Placental DNA Methylation Adaptation to MaternalGlycemic Response in PregnancyAndres Cardenas,1 Valerie Gagné-Ouellet,2 Catherine Allard,2 Diane Brisson,3 Patrice Perron,2

Luigi Bouchard,2 and Marie-France Hivert1,2,4

Diabetes 2018;67:1673–1683 | https://doi.org/10.2337/db18-0123

Maternal hyperglycemia during pregnancy is associatedwith excess fetal growth and adverse perinatal and de-velopmental outcomes. Placental epigenetic maladapta-tion may underlie these associations. We performed anepigenome-wide association study (>850,000 CpG sites)of term placentas and prenatal maternal glycemic re-sponse 2-h post oral glucose challenge at 24–30 weeksof gestation among 448mother-infant pairs.Maternal 2-hglycemia postload was strongly associated with lowerDNA methylation of four CpG sites (false discovery rate[FDR] q <0.05) within the phosphodiesterase 4B gene(PDE4B). Additionally, three other individual CpG siteswere differentially methylated relative to maternal glu-cose response within the TNFRSF1B, LDLR, and BLMgenes (FDR q <0.05). DNA methylation correlated withexpression of its respective genes in placental tissue atthree out of four independent identified loci: PDE4B (r =0.31,P < 0.01), TNFRSF1B (r =20.24,P = 0.013), and LDLR(r = 0.32, P < 0.001). In an independent replication cohort(N = 65–108 samples), resultswere consistent in directionbut not significantly replicated among testedCpG sites inPDE4B and TNFRSF1B. Our study provides evidence thatmaternal glycemic response during pregnancy is asso-ciated with placental DNA methylation of key inflamma-tory genes whose expression levels are partially underepigenetic control.

Prenatal nutritional, behavioral, and environmental con-ditions play a key role in fetal development by modulatingthe intrauterine environment, fetal nutrient availability,and growth. The Pedersen hypothesis states that prenatal

maternal glucose crosses the placenta and leads to in-trauterine hyperglycemia affecting fetal growth anddevelopment (1).

It is now well established that maternal hyperglycemiain pregnancy is associated with adverse maternal and birthoutcomes. For example, in the large prospective interna-tional multicenter Hyperglycemia and Adverse PregnancyOutcome (HAPO) study, robust linear associations werefound between greater maternal prenatal glucose levelsand higher birth weight or cord blood C-peptide levelsacross the whole spectrum of maternal glycemia, with nolower threshold (2). In the HAPO study, measures ofmaternal prenatal glycemia, both fasting and post oralglucose load, were linearly associated with adverse out-comes in mothers and in infants at birth (2). Of these,maternal glucose levels 2-h post 75-g glucose load showedslightly stronger associations with neonatal hypoglyce-mia (3) and with an increased risk of abnormal glucosetolerance in offspring at 7 years of age (4), suggestingthat maternal response to glucose loading in pregnancycontributes to offspring metabolic programming atbirth and later in life. Indeed, emerging evidence sug-gests that maternal hyperglycemia could have lastinghealth consequences due to metabolic programmingthat occurs during key fetal developmental stages andmay act through epigenetic programming and molecularmechanisms (5).

It is now recognized that the placenta is not simplya transferring organ for nutrients from the mother to fetusbut participates actively in maternal metabolism and likelycontributes to fetal programming. The placenta is the

1Division of Chronic Disease Research Across the Lifecourse, Department ofPopulation Medicine, Harvard Medical School and Harvard Pilgrim Health CareInstitute, Boston, MA2Centre de Recherche du Centre Hospitalier Universitaire de Sherbrooke, Sher-brooke, Quebec, Canada3Lipidology Unit, Community Genomic Medicine Centre and ECOGENE-21, De-partment of Medicine, Université de Montréal, Saguenay, Quebec, Canada4Diabetes Unit, Massachusetts General Hospital, Boston, MA

Corresponding author: Marie-France Hivert, mhivert@partners.org.

Received 26 January 2018 and accepted 4 May 2018.

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db18-0123/-/DC1.

© 2018 by the American Diabetes Association. Readers may use this article aslong as the work is properly cited, the use is educational and not for profit, and thework is not altered. More information is available at http://www.diabetesjournals.org/content/license.

Diabetes Volume 67, August 2018 1673

GENETIC

S/G

ENOMES/P

ROTEOMIC

S/M

ETABOLOMIC

S

master regulator of the fetal environment and is directlyresponsible for maternal-fetal nutrient and waste trans-port as well as hormone synthesis (6). Fetal gluconeogen-esis is relatively minimal, and thus the fetus relies heavilyon circulating maternal glucose transported into the fetalside by facilitated placental diffusion through membersof the glucose transporter (GLUT) family (6). Over 95% ofthe fetal glucose levels are estimated to be derived frommaternal plasma levels by diffusion through the placenta(7,8), and whenmaternal hyperglycemia occurs, this excessmaternal glucose is passed toward the fetus.

The placenta has been shown to adapt to nutrientavailability through epigenomic modifications in responseto gestational diabetes mellitus (9–11). Nevertheless, thesestudies reported epigenetic modifications with cases ofgestational diabetes mellitus and thus could not determinewhether the observed patterns were consequences of ges-tational diabetes mellitus (or of its treatments) or actualplacental adaptations to hyperglycemia in pregnancy. Toaddress this gap of the role of placental adaptation tomaternal glucose levels during fetal development, we con-ducted an epigenome-wide association study (EWAS) formaternal 2-h glucose post oral glucose tolerance test(OGTT) administered during the second trimester of preg-nancy and DNA methylation of full-term placenta tissue.We measured genome-wide DNA methylation of the fetalplacenta side among 448 mother-infant pairs and observedstrong associations for several inflammatory-relevant genes.Furthermore, we confirmed the functional role of DNAmethylation at the discovered CpG loci by quantifying geneexpression in placental tissue. Finally, we tested for externalreplication of our top differentially methylated loci in anindependent birth cohort.

RESEARCH DESIGN AND METHODS

Study PopulationSamples and study participants were selected from theGenetics of Glucose regulation in Gestation and Growth(Gen3G), a prospective Canadian prebirth cohort. Gen3Gwas designed to elucidate the biological, environmental,genetic, and epigenetic determinants of glucose regulationduring pregnancy and the impact on offspring develop-ment (12). Briefly, we recruited expecting mothers dur-ing the first trimester of their pregnancy, and we enrolledpregnant women in the study if they were at least 18 yearsof age or older with a singleton pregnancy and did nothave prepregnancy diabetes based on medical history andscreening during the first trimester blood sampling. Forthis study, mother-infant pairs were selected from thelarger cohort if they had placental tissue available for DNAisolation as well as .37 weeks of gestation at delivery.Study participants provided written informed consentprior to enrollment in accordance with the Declarationof Helsinki. All study protocols were approved by the ethicsreview board from the Centre Hospitalier Universitaire deSherbrooke.

Placental Tissue DNA and RNA ExtractionTrained research personnel collected fetal placenta tissuesamples immediately after delivery (,30 min postpar-tum). A 1 cm3 placenta tissue sample was collected ap-proximately 5 cm from the umbilical cord insertion, fromthe fetal side of the placenta for each delivery. Placentasamples were collected by trained study staff and stored inRNAlater (Qiagen) at280°C until DNA or RNA extractionoccurred. We purified DNA and RNA from the placentasamples using the AllPrep DNA/RNA/Protein Mini Kit(Qiagen). Purity of extracted DNA was evaluated usinga Spectrophotometer (Ultrospec 2000 UV/Visible; PharmaciaBiotech) with an absorbance ratio set at 260–280 nm asrecommended (13).

OGTTDuring the first trimester, enrolled participants completeda nonfasting 50-g glucose challenge test (GCT), and wemeasured glucose 1 h after the glucose load (to screen forpreexisting diabetes). During the second trimester clinicalvisit, all women performed a fasting 75-g OGTT, and wemeasured maternal glucose levels at fasting prior to the oralglucose challenge and at 1 h and 2 h post glucose challenge.Maternal blood glucose concentrations were all measured atthe Centre Hospitalier Universitaire de Sherbrooke centrallaboratory.

Epigenome-Wide DNA Methylation MeasurementsEpigenome-wide DNA methylation measurements wereperformed on DNA from placenta samples using bisulfiteconversion followed by quantification using the InfiniumMethylationEPIC BeadChip (Illumina, San Diego, CA) thatmeasures over 850,000 CpG sites at a single nucleotideresolution. Samples were randomly allocated to differentplates and chips to minimize confounding. Methylationdata were imported into R for preprocessing using minfi(14). We performed quality control at the sample level,excluding samples that failed (n = 8), mismatch on geno-type (n = 12) or sex (n = 1), and technical duplicates (n =10). A total of 448 high-quality samples were retained forsubsequent analyses. We performed quality control onindividual probes by computing a detection P value andexcluded 2,003 probes with nonsignificant detection (P .0.05) for 5% or more of the samples. We also excluded19,129 probes annotated to sex chromosomes, 2,836non-CpG probes, 5,552 single nucleotide polymorphism–associated probes at the single base extension with a minorallele frequency of $5%, and 4,453 probes with a singlenucleotide polymorphism at the target CpG site witha minor allele frequency of $5%. Finally, we excluded40,448 cross-reactive probes previously identified (15). Atotal of 791,131 CpG sites were included in the finalanalyses. We processed our data using functional normal-ization with the default of two principal components fromcontrol probes (16). We also adjusted for probe-type biasusing RCP, a regression method approach that uses geno-mic proximity to adjust the distribution of type 2 probes

1674 Placental Epigenetic Adaption Diabetes Volume 67, August 2018

(17). Last, we used the ComBat function from the svapackage to adjust for sample plate (18). We visualized thedata using density distributions at all processing stepsand performed principal component analyses to examinethe association of both technical and biological variables.We logit transformed the b-values to M-values for sta-tistical analyses (19). However, we report effect esti-mated and summary statistics on the b-value scale to easeinterpretability.

RNA Quantification of Top Differentially MethylatedLociWe selected a simple random sample of 104 participantswith DNA methylation data for RNA quantification. RNAconcentrations and RNA Integrity Number (RIN) wereassessed using Agilent 2100 Bioanalyzer and the AgilentRNA 6000 Nano Kit (Agilent Technologies). Complemen-tary DNA of placental RNA was generated using a ran-dom primer hexamer (High Capacity cDNA RT; AppliedBiosystems). Amplicons were generated in duplicate in20 mL with 10 mL of TaqMan Universial PCR Master Mix(Applied Biosystems). RNA expression for genes annotatedto the differentially methylated loci significantly associatedwith 2-h maternal glucose (PDE4B: Hs00277080_m1,TNFRSF1B: Hs00961748_m1, BLM: Hs00172060_m1,LDLR: Hs00181192_m1; Applied Biosystems) were mea-sured with quantitative real-time PCR (qRT-PCR) using7500 Real Time PCR system (Thermo Fisher Scientific).Expression levels are reported as a Ct ratio with respect to thereference gene, the tyrosine 3-monooxygenase/tryptophan5-monooxygenase activation protein z (YWHAZ: Hs01122445_g1; Applied Biosystems) previously shown to be stable inhuman placenta (20). We report 1/ratio values (to ease inter-pretation as a higher 1/ratio value reflect higher expression)for correlation analyses with DNA methylation.

Replication Cohort and Pyrosequencing: ECO-21The independent replication cohort (ECO-21) consisted ofmother-infant pairs recruited in the first trimester ofpregnancy, from a French Canadian population from theSaguenay-Lac-Saint-Jean region of Québec (Saguenay cityarea, Québec, Canada) (21). Pregnant women were ex-cluded if they were,18 or.40 years of age, had a historyof alcohol/drug abuse in pregnancy, or had diagnosedfamilial hypercholesterolemia, pregestational diabetes, orother prepregnancy disorders impairing glucose homeo-stasis. For this analysis, we excluded women who weretreated for gestational diabetes mellitus, to avoid con-founding by treatment. The Chicoutimi Hospital ethicscommittee reviewed and approved this project. Womenprovided written informed consent prior to enroll-ment in the study in accordance with the Declaration ofHelsinki.

In the replication cohort, DNAmethylation levels of thetop CpG site annotated to the TNFRSF1B gene (cg26189983)as well as three other CpG sites (cg07734160, cg13866577,and cg03442467) within the PDE4B region were measured

using pyrosequencing (PyroMark Q24; Qiagen). Althoughwe also intended to pyrosequence our other top loci,our primers failed. To pyrosequence TNFRSF1B and PDE4BCpG sites, DNA underwent a sodium bisulfite (NaBis)treatment (EpiTect Bisulfite Kits; Qiagen). Target NaBis-DNA loci were PCR-amplified with specific primers designedusing the PyroMark Assay Design software (version2.0.1.15; Qiagen) and were then pyrosequenced. Pyrose-quencing runs performed included a negative PCR andsodium bisulfite conversion controls. Additionally, pyrose-quencing quality control was assessed for each sample, asrecommended by the manufacturer, using PyroMark Q24Analysis Software (v1.0.10.134).

Statistical AnalysesFor the 448 participants eligible for analyses, we report oursample characteristics using means, SD, or proportions.We performed CpG-by-CpG analyses by fitting robustlinear regression models for each site adjusted for covar-iates with DNA methylation as the response variable onthe M-value scale using maternal 2-h glucose levels postOGTT as the main predictor. Robust linear regression wasused to protect against potential heteroskedasticity (22).To control for cell-type heterogeneity, we used the top10 components from ReFACTor, a reference-free methodthat adjusts for cell-type mixture in heterogenous tissues(23). Robust linear regression models were adjusted formaternal age in years, BMI, parity, smoking during preg-nancy, gestational age at birth, sex, and the first 10 prin-cipal components estimated from ReFACTor as proxyfor placenta cellular heterogeneity. CpG-by-CpG analyseswere controlled for the false discovery rate (FDR) at 5%(q,0.05). Quantile-quantile plots of P values were used toinspect genomic inflation. In epigenome-wide associationanalyses, bioinformatic adjustment reduced the geno-mic inflation (l) from 1.39 to 1.01 (Supplementary Fig.1). Regional and genome-wide Manhattan plots wereused to report results from epigenome-wide associationanalysis.

We calculated Pearson correlation coefficients to esti-mate the association between DNA methylation and geneexpression among a randomly selected sample of 104 par-ticipants for top differentially methylated placenta genesassociated with maternal glucose response (FDR q,0.05).Significant statistical correlations between DNA methyla-tion and gene expression were considered with P , 0.05.In additional analyses, we evaluated associations adjustingfor the same covariates among top differentially methylatedCpG sites and maternal 1-h glucose levels post 50-g GCTduring the first trimester as well as baseline fasting glucoselevels and 1-h glucose levels post 75-g OGTT at secondtrimester per SD of each glycemic trait to allow compara-bility between traits. We performed these additional anal-yses with other measures of glycemia taken in the firsttrimester or during OGTT to ensure robustness of resultsand to assess the specificity of our associations with regardto postchallenge glucose response.

diabetes.diabetesjournals.org Cardenas and Associates 1675

RESULTS

Participant CharacteristicsA total of 448 mother-infant pairs were eligible for anal-yses with complete DNA methylation and 2-h glucoselevels post OGTT. All mothers were Caucasian, 49.3%were primiparous, with a mean 6 SD age of 28.2 6 4.3years and BMI of 25.45 6 5.7 kg/m2, and 90.2% reportedto be nonsmokers during pregnancy. Mean 6 SD gesta-tional age at birth was 39.5 6 1.0 weeks, mean birthweight was 3,4486 428 g, and 52.7% of births were males.At enrollment and by design of the study, none of thepregnant women had pregestational diabetes evaluated byfirst trimester glycemic testing (GCT and HbA1c). Atthe second trimester 75-gOGTT,mean6 SD fasting glucoselevel was 4.20 6 0.37 mmol/L, 1-h glucose was 7.11 61.61 mmol/L, and 2-h glucose was 5.80 6 1.33 mmol/L.Additional participants characteristics are summarized inTable 1. Concentrations for 2-h glucose were approximatelynormally distributed (Supplementary Fig. 2).

Association of Maternal 2-Hour Glucose Levels WithDNA Methylation of PlacentaIn adjusted CpG-by-CpG analyses, we found seven CpGsites of placental DNA significantly associated (FDRq ,0.05) with maternal 2-h glucose levels post 75-gOGTT (Table 2). Epigenome-wide results are shown inFig. 1. We observed lower placental DNA methylation offour CpG sites annotated to PDE4B (phosphodiesterase 4Bgene) in response to higher maternal 2-h glucose levels.Specifically, a 1 mmol/L greater 2-h glucose level wasassociated with a 1.16%, 0.88%, 1.86%, and 0.58% lowerDNA methylation at cg07734160 (P = 1.20 3 1029),

cg13866577 (P = 1.11 3 1027), cg03442467 (P = 2.84 31027), and cg13349623 (P = 2.06 3 1029), respectively.The four CpG sites were annotated within the transcrip-tion start site (1,500–200 bps), first exon, or body ofPDE4B and surrounded by another four sites just belowour FDR statistical threshold (Fig. 2). Scatter plots forthe association between DNA methylation and 2-h glu-cose levels are shown in Fig. 3. In addition, a 1 mmol/Lgreater maternal 2-h glucose level was associated with1.22% (P = 1.70 3 1027) higher DNA methylation ofcg26189983 annotated to TNFRSF1B/TNFR2 (tumor ne-crosis factor receptor superfamily member 1B) and lo-cated within the gene body. We also observed an inverseassociation for cg20254265 annotated to the exon bound-ary of BLM (bloom syndrome RecQ like helicase gene) witha 1 mmol/L greater 2-h glucose level associated witha 0.63% lower DNA methylation (P = 7.583 1028). Last,we observed 0.27% lower DNA methylation at cg08483713per 1 mmol/L greater 2-h glucose level, and this CpG site wasannotated to the gene body of LDLR (low density lipoproteinreceptor).

Association of Additional Maternal Glucose MeasuresWith CpG Sites Associated With 2-Hour GlucoseAs additional analyses, we tested associations with 1-hglucose levels measured post 50-g GCT performed at firsttrimester as well as fasting glucose and 1-h glucose levelspost 75-g OGTT measured during the second trimester perSD increase for each glycemic trait to allow comparability(Table 3). Greater postload maternal glucose at either first(1-h glucose level post 50-g GCT) or second trimester (1-hglucose level post 75-g OGTT) was associated with lowerDNA methylation at previously discovered CpG sites inPDE4B, BLM, and LDLR genes (P, 0.05), consistent in thedirection of association found in discovery associations with2-h glucose levels at second trimester but with relativelysmaller effect sizes. Associations between maternal fast-ing glucose at second trimester and DNA methylation atCpG sites in PDE4B were in the same direction as initialfindings in the 2-h glucose analyses but were not signif-icant. Associations among other maternal glucose measuresand the TNFRSF1B CpG site were less consistent.

DNA Methylation and Gene Expression in PlacentaWe investigated potential functional expression adapta-tions of genes identified by our epigenetic investigationsusing placenta samples from 104 randomly selected par-ticipants. We observed that greater DNA methylation atcg03442467 (gene body and TSS200) within PDE4B wassignificantly correlated with higher PDE4B expression inplacental tissue (r = 0.31; P = 1.81 3 1023) (Fig. 4). Theother CpG sites in PDE4B were positively correlated butnot statically significant (Supplementary Fig. 3). GreaterDNA methylation of TNFRSF1B at cg26189983 (gene body)was negatively correlated with expression (r = –0.24; P =0.013), whereas greater DNA methylation at the LDLRcg08483713 (gene body) site was associated with greater

Table 1—Participant characteristics from the Gen3Gprospective cohort (N = 448)

Maternal age (years) 28.2 6 4.3

BMI (kg/m2) 25.45 6 5.7

ParityPrimiparous (%) 221 (49.3)

EthnicityCaucasian 448 (100)

Maternal smoking during pregnancyNo 404 (90.2)Yes 39 (8.7)Unknown 5 (1.1)

First trimester GCT1-h glucose (mmol/L) 5.55 6 1.41

Second trimester OGTTFasting glucose (mmol/L) 4.20 6 0.381-h glucose (mmol/L) 7.11 6 1.612-h glucose (mmol/L) 5.80 6 1.33

Child sexMale (%) 236 (52.7)

Gestational age at birth (weeks) 39.5 6 1.0

Birth weight (g) 3,448 6 428

Data are mean 6 SD or n (%).

1676 Placental Epigenetic Adaption Diabetes Volume 67, August 2018

expression (r = 0.32, P = 9.623 1024) in placenta tissue. Noassociation was observed between BLM DNA methylationand expression at the site (Fig. 4). There was a strong positivecorrelation between placental PDE4B and TNFRSF1B expres-sion (r = 0.82, P, 2.203 10216), and the expression levels ofall other genes were also positively associated (Supplemen-tary Fig. 4).

Replication of Results in an Independent Study Cohort:ECO-21In the external independent birth cohort with a smallersample size (N = 65–108), we sought to replicate of ourtop differentially methylated site (cg26189983) at theTNFRSF1B gene and three CpG sites (cg07734160,cg13866577, cg03442467) within the PDE4B region. Studycharacteristics for the replication cohort are described inSupplementary Table 1. In this external replication cohort,estimated adjusted associations were consistent in direc-tion of association but did not reach statistical significance(Supplementary Table 2). However, we had relatively low

statistical power and would have required at least 200 sam-ples to achieve adequate power for our strongest signal andmain finding at PDE4B (Supplementary Table 3).

DISCUSSION

In this prepregnancy birth cohort, we conducted thelargest EWAS of placenta for maternal prenatal glucoseresponse during pregnancy thus far. We found CpG sitesfor which DNA methylation levels were associated withmaternal glucose response 2-h post 75 g glucose loading,performed during the second trimester of pregnancy.Notably, there was evidence that multiple CpG sites ina genomic region of PDE4B were strongly and inverselyassociated with higher maternal 2-h glucose levels. Fur-thermore, greater methylation levels at identified CpGsites in the PDE4B locus correlated with higher gene ex-pression of PDE4B in placental tissue, supporting thefunctional role of these placental epigenetic markers influ-enced by maternal glucose response in pregnancy. Other

Figure 1—Manhattan plot for the EWAS of maternal 2-h glucose levels post OGTT with DNA methylation in placenta (solid line: Bonferronithreshold; dashed line: FDR q ,0.05).

Table 2—Adjusted differences in DNA methylation associated with a 1 mmol/L increase in prenatal 2-h glucose levels post OGTT

CpG

DNAmethylation (%),mean 6 SD Chromosome Position Gene Gene group

Percentdifferencein DNA

methylation 95% CI P

cg26189983 49.7 6 8.2 chr1 12227700 TNFRSF1B Body 1.22 (0.80, 1.7) 1.70 3 1027

cg07734160 13.0 6 7.5 chr1 66797378 PDE4B TSS1500, body 21.16 (21.5, 20.8) 1.20 3 1029

cg13866577 9.4 6 6.6 chr1 66797481 PDE4B Body, TSS1500 20.88 (21.2, 20.6) 1.11 3 1027

cg03442467 24.6 6 13.1 chr1 66797701 PDE4B Body, TSS200 21.86 (22.6, 21.2) 2.84 3 1027

cg13349623 5.8 6 3.6 chr1 66798221 PDE4B First exon, body 20.58 (20.8, 20.4) 2.06 3 1029

cg20254265 79.6 6 9.1 chr15 91306178 BLM ExonBnd, body 20.63 (20.9, 20.4) 7.58 3 1028

cg08483713 5.86 6 2.9 chr19 11241669 LDLR Body 20.27 (20.4, 20.2) 1.39 3 1026

Body, between the ATG and stop codon, irrespective of the presence of introns, exons, TSS, or promoters; ExonBnd, within 20 basesof an exon boundary, i.e., the start or end of an exon; TSS200, 0–200 bases upstream of the transcriptional start site (TSS); TSS1500,200–1,500 bases upstream of the TSS.

diabetes.diabetesjournals.org Cardenas and Associates 1677

discovered epigenomic loci mapped to three additionalgenes—TNFRSF1B, BLM, and LDLR—observed to be associ-ated with maternal 2-h glucose, of which DNA methylationat TNFRSF1B and LDLR loci correlated with expression ofthe respective gene in placental tissue. Our study providesevidence that maternal glucose response postchallenge inmidpregnancy is associated with differential methylationof genes within the placenta at birth and that these loci arepartially under epigenetic control for gene expression. Ourresults highlight the ability of the placenta to epigeneticallyadapt to the maternal nutritional environment that mightplay a functional role in metabolic programming of theoffspring.

PDE4B is a member of the cyclic nucleotide phospho-diesterases family responsible for the hydrolysis of cyclicAMP and GMP (24). The PDE4 family of enzymes catalyzesthe hydrolysis of second messenger cyclic AMP, a keysignaling molecule for immune response regulation (25).Inhibition of PDE4 decreases secretion of tumor necrosisfactor-a (TNF-a) (26), a potent proinflammatory cytokine(27). Specifically, the PDE4B isoform has been shown topredominantly mediate TNF-a release (26). Indeed, PDE4inhibition in mice has been shown to block intrauter-ine inflammation, decrease cytokine production, and de-lay preterm birth (28). For instance, in an experimentalmouse study, intrauterine injection with Escherichia coli

Figure 2—Regional Manhattan plot and correlation heat map for CpG sites near the differentially methylated loci in the PDE4B gene:associations with maternal 2-h glucose levels post OGTT.

1678 Placental Epigenetic Adaption Diabetes Volume 67, August 2018

LPS increased expression of PDE4B, eliciting an inflamma-tory response and triggering preterm delivery (29): inthis mouse study, investigators also showed that PDE4Binhibition blocked intrauterine inflammatory responseand prevented preterm delivery. Remarkably, this inflam-matory response was localized in glycogen trophoblast cellsof the placenta from the fetal compartment, suggestinga direct role of the placenta (29). The relationship betweenPDE4 inhibition and the proinflammatory cytokine TNF-a is of high interest, as TNF-a levels have been observedto be increased in adipose and placental tissue of obesepregnant women compared with nonobese pregnantwomen (30). Circulating levels of TNF-a are also a strongindependent predictor of insulin resistance in pregnancy,with placenta tissue being a primary contributor to maternalTNF-a levels (31). The association between gestationalinsulin resistance and TNF-a levels in pregnancy hasalso been previously demonstrated in our cohort (32).

There is emerging evidence that suggests that PDE4Bplays an important role in adiposity and metabolic func-tion. For example, PDE4B-null mice have been shown to beleaner, with lower fat pad weights, smaller adipocytes, anddecreased serum leptin levels compared with wild-typelittermates (33). Treatment with a PDE4 inhibitor reducedthe body weight of mice fed a Western-type diet mediatedby an increase in energy expenditure and PDE4B mRNA inwhite adipose tissue (34). Furthermore, chronic treatmentwith PDE4 inhibitors was shown to delay the progressionof diabetes in an experimental animal model for obesity,diabetes, and metabolic syndrome (db/db mice) (35). Inaddition, a randomized controlled trial of newly diagnosedpatients with type 2 diabetes demonstrated that treatmentwith a PDE4 inhibitor (roflumilast) successfully loweredglucose levels (36). Therefore, PDE inhibitors have beenproposed as a potential therapeutic agent for diabetes and

metabolic syndrome (37). Our results add to the growingbody of evidence suggesting that PDE4B plays a functionalrole in metabolic programming.

Higher DNA methylation of TNFRSF1B loci was asso-ciated with greater 2-h glucose levels, and a weaker positiveassociation was also observed with 1-h glucose levelspostload during the second trimester of pregnancy.TNFRSF1B encodes a high-affinity receptor for TNF-a.TNF-a is linked to metabolism and insulin sensitivity inhuman tissues as well as in experimental studies and ge-netically linked to hyperlipidemia (38,39). Expression ofTNFRSF1B, also known as TNFR2, has been detected inplacental trophoblasts and distributed across the cytosolwith multiple functions such as apoptosis, inhibition oftrophoblast cell fusion, and invasion or epithelial shedding(40). In an animal model, pharmacological attenuation ofTNF-a signaling with soluble TNFR2-IgG (etanercept)protected the placenta from deformities due to infection.In obese adult individuals, prolonged treatment withetanercept improved fasting glucose and adiponectin levels(41). Furthermore, among individuals with normal glu-cose levels, plasma soluble TNFR2 (sTNFR2) was nega-tively associated with insulin sensitivity (42), while highersTNFR2 has been observed in offspring of subjects withtype 2 diabetes (43). In line with these findings, plasmasTNFR2 concentration has been proposed to serve asa marker of TNF-a–related insulin resistance (44), butit is still unclear whether circulating sTNFR2 acts asa buffer in response to higher TNFa or also participatesin the inflammatory process. In our findings, higher ma-ternal 2-h glucose is associated with greater DNA meth-ylation of TNFRSF1B, which in turn is associated withlower expression levels, suggesting a possible maladaptationof the sTNFR2 buffering system and allowing greater in-flammation within the placenta. Epigenomic modifications

Figure 3—Scatter plots for the associations between DNA methylation of placenta CpG sites and 2-h glucose levels post OGTT.

diabetes.diabetesjournals.org Cardenas and Associates 1679

at the TNFRSF1B and PDE4B genes and our findings thatboth genes correlated highly in placental expression suggeststhat they act in common pathways in response to ma-ternal glucose and point toward a likely role of TNF-a,a proinflammatory cytokine amply associated with in-sulin sensitivity and metabolic dysregulation (45,46).

Another CpG site associated with prenatal maternalglucose response was annotated to the body of LDLR.LDLR encodes a lipoprotein receptor that mediates endo-cytosis of LDL particles into the cell and it is known tobe expressed in the placenta (47). Human studies haveshown that intrauterine growth restriction is associatedwith changes in placental LDLR expression compared withnormal pregnancies (48,49). Additionally, an increase inplacenta LDLR expression has been documented in placentasof full-term pregnancies with gestational diabetes mellitusand has been suggested to be attributed to maternal in-flammation (47). Supporting this hypothesis, in vitro studieshave shown that inflammatory cytokines such as TNF-aregulate cholesterol-mediated LDL receptor regulation (50).

Lower DNA methylation at a CpG site annotated to thebody of BLM was associated with greater maternal glucoselevels post oral glucose load. BLM codes for an enzyme thatrestores replication breaks in DNA and is associated withgenome stability and maintenance. Mutations of this geneare associated with an autosomal recessive syndrome, Bloomsyndrome (51). However, its role in glucose homeostasis orplacental functions is unknown. Given the limited litera-ture on BLM and glycemic and metabolic traits or potentialrole in placenta biology, this finding must be interpretedwith caution.

We did not find overlap between our differentiallymethylated CpG sites and those previously reported inplacentas in pregnancies with gestational diabetes mellitus(9–11). Additionally, among the discovered CpG sites, wetested for associations in paired cord blood samples, butwe did not observe consistent associations, suggesting thatresults are placenta specific (Supplementary Table 4).

Our study has strengths and limitations. Our strengthsinclude our prospective design, our relatively large samplesize, careful placenta collection for methylation and ex-pression studies, and the use of the most recent technologycovering.850,000 CpG sites across the genome. Althoughresults were not directly replicated in a smaller inde-pendent cohort of pregnant women without diabetes,estimated relationships were consistent in direction butnonsignificant, possibly due to sample size and low sta-tistical power. Despite our attempts, we were not success-ful in finding another cohort with a larger sample sizethat had appropriate phenotypes and tissue samples forreplication.

Given the prospective design, our findings suggest thatexposure to maternal hyperglycemia gives rise to DNAmethylation alterations as part of the placental adapta-tions reflected in placental DNA collected at birth. Alter-natively, the observed associations might be part of thepathophysiology of impaired glucose response during

Tab

le3—

Adjusted

differen

cesin

DNAmethy

lation(per

SD

chan

geforea

chglyce

mic

trait)am

ong

CpG

sitesdisco

veredin

theEWASofmaterna

l2-h

gluco

seleve

lspost

OGTT

Gen

eCpG

1-hgluc

osepos

t50

-gGCT*

Bas

elinegluc

osepriorto

75-g

OGTT

†1-hgluc

osepos

t75

-gOGTT

†2-hgluc

osepos

t75

-gOGTT

†

Perce

ntdifferen

cein

DNAmethy

latio

n(95%

CI)

P

Perce

ntdifferen

cein

DNAmethy

latio

n(95%

CI)

P

Perce

ntdifferen

cein

DNAmethy

latio

n(95%

CI)

P

Perce

ntdifferen

ceinDNAmethy

latio

n(95%

CI)

P

TNFR

SF1

Bcg

2618

9983

0.13

(20.6,

0.8)

0.72

20.06

(20.6,

0.5)

0.84

1.00

(0.2,1.8)

1.22

310

22

1.62

(1.0,2.2)

1.70

310

27

PDE4B

cg07

7341

6020.91

(21.6,

20.2)

7.08

310

23

20.62

(21.2,

0.01

)0.05

21.08

(21.6,

20.5)

1.42

310

24

21.55

(22.1,

21.1)

1.20

310

29

PDE4B

cg13

8665

7720.59

(21.1,

20.1)

3.07

310

22

20.50

(21.0,

20.01

)0.04

20.95

(21.4,

20.5)

4.58

310

25

21.17

(21.6,

20.7)

1.11

310

27

PDE4B

cg03

4424

6721.46

(22.08

,20.1)

2.95

310

22

20.78

(21.9,

0.4)

0.18

21.57

(22.7,

20.5)

4.56

310

23

22.48

(23.4,

21.5)

2.84

310

27

PDE4B

cg13

3496

2320.49

(20.8,

20.2)

4.30

310

23

20.26

(20.6,

0.1)

0.10

20.52

(20.8.

20.2)

5.22

310

24

20.77

(21.0,

20.5)

2.06

310

29

BLM

cg20

2542

6520.50

(20.9,

20.1)

1.85

310

22

0.03

(20.3,

0.4)

0.86

20.65

(21.0,

20.3)

1.45

310

24

20.84

(21.2,

20.5)

7.58

310

27

LDLR

cg08

4837

1320.22

(20.4,

20.1)

5.75

310

23

20.02

(20.2,

0.1)

0.80

20.17

(20.3,

20.02

)2.79

310

22

20.37

(20.5,

20.2)

1.39

310

26

*Firs

ttrim

ester,no

nfas

ting.

†Sec

ondtrim

ester,fasting.

1680 Placental Epigenetic Adaption Diabetes Volume 67, August 2018

pregnancy and, therefore, DNA methylation shifts couldbe seen as a biomarker of this physiological process oreven contributing to maternal glycemic regulation. Ourprospective design minimizes the latter but does not ruleout the possibility of reverse causality. In addition, evenwith bioinformatic adjustment for cell-type composition,there is currently no gold standard for cell-type adjustmentfor placental DNA methylation. However, the ReFACTormethod has been shown to perform well when comparedwith reference-based methods in blood (52). Yet, adjustmentwith reference-free methods or reference-based meth-ods does not guarantee that associations do not originatefrom cell-type differences, and results could reflect cellularlineage commitment and differentiation. The correlationsobserved between DNAmethylation and expression couldbe driven by or reflect chromatin configuration. Last, oursample is composed of women from European descentand may not be generalizable to other ethnicities.

CONCLUSION

In this prospective study of healthy expecting mothersand term births, we observed robust associations between

maternal glucose response and DNA methylation of theplacenta at several genes implicated in inflammatoryprocesses. Methylation levels at the discovered locicorrelated with functional changes in gene expression,potentially reflecting placental adaptions to maternalimpaired glucose response that could underlie fetal met-abolic programming.

Funding. This work was supported by American Diabetes Association accel-erator award 1-15-ACE-26 (M.-F.H.), Fonds de Recherche du Québec - Santé20697 (M.-F.H.), Canadian Institutes of Health Research MOP 115071 (M.-F.H.),and Diabetes Québec grants (P.P. and L.B.).Duality of Interest. No potential conflicts of interest relevant to this articlewere reported.Author Contributions. A.C. is the lead author of the study and carried outall epigenome-wide analyses with the guidance of M.-F.H., who conceived theoriginal study design and analyses plan with the help of P.P. and L.B. V.G.-O.performed the gene expression analyses and replication of results for ECO-21with the guidance of D.B. and L.B. C.A. helped with the data analyses. Allauthors helped write the manuscript and approved the final version. A.C. is theguarantor of this work and, as such, had full access to all the data in the studyand takes responsibility for the integrity of the data and the accuracy of thedata analysis.

Figure 4—Pearson correlation coefficients and fitted scatter plot lines for the association between placental DNA methylation and geneexpression among top loci associated with prenatal maternal 2-h glucose levels post OGTT (N = 104).

diabetes.diabetesjournals.org Cardenas and Associates 1681

Prior Presentation. Parts of this study were presented in abstract format the 77th Scientific Sessions of the American Diabetes Association, San Diego,CA, 9–13 June 2017.

References1. Pedersen J. The Pregnant Diabetic and Her Newborn: Problems andManagement. Copenhagen, Munksgaard, 19772. HAPO Study Cooperative Group, Metzger BE, Persson B, et al. Hyperglycemiaand adverse pregnancy outcome study: neonatal glycemia. Pediatrics 2010;126:e1545–e15523. Metzger BE, Persson B, Lowe LP, et al. Hyperglycemia and adversepregnancy outcome study: neonatal glycemia. Pediatrics 2010;126:e1545–e15524. Tam WH, Ma RCW, Ozaki R, et al. In utero exposure to maternal hyper-glycemia increases childhood cardiometabolic risk in offspring. Diabetes Care2017;40:679–6865. Symonds ME, Sebert SP, Hyatt MA, Budge H. Nutritional programming of themetabolic syndrome. Nat Rev Endocrinol 2009;5:604–6106. Brett KE, Ferraro ZM, Yockell-Lelievre J, Gruslin A, Adamo KB. Maternal-fetalnutrient transport in pregnancy pathologies: the role of the placenta. Int J Mol Sci2014;15:16153–161857. Staat BC, Galan HL, Harwood JE, et al. Transplacental supply of mannose andinositol in uncomplicated pregnancies using stable isotopes. J Clin EndocrinolMetab 2012;97:2497–25028. Illsley NP. Glucose transporters in the human placenta. Placenta 2000;21:14–229. Ruchat S-M, Houde A-A, Voisin G, et al. Gestational diabetes mellitusepigenetically affects genes predominantly involved in metabolic diseases. Epi-genetics 2013;8:935–94310. Finer S, Mathews C, Lowe R, et al. Maternal gestational diabetes is as-sociated with genome-wide DNA methylation variation in placenta and cord bloodof exposed offspring. Hum Mol Genet 2015;24:3021–302911. Rong C, Cui X, Chen J, Qian Y, Jia R, Hu Y. DNA methylation profiles inplacenta and its association with gestational diabetes mellitus. Exp Clin EndocrinolDiabetes 2015;123:282–28812. Guillemette L, Allard C, Lacroix M, et al. Genetics of Glucose regulation inGestation and Growth (Gen3G): a prospective prebirth cohort of mother-child pairsin Sherbrooke, Canada. BMJ Open 2016;6:e01003113. Taylor S, Wakem M, Dijkman G, Alsarraj M, Nguyen M. A practical approachto RT-qPCR-Publishing data that conform to the MIQE guidelines. Methods 2010;50:S1–S514. Aryee MJ, Jaffe AE, Corrada-Bravo H, et al. Minfi: a flexible and compre-hensive Bioconductor package for the analysis of Infinium DNA methylationmicroarrays. Bioinformatics 2014;30:1363–136915. Pidsley R, Zotenko E, Peters TJ, et al. Critical evaluation of the IlluminaMethylationEPIC BeadChip microarray for whole-genome DNA methylation pro-filing. Genome Biol 2016;17:20816. Fortin J-P, Labbe A, Lemire M, et al. Functional normalization of 450kmethylation array data improves replication in large cancer studies. Genome Biol2014;15:50317. Xu Z, Langie SA, De Boever P, Taylor JA, Niu L. RELIC: a novel dye-biascorrection method for Illumina Methylation BeadChip. BMC Genomics 2017;18:418. Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray ex-pression data using empirical Bayes methods. Biostatistics 2007;8:118–12719. Du P, Zhang X, Huang C-C, et al. Comparison of Beta-value and M-valuemethods for quantifying methylation levels by microarray analysis. BMC Bio-informatics 2010;11:58720. Meller M, Vadachkoria S, Luthy DA, Williams MA. Evaluation of house-keeping genes in placental comparative expression studies. Placenta 2005;26:601–60721. Desgagné V, Hivert MF, St-Pierre J, et al. Epigenetic dysregulation of the IGFsystem in placenta of newborns exposed to maternal impaired glucose tolerance.Epigenomics 2014;6:193–207

22. Fox J, Weisberg S. An R Companion to Applied Regression: Robust Re-gression in R. Thousand Oaks, CA, Sage Publications, 201123. Rahmani E, Zaitlen N, Baran Y, et al. Sparse PCA corrects for cell typeheterogeneity in epigenome-wide association studies. Nat Methods 2016;13:443–44524. Torphy TJ. Phosphodiesterase isozymes: molecular targets for novel anti-asthma agents. Am J Respir Crit Care Med 1998;157:351–37025. Serezani CH, Ballinger MN, Aronoff DM, Peters-Golden M. Cyclic AMP:master regulator of innate immune cell function. Am J Respir Cell Mol Biol 2008;39:127–13226. Jin S-LC, Conti M. Induction of the cyclic nucleotide phosphodiesterasePDE4B is essential for LPS-activated TNF-a responses. Proc Natl Acad Sci U S A2002;99:7628–763327. Horiuchi T, Mitoma H, Harashima S, Tsukamoto H, Shimoda T. Trans-membrane TNF-a: structure, function and interaction with anti-TNF agents.Rheumatology (Oxford) 2010;49:1215–122828. Méhats C, Schmitz T, Oger S, Hervé R, Cabrol D, Leroy M-J. PDE4 as a targetin preterm labour. BMC Pregnancy Childbirth 2007;7(Suppl. 1):S1229. Schmitz T, Souil E, Hervé R, et al. PDE4 inhibition prevents preterm deliveryinduced by an intrauterine inflammation. J Immunol 2007;178:1115–112130. Basu S, Haghiac M, Surace P, et al. Pregravid obesity associates with in-creased maternal endotoxemia and metabolic inflammation. Obesity (SilverSpring) 2011;19:476–48231. Kirwan JP, Hauguel-De Mouzon S, Lepercq J, et al. TNF-a is a predictor ofinsulin resistance in human pregnancy. Diabetes 2002;51:2207–221332. Guillemette L, Lacroix M, Battista M-C, et al. TNFa dynamics during the oralglucose tolerance test vary according to the level of insulin resistance in pregnantwomen. J Clin Endocrinol Metab 2014;99:1862–186933. Zhang R, Maratos-Flier E, Flier JS. Reduced adiposity and high-fat diet-induced adipose inflammation in mice deficient for phosphodiesterase 4B. En-docrinology 2009;150:3076–308234. Möllmann J, Kahles F, Lebherz C, et al. The PDE4 inhibitor roflumilastreduces weight gain by increasing energy expenditure and leads to improvedglucose metabolism. Diabetes Obes Metab 2017;19:496–50835. Vollert S, Kaessner N, Heuser A, et al. The glucose-lowering effects of thePDE4 inhibitors roflumilast and roflumilast-N-oxide in db/db mice. Diabetologia2012;55:2779–278836. Wouters EF, Bredenbröker D, Teichmann P, et al. Effect of the phospho-diesterase 4 inhibitor roflumilast on glucose metabolism in patients with treat-ment-naive, newly diagnosed type 2 diabetes mellitus. J Clin Endocrinol Metab2012;97:E1720–E172537. Plock N, Vollert S, Mayer M, Hanauer G, Lahu G. Pharmacokinetic/phar-macodynamic modeling of the PDE4 inhibitor TAK-648 in type 2 diabetes: earlytranslational approaches for human dose prediction. Clin Transl Sci 2017;10:185–19338. van Greevenbroek MM, van der Kallen CJ, Geurts JM, Janssen RG, BuurmanWA, de Bruin TW. Soluble receptors for tumor necrosis factor-a (TNF-R p55 andTNF-R p75) in familial combined hyperlipidemia. Atherosclerosis 2000;153:1–839. Uysal KT, Wiesbrock SM, Marino MW, Hotamisligil GS. Protection fromobesity-induced insulin resistance in mice lacking TNF-a function. Nature 1997;389:610–61440. Haider S, Knöfler M. Human tumour necrosis factor: physiological andpathological roles in placenta and endometrium. Placenta 2009;30:111–12341. Stanley TL, Zanni MV, Johnsen S, et al. TNF-a antagonism with etanerceptdecreases glucose and increases the proportion of high molecular weight adi-ponectin in obese subjects with features of the metabolic syndrome. J ClinEndocrinol Metab 2011;96:E146–E15042. Hivert M, Sullivan L, Shrader P, et al. The association of TNFr2 and TNFawithinsulin resistance and the influence of adipose tissue biomarkers in humans.Metabolism 2010;59:54043. Straczkowski M, Kowalska I, Stepien A, Dzienis-Straczkowska S, Sze-lachowska M, Kinalska I. Increased plasma-soluble tumor necrosis factor-a

1682 Placental Epigenetic Adaption Diabetes Volume 67, August 2018

receptor 2 level in lean nondiabetic offspring of type 2 diabetic subjects. DiabetesCare 2002;25:1824–182844. Dzienis-Straczkowska S, Straczkowski M, Szelachowska M, Stepien A,Kowalska I, Kinalska I. Soluble tumor necrosis factor-a receptors in young obesesubjects with normal and impaired glucose tolerance. Diabetes Care 2003;26:875–88045. Katsuki A, Sumida Y, Murashima S, et al. Serum levels of tumor necrosisfactor-a are increased in obese patients with noninsulin-dependent diabetesmellitus. J Clin Endocrinol Metab 1998;83:859–86246. Winkler G, Salamon F, Salamon D, Speer G, Simon K, Cseh K. Elevated serumtumour necrosis factor-alpha levels can contribute to the insulin resistance in type II(non-insulin-dependent) diabetes and in obesity. Diabetologia 1998;41:860–86147. Dubé E, Ethier-Chiasson M, Lafond J. Modulation of cholesterol transport byinsulin-treated gestational diabetes mellitus in human full-term placenta. BiolReprod 2013;88:16

48. Wadsack C, Tabano S, Maier A, et al. Intrauterine growth restriction isassociated with alterations in placental lipoprotein receptors and maternal lipo-protein composition. Am J Physiol Endocrinol Metab 2007;292:E476–E48449. Stepan H, Faber R, Walther T. Expression of low density lipoprotein receptormessenger ribonuclein acid in placentas from pregnancies with preterm deliveryand pregnancies with intrauterine growth retardation. BJOG 1999;106:1221–122250. Conover CA, Harrington SC, Bale LK. Differential regulation of pregnancyassociated plasma protein-A in human coronary artery endothelial cells andsmooth muscle cells. Growth Horm IGF Res 2008;18:213–22051. Kaneko H, Fukao T, Kondo N. The function of RecQ helicase gene family(especially BLM) in DNA recombination and joining. Adv Biophys 2004;38:45–6452. Rahmani E, Zaitlen N, Baran Y, et al. Correcting for cell-type heterogeneityin DNA methylation: a comprehensive evaluation. Nat Methods 2017;14:218–219

diabetes.diabetesjournals.org Cardenas and Associates 1683