UBC Faculty Research and Publications

Choice of surrogate tissue influences neonatal EWAS findings Lin, Xinyi; Teh, Ai L; Chen, Li; Lim, Ives Y; Tan, Pei F; MacIsaac, Julia L; Morin, Alexander M; Yap, Fabian; Tan, Kok H; Saw, Seang M; Lee, Yung S; Holbrook, Joanna D; Godfrey, Keith M; Meaney, Michael J; Kobor, Michael S; Chong, Yap S; Gluckman, Peter D; Karnani, Neerja Dec 5, 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12916_2017_Article_970.pdf [ 1.41MB ]
JSON: 52383-1.0362002.json
JSON-LD: 52383-1.0362002-ld.json
RDF/XML (Pretty): 52383-1.0362002-rdf.xml
RDF/JSON: 52383-1.0362002-rdf.json
Turtle: 52383-1.0362002-turtle.txt
N-Triples: 52383-1.0362002-rdf-ntriples.txt
Original Record: 52383-1.0362002-source.json
Full Text

Full Text

TECHNICAL ADVANCE Open AccessChoice of surrogate tissue influencesneonatal EWAS findingsXinyi Lin1,2†, Ai Ling Teh1†, Li Chen1, Ives Yubin Lim1, Pei Fang Tan1, Julia L. MacIsaac3, Alexander M. Morin3,Fabian Yap4, Kok Hian Tan4, Seang Mei Saw2,5,6, Yung Seng Lee1,7,8, Joanna D. Holbrook1,9, Keith M. Godfrey10,Michael J. Meaney1,11, Michael S. Kobor3, Yap Seng Chong1,12, Peter D. Gluckman1,13 and Neerja Karnani1,14*AbstractBackground: Epigenomes are tissue specific and thus the choice of surrogate tissue can play a critical role ininterpreting neonatal epigenome-wide association studies (EWAS) and in their extrapolation to target tissue. Todevelop a better understanding of the link between tissue specificity and neonatal EWAS, and the contributions ofgenotype and prenatal factors, we compared genome-wide DNA methylation of cord tissue and cord blood, two ofthe most accessible surrogate tissues at birth.Methods: In 295 neonates, DNA methylation was profiled using Infinium HumanMethylation450 beadchip arrays. Sitesof inter-individual variability in DNA methylation were mapped and compared across the two surrogate tissues at birth,i.e., cord tissue and cord blood. To ascertain the similarity to target tissues, DNA methylation profiles of surrogatetissues were compared to 25 primary tissues/cell types mapped under the Epigenome Roadmap project. Tissue-specific influences of genotype on the variable CpGs were also analyzed. Finally, to interrogate the impact of thein utero environment, EWAS on 45 prenatal factors were performed and compared across the surrogate tissues.Results: Neonatal EWAS results were tissue specific. In comparison to cord blood, cord tissue showed higher inter-individual variability in the epigenome, with a lower proportion of CpGs influenced by genotype. Both neonatal tissueswere good surrogates for target tissues of mesodermal origin. They also showed distinct phenotypic associations, witheffect sizes of the overlapping CpGs being in the same order of magnitude.Conclusions: The inter-relationship between genetics, prenatal factors and epigenetics is tissue specific, and requirescareful consideration in designing and interpreting future neonatal EWAS.Trial registration: This birth cohort is a prospective observational study, designed to study the developmental originsof health and disease, and was retrospectively registered on 1 July 2010 under the identifier NCT01174875.Keywords: Epigenome-wide association study, Tissue-specificity, DNA methylation, Prenatal factors, Genotype, NeonateBackgroundEpigenetic processes, such as DNA methylation, areimportant regulators of gene expression and thus play avital role in human development and disease. Inter-individual variation in infant DNA methylomes can arisefrom genetic [1, 2], environmental [3], or stochasticperturbations [4]. Epigenome-wide association studies(EWAS) using neonate tissues can help interrogate theinter-relationship between these factors and enhance ourunderstanding of the biological mechanisms underpin-ning disease predisposition and progression. Further,they are also instrumental in identifying diagnostic andprognostic biomarkers.Epigenomes are tissue specific [5] and thus the choiceof neonatal tissue is an important consideration in de-signing a neonatal EWAS. However, the target tissues ofdirect relevance to the outcome of interest are often im-possible or extremely difficult to collect. As an alternateapproach, surrogate tissues, such as cord blood, cord tis-sue, placenta, or buccal epithelium, are used as proxies* Correspondence: neerja_karnani@sics.a-star.edu.sg†Equal contributors1Singapore Institute for Clinical Sciences, A*STAR, Singapore 117609,Singapore14Department of Biochemistry, Yong Loo Lin School of Medicine, NationalUniversity of Singapore, Singapore 119228, SingaporeFull list of author information is available at the end of the article© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.Lin et al. BMC Medicine  (2017) 15:211 DOI 10.1186/s12916-017-0970-xfor target tissues. A number of studies have comparedDNA methylation markers across different surrogateneonatal tissues. For example, in a twin-study, Gordonet al. [4] reported the influence of genetic factors on asubset of variable CpGs to be higher in cord blood thanplacenta or human umbilical vein endothelial cells(HUVEC). Armstrong et al. [6] compared the DNAmethylation status of seven candidate gene loci and re-peat sequences (LINE-1 and ALUYb7) in the genomeacross three infant tissues (cord blood, placenta, andearly infancy buccal epithelium), and reported tissue-specific differences in DNA methylation levels for mostof the tested loci. Previous studies have also observedless concordance in the DNA methylation–prenatal fac-tor associations performed on different infant tissues.For example, Lesseur et al. [7] reported significant asso-ciations between leptin DNA methylation and geneticvariation, weight-for-gestational-age, maternal adiposity,and maternal smoking in cord blood, but not in pla-centa. Similarly, Novakovic et al. [8] found associationsbetween intrauterine cigarette smoke exposure and arylhydrocarbon receptor repressor DNA methylation inboth cord blood and infant blood at 18 months, but notin placenta and buccal epithelium at birth. Likewise,Nomura et al. [9] reported links between maternal gesta-tional diabetes, preeclampsia, and obesity with globalDNA methylation levels in placenta but not in cordblood. Contrary to these tissue-specific DNA methyla-tion–prenatal factor findings, Ruchat et al. [10] found agreater than 25% overlap in the genes differentiallymethylated in response to maternal gestational diabetesin placenta and cord blood.Even though previous comparisons have advanced ourunderstanding of the relationship of neonate DNAmethylomes with genetic variation, prenatal exposureperturbations, and tissue specificity, there are still out-standing questions that remain unanswered in contextof a neonatal EWAS. First, previous DNA methylation–prenatal factor investigations thus far have been limitedin statistical power and coverage of the epigenome asthey have been conducted on small sample sizes (N < 100)and/or have investigated only a few candidate genes orrepeat regions. It remains unclear how genome-widesite-specific DNA methylation from different neonatetissues would compare in larger sample sizes. Second,previous comparisons have mostly scrutinized DNAmethylation profiles from cord blood, buccal epithe-lium, and placenta, while there have been fewer reportsfrom cord tissue. Additionally, there is limited data onthe utility of cord tissue versus cord blood as a surro-gate tissue in a neonate EWAS. Third, previous reportshave largely restricted their comparisons to either a fewprenatal factors or just genetic variation. These single-faceted investigations provide a useful yet incompletepicture of the complex associations between genetics,prenatal factors, and epigenetics across different neo-natal tissues.To address the limitations of previous studies, weprovide a large-sample epigenome-wide comparison ofgenome-wide DNA methylation from two neonatal tis-sues, and their association with genetic and prenatal fac-tor influences. To accomplish this, we first measuredand compared the inter-individual variation in DNAmethylation of the two neonatal tissues. Second, by com-paring the DNA methylation profiles of these surrogatetissues with the DNA methylation profiles of differentfetal and adult tissues mapped under the EpigenomeRoadmap project, we determined the target tissues thatthese surrogate tissues can proxy for. Third, we investi-gated the extent to which inter-individual variation inthese tissues can be explained by genetic factors. Finally,we also examined the extent to which inter-individualvariation in these surrogate tissues can be explained byprenatal factors by comparing the neonate EWAS resultsfrom prenatal factors.MethodsStudy populationMother–offspring dyads were prospectively recruited aspart of the Growing Up in Singapore Towards HealthyOutcomes (GUSTO) birth cohort study, which has beenpreviously described [11]. Pregnant women in their firsttrimester of pregnancy and of at least 18 years of age wererecruited from the two major public hospitals with obstet-ric services in Singapore, namely the KK Women’s andChildren’s Hospital (KKH) and the National UniversityHospital (NUH). To be eligible, participants had to holdSingapore citizenship or permanent residency, or intent toreside in Singapore for the next 5 years, were of Chinese,Malay or Indian ethnicity, had homogeneous parentalethnic background, and had the intention to deliver ateither NUH or KKH. Women with significant health con-ditions such as those who were on chemotherapy or psy-chotropic drugs were excluded from the study. Thepresent analysis was restricted to live singleton full-termbirths, with an Apgar score of at least nine, and with in-fant genotype and DNA methylation data (cord tissue andcord blood) (Additional file 1: Figure A1). Gestational age(GA) was determined by ultrasonography in the first tri-mester. Child sex was extracted from the medical records.Prenatal factors – demographics, maternal smoking, andalcohol useAt enrolment, interviewer-administered questionnaireswere used to collect information on maternal age andeducation. An interviewer-administered questionnairewas also conducted at 26–28 weeks’ gestation to obtaininformation on maternal occupational activity duringLin et al. BMC Medicine  (2017) 15:211 Page 2 of 13pregnancy, maternal alcohol usage before and duringpregnancy, and maternal smoking behavior before andduring pregnancy. Parity (birth order) was extractedfrom medical records.Prenatal factors – maternal moodThe Spielberger State-Trait Anxiety Inventory (STAI)scale and the Edinburgh Postnatal Depression Scale(EPDS) were used to assess maternal anxiety and depres-sion, respectively, at 26–28 weeks’ gestation. The STAIinstrument contains 40 items scored on a 4-point Likertscale, with 20 items ascertaining the trait measure and20 items ascertaining the state measure. The trait meas-ure is a reflection of a more stable personality character-istic, such as an anxious personality, while the statemeasure is a reflection of transient characteristics ofanxiety such as anxiety disorders. The EPDS instrumentassesses 21 common depressive symptoms experiencedover the past week.Prenatal factors – maternal metabolic/anthropometryPre-pregnancy weight was self-reported during studyrecruitment in the first trimester of pregnancy. Maternalheight and weight were measured at 26–28 weeks’ gesta-tion. Gestational weight gain (GWG) was calculated asthe difference between pre-pregnancy and 26–28 weekweights. Maternal pre-pregnancy BMI (ppBMI) was de-rived as pre-pregnancy weight divided by height squared.Maternal glucose levels (2-h post-glucose (2-h PG) andfasting plasma glucose (FPG)) were ascertained at 26–28weeks using an oral glucose tolerance test of 75 g afteran overnight fast (8–14 h). Maternal peripheral systolicblood pressure (SBP) and diastolic blood pressure(DBP) at 26–28 weeks’ gestation were measured fromthe brachial artery at 30- to 60-second intervals.Prenatal factors – maternal fatty acids and vitaminsMaternal plasma fatty acids were measured using serumdrawn at 26–28 weeks’ gestation. The fatty acids wereexpressed as percentage contribution to total plasmaphosphatidylcholine fatty acid. We investigated the totaln-6 polyunsaturated fatty acids (PUFAs), the total n-3PUFAs, the total PUFAs (n-6 PUFAs + n-3 PUFAs), thetotal monounsaturated fatty acids (MUFAs), and thetotal saturated fatty acids (SFAs). We also investigatedindividual saturated fatty acids myristic acid, palmitic acid,and stearic acid; monounsaturated fatty acids oleic acid andgondoic acid; n-3 PUFAs eicosatetraenoic acid (ETA), eicosa-pentaenoic acid (EPA), docosapentaenoic acid (DPA), anddocosahexaenoic acid (DHA); and n-6 PUFAs linoleic acid,dihomo-gamma-linolenic acid (DGLA), and n-6 arachidonicacid (AA). Finally, the n-6:n-3 PUFA ratio, namely AA:DHAratio, AA:EPA ratio, DHA:DPA ratio, and AA:(DHA+EPA)ratio were also assessed. Maternal micronutrient levels,including vitamin D, vitamin B6, vitamin B12, and folate,were tested using serum drawn at 26–28 weeks’ gestation.Tissue collection and processingCord bloodUp to 40 mL of cord blood was collected from infantumbilical cords within 4 h post-delivery, either by dir-ectly dripping into EDTA tubes for normal deliveries,or extracted through a syringe for cords deliveredthrough cesarean section deliveries, then stored inEDTA tubes. Blood samples were then centrifuged at3000 g at 4 °C for 5 min to separate the blood into threedistinct layers – plasma, buffy coat, and erythrocytes.The top plasma layer was then carefully extracted(without disturbing the buffy coat), followed by extrac-tion of the buffy coat layer. The buffy coat was storedat −80 °C. DNA extraction from the buffy coat was per-formed using QIAsymphony DNA kits as per the man-ufacturer’s instructions.Cord tissueAfter the extraction of cord blood, sections of umbilicalcord tissue (~2 cm per section) were collected andcleaned with phosphate buffer saline solution. Each sec-tion was then cut into smaller pieces with a clean scalpeland stored into 2 mL cryovials. The cord samples werethen snap frozen in liquid nitrogen and stored at −80 °Cuntil subsequent DNA extraction. For DNA extraction,frozen umbilical cords were pulverized with a mortarand pestle, weighed, and allowed to equilibrate to roomtemperature before treatment with 10 U/mL hydraluro-nidase enzyme, ensuring that all tissue was submergedin the enzyme solution. Cord samples were then incu-bated at 37 °C for 30 min on a shaker (150 rpm) in anincubator. Then, 250 μL of Tris-NaCl-EDTA-SDS solu-tion was added before the tissue was homogenized sixtimes (10 seconds each cycle) using a Xiril Dispomixhomogenizer. Samples were then pulse spun to pelletthe tissue prior to adding proteinase K, and incubatedovernight at 55 °C. NaCl (250 μL, 5 M) was added andthe contents of the tube were mixed. Samples were cen-trifuged at 3500 g for 20 min and the supernatant trans-ferred to a fresh tube. An equal volume of 100% ethanolwas added to the supernatant with gentle mixing toallow DNA to precipitate. DNA was spooled and trans-ferred to a fresh tube containing 500 μL of water and5 μL of RNase A solution. Samples were then incubatedat 55 °C for 30 min to remove the RNA. The DNA solu-tion was transferred to MaXtract tubes, where an equalvolume of phenol/chloroform was added with gentlemixing, and then centrifuged at 20,000 g for 10 min. Thetop aqueous layer was extracted and the phenol/chloro-form wash step repeated. The final top aqueous layerwas extracted and a 10% volume of 3 M NaAc (pH 5.2)Lin et al. BMC Medicine  (2017) 15:211 Page 3 of 13and a 200% volume of 100% ethanol was added, gentlymixed, and allowed to precipitate DNA for 10 min at−80 °C. This solution was centrifuged at 20,000 g for10 min to pellet down the DNA. The supernatant wasremoved and the DNA pellet was washed with 70% etha-nol, spun down again, and the supernatant removed.The DNA pellet was air dried to the point of translu-cency and re-suspended in 100 μL TE buffer to dissolvethe DNA.DNA methylation data – infant cord tissue, infant cordbloodProfiling and downstream processing of DNA methylationdata from both tissues (infant umbilical cord tissue, infantcord blood) were conducted separately but followed simi-lar procedures. We used the Infinium HumanMethyla-tion450 array, following standard protocol and processedthe data using an in-house quality control procedure [12].Raw DNA methylation beta values were exported fromGenomeStudioTM. Probes with less than three beads foreither the methylated or unmethylated channel or with adetection P value above 0.01 were set to missing. Probeson sex chromosomes were removed. We further retainedprobes that had non-missingness in all samples. Coloradjustment and normalization of Type 1 and 2 probes wasperformed. To assess the presence and impact of technicalvariables, we performed a principal component analysis(PCA) on the raw DNA methylation data and regressedthe principal components against technical variables,including (1) chip-set (8 chips containing 96 samples perchip-set for cord tissue; 15 chips containing 180 samplesper chip-set for cord blood), (2) chip (12 samples perchip), (3) chip position, (4) bisulfite conversion batch (96samples per plate), and (5) DNA extraction batch (cordtissue only). Samples within a chip (12 samples) werenested within a chip-set (96 or 180 samples), but samplesin the same bisulfite conversion plate (96 samples) werenot necessarily nested within a chip-set. The top principalcomponents from the PCA of raw DNA methylation datawere most strongly associated with the chip variable. Wethus used COMBAT [13] to adjust for chip effects. DNAmethylation beta values were first converted to M-valuesbefore applying COMBAT to remove chip effects and theCOMBAT-corrected DNA methylation values were trans-formed back to beta-values. We then conducted anotherPCA on the COMBAT-corrected dataset. Position onchip, bisulfite conversion batch, chip-set (for cord bloodonly), and DNA extraction batch (for cord tissue only)were associated with the top principal components andthese were adjusted for as covariates in all regressionmodels. Because we did not have complete informationfor some of the potential sources, to allow for the possibilityof other technical artifacts besides the ones considered here,we used surrogate variable analysis [14, 15] to estimatesources of batch effects directly from the DNA methylationdata (surrogate variables). The surrogate variables can alsohelp account for cell type composition. We conducted add-itional sensitivity analyses, where we repeated all analysesadjusting for surrogate variables from the surrogate variableanalysis. Finally, cross-hybridizing probes [16, 17], CpGs lo-cated at single nucleotide polymorphisms (SNPs), andCpGs with multi-modal distribution were excluded fromthe analysis. After quality control filtering, 239,560 CpGsthat passed quality control in both datasets were availablefor analysis. For infant cord tissue, cellular proportions forfibroblasts, B-cells, and T-cells were estimated [18] using areference panel (accession number EGAD00010000460)[19], and their principal components were adjusted as co-variates in the regression models. Likewise, for infant cordblood, we used the reference panel reported by Bakulski etal. [20] to obtain estimated cellular proportions in nucleatedred blood cells, granulocytes, monocytes, natural killer cells,B-cells, CD4+ T-cells, and CD8+ T-cells. Their principalcomponents were then adjusted as covariates in all regres-sion models. As we have previously observed that the asso-ciation of cellular proportions with prenatal factors/DNAmethylation could be ethnicity dependent, interaction termsbetween (principal components of) cellular proportionsand ethnicity were included as covariates in all regressionmodels (in addition to their main effects).Genotype dataGenotyping for infant was performed using the IlluminaOmniExpressExome array. Non-autosomal SNPs as wellas SNPs with call rates less than 95% or minor allele fre-quency less than 10% or failed Hardy–Weinberg equilib-rium were excluded. PCA was used to confirm self-reported ethnicity/ancestry. Samples with a call rate lessthan 99%, cryptic relatedness, or sex/ethnic discrepan-cies were excluded. Alleles were expressed at the positivestrand of the human build (hg19). After quality controlfiltering, 487,176 SNPs that passed quality control wereavailable for analysis.Statistical analysisCpG sites that showed inter-individual variation in eachinfant tissueWe first quantified the number of CpGs that showedinter-individual variation in each tissue (infant umbilicalcord tissue, infant cord blood). For each CpG in eachtissue, a CpG was defined to show inter-individualvariation if the DNA methylation range (maximum–minimum, excluding outliers) was greater than 10% andthe DNA methylation 99th percentile–1st percentile wasgreater than 5%. The CpGs were segregated into fourdistinct categories as (1) CpGs which showed inter-individual variation in both tissues, (2) CpGs whichLin et al. BMC Medicine  (2017) 15:211 Page 4 of 13showed inter-individual variation only in infant cordblood, (3) CpGs which showed inter-individual variationonly in infant cord tissue, and (4) CpGs which did notshow inter-individual variation in either tissue. Eachgroup of CpGs was annotated in terms of their genomicfeatures (promoter, 5′-UTR, exon, intron, 3′-UTR, TTS,and intergenic) and CpG content (island, shores, shelves,open seas) using Homer annotatePeaks function (hg19).We also annotated the genomic location of each groupof CpGs in the enhancers predicted by either theEncyclopedia of DNA Elements (ENCODE) consortium[21] or the Functional Annotation of the MammalianGenome (FANTOM) consortium [22]. For predictedenhancers from ENCODE, we used the annotation thatwas included in the Infinium HumanMethylation450manifest file. The FANTOM5-predicted enhancer anno-tation was obtained by using FANTOM5 Phase 1 andPhase 2 data.Hierarchical clusteringTwo sets of hierarchical clustering analyses wereperformed. First, we performed hierarchical clusteringusing DNA methylation data of all the study (GUSTO)samples (295 infant cord tissue samples, 295 infant cordblood samples). The clustering was conducted using allCpGs that passed quality control filtering. The clusteringanalysis confirmed that all 295 infant cord tissue samplesclustered together as did all 295 infant cord blood sam-ples (Additional file 1: Figure B2). Second, we performedhierarchical clustering of the study (GUSTO) sampleswith 25 primary tissues/cells profiled using reducedrepresentation bisulfite sequencing in the EpigenomeRoadmap project [5]. For each of the GUSTO tissues(infant cord tissue, infant cord blood), the median valueacross all 295 samples was used to represent each CpGin each tissue. For DNA methylation data generated bythe Epigenome Roadmap project, we retained only DNAmethylation sites that had a minimum reads coverage of30X and reads from both strands were combined. Thehierarchical clustering was performed using CpG sitesthat passed quality control filtering in the GUSTO tis-sues (infant cord tissue, infant cord blood), were non-missing in at least 10 out of the 25 Epigenome Roadmapsamples, and had interquartile range greater than 10%across different Epigenome Roadmap tissues/cells. Wealso computed the Spearman correlation between eachGUSTO sample/tissue and each Epigenome Roadmaptissue/cell.Genetic influences on DNA methylationWe determined if inter-individual variation in DNAmethylation in each tissue could be explained by geno-type. CpGs whose inter-individual variation in DNAmethylation could be explained by SNPs were defined tobe influenced by genetic factors (SNPs) or genotype-associated factors. We regressed each CpG that showedinter-individual variation in each tissue, against allcis-SNPs (all SNPs that resided on the same chromo-some as the CpG), using an additive genotype model. Tohelp increase the precision of the estimates of effectsizes in assessing the association between genotype andDNA methylation, we adjusted for child sex, GA, ethni-city, cellular proportions, bisulfite conversion batch,hospital, DNA extraction batch (for cord tissue only),chip-set (for cord blood only), and chip position, asthese variables were associated with the top principalcomponents from a PCA of the COMBAT-correctedDNA methylation dataset. We also included interactionsbetween ethnicity and cellular proportions in the regres-sion models. DNA methylation outliers were truncatedto the boundary (next possible) value. For each CpG, wereported the most significant association (smallest Pvalue) between the CpG and cis-SNPs. A CpG was de-fined to be genotype-associated or have its inter-individual variation explained by SNPs if the most sig-nificant association between the CpG and cis-SNPsattained a P value < 5 × 10–8, the Bonferroni thresholdtypically used in genome-wide association studies (corre-sponding to testing for approximately 106 independentSNPs at a family-wise Type 1 error rate of 0.05). For eachtissue, we report the number and percentage of genotype-associated CpGs out of all CpGs that showed inter-individual variation in the tissue. We also report whetherthe CpG was genotype-associated in the other tissue.Prenatal factor influences on DNA methylationWe investigated whether inter-individual variation inDNA methylation in each tissue could be explained byprenatal factors. Linear regression models were used tostudy the association of DNA methylation with each ofthe 45 prenatal factor variables. To help increase theprecision of the estimates of effect sizes in assessing theassociation between prenatal factors and DNA methyla-tion, we adjusted for child sex, GA, ethnicity, cellularproportions, bisulfite conversion batch, hospital, DNAextraction batch (for cord tissue only), chip-set (for cordblood only), and chip position, as these variables wereassociated with the top principal components from aPCA of the COMBAT-corrected DNA methylation data-set. We also included interactions between ethnicity andcellular proportions in the regression models. To ensurethat results were robust to the presence of outliers, out-liers in DNA methylation and continuous prenatal factorvariables were truncated to boundary (next possible)value. We defined a CpG to be influenced by prenatalfactors if the most significant association with the 45prenatal factor variables had a P value < 1 × 10–3(Bonferroni threshold to maintain a family-wise type 1Lin et al. BMC Medicine  (2017) 15:211 Page 5 of 13error rate of 0.05 for testing 45 prenatal factor variables).For each tissue, we report the number and percentage ofCpGs whose inter-individual variation could be ex-plained by prenatal factors out of all CpGs that showedinter-individual variation in the tissue. We also reportwhether the CpG could be explained by prenatal factorsin the other tissue. Finally, we contrasted individualEWAS results across 45 prenatal factors for the twoinfant tissues.ResultsStudy populationThis study used 295 mother–offspring dyads from livesingleton term births, with Apgar score ≥ 9, and avail-ability of genotype and DNA methylation data(Additional file 1: Figure A1). Summary statistics of the295 mother–offspring participants are provided inAdditional file 1: Tables A1, 2, and include 49%, 20%,and 30% of subjects from Chinese, Indian, and Malayethnic groups, respectively. Further, 49% of the neonateswere male. We interrogated DNA methylation profilesderived from infant cord tissue and infant cord bloodusing the Infinium HumanMethylation450 array. Afterquality control filtering, 239,560 CpGs could be used forsubsequent analyses (Additional file 1: Table A3).Infant cord tissue DNA methylation showed more inter-individual variabilityAs the key focus of an EWAS is to examine the inter-individual variation in DNA methylation, we first charac-terized and compared the variable CpGs in the twoinfant tissues (Fig. 1, Additional file 1: Table A4). Of the239,560 CpGs that passed quality control, 20% exhibitedinter-individual variation in both tissues, 21% showedvariation in only one dataset, and the remaining 59% didnot show variation in any dataset (Fig. 1a). The non-variable CpGs were more likely to be located in pro-moter regions and CpG islands and were less likely to bein enhancers, while the variable CpGs were more likelyto be located in open seas and intronic/intergenic re-gions and more likely to be in enhancers (Additional file1: Figures A2–4). Among the tissue-specific CpGs, infantcord tissue had more variable CpGs (18%) (Fig. 1a). Incontrast, infant cord blood was less variable, with onlybacFig. 1 Infant cord tissue showed more inter-individual variation than infant cord blood: proportion of CpGs that showed inter-individual variationand interquartile range (IQR) in DNA methylation. a Pie chart shows the proportion of CpGs for four distinct categories: (1) CpGs which showedinter-individual variation in both tissues, (2) CpGs which showed inter-individual variation only in infant cord blood, (3) CpGs which showed in-ter-individual variation only in infant cord tissue, and (4) CpGs which did not show inter-individual variation in either tissue. A total of 239,560 CpGspassed quality control in both datasets. b Plot of proportion of CpGs (vertical axis) in each tissue (out of 239,560 CpGs) with DNA methylation IQRgreater than or equal to the value specified on the horizontal axis. c Boxplots show the distribution of the DNA methylation IQR, for CpGs in infant cordtissue (bright orange) and infant cord blood (bright blue), respectively, for each of the four categories. Outliers are not shown in the boxplots. A CpGwas defined to show inter-individual variation if the DNA methylation range (maximum–minimum, excluding outliers) was greater than 10% and DNAmethylation 99th percentile–1st percentile was greater than 5%Lin et al. BMC Medicine  (2017) 15:211 Page 6 of 133% of its CpGs exhibiting inter-individual variation spe-cific to cord blood (Fig. 1a). Additionally, as is evidentby the interquartile ranges of the CpGs (Fig. 1b, c), theinter-individual variation was higher in the cord tissuethan in cord blood. To reduce false positives and to in-crease statistical power, CpGs that do not exhibit suffi-cient inter-individual variation are typically excludedfrom EWAS analysis because their observed variabilitycan potentially be attributed to technical variability [23,24]. Thus, a point worth noting from this finding is thatan EWAS conducted using infant cord tissue would havemore CpGs retained for downstream analysis than infantcord blood.Neonatal surrogate tissues primarily proxy for tissues/cells of mesodermal originInfant cord tissue and cord blood are typically used assurrogates for other target tissues [25, 26]. This hasstrong implications for the clinical relevance of the iden-tified epigenetic signatures as phenotypic biomarkers.To evaluate the similarity of these surrogate tissues withprimary tissues, we performed a hierarchical clusteringanalysis of these infant tissues with 25 primary tissues/cells (Additional file 1: Table B1) profiled using reducedrepresentation bisulfite sequencing under the EpigenomeRoadmap project (Fig. 2). These 25 primary tissues/cellscomprised a good representation of tissues/cells derivedfrom the ectoderm (e.g., brain, represented in light pinkin dendrogram), endoderm (e.g., lung, pancreas, digest-ive, represented in light purple), mesenchymal stem cell(MSC)-derived mesoderm (e.g., muscle, heart, kidney,represented in light orange), and hematopoietic stem cell(HSC)-derived mesoderm (e.g., blood, represented inlight turquoise) germinal origins. Consistent with thefindings reported by the Epigenome Roadmap project,tissues/cells generally clustered by their germinal origins(Fig. 2, Additional file 1: Figure B1, Additional file 1:Table B2). Infant cord tissue clustered with MSC-derivedmesodermic tissues and fetal tissues, while infant cordblood clustered with the HSC-derived mesodermic tis-sues (blood).Genotype influences a greater proportion of variableCpGs in infant cord bloodWe assessed the extent to which genetic variation con-tributes to inter-individual variability in DNA methyla-tion levels (Fig. 3, Additional file 1: Table C1). Eachvariable CpG in each infant tissue was regressed againstall cis-SNPs (SNPs that resided on the same chromo-some as the CpG). We found 21% (19,126 CpGs) of the89,871 variable CpGs in infant cord tissue to be associ-ated with genetic variation (with at least one cis-SNP).The corresponding proportion in infant cord blood was31% (17,136 out of 55,810 CpGs), though infant cordtissue still had more genotype-associated CpGs (19,126vs. 17,136) due to more variable CpGs. This finding issupported by a previous twin-study which found thatgenetic factors explained more inter-individual variationin cord blood DNA methylation than in HUVEC DNAmethylation [4]. Of note, HUVEC are one of the celltypes present in cord tissue. The effect sizes for theCpG-SNP associations in both tissues were similar (Add-itional file 1: Figure C1). The results from a sensitivityanalysis where we adjusted for surrogate variables led tosimilar conclusions (Additional file 1: Figure C2), thoughthe percentage of SNP-associated CpGs was slightlyhigher for both tissues (28% for cord tissue and 35% forcord blood).We further examined the overlap in genotype-associated CpGs from the two tissues (Fig. 3b,Additional file 1: Table C1). The overlap in genotype-associated CpGs between infant cord tissue and cordblood was at 41% or 46%, depending on the numberused as denominator. We attempted to replicate thegenotype-associated CpGs with those previouslyreported in cord blood in the Avon Longitudinal Studyof Parents and Child (ALSPAC) cohort by Gaunt et al.[2]. Overall, 54% of the genotype-associated CpGs frominfant cord blood in our cohort could be replicated ininfant cord blood from the ALSPAC cohort (Additionalfile 1: Table C2). The lack of replication for the remain-der of genotype-associated CpGs could be due to ethnicdifferences in the two cohorts (Asian in GUSTO cohortvs. Caucasian in ALSPAC). Smith et al. [1] reported 131and 298 genotype-associated CpGs (out of 20,093 CpGsanalyzed) in African American and Caucasian infantcord blood, respectively, with a similar degree of overlapbetween African American and Caucasian infant cordblood (96 CpGs, 32% or 73% depending on the numberused as the denominator).EWAS associations in the two surrogate tissues weredistinctWe investigated the role of prenatal factors in contribut-ing to the inter-individual variability in DNA methyla-tion levels. For this, we regressed all variable CpGs ineach tissue with 45 prenatal factor variables separately.A list of these 45 variables and their pairwise correlationis shown in Fig. 4a. A CpG was defined to be associatedwith the prenatal factors if it was associated with at least1 of the 45 prenatal factor variables. As an overallcharacterization of the DNA methylation–prenatal factorrelationship, we computed the number/percentage ofvariable CpGs that were associated with the prenatal fac-tors in each tissue (Fig. 4b), and the overlap in prenatalfactor-associated CpGs in the two tissues (Fig. 4c). Theprenatal factors as a whole explained a similar propor-tion (4% of variable CpGs) of inter-individual variationLin et al. BMC Medicine  (2017) 15:211 Page 7 of 13in both tissues (Fig. 4b). The overlap in prenatal factor-associated CpGs in the two tissues was low (Fig. 4b,Additional file 1: Table D1). The DNA methylation–pre-natal factor effect sizes in both tissues were similar(Additional file 1: Figure D1). A subset of these prenatalfactor-associated CpGs also showed association withgenetic variation, at 22% and 32% for infant cord tissueand infant cord blood, respectively (Additional file 1:Table D2). In both the surrogate tissues, CpGs associ-ated or not with prenatal factors showed similar inter-individual variation and a similar distribution of genomicfeatures. Prenatal factor-associated CpGs in both thetissues did not show significant enrichment in any geneontology pathways. Finally, we also contrasted the effectsof individual prenatal factors on individual CpGs in theinfant tissues (Additional file 1: Figures D2, 3). For aCpG associated with a prenatal factor in infant cord tis-sue, we examined if this CpG was also associated withthe same prenatal factor in infant cord blood (Additionalfile 1: Figure D2). We also attempted the reverse ana-lyses (Additional file 1: Figure D3). In general, wenoticed a low concordance in EWAS results from thetwo neonatal tissues. Sensitivity analysis, where we ad-justed for surrogate variables, led to similar conclusions(Additional file 1: Figures D4–6).DiscussionThis study reports a comprehensive analysis of inter-individual variation in genome-wide DNA methylationin two routinely collected surrogate tissues at birth (cordE072 (Brain)E069 (Brain)E074 (Brain)E073 (Brain)E068 (Brain)E051 (Blood)E050 (Blood)E030 (Blood)GUSTO (Cord Blood)E031 (Cord Blood)E035 (Blood)E087 (Pancreas)E110 (Digestive)E101 (Digestive)E077 (Digestive)E102 (Digestive)E081 (Brain, fetal)E083 (Heart, fetal)E086 (Kidney, fetal)E088 (Lung, fetal)E075 (Digestive)E107 (Muscle)E108 (Muscle)GUSTO (Cord Tissue)E103 (Smooth Muscle)E111 (Smooth Muscle)E076 (Smooth Muscle)DNA methylation (%)0%50%100%Fig. 2 Infant cord tissue is a better surrogate for primary tissues of mesenchymal stem cell (MSC)-derived mesodermic germinal origins, while infantcord blood is a better surrogate for primary tissues of hematopoietic stem cell (HSC)-derived mesodermic germinal origins: hierarchical clustering ofGUSTO tissues (cord tissue, cord blood) with 25 primary tissues/cells profiled using reduced representation bisulfite sequencing in the EpigenomeRoadmap project. Infant cord tissue clustered more closely with primary tissues of MSC-derived mesodermic germinal origins, while infant cord bloodclustered more closely with primary tissues of HSC-derived mesodermic germinal origins. Left panel shows heatmap of DNA methylation values, witheach row representing each tissue type and each column representing each CpG. Color changes from yellow to blue as DNA methylation changesfrom 0% to 100%. Right panel of plot shows dendrogram, with tissue types of ectodermic, endodermic, HSC-derived mesodermic, and MSC-derivedmesodermic germinal origins represented in light pink, light purple, light turquoise, and light orange, respectively; GUSTO cord tissue and cord bloodare represented in bright orange and bright blue, respectively. DNA methylation values from GUSTO tissues were generated using Infinium 450 K array(for each CpG and tissue type, the median value across all samples was used). For tissues/cells profiled by the Epigenome Roadmap project, only DNAmethylation sites with a minimum reads coverage of 30X were retained and reads from both strands were combined. Hierarchical clustering wasperformed using only CpG sites that passed quality control filtering in GUSTO tissues, were non-missing in at least 10 out of the 25 EpigenomeRoadmap samples, and had interquartile range greater than 10% across different Epigenome Roadmap tissues/cellsLin et al. BMC Medicine  (2017) 15:211 Page 8 of 13tissue and cord blood). This comparison between infantcord tissue and cord blood highlights the importance ofconsidering tissue specificity in interrogating the rela-tionship between genetic and prenatal factors withepigenetic variation in neonatal tissues. Our findingssuggest that, for a neonatal EWAS conducted using thecord tissue versus the cord blood, there will be (1) morevariable CpGs retained for subsequent phenotype associ-ation analysis, (2) these variable CpGs will be less likelyto be associated with genotype, (3) but equally likely tobe associated with prenatal factors, and finally, (4) cordtissue will serve as a better surrogate for target tissues ofMSC origin.Our tissue-specific findings provide better insights intotissue selection and hypotheses that can be addressed infuture neonatal EWAS. For discovery-based studiesrelated to a phenotype of interest, examining more thanone surrogate tissue can provide a more comprehensiveunderstanding of the underlying biological mechanisms,and the future potential of the surrogate tissues in aclinical setting. EWAS in surrogate tissue can also beused for the purpose of identifying biomarkers, thoughthe identified biomarkers need not always reflect theunderlying biological mechanisms in the primary tissues,and might be surrogate tissue specific. Since it is quitelikely that contrasting results will be obtained whencomparing EWAS findings from different neonatal tis-sues, our findings also imply being careful whenattempting replication analyses, as it will be more repro-ducible when conducted on the same tissue type.a bInfant Cord TissueVariable CpGsYes: SNP-associatedNo: not SNP-associatedInfant Cord BloodFig. 3 SNPs explained a greater proportion of inter-individual variation in DNA methylation in infant cord blood (CB) than in infant cord tissue (CT):SNP-associated CpGs detected in each infant tissue. a Pie charts show the percentage of CpGs in each infant tissue whose inter-individual variationcould be explained by SNPs (out of all CpGs which showed inter-individual variation in the infant tissue). A CpG whose inter-individual variationcould be explained by SNPs (SNP-associated) was defined to be one where the most significant association between the CpG and cis-SNPs (all SNPson the same chromosome as CpG) attained a P value < 5 × 10–8, the commonly used Bonferroni threshold for genome-wide association studies(corresponding to testing for 106 independent SNPs across the genome at a family-wise Type 1 error rate of 0.05). b Overlap between SNP-associated,non-SNP-associated (but variable), and non-variable CpGs in the two tissues. Only CpGs which showed inter-individual variation in at least one tissuewere included (N = 98,124). Examining each tissue separately, each of these 98,124 CpGs can either be SNP-associated, not SNP-associated, or notvariable in each tissue. The number of CpGs in each of these three sets in each tissue is shown in the bottom left bar chart (for each tissue the numberof CpGs from the three sets will sum to 98,124). Collectively, the 98,124 CpGs can be grouped into eight categories. The bottom right panel identifieseach of these eight categories, with the solid black dots representing the sets being considered. For example, the extreme right column identifies thegroup of CpGs that are SNP-associated in both tissues. The top bar chart shows the number of CpGs in each of these eight categories. For example,7822 CpGs were SNP-associated in both tissuesLin et al. BMC Medicine  (2017) 15:211 Page 9 of 13ab cFig. 4 Prenatal factors (PFs) explained a similar proportion of inter-individual variation in infant cord blood (CB) and infant cord tissue (CT): CpGswhere the inter-individual variation in DNA methylation were explained by PFs. a Heatmap shows the pairwise Spearman correlation (absolutevalue) between 45 PFs. Each row/column represents each PF. Color changes from white to blue as correlation changes from zero to one. b Piecharts show the percentage of CpGs in each infant tissue whose inter-individual variation could be explained by PFs (out of all CpGs, whichshowed inter-individual variation in the infant tissue). A CpG whose inter-individual variation could be explained by PFs was defined to be onewhere the most significant association between the CpG and all 45 PFs attained a P value < 1 × 10–3, the Bonferroni threshold for testing 45 PFsat a family-wise Type 1 error rate of 0.05. c Overlap between PF-associated, non-PF-associated (but variable), and non-variable CpGs in the twotissues. Only CpGs which showed inter-individual variation in at least one tissue were included (N = 98,124). Examining each tissue separately,each of these 98,124 CpGs can either be PF-associated, non-PF-associated, or not variable in each tissue. The number of CpGs in each of thesethree sets in each tissue is shown in the bottom left bar chart. Collectively, the 98,124 CpGs can be grouped into eight categories. The bottomright panel identifies each of these eight categories, with the solid black dots representing the sets being considered. The top bar chart showsthe number of CpGs in each of these eight categoriesLin et al. BMC Medicine  (2017) 15:211 Page 10 of 13Likewise, replication studies attempted on different tis-sue types should be carefully interpreted and the caveatsdiscussed accordingly.Findings from this study also provide evidence for theutility of infant cord tissue in a neonatal EWAS. To date,large sample size EWAS (N > 100) have primarily beenattempted on infant cord blood. This study demon-strates that infant cord tissue can capture distinct DNAmethylation signatures and prenatal factor influencesfrom infant cord blood. Additionally, the closer cluster-ing of cord tissue with the MSC-derived mesodermictissues, such as skeletal muscle and smooth muscle, sug-gests that cord tissue is a better surrogate for these pri-mary tissues than cord blood, although this wouldrequire further experimental validation in future studies.This study has some limitations. First, even though weadjusted for cellular heterogeneity using a referencepanel, residual confounding effects could persist. In sucha scenario, DNA methylation–prenatal factor associa-tions will be more susceptible to these effects than theDNA methylation–genotype associations. Developingbetter reference panels will alleviate such limitations.Second, the higher number of variable CpGs in cord tis-sue could arise due to increased diversity of cell types incord tissue. For example, cord tissue probably consistsof a mixture of stromal, endothelial, epithelial, and bloodcontamination [27], while cord blood consists of differ-ent leukocytes. To adequately interrogate this possibility,future studies will require fractionating constituent celltypes of cord tissue and cord blood and comparing theirDNA methylation profiles. Third, we did not profileDNA methylation from placenta or buccal cells at birth,or additional tissues later in the life-course. Examinationof the buccal DNA methylome could be useful at subse-quent stages of child growth as buccal samples are non-invasive and more accessible, thus enabling comparisonof DNA methylation patterns across the life-course.Additionally, studies have reported that buccal cellsmight be a better surrogate for brain tissue than blood[28]. On the other hand, placenta and cord tissue canonly be examined for neonatal EWAS. Further, whilecord blood DNA methylation can be compared to DNAmethylation patterns in blood taken at later stages inlife-course, blood samples are typically not available inearly childhood. As of now, it remains unclear howEWAS findings from cord tissue and cord blood wouldrelate to those from placenta, buccal cells, or tissues de-rived later in the life-course. Future research is necessaryto address this question. Finally, while our study samplesize (N = 295) is larger than most sample sizes used inprevious tissue-specificity investigations (N < 50–100),our study could still be underpowered, especially in theexamination of DNA methylation–prenatal factor associ-ations with small effect sizes. In examining the DNAmethylation–prenatal factor associations, we have used aless conservative threshold of 1 × 10–3. However, ourstudy sample size is comparable to frequently utilizedsample sizes for most of the prenatal factors interrogatedin this study. Thus, our observations from neonateEWAS contributes to the current understanding of pre-natal factor influences on the fetus in utero.ConclusionThere has been a considerable increase in the use ofEWAS analysis to study the developmental origins ofhealth and disease. However, it is becoming increasinglyevident that EWAS studies are more complicated thangenome-wide association studies as epigenetic markersare dynamic, tissue specific, and influenced by geneticand environmental factors. Thus, designing an EWASwarrants multiple considerations to facilitate the identifi-cation of a reliable epigenetic signal, especially from thesurrogate tissues. This study emphasizes that theepigenetic-genetic-prenatal factor relationship is tissuespecific and the choice of neonatal tissues used forEWAS analyses is important to enhance the scope andreplication of the epigenetic findings in future studies.Additional filesAdditional file 1: Supplementary tables and figures. (PDF 2720 kb)Additional file 2: Tab-delimited file containing DNA methylation valuesfor 295 cord tissue samples, 239,560 CpGs. (TXT 615004 kb)Additional file 3: Tab-delimited file containing DNA methylation valuesfor 295 cord blood samples, 239,560 CpGs. (TXT 613902 kb)Additional file 4: Information on using DNA methylation data inAdditional files 2 and 3. (PDF 36 kb)AbbreviationsAA: Arachidonic acid; ALSPAC: Avon longitudinal study of parents and child;DBP: Diastolic blood pressure; DGLA: Dihomo-gamma-linolenic acid;DHA: Docosahexaenoic acid; DPA: Docosapentaenoic acid;ENCODE: Encyclopedia of DNA Elements; EPA: Eicosapentaenoic acid;EPDS: Edinburgh postnatal depression scale; ETA: Eicosatetraenoic acid;EWAS: Epigenome wide association studies; FANTOM: Functional Annotationof the Mammalian genome; FPG: Fasting plasma glucose; GA: Gestationalage; GWG: Gestational weight gain; GUSTO: Growing Up in SingaporeTowards Healthy Outcomes; HSC: Hematopoietic stem cell; HUVEC: Humanumbilical vein endothelial cells; KKH: KK Women’s and Children’s Hospital,Singapore; MSC: Mesenchymal stem cell; MUFA: Monounsaturated fatty acid;NUH: National University Hospital, Singapore; PCA: Principal componentanalysis; PUFA: Polyunsaturated fatty acid; ppBMI: Pre-pregnancy BMI;SBP: Systolic blood pressure; SFA: Saturated fatty acid; SNP: Single NucleotidePolymorphism; STAI: State-trait anxiety inventory; 2-h PG: 2-hr post-glucoseAcknowledgementsThe GUSTO study group includes Pratibha Agarwal, Arijit Biswas, Choon LooiBong, Birit F.P. Broekman, Shirong Cai, Jerry Kok Yen Chan, Yiong Huak Chan,Cornelia Yin Ing Chee, Helen Chen, Yin Bun Cheung, Amutha Chinnadurai,Chai Kiat Chng, Mary Foong-Fong Chong, Yap-Seng Chong, Shang CheeChong, Mei Chien Chua, Doris Fok, Marielle V. Fortier, Peter D. Gluckman,Keith M. Godfrey, Anne Eng Neo Goh, Yam Thiam Daniel Goh, Joshua J.Gooley, Wee Meng Han, Mark Hanson, Christiani Jeyakumar Henry, Joanna D.Holbrook, Chin-Ying Hsu, Neerja Karnani, Jeevesh Kapur, Kenneth Kwek, IvyYee-Man Lau, Bee Wah Lee, Yung Seng Lee, Ngee Lek, Sok Bee Lim, IlianaLin et al. BMC Medicine  (2017) 15:211 Page 11 of 13Magiati, Lourdes Mary Daniel, Michael Meaney, Cheryl Ngo, KrishnamoorthyNiduvaje, Wei Wei Pang, Anqi Qiu, Boon Long Quah, Victor Samuel Rajadurai,Mary Rauff, Salome A. Rebello, Jenny L. Richmond, Anne Rifkin-Graboi,Seang-Mei Saw, Lynette Pei-Chi Shek, Allan Sheppard, Borys Shuter, LeherSingh, Shu-E Soh, Walter Stunkel, Lin Lin Su, Kok Hian Tan, Oon Hoe Teoh,Mya Thway Tint, Hugo P S van Bever, Rob M. van Dam, Inez Bik Yun Wong,P. C. Wong, Fabian Yap, and George Seow Heong Yeo.FundingThis work was supported by the Translational Clinical Research (TCR) FlagshipProgram on Developmental Pathways to Metabolic Disease funded by theNational Research Foundation (NRF) and administered by the NationalMedical Research Council (NMRC), Singapore – NMRC/TCR/004-NUS/2008.Additional funding is provided by Strategic Positioning Fund (SPF) awardedby the Agency for Science, Technology and Research (A*STAR), Singapore.KMG is supported by the National Institute for Health Research through theNIHR Southampton Biomedical Research Centre and by the EuropeanUnion’s Seventh Framework Programme (FP7/2007-2013), projectsEarlyNutrition and ODIN under grant agreement numbers 289346 and613977.Availability of data and materialsDNA methylation datasets used in this study have been included as Additionalfiles 2 and 3. Additional file 4 provides further details on these two datasets. Datarelated to prenatal factors are not publicly available due to ethical restrictionsbut can be obtained from the authors upon reasonable request and subject toappropriate approvals, including from the GUSTO cohort’s Executive Committee.Authors’ contributionsXL contributed to the study design, analysis and interpretation of data,and writing the manuscript. ALT contributed to processing, analysis, andinterpretation of the data. LC contributed to processing of genotyping data. YILand PFT contributed to data analyses and interpretation. JLM, AMM, and MSKcontributed to generation of DNA methylation data. YSC and PDG contributedto cohort data acquisition and critical reading of the manuscript. MJM, FY, KHT,SMS, and YSL contributed to acquisition of phenotypic data. JDH and KMGcontributed to critical reading of the manuscript. NK supervised the study andcontributed to its conception and design, interpretation of the data and writingof the manuscript. All authors read and approved the final manuscript.Ethics approval and consent to participateWritten informed consent was obtained from all women who participated inthe study. Approval for the study was granted by the ethics boards of bothKK Women’s and Children’s Hospital and National University Hospital, whichare the Centralised Institute Review Board and the Domain Specific ReviewBoard, respectively.Consent for publicationNot applicable.Competing interestsYSC and KMG have received reimbursement for speaking at conferencessponsored by companies selling nutritional products. They are part of anacademic consortium that has received research funding from Abbott Nutrition,Nestec, and Danone. The other authors declare no competing interests.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Singapore Institute for Clinical Sciences, A*STAR, Singapore 117609,Singapore. 2Duke NUS Medical School, Singapore 169857, Singapore. 3Centrefor Molecular Medicine and Therapeutics, Child and Family Research Institute,Department of Medical Genetics, University of British Columbia, Vancouver,BC V5Z 4H4, Canada. 4KK Women’s and Children’s Hospital, Singapore229899, Singapore. 5Saw Swee Hock School of Public Health, NationalUniversity of Singapore, Singapore 117597, Singapore. 6Singapore EyeResearch Institute, Singapore 169856, Singapore. 7Department of Paediatrics,Yong Loo Lin School of Medicine, National University of Singapore,Singapore 119228, Singapore. 8Division of Paediatric Endocrinology andDiabetes, Khoo Teck Puat-National University Children’s Medical Institute,National University Health System, Singapore 119228, Singapore. 9NIHRBiomedical Research Centre, University of Southampton, Southampton SO166YD, UK. 10MRC Lifecourse Epidemiology Unit and NIHR SouthamptonBiomedical Research Centre, University of Southampton and UniversityHospital Southampton NHS Foundation Trust, Southampton SO16 6YD, UK.11Ludmer Centre for Neuroinformatics and Mental Health, Douglas UniversityMental Health Institute, McGill University, Montreal, Quebec H4H 1R3,Canada. 12Department of Obstetrics and Gynaecology, Yong Loo Lin Schoolof Medicine, National University of Singapore, Singapore 119228, Singapore.13Centre for Human Evolution, Adaptation and Disease, Liggins Institute,University of Auckland, Auckland 1142, New Zealand. 14Department ofBiochemistry, Yong Loo Lin School of Medicine, National University ofSingapore, Singapore 119228, Singapore.Received: 12 April 2017 Accepted: 31 October 2017References1. Smith AK, Kilaru V, Kocak M, Almli LM, Mercer KB, Ressler KJ, Tylavsky FA,Conneely KN. Methylation quantitative trait loci (meQTLs) are consistentlydetected across ancestry, developmental stage, and tissue type. BMCGenomics. 2014;15:145.2. Gaunt TR, Shihab HA, Hemani G, Min JL, Woodward G, Lyttleton O,Zheng J, Duggirala A, McArdle WL, Ho K, et al. Systematic identificationof genetic influences on methylation across the human life course.Genome Biol. 2016;17:61.3. Green BB, Marsit CJ. Select prenatal environmental exposures andsubsequent alterations of gene-specific and repetitive element DNAmethylation in fetal tissues. Curr Environmen Health Rep. 2015;2:126–36.4. Gordon L, Joo JE, Powell JE, Ollikainen M, Novakovic B, Li X, Andronikos R,Cruickshank MN, Conneely KN, Smith AK, et al. Neonatal DNA methylationprofile in human twins is specified by a complex interplay betweenintrauterine environmental and genetic factors, subject to tissue-specificinfluence. Genome Res. 2012;22:1395–406.5. Roadmap Epigenomics Consortium, Kundaje A, Meuleman W, Ernst J,Bilenky M, Yen A, Heravi-Moussavi A, Kheradpour P, Zhang Z, Wang J,et al. Integrative analysis of 111 reference human epigenomes. Nature.2015;518:317–30.6. Armstrong DA, Lesseur C, Conradt E, Lester BM, Marsit CJ. Global and gene-specific DNA methylation across multiple tissues in early infancy:implications for children’s health research. FASEB J. 2014;28:2088–97.7. Lesseur C, Armstrong DA, Paquette AG, Koestler DC, Padbury JF, Marsit CJ.Tissue-specific Leptin promoter DNA methylation is associated withmaternal and infant perinatal factors. Mol Cell Endocrinol. 2013;381:160–7.8. Novakovic B, Ryan J, Pereira N, Boughton B, Craig JM, Saffery R. Postnatalstability, tissue, and time specific effects of AHRR methylation change inresponse to maternal smoking in pregnancy. Epigenetics. 2014;9:377–86.9. Nomura Y, Lambertini L, Rialdi A, Lee M, Mystal EY, Grabie M, Manaster I,Huynh N, Finik J, Davey M, et al. Global methylation in the placenta andumbilical cord blood from pregnancies with maternal gestational diabetes,preeclampsia, and obesity. Reprod Sci. 2014;21:131–7.10. Ruchat SM, Houde AA, Voisin G, St-Pierre J, Perron P, Baillargeon JP, GaudetD, Hivert MF, Brisson D, Bouchard L. Gestational diabetes mellitusepigenetically affects genes predominantly involved in metabolic diseases.Epigenetics. 2013;8:935–43.11. Soh S-E, Tint MT, Gluckman PD, Godfrey KM, Rifkin-Graboi A, Chan YH,Stünkel W, Holbrook JD, Kwek K, Chong Y-S, et al. Cohort profile: GrowingUp in Singapore Towards healthy Outcomes (GUSTO) birth cohort study. IntJ Epidemiol. 2014;43:1401–9.12. Pan H, Chen L, Dogra S, Teh AL, Tan JH, Lim YI, Lim YC, Jin S, Lee YK, Ng PY,et al. Measuring the methylome in clinical samples: improved processing of theInfinium Human Methylation450 BeadChip Array. Epigenetics. 2012;7:1173–87.13. Johnson WE, Rabinovic A, Li C. Adjusting batch effects in microarrayexpression data using empirical Bayes methods. Biostatistics. 2007;8:118–27.14. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package forremoving batch effects and other unwanted variation in high-throughputexperiments. Bioinformatics. 2012;28:882–3.15. Leek JT, Scharpf RB, Bravo HC, Simcha D, Langmead B, Johnson WE, GemanD, Baggerly K, Irizarry RA. Tackling the widespread and critical impact ofbatch effects in high-throughput data. Nat Rev Genet. 2010;11:733–9.Lin et al. BMC Medicine  (2017) 15:211 Page 12 of 1316. Chen YA, Lemire M, Choufani S, Butcher DT, Grafodatskaya D, Zanke BW,Gallinger S, Hudson TJ, Weksberg R. Discovery of cross-reactive probes andpolymorphic CpGs in the Illumina Infinium HumanMethylation450microarray. Epigenetics. 2013;8:203–9.17. Price ME, Cotton AM, Lam LL, Farre P, Emberly E, Brown CJ, Robinson WP,Kobor MS. Additional annotation enhances potential for biologically-relevant analysis of the Illumina Infinium HumanMethylation450 BeadChiparray. Epigenetics Chromatin. 2013;6:4.18. Houseman EA, Accomando WP, Koestler DC, Christensen BC, Marsit CJ,Nelson HH, Wiencke JK, Kelsey KT. DNA methylation arrays as surrogatemeasures of cell mixture distribution. BMC Bioinf. 2012;13:86.19. Gutierrez-Arcelus M, Lappalainen T, Montgomery SB, Buil A, Ongen H,Yurovsky A, Bryois J, Giger T, Romano L, Planchon A, et al. Passive andactive DNA methylation and the interplay with genetic variation in generegulation. Elife. 2013;2:e00523.20. Bakulski KM, Feinberg JI, Andrews SV, Yang J, Brown S, LM S, Witter F,Walston J, Feinberg AP, Fallin MD. DNA methylation of cord blood celltypes: applications for mixed cell birth studies. Epigenetics. 2016;11:354–62.21. Siggens L, Ekwall K. Epigenetics, chromatin and genome organization:recent advances from the ENCODE project. J Intern Med. 2014;276:201–14.22. Lizio M, Harshbarger J, Shimoji H, Severin J, Kasukawa T, Sahin S, Abugessaisa I,Fukuda S, Hori F, Ishikawa-Kato S, et al. Gateways to the FANTOM5 promoterlevel mammalian expression atlas. Genome Biol. 2015;16:22.23. Meng H, Joyce AR, Adkins DE, Basu P, Jia Y, Li G, Sengupta TK, Zedler BK,Murrelle EL, van den Oord EJ. A statistical method for excluding non-variable CpG sites in high-throughput DNA methylation profiling. BMCBioinf. 2010;11:227.24. Chen J, Just AC, Schwartz J, Hou L, Jafari N, Sun Z, Kocher JP, Baccarelli A,Lin X. CpGFilter: model-based CpG probe filtering with replicates forepigenome-wide association studies. Bioinformatics. 2016;32:469–71.25. Rakyan VK, Down TA, Balding DJ, Beck S. Epigenome-wide associationstudies for common human diseases. Nat Rev Genet. 2011;12:529–41.26. Chadwick LH, Sawa A, Yang IV, Baccarelli A, Breakefield XO, Deng H-W,Dolinoy DC, Fallin MD, Holland NT, Houseman EA, et al. New insights andupdated guidelines for epigenome-wide association studies.Neuroepigenetics. 2015;1:14–9.27. Hayward CJ, Fradette J, Galbraith T, Remy M, Guignard R, Gauvin R, GermainL, Auger FA. Harvesting the potential of the human umbilical cord: isolationand characterisation of four cell types for tissue engineering applications.Cells Tissues Organs. 2013;197:37–54.28. Lowe R, Gemma C, Beyan H, Hawa MI, Bazeos A, Leslie RD, Montpetit A,Rakyan VK, Ramagopalan SV. Buccals are likely to be a more informativesurrogate tissue than blood for epigenome-wide association studies.Epigenetics. 2013;8:445–54.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:Lin et al. BMC Medicine  (2017) 15:211 Page 13 of 13


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items