UBC Faculty Research and Publications

Maternal blood contamination of collected cord blood can be identified using DNA methylation at three… Morin, Alexander M; Gatev, Evan; McEwen, Lisa M; MacIsaac, Julia L; Lin, David T S; Koen, Nastassja; Czamara, Darina; Räikkönen, Katri; Zar, Heather J; Koenen, Karestan; Stein, Dan J; Kobor, Michael S; Jones, Meaghan J Jul 25, 2017

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

Item Metadata


52383-13148_2017_Article_370.pdf [ 2.03MB ]
JSON: 52383-1.0362000.json
JSON-LD: 52383-1.0362000-ld.json
RDF/XML (Pretty): 52383-1.0362000-rdf.xml
RDF/JSON: 52383-1.0362000-rdf.json
Turtle: 52383-1.0362000-turtle.txt
N-Triples: 52383-1.0362000-rdf-ntriples.txt
Original Record: 52383-1.0362000-source.json
Full Text

Full Text

METHODOLOGY Open AccessMaternal blood contamination of collectedcord blood can be identified using DNAmethylation at three CpGsAlexander M. Morin1, Evan Gatev1, Lisa M. McEwen1, Julia L. MacIsaac1, David T. S. Lin1, Nastassja Koen2,Darina Czamara3, Katri Räikkönen4, Heather J. Zar5, Karestan Koenen6, Dan J. Stein2, Michael S. Kobor1,7and Meaghan J. Jones1*AbstractBackground: Cord blood is a commonly used tissue in environmental, genetic, and epigenetic population studiesdue to its ready availability and potential to inform on a sensitive period of human development. However, theintroduction of maternal blood during labor or cross-contamination during sample collection may complicatedownstream analyses. After discovering maternal contamination of cord blood in a cohort study of 150 neonatesusing Illumina 450K DNA methylation (DNAm) data, we used a combination of linear regression and random forestmachine learning to create a DNAm-based screening method. We identified a panel of DNAm sites thatcould discriminate between contaminated and non-contaminated samples, then designed pyrosequencingassays to pre-screen DNA prior to being assayed on an array.Results: Maternal contamination of cord blood was initially identified by unusual X chromosome DNAmethylation patterns in 17 males. We utilized our DNAm panel to detect contaminated male samples anda proportional amount of female samples in the same cohort. We validated our DNAm screening methodon an additional 189 sample cohort using both pyrosequencing and DNAm arrays, as well as 9 publically available cordblood 450K data sets. The rate of contamination varied from 0 to 10% within these studies, likely related to collectionspecific methods.Conclusions: Maternal blood can contaminate cord blood during sample collection at appreciable levels across multiplestudies. We have identified a panel of markers that can be used to identify this contamination, either post hoc after DNAmarrays have been completed, or in advance using a targeted technique like pyrosequencing.Keywords: Cord blood, Contamination, DNA methylation, 450K, Genotyping, Maternal blood, Blood bankingBackgroundNeonatal blood from the umbilical cord at the time of de-livery is increasingly being collected for both research andmedical purposes. In research, interest in the developmen-tal origins of health and disease has made cord blood apopular choice for genetic, epigenetic, and environmentalstudies [1]. Cord blood has several physiological differ-ences from adult blood, such as the presence of nucleatedred blood cells and fetal hemoglobin, and is an excellentwindow into the in utero environment, free of confound-ing post-natal exposures [2, 3]. Medically, cord blood isbanked for transplantation as a source of progenitor cellsfor replenishing the hematopoietic system [4]. Cord bloodcan be collected after caesarian or vaginal delivery, eitherpreceding or following delivery of the placenta. Both pro-cesses typically involve venipuncture of the umbilical ar-tery and collection into a blood bag by gravity [4].Problems can arise when the collected cord blood be-comes contaminated with other cells, most frequently ma-ternal white blood cells [5, 6]. In some cases, maternalblood cells may enter fetal circulation through the placenta.Previous studies have shown that such contamination can* Correspondence: mjones@cmmt.ubc.ca1Centre for Molecular Medicine and Therapeutics, BC Children’s Hospital,Department of Medical Genetics, University of British Columbia, 950 W 28thAve, Vancouver, BC V5Z 4H4, CanadaFull list of author information is available at the end of the article© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.Morin et al. Clinical Epigenetics  (2017) 9:75 DOI 10.1186/s13148-017-0370-2occur relatively frequently, estimated at 2–20% of collectedsamples, but it makes up a very small fraction of fetal blood,with ~10−4 to 10−5 fetal nucleated cells estimated as mater-nal [7–10]. This small amount of contamination shouldhave negligible effects on the assessment of DNA or RNA.However, contamination in larger amounts, which couldoccur through mixing of blood during collection, is ofgreater concern.Previous techniques for identifying larger amounts ofmaternal contributions to collected cord blood have in-cluded PCR on highly variable mini satellites or specificpolymorphic alleles and fluorescent in situ hybridization(FISH) or TaqMan assay to detect two X chromosomes[7, 9, 11]. Neither technique is universally unambiguous,as mother/child pairs may not be informative for tar-geted genetic variants, and FISH or TaqMan analysis canonly be performed on male children, as they differentiateXX maternal cells from XY child cells [5, 7–9, 11, 12].DNA methylation (DNAm) is another potentialmethod by which to identify maternal contaminationof cord blood, as it is highly different between new-borns and adults [13, 14]. DNAm is an epigeneticmark where a methyl group is covalently bound toDNA, primarily at CpG dinucleotides. It is stableunder a variety of collection and storage methods,and often employed to identify epigenetic patternsassociated with specific environmental or develop-mental exposures [15–17]. If present at considerableamounts, maternal contamination of cord blood is ofconcern to studies of DNAm data, as it could masksignals from cord blood or introduce signals presentin the maternal blood. This contamination would bedifferentially observable in male and female children.Since the X chromosome has highly distinct male-and female-specific patterns of DNAm, XX bloodfrom mothers would be more apparent when mixedwith XY male children than XX females.In this study, we initially observed a high propor-tion of cord blood samples evidently contaminatedwith maternal blood in the quality control phase ofan epigenome-wide association study. Using DNAmdata from the genome-wide Illumina 450K array, wecreated a method by which to identify contaminatedsamples using 10 CpGs that correctly discriminatedcontamination status. We also showed that a subsetof three CpGs were sufficient for screening DNAusing pyrosequencing. While it cannot accuratelypredict the proportion of contamination, this processis capable of detecting levels that appreciably affectthe output of common methods for assessment ofDNA methylation. This method can be used to pre-screen prior to running the samples on a DNAmarray, or in cases where it is important to identifymaternal contamination, such as cord blood banking.ResultsDetection of maternal contaminationOur first indication of potential maternal contamin-ation of cord blood came from unusual patterns inthe DNAm data during quality control. Quality con-trol MDS plots of un-normalized data showed 17 of86 male participants’ DNAm profiles clustered withfemale children or in between male and female,which was confirmed by plotting principal compo-nents 1 and 2 (Fig. 1a). Investigating the X and Ychromosome probes prior to probe filtering andnormalization in more detail, we observed that thesemale children showed a DNAm pattern on the Xchromosome that was intermediate between the nor-mal male and normal female patterns (Fig. 1b). To-gether, this was suggestive of female blood beingmixed with the cord blood of the newborn males,which could have occurred across the placenta dur-ing labor or after delivery.Investigation of the cord blood collection procedurerevealed that maternal contamination of the resultingcord blood after delivery was the most likely hypothesisto explain these unexpected DNAm patterns. With thisinsight, we then divided samples into three groups basedon principal component 2 (PC2) of the full data andDNAm at cg05533223 on the X chromosome. As ini-tially observed, PC2 clearly separated male from femalesamples, but was not associated with the major variablesin the sub-study, ethnicity (ANOVA p > 0.8) or traumaexposure (t test p > 0.3). The CpG used, cg05533223, inthe X-inactivation specific transcript (XIST) should behighly methylated in males and ~50% methylated in fe-males [18]. Based on these two criteria, 17 males werecontaminated (C), 64 were not contaminated (NC) and5 were unclear (U) (Additional file 1: Figure S1 in Add-itional file 1). As we relied on X chromosome methyla-tion levels, which would not differ between XX mothersand their XX daughters, this method was only applicableto XY male children. Since it called approximately 20%of male samples contaminated, we hypothesized that asimilar proportion (approximately 13/64) of female chil-dren would also be contaminated. There was no reasonto expect that the amount of maternal contaminationdue to sample collection would differ by sex, as all col-lection occurred in the same hospital using the samestandard procedures.Using epigenetic age and genotyping no-calls to identifycontaminated samplesWe thus sought a way of discriminating contaminatedfemales using other data. First, we tested epigenetic ageby comparing the C and NC male samples using pub-lished methods [19]. As epigenetic age of cord bloodsamples has been demonstrated to be below 1 year, weMorin et al. Clinical Epigenetics  (2017) 9:75 Page 2 of 9hypothesized that mixing with maternal blood would re-sult in an increase in epigenetic age of the whole sample.Though the DNAm age means were significantly differ-ent between C and NC, (two-sided Student’s t testp = 0.025), the large confidence intervals (−14.714880 to−1.077678) meant that this was not a sufficiently accur-ate test, despite the identification of at least 4 femaleswho were likely contaminated (Additional file 1: FigureS2A). Using a similar method that estimates gestationalage from DNAm data, we found similarly poor predict-ive value (Additional file 1: Figure S2B) [20].Next, we used genotyping data to see whether a highernumber of “no calls” from the Illumina PsychChip wasassociated with contamination. Our rationale was thatmixing two blood samples together, even if geneticallyrelated, would result in a higher number of un-callablegenotypes with signals falling between the three normalgenotype groups. While performing better than epigen-etic age, the extreme confidence intervals (34,281.73–10,811.97, p value <0.001), difference in basal number ofno calls between males and females, and potential lackof genotyping data in other studies meant, in our opin-ion, this was not a suitable discriminatory screen either(Additional file 1: Figure S2C).Identification of CpGs indicative of contaminationWe next reasoned that since DNAm has been shown tobe highly different between neonates and adults, it mightserve to discriminate contaminated samples. Using linearmodeling followed by a random forests approach, wedetermined that 10 CpGs could discriminate between con-taminated and non-contaminated male samples at 99%confidence (Additional file 1: Figure S2A, Additional file 1:Table S2). Importantly, the calculated thresholds for identi-fying contaminated samples were sensitive to normalizationmethod, and so we present thresholds for two commonnormalization methods; SWAN and BMIQ [21, 22].To identify the contaminated female samples, we ap-plied the thresholds of these 10 CpGs to all of our samples(Fig. 2b). This method identified 13 females as contami-nated, including the 4 previously identified by epigeneticage, in line with the approximately 20% expected based onproportion of contaminated males, and all 5 unclear maleswere categorized as non-contaminated (Fig. 2b). Thisshowed that these 10 CpGs were sufficient for screeningpreviously generated DNAm data to identify maternalblood contamination in male and female children. How-ever, we wished to refine this panel so that samples couldbe screened prior to being run on an array in cases wherecontamination might be expected.Verification of screening CpGs using pyrosequencingTo ensure that this pre-screening method was quick andcost-effective, we focused on pyrosequencing and re-duced the 10 identified CpGs to 3. These three CpGshad the best discrimination between contaminated andnon-contaminated male samples and were sites forwhich a robust pyrosequencing assay could be designed(Table S2). After selecting cg25556035, cg15931839, andcg02812891, we performed pyrosequencing of these 3−0.10−−0.090 −0.085 −0.080 −0.075 −0.070PC1PC2 FemaleMale0.00 0.25 0.50 0.75 1.00012301230123X chromosome beta valuedensityUncontaminated maleContaminated maleFemaleBAFig. 1 Principal component and X chromosome DNA methylation (DNAm) patterns revealed maternal blood contamination in cord blood. aPlotting the first two principal components of 450K DNAm data identified a number of male samples with DNAm patterns similar to femaleparticipants or intermediate between male and female. b Examining the distribution of X chromosome DNAm beta values in these samplesrevealed that the intermediate male samples clearly showed patterns indicative of a mixture of male (top) and female (bottom) distributionsMorin et al. Clinical Epigenetics  (2017) 9:75 Page 3 of 9sites on our original 150 samples (Fig. 2c). Interestingly,the assay that measured cg02812891 also measuredcg13138089 as these CpGs are in close proximity. Asthese two CpGs were strongly correlated (r = 0.977)within the assay, we deemed cg13138089 to be redun-dant for the purpose of designing a minimal screen,though other groups may consider its inclusion in thescreening process. A strict cut-off requiring all 3 CpGsto surpass the contamination threshold identified 14male samples as contaminated, all consistent with thearray and X chromosome data. A less stringent cut-offof 2 CpGs identified 17 male samples, with 1 false posi-tive and 1 false negative. In females, the less stringent2 CpG cut-off predicted 11 of the 13 samples calledcontaminated using the 450K array data, and the strictmethod predicted 6; neither had false positives. Whilethis screen is not as accurate as the 10 CpG methodfrom the 450K array data, it is sufficient to identify andeliminate the worst contaminated samples. All predic-tion methods and results are summarized in Fig. 3.Validation on second data setTo validate this screening method, 189 additional sam-ples from the same cohort study were screened usingthe pyrosequencing assays. Eighteen males and 15 fe-males were identified as contaminated using the 2 CpGcut-off, again approximating the 20% contamination ratewe initially observed (Fig. 4a). We ran all 156 uncontam-inated samples and 2 contaminated male samples on theEPIC array. We chose male samples as validation, as wecould use sex-specific differences in DNA methylation atXIST on the X chromosome as independent confirm-ation of our screening method. Initial principal compo-nents plots showed that only the two knowncontaminated male samples demonstrated the inter-mediate DNAm pattern indicative of contamination (Fig.4b). We then examined the 10 CpGs identified in ourdiscovery data set and, as expected, only the 2 knownmale samples were identified as contaminated (Fig. 4c).This supports that 3 CpGs are sufficient to correctlyeliminate contaminated samples prior to running onan array.Validation on publicly available dataTo address the frequency with which maternal blood con-tamination occurs in DNAm studies, we used nine pub-lished cord blood DNAm data sets (GSE30870,GSE54399, GSE62924, GSE66459, GSE74738, GSE79056,GSE80310, GSE83334, and PREDO). We applied our posthoc maternal contamination assay with 10 CpGs acrossthese studies and identified 2 data sets with contaminatedsamples (Fig. 5). GSE54399 had 2/24 (~10%, 1 male and 1female) samples indicating contamination, and PREDO 8/834 (~1%, 4 males and 4 females). Across all studies, ma-ternal blood contamination was present at a frequency ofapproximately 1% (10/1014), but the study-specific patternsuggests that contamination may be related to specificcollection methods.Finally, we examined our discovery samples, validationsamples, and the publicly available data together to de-termine whether our 10 CpG method was affected bybatch or technology. We compared the residuals of eachsample’s methylation to thresholds of each of our 10CpGs (Additional file 1: Figure S3). We observed similardistributions for each CpG in all studies except for thevalidation cohort, the only one to use the EPIC array.These data were normalized with methods consistentwith the GEO data, so the effect is due to technologycg02812891 cg12634306 cg13138089 cg15645660cg15931839 cg16617301 cg19509778 cg24767131cg25241559cg255560350.000.250.500.751. categoryBeta valueN C UCordAdultN C UCordAdultN C UCordAdultN C UCordAdultN C UCordAdult0510Contamination category# CpGs at Threshold450k data0123Contamination category# CpGs at ThresholdPyrosequencing dataN C U N C UACBFemaleMaleFig. 2 DNA methylation at 10 autosomal CpGs was sufficient tocorrectly identify all known contaminated male samples, and found 13contaminated female samples. a The 10 CpGs selected by the randomforest method clearly separate cord and adult samples, and also clearlydiscriminate non-contaminated (N) from contaminated (C) male samples,and divide unknown (U) samples into two groups. b Counting thenumber of sites over thresholds per sample (x axis), contamination wascalled if at least 5 of the 10 CpGs were above the threshold. Unclearmales were all non-contaminated, and 13 females were identified asbeing contaminated. c A subset of 3 out of the 10 CpGs can be used forpyrosequencing screening. Two thresholds are shown—one requiringtwo of the three CpGs to be above the threshold to be calledcontaminated (yellow), and one requiring all three (red)Morin et al. Clinical Epigenetics  (2017) 9:75 Page 4 of 9and not normalization method. This suggests that, des-pite successfully identifying the known contaminatedsamples in our EPIC cohort, the 10 CpG method is in-fluenced by array technology and thus using all 10 CpGsis highly recommended when working with EPIC data.DiscussionThe popularity of cord blood collection for both re-search and medical purposes means that it is more im-portant than ever to ensure that the collected blood isfree of contaminating maternal white blood cells. In thisstudy, we initially observed unusual patterns in a pre-normalization MDS plot driven by X chromosomeDNAm in male cord blood samples. After consulting thecollection procedure, we strongly suspected that mater-nal blood contamination was present in a subset of thecohort. We developed a universal screen for identifyingmaternal contamination of cord blood using DNAm at asubset of CpGs in the genome. This screen can be ap-plied to already-generated DNAm data from the 450Kor EPIC microarray platforms, but perhaps more inter-estingly, simple pyrosequencing at a subset of CpGs washighly efficient at identifying contaminated samples. Thisapproach could then be used to screen DNA from sam-ples destined for many purposes, including genotypingor gene expression methods or even cord blood banking.The described methods can reliably detect maternalblood contamination at levels that would confound gen-etic or epigenetic analyses. The amount of contamin-ation observed in all three studies could interfere withDNAm data analysis, but our proposed 10 CpG post hocscreen accurately identified and removed contaminatedmale and female samples. The three CpG pyrosequenc-ing screen will be useful primarily for: (a) cord bloodthat is not destined for DNAm assessment, such asgenotyping or gene expression studies, (b) when the ex-pected rate of contamination is high, or (c) if it is par-ticularly disadvantageous to run a possibly contaminatedsample. Our method has significant advantages comparedto other methods of detection of maternal contamination.MalesFemalesX Chromosome DNAm - True call>10,000 No calls in genotypingEpigenetic age > 1 year10 CpG method (array data)Pyrosequencing, 2 CpGsX Chromosome DNAm - Unknown call>10,000 No calls in genotypingEpigenetic age > 1 year10 CpG method (array data)% contam identified65%Of totalOf contam13%53% 11%100% 20%82% 16%% contamof total8%8%20%17%Pyrosequencing, 3 CpGsPyrosequencing, 2 CpGsPyrosequencing, 3 CpGs94% 20%9%Not contaminatedUnknown/unclearContaminatedMaleFemaleFig. 3 Summary of performance of all methods used to predict cord blood contamination. Each column represents the same participantacross each method. The 10 CpG method using 450K array data was the most reliable, but using a subset of three CpGs was sufficientto identify at least 82% of contaminated samples−0.10−−0.084 −0.080 −0.076 −0.072PC1PC2FemaleMale0123Male FemaleSex# CpGs at ThresholdAll validation samplesPyro screening n=189CBA0510Male FemaleSex# CpGs at ThresholdScreened validation samplesEPIC data n=158Screened validation samplesMDS n=158Fig. 4 Pre-screening using the pyrosequencing method correctly identified contaminated male samples. a Applying a cut-off of 2 CpGs above the thresh-old (yellow line) to the 3 CpG pyrosequencing method on validation data, 18 males and 15 females were identified as contaminated. b Principal compo-nent plot of EPIC DNA methylation data on all non-contaminated samples with two male samples that had been called contaminated by pyrosequencingshowed that contaminated male samples had been correctly identified. c Using the 10 CpG method from EPIC data, only the 2 male samples known tobe contaminated had more than 5 CpGs above the threshold (red line)Morin et al. Clinical Epigenetics  (2017) 9:75 Page 5 of 9For example, FISH requires whole cells, and mostTaqMan assays require DNA samples from bothmother and child [5–8, 11, 23]. For our DNAm-baseddetection of contamination, neither is required, how-ever, this does mean that we were not able to bench-mark our method against these others, as we did nothave the required sample types.While standard procedures exist for the collection ofcord blood, our results suggest that maternal contamin-ation is still observed. In our cohort study, the rate of con-tamination was 20%, and we observed two other studieswith appreciable levels of contamination, at 10% and 1%of samples. This suggests that maternal contamination isconsiderable overall, but importantly might occur morefrequently in some studies. Our samples were collectedfrom rural communities in a region near Cape Town,South Africa, and the publically available study with thehighest ratio of contaminated samples (GSE54399) wascollected in the Congo [24]. Collection procedures used instudies with less experience, many collections per day, orwith fewer resources may be more prone to introducingmaternal contamination in cord blood.As our study used real collected cord blood samples, itis difficult to estimate the specific detection limit of ourscreening method. Since the differences in DNAm areproportional to the amount of contamination, any samplesthat fail to meet the recommended cut-offs must containat most a small contribution of maternal blood. This un-certainty is reflected in our attempt to use either epigen-etic age or number of no calls in genotyping data toscreen for maternal contamination. Both methods identi-fied some but not all contaminated samples, and had veryhigh variability. It is thus unclear whether these methodsare inherently less predictive than the 10 CpGs we identi-fied, or if the amount of contamination in our sampleswas too small to detect by these methods. To determineexact proportions of contamination detectable by thesemethods, a follow-up study may consider creating knowndilutions of cord blood spiked with maternal blood, andassessing epigenetic age, genotyping no calls, as well asour 10 and 3 CpG methods. Thus while our proposedmethod cannot guarantee that all maternal contaminationis eliminated, it should assure that the most contaminatedsamples are identified and that any remaining contamin-ation has a minimal impact on downstream applications.Finally, given that we recognized the contaminationissue during routine quality control, it is possible thatmany researchers already find and remove some contami-nated samples from their cord blood DNAm studies.However, our inability to identify contaminated femalesamples during QC and the fact that we detected contami-nated samples in published data demonstrate that normalQC is not sufficient to completely eliminate contamin-ation, particularly of female samples. The 10 CpG panel isthen useful to ensure the removal of any contaminatedsamples once DNAm data has been generated.ConclusionsIn conclusion, we have created a screen to test for ma-ternal contamination in cord blood that has two inde-pendent applications: first, a simple and cost-effectivemethod to screen DNA from cord blood using pyrose-quencing, and second, a way to identify contaminatedsamples post hoc from DNAm arrays. Both cliniciansand researchers should be aware of the possibilities ofcross-contamination of maternal and cord blood, andthe CpGs we have identified will allow for easy identifi-cation and removal of contaminated samples.MethodsCord blood collectionIn the Drakenstein study, cord blood was collected bytrained staff after delivery of the baby but before deliveryof the placenta. The cord was clamped and cut, then theclamp was released and cord blood drained by gravityinto a kidney dish, then collected using a syringe forprocessing and storage.Samples used in this analysis were selected from thefull Drakenstein cohort for a sub-study on exposure tomaternal traumatic stress, and approximately 30% ofGSE30870GSE54399GSE62924GSE66459GSE74738GSE79056GSE80310GSE83334PredoStudy20 24 38 22 1 36 24 15 834n0255075100Percent of sites above thresholdFig. 5 Identification of studies with significant contaminationlevels in public data. Using available data, we examined the 10CpGs chosen to identify contamination, though some studieshad previously filtered their data and some CpGs were notavailable. We called maternal contamination of samples if morethan 50% of the available CpGs were above our contaminationthresholds, and identified two studies (GSE54399 and PREDO)with contaminated samplesMorin et al. Clinical Epigenetics  (2017) 9:75 Page 6 of 9children had been exposed to maternal trauma. TheDrakenstein cohort general inclusion criteria are de-scribed elsewhere [25]. Study participants with availableneuroimaging data were preferentially selected wherefeasible. Only samples of offspring whose mothers hadprovided informed consent for the collection, storage,and future analyses of DNA were eligible for inclusion.DNA methylation dataIn the discovery data set, DNAm was measured on 150samples (86 males, 64 females) using the Illumina Infi-nium HumanMethylation450 bead array (Illumina, SanDiego, USA), per manufacturer’s instructions and previouswork [26]. Next, we imported the raw data into IlluminaGenomeStudio Software for background subtraction andcolor correction, then exported it for processing using thelumi package in R (version 3.2.3) [27]. Initial quality con-trol and identification of maternal contamination in malesamples by multi-dimensional scaling (MDS) plotting andX chromosome DNAm occurred prior to removal of anyprobes. We then removed rs probes, X and Y chromo-some probes, probes with detection p values above 0.05,probes with less than three beads contributing to signal,and previously identified cross-reacting probes, for a totalof 421,993 probes remaining [28]. Quantro analysis indi-cated that quantile normalization was allowable, so wefirst normalized with the lumi quantile method, then withSWAN for probe type correction [21]. Finally, we usedComBat to remove chip and row effects [29].For validation data, analysis was identical with threeexceptions: first, data were generated using the InfiniumHumanMethylationEPIC (Illumina, San Diego, USA) on158 samples (89 males, 69 females). Second, we usedBMIQ normalization, and only performed ComBat onthe chip effects [22]. Third, we only retained the 10probes identified as indicators of contamination.Publicly available data were downloaded from GEO(GSE30870, GSE54399, GSE62924, GSE66459, GSE74738,GSE79056, GSE80310, and GSE83334), pre-processed asabove, and data from the PREDO study were provided bycoauthors [30].Genotyping data and no calls analysisGenotyping data were generated using the Illumina Psy-chChip (Illumina, San Diego, USA) per manufacturer’sinstructions then raw data were imported into Geno-meStudio using the PsychChip cluster file. Genotypeswere called by default methods in the GenomeStudiosoftware by comparing the sample intensities at eachlocus to expected genetic clusters, and a default qualitymetric represented a sample’s distance from the ex-pected cluster. The standard cut-off of 0.15 was used toestablish a threshold, outside of which samples were toofar from the cluster and the GenomeStudio software didnot call a genotype at that locus. p values and 90% confi-dence intervals for differences between contaminatedand non-contaminated samples were assessed using twosided Student’s t test with the t.test function in R statis-tical software [27].Epigenetic and gestational age analysisEpigenetic age was determined using two epigeneticclocks, one which outputs chronological age and is de-signed for adults, and the other which outputs gesta-tional age and is designed for newborns [19, 20]. Bothmethods use a panel of CpGs whose collective DNAmethylation status is strongly predictive of chronologicalage. As above, p values and confidence intervals for thedifference between contaminated and non-contaminatedsamples was calculated using two sided Student’s t testwith the t.test R package [27].Identification of sites used to detect contaminationTo discover CpGs capable of identifying maternal con-tamination, we first performed linear modeling on wholecord (GSE## to be determined) and adult (Flow.sorted.-blood.450K R package) blood DNAm data to identifysites that were most different between cord and adultblood [31, 32]. With thresholds of adjusted p value<1 × 10−20 and mean beta value difference greater than0.2, we identified 2250 DNAm sites that were differen-tially methylated between cord and adult. Though thesesites were all statistically significant, they were redun-dant in their multiplicity, and we wished to reduce thenumber of sites to make assessment more feasible. Thus,we analyzed this large set of 2250 CpGs with a randomforest approach from machine learning [33]. This en-semble learning method is designed to take advantage ofmultiple predictors, while also addressing small-sampleover-fitting. The random forest method ranked theDNAm sites by mean decrease in accuracy, a measure oftheir importance. We then applied binary recursive par-titioning to choose the threshold values separating con-taminated from non-contaminated samples [34].Pyrosequencing verificationWe used PyroMark Assay Design 2.0 (Qiagen, Inc.) softwareto design bisulfite pyrosequencing assays covering threeidentified CpGs (sequences in Additional file 1: Table S1).DNA was bisulphite converted using the EZ DNA Methyla-tion Kit (Zymo Research), and PCR and pyrosequencingperformed as previously described [35]. Streptavidin-coatedsepharose beads were bound to the biotinylated strand ofthe PCR product and were then washed and denatured toyield single-stranded DNA. Sequencing primers were thenadded for pyrosequencing per manufacturer’s instructions(Pyromark™ Q96 MD Pyrosequencer, Qiagen, Inc.).Morin et al. Clinical Epigenetics  (2017) 9:75 Page 7 of 9Additional fileAdditional file 1: Table 1. Primer sequences used for pyrosequencing.Table 2. Beta value thresholds used for DNA methylation arrays andpyrosequencing. Figure S1. Strategy used to identify contamination in malesamples. Plotting PC2, which separated male from female samples in our data,against DNA methylation at a CpG in XIST on the X chromosome, revealedthree populations of male samples: contaminated, non-contaminated, and agroup of five samples which were unclear. Figure S2. Neither epigenetic age(A), gestational epigenetic age (B) nor number of genotyping “no calls” (C)were sufficient to identify maternal blood contamination of cord blood. In allcases, contaminated and non-contaminated males showed high overlap, indi-cating insufficient discrimination. Figure S3. Across-batch differences in DNAmethylation level support the use of multiple predictive CpGs for identificationof contamination. Residual plot of discovery data (A), validation data (B),publically-available data (C), and PREDO (D) indicate technical spread acrosssamples and studies. In particular, EPIC data (second cohort, topright) shows greater variability and higher baseline levels than the450K data sets. (PDF 759 kb)AbbreviationsDNAm: DNA methylation; FISH: Fluorescent in-situ hybridization; GEO: Geneexpression omnibus; MDS: Multi-dimensional scaling; PCR: Polymerase chainreactionAcknowledgementsWe would like to thank Whitney Barnett for her assistance with thismanuscript.We thank the Drakenstein study staff, the clinical and administrative staff ofthe Western Cape Government Health Department at Paarl Hospital and atthe clinics for support of the study. We also thank our collaborators and themasters, doctoral, and postdoctoral students for their work on the study.Finally, we thank all mothers and children enrolled in the Drakenstein ChildHealth Study.The PREDO study would not have been possible without the dedicatedcontribution of the PREDO Study group members: Esa Hämäläinen, EeroKajantie, Jari Lahti, Hannele Laivuori, Anu-Katriina Pesonen, and Pia Villa. Wealso thank the PREDO study hospitals, and cohort mothers, fathers, andchildren for their enthusiastic participation.The PREDO study has received funding from the Academy of Finland,EraNet, EVO (a special state subsidy for health science research), University ofHelsinki Research Funds, the Signe and Ane Gyllenberg foundation, the EmilAaltonen Foundation, the Finnish Medical Foundation, the Jane and AatosErkko Foundation, the Novo Nordisk Foundation, the Päivikki and SakariSohlberg Foundation, and the Sigrid Juselius Foundation.FundingThis work was supported by the National Institutes of Health (grant numberR21HD085849 to KK); the AllerGen Network of Centers of Excellence (grantnumber GE-2 to MSK); the Bill and Melinda Gates Foundation (grant numberOPP 1017641 to HJZ). Individual team members were supported by the Na-tional Research Foundation and the South African Medical Research Council(DJS and NK), and the Canadian Institute of Health Research (LMM). No fund-ing agencies had any role in the design of the study and collection, analysis,interpretation of data or in writing the manuscript.Availability of data and materialsThe Drakenstein datasets generated during the current study are notpublicly available due to lack of consent to release data publically, but areavailable from the corresponding author on reasonable request. The PREDOdata that support the findings of this study are available from Dr. Raikkonnen,but restrictions apply to the availability of these data, which were used withpermission for the current study, and so are not publicly available. Data arehowever available from the authors upon reasonable request.Authors’ contributionsAMM performed the pyrosequencing, analyzed the data, created the figures,and co-wrote the manuscript. EGG performed the feature selection and gen-erated thresholds for screening. AMM, LMM, JLM, and DTSL generated anddid QC on the 450K and EPIC data. NK contributed to study design andinterpretation of data, and edited the manuscript. DC and KR generated,processed, and provided the PREDO data. HJZ, KK, and DJS contributed tothe acquisition of data and edited the manuscript. MSK conceived the study,advised on study design, and edited the manuscript. MJJ conceivedthe study, performed the 450K and EPIC data analysis, and co-wrotethe manuscript. All authors read and approved the final manuscript.Ethics approval and consent to participateAll Drakenstein study participants gave informed consent to participate andthe study was approved by University of Cape Town IRB. All other data waspublically available.Consent for publicationNot applicable.Competing interestsMJJ, MSK, AMM, EGG, JLM, and LMM are in the process of applying for apatent relating to the work presented.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Centre for Molecular Medicine and Therapeutics, BC Children’s Hospital,Department of Medical Genetics, University of British Columbia, 950 W 28thAve, Vancouver, BC V5Z 4H4, Canada. 2Department of Psychiatry and MentalHealth, South African Medical Research Council (SAMRC) Unit on Anxiety andStress Disorders, University of Cape Town, Groote Schuur Hospital, J2, AnzioRoad, Observatory, Cape Town, South Africa. 3Max Planck Institute ofPsychiatry, Department of Translational Research in Psychiatry,Kraepelinstraße 2-10, 80804 Munich, Germany. 4Department of Psychologyand Logopedics, Faculty of Medicine, University of Helsinki, P.O.Box 63, 00014Helsinki, Finland. 5Department of Paediatrics, MRC Unit on Child andAdolescent Health, University of Cape Town, Room 513 ICH Building RedCross Children’s Hospital Klipfontein Road, Cape Town, South Africa.6Department of Epidemiology, Harvard T. H. Chan School of Public Health,677 Huntington Avenue, Kresge Building, 505, Boston, MA 02115, USA.7Human Early Learning Partnership, University of British Columbia, 2208 EastMall, Vancouver, BC 02115, Canada.Received: 24 April 2017 Accepted: 11 July 2017References1. Hodyl NA, Roberts CT, Bianco-Miotto T. Cord blood DNA methylationbiomarkers for predicting neurodevelopmental outcomes. Genes(Basel). 2016;7.2. Küpers LK, Xu X, Jankipersadsing SA, Vaez A, la Bastide-van Gemert S,Scholtens S, et al. DNA methylation mediates the effect of maternalsmoking during pregnancy on birthweight of the offspring. Int J Epidemiol.2015;44:1224–37.3. Hermansen MC. Nucleated red blood cells in the fetus and newborn. ArchDis Child Fetal Neonatal Ed. 2001;84:211F–215.4. Armson BA. Maternal/Fetal Medicine Committee, Society of Obstetriciansand Gynaecologists of Canada. Umbilical cord blood banking: implicationsfor perinatal care providers. J Obstet Gynaecol Can. 2005;27:263–90.5. Lo YM, Lau TK, Chan LY, Leung TN, Chang AM. Quantitative analysis of thebidirectional fetomaternal transfer of nucleated cells and plasma DNA. ClinChem. 2000;46:1301–9.6. Masuzaki H, Miura K, Miura S, Yoshiura K-I, Mapendano CK, Nakayama D, etal. Labor increases maternal DNA contamination in cord blood. Clin Chem.2004;50:1709–11.7. Hall JM, Lingenfelter P, Adams SL, Lasser D, Hansen JA, Bean MA. Detectionof maternal cells in human umbilical cord blood using fluorescence in situhybridization. Blood. 1995;86:2829–32.8. Bauer M, Orescovic I, Schoell WM, Bianchi DW, Pertl B. Detection ofmaternal deoxyribonucleic acid in umbilical cord plasma by usingfluorescent polymerase chain reaction amplification of short tandem repeatsequences. Am J Obstet Gynecol. 2002;186:117–20.Morin et al. Clinical Epigenetics  (2017) 9:75 Page 8 of 99. Petit T, Dommergues M, Socié G, Dumez Y, Gluckman E, Brison O. Detectionof maternal cells in human fetal blood during the third trimester ofpregnancy using allele-specific PCR amplification. Br J Haematol.1997;98:767–71.10. Cairo MS, Wagner JE. Placental and/or umbilical cord blood: an alternativesource of hematopoietic stem cells for transplantation. Blood. 1997;90:4665–78.11. Socié G, Gluckman E, Carosella E, Brossard Y, Lafon C, Brison O. Search formaternal cells in human umbilical cord blood by polymerase chain reactionamplification of two minisatellite sequences. Blood. 1994;83:340–4.12. Guerrero-Preston R, Goldman LR, Brebi-Mieville P, Ili-Gangas C, LebronC, Witter FR, et al. Global DNA hypomethylation is associated with inutero exposure to cotinine and perfluorinated alkyl compounds.Epigenetics. 2010;5:539–46.13. Heyn H, Li N, Ferreira HJ, Moran S, Pisano DG, Gomez A, et al. Distinct DNAmethylomes of newborns and centenarians. Proc Natl Acad Sci U S A. 2012;109:10522–7.14. Florath I, Butterbach K, Müller H, Bewerunge-Hudler M, Brenner H. Cross-sectional and longitudinal changes in DNA methylation with age: anepigenome-wide analysis revealing over 60 novel age-associated CpG sites.Hum Mol Genet. 2014;23:1186–201.15. Ladd-Acosta C. Epigenetic signatures as biomarkers of exposure. Curr EnvirHealth Rpt. 2015;2:117–25.16. Marsit CJ. Influence of environmental exposure on human epigeneticregulation. J Exp Biol. 2015;218:71–9.17. Joubert BR, Felix JF, Yousefi P, Bakulski KM, Just AC, Breton C, et al. DNAMethylation in newborns and maternal smoking in pregnancy: genome-wide consortium meta-analysis. Am J Hum Genet. 2016;18. Cotton AM, Price EM, Jones MJ, Balaton BP, Kobor MS, Brown CJ.Landscape of DNA methylation on the X chromosome reflects CpGdensity, functional chromatin state and X-chromosome inactivation.Hum Mol Genet. 2015;24:1528–39.19. Horvath S. DNA methylation age of human tissues and cell types. GenomeBiol. 2013;14:R115.20. Knight AK, Craig JM, Theda C, Bækvad-Hansen M, Bybjerg-Grauholm J,Hansen CS, et al. An epigenetic clock for gestational age at birth based onblood methylation data. Genome Biol. 2016;17:206.21. Maksimovic J, Gordon L, Oshlack A. SWAN: subset-quantile within arraynormalization for illumina infinium HumanMethylation450 BeadChips.Genome Biol. 2012;13:R44.22. 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 450 k DNA methylation data. Bioinformatics.2013;29:189–96.23. Petit T, Gluckman E, Carosella E, Brossard Y, Brison O, Socié G. A highlysensitive polymerase chain reaction method reveals the ubiquitouspresence of maternal cells in human umbilical cord blood. Exp Hematol.1995;23:1601–5.24. Stein DJ, Koen N, Donald KA, Adnams CM, Koopowitz S, Lund C, et al.Investigating the psychosocial determinants of child health in Africa: TheDrakenstein Child Health Study. J Neurosci Methods. 2015;252:27–35.25. Zar HJ, Barnett W, Myer L, Stein DJ, Nicol MP. Investigating the early-lifedeterminants of illness in Africa: The Drakenstein Child Health Study. Thorax.2015;70:592–4.26. 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;28:1385–99.27. R DCT. R: a language and environment for statistical computing. [Internet].Vienna: R Foundation for Statistical Computing; 2008. Available from: http://www.R-project.org/28. 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;6:4.29. Teschendorff AE, Zhuang J, Widschwendter M. Independent surrogatevariable analysis to deconvolve confounding factors in large-scalemicroarray profiling studies. Bioinformatics. 2011;27:1496–505.30. Girchenko P, Hämäläinen E, Kajantie E, Pesonen A-K, Villa P, Laivuori H, et al.Prediction and prevention of preeclampsia and intrauterine growthrestriction (PREDO) study. Int J Epidemiol. 2016;31. de Goede OM, Razzaghian HR, Price EM, Jones MJ, Kobor MS, Robinson WP,et al. Nucleated red blood cells impact DNA methylation and expressionanalyses of cord blood hematopoietic cells. Clin Epigenetics. 2015;7:95.32. Reinius LE, Acevedo N, Joerink M, Pershagen G, Dahlen S-E, Greco D, et al.Differential DNA methylation in purified human blood cells: implications forcell lineage and studies on disease susceptibility. PLoS One. 2012;7:e41361.33. Breiman L. Random forests. Machine learning, vol. 45; 2001. p. 5–32.34. Breiman L, Friedman J, Stone CJ, Olshen RA. Classification and regressiontrees. Florida: Chapman and Hall/CRC press; 1984.35. Clifford RL, Jones MJ, MacIsaac JL, McEwen LM, Goodman SJ, Mostafavi S, etal. Inhalation of diesel exhaust and allergen alters human bronchialepithelium DNA methylation. J Allergy Clin Immunol. 2017;139:112–21.•  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:Morin et al. Clinical Epigenetics  (2017) 9:75 Page 9 of 9


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