#epiroadmap-20 スライド① @n0rr ARTICLE Received 7 Jan 2014 | Accepted 8 Sep 2014 | Published 10 Oct 2014 DOI: 10.1038/ncomms6187 Sexual dimorphism in epigenomic responses of stem cells to extreme fetal growth Fabien Delahaye1, N. Ari Wijetunga2, Hye J. Heo1, Jessica N. Tozour1, Yong Mei Zhao1, John M. Greally2 & Francine H. Einstein1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6187 Extreme fetal growth is associated with increased susceptibility to a range of adult diseases through an unknown mechanism of cellular memory. We tested whether heritable epigenetic IUGR/control LGA/control processes in long-lived CD34 þ haematopoietic stem/progenitor cells showed evidence for Female 200 400 re-programming associated with the extremes of fetal growth. Here we show that both fetal Male growth restriction and over-growth are associated with global shifts towards DNA 150 300 hypermethylation, targeting cis-regulatory elements in proximity to genes involved in glucose homeostasis 100 and stem cell function. We find a sexually dimorphic 200 response; intrauterine growth Density restriction is associated with substantially greater epigenetic dysregulation in males, whereas larg 50 e for gestational age growth predominantly affects females. 100 The findings are consistent with extreme fetal growth interacting with variable fetal susceptibility to influence 0 cellular ag 0 eing and metabolic characteristics through epigenetic mechanisms, potentially 7 7 generating biomarkers that could identify infants at higher risk for chronic disease later in life. 6 6 5 5 value) 4 4 (P 10 3 3 –Log 2 2 1 1 0 0 –60 –40 –20 0 20 40 60 –60 –40 –20 0 20 40 60 Methylation score difference Methylation score difference Figure 2 | Sexual dimorphism in IUGR males and LGA females for differentially methylated loci. The lower panels show volcano plots of DNA methylation score differences, the upper panels quantify the densities of differentially methylated loci (P valueo0.05 using analysis of variance with pairwise two-tailed Tukey 1 Departm -tests, ent of Obstetrics methylation & Gynecology difference and 4 Women’s |20|). ( Health, a) Albert IUGR Einstein Co compar lleg ed e of Medicine with , 1300 controls, Morris (b) P L ark A GA venue, Block compareBuilding, d R with oom 631, controls. Bronx, New York 10461, USA. 2 Department of Genetics, Albert Einstein College of Medicine, 1301 Morris Park Avenue, Price Building, Room 322, Bronx, New York 10461, USA. Correspondence and requests for materials should be addressed to J.M.G. (email: firstname.lastname@example.org) or to F.H.E. (email: email@example.com). NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms618 chr9 7 | www.nature.com/natur 137,215,000 1 ecommunications 37,220,000 137,225,000 137,230,000 137,235,000 137,240,000 1 50 & 2014 Macmillan Publishers Limited. All rights reserved. 0 Repressive 1 2 Transcribed control) – 3 0 difference 4 DNA methylation (IUGR Enhancer 5 * –30 50 6 Promoter * f.test 0 <2.2e–16 difference * DNA methylation (LGA – control) –30 50 Significant loci 0 HELP-tagging assay difference DNA methylation (cases – control) –30 Candidate differentially-methylated locus Repressive Transcribed Enhancer Promoter RefSeq genes RXRA HpaII CpG islands Increased methylation in cases Exon Intron Decreased methylation in cases Figure 3 | Candidate differentially methylated loci are enriched at cis-regulatory elements. (a) Based on empirical annotation of promoter, enhancer, repressive and transcribed regions, enrichment of candidate differentially methylated loci (n ¼ 10,043) in cases (IUGR and LGA) compared with controls is illustrated with significance values shown for enriched sequence features. The bar on the left represents the proportional representation of each feature in terms of loci tested by HELP-tagging, whereas the bar on the right shows the proportions of features at which differentially methylated loci are found. Significant enrichment for differential methylation at candidate promoters and enhancers is observed. (b) An example of the RXRA gene with a candidate differentially methylated locus is shown. The DNA methylation score differences between controls and IUGR (top), LGA (middle) and cases (bottom, IUGR and LGA combined) are depicted, with a site identified as being a candidate differentially methylated locus in the CpG island promoter region shown in grey. Blue, positive values represent decreased DNA methylation in the cases of extreme fetal growth; yellow, negative value increased methylation. 4 NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms6187 | www.nature.com/naturecommunications & 2014 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6187 Environmental factors have the potential for significant functionsatthematernal–fetalinterfaceandmaybeapotential impact on normal development and health throughout the mediator of intrauterine environmental conditions17–19. life span. Suboptimal intrauterine conditions represent a However, testing the placenta does not address the latent risk specific type of environmental exposure that is associated with in adulthood of chronic disease, which has to be mediated by increased risk for cardiovascular disease1,2 and premature death somatic cells of the offspring. Furthermore, the use of samples of in adulthood3. A substantial amount of evidence has also mixed populations of cells in DNA methylation studies, such as demonstrated the relationship between poor maternal nutrition those sampled from highly heterogeneous placental tissue, is now or low birth weight with a range of metabolic disorders and recognized as a major source of experimental artefact that limits obesity in humans4,5 with animal studies further corroborating interpretability of results20,21. these findings6. At the opposite end of the spectrum of extreme We focus on haematopoietic stem/progenitor cells (HSPCs), fetal growth, excess nutrition leading to large for gestational age purified using the CD34 surface marker to reduce cell subtype (LGA) birth weights is associated with similar adult phenotypes, effects20. HSPCs include a subset of long-term, self-renewing stem with increased risk for premature mortality3 and a range of other cells that persist through the life of the individual22, allowing the age-related diseases7. Fetal growth restriction and over-growth cellular propagation or the ‘memory’ of exposure to temporally show a decline in resistance to chronic disease in adulthood and remote suboptimal conditions. The role of CD34 þ HSPCs in the involvement of multiple organ systems, which is typical of normal maintenance of vascular integrity23,24 is mechanistically relevant ageing and may represent a precocious ageing phenotype for the adult phenotype associated with increased risk for associated with both extremes of the fetal growth spectrum8. cardiovascular disease4,25. We studied infants born with the two Adverse exposures appear to be particularly consequential in extremes of fetal growth, IUGR and LGA, compared with control early life, possibly due to the rapid expansion of cell populations infants with appropriate weight for gestational age. Owing to the necessary for growth, and the dynamism of cellular differentiation thorough characterization of CD34 þ HSPCs by the Roadmap and lineage commitment that occurs during this period of Epigenomics Program, we were able to exploit the mapping of development. Inherent to the differentiation process is the chromatin constituents to define empirically the cis-regulatory modification of transcriptional regulatory patterns. These include elements, such as promotors and enhancers, specific to this cell epigenetic regulators that are capable of transmitting newly type26 allowing us to interpret changes in DNA methylation at established regulatory marks through cell replication9. otherwise unannotated loci in the genome. Environmentally induced perturbations of the cell’s normal epigenetic regulatory controls may be maintained in long-lived, self-renewing cells, maintained through proliferation and Results resulting in functional consequences later in life. Although Genome-wide DNA methylation profiling. We perform gen- alterations in DNA methylation has been associated with the ome-wide DNA methylation profiling on purified CD34 þ HSPCs cumulative exposures inherent to ageing10, environmental from 60 subjects, 20 in each of three groups defined by appropriate exposures early in life may induce addition dysregulation of the or excessively large or small birth weight and ponderal index for epigenome conferring increased susceptibility for age-related gestational age and sex (Table 1). The HELP-tagging assay is used disease at a younger age. as a survey technique testing B1.8 million loci quantitatively at We11,12 and others13–15 have explored the possibility that non- nucleotide resolution and including relatively CG dinucleotide- random epigenetic changes are associated with intrauterine depleted loci27. This assay generates a methylation score that is growth restriction (IUGR). In studies of disease or phenotype- inversely correlated to DNA methylation level, with a methylation associated epigenetic changes, the choice of cell type generally score of 0 indicating full methylation and 100 indicating complete represents a compromise between accessibility, purity, quantity lack of methylation, based on a normalized ratio between tag IUGR 小さい #and epi mechanistic roadma relevance. p-20 Unpurified ス peripheral ライド⑤ @n blood 0rr counts generated by the methylation-sensitive enzyme HpaII and leukocytes have previously been studied in individuals whose its methylation-insensitive isoschizomer MspI28. Based on quality LGA 大きい mothers were exposed prenatally to famine. Altered DNA control measures (Methods and Supplementary Fig. 1), 993,514 比較区の設定 methylation of multiple sites within the differentially loci are selected for further analyses. Of these, 10,043 loci methylated region of the imprinted Insulin-like growth factor 2 are defined as candidate differentially methylated loci using (IGF2) gene was found in subjects decades later13. Cord blood batch-adjusted significance and degree of methylation difference leukocytes have also been used to demonstrate associations of thresholds in comparisons of IUGR and LGA infants with the DNA methylation with in utero conditions and birth weight15,16. normal birth weight controls. We observe a global relative Another commonly studied tissue type is the placenta, which shift towards DNA hypermethylation in CD34 þ HSPCs in both Table 1 | Clinical cohort characteristics. Cohort 1 Cohort 2 Control (n ¼ 20) IUGR (n ¼ 20) LGA (n ¼ 20) Control (n ¼ 8) IUGR (n ¼ 8) LGA (n ¼ 8) Gestational age, weeks 39.3±0.3 38.7±0.4 39.2±0.2 39.1±0.5 38.7±0.4 39.8±0.3 Birth weight, g 3,150±64 2,515±69* 3,996 ±81* 3,150±161 2,498±78w 4,053±67* Ponderal index, g cm À 3 2.8±0.03 2.3±0.04* 3.2±0.03* 2.8±0.07 2.3±0.04* 3.4±0.07* % Male 50 50 50 50 50 50 Maternal age, years 24.6±0.9 26.5±1.3 31.1±1.1* 33.1±1.7 27.7±2.6w 31.5±1.6 Pre-pregnancy BMI, kg m À 2 26.6±1.3 25.5±0.9 29.0±1.3 27.0±2.1 26.5±2.8 28.5±1.4 Weight gain, pounds 29.4±2.7 23.0±2.7 29.9±2.1 27.0±5.1 12.7±5.1 24.8±2.9 BMI, body mass index; IUGR, intrauterine growth restriction; LGA, large for gestational age. Showing mean±standard deviation values. *Po0.001 compared with Control. wPo0.05 compared with Control. 2 NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms6187 | www.nature.com/naturecommunications Cohort 1 Cohort 2 & 2014 Macmillan Publishers Limited. All rights reserved. ゲノムワイドな解析に使う Cohort 1 の検証に使う HELP-tagging assay MassArray IUGR 小さい児 バイサルファイト-seq LGA 大きい児
#epiroadmap-20 スライド⑥ @n0rr 今回調べた細胞 wikipedia より 臍帯血由来の造血幹細胞
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6187 ARTICLE IUGR and LGA subjects when compared with the controls functional consequences of these candidate differentially methy- (Fig. 1a). The clustering of cases (LGA/IUGR) is not uniform, lated loci, we take advantage of the mapping of chromatin with a subset of cases clustering with controls (Fig. 1b), indicating components in CD34 þ HSPCs performed as part of the Road- that epigenetic dysregulation does not occur universally as a map Epigenomics Program. The details of this annotation are response to extreme fetal growth. Although there exists a described in a separate report26 and involve the use of the Segway subset of common loci altered in both IUGR and LGA algorithm29 to generate genomic features (Methods) that are then neonates, most of the dysregulated loci are distinctive between interpreted using Self-Organizing Maps30. We are thus able to these groups (Fig. 1c,d). We also see an overlap of genes define candidate promoters, enhancers, transcribed sequences (as opposed to loci) undergoing differential DNA methylation and repressive chromatin in the epigenome specific to the (Supplementary Fig. 2). CD34 þ HSPC population. Every HpaII site is then assigned to a candidate feature based on its genomic position. The HELP- Sexual dimorphism associated with the extremes of fetal tagging assay represents each of the candidate genomic features growth. Sex-specific comparisons for DNA methylation patterns (based on 993,514 loci) and the candidate differentially are shown between control and IUGR and LGA subjects (Fig. 2). methylated loci (10,043) are significantly enriched in Segway Both IUGR males and females show a shift in DNA methylation features 4 (enhancers, Po0.001) and 6 (promoters, Po0.001), profiles compared with controls, but the number of hyper- indicating preferential targeting to transcriptional regulatory methylated loci is markedly higher in males compared with elements (Fig. 3a). We show an example of the mapping of the females (Fig. 2a). Sex-specific differences are also seen in the one the candidate differentially methylated loci, to the promoter comparison of LGA to controls, with LGA females showing an of the Retinoid X receptor, alpha (RXRA) gene, at an annotated increase in the overall number of candidate differentially CpG island, and within the Segway feature 6 annotation methylated loci compared with males (Fig. 2b). These findings indicating candidate promoter function. The HELP-tagging indicate a sexual dimorphism in the epigenetic responses of derived methylation scores for cases (IUGR and LGA HSPCs to the extremes of growth conditions in utero. combined) are compared with controls to demonstrate the #epiroadmap-20 スライド⑬ @n0rr magnitude of the change at this locus (Fig. 3b). Targeting of DNA methylation changes to specific genomic contexts. Although the consequences of DNA methylation Targeting of DNA methylation changes to genes with specific 大きい/小さい児はメチル化されてた changes at recognized promoter sequences are generally pre- properties. We test whether the subset of loci affected by DNA dictable, a genome-wide study of this type can generate a majority methylation changes are enriched at a specific subset of genes of findings in un-annotated genomic locations. To predict the characterized by concordance of function of their protein a b IUGR 小さい Hypermethylation Hypomethylation Control LGA 大きい IUGR 0.04 Control LGA 0.02 Density 0 0.04 IUGR 0.02 Density 0 0.04 LGA 0.02 Density 0 0 50 100 Methylation score c d IUGR/control IUGR/control LGA/control IUGR/LGA 7 LGA/control 980 2,699 5 4,645 223 5,560 588 1,051 1,571 value) (P 3 815 1,189 618 10 1 4,353 IUGR/LGA –Log –50 0 50 –50 0 50 –50 0 50 Methylation score Methylation score Methylation score difference difference difference Figure 1 | Genome-wide DNA methylation profiles. (a) Density plots of methylation scores for IUGR or LGA compared with controls. The distributions of DNA methylation scores are shown in red. (b) A self-organizing heatmap of candidate differentially methylated loci showing clustering by sample. (c) Volcano plots of DNA methylation score differences for IUGR compared with control, LGA compared with control and IUGR compared with LGA, based on 993,514 loci throughout the genome. Differentially methylated loci with P value o0.05 and methylation difference 4|20| are shown in black. (d) Differentially methylated loci meeting threshold criteria are quantified in a proportional Venn diagram for each comparison. NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms6187 | www.nature.com/naturecommunications 3 & 2014 Macmillan Publishers Limited. All rights reserved.
IUGR 小さい #epiroadmap-20 スライド⑭ @n0rr LGA 大きい ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6187 性的２型 IUGR/control LGA/control Female 200 400 Male 150 300 100 200 Density 50 100 0 0 7 7 6 6 5 5 value) 4 4 (P 10 3 3 –Log 2 2 1 1 0 0 –60 –40 –20 0 20 40 60 –60 –40 –20 0 20 40 60 Methylation score difference Methylation score difference Figure 2 | Sexual dimorphism in IUGR males and LGA females for differentially methylated loci. The lower panels show volcano plots of DNA methylation score differences, the upper panels quantify the densities of differentially methylated loci (P valueo0.05 using analysis of variance with pairwise two-tailed Tukey-tests, methylation difference 4|20|). (a) IUGR compared with controls, (b) LGA compared with controls. chr9 137,215,000 137,220,000 137,225,000 137,230,000 137,235,000 137,240,000 0 Repressive 50 1 2 Transcribed control) – 3 0 difference 4 DNA methylation (IUGR Enhancer 5 * –30 50 6 Promoter * f.test 0 <2.2e–16 difference * DNA methylation (LGA – control) –30 50 Significant loci 0 HELP-tagging assay difference DNA methylation (cases – control) –30 Candidate differentially-methylated locus Repressive Transcribed Enhancer Promoter RefSeq genes RXRA HpaII CpG islands Increased methylation in cases Exon Intron Decreased methylation in cases Figure 3 | Candidate differentially methylated loci are enriched at cis-regulatory elements. (a) Based on empirical annotation of promoter, enhancer, repressive and transcribed regions, enrichment of candidate differentially methylated loci (n ¼ 10,043) in cases (IUGR and LGA) compared with controls is illustrated with significance values shown for enriched sequence features. The bar on the left represents the proportional representation of each feature in terms of loci tested by HELP-tagging, whereas the bar on the right shows the proportions of features at which differentially methylated loci are found. Significant enrichment for differential methylation at candidate promoters and enhancers is observed. (b) An example of the RXRA gene with a candidate differentially methylated locus is shown. The DNA methylation score differences between controls and IUGR (top), LGA (middle) and cases (bottom, IUGR and LGA combined) are depicted, with a site identified as being a candidate differentially methylated locus in the CpG island promoter region shown in grey. Blue, positive values represent decreased DNA methylation in the cases of extreme fetal growth; yellow, negative value increased methylation. 4 NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms6187 | www.nature.com/naturecommunications & 2014 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6187 ARTICLE products. A candidate differentially methylated locus is linked to respectively, even though the loci targeted for DNA methylation a specific gene if the site is (i) located in proximity to the changes are not necessarily the same in each group. transcription start site of the RefSeq gene and (ii) overlapping candidate regulatory loci (features 4 or 6). We select only those candidate promoters (feature 6) within ±2 kb and candidate Verification and validation. To test the robustness of our enhancers (feature 4) within ±5 kb of RefSeq transcription start genome-wide technique, we assess the reproducibility of DNA sites. Although enhancers can act over substantially longer dis- methylation differences at our candidate differentially methylated tances than 5 kb (ref. 31), we are deliberately conservative in loci using single-locus quantitative validation studies. We first restricting the distance so that we would be more likely to perform verification studies on samples from Cohort 1, on whom associate an enhancer with the gene upon which it exerts its the genome-wide studies had been performed, testing four loci effects. The resulting list of genes is used to perform a gene set selected for differing levels of DNA methylation in 24 subjects. A enrichment analysis (GSEA). Traditional GSEA does not take strong correlation between bisulphite MassArray and HELP-tag- into account the physical characteristics of the gene and has ging is found (R2 ¼ 0.98, Supplementary Fig. 4). In a second, been shown to be biased by factors such as the numbers of CG independent set of CD34 þ HSPC samples (Cohort 2) consisting of dinucleotide sites associated with different classes of gene and eight new subjects per group (control, IUGR, LGA) with equal gene promoters32. To address this, the Bioconductor package numbers of males and females in each group, we perform a tar- GoSeq33 was developed to control for variability of length of geted bisulphite sequencing (TBS) assay, using bisulphite treat- genes. We adapted GoSeq to normalize our data to control for ment, targeted PCR and massively parallel sequencing to measure the number of CG dinucleotides linked to each gene by the DNA methylation at 72 loci in the 24 subjects (see Methods and above criteria. Detailed information describing the results of the Supplementary Table 3). The correlations between HELP-tagging normalized GSEA is shown in Supplementary Tables 1 and 2. with MassArray in Cohort 1 (R2 ¼ 0.98, Supplementary Fig. 4) Among the different significant pathways from KEGG (Kyoto and with TBS in Cohort 2 (R2 ¼ 0.72, Supplementary Fig. 5 and Encyclopedia of Genes and Genomes), two pathways of interest Supplementary Table 3) are both strong. These highly quantitative emerge as significant regardless of group comparison: the KEGG verification and validation studies demonstrate the technical pathways for Maturity onset diabetes of the young, relevant to robustness of the genome-wide HELP-tagging assay, as well as the glucose homeostasis and Hedgehog (HH) signalling. Both of potential to validate DNA methylation differences, even when these pathways contain genes involved in proliferation, using a new cohort of subjects. Of the 54 candidate differentially differentiation and self-renewal capabilities of stem cells. methylated loci from the HELP-tagging group comparisons, we Permutation analysis was performed to confirm the focus on loci implicated by our GOseq-normalized GSEA results, significance of these results. Based on the criteria for assigning using primers for candidate differentially methylated loci proximal HpaII sites to RefSeq genes described above, the HELP-tagging to WNT6 (Fig. 5a) and PTCH1 (Fig. 5b) from the HH signalling assay represents 97.6% of RefSeq genes, so we randomly select pathway and MAFA from the Maturity onset diabetes of the young from within this group of genes the same number of genes used pathway (Supplementary Fig. 6). We find the direction of DNA to define our pathways, and perform the GSEA analysis 1,000 methylation changes to be concordant between genome-wide and IUGR 小さい #epiroadmap-20 スライド⑮ @n0rr and 3,000 times to test how frequently the same pathways are targeted assays for all three loci, with statistically significant dif- identified as, defining the significance of our detection of these ferences demonstrable for TBS data from WNT6 (P ¼ 0.023) and LGA 大きい pathways as Po10 À 3. The same pathways are targeted by IUGR PTCH1 (P ¼ 0.014; Supplementary Table 4). We also show the 機能について and LGA even when the loci involved are not identical (Fig. 4). PTCH1 and WNT6 genes to have increased DNA methylation by A similar effect is seen for the loci affected differentially between TBS at local cis-regulatory elements in cases (IUGR and LGA) males and females (Supplementary Fig. 3). These findings compared with controls (Supplementary Table 4, Po0.05). Finally, combine to show convergence of dysregulation of the same we interrogate loci associated with genes that are differentially pathways by IUGR and LGA and in male and female subjects, methylated on average between cases (IUGR plus LGA) and KEGG: maturity onset diabetes of the young KEGG: HH (Hedgehog) signaling pathway WNT family PTCH1 MAFA HH homologues PDX1 Gene associated with LGA Gene associated with IUGR Gene associated with IUGR and LGA Figure 4 | Network analysis. A network representation of KEGG pathways for (a) Maturity onset diabetes of the young and (b) Hedgehog (HH) signalling. Nodes are colour- and size-coded based on the association of genes represented by each node with LGA or IUGR, or with both LGA and IUGR. Edges (solid lines) represent known physical interaction between genes. NATURE COMMUNICATIONS | 5:5187 | DOI: 10.1038/ncomms6187 | www.nature.com/naturecommunications 5 & 2014 Macmillan Publishers Limited. All rights reserved.