class at birth which were independent risk factors of MetS. The fact that the link between childhood socioeconomic conditions and MetS persisted in the full model, after the inclusion of a large range of confounders and mediators, questions the mechanisms involved and not considered in the study. Pathways likely to explain this association deserve to be explored in more depth.