UBC Faculty Research and Publications

DNA methylation as a predictor of fetal alcohol spectrum disorder Lussier, Alexandre A; Morin, Alexander M; MacIsaac, Julia L; Salmon, Jenny; Weinberg, Joanne; Reynolds, James N; Pavlidis, Paul; Chudley, Albert E; Kobor, Michael S Jan 12, 2018

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

Item Metadata


52383-13148_2018_Article_439.pdf [ 1.14MB ]
JSON: 52383-1.0363041.json
JSON-LD: 52383-1.0363041-ld.json
RDF/XML (Pretty): 52383-1.0363041-rdf.xml
RDF/JSON: 52383-1.0363041-rdf.json
Turtle: 52383-1.0363041-turtle.txt
N-Triples: 52383-1.0363041-rdf-ntriples.txt
Original Record: 52383-1.0363041-source.json
Full Text

Full Text

RESEARCH Open AccessDNA methylation as a predictor of fetalalcohol spectrum disorderAlexandre A. Lussier1,2, Alexander M. Morin1, Julia L. MacIsaac1, Jenny Salmon3,4, Joanne Weinberg2,James N. Reynolds5, Paul Pavlidis6,8, Albert E. Chudley3,4 and Michael S. Kobor1,7*AbstractBackground: Fetal alcohol spectrum disorder (FASD) is a developmental disorder that manifests through a range ofcognitive, adaptive, physiological, and neurobiological deficits resulting from prenatal alcohol exposure. Althoughthe North American prevalence is currently estimated at 2–5%, FASD has proven difficult to identify in the absenceof the overt physical features characteristic of fetal alcohol syndrome. As interventions may have the greatestimpact at an early age, accurate biomarkers are needed to identify children at risk for FASD. Building on ourprevious work identifying distinct DNA methylation patterns in children and adolescents with FASD, we haveattempted to validate these associations in a different clinical cohort and to use our DNA methylation signature todevelop a possible epigenetic predictor of FASD.Methods: Genome-wide DNA methylation patterns were analyzed using the Illumina HumanMethylation450 arrayin the buccal epithelial cells of a cohort of 48 individuals aged 3.5–18 (24 FASD cases, 24 controls). The DNAmethylation predictor of FASD was built using a stochastic gradient boosting model on our previously publisheddataset FASD cases and controls (GSE80261). The predictor was tested on the current dataset and an independentdataset of 48 autism spectrum disorder cases and 48 controls (GSE50759).Results: We validated findings from our previous study that identified a DNA methylation signature of FASD,replicating the altered DNA methylation levels of 161/648 CpGs in this independent cohort, which may represent arobust signature of FASD in the epigenome. We also generated a predictive model of FASD using machine learningin a subset of our previously published cohort of 179 samples (83 FASD cases, 96 controls), which was tested in thisnovel cohort of 48 samples and resulted in a moderately accurate predictor of FASD status. Upon testing thealgorithm in an independent cohort of individuals with autism spectrum disorder, we did not detect any biastowards autism, sex, age, or ethnicity.Conclusion: These findings further support the association of FASD with distinct DNA methylation patterns, whileproviding a possible entry point towards the development of epigenetic biomarkers of FASD.Keywords: Fetal alcohol spectrum disorder, Epigenetics, DNA methylation, Biomarkers, Neurodevelopmentaldisorders* Correspondence: msk@cmmt.ubc.ca1Department of Medical Genetics, Centre for Molecular Medicine andTherapeutics, British Columbia Children’s Hospital Research Institute,University of British Columbia, Vancouver, British Columbia, Canada7Human Early Learning Partnership, University of British Columbia, Vancouver,British Columbia, CanadaFull list of author information is available at the end of the article© The Author(s). 2018 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.Lussier et al. Clinical Epigenetics  (2018) 10:5 DOI 10.1186/s13148-018-0439-6BackgroundFetal alcohol spectrum disorder (FASD) is a leading pre-ventable cause of developmental disability, with a NorthAmerican prevalence currently estimated at 2–5% [1–3].FASD presents through a wide spectrum of phenotypes,ranging from growth deficits and physical abnormalitiesto cognitive and behavioral deficits, as well as motor andsensory impairments, immune dysfunction, and in-creased vulnerability to mental health problems in adult-hood [4–6]. On the most severe end of the spectrum liesfetal alcohol syndrome (FAS), which is characterized bygrowth retardation, a distinct set of facial dysmorphisms,and central nervous system abnormalities [7, 8]. By con-trast, Alcohol-Related Neurodevelopmental Disorder(ARND) describes the less visible and largest groupwithin the spectrum, where individuals with confirmedalcohol exposure during pregnancy show primarilybehavioral, adaptive, and/or cognitive abnormalitieswithout obvious facial dysmorphisms [9]. Of note, indi-viduals across the spectrum show cognitive and behav-ioral deficits, which can be as serious in those withoutany physical features as in those with full FAS [10].Although children with FAS are often diagnosed ininfancy or in early life, FASD in general has proven diffi-cult to identify, particularly in the absence of the overtfacial features characteristic of FAS. As such, many indi-viduals with FASD are not identified until they reachschool age, where they begin to struggle with increasedsocial pressure and cognitive challenges [11]. However,early cognitive and behavioral interventions may poten-tially attenuate some of the deficits associated withFASD and improve the long-term outcomes of theseindividuals [12]. As early diagnosis is a strong predictorof positive outcome, early screening tools are necessaryto help identify at-risk children at a young age andpotentially buffer some of the deficits associatedwith prenatal alcohol exposure (PAE) [13, 14].While self-report methods are most commonly usedfor assessing PAE and a child’s risk for FASD, these arenot always accurate and can underestimate alcohol con-sumption during pregnancy [15–17]. Over the past de-cades, various biomarkers of alcohol exposure have beendeveloped to complement self-report measures, focusingprimarily on the direct or indirect products of ethanolmetabolism, which can be measured in biological speci-mens from both the mother and infant [18]. Althoughthese biomarkers are very sensitive to alcohol exposure,they present a number of limitations when attempting todetermine whether prenatal alcohol exposure has oc-curred or to gain insight into the biological underpin-nings of alcohol-induced deficits and the developmentalprofiles associated with FASD. For example, many ofthese biomarkers have short windows of detection (e.g.,urine, blood, plasma) or are limited by specimenavailability (e.g., placenta, meconium), making them use-ful for identification of alcohol exposure around the timeof parturition, but not in infants and children over thecourse of development [19]. As such, objective and per-sistent measures are needed to aid in the screening anddiagnosis of children at risk for FASD.Epigenetic marks are now emerging as potentialbiomarkers or signatures of early-life exposures. Broadlydefined, epigenetics refers to modifications of DNA andits regulatory components, including chromatin andnon-coding RNA, that potentially modulate gene tran-scription without changing underlying DNA sequences[20–22]. In addition to their role in the regulation of cel-lular processes, these may also bridge environmental fac-tors and genetic regulation to capture a lasting signatureof early life exposures. In particular, DNA methylation isemerging as a candidate biomarker for environmentalexposures and disease. Typically found on the cytosineresidues of cytosine-guanine dinucleotides (CpG), thisepigenetic mark is both stable over time and dynamic inresponse to environmental factors [23]. Several pre- andpostnatal environmental influences have been associatedwith altered DNA methylation patterns, hinting at pos-sible malleability by early-life environments and suggest-ing a potential utility as biomarkers [24, 25]. For example,prenatal exposure to cigarette smoke is associated withlasting alterations to DNA methylation patterns, whichare now being used as biomarkers of cigarette smokeexposure in infants [26].While in its infancy in relation to FASD, epigeneticbiomarkers show promise for early screening of at-riskindividuals, as the DNA methylome retains a lasting sig-nature of prenatal alcohol exposure in both the centralnervous system and peripheral tissues (reviewed in [27]).Numerous studies performed in animal and cell culturemodels have identified both short-term and persistentalterations to DNA methylation patterns following PAE.Although some of these models reflect supra-physiological levels of alcohol exposure or display mod-est effect sizes in response to PAE, the findings fromthese pre-clinical models suggest the possibility that PAEmay directly influence epigenetic patterns and that thesemay play a role in PAE-induced deficits [27–33]. Bycontrast, fewer studies have investigated DNA methyla-tion patterns in individuals with FASD. More targetedmethods identified differences in DNA methylationlevels in the promoter region of Drd4 in a largeAustralian cohort of children exposed to alcohol duringbreastfeeding [34]. Others have employed discovery-driven approaches, assessing genome-wide DNA methy-lation patterns in case-control studies of FASD. The firstof these came from a small cohort of children, whereslight differences in DNA methylation patterns withinthe protocadherin (PCDH) gene clusters reported with aLussier et al. Clinical Epigenetics  (2018) 10:5 Page 2 of 14rather modest significance threshold [35]. Recently, weanalyzed DNA methylation profiles in a large cohort ofchildren with FASD recruited by NeuroDevNet (NDN),a Canadian Networks of Centres of Excellence, wherewe identified a signature of 658 differentially methylatedCpGs [36]. Although few results have been validatedacross different cohorts, these findings set the stage forbroader applications of DNA methylation in the contextof FASD, creating a framework upon which to build fu-ture epigenomic studies of FASD.To validate the findings from our previous DNAmethylation study, we assessed the genome-wide DNAmethylation profiles of buccal epithelial cells (BEC) froman independent cohort of 24 individuals with FASD,aged 3.5–18, and 24 typically developing controls, aged5–17. Given that our initial study provided a frameworkfor genome-wide assessment of DNA methylation pat-terns in individuals with FASD, we used the findingsfrom the NDN study as a foundation for the identifica-tion of replicable epigenetic differences associated withFASD. Notably, nearly 25% of statistically significantassociations from the NDN cohort were validated in thisnew cohort at a false-discovery rate (FDR) < 0.05 [37]. Inaddition to the validation analyses, we also assessedwhether DNA methylation profiles could be used toidentify individuals with FASD, generating a classifica-tion algorithm that uses DNA methylation levels toaccurately predict FASD status. Taken together, theseresults suggested that there were replicable differencesin DNA methylation patterns between individuals withFASD and controls, which could potentially contributeto the development of a screening tool for at-riskchildren.MethodsThe Kids Brain Health Network cohort of children withFASDThe present cohort was collected as a replication studyby Kids Brain Health Network (KBHN), formerlyNeuroDevNet, and is hereby referred to as the KBHNcohort [38]. Ethics for this study were reviewed andapproved by the “Children’s and Women’s ResearchEthics Board – Clinical” (H10-01149). All experimentalprocedures were reviewed and approved by the Univer-sity of Manitoba and the University of British Columbia.Written informed consent was obtained from a parentor legal guardian, and assent was obtained from eachchild before study participation. The clinics used previ-ously described guidelines for the diagnosis of FASD[39]. Children with FASD and typically developingcontrols were recruited from the Manitoba FASDdiagnostic clinic in Winnipeg, Manitoba, Canada. Briefly,buccal epithelial cell (BEC) samples were collected forDNA methylation analysis from 25 FASD and 26 age-and sex-matched control children aged between 3.5 and18, prior to pre-processing (Table 1). BECs were col-lected using the Isohelix buccal swabs and Dri-Capsule(Cell Projects Ltd., Kent, UK). To collect buccal cells,the swab was inserted into the participants’ mouth andrubbed firmly against the inside of the left cheek for1 min. The swab was then placed into a sterile tube witha Dri-Capsule and the tube sealed. An identical proced-ure was followed for the right cheek. Participants didnot have any dental work performed 48 h prior tocollection, and no food was consumed less than 60 minprior to collection to avoid contamination.DNA methylation 450K assayDNA was extracted from BECs using the Isohelix DNAisolation kit (Cell Projects, Kent, UK). Seven hundredfifty nanograms of genomic DNA was subjected to bisul-fite conversion using the Zymo EZ DNA MethylationKit (Zymo Research, Irvine, California), which convertsDNA methylation information into sequence base differ-ences by deaminating unmethylated cytosines to uracilwhile leaving methylated cytosines unchanged. One hun-dred sixty nanograms of converted DNA was applied tothe HumanMethylation450 BeadChip array from Illu-mina (450K array), which enables the simultaneousTable 1 Characteristics of the NeuroDevNet II FASD cohortFASD cases ControlsN 24 24ARND 18Partial FAS 6FAS 1FASD 1Age (years)Range 3.5–18 5–17Mean 9.1 11.6SexFemale 9 13Male 15 11Self-declared ethnicityCaucasian 4 (2)a 22First Nations 17 (20)a 1Asian 1 (0)a 1Not reported 2 0Caregiver statusBiological parents 7 24Biological grandparents 3 0Adopted/legal guardian 8 0Foster care 6 0aIncluding mixed lineage First NationsLussier et al. Clinical Epigenetics  (2018) 10:5 Page 3 of 14quantitative measurement of 485,512 CpG sites acrossthe human genome, following the manufacturer’sinstructions. Chips were scanned on an Illumina HiScan,with the 51 samples run in two batches and eachcontaining a similar number of FASD and control sam-ples, randomly distributed across the chips. Two pairs oftechnical replicates were also included and showed aPearson correlation coefficient r > 0.994 in both cases,highlighting the technology’s reproducibility on our in-house platform. Inter-sample correlations ranged from0.926–0.99.DNA methylation data quality control and normalizationThe raw DNA methylation data were subjected to arigorous set of quality controls, first of the samples, andthen of the probes. Of the 51 initial samples, 3 were re-moved from the final dataset based on poor quality data,which was identified through skewed internal controlsand/or > = 5% of probes with a detection p value > 0.05(2 controls and 1 FASD). Next, probes were removedfrom the dataset according to the following criteria: (1)probes on X and Y chromosomes (n = 11,648), (2) SNPprobes (n = 65), (3) probes with bead count < 3 in 10% ofsamples (n = 726), (4) probes with 10% of samples with adetection p value > 0.01 (n = 11,864), and (5) probes witha polymorphic CpG and non-specific probes (N = 19,337SNP-CpG and 10,484 non-specific probes) [40]. A finalfiltering step was performed to set the methylationvalues to NA for any remaining probe-sample pair wherebead count < 3 or detection p value > 0.01. Datanormalization was performed using the SWAN methodon the final dataset, composed of 48 samples (24 FASDand 24 controls) and 431,544 probes [41]. Finally, batcheffects (chip number and chip position) were removedusing the ComBat function from the SVA package in R[42]. Statistical analyses were performed using onComBat-corrected M values, which represent the log2ratio of methylated/unmethylated, where negative valuesindicate less than 50% methylation and positive valuesindicate more than 50% methylation [43]. Percentmethylation differences (beta-values) were used ingraphical representations of the data and indicate thepercentage of methylation calculated by methylated/(methylated + unmethylated), ranging from 0 (fullyunmethylated) to 1 (fully methylated).Differential methylation analysis and validation ofNeuroDevNet (NDN) findingsCell type deconvolution was performed to assess theproportions of CD14, CD34, and buccal epithelial cellsin each sample using DNA methylation levels at CpGshighly correlated with these cell types [44]. Surrogatevariable analysis (SVA) was also performed on ComBat-corrected, normalized data using the SVA package in Rto identify surrogate variables (SVs) representative of un-wanted heterogeneity [42]. Using DNA methylation datafrom all 48 samples, SVA identified 6 SVs not associatedwith clinical status (FASD vs control). As these werepartially associated with known covariates, such as celltype proportions and age, the SVs were included in thelinear regression analysis to account for their effects.More specifically, linear modeling was performed on the648 differentially methylated probes identified in the ini-tial NDN study and found in the present dataset usingthe limma package in R and a model that included clin-ical status and all identified SVs as covariates [36, 45].Significant differentially methylated probes betweengroups were identified at a false-discovery rate (FDR) <0.05 following multiple test correction by the Benjamini-Hochberg method and were required to show the samedirection of change as the NDN cohort’s findings [46].Further evaluation of potential biological significancewas performed using an arbitrary threshold of > 5%mean percent DNA methylation difference betweenFASD and controls.DNA methylation pyrosequencing assayThe bisulfite pyrosequencing assay was designed withPyroMark Assay Design 2.0 (Qiagen; Additional file 1:Table S1). The region of interest was amplified by PCRusing the HotstarTaq DNA polymerase kit (Qiagen) asfollows: 15 min at 95 °C, 45 cycles of 95 °C for 30s, 58 °C for 30s, and 72 °C for 30s, and a 5-min 72 °C finalextension step. For pyrosequencing, single-strandedDNA was prepared from the PCR product with thePyromark™ Vacuum Prep Workstation (Qiagen) and thesequencing was performed using sequencing primers ona Pyromark™ Q96 MD pyrosequencer (Qiagen). Thequantitative levels of methylation for each CpGdinucleotide were calculated with Pyro Q-CpG software(Qiagen).The NDN cohort of children with FASDDNA methylation data from the previous cohort of chil-dren with FASD were obtained from GEO (GSE80261)and normalized as described in our original publication[36]. This cohort was collected by NeuroDevNet, aCanadian Network of Centres for Excellence, and ishereby referred to as the NDN cohort [36]. Briefly, weselected the individuals with a confirmed diagnosis ofFASD from this dataset, as well as age- and sex-matchedtypically developing controls, resulting in dataset com-posed of 83 children with FASD (55 ARND, 18 partialFAS, 10 FAS) and 96 typically developing controls. Themean age (in years) for individuals with FASD was 11.88and 11.28 for controls, both ranging from 5 to 18 yearsold. The proportions of males and female differedslightly between groups, with 42 females and 41 males inLussier et al. Clinical Epigenetics  (2018) 10:5 Page 4 of 14the FASD cases and 57 females and 39 males in the con-trol group. A skew in self-declared ethnicity was presentbetween the groups, as the majority of controls identi-fied as Caucasian, while the majority of children in theFASD group identified as First Nations. This skew wasaddressed in the initial epigenome-wide associationstudy through the use of a more ethnically homogeneoussubset of the cohort. DNA methylation data wereobtained from BEC using the Illumina 450K array andwere normalized using the beta-mixture quantilenormalization method.Cohort of individuals with autism spectrum disorderProcessed DNA methylation data from a publically avail-able dataset of individuals with autism spectrum dis-order (ASD) were obtained from GEO (GSE50759).Briefly, this dataset was composed of 48 individuals withASD and 48 typically developing controls. As per theauthors’ description of the GEO data, these were prepro-cessed using the R packages minfi and sva to obtainnormalized M values [47]. The samples consisted of 58males and 38 females, consistent with the skew towardsmales in ASD. The mean age (8.84) and range (1–28 years old) differed from the NDN and KBHN studies,and the genetic ancestry of most individuals was Caucasian(European), though a proportion of the cohort was ofNigerian ancestry. DNA methylation data of these sampleswere obtained from BEC using the Illumina 450K array.DNA methylation as a predictor of FASD statusA predictive model of FASD status was created usingDNA methylation data and the caret package in R[48]. First, a predictive model was created using sto-chastic gradient boosting on the NDN cohort (83FASD cases, 96 controls) using the beta-values of thedifferentially methylated probes identified in the NDNstudy (648 probes) [36]. The parameters of the mod-eling were optimized for area under the receiver oper-ating characteristic (ROC) curve by grid tuning forrepeated cross-validation (number of trees 50–1500;1, 5, or 9 interaction depth; 0.1 shrinkage). The opti-mal model for predicting clinical FASD status using648 probes was 550 trees, 1 of interaction depth, and20 minimum observations per node. The KBHN co-hort (24 FASD, 24 controls) was then used to verifythe predictive sensitivity and specificity of the model.In parallel, 450K data from a cohort of children withASD were tested to verify the predictive specificity ofthe model for FASD. The predictor was tested usingnormalized data that was uncorrected for batch ef-fects to better mimic the potential use of the predict-ive model by independent groups.ResultsThe KBHN cohort of children with FASDAs noted, we analyzed genome-wide DNA methyla-tion patterns from 24 children with FASD and 24typically developing controls, matched for sex andage, ranging from 3.5 to 18 years of age (Table 1).We found that self-declared ethnicity, primarycaregiver, and mean age were significantly differentbetween the FASD and control participants (Student’st test; p < 0.05). We corrected for the potential effectsof age on DNA methylation through the statisticalmethods outlined below. However, given the confoundin self-declared ethnicity and caregiver status, wecould not correct for these effects and relied on theprevious correction of ethnic bias in the initial NDNstudy (see below) [36]. Furthermore, we could not ac-count for the different life experiences of individualswith FASD, including potential exposure to adverseearly life events at considerably higher levels thanthose in the general population. It is possible thatthese distinct experiences in themselves may poten-tially be associated with DNA methylation patterns.Children with FASD and typically developing controlsshowed differential DNA methylation patternsFollowing quality control and normalization, 431,544sites of the 485,512 sites remained in the final dataset of48 samples, which were corrected for batch effects usingComBat. While BECs are mostly a homogeneous popu-lation of cells, they contain small proportions of CD34-and CD14-positive white blood cells, which can poten-tially skew DNA methylation analyses. As such, cell typedeconvolution was performed to identify any blood con-tamination in the samples, identifying a trend towardssignificance in the proportions of different cells typesbetween groups (CD34+, p = 0.115; CD14+, p = 0.224;BEC, p = 0.068). To account for this factor in addition toother potential confounding variables within the dataset,we performed SVA to identify patterns of variation,identifying 6 surrogate variables when protecting the ef-fects of group (FASD vs control). These were correlatedwith known sources of variation within the data, includ-ing cell type proportions and age (Additional file 2: Fig-ure S1).To identify DNA methylation patterns specific to theFASD group, we coupled differential DNA methylationanalysis using a two-group design with the surrogatevariables to correct for undesirable variation in the data.Given that we already accounted for ethnicity-relatedprobes as much as possible in the NDN study, it wasconcluded that the effects of ethnic background wouldbe lessened by using the final 658 differentially methyl-ated CpGs [36]. As such, we performed linear modelingon the probes that were differentially methylated in theLussier et al. Clinical Epigenetics  (2018) 10:5 Page 5 of 14first study and remained in the dataset after pre-processing (648 CpGs of 658 from NDN). Of these, 161CpGs displayed statistically significant differentialmethylation in the same direction as the initial cohort inthe KBHN FASD group compared to the controls at anFDR < 0.05 (Fig. 1a; Additional file 1: Table S2). Toassess the probability of validating this many probes,random group subsampling was performed 10,000 times.As none showed more differentially methylated probesthan the original replication cohort (maximum = 31differentially methylated probes), the probability of valid-ating 161/648 probes was < 1e−4 (Additional file 2:Figure S2). Of the 161 validated probes, 82 were up-methylated, while 79 were down-methylated in FASDcompared to control samples. Several genes containedmultiple differentially methylated CpGs across both co-horts, including Hla-dpb1 (5), Fam59b (4), Capn10 (3),Des (3), Slc6a3 (3), Slc38a2 (3), Fam24a (2), H19 (2),and Tgfb1i1 (2) (Table 2). Moreover, 53 CpGs showed >5% difference in methylation, an arbitrary cutoff oftenused to gauge potential biological significance [49].Three genes contained 2 or more differentially methyl-ated (DM) probes that showed both an FDR < 0.05 and> 5% difference in percent methylation, includingFam59b (4 probes), Hla-dpb1 (2 probes), and Slc6a3 (2probes). In particular, the Fam59b CpGs were locatedwithin a CpG island and showed substantial differences inDNA methylation levels between FASD and controlgroups, with an average 13% methylation difference acrossthe array probes in the CpG island (Fig. 2). Three add-itional sites located in intergenic regions showed > 10%percent DNA methylation difference between groups.Overall, the percent methylation differences betweengroups of the 648 analyzed probes were highly correlatedbetween the NDN and KBHN cohorts, as determined bylinear modeling (r = 0.638, p < 2.2e−16; Fig. 1b). We alsocompared the ranking of probes by p value from linearFig. 1 Visualization and verification of the differentially methylated probes. a Heatmap of the 161 validated probes validated in the KBHN cohortat an FDR < 0.05 (79 hypermethylated in FASD; 82 hypomethylated in FASD). The percent methylation values (ranging from 0 to 100) werecentered, scaled, and trimmed, resulting in a standardized DNA methylation level ranging from − 2 to + 2 (blue-red scale). b Scatter plot of thedifferences in percent methylation between FASD and controls for the 648 differentially probes identified in the NDN cohort. The meandifferences between groups were highly correlated between both the NDN and KBHN cohorts (r = 0.638, p < 2.2e-16). The red points show theprobes that were statistically significant (FDR < 0.05) and showed the same direction of change across both studies c Verification by bisulfitepyrosequencing in FASD (blue) and control (gray) samples. The left panel shows the DNA methylation levels from the pyrosequencing assay,while the right panel shows the results from the 450K array. The CpG assayed was located in the CACNA1A gene body (cg24800175) and showedstatistically significant differences between groups (p = 0.04)Table 2 Genes containing multiple differentially methylatedCpGs in FASDGene No. of CpGs Direction of changeHla-dpb1 5 UpFam59b 4 DownDes 3 DownSlc6a3 3 UpSlc38a2 3 DownCapn10 3 UpFam24a 2 UpH19 2 DownTgfb1i1 2 DownLussier et al. Clinical Epigenetics  (2018) 10:5 Page 6 of 14modeling between the NDN and KBHN cohorts; nosignificant similarities were identified (p= 0.91). Of note, 21of the significant probes with > 5% methylation differencebetween groups from the NDN study were validated in thepresent analysis (39 of 41 were present in the fil-tered KBHN dataset). This proportion (54%) was muchhigher than all validated probes (25%), suggesting that thesepotentially represented more robust associations with FASD.Bisulfite pyrosequencing verified the differential DNAmethylation of CACNA1ATo verify that the differential DNA methylation resultsdid not depend on the method used to measure them,we assessed DNA methylation levels of the cg24800175probe in CACNA1A. We selected this probe as it wasalso verified in the initial NDN study, where it similarlyshowed a > 5% difference in DNA methylation betweenindividuals with FASD and controls. Pyrosequencing re-sults confirmed the DNA methylation levels observed onthe 450K array, showing similar DNA methylation levelsand differences between groups for CpGs located inCACNA1A (Fig. 1c). The Pearson correlation betweenthese two methods was 0.826 and the Bland–Altmanplot showed little difference when comparing the 450Karray to pyrosequencing, suggesting good concordancebetween DNA methylation data from the two methods(Additional file 2: Figure S3). Linear regression analysisof pyrosequencing data between FASD cases and con-trols confirmed differential DNA methylation in this site,even without correcting for covariates (p = 0.04).DNA methylation patterns classified individuals withFASD versus controlsTo assess whether DNA methylation data could be usedto predict FASD status, we created a predictive algo-rithm of FASD using machine learning approaches. First,we selected the normalized DNA methylationdata (beta-values) of 179 samples from the NDN cohort(83 FASD; 96 control) in the 648 initial probes that werealso found in the KBHN data. Our strategy was to buildthe predictor using an initial training cohort (NDN),followed by subsequent evaluation in the test cohort(KBHN). See Fig. 3 for an overview of steps used tobuild the FASD predictor.Using a gradient boosting model in the caret packageto optimize both sensitivity and specificity (area underthe receiver operating characteristic (ROC) curve), weFig. 2 Several differentially methylated CpGs were located in the Fam59b gene body. DNA methylation levels for FASD (blue) and controls (gray)are shown for 10 CpGs within the gene, with the red circles representing the validated hits in KBHN (FDR < 0.05). These were located in a CpGisland, illustrated by the green bar at the bottom, which showed an average 13% difference in DNA methylation levels in individuals with FASDversus controls across all five CpGs covered by the 450K arrayFig. 3 Flowchart of bioinformatic analyses for the DNA methylationpredictor of FASD. Briefly, samples from the NDN cohort were usedas the training set, and machine learning was performed on theDNA methylation signature of FASD identified in the initial NDNstudy. The resulting FASD predictor was tested on the KBHN test set,as well as an independent cohort composed of individuals withautism spectrum disorder and typically developing controls to testthe specificity of the predictor for FASDLussier et al. Clinical Epigenetics  (2018) 10:5 Page 7 of 14created a predictive model to assess the probability ofFASD based on DNA methylation patterns [48]. Thismethod provided weighted values for the different fea-tures (CpGs) of the model to determine their import-ance in classifying the samples. Of the 648 initialfeatures, 183 had non-zero influence on the predictivemodel and could be used to predict FASD status(Additional file 1: Table S3). As the number of non-zerofeatures was similar to the total number of samples,concerns of model over-fitting were reduced.Through this approach, the predicted sensitivity andspecificity for the training cohort were 0.879 and 0.944, re-spectively, for an area under the curve of 0.977 (95% con-fidence intervals, 0.972–0.982; Fig. 4a). The performanceof the predictor on the training data indicated that DNAmethylation could be used to distinguish FASD cases andcontrols, although these results will need to be carefullyassessed in independent test sets or clinical settings.To get a better understanding of the utility of this tool,we next assessed the predictive model using the normal-ized, batch-corrected DNA methylation data of the KBHNcohort as a test set. Of note, these data were not correctedfor any covariates or surrogate variables other than batchcorrection. As expected for analysis of an independent testset, the model performed at a lower level in this cohort,displaying 0.917 sensitivity, 0.75 specificity, and 0.920 areaunder the ROC curve (Table 3; Fig. 4a). The balancedaccuracy of the model in this cohort was 0.833 (95% CI0.698–0.925), and the ROC curve was not significantlydifferent from the one obtained in the training cohort (p= 0.192). Overall, 2 controls were misclassified as FASDand 6 children with FASD were misclassified as controls,giving a negative predictive value (NPV) of 78.6% and apositive predictive value (PPV) of 90%. Given the discrep-ancies in ethnic backgrounds between FASD and controlgroups, the misclassified samples were assessed for differ-ences in self-reported ethnicity, caregiver status, age, andbuccal cell-type proportions in the classification. We didnot identify any skew of these data in the misclassifiedcontrols, which were both Caucasian males aged 15 and16, respectively. Although every misclassified individualwith FASD had a previous diagnosis of ARND, a categorythat was present in high proportion within this cohort, noother patterns emerged between the correctly and incor-rectly classified individuals with FASD (3 females/3 males;3 First Nations/1 Métis/2 Caucasian; aged 6–18). Takentogether, these findings suggested that differences in thesedemographic variables between the groups did not drivetheir classification.The DNA methylation predictors were not biased by ASDin an independent cohortBEC samples from an independent published autismspectrum disorder (ASD) cohort were used to assess thespecificity of the model in the FASD cohorts. To thisend, we used a publically available dataset of 450K arraydata from the BECs of 48 individuals with ASD and 48typically developing controls from the gene expressionomnibus (GSE50759) [47]. Using processed GEO datafrom this cohort, the predictor correctly identified thevast majority of individuals in the cohort as non-FASD.The model only misclassified 1 individual as FASD, for aspecificity of 0.990 (95% CI 0.943–0.9997), higher thanthe predicted specificity in the training set. This sample,a 3-year-old female with ASD (51% African ancestry,41% European ancestry) did not have any particulardistinguishing features compared to the correctly classi-fied samples, suggesting that the predictive model wasnot biased for ASD, sex, age, or African ancestry in thisindependent cohort.DiscussionEpigenetic marks are emerging as potential biomarkersand mediators of environmental exposures, and a grow-ing body of literature suggests that epigenetic factorsmay be involved in the etiology of FASD. In particular,our recent study using a large cohort of children withFASD to date identified a signature of 658 differentiallymethylated CpGs in the BEC of individuals with FASDcompared to typically developing controls [36]. Here, wepresent a validation of genome-wide DNA methylationdata in a small cohort of individuals with FASD, wherewe successfully replicated 161 of the 658 differentiallymethylated CpGs identified in the initial NDN cohort.Furthermore, we demonstrated that DNA methylationdata could be utilized to generate a predictive algorithmto classify individuals as FASD or control with highaccuracy. These results indicated that DNA methylationin BECs could potentially be used towards developing ascreening tool for children at risk for FASD.Our present findings represent the initial validation ofgenome-wide DNA methylation differences in individ-uals with FASD. Of the 161 validated CpGs at an FDR <0.05, 53 had > 5% difference in DNA methylation levelsbetween groups. This 5% threshold is often used forassessing potential biological relevance in epigeneticstudies of psychiatric and neurodevelopmental disorders,and therefore, we confined our interpretation of possiblefunctional implications to CpGs with this effect size [47,49–51]. Importantly, the biological significance of a 5%difference in DNA methylation is poorly understood,and its functional relevance may be limited in relation togene expression or cellular physiology. Nevertheless, 21CpGs showed a > 5% difference between FASD casesand controls at an FDR < 0.05 in both the KBHN andNDN cohorts, suggesting that these may represent thestrongest associations with FASD. Although the DNAmethylation differences between FASD and controlsLussier et al. Clinical Epigenetics  (2018) 10:5 Page 8 of 14were highly correlated between the NDN and KBHNcohorts (p < 2.2e−16), the majority of CpGs showing thesame direction of change did not achieve statisticalsignificance (301/648), potentially due to the replicationcohort’s small size or the absence of individuals withonly PAE in this cohort. In addition, we verified the re-sults from the 450K array by bisulfite pyrosequencing,confirming the differential DNA methylation results fora CpG located in CACNA1A and supporting that ourfindings were not an artifact of array technology. Asdiscussed in a recent comprehensive review, we notethat the functional relevance of these differences ishighly dependent on multiple factors, including subcel-lular differences, transcription factor binding regulation,density and cooperativity of DNA methylation, or othercis-regulatory elements [52].Several genes previously associated with PAE or FASDcontained multiple differentially methylated CpGs with> 5% difference in DNA methylation between groups,including Fam59b, Hla-dpb1, and Slc6a3. The Hla-dpb1Fig. 4 Visualization of the training and test set performance for the DNA methylation predictor of FASD. a The DNA methylation predictorcreated using the 648 probes identified in NDN showed high accuracy in the training cohort (dark gray; area under the curve = 0.977) and slightlypoorer accuracy in the KBHN test set (blue; area under the curve = 0.920). These curves were not significantly different (p = 0.192). b The confusionmatrix displays number of samples classified correctly or incorrectly. Of note, six individuals with FASD in the test set were classified as controls,while only two control samples were misclassified as FASDLussier et al. Clinical Epigenetics  (2018) 10:5 Page 9 of 14locus, a member of the major histocompatibility com-plex proteins, contained several differentially methylatedCpGs, which overlapped with a differentially methylatedregion identified in the NDN study. Given its importantfunction in immune regulation and potential role inrheumatoid arthritis, these differences could potentiallyreflect some of the immune deficits associated withFASD [53]. Furthermore, the Fam59b gene containedseveral CpGs with substantial (> 10%) differences inDNA methylation levels between individuals with FASDand controls, potentially representing a particularly sen-sitive locus with regard to FASD. Of note, only one vali-dated CpG was located in one of the protocadherin geneclusters (Pcdhb18), which were considerably enriched inprevious genome-wide studies of DNA methylation inindividuals with FASD [35, 36]. Given that these differ-ent studies only showed one overlapping probe, thiscould indicate higher variability within these gene clus-ters that may be associated with other variables notpresent in the current dataset, such as differences in age,body mass index, ethnicity, and socio-economic status.Of particular interest, we replicated the differentialDNA methylation patterns of the two genes involved indopamine signaling from the NDN cohort, the dopaminetransporter Slc6a3 and the dopamine receptor D4(Drd4). Given the important role of the dopaminergicsystem in brain development and its interactions withneuroendocrine and immune systems, these differencescould potentially reflect broader alterations to signalingpathways in the organism. Of note, the BEC of childrenexposed to alcohol during prenatal life and breastfeedingalso display altered DNA methylation patterns in thepromoter region of Drd4 [34]. Furthermore, several dis-orders previously associated with allelic variation andDNA methylation in this gene show either overlaps orco-morbidities with FASD, including attention deficithyperactivity disorder, bipolar disorder, anxiety disorder,schizophrenia, and substance abuse [54–64]. Although itis tempting to interpret these findings in the context ofPAE-induced deficits, DNA methylation differences inBEC likely do not fully reflect alterations in the centralnervous system. Nevertheless, it has been suggested thatBECs may act as a suitable surrogate tissue in humanstudies of DNA methylation, as they are also derivedfrom the ectoderm [65]. While we did not measure thesegenes in additional tissues, evidence from animal modelssuggests that PAE can cause lasting alterations to theepigenome of central nervous system tissues, and assuch, these results could potentially represent broaderassociations with epigenomic patterns in the brain [27].Although these findings represent the initial validationof genome-wide DNA methylation data in children withFASD, a few particularities of the KBHN cohort limit theinterpretability and generalizability of these results. Simi-lar to the original cohort, the KBHN replication cohortwas confounded by ethnicity, as the vast majority ofFASD cases were from First Nations communities, whilecontrols were mainly Caucasian. Given that geneticbackground influences DNA methylation patterns, dif-ferences between groups may have been, at least partly,due to ethnicity. Unfortunately, the KBHN cohort wastoo small to separate the groups into more ethnicallyhomogeneous subsets, a method we had previously usedto account for ethnicity-related differences in DNAmethylation [37]. As such, we performed linear modelingon the sites that had been previously identified in theNDN study, which were partially filtered for ethnicity-related differences during the analysis of the first cohort.However, some of the top differentially methylated genescould potentially be influenced by ethnicity differencesbetween groups in spite of our best efforts. For instance,three known polymorphisms are located within theFam59b locus (dbSNP minor allele frequencies:rs774397935, 1.04%; rs4665833, 5.1%; rs181971256,21.4%). Although, as of now, none of these are knownmethylation quantitative trait loci (mQTL), the Fam59bgene body contains several mQTLs in the developinghuman brain, and genetic variation outside the regioncould potentially influence DNA methylation levels [66].In addition, nearby genetic variation can also influenceDNA methylation patterns in the promoter of Drd4,which may be reflected in this cohort through the skewin ethnicity between groups [59]. Although theTable 3 Summarized results from the classification algorithmTraining set (NDN)AUC 0.977Accuracy 0.914Sensitivity 0.879Specificity 0.944Test set (KBHN)AUC 0.920Accuracy 0.833Sensitivity 0.75Specificity 0.917False positives 2False negatives 6PPV 0.900NPV 0.786Negative control (ASD)Accuracy 0.990Sensitivity NASpecificity 0.990False positives 1Lussier et al. Clinical Epigenetics  (2018) 10:5 Page 10 of 14frequencies of these alleles in First Nation populationshave not been assessed, genetic differences betweengroups could potentially influence DNA methylationlevels within this differentially methylated region.In addition to self-declared ethnicity, significant differ-ences in the primary caregiver were present betweengroups, as all controls lived with their biological families,while the majority of children with FASD were generallyin foster care or adoptive families. While the effects ofthis disparity on the epigenome are unclear, they couldinfluence DNA methylation patterns through a numberof factors, including nutrition, early-life adversity, andsocio-economic status [67]. Individuals with FASD alsotend to have life experiences different from those of typ-ically developing children, which include early life adver-sity (e.g., maltreatment or neglect), separation from thebiological family/placement in foster care (as occurred inour cohorts), poverty, and familial stress [13, 68].Importantly, both pre- and postnatal experiences areknown to play a role in early programming and thusmay also influence DNA methylation patterns. As such,it may be difficult to separate the impacts of PAE andenvironmental adversity, and studies evaluating FASDmay in many instances assess a combination of differentfactors and exposures, which is often the reality in thispopulation. Nevertheless, our findings demonstratedclear and replicable associations between FASD andDNA methylation patterns across two independent co-horts. We believe that our use of SVA to partiallyaccount for unknown factors that could influence DNAmethylation reduced some of the potential confoundsassociated with the cohort design. Future studies withlarger groups that are balanced for ethnicity, age, andadditional variables, including a focus on environmentalstress/adversity, will be necessary to tease out these dif-ferences and further validate our findings.Finally, we show here that DNA methylation patternscan be utilized as predictive variables for FASD in clin-ical populations. These findings complement and extendprevious studies that investigated different molecularand physiological markers to help screen children forpotential prenatal alcohol exposure, including alcoholmetabolites in mothers and children, circulating miRNAin mothers, and cardiac orienting response in children[69–72]. In particular, eye tracking measures have beenused in a small cohort of children to distinguish childrenwith FASD, ADHD, or typically developing controls withrelatively good accuracy [73]. In contrast to these stud-ies, we selected only individuals with confirmed FASDfrom the initial NDN training cohort to create a DNAmethylation-based predictor that was specific to individ-uals with an FASD diagnosis. The classification modelwas tuned to screen children at a higher risk for FASDwith high sensitivity and specificity in an attempt tobalance the false-positive and false-negative rates. Im-portantly, our results suggest that DNA methylation pre-dictors can achieve high accuracy in the classification ofindividuals with FASD versus controls across multiplecohorts. Moreover, the predictive algorithm appeared tobe largely independent of typical confounding factors,such as age, sex, ethnicity, and cell type composition ofthe samples. The predictor was also unbiased towardsindividuals with ASD and although there was no reportof FASD in this independent cohort, it is possible thatour reported false-positive could be due to an undiag-nosed FASD case [74]. Collectively, these results supportthe use of DNA methylation as a potential screening toolfor FASD.ConclusionsGiven the broad spectrum of cognitive, behavioral, andbiological deficits associated with PAE, FASD places animportant strain on both societal resources and theaffected individuals and families. As such, accuratescreening methods are necessary to identify children atrisk for FASD at an early age, when interventions aremost effective. Our findings provide an initial stepping-stone towards epigenetic biomarkers of FASD and couldpotentially be adapted for the development of relatedscreening tools for neurodevelopmental disorders. Valid-ation of these tools across different cohorts, with in-creased sample sizes, varying ages, ethnicities, and betterdocumented environmental exposures will be essentialto parse out the strongest associations and to developreliable epigenetic screening tools for FASD.Additional filesAdditional file 1: Supplementary tables. (XLSX 69 kb)Additional file 2: Supplementary figures. (DOCX 130 kb)Abbreviations450K array: Illumina HumanMethylation450 BeadChip array; ARND: Alcohol-related neurodevelopmental disorder; ASD: Autism spectrum disorder;BEC: Buccal epithelial cell; CpG: Cytosine-guanine dinucleotide;DNA: Deoxyribonucleic acid; FAS: Fetal alcohol syndrome; FASD: Fetalalcohol spectrum disorder; FDR: False discovery rate; GEO: Gene ExpressionOmnibus; KBHN: Kids Brain Health Network; NDN: NeuroDevNet;PAE: Prenatal alcohol exposure; PCDH: Protocadherin; RNA: Ribonucleic acid;ROC: Receiver operator characteristic; SV: Surrogate variable; SVA: Surrogatevariable analysisAcknowledgementsMSK is the Canada Research Chair in Social Epigenetics, the Sunny Hill BCLeadership Chair in Child Development, and a Senior Fellow of the CanadianInstitute For Advanced Research (CIFAR).FundingThis research is supported by the Kids Brain Health Network (formerlyNeuroDevNet), a Canadian Network of Center’s for Excellence and a programof the federal government to advance science and technology. AAL issupported by a Developmental Neurosciences Research Training award fromBrain Canada & NeuroDevNet. JW is supported by grants from the USLussier et al. Clinical Epigenetics  (2018) 10:5 Page 11 of 14National Institutes of Health/National Institute on Alcohol Abuse andAlcoholism (R37 AA007789 and RO1 AA022460); and the CanadianFoundation for Fetal Alcohol Research. PP is supported by an NSERCDiscovery Grant.Availability of data and materialsThe KBHN DNA methylation data are deposited into the GEO databaseunder accession number GSE109042. The predictive algorithm can be foundand used here: https://fasdpredictor.shinyapps.io/fasdpredictorapp/.Authors’ contributionsAAL all bioinformatic analyses and wrote the manuscript. AMM performedthe pyrosequencing. JS and AEC collected the samples. AEC, JW, PP, JNR,and MSK helped with the study design, interpretation, and writing. Allauthors read and approved the final manuscript.Ethics approval and consent to participateEthics for this study were reviewed and approved by the “Children’s andWomen’s Research Ethics Board – Clinical” (H10-01149). All experimentalprocedures were reviewed and approved by the University of Manitoba andthe University of British Columbia.Consent for publicationWritten informed consent was obtained from a parent or legal guardian andassent was obtained from each child before study participation.Competing interestsThe authors declare that they have no competing interests.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Department of Medical Genetics, Centre for Molecular Medicine andTherapeutics, British Columbia Children’s Hospital Research Institute,University of British Columbia, Vancouver, British Columbia, Canada.2Department of Cellular and Physiological Sciences, Life Sciences Institute,University of British Columbia, Vancouver, British Columbia, Canada.3Department of Pediatrics and Child Health, Faculty of Medicine, Universityof Manitoba, Winnipeg, Manitoba, Canada. 4Department of Biochemistry andMedical Genetics, Faculty of Medicine, University of Manitoba, Winnipeg,Manitoba, Canada. 5Department of Biomedical and Molecular Sciences,Centre for Neuroscience Studies, Queen’s University, Kingston, Ontario,Canada. 6Michael Smith Laboratories, University of British Columbia,Vancouver, British Columnbia, Canada. 7Human Early Learning Partnership,University of British Columbia, Vancouver, British Columbia, Canada.8Department of Psychiatry, University of British Columbia, Vancouver, BritishColumbia, Canada.Received: 22 August 2017 Accepted: 4 January 2018References1. May PA, Gossage JP, Kalberg WO, Robinson LK, Buckley D, Manning M, et al.Prevalence and epidemiologic characteristics of FASD from various researchmethods with an emphasis on recent in-school studies. Dev Disabil Res Rev.2009;15:176–92.2. May PA, Baete A, Russo J, Elliott AJ, Blankenship J, Kalberg WO, et al.Prevalence and characteristics of fetal alcohol spectrum disorders. Pediatrics.2014;134:855–66. https://doi.org/10.1542/peds.2013-3319.3. May PA, Keaster C, Bozeman R, Goodover J, Blankenship J, Kalberg WO,et al. Prevalence and characteristics of fetal alcohol syndrome and partialfetal alcohol syndrome in a Rocky Mountain Region City. Drug AlcoholDepend. 2015;155:118–27. https://doi.org/10.1016/j.drugalcdep.2015.08.006.4. Zhang X, Sliwowska JH, Weinberg J. Prenatal alcohol exposure and fetalprogramming: effects on neuroendocrine and immune function. Exp BiolMed. 2005;230:376–88.5. Pei J, Denys K, Hughes J, Rasmussen C. Mental health issues in fetal alcoholspectrum disorder. J Ment Health. 2011;20:473–83.6. Mattson SN, Crocker N, Nguyen TT. Fetal alcohol spectrum disorders:neuropsychological and behavioral features. Neuropsychol Rev. 2011;21:81–101.7. Jones KL, Smith DW. Recognition of the fetal alcohol syndrome in earlyinfancy. Lancet. 1973;302:999–1001. https://doi.org/10.1016/S0140-6736(73)91092-1.8. Astley SJ, Clarren SK. Diagnosing the full spectrum of fetal alcohol-exposed individuals: introducing the 4-digit diagnostic code. AlcoholAlcohol. 2000;35:400–10.9. Jacobson SW, Jacobson JL, Stanton ME, Meintjes EM, Molteno CD.Biobehavioral markers of adverse effect in fetal alcohol spectrum disorders.Neuropsychol Rev. 2011;21:148–66.10. Pollard I. Neuropharmacology of drugs and alcohol in mother and fetus.Semin Fetal Neonatal Med. 2007;12:106–13.11. Senturias Y, Baldonado M. Fetal spectrum disorders: an overview of ethicaland legal issues for healthcare providers. Curr Probl Pediatr Adolesc HealthCare. 2014;44:102–4. doi:https://doi.org/10.1016/j.cppeds.2013.12.01012. Paley B, O’Connor MJ. Behavioral interventions for children and adolescentswith fetal alcohol spectrum disorders. Alcohol Res Heal. 2011;34:64–75.http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3860556/13. Streissguth AP, Bookstein F, Barr H, Sampson P, O’Malley K, Young J. Riskfactors for adverse life outcomes in fetal alcohol syndrome and fetal alcoholeffects. J Dev Behav Pediatr. 2004;25:228–38. https://doi.org/10.1097/00004703-200408000-00002.14. Fox SE, Levitt P, Nelson CA III. How the timing and quality of earlyexperiences influence the development of brain architecture. Child Dev.2010;81:28–40. https://doi.org/10.1111/j.1467-8624.2009.01380.x.15. Russell M, Martier SS, Sokol RJ, Mudar P, Jacobson S, Jacobson J. Detectingrisk drinking during pregnancy: a comparison of four screeningquestionnaires. Am J Public Health. 1996;86:1435–9. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1380656/16. Jones TB, Bailey BA, Sokol RJ. Alcohol use in pregnancy: insights inscreening and intervention for the clinician. Clin Obstet Gynecol. 2013;56:114–23. https://doi.org/10.1097/GRF.0b013e31827957c0.17. Burns E, Gray R, Smith LA. Brief screening questionnaires to identify problemdrinking during pregnancy: a systematic review. Addiction. 2010;105:601–14.https://doi.org/10.1111/j.1360-0443.2009.02842.x.18. Concheiro-Guisan A, Concheiro M. Bioanalysis during pregnancy: recentadvances and novel sampling strategies. Bioanalysis. 2014;6:3133–53. https://doi.org/10.4155/bio.14.278.19. Cabarcos P, Álvarez I, Tabernero MJ, Bermejo AM. Determination of directalcohol markers: a review. Anal Bioanal Chem. 2015;407:4907–25. https://doi.org/10.1007/s00216-015-8701-7.20. Bird A. Perceptions of epigenetics. Nature. 2007;447:396–8.21. Meaney MJ. Epigenetics and the biological definition of gene Xenvironment interactions. Child Dev. 2010;81:41–79.22. Henikoff S, Greally JM. Epigenetics, cellular memory and gene regulation.Curr Biol. 2016;26:R644–8. http://dx.doi.org/10.1016/j.cub.2016.06.01123. Boyce WT, Kobor MS. Development and the epigenome: the “synapse” ofgene-environment interplay. Dev Sci. 2015;18:1–23. https://doi.org/10.1111/desc.12282.24. Joubert BR, Håberg SE, Nilsen RM, Wang X, Vollset SE, Murphy SK, et al. 450Kepigenome-wide scan identifies differential DNA methylation in newbornsrelated to maternal smoking during pregnancy. Environ Health Perspect.2012;120:1425–31. https://doi.org/10.1289/ehp.1205412.25. Heijmans BT, Tobi EW, Stein AD, Putter H, Blauw GJ, Susser ES, et al.Persistent epigenetic differences associated with prenatal exposure tofamine in humans. Proc Natl Acad Sci U S A. 2008;105:17046–9. https://doi.org/10.1073/pnas.0806560105.26. Reese SE, Zhao S, Wu MC, Joubert BR, Parr CL, Håberg SE, et al. DNAMethylation score as a biomarker in newborns for sustained maternalsmoking during pregnancy. Environ Health Perspect. 2017;125:760–6.https://doi.org/10.1289/EHP333.27. Lussier AA, Weinberg J, Kobor MS. Epigenetics studies of fetal alcoholspectrum disorder: where are we now? Epigenomics. 2017;9:291–311.https://doi.org/10.2217/epi-2016-0163.28. Chater-Diehl EJ, Laufer BI, Castellani CA, Alberry BL, Singh SM. Alteration ofgene expression, DNA methylation, and histone methylation in free radicalscavenging networks in adult mouse hippocampus following fetal alcoholexposure. PLoS One. 2016;11:e0154836.29. Laufer BI, Mantha K, Kleiber ML, Diehl EJ, Addison SMF, Singh SM. Long-lasting alterations to DNA methylation and ncRNAs could underlie theLussier et al. Clinical Epigenetics  (2018) 10:5 Page 12 of 14effects of fetal alcohol exposure in mice. Dis Model Mech. 2013;6:977–92.https://doi.org/10.1242/dmm.010975.30. Liu Y, Balaraman Y, Wang G, Nephew KP, Zhou FC. Alcohol exposure altersDNA methylation profiles in mouse embryos at early neurulation.Epigenetics. 2009;4:500–11.31. Hicks SD, Middleton FA, Miller MW. Ethanol-induced methylation of cellcycle genes in neural stem cells. J Neurochem. 2010;114:1767–80.32. Zhou FC, Chen Y, Love A. Cellular DNA methylation program duringneurulation and its alteration by alcohol exposure. Birth Defects Res Part A -Clin Mol Teratol. 2011;91:703–15.33. Otero NKH, Thomas JD, Saski CA, Xia X, Kelly SJ. Cholinesupplementation and DNA methylation in the hippocampus andprefrontal cortex of rats exposed to alcohol during development.Alcohol Clin Exp Res. 2012;36:1701–9.34. Fransquet PD, Hutchinson D, Olsson CA, Wilson J, Allsop S, Najman J, et al.Perinatal maternal alcohol consumption and methylation of the dopaminereceptor DRD4 in the offspring: the triple B study. Environ Epigenetics. 2016;2:dvw023. http://dx.doi.org/10.1093/eep/dvw02335. Laufer BI, Kapalanga J, Castellani CA, Diehl EJ, Yan L, Singh SM. AssociativeDNA methylation changes in children with prenatal alcohol exposure.Epigenomics. 2015;7 August:1–16. https://doi.org/10.2217/epi.15.60.36. Portales-Casamar E, Lussier AA, Jones MJ, MacIsaac JL, Edgar RD, Mah SM,et al. DNA methylation signature of human fetal alcohol spectrum disorder.Epigenetics Chromatin. 2016;9:81–101.37. Portales-Casamar E, Lussier AA, Jones MJ, MacIsaac JL, Edgar RD, MahSM, et al. DNA methylation signature of human fetal alcohol spectrumdisorder. Epigenetics Chromatin. 2016;9:25. https://doi.org/10.1186/s13072-016-0074-4.38. Reynolds JN, Weinberg J, Clarren S, Beaulieu C, Rasmussen C, Kobor M, et al.Fetal alcohol spectrum disorders: gene-environment interactions, predictivebiomarkers, and the relationship between structural alterations in the brainand functional outcomes. Semin Pediatr Neurol. 2011;18:49–55.39. Chudley AE, Conry J, Cook JL, Loock C, Rosales T, LeBlanc N. Fetal alcoholspectrum disorder: Canadian guidelines for diagnosis. Can Med Assoc J.2005;172(5 Suppl):S1–21.40. Price ME, Cotton AM, Lam LL, Farré P, Emberly E, Brown CJ, et al. Additionalannotation enhances potential for biologically-relevant analysis of theIllumina Infinium HumanMethylation450 BeadChip array. EpigeneticsChromatin. 2013;6:4. https://doi.org/10.1186/1756-8935-6-4.41. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-CabreroD, et al. A Beta-mixture quantile normalization method for correcting probedesign bias in Illumina Infinium 450k DNA methylation data. Bioinformatics.2012;29:189–96.42. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package forremoving batch effects and other unwanted variation in high-throughputexperiments. Bioinformatics. 2012;28:882–3.43. Du P, Zhang X, Huang C-C, Jafari N, Kibbe WA, Hou L, et al. Comparison ofBeta-value and M-value methods for quantifying methylation levels bymicroarray analysis. BMC Bioinformatics. 2010;11:587. https://doi.org/10.1186/1471-2105-11-587.44. Smith AK, Kilaru V, Klengel T, Mercer KB, Bradley B, Conneely KN, et al. DNAextracted from saliva for methylation studies of psychiatric traits: evidencetissue specificity and relatedness to brain. Am J Med Genet Part BNeuropsychiatr Genet. 2015;168:36–44.45. Smyth GK. Linear models and empirical bayes methods for assessingdifferential expression in microarray experiments. Stat Appl Genet Mol Biol.2004;3:Article3.46. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical andpowerful approach to multiple testing. J R Stat Soc Ser B. 1995;57:289–300.https://doi.org/10.2307/2346101.47. Berko ER, Suzuki M, Beren F, Lemetre C, Alaimo CM, Calder RB, et al. Mosaicepigenetic dysregulation of ectodermal cells in autism spectrum disorder.PLoS Genet. 2014;10:e1004402. https://doi.org/10.1371/journal.pgen.1004402.48. Kuhn M. Building predictive models in R using the caret package. J StatSoftware. 2008;1(5) https://doi.org/10.18637/jss.v028.i05.49. Breton CV, Marsit CJ, Faustman E, Nadeau K, Goodrich JM, Dolinoy DC, et al.Small-magnitude effect sizes in epigenetic end points are important inchildren’s environmental health studies: the Children’s environmental healthand disease prevention research center’s epigenetics working group.Environ Health Perspect. 2017;125:511–26. https://doi.org/10.1289/EHP595.50. Ladd-Acosta C, Hansen KD, Briem E, Fallin MD, Kaufmann WE, Feinberg AP.Common DNA methylation alterations in multiple brain regions in autism.Mol Psychiatry. 2014;19:862–71.51. Rakyan VK, Down TA, Balding DJ, Beck S. Epigenome-wide associationstudies for common human diseases. Nat Rev Genet. 2011;12:529–41.52. Lappalainen T, Greally JM. Associating cellular epigenetic models withhuman phenotypes. Nat Rev Genet. 2017;18:441–51. http://dx.doi.org/10.1038/nrg.2017.3253. Liu Y, Aryee MJ, Padyukov L, Fallin MD, Hesselberg E, Runarsson A, et al.Epigenome-wide association data implicate DNA methylation as anintermediary of genetic risk in rheumatoid arthritis. Nat Biotechnol. 2013;31:142–7. https://doi.org/10.1038/nbt.2487.54. Sánchez-Mora C, Ribasés M, Casas M, Bayés M, Bosch R, Fernàndez-CastilloN, et al. Exploring DRD4 and its interaction with SLC6A3 as possible riskfactors for adult ADHD: a meta-analysis in four European populations. Am JMed Genet Part B, Neuropsychiatr Genet. 2011;156B:600–12.55. Dadds MR, Schollar-Root O, Lenroot R, Moul C, Hawes DJ. Epigeneticregulation of the DRD4 gene and dimensions of attention-deficit/hyperactivity disorder in children. Eur Child Adolesc Psychiatry. 2016;25:1081–9. https://doi.org/10.1007/s00787-016-0828-3.56. Ji H, Wang Y, Jiang D, Liu G, Xu X, Dai D, et al. Elevated DRD4 promotermethylation increases the risk of Alzheimer’s disease in males. Mol Med Rep.2016;14:2732–8.57. Cheng J, Wang Y, Zhou K, Wang L, Li J, Zhuang Q, et al. Male-specificassociation between dopamine receptor D4 gene methylation andschizophrenia. PLoS One. 2014;9:e89128. https://doi.org/10.1371/journal.pone.008912858. Kordi-Tamandani DM, Sahranavard R, Torkamanzehi A. Analysis ofassociation between dopamine receptor genes’ methylation and theirexpression profile with the risk of schizophrenia. Psychiatr Genet. 2013;23:183–7. http://journals.lww.com/psychgenetics/Abstract/2013/10000/Analysis_of_association_between_dopamine_receptor.1.aspx.59. Docherty SJ, Davis OSP, Haworth CMA, Plomin R, D’Souza U, Mill J. Agenetic association study of DNA methylation levels in the DRD4 generegion finds associations with nearby SNPs. Behav Brain Funct. 2012;8:31.https://doi.org/10.1186/1744-9081-8-31.60. Ptáček R, Kuželová H, Stefano GB. Dopamine D4 receptor gene DRD4 andits association with psychiatric disorders. Med Sci Monit. 2011;17:RA215–20.https://doi.org/10.12659/MSM.881925.61. Bau CH, Almeida S, Costa FT, Garcia CE, Elias EP, Ponso AC, et al. DRD4 andDAT1 as modifying genes in alcoholism: interaction with novelty seeking onlevel of alcohol consumption. Mol Psychiatry. 2001;6:7–9.62. Zhang H, Herman AI, Kranzler HR, Anton RF, Zhao H, Zheng W, et al. Array-based profiling of DNA methylation changes associated with alcoholdependence. Alcohol Clin Exp Res. 2013;37(Suppl 1):E108–15.63. Faraone SV, Bonvicini C, Scassellati C. Biomarkers in the diagnosis ofADHD––promising directions. Curr Psychiatry Rep. 2014;16:497. https://doi.org/10.1007/s11920-014-0497-1.64. Chen D, Liu F, Shang Q, Song X, Miao X, Wang Z. Association betweenpolymorphisms of DRD2 and DRD4 and opioid dependence: evidence fromthe current studies. Am J Med Genet Part B Neuropsychiatr Genet. 2011;156:661–70. https://doi.org/10.1002/ajmg.b.31208.65. Lowe R, Gemma C, Beyan H, Hawa MI, Bazeos A, Leslie RD, et al. Buccals arelikely to be a more informative surrogate tissue than blood for epigenome-wide association studies. Epigenetics. 2013;8:445–54.66. Hannon E, Spiers H, Viana J, Pidsley R, Burrage J, Murphy TM, et al.Methylation QTLs in the developing brain and their enrichment inschizophrenia risk loci. Nat Neurosci. 2015;19:48–54. https://doi.org/10.1038/nn.4182.67. Esposito EA, Jones MJ, Doom JR, MacIsaac JL, Gunnar MR, Kobor MS.Differential DNA methylation in peripheral blood mononuclear cells inadolescents exposed to significant early but not later childhood adversity.Dev Psychopathol. 2016;28 4pt2:1385–99. https://doi.org/10.1017/S0954579416000055.68. Coggins TE, Timler GR, Olswang LB. A state of double jeopardy: impact ofprenatal alcohol exposure and adverse environments on the socialcommunicative abilities of school-age children with fetal alcohol spectrumdisorder. Lang Speech Hear Serv Sch. 2007;38:117–27. http://dx.doi.org/10.1044/0161-1461(2007/012)69. Balaraman S, Schafer JJ, Tseng AM, Wertelecki W, Yevtushok L, Zymak-Zakutnya N, et al. Plasma miRNA profiles in pregnant women predict infantLussier et al. Clinical Epigenetics  (2018) 10:5 Page 13 of 14outcomes following prenatal alcohol exposure. PLoS One. 2016;11:e0165081. https://doi.org/10.1371/journal.pone.0165081.70. Mesa DA, Kable JA, Coles CD, Jones KL, Yevtushok L, Kulikovsky Y, et al. Theuse of cardiac orienting responses as an early and scalable biomarker ofalcohol-related neurodevelopmental impairment. Alcohol Clin Exp Res.2017;41:128–38. https://doi.org/10.1111/acer.13261.71. Goh PK, Doyle LR, Glass L, Jones KL, Riley EP, Coles CD, et al. A decision treeto identify children affected by prenatal alcohol exposure. J Pediatr. 2016;177:121–127.e1. https://doi.org/10.1016/j.jpeds.2016.06.04772. McQuire C, Paranjothy S, Hurt L, Mann M, Farewell D, Kemp A. Objectivemeasures of prenatal alcohol exposure: a systematic review. Pediatrics. 2016;138 https://doi.org/10.1542/peds.2016-0517.73. Tseng P-H, Cameron IGM, Pari G, Reynolds JN, Munoz DP, Itti L. High-throughput classification of clinical populations from natural viewingeye movements. J Neurol. 2013;260:275–84. https://doi.org/10.1007/s00415-012-6631-2.74. Kelleher E, Corvin A. Overlapping etiology of neurodevelopmental disorders.In: The genetics of Neurodevelopmental disorders: Wiley; 2015. p. 29–48.https://doi.org/10.1002/9781118524947.ch2.•  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:Lussier et al. Clinical Epigenetics  (2018) 10:5 Page 14 of 14


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