Developmentally dynamic changes in DNA methylation in the human pancreas

Abstract

Development of the human pancreas requires the precise temporal control of gene expression via epigenetic mechanisms and the binding of key transcription factors. We quantified genome-wide patterns of DNA methylation in human fetal pancreatic samples from donors aged 6 to 21 post-conception weeks. We found dramatic changes in DNA methylation across pancreas development, with > 21% of sites characterized as developmental differentially methylated positions (dDMPs) including many annotated to genes associated with monogenic diabetes. An analysis of DNA methylation in postnatal pancreas tissue showed that the dramatic temporal changes in DNA methylation occurring in the developing pancreas are largely limited to the prenatal period. Significant differences in DNA methylation were observed between males and females at a number of autosomal sites, with a small proportion of sites showing sex-specific DNA methylation trajectories across pancreas development. Pancreas dDMPs were not distributed equally across the genome and were depleted in regulatory domains characterized by open chromatin and the binding of known pancreatic development transcription factors. Finally, we compared our pancreas dDMPs to previous findings from the human brain, identifying evidence for tissue-specific developmental changes in DNA methylation. This study represents the first systematic exploration of DNA methylation patterns during human fetal pancreas development and confirms the prenatal period as a time of major epigenomic plasticity.

Keywords: Brain; DNA methylation; Development; Fetal; Monogenic diabetes; Neonatal diabetes; Pancreas; Sex differences.

Fig. 1

Characterization of fetal pancreas samples. A An overview of the age and sex distribution of fetal pancreas samples profiled in this study. B Chromogranin A (brown) and Cytokeratin 19 (pink) staining of FFPE tissue from a 17 PCW donor showing both endocrine and ductal tissue. C Immunofluorescent staining analysis of FFPE tissue from a 17 PCW donor confirming the presence of endothelial cells (CD31 + , red), alpha cells (GLU + , green), beta cells (INS + , cyan), and delta cells (SST + , magenta). D Age estimates derived from an epigenetic clock calibrated on fetal tissue samples confirm a strong correlation between actual and estimated gestational age across the 99 fetal pancreas samples included in this study (corr = 0.89, P = 8.31 × 10^–35)

Fig. 2

Changes in DNA methylation associated with development of the human pancreas. A Mean levels of DNA methylation in early fetal pancreas samples (6—8 PCW) and later fetal pancreas samples (19—21 PCW) across all significant dDMPs. B cg08125539, annotated to the IGF2BP1 gene on chromosome 17, was the most significant hypermethylated dDMP across pancreas development (change in DNA methylation proportion per gestational week = 0.049, P = 1.35 × 10^–61). C cg20554008, annotated to the MAP2K3 gene on chromosome 17, was the most significant hypomethylated dDMP across pancreas development (change in DNA methylation proportion per gestational week = -0.036, P = 1.48 × 10^–61). Many dDMPs are colocalized into larger regions of differential DNA methylation (dDMRs) associated with pancreas development. Shown are (D) the top-ranked hypermethylated dDMR located in the BLCAP and NNAT genes on chromosome 20 (regression coefficient = 0.253, P =  < 1.14 × 10^–304) and E) the top-ranked hypomethylated dDMR located in the transcription start site of NWD1 on chromosome 19 (regression coefficient = -0.239, P = P =  < 1.14 × 10^–304)

Fig. 3

Gene comethylation modules associated with development of the human pancreas. Across DNA methylation sites located in promoter regions, the module eigengene of the (A) Promoter4 module was most positively correlated with developmental age and the module eigengene of the (B) Promoter3 module had the strongest negative correlation with developmental age. Shown for each module is the module eigengene for each sample (left panel) and a heatmap depicting the mean DNA methylation for core sites in each module across development (right panel) with colour corresponding to DNA methylation level at sites with a module membership > 0.85. Across DNA methylation sites located in gene body regions, the module eigengene of the (C) GeneBody3 module was was most positively correlated with developmental age and the module eigengene of the (D) GeneBody4 module had the strongest negative correlation with developmental age

Fig. 4

The distribution and direction of fetal pancreas dDMPs differs markedly across genomic features. A Compared to the genome average, dDMPs are significantly underrepresented in CpG islands, transcription start-sites, 5′ UTRs, and first exons, but significantly enriched in CpG island shores, CpG island shelves, and the gene body. Relative enrichment of dDMPs in CpG island features. B The depletion of dDMPs in CpG islands is largely driven by a dramatic paucity of sites becoming hypomethylated across development. Sites becoming hypomethylated across development are shown in blue and sites becoming hypermethylated across development are shown in red. Shown is the relative enrichment of sites in each category based on their P-value for association with developmental age (from 1 × 10^–1, to 1 × 10^–35) determined using a Fisher’s exact test. The genome-wide significance threshold for dDMPs (P < 9 × 10^–8) is denoted using a grey line. C Compared to the genome average, dDMPs are also depleted in regulatory domains defined by key transcription factor binding site occupancy and chromatin accessibility in pancreatic islets from human donors. In contrast to the pattern seen in CpG islands, these regulatory domains are characterized by a dramatic depletion of hypermethylated dDMPs for both regions of (D) transcription factor binding occupancy and (E) open chromatin in pancreatic islet cell-types. F The depletion of dDMPs in domains bound by CTCF, for example, is largely driven by a dramatic paucity of sites becoming hypomethylated across development

Fig. 5

Developmental changes in pancreatic DNA methylation annotated to neonatal diabetes genes. A There is a significant enrichment of sites becoming hypermethylated across pancreas development annotated to neonatal diabetes genes. Sites becoming hypomethylated across development are shown in blue and sites becoming hypermethylated across development are shown in red. Shown is the relative enrichment of sites in each category based on their P-value for association with developmental age (from 1 × 10^–1, to 1 × 10^–35) determined using a Fisher’s exact test. The genome-wide significance threshold for dDMPs (P < 9 × 10^–8) is denoted using a grey line. B cg23425348 shows the most significant developmental change in DNA methylation associated across sites annotated to NKX2-2 (change in DNA methylation proportion per gestational week = 2.83 × 10^–2, P = 8.26 × 10^–29) which is the neonatal gene with the highest proportion of dDMPs and C) contains a discrete region of significant hypermethylation (regression coefficient = 0.195, P = 1.14 × 10^–51). D cg25995955 shows the most significant developmental change in DNA methylation associated across sites annotated to LRBA (change in DNA methylation proportion per gestational week = 3.33 × 10^–2, P = 5.05 × 10^–50). F The LRBA gene is another neonatal diabetes gene with an enrichment of dDMPs clustered into a discrete region characterized by increasing DNA methylation across development (regression coefficient = 0.224, P = 1.14 × 10^–304)

Fig. 6

The dramatic temporal changes in DNA methylation at fetal dDMPs is limited to the prenatal period. A Correlation of DNA methylation changes during fetal development (X-axis) and postnatal aging (Y-axis) at 176,624 dDMPs tested in both fetal and adult pancreas. No pancreas dDMPs were significantly associated with age in the postnatal pancreas. B DNA methylation at pancreas dDMPs was dramatically more variable in fetal pancreas samples than postnatal pancreas samples. Shown is the distribution of variance in DNA methylation across the 10,000 most significantly dDMPs in fetal pancreas samples (pink) (variance = 0.0150) compared to the variance in DNA methylation at the same sites in adult pancreas samples (blue) (0.0013). C DNA methylation at pancreas dDMPs was dramatically different between early fetal (6—8 PCW) and adult pancreas samples at the 10,000 most significant dDMPs. D In contrast, DNA methylation at pancreas dDMPs was highly correlated between early fetal (6—8 PCW) and adult pancreas samples at the same dDMPs (corr = 0.70)

Fig. 7

Overlap between developmental changes in DNA methylation between pancreas and brain. A Correlation of effect sizes at pancreas dDMPs with those at the same sites in the developing brain. B A site at which developmental changes in DNA methylation is highly similar between pancreas and brain. Shown is cg08486065 which is characterized by significant decrease in DNA methylation across development in both pancreas (effect size = 0.0362, P = 1.06 × 10^–54) and brain (effect size = 0.0341, P = 1.38 10^–12). C A site at which DNA methylation changes significantly across development in the pancreas and brain but in the opposite direction. Shown is cg16094352 which is characterized by significant hypermethylation across development in the pancreas (effect size = 0.0151, P = 3.13 × 10^–41) but significant hypomethylation across development in the brain (effect size = -0.00484, P = 1.68 × 10^–8). D A site at which developmental changes in DNA methylation are specific to the pancreas. Shown is cg08219218 annotated to LBRA that is characterized by significant hypermethylation across development in the pancreas (effect size = 0.0490, P = 4.52 × 10^–41) but no change in the brain. E Correlation of effect sizes at brain dDMPs with those at the same sites in the developing pancreas. F A site at which developmental changes in DNA methylation are specific to the brain. Shown is cg02192555 annotated to MSRA that is characterized by significant hypomethylation across development in the brain (effect size = -0.0371, P = 3.86 × 10^–12) but no change in the pancreas