Study participants
This current study uses data from the U.S. National Institute of Child Health and Human Development (NICHD) Scandinavian Successive Small-for-Gestational Age births study (The SGA Study) [20]. The SGA Study is a large multi-center prospective cohort study conducted in Trondheim and Bergen (Norway) and Uppsala (Sweden) from 1986 to 1988. The SGA Study was designed to study longitudinal fetal growth, as well as perinatal and postnatal outcomes among mother and child [20]. In brief, all pregnant women (< 20 weeks gestation) in the study catchment areas who were expecting their 2nd or 3rd child were eligible for study inclusion and made the first appointment (n = 5722) (Fig. 1). Women with elevated risk for an SGA birth were intentionally oversampled. Risk factors for SGA birth included a previous low birth weight child, previous perinatal death, low maternal pre-pregnancy weight (< 50 kg), maternal smoking at conception and/or chronic maternal hypertension or renal disease. All high-risk pregnancies resulting in an SGA birth (birth weight below 10th percentile adjusted for sex and parity), and a 10% random sample of the study population were invited for follow-up when children were five years of age (n = 791). Of these, 534 (68%) attended the 5-year evaluation. In the current study, 412 mother-child pairs (137 SGA births and 275 non-SGA births) were included in the analyses (Fig.1).
Exposure assessment of maternal serum POP concentrations
According to study protocol (1986–88), maternal serum samples were collected in the 2nd trimester (gestational week 17–20) and stored at minus 80 °C for later analysis. Analyses of maternal serum PFAS and OC concentrations were performed.
PFAS analyses
The PFAS analyses were performed at the laboratories of Norwegian Institute for Air Research, Tromsø, Norway (NILU). Maternal serum samples were quantified for two target analytes including perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS). Detailed information about the sample preparation, extraction method, analytical method, reagents and instrumentation is previously reported [21, 22]. Maternal serum PFAS concentrations were determined using sonication-facilitated liquid-liquid extraction, activated ENVI-carb clean-up [23], and analyzed by ultrahigh pressure liquid chromatography triple-quadruple mass-spectrometry (UHPLC-MS/MS). Participation in the AMAP Ring Test [24] ensures that the uncertainties of the analysis are within ±15–20% of the assigned values.
OC analyses
Maternal serum OC concentrations were analysed at the Institut National de Santé Publique du Quebec, Centre Toxicologie, Quebec. Several OCs were measured, including hexachlorobenzene (HCB), oxychlordane, polychlorinated biphenyl (PCB) 52, 101, 118, 153, 156, 170, and 180, p,p’-dichlorodiphenyldichloroehylene (p,p’-DDE), p,p’-dichlorophenyltrichloroethane (p,p’-DDT), β-hexachlorohexane (β-HCH) and trans-nonachlor (t-NC). In short, 0.5–1 ml serum sample was extracted using hexane (2 × 6 ml), ethanol (2 ml) and saturated ammonium sulphate solution (2 ml). This method is a slight modification of the one described by Sandanger et al. [25], where the samples were cleaned up using 1 g of activated fluorisil on an automated Liquid handler system before GC-MS analysis [26]. Uncertainties of the analyses are within ±15–20% of the assigned values, which are confirmed by participation in the AMAP Ring Test [24]. Lipids were enzymatically determined and the summed lipid amounts were calculated based on triglycerides and cholesterol measurements using the following formula:
Total lipids = 1.33*triglycerides +1.12*cholesterol +1.48 (g/l) [27]. This formula showed good correlation with complete formulas including phospholipids [28].
We report PCB 153 as a proxy for total PCBs, and excluded p,p-DDT because of low limit of detection (LOD) (> 50% of samples < LOD). LODs are listed in Table 2. Values below LOD were replaced by LOD/√2.
Outcome assessment of child overweight/obesity
Child weight was recorded at 5-year follow-up in the clinic by trained professionals using a standard procedure. Standing weight was recorded to the nearest 100 g. Standing height was measured according to standard procedures and recorded to the nearest 0.1 cm [29]. BMI was calculated from weight in kilograms (kg) divided by height in meters squared (kg/m2). We used offspring age (in months), offspring sex and offspring BMI to calculate age-and-sex-specific BMI z-scores. BMI percentiles were based on the 2006 WHO child growth standards for children 5 years or younger [30], and the 2007 WHO growth standards for children and adolescents aged 5 to 19 years [18]. We assessed child age-and-sex-specific BMI z-scores as a continuous outcome at 5-year follow-up. We also analyzed child overweight/obesity at 5-year follow-up categorically (BMI ≥ 85th percentile for age and sex compared to BMI below the 85th percentile) [31]. Skinfold thickness was measured once using a Harpenden caliper (John Bull, British Indicators Ltd.) to the nearest 0.10 mm and 60 s after release of the grip to allow full tension placed on the compressed skinfold. Subscapular skinfold thickness was measured below the inferior angle of the left scapula, and triceps skinfold thickness was measured over the triceps in the middle of the left upper arm [29, 32]. We calculated age-and-sex-specific z-scores for triceps and subscapular skinfolds according to the Center for Disease Control and Prevention (CDC) 2000 Growth Charts for children from 1.5 to 20 years of age [33].
Covariates
Information on maternal age, height, pre-pregnancy weight, education, smoking habits, previous breastfeeding duration and inter-pregnancy interval were collected via in-person interviews and self-report questionnaires during the original study period as per SGA Study protocol. Maternal pre-pregnancy BMI was calculated based on self-reported height and weight at first study visit. We calculated maternal weight gain up to 17 weeks of gestation as the difference between self-reported pre-pregnancy weight and clinically recorded weight closest to gestational week 17 (done by the woman’s own midwife or GP). Based on a known J-shaped association with adverse perinatal outcomes including restricted fetal growth [34], we categorized the inter-pregnancy interval as <18 months, 19–60 months and > 60 months since their last birth.
Statistical analyses
Maternal serum PFAS and OC concentrations were logarithmically (ln) transformed to obtain normal distribution. We used wet weight maternal serum PFAS concentrations, and lipid-adjusted maternal serum OC concentrations [27].
We used multivariable linear regression with 95% confidence intervals (CIs) to examine the association between ln-transformed maternal serum concentrations of seven separate POPs (PFASs: PFOA, PFOS; OCs: PCB153, p,p’-DDE, HCB, t-NC and β-HCH) and offspring i) sex-and-age-specific z-scores for BMI at 5-year follow-up, and ii) sex-and-age-specific triceps and subscapular skinfolds at 5-year follow-up. We used multivariable logistic regression to estimate adjusted odds ratios (ORs) and 95% CIs for the association between maternal serum POP concentrations and child overweight/obesity (BMI z-scores ≥ 85th percentile for age and sex) at 5-year follow-up. We constructed a directed acyclic graph (DAG) to assess and select potential confounders (Additional file 1: Fig. S1). Prenatal growth was considered a mediator in the pathway between exposure to POPs and childhood overweight, due to positive associations between increasing prenatal levels and POPs and SGA birth in our study sample [35]. As adjustment for a mediator may introduce collider bias if there are shared unmeasured causes of both the mediator (SGA status) and the outcome (childhood overweight) [36], we did not include prenatal growth or SGA status in the multivariate analyses. The following variables were included in multivariable analyses as potential confounders: maternal age (continuous; years), maternal pre-pregnancy body mass index (BMI) (continuous: kg/m2), maternal education (categorical: < 9 years, 10–12 years, or ≥ 13 years), maternal smoking status at conception (categorical: 0, 1–9 or ≥ 10 cigarettes per day), previous breastfeeding duration (continuous: months), inter-pregnancy interval between the last two children (categorical: ≤ 18 months, 19–60 months, ≥ 61 months), and maternal weight gain from conception up to gestational week 17 (continuous: kilograms). The pooled analyses were further adjusted by country (Norway or Sweden). All models were tested for normality of residuals, heteroscedasticity, and multi-collinearity.
We examined linearity by scatter plots, assigning maternal serum POP concentrations to the horizontal axis, and measures of child adiposity to the vertical axis. Marginal relationships between maternal serum POP concentrations and offspring BMI z-scores at 5-year follow-up were assessed by non-linear regression using 3-knot restricted cubic splines and 95% CIs. We determined non-linear associations by examination of cubic spline graphs, and by the Wald test.
We had some missing data including 7.2% missing for both maternal weight gain up to gestational week 17 and previous breastfeeding duration. Among children, we had 7.0% missing data on subscapular skinfold thickness and 6.1% missing data on triceps skinfold thickness. Overall, 80% of participants had complete data on all variables. Missing data were assumed missing at random. We used chained multiple imputation [37, 38] to generate and compare five complete data sets. Complete case analyses widened the 95% CIs, but did not change the estimates substantially.
We evaluated possible effect modification by country and offspring sex based on a priori evidence from the literature [12, 35]. We conducted several sensitivity analyses to assess the robustness of our results. First, we did stratum-weighted analyses to ensure generalizability of our reported estimates to the contemporary pregnant population according to the prevalence of i) SGA births, ii) maternal pre-pregnancy overweight, and iii) maternal smoking at conception (See Additional file 1: Supplementary description S1 for details). Such weighted analyses are recommended for analyses with case-control data or in other way unbalanced populations that may be subject to selection bias [39]. Second, we additionally adjusted for maternal fish consumption during pregnancy among Norwegian participants (See Additional file 1: Supplementary description S2 for details). Finally, we considered a multi-pollutant model approach by mutually adjusting for maternal serum POPs that were found to be associated with offspring BMI.
All statistical analyses were conducted with SPSS statistical software, version 22 (IBM SPSS Inc. Chicago, IL) and Stata IC/13.1.