Epigenome-wide association analyses

Two epigenome-wide association analyses were conducted, first with birthweight as a continuous exposure variable and, second, with small for gestational age (SGA) as a binary exposure variable. The latter was defined as the lowest 10^th percentile of the World Health Organization birthweight for sex and gestational age charts [31], with the remaining population above the 10th percentile as the reference group. As in our study population only 7 children (5%) were classified as large for gestational age (based on the 90^th percentile), we had no power to analyse this trait separately and opted for a more conservative approach by including them in the reference group. In both EWAS analyses, we used robust linear regression models adjusted for sex, age at saliva sampling, gestational age, maternal age, parity, maternal pre-pregnancy BMI, maternal education, batch, estimated surrogate variables, and case–control status of the original nested case–control study. p-Values adjusted for multiple comparisons were calculated using the Benjamini and Hochberg false discovery rate (FDR), and Volcano plots were used to visually present the results. In the EWAS of SGA, as a sensitivity analysis we provisionally excluded children born pre-term (< 37 gestational weeks at birth).

To assess whether the age at saliva sampling could have influenced the findings on the top CpG sites identified in the EWAS analyses, we tested the associations of the age at saliva sampling as a continuous variable (in months) with the methylation levels in the top CpG sites using the robust linear regression models adjusted for sex, batch and cell type composition estimated with the reference-based projections for saliva proposed by Zheng [29].

Finally, we performed a look-up of the saliva top findings in the PACE summary results, publicly available via Zenodo: https://zenodo.org/record/2222287#.YDOhCtWSmUk.