Study design
This is a secondary analysis of the RADIEL study, a multi-center, randomized controlled intervention trial aiming at prevention of GDM through lifestyle modification. The study enrolled a total of 724 women at high diabetes risk, in the Helsinki Metropolitan area or in Lappeenranta, Finland, during years 2008 to 2011. The recruited women had a history of GDM and/or obesity (BMI \(\ge\) 30kg/m2). Exclusion criteria were age < 18 years, current type 1 or 2 diabetes, medication altering glucose metabolism, multiple pregnancy, severe psychiatric problems, physical disabilities, substantial communication difficulties, and current substance abuse. The participants were randomized to a control group or to a combined lifestyle intervention group, emphasizing dietary goals following Finnish nutrition guidelines, physical activity (150 min/week), and limiting gestational weight gain. During pregnancy the study visits took place once in each trimester and 6 weeks, 6 and 12 months after delivery. Previous publications have provided detailed information on the study design and methods (21–23).
The study complies with the Declaration of Helsinki and was approved by the Ethical Boards of Helsinki University Hospital (HUS) and South-Karelia Central Hospital (SKCH). All the participants entered the study voluntarily and gave written informed consent. They were also free to discontinue at any point.
Measurements
Pregnancy
Every study visit included anthropometric measurements (height, weight, and blood pressure) and blood samples for analyzing markers of glucose and lipid metabolism as well as inflammatory markers. A 2-hour 75g oral glucose tolerance test (OGTT) was performed at enrollment (if before pregnancy), in the first and second trimesters of pregnancy, and 6 weeks and 12 months postpartum. GDM diagnosis was based on one or more pathological values in the OGTT (normative values: 0h < 5.3 mmol/l, 1h < 10.0 mol/l, and 2h < 8.6 mmol/l). Prepregnancy weight was self- reported for those recruited in early pregnancy and the weight at the last visit before pregnancy for those recruited before pregnancy. Gestational weight gain (GWG) was defined as the difference between prepregnancy weight and weight at the third trimester study visit.
Questionnaires provided data on maternal years of education and lifestyle such as substance use (smoking, alcohol) and moderate-intensity physical activity (PA) (min/week). Food frequency questionnaires (FFQs) offered information for calculating a Healthy Food Intake Index (HFII), which served as a description of maternal diet as a whole (24). Maximum score was 18 and fulfilling nutritional goals contributed points as follows: fast food (0–1 points), bread fat spread (0–2 points), low-fat cheese (0–1 points), intake of high-energy/low-nutrient snacks (0–2 points), sugar-sweetened beverages (0–1 points), high-fiber grains (0–2 points), low-fat milk (0–2 points), fish (0–2 points), red and processed meat (0–2 points), vegetables (0–2 points), and fruits and berries (0–1 points). A higher score indicated a healthier diet.
Offspring
The data from birth such as date of delivery, gestational weeks at delivery, delivery mode and placental weight were obtained from the hospital birth records. They also provided data on newborn characteristics at birth (weight, length, head circumference, and Apgar points). Cord blood samples from umbilical vein were collected at delivery for metabolomics analyses.
Communal childcare clinic registries from first two years of life provided additional data including growth (weight, height) and information on early feeding: total duration of breast feeding and exclusive breastfeeding, and the age at introduction of solid foods.
Our previous study used latent class mixed modelling and identified three distinct growth profiles (ascending n = 80, intermediate n = 346, and descending n = 146) based on the development of ponderal index (PI) during the first two years of life (12).
Metabolomic Profiling
Metabolomics analysis was performed using targeted nuclear magnetic resonance (NMR) based technique (Nightingale Ltd, Helsinki, Finland). Only the metabolites detected in > 70% of the participants were included in the analyses. Thus, 220 metabolites in each trimester and 70 metabolites from cord blood were included in this study. These metabolites cover multiple metabolic pathways, including lipoprotein lipids and subclasses, apolipoproteins, fatty and amino acids, ketone bodies, glycolysis and gluconeogenesis-related metabolites, fluid balance, and inflammation. NMR platform has been used in large-scale epidemiological samples and in samples of pregnant and non-pregnant women (25). The cord blood metabolome platform became available in 2019.
Statistical Analysis
The data are presented as means with standard deviations (SD), medians with interquartile range (IQR), or as frequencies with percentages. The Shapiro–Wilk test was used to examine normal distribution of the variables. The Fisher’s test, Mann–Whitney U test, Kruskal-Wallis test, Chi square test, ANOVA, or the independent samples T test were used for between-groups comparisons, as appropriate. We excluded from the analysis extreme outliers (deviating 1.5 times IQR from Q1 and Q3). Metabolites were log-transformed to achieve normal distribution. Principal component analysis (PCA) was used to assess the overall variability in the metabolomic data and 25 PCs explained 95% of the variance in the dataset. After correcting for multiple testing, associations with p < 0.002 (0.05/25 to account for 25 PCs) were considered significant.
Logistic regression analysis was used to assess associations between metabolites levels (maternal and cord blood) and growth trajectories. The analyses were adjusted for gestational age, birth weight SD (standardized for gestational age), gender, delivery mode, and maternal smoking. Mixed model was used for studying the longitudinal associations between metabolomics throughout pregnancy and early growth, and these analyses were also adjusted for gestational age, birth weight SD, gender, delivery mode, and maternal smoking. All analyses were performed with the SPSS 24.0 software program (SPSS Inc., Chicago, IL, USA) and figures produced with R studio 2022.07.1.