Association between fetal abdominal growth trajectories, maternal metabolite signatures early in pregnancy, and childhood growth and adiposity: prospective observational multinational INTERBIO-21st fetal study

Background: Obesity predominantly affects populations in high-income countries and those countries facing epidemiological transition. The risk of childhood obesity is increased among infants who had overweight or obesity at birth, but in low-resource settings one in five infants are born small for gestational age. We aimed to study the relationships between: (1) maternal metabolite signatures; (2) fetal abdominal growth; and (3) postnatal growth, adiposity, and neurodevelopment. Methods: In the prospective, multinational, observational INTERBIO-21st fetal study, conducted in maternity units in Pelotas (Brazil), Nairobi (Kenya), Karachi (Pakistan), Soweto (South Africa), Mae Sot (Thailand), and Oxford (UK), we enrolled women (≥18 years, with a BMI of less than 35 kg/m2, natural conception, and a singleton pregnancy) who initiated antenatal care before 14 weeks’ gestation. Ultrasound scans were performed every 5±1 weeks until delivery to measure fetal growth and feto–placental blood flow, and we used finite mixture models to derive growth trajectories of abdominal circumference. The infants’ health, growth, and development were monitored from birth to age 2 years. Early pregnancy maternal blood and umbilical cord venous blood samples were collected for untargeted metabolomic analysis. Findings: From Feb 8, 2012, to Nov 30, 2019, we enrolled 3598 pregnant women and followed up their infants to 2 years of age. We identified four ultrasound-derived trajectories of fetal abdominal circumference growth that accelerated or decelerated within a crucial 20–25 week gestational age window: faltering growth, early accelerating growth, late accelerating growth, and median growth tracking. These distinct phenotypes had matching feto–placental blood flow patterns throughout pregnancy, and different growth, adiposity, vision, and neurodevelopment outcomes in early childhood. There were 709 maternal metabolites with positive effect for the faltering growth phenotype and 54 for the early accelerating growth phenotype; 31 maternal metabolites had a negative effect for the faltering growth phenotype and 76 for the early accelerating growth phenotype. Metabolites associated with the faltering growth phenotype had statistically significant odds ratios close to 1·5 (ie, suggesting upregulation of metabolic pathways of impaired fetal growth). The metabolites had a reciprocal relationship with the early accelerating growth phenotype, with statistically significant odds ratios close to 0.6 (ie, suggesting downregulation of fetal growth acceleration). The maternal metabolite signatures included 5-hydroxy-eicosatetraenoic acid, and 11 phosphatidylcholines linked to oxylipin or saturated fatty acid sidechains. The fungicide, chlorothalonil, was highly abundant in the early accelerating growth phenotype group. Interpretation: Early pregnancy lipid biology associated with fetal abdominal growth trajectories is an indicator of patterns of growth, adiposity, vision, and neurodevelopment up to the age of 2 years. Our findings could contribute to the earlier identification of infants at risk of obesity. Funding: Bill & Melinda Gates Foundation. The prospective, multinational, observational INTERBIO-21st fetal study was conducted in maternity units in Pelotas (Brazil), Nairobi (Kenya), Karachi (Pakistan), Soweto (South Africa), Mae Sot (Thailand), and Oxford (UK) between Feb 8, 2012, and Nov 30, 2019.10 We enrolled 3598 women who initiated antenatal care before 14 weeks’ gestation, identified by ultrasound dating,11 and monitored their pregnancies to delivery. Their children’s health, growth, and development were monitored until the age of 2 years; a corrected age was used for infants who were born preterm. The inclusion criteria for the women were: being 18 years or older, a BMI of less than 35 kg/m2, natural conception, and singleton pregnancy. A sample of non-fasting venous blood was obtained at the earliest possible opportunity, at a median (IQR) gestational age of 13·2 weeks (11·9–17·2), plus a sample of umbilical cord venous blood at delivery; all samples were processed and stored as per protocol.12 The study was approved by the Oxfordshire Research Ethics Committee, the research ethics committees of the participating institutions, and their regional health authorities. All mothers provided written informed consent. After the dating ultrasonography, the women underwent an ultrasonography every 5±1 weeks until delivery using identical equipment (Philips HD-9, Philips Ultrasound, Andover, MA, USA). The methods to measure abdominal circumference growth have previously been described.13 From 22 weeks’ gestation, the umbilical artery Doppler Pulsatility Index, a marker of feto–placental perfusion, was measured. Mean umbilical artery Doppler Pulsatility Index values were expressed as z-scores of the international standard (reference).14 The anthropometric measurement methods used have previously been described.15 Infant age-specific and sex-specific z-scores and centiles were compared with the WHO Child Growth Standards.16 We assessed neurodevelopment at the age of 2 years using the INTERGROWTH-21st Neurodevelopmental Assessment (INTER-NDA), a multicultural, psychometric tool for children aged 22–30 months, designed to be implemented by non-specialists across international settings, which measures multiple dimensions of early development using directly administered, concurrently observed, and caregiver reported items.17 Metabolite signatures were measured using untargeted mass spectrometry (Sapient Bioanalytics, San Diego, CA, USA) and compound identification was performed by spectral matching to open access and private spectral databases. To visualise the metabolite feature data, we used Uniform Manifold Approximation and Projection (UMAP) with the Python package UMAP-learn.18 For more details of the metabolomic analysis, see the appendix (pp 3–5). To establish the abdominal circumference growth trajectories (the primary outcome of the study), we constructed models for the repeated measurement of abdominal circumference z-scores using finite mixture models.19 Group mean growth patterns were modelled using Gaussian distributions by applying a quadratic B-spline with one internal knot placed at the median.20 A z-score of 0 at 10 weeks’ gestation for each infant was added to highlight the growth trajectories relative to the initial ultrasonography and prevent grouping strictly by size. All mixture modelling was done using the hlme function of the R package lcmm.21, 22 Posterior group probabilities for models with three to five growth pattern groups were estimated,23 allowing for a low number of trajectories following recognised fetal growth patterns. Trajectories were established using only fetal measures without considering their 2-year outcomes or metabolomic data, to which the statistician (SAR) was masked during the analysis. The optimal number of groups for each growth measure was selected on the basis of the best model fit using the Bayesian Information Criterion, and the number of infant in the smallest group that comprised at least 2·5% of the total sample. Every infant received a posterior probability of being in each group and was then assigned to the definitive group with the highest probability (appendix p 9). Abdominal circumference growth trajectories comprised the primary independent variables. The trajectory tracking the 50th centile of the INTERGROWTH-21st standard13 was the reference group. Outcomes from the neurodevelopmental assessment were based on normative INTER-NDA scores.17, 24 Categories of low visual acuity (logarithm of the Minimum Angle of Resolution [logMAR] >0·4) and high contrast sensitivity (>3%) were based on Cardiff normal values.25 We used linear regression models to assess the relationships with continuous and normally distributed measures of growth, cognition, language, and fine motor and gross motor domains stratified by the duration of any breastfeeding (<7 months or ≥7 months). The positive affect domain was reverse coded, and both positive and negative scales were modelled as count data using generalised linear models, with a Poisson distribution and log-link function and a variance correction for over-dispersion. Vision outcomes were modelled as binary variables using Poisson regression with robust standard errors because they were not rare outcomes and negative binomial regression models often did not converge. The results represent a change in score or relative risk for each trajectory compared with the reference median growth trajectory. A p value of less than 0·05 was deemed statistically significant. We corrected our results for multiple comparisons using the Benjamini-Hochberg False Discovery Rate correction to control for type 1 error rate at <0·05. We selected covariates suspected a priori to be in the causal pathway, using separate directed acyclic graphs for growth and neurodevelopmental outcomes (appendix p 21). In all adjusted models, we included the infant's sex and age at assessment, preterm birth, maternal age, maternal education, and smoking during pregnancy. As done previously,26 we conducted a primary stratified analysis to evaluate the effect modification by the duration of breastfeeding (<7 months or ≥7 months). We included abdominal circumference and head circumference trajectories in the models to explore the independent effect of abdominal circumference phenotypes.10 All analyses were performed using R (version 3.6.0; The R Core Team, Vienna, Austria) and STATA (version 15.1; StataCorp, College Station, TX, USA). The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.