The current study is novel in providing an overview of the shared molecular pathways arising from in-utero adversities in influencing different psycho-cardiometabolic measures as a multimorbidity in adolescence. Importantly, we employed a robust approach by validating our comprehensive model in both early life and adolescent factors in two independent birth cohorts from two culturally diverse countries (Finland and Australia). We observed both direct and indirect effects of prenatal latent factors on adolescent psycho-cardiometabolic multimorbidity through composite epigenetic scores, displaying persistent role of epigenomes influenced by early environment until adolescence. Our objective was to further expand on the understanding of the structure of prenatal adversities and mental and physical health especially in adolescence which has been lacking from the existing literature.
In line with our previous findings [8], the prenatal latent factor model fitted the data well and revealed similar correlations between both cohorts. Various biological and psycho-social factors are known to aggregate during gestation influencing birth outcomes cumulatively and increasing the magnitude of the risk. A distinct pattern was observed for adolescents’ psycho-cardiometabolic traits, showing separate biological and psychological groupings (Fig. 2), as seen in a previous study [32]. Importantly, similar structure and factor loadings were observed for both cohorts when constructed independently. While, biological patterns were quite comparable amongst each other, heterogeneity was observed between ‘F4adolescent-Mental health’ and its correlation with other factors across the cohorts. For NFBC1986, ‘F4adolescent-Mental health’ was largely represented by anxious-depressed measure, and it was negatively correlated with ‘F1adolescent-Anthropometric’ and ‘F3adolescent-BP’. On the other hand, in the Raine Study, ‘F4adolescent-Mental health’ was equally characterised by each psychological symptom subscale (anxious-depressed, withdrawal depressed and somatic complaints) and was positively correlated with other latent factors. This suggests that biological parameters behave similarly between different populations but not psychological aspects. The reason may be that these patterns in adolescent period behave differently from adult patterns, attributable to their rapid hormonal changes, and wide range of biological, psychological and social challenges occurring in adolescence [33]. Additionally, this study comprised two culturally and socially different populations from different continents, where dynamics of perceiving health may also vary largely [34].
Our multimorbidity second order factor sheds important insight on the relationship between and with psycho-cardiometabolic traits in its entirety. The factor structure was largely representative of biological measures and less of mental states. This was expected as the correlations between biological factors and mental states were weak in the CFA model. Despite the imbalance, it is very interesting to note that all the psycho-cardiometabolic comorbidity factors loaded into one factor that replicates between cohorts. Individuals with poor mental disorder have up to 14 years of shorter life expectancy, which is often partly accounted by the co-occurring physical diseases [35]. Moreover, heritability studies suggests that the causes of multimorbidity have both genetic and environmental components shared between physical and mental disorders [36, 37]. Therefore, it was worthwhile to unravel the shared relationships between psycho-cardiometabolic multimorbidity, which is not captured when looking at the traits individually.
Together, our SEMs revealed a comprehensive approach to understand shared pathways to multimorbidity. Specifically, amongst prenatal latent factors ‘F1prenatal-BMI’ had the strongest direct influence on adolescent psycho-cardiometabolic multimorbidity. Maternal BMI embodies both a biological dimension as well as lifestyle and social factors such as maternal age, marital status, smoking and alcohol use [38]. These correlations were also reflected in our correlation matrix (Supplementary Fig S1, Additional File 1).
In the same way, ‘F2prenatal-SOP’ showed a direct effect on multimorbidity, but here the direction was negative, and no effect was modulated through epigenetic factors. This suggest that not all early life influences have epigenetic influence, particularly social factors (maternal age, marital status, parity). Additionally, the negative effect on the psycho-cardiometabolic multimorbidity factor highlights the protective dimension of social factors such as decreased parity, younger maternal age, and married status. The ‘F3prenatal-Lifestyle’, while not showing a direct effect, showed a strong indirect effect, primarily through DNAmMSS, which was derived using a machine learning approach to predict maternal smoking during pregnancy in a previous study [17]. Its strong intermediary role from ‘F3maternal-Lifestyle’ in our pathway analysis (Fig. 3) confirms the validity of the score as a proxy of ‘in-utero adversity’ since it mirrored the known association of prenatal smoking related epigenetic changes on cardio-metabolic health of the offspring in observational studies [13, 14, 16].
Epigenetic markers are important molecular readout of diverse environmental exposures across the lifespan. In our study, we observed that DNAmTL and DNAmMSS showed direct as well indirect influence going through PhenoAge marker. Increasing evidence supports the concept of molecular ageing as a component of chronic diseases and an important tool for predicting biological age of an individual [18]. Biological age evaluated using these epigenetic markers has been shown to vary within individuals of same chronological age based on the incidence of chronic mental and physical diseases [39] and is also significantly influenced by intrauterine conditions [40]. DNAmTL showed negative relationship with all the path variables in our study, corroborating previous evidence that telomere length shortens with age (TL) [20]. Telomere length is largely determined already during early foetal development and associates with several maternal factors during pregnancy, including maternal smoking, stress, socioeconomic status, BMI and gestational diabetes [41]. It is speculated that shorter TL may weaken the replicative potential and diminish somatic repair contributing to degenerative diseases such as cardio-metabolic diseases [42]. Correspondingly, our findings regarding PhenoAge were consistent with previous studies in showing the association with cardiometabolic risk factors. Importantly, it was observed to mediate the indirect effect of all the other epigenetic biomarker path factors in our study, highlighting its importance on phenotypic outcomes. Studies from both European and African-American cohorts have reported association of PhenoAge with a wide range of phenotypes such as blood pressure, insulin, glucose, triglycerides, and low density lipid cholesterol [19, 43]. Nonetheless, PhenoAge is relatively new DNAm age estimate, so further replication is required to fully understand its association with the range of health outcomes.
A further point to note is that in both prenatal biopsychosocial and in adolescent psycho-cardiometabolic constructs, metabolic factor (F1maternal-BMI and F1adolescent-Anthropometric) showed consistently strongest correlations and largest representation in the latent factors. Thus, suggests that in our study adiposity, a potentially modifiable factor, had a predominant role over other predictors in defining adolescents’ health.
Strengths and limitations
This is the first study to use a comprehensive factor structure approach to examine the latent relationship between prenatal adversities factors and later adolescent psycho-cardiometabolic health. The benefits of using a factor structure instead of an individual measure of biological health or psychological status is that it allows us to account for the different aspects of these variables, represented by the sub-factors of commonality [29]. A further advantage is that the magnitude of factor loadings is determined empirically and does not comply with the assumption that all component measures have equal weighting. We used data from two cohorts from two culturally different continents (Europe and Australia). The consistent and comparable findings from the cohorts further validate and increase generalizability and robustness.
We acknowledge the limitations of this study. In NFBC1986 methylation sample size was much smaller than the full cohort sample, however, the characteristics of both samples were relatively comparable (Supplementary Table S3, Additional File 1). There are analytical difficulties associated with measuring latent exposures. We included the most closely related and easily accessible prenatal measures as it can be challenging to develop a comprehensive model with maximum available measures, for which, we had limited similar prenatal measures harmonized between both cohorts. There are other DNAm age estimates such as Horvath,[25] Hannum [24], and Horvath’s estimate for skin and blood [26] developed to predict biological age, however, in our study DNAmTL and PhenoAge were more closely related to phenotypic markers compared to the other DNAm age estimates. Both these markers were also recently developed and suggested to correlate better with mortality and morbidity [19]. A common limitations of biological age estimates is that they rely on specific organs or tissue, however PhenoAge has been observed to relate with the wider range of tissue and cell types than other markers [19].