The Growing Up in Singapore Towards healthy Outcomes (GUSTO) study recruited pregnant women aged ≥ 18 years from Singapore’s two major public maternity hospitals (National University Hospital and KK Women’s and Children’s Hospital) between June 2009 and September 2010. Inclusion criteria for pregnant women were as follows: (1) Singaporean residents aged 18 years and above, (2) attending either KK Women’s and Children’s Hospital (KKH) or National University Hospital (NUH), and (3) intending to deliver and reside in Singapore for the next 5 years. Of 3751 women screened, 2034 met eligibility criteria, and 1344 were recruited (response rate 66.1%). These women gave birth to 1098 singleton infants. GUSTO mothers and children have been followed up since birth. At postpartum year-6 follow-up, 953 (86.8%) offspring were assessed, of whom 460 (48.3%) provided blood samples. Participants were included in the analytic sample if they had maternal glucose data and at least one AA as an outcome (n = 422, see Additional file 2: Fig. S1). Informed written consent was obtained from the women at the study entry, and the National Healthcare Group Domain Specific Review Board and SingHealth Centralized Institutional Review Board approved the study. Detailed study designs and recruitment have been published elsewhere [11].

At 26–28 weeks of gestation, pregnant women without pre-existing diabetes underwent a 2-h 75 g oral glucose tolerance test (OGTT) after an overnight fast [11]. Fasting and 2-h post-challenge venous blood samples were collected in fluoride-containing tubes, and glucose concentrations were assessed quantitatively (Advia 2400 Chemistry system and Beckman LX20 Pro analyzer). Women were diagnosed with GDM based on World Health Organization’s (WHO) 1999 guidelines (fasting plasma glucose ≥ 7.0 mmol/L and/or 2-h glucose ≥ 7.8 mmol/L]) [12]. Mothers who were diagnosed as having GDM were either managed by diet and/or medication (i.e., metformin and insulin) according to standard protocols practiced at study sites.

We collected fasted peripheral blood from 460 children at approximately six years of age. Blood samples were immediately fractionated, aliquoted, and stored at – 80 °C until transported on dry ice to Nightingale Health (Helsinki, Finland) for further analyses [13]. Circulating metabolite concentrations were quantified using an automated nuclear magnetic resonance (NMR)-based high throughput metabolomics platform [14]. The software of NMR platform undertakes automatic quality control [14]. After quality control, metabolomic data were available for 457 children. As the AAs we investigated were part of the metabolomic profile measured by the Nightingale platform, our analysis is confined to the nine AAs analyzed in this study, which included alanine, glutamine, glycine, histidine, isoleucine, leucine, valine, phenylalanine, and tyrosine. We calculated total BCAAs as the sum of isoleucine, leucine, and valine concentrations and aromatic AAs as the sum of phenylalanine and tyrosine [14]. Also, among the metabolomic data, fatty acids were measured. Measurements of other cardiometabolic measures implicated in the development of cardiometabolic diseases, such as homeostasis model assessment of insulin resistance (HOMA-IR), interleukin 6 (IL-6), high-sensitivity C-reactive protein (hsCRP), low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and triglyceride, are shown in Additional file 1.

At recruitment, maternal age, ethnicity, highest education level, parity, pre-pregnancy weight, and family history of diabetes were obtained through interviewer-administered questionnaires. Maternal height was measured using SECA 213 Stadiometer (SECA Corp, Hamburg, Germany) [15]. Maternal weight during pregnancy was measured using SECA 803 Weighing Scale (SECA Corp, Hamburg, Germany). Body weight and height were used to calculate pre-pregnancy body mass index (ppBMI, in kg/m2). Gestational weight gain (GWG) was calculated as the difference between the final measured weight before delivery and self-reported pre-pregnancy weight. Excessive GWG was classified according to the Institute of Medicine recommendation [16]. Information on dietary intakes of women was collected at 26–28 weeks of gestation using 24-h recalls and 3-day food diaries, from which diet quality (score range: 0–100) was measured by the Healthy Eating Index [17].

Date of child age, weight, and height were collected at their year-6 follow-up. Child BMI was calculated, and sex-specific BMI z-scores were generated using the WHO references [18]. Since some AAs, such as BCAAs, cannot be synthesized from other metabolites by the human body but are derived from diet intake [19], we also considered the influence of food intake on blood AA concentrations. At the year-5 visit, about 1 year before AA measurements, child protein intake in the previous month was assessed using an interviewer-administered 112 food items semi-quantitative food frequency questionnaire completed by the caregivers [20].

Distributions for all variables were checked for skewness and kurtosis. Maternal fasting glucose, 2-h glucose concentrations, and all offspring mid-childhood AAs were analyzed as continuous variables, and GDM status was analyzed as a binary variable (present/absent). Comparisons of characteristics between GDM and non-GDM participants were analyzed by Student’s t-test, non-parametric comparison test, or χ2-test when applicable.

Multi-variable linear regressions were applied to examine the associations of maternal fasting glucose concentration at test, 2-h glucose concentrations, and GDM diagnosis with child AA profiles, using three models: unadjusted model; model 1, adjusting for maternal age, ethnicity, maternal education, parity, family history of diabetes, ppBMI and child sex; and model 2: model 1 and additionally adjusting for child age and BMI z-score at mid-childhood. Furthermore, we tested the interactions between maternal GDM status with maternal age and ppBMI, respectively.

In sensitivity analysis, we considered potential confounders that were associated with maternal glycemic levels and child AAs. Linear regression models were performed with additional adjustments for excessive GWG, hypertension diagnosed during pregnancy, maternal Health Eating Index during pregnancy, child fatty acids, and child protein intake assessed at mid-childhood in addition to model 2. As child birthweight and gestational age at birth are recognized as significant risk factors for child cardiometabolic health, we conducted a sensitivity analysis with further adjustment for these two variables, in addition to model 2. Also, GDM mothers treated with medication or missing treatment data were excluded to investigate the direct effect without medication intervention between GDM and child AAs concentrations. We further corrected for multiple comparison using the Benjamini–Hochberg method to control false-discovery rate (FDR) [21].

To assess the potential relationships among mid-childhood offspring plasma AAs and cardiometabolic risks within our cohort, partial Spearman rank correlation was performed a posteriori after adjusting for maternal age, ppBMI, GDM status, GWG, child sex, age and BMI z-score at mid-childhood. The cardiovascular phenotypes (i.e., SBP, DBP, carotid intima-media thickness, pulse wave velocity, and augmentation index) and metabolic phenotypes (i.e., child HOMA-IR, IL-6, hsCRP, total fatty acids, total polyunsaturated fatty acids, total saturated fatty acids, total monounsaturated fatty acids, LDL cholesterol, HDL cholesterol, and triglyceride) were assessed accordingly.

Analyses were performed in Stata 16.0 SE (StataCrop LP, TX, USA). For all analyses, we standardized exposures (fasting glucose and 2-h glucose) and outcomes (each AA) to present effect size in standardized regression coefficients. P values and 95% confidence intervals (CIs) are presented accordingly. A significant P-value (two-tailed) was defined as < 0.05.