UBC Faculty Research and Publications

Developmental pathways to adiposity begin before birth and are influenced by genotype, prenatal environment… Lin, Xinyi; Lim, Ives Y; Wu, Yonghui; Teh, Ai L; Chen, Li; Aris, Izzuddin M; Soh, Shu E; Tint, Mya T; MacIsaac, Julia L; Morin, Alexander M; Yap, Fabian; Tan, Kok H; Saw, Seang M; Kobor, Michael S; Meaney, Michael J; Godfrey, Keith M; Chong, Yap S; Holbrook, Joanna D; Lee, Yung S; Gluckman, Peter D; Karnani, Neerja Mar 7, 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_800.pdf [ 4.41MB ]
JSON: 52383-1.0362035.json
JSON-LD: 52383-1.0362035-ld.json
RDF/XML (Pretty): 52383-1.0362035-rdf.xml
RDF/JSON: 52383-1.0362035-rdf.json
Turtle: 52383-1.0362035-turtle.txt
N-Triples: 52383-1.0362035-rdf-ntriples.txt
Original Record: 52383-1.0362035-source.json
Full Text

Full Text

RESEARCH ARTICLE Open AccessDevelopmental pathways to adipositybegin before birth and are influenced bygenotype, prenatal environment andepigenomeXinyi Lin1†, Ives Yubin Lim1,2†, Yonghui Wu1, Ai Ling Teh1, Li Chen1, Izzuddin M. Aris1, Shu E. Soh1,3,Mya Thway Tint2,3, Julia L. MacIsaac4, Alexander M. Morin4, Fabian Yap5, Kok Hian Tan5, Seang Mei Saw6,7,8,Michael S. Kobor4, Michael J. Meaney1,9, Keith M. Godfrey10, Yap Seng Chong1,2, Joanna D. Holbrook1,Yung Seng Lee1,3,11, Peter D. Gluckman1,12, Neerja Karnani1,13* and on behalf of the GUSTO study groupAbstractBackground: Obesity is an escalating health problem worldwide, and hence the causes underlying itsdevelopment are of primary importance to public health. There is growing evidence that suboptimal intrauterineenvironment can perturb the metabolic programing of the growing fetus, thereby increasing the risk of developingobesity in later life. However, the link between early exposures in the womb, genetic susceptibility, and perturbedepigenome on metabolic health is not well understood. In this study, we shed more light on this aspect byperforming a comprehensive analysis on the effects of variation in prenatal environment, neonatal methylome,and genotype on birth weight and adiposity in early childhood.Methods: In a prospective mother-offspring cohort (N = 987), we interrogated the effects of 30 variables thatinfluence the prenatal environment, umbilical cord DNA methylation, and genotype on offspring weight andadiposity, over the period from birth to 48 months. This is an interim analysis on an ongoing cohort study.Results: Eleven of 30 prenatal environments, including maternal adiposity, smoking, blood glucose and plasmaunsaturated fatty acid levels, were associated with birth weight. Polygenic risk scores derived from geneticassociation studies on adult adiposity were also associated with birth weight and child adiposity, indicating anoverlap between the genetic pathways influencing metabolic health in early and later life. Neonatal methylationmarkers from seven gene loci (ANK3, CDKN2B, CACNA1G, IGDCC4, P4HA3, ZNF423 and MIRLET7BHG) were significantlyassociated with birth weight, with a subset of these in genes previously implicated in metabolic pathways inhumans and in animal models. Methylation levels at three of seven birth weight-linked loci showed significantassociation with prenatal environment, but none were affected by polygenic risk score. Six of these birth weight-linked loci continued to show a longitudinal association with offspring size and/or adiposity in early childhood.(Continued on next page)* Correspondence: neerja_karnani@sics.a-star.edu.sg†Equal contributors1Singapore Institute for Clinical Sciences, A*STAR, 30 Medical Drive,Singapore 117609, Singapore13Department 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) 15:50 DOI 10.1186/s12916-017-0800-1(Continued from previous page)Conclusions: This study provides further evidence that developmental pathways to adiposity begin before birthand are influenced by environmental, genetic and epigenetic factors. These pathways can have a lasting effect onoffspring size, adiposity and future metabolic outcomes, and offer new opportunities for risk stratification andprevention of obesity.Clinical Trial Registration: This birth cohort is a prospective observational study, designed to study thedevelopmental origins of health and disease, and was retrospectively registered on 1 July 2010 under theidentifier NCT01174875.Keywords: Epigenome-wide association study, Offspring adiposity, DNA methylation, Prenatal environment,Birth weightBackgroundThe epidemic of obesity is a major public health issue.The risk of obesity appears to begin in utero, as a sub-optimal intrauterine environment can have a lasting im-pact on metabolic control [1–5]. A major mechanism bywhich the effects of an adverse in utero environment ap-pear to be transmitted is by perturbation of the off-spring’s DNA methylome [6–8]. Because the DNAmethylome is susceptible to both genetic [9–11] andenvironmental [12, 13] influences, both factors mayact during development to program pathways to obes-ity [14]. Advances in microarray technology havemade it feasible for DNA methylation to be quantifiedat multiple CpG sites across large samples, and haspaved the way for epigenome-wide association studies(EWAS) [15].Birth weight is often used as a surrogate outcome toevaluate the overall quality of the in utero environment.Firstly, both low and high birth weights have been impli-cated in childhood and adult onset of chronic diseasessuch as obesity, impaired glucose tolerance, type 2diabetes mellitus (T2DM) and coronary artery disease[16]. Secondly, modifiable prenatal environmental factors(themselves being determinants of the intrauterine envir-onment), such as maternal obesity and dietary intake, havebeen linked with birth weight [17]. Thirdly, there isevidence to suggest that birth weight and metabolicdiseases share some common genetic determinants [18].To date, only four epigenome-wide studies havebeen reported that examined the association betweenmethylation marks in neonate tissues and birth weight[19–22], and all these studies have been conductedon Caucasian populations. All four studies focused largelyon the associations between offspring size/adiposity andvariations in the neonate DNA methylome. The onlystudy [22] which included genetic information in theanalysis had a small sample size. Also, Engel et al. [19],Haworth et al. [22] and Simpkin et al. [20] did notconsider the influence of prenatal environments on theidentified associations, while Sharp et al. [21] focusedexclusively on the contribution of maternal adiposity(pre-pregnancy body mass index (ppBMI) and pregnancyweight gain) to the variation in offspring’s methylome.We previously reported that RXRA promoter methyla-tion in umbilical cord DNA correlates with childhoodobesity in replicate cohorts, and that the level of methy-lation is associated with maternal nutrition in the firsttrimester [23]. Using a candidate gene approach, we alsoreported that umbilical cord DNA methylation in thehypoxia inducible factor 3A (HIF3A) gene (a gene previ-ously associated with adult adiposity [24]) associateswith infant weight and adiposity [25]. It follows thatepigenetic alterations associated with the developmentof adiposity may arise during in utero development.These findings, along with previous findings on the influ-ence of SNPs and in utero environment on the epigenome[14], prompted an EWAS to be performed to comprehen-sively search for DNA methylome changes in utero in re-sponse to variations in prenatal environments and genotype.In the current study we take a comprehensive ap-proach to understand the genesis of adiposity in earlylife by interrogating the effects of prenatal environment,genotype and DNA methylation, and we report four im-portant findings. First, we identify prenatal environmentsthat influence birth weight. Second, we report associa-tions between child weight/adiposity (at birth and duringearly childhood) and polygenic risk score derived fromadult adiposity genetic association studies. Third, we findvariations in the neonate DNA methylome that associatewith birth weight and size/adiposity measures in earlychildhood. Last, we determine SNPs and specific pre-natal environments contributing to this variability in theneonate epigenome. This study is the first large samplesize EWAS (N = 987) that assesses the impact of prenatalenvironment and genetic and epigenetic factors on birthweight and size/adiposity in early childhood. It is also thefirst neonate EWAS conducted in an Asian population.MethodsStudy populationThis work is part of the Growing Up in SingaporeTowards healthy Outcomes (GUSTO) study, a prospectiveLin et al. BMC Medicine  (2017) 15:50 Page 2 of 18mother-offspring birth cohort designed to investigate de-velopmental origins of health and disease (DOHaD). TheGUSTO cohort has been described previously [26]. Preg-nant women of at least 18 years of age and in their firsttrimester of pregnancy were recruited from the two majorpublic hospitals in Singapore with obstetric services (KKWomen’s and Children’s Hospital and the NationalUniversity Hospital) between 2009 and 2010. Eligible par-ticipants were Singaporean citizens, permanent residents,or those who planned to reside in Singapore for the next5 years, and intended to deliver the baby at the NationalUniversity Hospital or KK Women’s and Children’sHospital. They could be of Chinese, Malay or Indian eth-nicity, but with homogeneous parental ethnic background.Women who were on chemotherapy or psychotropicdrugs were excluded from the study. Interviewer-administered questionnaires were used to assess maternalpre-pregnancy weight, demographics (including maternalage and education) and maternal obstetric and medicalhistory at enrolment. All pregnant women underwent fourultrasound scans during pregnancy to measure fetalgrowth. Extensive maternal assessments were conductedat 26–28 weeks gestation. All offspring were assessed atbirth and at different later time points (3, 6, 9, 12, 15, 18,24, 36 and 48 months). This study is still active with plansto collect data up to adolescence. This is an interim ana-lysis on an ongoing cohort study. Of the 1177 singletondeliveries, 987 subjects were selected as fulfilling the fol-lowing inclusion criteria: full-term births with Apgarscore ≥ 9, and availability of at least one child weightmeasurement, infant genotype and methylation data(Additional file 1: Supplementary Figures A1–A3).Child characteristics and anthropometryChild weight and recumbent length/standing heightwere measured at birth and at nine subsequent timepoints (3, 6, 9, 12, 15, 18, 24, 36 and 48 months). Childweight was measured using calibrated scales (birth to18 months: SECA 334 Weighing Scale; 24 to 48 months:SECA 803 Weighing Scale, SECA Corp) and recorded tothe nearest gram. Recumbent length (birth, 3, 6, 9, 12,15, 18, 24 months) was measured using a SECA infantmat (SECA 210 Mobile Measuring Mat, SECA Corp)and recorded to the nearest 0.1 cm. Standing height (36and 48 months) was measured using a stadiometer(SECA stadiometer 213, SECA Corp) from the top ofthe child’s head to his or her heels, and recorded to thenearest 0.1 cm. Weight and length/height measurementswere taken in duplicates for reliability. BMI was derivedas weight (kg) divided by length2 (m2) at all time points.Subscapular and triceps skinfolds were measured atbirth, 18, 24, 36 and 48 months, and taken in triplicateusing the Holtain skinfold callipers (Holtain Ltd,Crymych, UK) on the right side of the body, andrecorded to the nearest 0.2 mm. Subscapular to tricepsskinfold ratio was derived by dividing subscapular skin-fold (mm) by triceps skinfold (mm). BMI is used as aproxy for adiposity in the analyses and to be concise, theterms BMI and adiposity have been used interchangeablyin this study. Some caution should be exercised in inter-preting the findings on BMI, because while BMI iswidely accepted as an indirect measure of adiposity, ithas its limitations. For example, elevated BMI levels mayarise as a result of extra muscle mass or stunted lineargrowth [27]. We have also included additional analysesusing skinfolds to capture adiposity. However, skinfoldswere measured at fewer time points and were generallyassociated with larger measurement error. Gestationalage (GA) was determined by ultrasonography in the firsttrimester. Child sex was taken from the medical records.Prenatal environment exposuresAn interviewer-administered questionnaire was con-ducted at 26–28 weeks of gestation to obtain informa-tion on occupational activity during pregnancy, alcoholusage before and during pregnancy, and smoking pat-terns before and during pregnancy. Maternal height andweight were measured during the same time period. Pre-pregnancy weight was self-reported during study recruit-ment in the first trimester of pregnancy. Gestationalweight gain (GWG) was calculated as the difference be-tween the pre-pregnancy and 26–28 week weights. Ma-ternal ppBMI was derived as pre-pregnancy weightdivided by height squared. Maternal glucose levels (2-hpost-glucose and fasting) were ascertained at 26–28weeks using an oral glucose tolerance test of 75 g afteran overnight fast (8–14 hours). Maternal plasma fattyacids, including n-6 polyunsaturated fatty acids (PUFA),n-3 PUFA, monounsaturated fatty acids (MUFA), andsaturated fatty acids, were measured using gas chroma-tography–mass spectrometry, and expressed as percent-age contribution to total plasma phosphatidylcholine(PC) fatty acid. Specifically, plasma lipids were extractedusing chloroform–methanol (2:1, v/v) and PC was iso-lated by solid phase extraction. Fatty acid methyl esterswere generated from PC after reaction with methanolcontaining 2% (v/v) sulfuric acid, extracted into hexaneand separated by gas chromatography. Fatty acid methylesters were identified by comparison with retentiontimes of previous standard runs and quantified usingChemStation software (Agilent Technologies). Maternalmicronutrient levels (vitamin D, vitamin B6, vitaminB12, folate, zinc, iron and magnesium) were measuredfrom serum drawn at 26–28 weeks of gestation. Mater-nal calorie intake at 26–28 weeks gestation wascalculated from both 24-h dietary recall and 3-day fooddiary. Maternal depressive symptoms were assessedusing the Edinburgh Postnatal Depression Scale, whichLin et al. BMC Medicine  (2017) 15:50 Page 3 of 18was designed and normed expressly for depressive symp-toms over the peripartum period [28] and is validatedfor prenatal screening for depression in Singaporeanwomen [29, 30]. Symptoms of anxiety were assessedusing the State–Trait Anxiety Inventory [31], which is acomprehensive and common research tool that mea-sures both stable (trait) and more transient (state) symp-toms. Importantly, translation and back-translation of allquestionnaires into individual languages, includingChinese, Tamil and Malay, have been performed andvalidated to ensure consistency to the English version.This study included administration of questionnaires inall three languages according to the language preferenceindicated by the mother [32]. Birth order and mode ofdelivery were extracted from hospital medical records.We note that all prenatal exposures/factors listed herecontribute to the prenatal environment; to be concise,we have used the term “prenatal environment” to referto these exposures/factors in the subsequent sections.Infant methylation dataMethylation profiling of umbilical cord samples wasperformed using the Infinium HumanMethylation450array, following standard protocol, and processed usingin-house quality control procedure [33]. Raw methyla-tion beta values were exported from GenomeStudio™.Probes with less than three beads for methylated orunmethylated channel or with detection P > 0.01 wereset to missing. Probes from sex chromosomes wereremoved. Colour adjustment and normalisation of Type 1and 2 probes was performed. Methylation beta valueswere first converted to M-values before applying COM-BAT to remove batch (plate) effects [34], and the batch-corrected methylation values transformed back to betavalues. Finally cross-hybridising probes [35, 36], as wellas probes where the methylation range (maximum-minimum, excluding outliers) was less than 10%, wereexcluded, giving a total of 174,211 CpGs for analysis. Wedid not filter the probes that were annotated to be locatedwithin SNPs before analysis. Instead, a post-hoc analysiswas performed on the top birth weight-associated CpGsto ensure that (1) no common SNP was located at theCpG and the single base extension, and (2) scatterplots ofthe methylation values showed a “cloud-like” distributionand not a multi-modal distribution [37, 38].Infant genotype dataGenotyping was performed using the IlluminaOmniexpress + exome array. Non-autosomal SNPs, SNPswith call rates < 95%, or minor allele frequency < 5%, orthose that failed Hardy–Weinberg Equilibrium wereexcluded from the analysis. Principal components analysiswas used to confirm self-reported ethnicity/ancestry.Samples with call rate < 99%, cryptic relatedness and sex/ethnic discrepancies were excluded. Alleles on the positivestrand were reported as per the hg19 build of the humangenome assembly. After quality control filtering 577,204SNPs were available for downstream analysis.Statistical analysisThe overall analysis framework is summarised inAdditional file 1: Supplementary Figure A4 and eachanalysis is elaborated below. Information on covariates/confounders was available for all 987 infants; where rele-vant, these variables were adjusted for in the statisticalmodels. These variables included infant ethnicity, infantsex, gestational age and cellular proportions (estimatedfrom DNA methylation data). A complete-case analysiswas conducted, i.e. for each model, all infants withcomplete information for the outcome(s) and predictor(s)were included in the analysis.Prenatal environment influences on birth weightLinear regression models were used to examine the asso-ciation of 30 prenatal environment variables with infantbirth weight. Eleven of these 30 prenatal environmentvariables that associated with birth weight were used forsubsequent analysis. We first separately studied theassociation of each prenatal environment variable withbirth weight, adjusted for infant sex, ethnicity and GA.This was followed by the association of prenatal environ-ment variables with birth weight, adjusted for each otheralong with infant sex, ethnicity and GA. We examinedthe distribution of infant birth weight, and subsequentlydecided to use a log-transformation on infant birthweight to improve normality and reduce the impact ofoutliers. Following Gelman [39], binary environmentvariables were not scaled so that their estimates could bedirectly interpreted. Since the unscaled binary environ-ment variables generally have a standard deviation (SD)of approximately 0.5, continuous prenatal environmentvariables were standardised to have a SD of 0.5 (centredand divided by two times SD), so that effect estimatesfrom both continuous and binary prenatal environmentvariables were comparable. Note that this is differentfrom the Z-score, which is obtained by centring anddividing by one SD. Due to the standardisation of prenatalenvironment variables and log-transformation on birthweight, effect estimates are interpreted as percentagechange in birth weight for a 2 SD increase in prenatalenvironment variable (for continuous prenatal environ-ment variables), or percentage change in birth weightfor comparing two categories of prenatal environmentvariable (for binary prenatal environment variables).Genetic influences on birth weightTo determine whether genetic variation at loci previ-ously associated with adult adiposity was associated withLin et al. BMC Medicine  (2017) 15:50 Page 4 of 18newborn size/adiposity, polygenic risk score (PRS), orcumulative genetic risk profile, was computed for eachinfant in the GUSTO cohort using regression coeffi-cients and P values for adult BMI reported by theGenetic Investigation of ANthropometric Traits (GIANT)consortium [40]. PRS was computed using the P value-informed clumping procedure implemented in PLINK. Toreduce the inclusion of SNPs in linkage disequilibrium(LD), two rounds of clumping were performed. We firstused a cut-off of R2 = 0.5 within a 250-kb window to iden-tify potential index SNPs; in each 250-kb window, theSNP with the smallest P value from GIANT was keptwhile SNPs in LD (R2 > 0.5) were removed. Second, to fur-ther exclude SNPs in long-range LD with potential indexSNPs, the clumping procedure was repeated with a cut-offof R2 = 0.2 within a 5-Mbp window. For each individual,the cumulative score was computed by summing thenumber of score alleles, weighted by the regression coeffi-cients reported by the GIANT consortium. We computedPRS for each ethnic group separately, and at differentP value thresholds pT for the index SNPs (pT from10–10 to 1). For each ethnic group, PRS was standar-dised to mean 0 and variance 1 (Z-score) separately.We then regressed PRS against log-transformed childanthropometric measures, adjusted for child sex andGA, for each ethnic group. We examined ethnicity-stratified associations of PRS with child anthropomet-ric outcomes for different P value thresholds. Foreach ethnic group, we selected the P value thresholdthat gave the best-fit score (defined as PRS showingconsistent associations with child weight and BMI atmultiple time points). This best-fit PRS was then usedfor subsequent analysis. We did not use other childoutcomes (subscapular skinfolds, triceps skinfolds andsubscapular:triceps ratio) for evaluating the best-fitscore as these outcomes were measured at fewer timepoints and generally had larger measurement error.We also did not consider child length for evaluatingthe best-fit score as it did not capture adiposity.However, we report associations between PRS for allchild outcomes. For concision, the result sections re-port only the conclusive findings (child weight andBMI), while the rest (e.g. skinfolds) are providedunder Additional file 1: Supplementary File B.Birth weight and neonatal DNA methylomeTo interrogate the association between perinatal methy-lome and birth weight, we performed linear regressionof log-transformed birth weight against methylation ateach CpG site, adjusted for child sex, GA, ethnicity, cel-lular proportions and interactions between ethnicity andcellular proportions. Cellular proportions for fibroblasts,B-cells and T-cells were estimated [41] using a cell-specific methylation profile reference panel (accessionnumber EGAD00010000460) [42]. A principal componentsanalysis was performed on the three estimated cellular pro-portions and the first two principal components adjustedas covariates in all subsequent regression models. Since theassociations of estimated cellular proportions with birthweight were ethnicity dependent (data not shown), inter-action terms between principal components of cellularproportions and ethnicity were included as covariates in allregression models. For sensitivity analysis, we applied anadditional method (surrogate variable analysis) to correctfor cellular heterogeneity that did not require a reference-panel of cell-specific methylation profiles [43–45]. We fur-ther applied genomic control to the P values if the genomicinflation factor computed across 174,211 CpGs was greaterthan 1. A genomic control correction could help correctfor residual confounding due to cellular heterogeneity;however, it could also be too conservative, as a global infla-tion in epigenome-wide P values in response to increasedadiposity in adults has been previously reported [46] andcould be a true biological phenomenon. To adjust formultiple testing across 174,211 CpGs, we report all CpGsassociated with birth weight at a false discovery rate (FDR)< 0.05 [47]. This subset of CpGs identified at FDR 0.05 wasfurther investigated below. For CpGs significantly associ-ated with birth weight at FDR 0.05, we also examined if theassociations differed among the three ethnic groups byassessing interactions with ethnicity. This analysis wasdone by regression of log-transformed birth weight againstinteraction terms between methylation and ethnicity,adjusted for main effects of methylation, main effects ofethnicity, child sex, GA, cellular proportions, and interac-tions between ethnicity and cellular proportions. Interac-tions with infant sex were assessed in a similar manner.Genetic and environmental influences on top CpGsWe then characterised the influences of the prenatalenvironment and SNPs on the variability in methylationat CpGs showing association with birth weight. First, toinvestigate the influence of the prenatal environment onmethylation levels, we regressed methylation at eachCpG site against (standardised) prenatal environmentvariables, adjusting for child sex, GA, ethnicity, cellularproportions and interactions between ethnicity and cel-lular proportions. To adjust for multiple testing acrosseight CpGs and 11 prenatal environments, a CpG wasdefined to be influenced by the prenatal environment ifthe most significant association with the prenatal envir-onment variables had genomic control-adjusted P < 0.05/(8 × 11) ~ 5 × 10–4. Genomic inflation factor was com-puted for each prenatal environment across all 174,211variable CpGs, and genomic control was applied if theinflation factor for the variable was above 1. This simpleBonferroni correction for multiple testing across the 11prenatal environments was likely to be conservative as theLin et al. BMC Medicine  (2017) 15:50 Page 5 of 1811 prenatal environments were associated with one another(for example, a mother who smokes during pregnancy ishighly likely to be smoking before pregnancy). FDR was notused for multiple testing adjustments here and for the topCpGs and offspring size/adiposity in early childhood be-cause of the relatively small number of tests (88 and 360,respectively) and dependency between the tests [48, 49].For each CpG, we also report the prenatal environmentvariable that showed the strongest association (smallestP value) with the CpG.Second, to interrogate the influence of SNPs onmethylation levels, we regressed each CpG against cis-SNPs (defined here as SNPs on the same chromosomeas CpG), using an additive genetic model, adjusted forchild sex, GA, ethnicity, cellular proportions, and inter-actions between ethnicity and cellular proportions. ForSNPs where the minor homozygote genotype grouphad ≤ 50 individuals, the minor homozygote and hetero-zygote genotype groups were combined (dominant gen-etic model). A total of approximately 5 × 105 CpG-SNPtests were conducted, corresponding to testing eightCpGs across 8,392 to 47,298 cis-SNPs for each CpG(each CpG was tested against 8,392 to 47,298 cis-SNPsdepending on the chromosome of the CpG). A CpG wasdefined to be influenced by the genotype (SNPs) if themost significant association between the CpG and cis-SNPs attained P < 1 × 10–7, the Bonferroni threshold tomaintain a family-wise Type 1 error rate of 0.05 acrossapproximately 5 × 105 tests.Top CpGs and offspring size/adiposity in early childhoodFinally, we examined whether these methylation marksat birth were associated with offspring weight in earlychildhood (3–48 months) and offspring length andadiposity (BMI, subscapular skinfold, triceps skinfoldand subscapular:triceps ratio) from birth to 48 months.We also examined BMI change in early childhood(calculated as the difference between BMI Z-score at48 months and birth), where BMI Z-score at birth and48 months were calculated using WHO child growthcharts. Child anthropometric measures (except BMIchange) were log-transformed to improve normality andreduce the impact of outliers. Each offspring anthropo-metric measure at each assessment time point wasanalysed separately. This was done by linear regressionof (log-transformed) anthropometric measures at eachtime point against the methylation at each CpG site, ad-justed for child sex, GA, ethnicity, cellular proportionsand interactions between ethnicity and cellular propor-tions. To account for multiple testing across the eightCpGs and 45 child size/adiposity measures, a CpGwould be associated with offspring size/adiposity if thegenomic control-adjusted P < 0.05/(8 × 45) = 1 × 10–4.The genomic inflation factor was computed for eachoffspring anthropometry measure across all 174,211 CpGs,and genomic control was applied if the inflation factor forthe anthropometry outcome was above 1. This simple Bon-ferroni correction for multiple testing across differentsize/adiposity measures was likely to be extremelyconservative as the 45 size/adiposity measures werestrongly associated with each other. For concision, theresult sections describe only the conclusive findings,while the rest are reported in Additional file 1: Supple-mentary File F.Multiple testing correctionsWe used different multiple testing methods (FDR vs. Bon-ferroni) at different analysis steps in the sections above. Thereason for the use of different methods was due to thevastly different number of tests to be adjusted for multipletesting in each analysis step. To adjust for multiple testingacross 174,211 CpGs in birth weight and neonatal DNAmethylome, we used FDR. For the environmental influ-ences on top CpGs and for the top CpGs and offspringsize/adiposity in early childhood sections, FDR could not beused for multiple testing adjustments because of therelatively small number of tests (88 and 360, respectively)and dependency between the tests [48, 49]. Instead, weused Bonferroni threshold in order to maintain a family-wise Type 1 error rate of 0.05 at each analysis step.Accessing DNA methylation dataThe infant methylation data analyzed in the currentstudy is available as Additional files 2 and 3.ResultsBirth weight is associated with 11 prenatal environmentsThis analysis used 987 of 1177 singleton deliveries in theGUSTO cohort study. The subject selection criterion in-cluded live singleton term births with Apgar score ≥ 9,and availability of anthropometric measures, covariates/confounder information, as well as genotyping andmethylation data for all subjects (Additional file 1:Supplementary Figures A1–A3). Summary statistics ofthese 987 mother-offspring participants are provided inTables 1 and 2; 58%, 17% and 25% of the participantswere from Chinese, Indian and Malay ethnicity, respect-ively; and 52% of the infants were male. The number ofchildren with age- and sex-specific BMI Z-score exceed-ing 2 and 3 at each time point are reported in Additionalfile 1: Supplementary Table A1. The number of motherswho were underweight, normal weight, overweight andobese before pregnancy is reported in Additional file 1:Supplementary Table A2. Using non-Asian BMI cut-offs,12%, 64%, 17% and 7% of the mothers were underweight,normal weight, overweight and obese, respectively,before pregnancy. When Asian-specific BMI cut-offs wereLin et al. BMC Medicine  (2017) 15:50 Page 6 of 18used, more women were classified as being overweight(22%) and obese (14%) before pregnancy.We assessed 30 prenatal environment variablesfor association with birth weight (Additional file 1:Supplementary Table A3). Of the 30 prenatal environ-ment variables analysed (Tables 1 and 2), infant birthweight was associated with maternal ppBMI, maternalGWG, maternal height, maternal glucose levels (fastingand 2-h post-75 g-glucose challenge), maternal plasman-6 PUFA and MUFA levels at 26 weeks gestation,maternal age, and maternal smoking before and dur-ing pregnancy (Fig. 1; Additional file 1: SupplementaryTables A3 and A4; P < 0.05). There was also a suggestiveassociation with parity (Fig. 1; Additional file 1: Supple-mentary Tables A3 and A4; P = 0.059). Greater maternaladiposity (ppBMI and GWG), height, glucose levels (fast-ing and 2-h post-glucose), n-6 PUFA levels, age and paritywere associated with higher birth weight, while higherMUFA levels and maternal smoking (before and duringpregnancy) were associated with lower birth weight. Birthweight changed by 2.2–5.5% for every 2 SD change in ma-ternal adiposity (ppBMI and GWG), height, glucose levels(fasting and 2-h post-glucose) or FA levels (n-6 PUFA andMUFA). The effect sizes for parity (non-first born vs. firstborn), maternal age (≥35 years vs. < 35 years) and smoking(yes vs. no) were similar and ranged from 1.4% to 6.5%(Fig. 1c; Additional file 1: Supplementary Table A3). Sevenof 11 of these prenatal environment variables, includingmaternal adiposity (ppBMI and GWG), glucose levels(fasting and 2-h post-glucose), FA levels (n-6 PUFAand MUFA), and smoking during pregnancy, alsoshowed association with child BMI at birth (P < 0.05;Additional file 1: Supplementary Figure A6). Consistentwith earlier findings [50, 51], maternal ppBMI, GWG andglucose levels were also significantly associated with bothchild weight and BMI at 48 months of age (Additional file 1:Supplementary Figures A5–A8). For subsequent analysesTable 1 Offspring characteristics of the GUSTO cohort studiedin the analysisTime point N (%) Mean (SD)Ethnicity Chinese Delivery 570 (58%)Malay 247 (25%)Indian 170 (17%)Child sex Male 517 (52%)Female 470 (48%)Gestational age (weeks) 987 39 (1)Weight (g) Delivery 959 3130.5 (380.9)3 months 904 6150.6 (778.7)6 months 864 7717.1 (914.3)9 months 829 8615.0 (1001.4)12 months 846 9380.2 (1078.6)15 months 851 10086.2 (1164)18 months 804 10742.4 (1298.7)24 months 818 11981.6 (1552.8)36 months 824 14249.8 (2028.2)48 months 718 16442.1 (2692.4)Length/height (cm) Delivery 959 48.7 (1.8)3 months 904 60.9 (2.4)6 months 868 67.1 (2.7)9 months 830 71.6 (2.8)12 months 848 75.4 (3.1)15 months 843 78.9 (3.2)18 months 689 82.1 (3.3)24 months 718 87.6 (3.6)36 months 817 94.8 (3.8)48 months 716 102.3 (4.2)Body mass index (kg/m2) Delivery 959 13.2 (1.2)3 months 904 16.5 (1.6)6 months 864 17.1 (1.6)9 months 829 16.8 (1.5)12 months 845 16.5 (1.4)15 months 843 16.2 (1.4)18 months 687 15.9 (1.3)24 months 718 15.5 (1.4)36 months 817 15.8 (1.5)48 months 716 15.6 (1.8)Subscapular skinfold (mm) Delivery 959 5.0 (1.2)18 months 671 6.4 (1.4)24 months 757 6.4 (1.6)36 months 792 6.6 (1.9)48 months 674 6.8 (2.7)Table 1 Offspring characteristics of the GUSTO cohort studiedin the analysis (Continued)Triceps skinfold (mm) Delivery 960 5.5 (1.3)18 months 709 8.6 (1.7)24 months 733 8.8 (1.8)36 months 786 9.3 (2.3)48 months 684 9.8 (2.9)Subscapular:Triceps Delivery 959 0.9 (0.2)18 months 646 0.8 (0.1)24 months 722 0.7 (0.1)36 months 780 0.7 (0.1)48 months 671 0.7 (0.1)Lin et al. BMC Medicine  (2017) 15:50 Page 7 of 18Table 2 Maternal characteristics of the GUSTO cohort studied in the analysisTime point N (%) Mean (SD)Pre-pregnancy BMI (kg/m2) Self-reported at first clinic visit 906 22.7 (4.4)Gestational weight gain (kg) 26–28 weeks gestation 902 8.7 (4.7)Maternal height (m) 971 158.3 (5.6)Fasting glucose (mmol/L) 920 4.3 (0.5)2-h post-glucose (mmol/L) 920 6.5 (1.5)n-6 PUFA (%) 863 34.2 (3.3)n-3 PUFA (%) 863 6.4 (1.8)MUFA (%) 863 13.6 (2.3)SFA (%) 863 45.8 (3.3)EPDS score 955 7.4 (4.4)STAI state score 957 33.8 (10.0)STAI trait score 957 35.7 (9.6)Caloric intake 3-day food diary (kcal) 550 1871.2 (476.3)Caloric intake 24-h recall (kcal) 960 1843.6 (550.6)Parity >0 Delivery 536 (54%)0 451 (46%)Maternal age (years) ≥35 Self-reported at first clinic visit 251 (25%)<35 736 (75%)Smoking before pregnancy Yes Interviewer-administered questionnaire at 26–28 weeks gestation 121 (12%)No 855 (88%)Smoking during pregnancy Yes 24 (2%)No 951 (98%)Plasma vitamin D >50 nmol/L 26–28 weeks gestation 718 (87%)≤50 nmol/L 108 (13%)Plasma folate ≥6 ng/mL 774 (90%)<6 ng/mL 90 (10%)Plasma vitamin B12 ≥300 pg/mL 373 (43%)<300 pg/mL 492 (57%)Plasma vitamin B6 <20 nmol/L 137 (16%)≥20 nmol/L 727 (84%)Plasma iron ≥560 μg/L 403 (92%)<560 μg/L 36 (8%)Plasma zinc ≥700 μg/L 417 (95%)<700 μg/L 22 (5%)Plasma magnesium ≥18.25 mg/L 304 (69%)<18.25 mg/L 135 (31%)IVF birth Yes Self-reported at first clinic visit 69 (7%)No 918 (93%)Maternal education (years) ≥12 596 (61%)<12 379 (39%)Working during pregnancy Yes Interviewer-administered questionnaire at 26–28 weeks gestation 681 (70%)No 297 (30%)Lin et al. BMC Medicine  (2017) 15:50 Page 8 of 18on the associations of prenatal environment with neo-nate DNA methylation, we restricted the analyses to the11 birth weight associated prenatal environment variablesshown in Fig. 1. We note that these 11 prenatal environ-ments are not distinct/independent of each other, forexample, a mother who smokes during pregnancy ishighly likely to have been smoking before pregnancy.Birth weight and early childhood adiposity weresignificantly associated with polygenic risk score derivedfrom adult population studiesTo interrogate if the genetic variation at loci previouslyassociated with adult adiposity was associated withnewborn size/adiposity, PRS or cumulative genetic riskprofile was constructed using genetic variants previouslyreported to be associated with adult BMI by the GIANTconsortium [40]. This PRS showed a significant associationwith birth weight, supporting an overlap in the geneticfactors contributing to birth weight and adult adiposity(Fig. 2; Additional file 1: Supplementary Figure B1 andSupplementary Table B1). Birth weight increased by 1.6%for every 2 SD increase in PRS (Fig. 2a; Additional file 1:Supplementary Table B1). The association of PRS withbirth weight remained even after adjusting for the 11prenatal environment variables (Additional file 1:Supplementary Table A5 and Supplementary FigureA9), and the association of the prenatal environmentvariables with birth weight was not PRS dependent(Additional file 1: Supplementary Table A6), indicating theindependent influences of genotype and prenatal environ-ment on birth weight; 18% of the total variation in birthweight was explained by infant sex, ethnicity and GA,Table 2 Maternal characteristics of the GUSTO cohort studied in the analysis (Continued)Alcohol use before pregnancy Yes 338 (35%)No 636 (65%)Alcohol use during pregnancy Yes 19 (2%)No 938 (98%)BMI body mass index, EPDS Edinburgh Postnatal Depression Scale, IVF in vitro fertilisation, MUFA monounsaturated fatty acids, PUFA polyunsaturated fatty acids,SFA saturated fatty acids, STAI State–Trait Anxiety Inventorya bc dFig. 1 Prenatal environment influences on birth weight. a Scatterplots of birth weight (vertical axis) against significantly associated continuousprenatal environment variables (horizontal axis). b Boxplots of birth weight (vertical axis) against significantly associated binary prenatalenvironment variables (horizontal axis). c Univariate association between birth weight and each significantly associated prenatal environmentvariable, adjusted for infant sex, ethnicity and gestational age. Point estimates (height of bars) and 95% confidence intervals (top and bottomwhiskers), show percentage change in birth weight for two standard deviations increase in continuous prenatal environment variable, or forcomparing the two categories of binary prenatal environment variables. d Multivariate association between birth weight and significantlyassociated prenatal environment variables, adjusted for infant sex, ethnicity, gestational age and for each other. Point estimates (height of bars)and 95% confidence intervals (top and bottom whiskers), show percentage change in birth weight, for two standard deviations increase in acontinuous prenatal environment variable, or for comparing the two categories of binary prenatal environment variablesLin et al. BMC Medicine  (2017) 15:50 Page 9 of 18while an additional 14% was explained by 11 prenatalenvironment variables and PRS together. An associationbetween PRS and child BMI at birth was also observed(Fig. 2b) and PRS was longitudinally associated withweight and BMI in early childhood (Additional file 1:Supplementary Figure B1 and Supplementary Table B1).This longitudinal association of PRS with weight and BMIfrom 3 to 48 months of age remained after adjustment forbirth weight or BMI (Additional file 1: SupplementaryFigure B2).Birth weight was significantly associated with methylationat eight CpGs within seven gene lociDNA from umbilical cord tissue of the 987 neonateswas interrogated on Infinium HumanMethylation450BeadChip arrays. A total of 174,211 CpGs were iden-tified to vary in methylation by more than 10% acrossthe subjects. These CpGs were more likely to be lo-cated in open seas and intronic/intergenic regions(Additional file 1: Supplementary Figure C1). An EWASon birth weight was performed using these variably meth-ylated CpGs, and adjusted for child sex, GA, ethnicity,cellular proportions and interactions between ethnicityand cellular proportions. Methylation levels at eight CpGswere identified to be significantly associated with birthweight at a FDR of 0.05 (Table 3; Additional file 1:Supplementary Figures C2 and C3). Among them, sixCpGs were located within the protein coding genes:(1) 5’-UTR of Ankyrin-3 (ANK3; P = 4.6 × 10–8); (2) 3’-UTR of Cyclin-Dependent Kinase Inhibitor 2B (CDKN2B;P = 4.9 × 10–8); (3) intron of Immunoglobulin Superfamily,DCC Subclass, Member 4 (IGDCC4; P = 1.6 × 10–7);(4) intron of Prolyl 4-hydroxylase, Alpha PolypeptideIII (P4HA3; P = 4.0 × 10–7); (5) intron of CalciumChannel, Voltage-Dependent, T Type, Alpha 1G Subunit(CACNA1G; P = 1.2 × 10–6); and (6) intron of Zinc FingerProtein 423 (ZNF423; P = 1.9 × 10–6), while the remainingtwo CpGs mapped to the non-coding gene MIRLET7BHG(P = 9.9 × 10–7 and P = 2.2 × 10–6).Variability in methylation at these seven gene loci(eight CpGs) was modest with an interquartile range(IQR) of 4.6% to 9.6%. DNA methylation levels at five ofseven loci (five CpGs) were positively associated withbirth weight, while DNA methylation at the remainingtwo loci (three CpGs) were negatively associated withbirth weight. The effect sizes were modest, with a3.7–9.2% change in birth weight associated with a10% increase in methylation (corresponding to ap-proximately 0.4–2 IQR). Together, these eight CpGsaccounted for an additional 9.5% of the total variationin birth weight, in addition to the 32% accounted byinfant sex, ethnicity, GA, 11 prenatal environmentsa bc dFig. 2 Genetic influences on birth weight: Associations of child weight (a and b) and body mass index (c and d) at different time points withbest-fit polygenic risk score (PRS). Best-fit PRS for Chinese, Malay and Indian ethnic groups used clumping P value thresholds pT = 0.5, 0.1 and10–4, respectively. PRS was standardised to mean zero and unit variance within each ethnic group. Left panel (a and c) shows point estimates(height of bars) and 95% confidence intervals (top and bottom whiskers), for percentage change in child outcome, for a 2 SD increase in PRS,adjusted for child sex, gestational age and ethnicity. Analysis was done by linear regression of log-transformed child anthropometric outcome ateach time point against PRS, adjusted for child sex, gestational age and ethnicity. Right panel (b and d) shows scatterplot of standardised (meanzero and unit variance) log-transformed child outcome (vertical axis) against PRS (horizontal axis)Lin et al. BMC Medicine  (2017) 15:50 Page 10 of 18and PRS. Sensitivity analysis using a reference-freemethod to adjust for cellular heterogeneity gave similarresults (Additional file 1: Supplementary Table C1). Theassociations between birth weight and methylation atthese sites did not depend on ethnicity (Additional file 1:Supplementary Table C2) or infant sex (Additional file 1:Supplementary Table C3). For subsequent analyses, weused all seven loci (eight CpGs) identified at FDR < 0.05.Methylation levels at three of the seven birth weight-linked loci were significantly associated with prenatalenvironmentWe interrogated the contribution of individual prenatalenvironments on variability in the epigenome at theseseven loci (eight CpGs). Methylation levels at three ofseven loci (IGDCC4, MIRLET7BHG, CACNA1G) weresignificantly associated with the prenatal environmentafter adjusting for multiple testing (Fig. 3; Additionalfile 1: Supplementary Table D1; P < 5 × 10–4). Methyla-tion levels at cg25685359 (MIRLET7BHG) showed asignificant inverse association with maternal n-6 PUFAlevels (Fig. 3; P = 4.2 × 10–4), and a significant positive as-sociation with maternal smoking before pregnancy (Fig. 3;P = 2.3 × 10–4). Methylation levels at cg25487405, whichalso mapped to MIRLET7BHG, showed modest associa-tions (P < 0.05) with these two prenatal environmentvariables, though the associations did not survive multipletesting adjustments. The directionality of the associationsbetween methylation and prenatal environments is con-sistent (Fig. 3) as cg25685359 (MIRLET7BHG) showed anegative association with birth weight (Table 3), and birthweight was positively associated with maternal n-6 PUFAlevels but negatively associated with maternal smoking(Fig. 1).Methylation at cg23671997 (IGDCC4) showed a sig-nificant positive association with maternal fasting glu-cose levels (Fig. 3; P = 2.7 × 10–4), and it also showedpositive association with maternal ppBMI (P = 8.1 × 10–4).Likewise, methylation at cg22383874 (CACNA1G) wassignificantly and positively associated with maternal fast-ing glucose levels (Fig. 3; P = 1.7 × 10–4), and was alsopositively associated with maternal ppBMI (P = 2.9 × 10–2)and maternal 2-h post-glucose levels (P = 4.2 × 10–3).The directionality of associations between cg23671997(IGDCC4) and cg22383874 (CACNA1G) and the pre-natal environments is consistent (Fig. 3), as methyla-tion levels at both CpGs were positively associatedwith birth weight (Table 3), and birth weight was posi-tively associated (Fig. 1) with maternal adiposity-related influences (ppBMI, fasting and 2-h post-glucoselevels at mid-pregnancy). After adjustment for mater-nal ppBMI, the associations of cg23671997 (IGDCC4)and cg22383874 (CACNA1G) with maternal fastingmaternal glucose levels were similar but slightly re-duced (P = 2.7 × 10–4 vs. P = 2.3 × 10–3 for cg23671997;P = 1.7 × 10–4 vs. P = 4.0 × 10–4 for cg22383874; Fig. 3a vs.Additional file 1: Supplementary Figure D1).Methylation levels at three of the seven birth weightlinked loci were significantly associated with SNPsTo investigate the influence of genetic polymorphismson methylation at the seven birth weight associated loci(eight CpGs), we regressed each CpG against all cis-SNPs (SNPs on the same chromosome as the CpG).Three loci were significantly associated with cis-SNPsafter adjusting for multiple testing (Additional file 1:Supplementary Table E1). These three loci includedP4HA3, ZNF423 and MIRLET7BHG (only one of the twoMIRLET7BHG CpGs was significantly associated withSNPs). The CpG-SNP distances ranged from 12 to 168 kb(Additional file 1: Supplementary Table E1). For thesethree CpG-SNP pairs, the association of methylation withbirth weight (effect sizes and P values) was similar withand without adjustment for genotype at the SNP, and theTable 3 Methylome-CpGs associated with birth weight at a false discovery rate of 0.05CpG CHR POS IQR Est 95% CI P Gene Annotationcg00510507 10 61900413 8.4 4.9 (3.5 to 6.2) 4.6 × 10–8 ANK3 5’ UTRcg08390209 9 22005563 6.6 7.1 (5.1 to 9.0) 4.9 × 10–8 CDKN2B 3’ UTRcg23671997 15 65677753 4.6 9.2 (6.5 to 12) 1.6 × 10–7 IGDCC4 Introncg14300531 11 73969506 9.6 –3.9 (–5.0 to –2.8) 4.0 × 10–7 P4HA3 Introncg25685359 22 46473721 8.8 –3.7 (–4.8 to –2.6) 9.9 × 10–7 MIRLET7BHG Non-codingcg22383874 17 48670670 4.8 7.6 (5.2 to 10) 1.2 × 10–6 CACNA1G Introncg02729344 16 49888237 6.6 6.8 (4.7 to 9.0) 1.9 × 10–6 ZNF423 Introncg25487405 22 46473039 5.5 –5.6 (–7.2 to –3.9) 2.2 × 10–6 MIRLET7BHG Non-codingEight CpGs were significantly associated with birth weight at a false discovery rate (FDR) of 0.05. The eight CpGs mapped to seven loci (two CpGs mapped toMIRLET7BHG). Regression coefficients (Est), 95% confidence intervals (CI) and P values are reported as percentage change in birth weight for 10% increase inpercent methylation. Interquartile range (IQR), chromosome (CHR) and position (POS) of CpG are also shown. Analysis was done by linear regression of log-transformed birth weight against methylation at each CpG site, adjusted for child sex, gestational age, ethnicity, cellular proportions and interactions betweenethnicity and cellular proportionsLin et al. BMC Medicine  (2017) 15:50 Page 11 of 18genotype at the SNP was not associated with birth weight(Additional file 1: Supplementary Table E2). Finally, wealso investigated if the PRS was associated with methyla-tion levels at these eight CpGs, and also if the PRSmoderates the associations between methylation and birthweight/environment, but no significant associations wereobserved (Additional file 1: Supplementary Tables E3–E5).Methylation levels at six of the seven birth weight linkedloci predicted offspring size/adiposity at 48 monthsMethylation levels at all seven loci (eight CpGs) showedassociation with child weight in at least one time pointin early childhood (3–48 months), even though theseassociations did not survive multiple testing adjustments(Fig. 4a; P < 0.05). The effect sizes (associations betweenmethylation and child weight) were either (1) strongest atbirth and decreased from 3 to 48 months, or (2) strong atbirth, decreased initially, and then increased from 18 to48 months, or (3) strongest at birth, but remained the same(approximately) from 3 to 48 months (Fig. 4a). Methylationlevels at six of seven loci (six CpGs) were also significantlyassociated (P < 1 × 10–4) with BMI at birth (the remainingtwo CpGs showed suggestive associations; P < 0.005); thechange in effect sizes of BMI with child age showed asimilar pattern as that of child weight (Fig. 4b). At age48 months, methylation levels at six of seven loci (sixCpGs) and two of seven loci (two CpGs) showed moderateassociations with child weight and BMI, respectively (P <0.05). The associations between neonate methylation andchild size/adiposity in early childhood (3–48 months) werenot independent of birth weight (data not shown). Methyla-tion levels at cg25685359 (MIRLET7BHG) showed asuggestive association with BMI change in early child-hood (Additional file 1: Supplementary Table F1), whereBMI change was calculated as the difference betweenage- and sex-specific Z-score at 48 months and birth;this association did not survive adjustment for birthweight either (P > 0.05).a bFig. 3 Influence of prenatal environment on methylome at birth. a Associations of DNA methylation at birth with prenatal environment. Colourin heatmap represents regression coefficients for associations between methylation and each prenatal environment variable. Each row representsa CpG and each column represents a prenatal environment variable. With increasing magnitudes, colour changes from white to red (for negativecoefficients) or from white to blue (for positive coefficients). Asterisks within each square represent P values for associations between methylationand each prenatal environment variable (P < 5 × 10–8 is represented with eight asterisks, 5 × 10–8 ≤ P < 5 × 10–7 is represented with seven asterisks,5 × 10–3 ≤ P < 5 × 10–2 is represented with two asterisks, P ≥ 5 × 10–2 is represented with a blank square). Analysis was done by linear regression ofmethylation at each CpG site against each prenatal environment variable, adjusted for child sex, gestational age, ethnicity, cellular proportionsand interactions between ethnicity and cellular proportions. Regression coefficients and P values are reported as an increase in percent methylationfor a 2 SD increase in continuous prenatal environment variable, or for comparing the two categories of binary prenatal environment variables.b Flow chart summarises associations between birth weight, methylation and prenatal environment for three CpGs (three loci) influenced by theprenatal environment. A CpG was defined to be influenced by the prenatal environment if the most significant association between the CpG andprenatal environment attained a P value of < 5 × 10–4 the Bonferroni threshold to maintain a family-wise Type 1 error rate of 0.05 across approxi-mately 100 tests (8 CpGs x 11 prenatal environment variables). Directions in arrows indicate temporal sequence, measurements obtained at thesame time are indicated with two-headed arrowsLin et al. BMC Medicine  (2017) 15:50 Page 12 of 18DiscussionWe have demonstrated that genetic, epigenetic andprenatal environmental factors are linked to offspringsize and adiposity at birth and in early childhood. Firstly,we identified individual prenatal environmental influ-ences on birth weight; we have previously reported thatsome of these prenatal environment variables (maternalppBMI, GWG and glucose levels) continued to associatewith offspring size and adiposity in early childhood[50, 51]. Secondly, genetic variation, as captured byPRS, not only influenced birth weight, but also childsize and adiposity up to 48 months of age, independ-ent of birth weight. The PRS was constructed usingadiposity-linked genetic risk variants previously re-ported in an adult population. The association ofadult adiposity risk score with size and adiposity inour paediatric population indicates that the effects ofgenetic risk variants can be detected as early as birth.This finding is also in confirmation with the earlierstudy that reported an association between newbornweight and adiposity with adult adiposity-derived PRS[52]. Thirdly, neonatal methylation levels at seven lociwere associated with birth weight. At six of the sevenloci, there was suggestive evidence that the associa-tions continued to persist up to 48 months of age.Among them, two of the loci (CDKN2B/P4HA3) alsoshowed suggestive association with child BMI at 48months. Even though the associations in early child-hood did not survive multiple testing corrections,these CpGs still hold potential as biomarkers of ad-verse metabolic trajectory as the prevalence of obesityincreases with age and might become more apparentlater in the life-course. Lastly, methylation levels at threeof seven loci associated with birth weight (IGDCC4,MIRLET7BHG, CACNA1G) also showed significant asso-ciations with the prenatal environment; however, similaranalyses with childhood weight and adiposity measuresshowed suggestive associations. Together, these findingsprovide evidence that birth weight is influenced by bothgenetic and prenatal environment factors, possibly actingthrough different mechanisms, either by altering the epi-genome (evidenced by CpGs that were associated withprenatal environment and/or SNPs) or independently ofthe epigenome (e.g. the PRS).Notably, four of seven methylation loci were located incoding genes (ANK3, CDKN2B, CACNA1G) and thea bFig. 4 Influence of methylome at birth on adiposity outcomes in early childhood: Associations of child weight (a) and body mass index (b) atdifferent time points with DNA methylation at birth. Colour in heatmap represents regression coefficients for associations between childanthropometric outcome and methylation. Each row represents a CpG and each column represents a time point. With increasing magnitudes,colour changes from white to red (for negative coefficients) or from white to grey (for positive coefficients). Asterisks within each squarerepresent P values for associations between child anthropometric outcome and methylation (P < 5 × 10–8 is represented with eight asterisks,5 × 10–8 ≤ P < 5 × 10–7 is represented with seven asterisks, 5 × 10–3≤ P < 5 × 10–2 is represented with two asterisks, P ≥ 5 × 10–2 is represented witha blank square). Analysis was done by linear regression of log-transformed child anthropometric outcome at each time point against methylationat each CpG site, adjusted for child sex, gestational age, ethnicity, cellular proportions and interactions between ethnicity and cellular proportions.Regression coefficients and P values are reported as percentage change in child anthropometric outcome for 10% increase in percent methylationLin et al. BMC Medicine  (2017) 15:50 Page 13 of 18miRNA let-7b host gene (MIRLET7BHG) that have beenpreviously implicated in metabolic disorders in humanadults and animal model systems. ANK3 encodes a pro-tein from ankyrin family, and ankyrins have been associ-ated with age dependent adiposity and insulin resistancein a rat model system [53]. CDKN2B is known to beinvolved in metabolic processes since it is highlyexpressed in subcutaneous adipose tissue, and its expres-sion alters with energy balance (higher expression inobese subjects and down-regulated expression duringcalorie restriction-induced weight-loss) [54]. Further-more, genetic variants near the CDKN2A/B 9p21.3 locuswere previously found to be associated with risk forCVD and T2DM in adults [55]. T-type calcium channelsare implicated in maintaining body weight in a ratmodel, where the administration of CACNA1G antago-nists to obese rodents results in reduced body weightand fat mass, and increased lean muscle mass [56].MicroRNA let-7B, transcribed from the MIRLET7BHGhost gene, belongs to the let-7 family of miRNA that isknown to play an important role in adipocyte differenti-ation (3T3-L1 mouse cells) by targeting HMGA2, a tran-scription factor that regulates growth and proliferation[57, 58]. Furthermore, transgenic mouse experimentshave shown that let-7 is a potent regulator of glucosemetabolism and peripheral insulin receptors, by target-ing insulin-like growth factor 1 (Igf1r), insulin receptor(Insr) and insulin receptor substrate-2 (Irs-2) in skeletalmuscle and liver tissues [59]. Let-7 is also a potentialbiomarker for metabolic disease. In a human interven-tional study reducing the glycemic load in the diet ofhealthy premenstrual women, let-7b was the mostdramatically altered miRNA, with nearly an eightfoldincrease of plasma let-7b after 12 months [60].As mentioned earlier, some of these loci also showedassociation with either the prenatal environment(MIRLET7BHG, IGDCC4, CACNA1G) or suggestiveassociation with child BMI at age 48 months (CDKN2B,P4HA3). Collectively, our findings fit within the para-digm of epigenetic mediation in the DOHaD hypothesis.According to the DOHaD hypothesis, the predispositionto adulthood diseases is primed in utero by specific ante-natal environments [6], and the mechanistic underpin-nings of this phenomenon includes alterations in theepigenome [6]. Here, our discovery of an altered neo-natal epigenetic profile at metabolism-linked gene lociand its associations with prenatal environment and theonset of adiposity in utero fit with this paradigm.However, we note that the longitudinal contributions ofprenatal environment and the associated changes in themethylome were observed to be moderate for childhoodadiposity. Obesity is a complex multifactorial diseasethat is responsive to environmental changes. Likewise,the epigenome is a modifiable factor and sensitive todevelopmental and environment cues. In the future, itwould be critical to test how these prenatal environmentinduced changes in the child’s methylome interact oralter with the postnatal environment and developmentalchanges. It is evident that the associations of methyla-tion (at birth weight-linked loci) with child weight andadiposity either (1) stayed strongest at birth and declinedby 48 months, or (2) stayed strong at birth, decreasedinitially, and then increased from 18 to 48 months, or(3) were strong at birth, but remained the same(approximately) from 3 to 48 months (Fig. 4a and b).These observations very well indicate that epigeneticprograming of obesity in early life is dynamic, and caneither weaken, strengthen, or stay unchanged with time.Hence, it is possible that some of these epigenetic varia-tions acquired at birth will either become benign, or stayactive and become more detrimental later in the life-course. This is further supported by published literaturewhich shows that childhood obesity increases with age;the prevalence of childhood obesity among children aged7–11 years is almost double than that of children aged2–6 years [61]. Evaluation of these candidate loci forsubject risk stratification or obesity prevention requiresfurther work to examine how DNA methylation levels atthese loci changes with age and environmental expo-sures during childhood.This study has several strengths, including its pro-spective and longitudinal study design with a relativelylarge sample size. The previous three birth weightEWAS [19–21] that had comparable sample sizes didnot incorporate genetic or extensive prenatal environ-ment information. Also, the longitudinal offspringanthropometric measures allowed us to study theassociation of perinatal methylation with both birth andpostnatal outcomes for up to 48 months of age. Simpkinet al. [20] and Sharp et al. [21] also examined adipositymeasures (and methylation measures) in childhood andadolescence, but did not provide detailed information inearly childhood. Additionally, our study population iscomprised of three major Asian ethnic groups that makeup more than 40% of the world’s population, while previ-ous investigations were conducted primarily amongCaucasian participants. On examination of the CpGspreviously reported to be associated with birth weight[19–22], cg04521626, which mapped to the phospholip-ase D2 (PLD2) gene, was statistically significant afteradjustment for multiple testing in our cohort (Additionalfile 1: Supplementary Table C4). For other CpGs that wecould not replicate in our study, deviations from previ-ous findings could be due to the underlying differencesin the populations examined (different genetic and/orprenatal environment influences in different ethnicgroups). Deviations could also be due to differences intissues assayed (cord tissue vs. cord blood) as DNALin et al. BMC Medicine  (2017) 15:50 Page 14 of 18methylation is cell type-specific, and cord tissue andcord blood have different cellular composition and celllineages. For example, cord tissue contains stromal cellsfrom mesenchymal stem cell lineage [62, 63], while cordblood contains mostly cell types from hematopoieticstem cell lineage. This further suggests that neonateEWAS findings may be ethnicity and/or tissue-specific.Cross-tissue/cross-population studies are needed to gen-eralise the findings to other tissues/populations. Add-itionally, cross-tissue comparisons will enable us todistinguish between common and tissue-specific signals.There are limitations of this study. First, residual con-founding is a concern in any epidemiological investiga-tion. In the context of EWAS, one of the major sourcesstems from cellular heterogeneity of the tissue being sur-veyed, as different cell types can have distinct methyla-tion profiles. Cord tissue, like other infant tissuesexamined in a neonate EWAS, is heterogeneous in itscellular content and consists of stromal, epithelial andendothelial cells (and possibly cord blood contamin-ation) [62, 63]. To combat the issue of cellular hetero-geneity, we employed two independent methods ofanalyses; however, this does not completely rule out theconfounding effects of cellular heterogeneity. Availabilityof better cell type reference sets developed by fraction-ation of cell types in infant cord tissue, in a population-specific manner, will alleviate this limitation in future. Inspite of the lack of comprehensive reference sets, an im-portant observation is that we did not find associationbetween the estimated cellular proportions and birthweight for the majority of the study individuals investi-gated (Chinese and Indian, 75% of sample size), thus re-ducing the possible impact of residual confounding dueto cellular heterogeneity. Second, we acknowledge thatin investigating genetic influences on birth weight, ourstudy was not designed to have sufficient power for agenome-wide association study. Indeed, such a studyperformed on child anthropometric outcomes from birthto 48 months of age (data not shown) revealed no singlelocus significant at the commonly used genome-widesignificance threshold (P = 5 × 10–8). However, the ab-sence of any single loci achieving the conventionalgenome-wide significance at 5 × 10–8 was more likely tobe due to a lack of statistical power than caused by alack of genetic influences on birth weight. Therefore, weused a genetic risk profiling approach and genetic vari-ants reported by the GIANT consortium to form a singlecomposite measure/score of genetic risk, and used thisrisk score to investigate genetic influences on birthweight. Third, the GUSTO cohort study was primarilydesigned to obtain extensive prenatal environment mea-sures at mid-pregnancy. Consequently, we were unableto examine trimester-specific effects on the growingfetus. Since late pregnancy weight gain has been linkedto suboptimal metabolic outcomes in offspring, we ana-lysed maternal weight measures in late pregnancy de-rived from medical records (36–41 weeks, N = 803 of987 subjects). As gestational weight gain from pre-pregnancy to mid-pregnancy already showed a signifi-cant association with birth weight, we restricted the latepregnancy analysis to the weight gain between mid-pregnancy and 36–41 weeks. Unlike the gestationalweight gain up to mid-pregnancy, the additional weightgain during late pregnancy did not associate with infantbirth weight (P = 0.12). It is unclear whether the absenceof significance is an indication of trimester-specific ef-fects or a result of low statistical power due to thereduced sample size. Future studies require detailed pre-pregnancy and trimester-specific information to reflectbetter on the temporal influences of prenatal environ-ment on the growing fetus. Lastly, while we have longi-tudinal measures of anthropometry, we do not havelongitudinal measures of methylation in early childhoodand during fetal development, which would be import-ant for determining causality and directionality of the ef-fects. For example, to investigate if DNA methylationmediates effects of the prenatal environment on off-spring adiposity one would need to first establish thetemporality/directionality of the effects, i.e. whether (1)increased child adiposity leads to the alterations in DNAmethylation, or (2) DNA methylation changes lead to in-creased child adiposity. DNA methylation is a possiblemediator in the latter scenario but not the former.Moreover, examining further how DNA methylationlevels at these loci change with age, body size, adiposityduring childhood and environmental exposures duringchildhood will allow for better evaluation of these candi-date loci for stratification and obesity prevention strat-egies. A comparison of methylation measurementscollected in utero, at birth and in early childhood, acrossdifferent tissue types, is an important area of investiga-tion for future studies.Childhood obesity has both immediate and long-termeffects on the health and well-being of an individual.Children who are obese are more likely to become obeseadults [64–66]. In the Bogalusa Heart study [65], child-hood levels of both BMI and triceps skinfolds wereassociated with adult BMI and adiposity. The magni-tudes of these associations increased with childhood age,but were evident from as early as 2 years of age. Over-weight children (age 2– 5 years) with BMI ≥ 95th per-centile had more than four times the risk of becomingoverweight adults compared with children < 50th per-centile. Childhood obesity is also linked with several ad-versities and co-morbidities in the life-course [67]. It canlead to type 2 diabetes, cardiovascular risk and increasedincidence of metabolic syndrome in youth and adults. Itis also associated with earlier pubertal maturation inLin et al. BMC Medicine  (2017) 15:50 Page 15 of 18girls, and early maturing girls tend to have higher BMIsand body fat at the time of menarche [68, 69]. Co-morbidities developed during the life-course in obese chil-dren include bone and joint problems, as well as social andpsychological issues such as stigmatisation and poor self-esteem [70, 71]. A deeper understanding on how differentfactors contribute to adiposity, especially early in life, couldbe useful in troubleshooting the obesity epidemic.ConclusionsDevelopmental pathways to adiposity begin before birthand are influenced by genetic, epigenetic and prenatalenvironment factors. These pathways may have lastingeffects on offspring size, adiposity and metabolic trajec-tory, and have utility in identifying individuals who aresusceptible to obesity and metabolic disease later in life.Additional filesAdditional file 1: Supplementary tables and figures. (PDF 2610 kb)Additional file 2: Compressed folder containing a tab-delimited filewith methylation values for 987 samples, 174,211 CpGs. (GZ 560574 kb)Additional file 3: Information on using methylation data in Additionalfile 2. (PDF 42 kb)AbbreviationsDOHaD: Developmental origins of health and disease; EWAS: Epigenome-wide association studies; FDR: False discovery rate; GA: Gestational age;GUSTO: Growing Up in Singapore Towards Healthy Outcomes;GWG: Gestational weight gain; LD: Linkage disequilibrium;MUFA: Monounsaturated fatty acids; PC: Phosphatidylcholine;PUFA: Polyunsaturated fatty acids; ppBMI: Pre-pregnancy BMI; SD: Standarddeviation; PRS: Polygenic risk score; T2DM: Type 2 diabetes mellitusAcknowledgementsThe 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, Ivy Yee-Man Lau,Bee Wah Lee, Yung Seng Lee, Ngee Lek, Sok Bee Lim, Iliana Magiati, LourdesMary Daniel, Michael Meaney, Cheryl Ngo, Krishnamoorthy Niduvaje, Wei WeiPang, Anqi Qiu, Boon Long Quah, Victor Samuel Rajadurai, Mary Rauff, SalomeA. Rebello, Jenny L. Richmond, Anne Rifkin-Graboi, Seang-Mei Saw, LynettePei-Chi Shek, Allan Sheppard, Borys Shuter, Leher Singh, Shu-E Soh, WalterStunkel, Lin Lin Su, Kok Hian Tan, Oon Hoe Teoh, Mya Thway Tint, HugoP S van Bever, Rob M. van Dam, Inez Bik Yun Wong, P. C. Wong, FabianYap, 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 the Singapore Institute for ClinicalSciences (SICS) – Agency for Science, Technology and Research (A*STAR),Singapore. KMG was supported by the National Institute for Health Researchthrough the NIHR Southampton Biomedical Research Centre and by theEuropean Union’s Seventh Framework Program (FP7/2007-2013), projectEarly Nutrition under grant agreement n°289346.Authors’ contributionsXL, PDG and NK conceived the study. XL and IYL performed the statisticalanalysis. XL, IYL and NK interpreted the results and wrote the manuscript.IMA, SES, MTT, FY, KHT, SMS, MJM, KMG, YSC, JDH, YSL and PDG wereresponsible for the conception and recruitment of the GUSTO cohort. JLM,AMM and MSK generated the Infinium 450 K methylation data. YW, ALT andLC processed the methylation and genotype data. All authors criticallyrevised the manuscript for intellectual and scientific content and approvedthe final manuscript.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 AbbottNutrition, Nestec and Danone. The other authors declare no competinginterests.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 (KKH) and National University Hospital(NUH), which are the Centralized Institute Review Board (CIRB) and theDomain Specific Review Board (DSRB), respectively.Author details1Singapore Institute for Clinical Sciences, A*STAR, 30 Medical Drive,Singapore 117609, Singapore. 2Department of Obstetrics and Gynaecology,Yong Loo Lin School of Medicine, National University of Singapore,Singapore 119228, Singapore. 3Department of Pediatrics, Yong Loo LinSchool of Medicine, National University of Singapore, Singapore 119228,Singapore. 4Centre for Molecular Medicine and Therapeutics, Child andFamily Research Institute, Department of Medical Genetics, University ofBritish Columbia, Vancouver, BC V5Z 4H4, Canada. 5KK Women’s andChildren’s Hospital, Singapore 229899, Singapore. 6Saw Swee Hock School ofPublic Health, National University of Singapore, Singapore 117597, Singapore.7Singapore Eye Research Institute, Singapore 169856, Singapore. 8Duke NUSMedical School, Singapore 169857, Singapore. 9Ludmer Centre forNeuroinformatics and Mental Health, Douglas University Mental HealthInstitute, McGill University, Montreal, Quebec H4H 1R3, Canada. 10MRCLifecourse Epidemiology Unit and NIHR Southampton Biomedical ResearchCentre, University of Southampton and University Hospital SouthamptonNHS Foundation Trust, Southampton SO16 6YD, UK. 11Division of PaediatricEndocrinology and Diabetes, Khoo Teck Puat-National University Children’sMedical Institute, National University Health System, Singapore 119228,Singapore. 12Centre for Human Evolution, Adaptation and Disease, LigginsInstitute, University of Auckland, Auckland 1142, New Zealand. 13Departmentof Biochemistry, Yong Loo Lin School of Medicine, National University ofSingapore, Singapore 119228, Singapore.Received: 6 September 2016 Accepted: 21 January 2017References1. Stein Z, Susser M, Saenger G, Marolla F. Famine and Human Development.The Dutch Hunger Winter of 1944-1945. New York: Oxford University Press; 1975.2. Desai M, Hales CN. Role of fetal and infant growth in programmingmetabolism in later life. Biol Rev. 1997;72:329–48.3. Gluckman PD, Hanson M, Zimmet P, Forrester T. Losing the war againstobesity: the need for a developmental perspective. Sci Transl Med.2011;3(93):93cm19.4. Godfrey KM, Reynolds RM, Prescott SL, Nyirenda M, Jaddoe VW, Eriksson JG,Broekman BF. Influence of maternal obesity on the long-term health ofoffspring. Lancet Diabetes Endocrinol. 2017;5(1):53–64.5. Forsen T, Eriksson JG, Tuomilehto J, Teramo K, Osmond C, Barker DJ.Mother’s weight in pregnancy and coronary heart disease in a cohort ofFinnish men: follow up study. BMJ. 1997;315(7112):837–40.6. Gluckman PD, Hanson MA, Buklijas T, Low FM, Beedle AS. Epigenetic mechanismsthat underpin metabolic and cardiovascular diseases. Nat Rev Endocrinol.2009;5(7):401–8.7. Lawlor DA, Relton C, Sattar N, Nelson SM. Maternal adiposity - a determinantof perinatal and offspring outcomes? Nat Rev Endocrinol. 2012;8(11):679–88.Lin et al. BMC Medicine  (2017) 15:50 Page 16 of 188. Lillycrop KA, Burdge GC. Epigenetic changes in early life and future risk ofobesity. Int J Obes (Lond). 2011;35(1):72–83.9. Fraser HB, Lam LL, Neumann SM, Kobor MS. Population-specificity of humanDNA methylation. Genome Biol. 2012;13(2):R8.10. Gibbs JR, van der Brug MP, Hernandez DG, Traynor BJ, Nalls MA, Lai SL,Arepalli S, Dillman A, Rafferty IP, Troncoso J, et al. Abundant quantitativetrait loci exist for DNA methylation and gene expression in human brain.PLoS Genet. 2010;6(5):e1000952.11. Bell JT, Pai AA, Pickrell JK, Gaffney DJ, Pique-Regi R, Degner JF, Gilad Y,Pritchard JK. DNA methylation patterns associate with genetic and geneexpression variation in HapMap cell lines. Genome Biol. 2011;12(1):R10.12. Tobi EW, Lumey LH, Talens RP, Kremer D, Putter H, Stein AD, Slagboom PE,Heijmans BT. DNA methylation differences after exposure to prenatalfamine are common and timing- and sex-specific. Hum Mol Genet.2009;18(21):4046–53.13. Heijmans BT, Tobi EW, Stein AD, Putter H, Blauw GJ, Susser ES,Slagboom PE, Lumey LH. Persistent epigenetic differences associatedwith prenatal exposure to famine in humans. Proc Natl Acad Sci U S A.2008;105(44):17046–9.14. Teh AL, Pan H, Chen L, Ong ML, Dogra S, Wong J, MacIsaac JL, Mah SM,McEwen LM, Saw SM, et al. The effect of genotype and in uteroenvironment on interindividual variation in neonate DNA methylomes.Genome Res. 2014;24(7):1064–74.15. Rakyan VK, Down TA, Balding DJ, Beck S. Epigenome-wide associationstudies for common human diseases. Nat Rev Genet. 2011;12(8):529–41.16. Hanson MA, Gluckman PD. Early developmental conditioning of laterhealth and disease: physiology or pathophysiology? Physiol Rev.2014;94(4):1027–76.17. Kramer MS. Determinants of low birth weight: methodological assessmentand meta-analysis. Bull World Health Organ. 1987;65(5):663–737.18. Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh PR, Duncan L,Perry JR, Patterson N, Robinson EB, et al. An atlas of genetic correlationsacross human diseases and traits. Nat Genet. 2015;47(11):1236–41.19. Engel SM, Joubert BR, Wu MC, Olshan AF, Haberg SE, Ueland PM, Nystad W,Nilsen RM, Vollset SE, Peddada SD, et al. Neonatal genome-wide methylationpatterns in relation to birth weight in the Norwegian Mother and ChildCohort. Am J Epidemiol. 2014;179(7):834–42.20. Simpkin AJ, Suderman M, Gaunt TR, Lyttleton O, McArdle WL, Ring SM,Tilling K, Davey Smith G, Relton CL. Longitudinal analysis of DNA methylationassociated with birth weight and gestational age. Hum Mol Genet. 2015;24(13):3752–63.21. Sharp GC, Lawlor DA, Richmond RC, Fraser A, Simpkin A, Suderman M,Shihab HA, Lyttleton O, McArdle W, Ring SM, et al. Maternal pre-pregnancyBMI and gestational weight gain, offspring DNA methylation and lateroffspring adiposity: findings from the Avon Longitudinal Study of Parentsand Children. Int J Epidemiol. 2015;44(4):1288–304.22. Haworth KE, Farrell WE, Emes RD, Ismail KM, Carroll WD, Hubball E, Rooney A,Yates AM, Mein C, Fryer AA. Methylation of the FGFR2 gene is associated withhigh birth weight centile in humans. Epigenomics. 2014;6:477–91.23. Godfrey KM, Sheppard A, Gluckman PD, Lillycrop KA, Burdge GC, McLean C,Rodford J, Slater-Jefferies JL, Garratt E, Crozier SR, et al. Epigenetic genepromoter methylation at birth is associated with child’s later adiposity.Diabetes. 2011;60(5):1528–34.24. Dick KJ, Nelson CP, Tsaprouni L, Sandling JK, Aissi D, Wahl S, Meduri E,Morange PE, Gagnon F, Grallert H, et al. DNA methylation and body-massindex: a genome-wide analysis. Lancet. 2014;383(9933):1990–8.25. Pan H, Lin X, Wu Y, Chen L, Teh AL, Soh SE, Lee YS, Tint MT, MacIsaac JL,Morin AM, et al. HIF3A association with adiposity: the story begins beforebirth. Epigenomics. 2015;7(6):937–50.26. 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.Int J Epidemiol. 2014;43(5):1401–9.27. Rolland-Cachera MF. Childhood obesity: current definitions andrecommendations for their use. Int J Pediatr Obes. 2011;6(5-6):325–31.28. Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression.Development of the 10-item Edinburgh Postnatal Depression Scale.Br J Psychiatry. 1987;150:782–6.29. Chee CYI, Lee DTS, Chong YS, Tan LK, Ng TP, Fones CSL. Confinement andother psychosocial factors in perinatal depression: A transcultural study inSingapore. J Affect Disord. 2005;89(1):157–66.30. Chen H, Wang J, Ch’ng YC, Mingoo R, Lee T, Ong J. Identifying motherswith postpartum depression early: integrating perinatal mental health careinto the obstetric setting. ISRN Obstet Gynecol. 2011;2011:309189.31. Spielberger CD, Gorsuch RL, Lushene R, Vagg PR, Jacobs GA. Manual for theState-Trait Anxiety Inventory. Palo Alto: Consulting Psychologists Press; 1983.32. Chong S-C, Broekman BF, Qiu A, Aris IM, Chan YH, Rifkin-Graboi A, Law E,Chee CYI, Chong Y-S, Kwek KYC, et al. Anxiety and depression duringpregnancy and temperament in early infancy: findings from a multi-ethnic,Asian prospective birth cohort study. Infant Mental Health J. 2016;37(5):584–98.33. 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 ofthe Infinium Human Methylation450 BeadChip Array. Epigenetics.2012;7(10):1173–87.34. Johnson WE, Rabinovic A, Li C. Adjusting batch effects in microarrayexpression data using Empirical Bayes methods. Biostatistics. 2007;8(1):118–27.35. 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(2):203–9.36. 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(1):4.37. Daca-Roszak P, Pfeifer A, Zebracka-Gala J, Rusinek D, Szybinska A, Jarzab B,Witt M, Zietkiewicz E. Impact of SNPs on methylation readouts by IlluminaInfinium HumanMethylation450 BeadChip Array: implications forcomparative population studies. BMC Genomics. 2015;16:1003.38. Xu Z, Niu L, Li L, Taylor JA. ENmix: a novel background correctionmethod for Illumina HumanMethylation450 BeadChip. Nucleic Acids Res.2016;44(3):e20.39. Gelman A. Scaling regression inputs by dividing by two standard deviations.Stat Med. 2008;27(15):2865–73.40. Locke AE, Kahali B, Berndt SI, Justice AE, Pers TH, Day FR, Powell C,Vedantam S, Buchkovich ML, Yang J, et al. Genetic studies of body massindex yield new insights for obesity biology. Nature. 2015;518(7538):197–206.41. 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 Bioinformatics. 2012;13:86.42. 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:e00523.43. 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(6):882–3.44. Leek JT, Scharpf RB, Bravo HC, Simcha D, Langmead B, Johnson WE,Geman D, Baggerly K, Irizarry RA. Tackling the widespread and criticalimpact of batch effects in high-throughput data. Nat Rev Genet.2010;11(10):733–9.45. Leek JT, Storey JD. Capturing heterogeneity in gene expression studies bysurrogate variable analysis. PLoS Genet. 2007;3(9):1724–35.46. Agha G, Houseman EA, Kelsey KT, Eaton CB, Buka SL, Loucks EB. Adiposity isassociated with DNA methylation profile in adipose tissue. Int J Epidemiol.2014;44(4):1277–87.47. Yoav B, Yosef H. Controlling the false discovery rate: a practical andpowerful approach to multiple testing. J Royal Stat Soc Series B.1995;57(1):289–300.48. Benjamini Y, Yekutieli D. The control of the false discovery rate in multipletesting under dependency. Annals Statistics. 2001;29(4):1165–88.49. Clarke S, Hall P. Robustness of multiple testing procedures againstdependence. Annals Statistics. 2009;37(1):332–58.50. Lin X, Aris IM, Tint MT, Soh SE, Godfrey KM, Yeo GS, Kwek K, Chan JK,Gluckman PD, Chong YS, et al. Ethnic differences in effects of maternalpre-pregnancy and pregnancy adiposity on offspring size and adiposity.J Clin Endocrinol Metab. 2015;100(10):3641–50.51. Aris IM, Soh SE, Tint MT, Saw SM, Rajadurai VS, Godfrey KM, Gluckman PD,Yap F, Chong YS, Lee YS. Associations of gestational glycemia andprepregnancy adiposity with offspring growth and adiposity in an Asianpopulation. Am J Clin Nutr. 2015;102(5):1104–12.52. Chawla R, Badon SE, Rangarajan J, Reisetter AC, Armstrong LL, Lowe LP,Urbanek M, Metzger BE, Hayes MG, Scholtens DM, et al. Genetic risk scoreLin et al. BMC Medicine  (2017) 15:50 Page 17 of 18for prediction of newborn adiposity and large-for-gestational-age birth.J Clin Endocrinol Metab. 2014;99(11):E2377–86.53. Lorenzo DN, Healy JA, Hostettler J, Davis J, Yang J, Wang C, Hohmeier HE,Zhang M, Bennett V. Ankyrin-B metabolic syndrome combines age-dependent adiposity with pancreatic beta cell insufficiency. J Clin Invest.2015;125(8):3087–102.54. Svensson PA, Wahlstrand B, Olsson M, Froguel P, Falchi M, Bergman RN,McTernan PG, Hedner T, Carlsson LM, Jacobson P. CDKN2B expression andsubcutaneous adipose tissue expandability: possible influence of the 9p21atherosclerosis locus. Biochem Biophys Res Commun. 2014;446(4):1126–31.55. Hannou SA, Wouters K, Paumelle R, Staels B. Functional genomics of theCDKN2A/B locus in cardiovascular and metabolic disease: what have welearned from GWASs? Trends Endocrinol Metab. 2015;26(4):176–84.56. Uebele VN, Gotter AL, Nuss CE, Kraus RL, Doran SM, Garson SL, Reiss DR,Li Y, Barrow JC, Reger TS, et al. Antagonism of T-type calcium channelsinhibits high-fat diet-induced weight gain in mice. J Clin Invest.2009;119(6):1659–67.57. McGregor RA, Choi MS. microRNAs in the regulation of adipogenesis andobesity. Curr Mol Med. 2011;11(4):304–16.58. Sun T, Fu M, Bookout AL, Kliewer SA, Mangelsdorf DJ. MicroRNA let-7regulates 3 T3-L1 adipogenesis. Mol Endocrinol. 2009;23(6):925–31.59. Deiuliis JA. MicroRNAs as regulators of metabolic disease: pathophysiologicsignificance and emerging role as biomarkers and therapeutics. Int J Obes.2016;40(1):88–101.60. McCann SE, Liu S, Wang D, Shen J, Hu Q, Hong CC, Newman VA, Zhao H.Reduction of dietary glycaemic load modifies the expression of microRNApotentially associated with energy balance and cancer pathways in pre-menopausalwomen. Br J Nutr. 2013;109(4):585–92.61. Ogden CL, Carroll MD, Fryar CD, Flegal KM. Prevalence of Obesity AmongAdults and Youth: United States, 2011–2014. Hyattsville: National Center forHealth Statistics; 2015.62. Hayward CJ, Fradette J, Galbraith T, Remy M, Guignard R, Gauvin R,Germain L, Auger FA. Harvesting the potential of the human umbilicalcord: isolation and characterisation of four cell types for tissueengineering applications. Cells Tissues Organs. 2013;197(1):37–54.63. Lim IJ, Phan TT. Epithelial and mesenchymal stem cells from the umbilicalcord lining membrane. Cell Transplant. 2014;23(4-5):497–503.64. Guo SS, Chumlea WC. Tracking of body mass index in children in relation tooverweight in adulthood. Am J Clin Nutr. 1999;70(1):145S–8S.65. Freedman DS, Khan LK, Serdula MK, Dietz WH, Srinivasan SR, Berenson GS.The relation of childhood BMI to adult adiposity: the Bogalusa Heart Study.Pediatrics. 2005;115(1):22–7.66. Deshmukh-Taskar P, Nicklas TA, Morales M, Yang SJ, Zakeri I, Berenson GS.Tracking of overweight status from childhood to young adulthood: theBogalusa Heart Study. Eur J Clin Nutr. 2006;60(1):48–57.67. Biro FM, Wien M. Childhood obesity and adult morbidities. Am J Clin Nutr.2010;91(5):1499S–505S.68. Kaplowitz PB, Slora EJ, Wasserman RC, Pedlow SE, Herman-Giddens ME.Earlier onset of puberty in girls: relation to increased body mass index andrace. Pediatrics. 2001;108(2):347–53.69. Adair LS, Gordon-Larsen P. Maturational timing and overweight prevalencein US adolescent girls. Am J Public Health. 2001;91(4):642–4.70. Daniels SR, Arnett DK, Eckel RH, Gidding SS, Hayman LL, Kumanyika S,Robinson TN, Scott BJ, St Jeor S, Williams CL. Overweight in children andadolescents: pathophysiology, consequences, prevention, and treatment.Circulation. 2005;111(15):1999–2012.71. Dietz WH. Overweight in childhood and adolescence. N Engl J Med.2004;350(9):855–7.•  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) 15:50 Page 18 of 18


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