Detecting cord blood cell type-specific epigenetic associations with gestational diabetes mellitus and early childhood growth

Abstract

Background: Epigenome-wide association studies (EWAS) have provided opportunities to understand the role of epigenetic mechanisms in development and pathophysiology of many chronic diseases. However, an important limitation of conventional EWAS is that profiles of epigenetic variability are often obtained in samples of mixed cell types. Here, we aim to assess whether changes in cord blood DNA methylation (DNAm) associated with gestational diabetes mellitus (GDM) exposure and early childhood growth markers occur in a cell type-specific manner.

Results: We analyzed 275 cord blood samples collected at delivery from a prospective pre-birth cohort with genome-wide DNAm profiled by the Illumina MethylationEPIC array. We estimated proportions of seven common cell types in each sample using a cord blood-specific DNAm reference panel. Leveraging a recently developed approach named CellDMC, we performed cell type-specific EWAS to identify CpG loci significantly associated with GDM, or 3-year-old body mass index (BMI) z-score. A total of 1410 CpG loci displayed significant cell type-specific differences in methylation level between 23 GDM cases and 252 controls with a false discovery rate < 0.05. Gene Ontology enrichment analysis indicated that LDL transportation emerged from CpG specifically identified from B-cells DNAm analyses and the mitogen-activated protein kinase pathway emerged from CpG specifically identified from natural killer cells DNAm analyses. In addition, we identified four and six loci associated with 3-year-old BMI z-score that were specific to CD8+ T-cells and monocytes, respectively. By performing genome-wide permutation tests, we validated that most of our detected signals had low false positive rates.

Conclusion: Compared to conventional EWAS adjusting for the effects of cell type heterogeneity, the proposed approach based on cell type-specific EWAS could provide additional biologically meaningful associations between CpG methylation, prenatal maternal GDM or 3-year-old BMI. With careful validation, these findings may provide new insights into the pathogenesis, programming, and consequences of related childhood metabolic dysregulation. Therefore, we propose that cell type-specific analyses are worth cautious explorations.

Keywords: Cell type specificity; DNA methylation; Early childhood growth; Epigenome-wide association study; Gestational diabetes mellitus.

Fig. 1

Manhattan plots summarizing epigenome-wide CpG methylation associated with gestational diabetes mellitus. (a–g) Cell type-specific differentially methylated CpG loci are indicated for seven cell types. Results obtained from standard EWAS adjusting for estimated cell type proportions are summarized in (h). CpG loci are aligned on the x-axis according to genomic coordinate and are colored by chromosome. The y-axis represents − log₁₀ (p value). Red dashed lines denote Bonferroni-corrected genome-wide significance threshold (p value < 6.3 $\times$ 10^–8)

Fig. 2

Manhattan plots summarizing epigenome-wide CpG methylation associated with 3-year-old BMI z-score. (a–g) Cell type-specific differentially methylated CpG loci are indicated for seven cell types. Results obtained from standard EWAS adjusting for estimated cell type proportions are summarized in (h). CpG loci are aligned on the x-axis according to genomic coordinate and are colored by chromosome. The y-axis represents − log₁₀ (p value). Red dashed lines denote Bonferroni-corrected genome-wide significance threshold (p value < 6.3 $\times$ 10^–8)

Fig. 3

Quantile–Quantile plots of p values in epigenome-wide permutation tests. Ten permutations for association tests for (a) gestational diabetes mellitus and (b) 3-year-old BMI z-score were performed respectively. Distributions of p values obtained in these permutations are compared to those (red dots) obtained in the original analysis of cell type-specific differential methylation. All significant CpG loci associated with gestational diabetes mellitus (FDR < 0.05) reside on the right side of the curve after inflexion. (c) Association between 3-year-old BMI z-score and methylation level at cg12586150 in SERPINB1. Solid lines indicate predicted effects; Dotted lines delineate 95% confidence intervals. For visualization, predictions were based on median maternal age, non-smoker, no parity, median gestational age, and child being female. Cell type proportions of other six cell types were set to $\frac{1-ProportionofMonocytes}{6}$