Conception by fertility treatment and offspring deoxyribonucleic acid methylation

Abstract

Objective:

To investigate whether deoxyribonucleic acid (DNA) methylation at birth and in childhood differ by conception using assisted reproductive technologies (ART) or ovulation induction compared with those in children conceived without fertility treatment.

Design:

Upstate KIDS is a matched exposure cohort which oversampled on newborns conceived by treatment.

Setting:

New York State (excluding New York City).

Patient(s):

This analysis included 855 newborns and 152 children at approximately 9 years of age.

Intervention(s):

None.

Main Outcome Measure(s):

DNA methylation levels were measured using the Illumina EPIC platform. Single CpG and regional analyses at imprinting genes were conducted.

Result(s):

Compared to no fertility treatment, ART was associated with lower mean DNA methylation levels at birth in 11 CpGs (located in/near SYCE1, SPRN, KIAA2013, MYO1D, GET1/WRB-SH4BGR, IGF1R, SORD, NECAB3/ACTL10, and GET1) and higher mean methylation level in 1 CpG (KLK4; all false discovery rate P< .05). The strongest association (cg17676129) was located at SYCE1, which codes for a synaptonemal complex that plays a role in meiosis and therefore infertility. This CpG remained associated with newborn hypomethylation when the analysis was limited to those conceived with ICSI, but this may be because of underlying male infertility. In addition, nine regions in maternally imprinted genes (IGF1R, PPIEL, SVOPL GNAS, L3MBTL, BLCAP, HYMAI/PLAGL1, SNU13, and MEST) were observed to have decreased mean DNA methylation levels among newborns conceived by ART. In childhood, hypomethylation of the maternally imprinted gene, GNAS, persisted. No CpGs or regions were associated with ovulation induction.

Conclusion(s):

ART but not ovulation induction was associated with hypomethylation at birth, but only one difference at an imprinting region appeared to persist in childhood.

The number of children conceived by assisted reproductive technologies (ARTs) worldwide continues to increase since its first use over 40 years ago (1–3). ART consists of manipulation of gametes outside of the body with in vitro fertilization (IVF) involving intracytoplasmic sperm injection (ICSI) being commonly used. In 2016, 1.8% of all births in the United States were conceived by ART, translating to approximately 70,000 births per year, of which 66% used ICSI (3). Other types of non-ART fertility treatments, such as ovulation induction (OI) and intrauterine insemination (IUI), are estimated to be involved in 3%–7% of births (4).

ART consists of multiple steps; beginning with controlled ovarian hyperstimulation for oocyte retrieval, followed by fertilization of eggs and culture of embryos in vitro, and ending with embryo transfer which, if successful, leads to pregnancy. Epigenetic modifications induced by any treatments, such as hormones, can occur in any of these steps (5). Even small changes in embryo culture media could lead to epigenetic changes (6) as oocyte development, fertilization, and embryo development are critical times for chromosome remodeling and deoxyribonucleic acid (DNA) demethylation and remethylation (7). Superovulation has been observed to be associated with methylation differences, particularly in imprinting regions (8, 9). Altered placental function could also play a role because the trophoblast is very sensitive to epigenetic changes (10). Rare imprinting disorders (such as Beckwith-Wiedemann and Angelman syndrome) have been observed in children conceived from ART (11) and from ovarian stimulation (12). Research on imprinting disorders suggested that the normal DNA methylation from each parent is not being correctly maintained during the ART procedures, but the findings have been mixed (13–18). There are also concerns related to the differences in birth outcomes, even among singletons born to the same mother, who were conceived by ART compared with siblings who were not (19).

Although imprinting disorders remain rare, their associations point to the potential for genome-wide differences with DNA methylation. Methylation is the attachment of a methyl group to the cytosine of a base pair in the DNA sequence. Genes which are imprinted are methylated (either on the paternal or maternal allele) in order to silence their expression. More broadly speaking, it has been similarly observed in non-imprinted genes that hypermethylation is related to decreased gene expression while hypomethylation to increased gene expression. In terms of genomic architecture, CpGs are not randomly distributed throughout the genome but tend to cluster in what are termed CpG islands. The first study in humans evaluated approximately 1,600 CpGs among 10 exposed IVF children and 13 controls who were spontaneously conceived, finding differences for CpGs in genes associated with the cardiometabolic pathways (20). Since then, subsequent studies have been largely inconsistent because of small sample sizes and differences in the methods used (21–25) Most recently, one study followed children conceived by IVF in the 1990s until they were adults, finding that DNA methylation in adulthood did not differ from that in those conceived naturally(26). The use of ICSI was not represented because conception took place during the 1990s before ICSI was widely adopted. Contemporary ART uses ICSI for the majority of births.

The current study examined DNA methylation differences at birth and in early childhood (8–10 years) in a cohort of children born in New York from 2008 to 2010. The study improves upon previous investigations in several ways: the ability to evaluate ART children conceived predominantly with ICSI reflecting current practice (>70% [27]); the ability to evaluate methylation differences at birth and in middle childhood; and the inclusion of a group conceived by OI to account for underlying infertility and examine the potential for independent associations with methylation.

MATERIALS AND METHODS

Study Design and Population

Participants in the Upstate KIDS study were sampled among women who delivered between 2008 and 2010 in New York State, excluding New York City (28). All mothers of liveborn singletons whose birth certificates indicated any type of fertility treatment were invited to participate approximately 2–4 months after delivery. During the same period, mothers of three singletons not conceived with fertility treatment were recruited for every singleton conceived with treatment, frequency matched on seven perinatal regions of birth designated by the state (28). Mothers of all multiples (twins and higher order) were also invited to participate regardless of treatment. Of 18,479 mothers approached, 5,034 (27%) mothers enrolled with their 6,171 newborns (28). Thirty percent of newborns were conceived by fertility treatment, with 14% by ART and 16% by OI/IUI. Details on the first phase of follow-up through 3 years of age can be found elsewhere, including that fertility treatment was not associated with differences in early childhood growth or development (28–30). Follow-up of the children continued until 2019, but beginning with the second phase of the study in 2014, excluded triplets and quadruplets because of low numbers (n = 134). The current investigation includes children with newborn dried blood spots (DBS) available (Supplemental Fig. 1, available online). Beginning in 2017, 4,644 (84%) children were invited for clinic visits by their proximity (within 2 hours) to 1 of 4 clinic sites. Children who agreed to a blood draw and had samples collected by June 2019 had DNA methylation measured (n = 152, 27%). Eighty-one (53%) of these 152 also had newborn DNA methylation analyses available. With the limited number (n = 81) with samples overlapping from the same child at birth and in childhood, all samples were analyzed without regard to newborn findings. The New York State Department of Health and the University of Albany Institutional Review Board (IRB) (NYSDOH IRB #07–097; UAlbany #08–179 and #15E-122) approved the study and served as the IRB designated by the National Institutes of Health under a reliance agreement. Mothers provided written informed consent. Children also assented to clinic visits.

Fertility Treatment Exposure

In the baseline questionnaire at approximately 4 months postpartum, mothers were asked to indicate all medical services or medications used to assist them in becoming pregnant. We defined ART use as use of IVF or ICSI, assisted hatching, frozen embryo transfer, gamete intrafallopian transfer (GIFT), zygote intrafallopian transfer (ZIFT), with or without the use of donor eggs or embryos. We defined OI as the use of oral or injectable medications (e.g., clomiphene citrate [CC; Clomid] or gonadotropins) with or without IUI (28). As reported, the sensitivity (93%) and specificity (99%) of maternal report of ART use were high as compared with ART use identified by linkage with the Society for Assisted Reproductive Technology Clinic Outcome Reporting System (SART-CORS) (31). Given the validity of maternal report of ART and lacking any available registry on OI/IUI use, we relied on maternal report for all fertility treatment information. Among those with ART, ICSI and male factor infertility was based on the information from SART-CORS. In addition, no one in the ART group used GIFT or ZIFT.

DNA Methylation

Mothers provided consent to allow the study to retrieve remaining DBS from the State’s Newborn Screening Program for analysis of biomarkers when the children were 8 months old (32). We have previously detailed this process (32). Three 3.2 mm DBS punches were eluted with buffer solution and the eluants processed for biomarker analyses (33), after which the remaining eluted DBS punches were returned to −20°C storage. As the original DBS consent did not cover genetic analyses, the study reconsented parents for the use of genetic analyses in 2016–2017. The study processed previously eluted DBS punches from 1,055 children. DNA was extracted using the GenSolve DNA recovery kit followed by purification using the Qiagen QIAmp DNA blood kits (#51104), a silica membrane-based purification system. Between 2018 and 2019, children attended clinic visits at 1e of 4 sites across the state (i.e., University at Albany, University at Buffalo, University of Rochester, or New York University Langone Health) when they were 8–10 years old. Blood was processed into buffy coat samples and frozen in −80°C storage until DNA extraction of samples from 177 children in July 2019 at the University of Minnesota. DNA from DBS and from buffy coat samples underwent bisulfite conversion with standardized kits (e.g., Zymo EZ DNA Methylation kit; Zymo, Irvine, CA), followed by measurements of DNA methylation using the Infinium MethylationEPIC BeadChip microarray (34). There were 30 (3%) newborn DBS samples that did not have sufficient DNA, 19 (2%) that failed quality control checks (specified following), and 151 (14%) that were excluded as twins of the same family, resulting in methylation data from 855 unrelated newborns. At follow-up, 18 (10%) samples did not have sufficient DNA because of errors in blood processing and 7 (4%) failed quality control checks, resulting in data from 152 children at 8–10 years.

Statistical Analysis

Data cleaning.

Methylation data were processed using the minfi package in R (35), including the identification of failed probes and scaling with Illumina control probes to determine methylation values. The beta value (β) was determined for each of the CpG sites by the fluorescent signals (β = Max (M, 0)/[Max(M,0)+Max(U, 0)+100)] (36). The β values approaching one are completely methylated and those close to zero are unmethylated. Background and dye-bias corrections were applied. Quantile normalization was used to normalize the β values between the two types of probes (37). Principal component analysis (PCA) was performed to further detect outliers and samples mismatched in sex compared against information from vital records. We extracted the detection P value for each methylation measure (per site per sample) and filtered data that failed detection P value (P>.01). We removed samples and CpG sites with low passing rate (<97%) based on detection P value and bead counts (<3). Methylation data at birth and at follow-up were excluded for sex mismatch (<1%) where the principal component for sex was not the same as noted in the records. After probe removal, 836,001 CpG probes for newborn samples and 833,355 probes for childhood samples remained.

Modeling exposure.

Fertility treatment exposure was first examined (i.e., yes if either OI/IUI or ART was used vs. no). This comparison helped to delineate whether underlying infertility may have been related to any DNA methylation differences regardless of the type of treatment. Comparisons were then made between children not conceived with any treatment and those with OI/IUI and ART, separately. In the secondary analysis, we ran the same adjusted model comparing those conceived with ICSI (n = 100) and those conceived without any treatment, keeping the latter as the referent group (n = 518). As the use of ICSI is not an indication of male factor infertility, we also performed an analysis limiting those who had ICSI without male factor infertility, to verify that any differences would not be driven by male infertility.

Modeling outcomes.

Robust linear regression models were used to test associations between methylation β values at each CpG site and exposures of interest with adjustment for covariates. Batch effects (as covariates of chip and row) were accounted for through random effects. False discovery rate (FDR) correction was applied to account for multiple testing (38).

Covariates.

The covariates evaluated included maternal age, race/ethnicity, education, pregnancy smoking, private insurance, plurality, infant sex, estimated cell type, and batch. The covariate information came from vital records (i.e., maternal age, insurance status, plurality, birth weight, and gestational age) or by maternal report at baseline (i.e., maternal race, education, pregnancy smoking). Cell type mixture was estimated on the full set of normalized data (FlowSorted.CordBlood.450K package) (39). Cell counts at birth using DNA from newborn DBS samples were estimated based on a cord blood reference using minfi in R, including B cells, CD4⁺ T cells, CD8⁺ T cells, granulocytes, monocytes, natural killer cells, and nucleated red blood cells (40). A major difference previously identified between adult vs. newborn cell count is the proportion of nucleated red blood cells (40). Cell type estimation for buffy coat samples from childhood were calculated using the Houseman method (41).

Regional analyses.

Methylation at imprinting regions were additionally examined for differences by exposure categories. Hernandez Mora et al (42) cataloged the probes from the EPIC microarray corresponding to these differentially methylated regions. DMRff (43) with a 5kb sliding window was used in R for testing these 888 CpGs (separately by maternal vs. paternal imprinting) jointly for each imprinting gene, and with the same covariates as previously, based on the outputs from the robust linear regression models. DMRff provides Bonferroni adjusted P values. In comparison with single CpG analyses, Bonferroni adjusted P values were implemented to be more stringent than FDR correction as there were now fewer multiple tests being accounted for (i.e., not >800,000 CpGs).

Sensitivity Analysis

The New York State Congenital Malformations Registry was linked to the participant data. Mothers also reported on congenital conditions at 4 months postpartum. We verified that there were no newborns with imprinting disorders. However, we identified 19 newborns with potential conditions that could lead to outlying DNA methylation values because of potential genetic etiologies (i.e., hereditary anemia, trisomy 21, cerebral palsy, limb deficiency, hypospadias, hemophilia, supernumerary digit, hemolytic Rh disease, congenital cyst, double right kidney, congenital hypoplastic left heart, cranio-synostosis, chondrodystrophy, factor VIII, congenital hydrocephalus). An additional seven newborns were excluded by parental report of omphalocele, limb reductions, Edwards syndrome, hypospadias, or cleft lip/palate. All analyses were rerun excluding these 26 newborns as a sensitivity analysis.

RESULTS

Newborns not conceived with fertility treatment (61%) served as the reference group for all analyses, in comparisons with newborns conceived with any type of treatment (39%) and also specifically by ART (18%) and by OI/IUI (21%). Differences in baseline characteristics by fertility treatment exposure were as expected (Supplemental Table 1). Older parental age, bearing twins, being non-Hispanic white, having higher socioeconomic status, and less smoking during pregnancy were associated with using fertility treatments. Compared with the original cohort of children, those included had higher education, were more likely white, and less likely to smoke (44).

In newborns, 12 CpGs were associated with ART treatment (Table 1) but none were associated with OI/IUI or the combination of any fertility treatment (Supplemental Table 2). Of the 12 CpGs, all except 1 (cg08681856) were associated with decreased methylation observed among ART newborns. The top association (cg17676129) was within SYCE1 with mean levels of methylation among ART newborns 1.7% lower than that of those conceived without treatment. Unadjusted analyses were very similar suggesting a minimal impact of covariates.

TABLE 1.

Newborn DNA methylation differences by ART treatment exposure (Upstate KIDS study; n = 855.). ART (n = 157) vs. no fertility treatment (n = 520).

CpG	Beta	SE	P value	FDR-P	Chr	Position	Gene	Relation to Island
cg17676129	−0.01754	0.0025	1.77E-12	1.48E-06	10	135382545	SYCE1	Island
cg24413339	−0.01561	0.0028	1.37E-08	.0057	10	135237754	SPRN	Island
cg01061626	−0.01114	0.0021	6.53E-08	.018	1	11986394	KIAA2013	Island
cg01050010	−0.04497	0.0084	9.26E-08	.019	17	31149877	MYO1D	Island
cg27119318	−0.04852	0.0092	1.19E-07	.020	21	40759574	GET1; WRB-SH4BGR	North Shore
cg19322380	−0.03368	0.0065	2.18E-07	.030	15	99408636	IGF1R	Open SeaSea
cg27073142	−0.02658	0.0052	3.02E-07	.034	15	45314933	SORD	North Shore
cg13403462	−0.02975	0.0058	3.23E-07	.034	20	32256071	NECAB3; ACTL10	South Shore
cg08681856	0.05493	0.0109	4.46E-07	.041	19	51411737	KLK4	Island
cg02289287	−0.03929	0.0079	5.65E-07	.044	17	20989155	LINC01563	Open Sea
cg14921437	−0.02608	0.0052	5.77E-07	.044	20	32255988	NECAB3; ACTL10	Island
cg19684207	−0.04848	0.0098	6.93E-07	.048	21	40759686	GET1	North Shore

Note: Models adjusted for maternal age (continuous), race (white and non-white), education (less college, college, and advanced degree), pregnancy smoking (y/n), private insurance (y/n), plurality (twin/not), infant sex(girl/boy), estimated cell type (B cells, CD4⁺ T cells, CD8⁺ T cells, granulocytes, monocytes, natural killer cells, and nucleated red blood cells), and batch (categorical, 12 levels). Only FDR significant (P < .05) CpGs are shown. ART = assisted reproductive technology; Chr = chromosome; DNA = deoxyribonucleic acid; FDR = false discovery rate.

Yeung. ART and DNA methylation. Fertil Steril 2021.

One CpG in IGF1R, a known imprinting gene, was associated with ART in single CpG-by-CpG analysis. Using a sliding window approach to detect differentially methylated regions (DMRs), a region corresponding to seven CpGs within IGF1R was found to be hypomethylated among newborns conceived with ART (Table 2). In total, nine imprinted gene regions were associated with ART. All regions consistently exhibited hypomethylation rather than hypermethylation in newborns conceived with ART. All associated regions were maternally rather than paternally imprinted. Figure 1 shows the top 3 regions identified (i.e., IGF1R, PPIEL, and SVOPL).

TABLE 2.

Differences in newborn DNA methylation of imprinted genes by ART exposure (Upstate KIDS; n = 855). ART (n = 157) vs. no fertility treatment (n = 520).

Gene	Chr	Start	End	CpGs No.a	Betab	SE	P adjustedc
IGF1R	15	99408636	99409506	7	−0.02042	0.002032	2.46E-20
PPIEL	1	40024971	40025415	4	−0.02271	0.002999	9.58E-11
SVOPL	7	138348774	138349443	6	−0.01855	0.00275	4.06E-08
GNAS	20	57426538	57427443	28	−0.00302	0.000457	9.49E-08
L3MBTL	20	42142417	42143174	22	−0.00483	0.000738	1.52E-07
BLCAP	20	36148994	36149271	13	−0.0096	0.001577	3.03E-06
HYMAI; PLAGL1	6	144329172	144329829	12	−0.00263	0.00051	.000658
SNU13	22	42078217	42078567	5	−0.00676	0.001339	.001173
MEST	7	130131676	130131905	10	−0.00493	0.001115	.026511

Note: ART = assisted reproductive technology; Chr = chromosome.

^aCpGs correspond to cg number (location) for each gene listed: IGF1R: cg19322380, cg21746425, cg12553689, cg07615383, cg00098799, cg13812291, cg03380198 (all located in the same intron); PPIEL: cg10243676 (exon), cg11704876 (intron), cg22862450 (post transcript), cg15057250 (post transcript); SVOPL cg05719902 (intron), cg16111924 (intron/post transcript), cg19079272 (intron/post transcript), cg13273418 (intron/post transcript), cg10184328 (intron/post transcript), cg23085143 (intron/post transcript); GNAS: cg22860367, cg07947033, cg06693667, cg23484981, cg04677683, cg15160445, cg25326570, cg23249369, cg01715551, cg14176797, cg04457481, cg27262796, cg21938532, cg03606258, cg22989942, cg18224653, cg14309385, cg05309239, cg12573482, cg24617313, cg22290117, cg03908391, cg26496204, cg03010274, cg25652859, cg12321149, cg06065549 (all located in the same intron); L3MBTL: cg22457903, cg20091959, cg17091610, cg23626798, cg15388309, cg16862791, cg06446163, cg04984575, cg02177317, cg20529070, cg10360552, cg18911945, cg01877937, cg08633313, cg12699433, cg11319028, cg22601123, cg22330467, cg14306330, cg01071811, cg02611863, cg15330298 (variant dependent locations); BLCAP: cg15473473, cg07156273, cg01466133, cg24338351, cg24675557, cg20479660, cg22421148, cg22510412, cg3061677, cg00576435, cg18433380, cg16648571, cg04810287 (all located in the same intron); HYMAI/PLAGL1: cg22378065, cg22352234, cg00702231, cg12757684, cg21526238, cg23541304, cg12820609, cg05326984, cg08263357, cg11532302, cg27216384, cg17895149 (variant dependent locations); SNU13: cg18152773, cg05871614, cg22083753, cg15284719 (exon/intron), cg08686092 (exon/intron); MEST: cg12347392, cg04786207, cg26708559, cg22705386, cg06212135, cg10249538, cg16823958, cg27338480, cg09080913, cg13104298 (all located in the same intron).

^bModels adjusted for maternal age, race, education, pregnancy smoking, private insurance, plurality, infant sex, estimated cell type, and batch. Regional analyses of imprinted gene locations using DMRff (5kb window).

^cBonferroni adjustment by DMRff.

FIGURE 1.

Regional differences in maternally imprinted genes by ART status. Plots showing mean (± 95% CI) DNA methylation in regions of top 3 maternally imprinted genes for ART (red) and no treatment (blue) newborns at A, IGF1R, B, PPIEL, and C, SVOPL. ART = assisted reproductive technology; CI = confidence interval; DNA = deoxyribonucleic acid.

Results were similar in the sensitivity analyses when 26 newborns with congenital conditions were removed (Supplemental Table 3). Of these 26 newborns, 5 (2.5% of the sample) were conceived with ART, 8 (4.8%) with OI/IUI, and 13 (3.1%) without treatment. Seven of the single CpG associations with ART at birth remained FDR significant. The other associations, although reduced in statistical significance, were otherwise similar. Imprinting results remained virtually the same for all except two regions that were nonsignificant in the sensitivity analyses (i.e., HYMAI/PLAGL1, SNU13, Supplemental Table 4). As before, no CpGs were associated with OI/IUI or any combined treatment.

In the secondary analysis, newborn DNA methylation differences were also investigated specifically among those conceived with ICSI (n = 100, 63% of ART) in comparison with newborns conceived without treatment (Supplemental Table 5). Four CpGs were FDR significant, including two already identified in the larger ART cohort (cg17676129 at SYCE1 and cg27119318 on chromosome 21 in WRB). Of the 12 identified with ART, 6 were among the top 20 hits listed including SYCE1, suggesting great overlap. After removing those diagnosed with male infertility (n = 32), no CpGs were FDR significant (Supplemental Table 6). The top hit remained cg27119318 on chromosome 21 in WRB, an imprinted gene, but cg17676129 at SYCE1 was no longer among the significant CpGs (β = −0.01, SE = 0.003, FDR- p= .58).

Follow-up DNA samples were collected from 152 children at 8–10 years old. Among the 152 children, 34 were conceived by OI/IUI and 23 with ART (Supplemental Table 7). Only one significant CpG was associated with ART (cg04061372) and none with OI/IUI or any treatment (Supplemental Table 8). Of the 12 CpGs associated with ART conception at birth, mean differences generally did not persist except for three CpGs with similar differences, though these were no longer significant (cg27119318, cg08681856, cg02289287) (Supplemental Table 9). Although the CpG-by-CpG analysis did not replicate in childhood, there were indications that the hypomethylation in imprinting regions of GNAS on chromosome 20 remained (Supplemental Table 10). Figure 2 shows the overlap between the lower DNA methylation identified at birth and the differences at childhood. Although the regions identified were not completely the same, the hypomethylation was consistent (i.e., the 905 bp region identified at birth overlaps with the 874 bp region identified in childhood by a 340 bp section covered by 9 CpGs). Models did not seem overinflated with λs between 0.97 and 1.04 (Supplemental Table 11).

FIGURE 2.

GNAS DNA methylation differences identified at birth and at childhood. Plots showing mean (± 95% CI) DNA methylation in GNAS regions for ART (red) and no treatment (blue) among newborns (upper panel) and children at approximately 9 years of age (lower panel). Shaded region denotes the significant DMRs identified. ART = assisted reproductive technology; CI = confidence interval; DMRs = differentially methylated regions; DNA = deoxyribonucleic acid.

DISCUSSION

We observed lower methylation in 12 CpGs at birth among newborns conceived with ART compared with levels in spontaneously conceived newborns in the largest fertility cohort study in the U.S. to examine DNA methylation to date. Among the 12 CpGs, 3 were in the maternally imprinted genes IGF1R and GET1/WRB. Regional analyses further confirmed a region of seven CpGs in IGF1R that was hypomethylated along with eight other maternally imprinted gene regions. One such DMR on chromosome 20 in GNAS was also observed to be hypomethylated at age 8–10 years, suggesting persistence of the change. Nevertheless, all CpGs and other imprinting regions identified at birth were not observed to be associated with ART among the smaller group of 8–10- year-old children. Subgroup analysis among those conceived by ICSI did not identify additional CpGs aside from one (at MLX) with a small effect size. Comparisons of those conceived by OI/IUI with those with no fertility treatment also did not reveal differences.

The top hit (cg17676129) identified with ART was in a CpG island, located in an intronic region at one variant of SYCE1. SYCE1 codes for a synaptonemal complex that is important for linking homologous chromosomes together during meiosis, such that knockout mice of this gene are infertile (45). Microdeletions in humans have been associated with primary ovarian insufficiency, a cause of female infertility, in multiple populations (46). In men, this gene is associated with nonobstructive azoospermia (47). We lacked genotype data to verify that the CpG we identified (cg17676129) was not influenced by genetic mutations, but the methylation levels of the gene exhibited a normal distribution with no clear clustering. We did observe that removing the children conceived by ICSI whose fathers were diagnosed with male infertility virtually wiped away this association, suggesting that it is likely driven by underlying infertility. Previous studies of ART, however, did not identify this CpG/gene as being differentially methylated (48).

In fact, when comparing the results of five previous studies investigating ART using Illumina’s 27K/450K microarrays, a review noted that only four genes were identified in >1 study (i.e., GNAS, PEG10, PRCP, and RUNX3) (48). Similarly, a meta-analysis of 18 studies also found inconsistent differences (49). A longitudinal analysis of DNA at birth using DBS compared with DNA from the same adults at 22–35 years of age identified yet other genes (26). The authors concluded that few differences detected at birth persisted (i.e., CHRNE, PRSS16, TMEM18) (26). In contrast to our findings, they observed the GNAS antisense RNA (noncoding) paternally methylated gene, GNASAS, to be hypermethylated at birth, and no differences were found in adulthood (26). On the other hand, they found IGF1R to be hypomethylated at birth and in adulthood but at <4% difference, which seemed inconsequential as it was less than the difference observed in other CpGs (approximately 9%) (26). Heterogeneity, including stochastic changes to ART treatment techniques over the years, different reference panels for accounting for cell type composition, or sample sizes of the studies are likely reasons for the inconsistency, and pooling of results across studies in future work may be needed to resolve these differences.

Nevertheless, the fact that most studies observed DNA methylation differences at birth (21–25) and, separately, evidence of imprinting disorders in ART populations (49), remain of concern. We tried to rule out several factors. Although superovulation has been suggested as a mechanism of effect (50), the OI/IUI group consistently did not differ from the group with no treatment. This group underwent OI by injection or by oral agents such a CC. As no global difference was observed and given the limited sample size for subgroups, no further analyses were made to separate those conceived by CC compared with gonadotropins in the OI/IUI group. Nevertheless, our findings also suggested that underlying fertility alone might not explain the differences, although the severity of infertility with ART cannot be ruled out (e.g., for male infertility and SYCE1). Many studies have also observed differences in the DNA methylation of spermatozoa from infertile men. Differences in H19, MEST, and SNRPN were noted from a meta-analysis of 24 studies (51). H19, a paternally imprinted gene, has frequently been associated with ART, although not by studies using the Illumina microarrays (48). We were also unable to observe differences in H19 by either single CpG analysis or by regions, even among those using ICSI, albeit with its loose correlation with male infertility given its wide use. However, some previous studies primarily found differences when placental tissue (15, 16) rather than cord blood was used, and increasing evidence suggests that paternal imprinting plays a greater role in placental development. What remains to be teased apart in a larger consortium of studies are whether specific techniques, including frozen vs. fresh embryo transfer, culture media, and other specific techniques of ART may be playing a role. It may well be that they all play different roles, leading to the heterogeneity of results observed.

Although the remaining genes identified by single CpG analysis have no known ties to infertility, two have been associated with birth defects. cg01050010 is located in a CpG is-land of MYO1D, as a myosin gene, found to play a role in establishing and maintaining actin filaments (52). Suppressing its gene expression by small interfering RNA is associated with lower branching in oligodendrocytes and most research is tied to its role in neurodevelopment (52). Methylation at a different CpG site (cg18200510) i n MYO1D than the one that we identified was higher among those with hypospadias (although not robust to multiple testing, P = 4 × 10⁻⁵) (53). Two CpGs (cg19684207 and cg27119318) on chromosome 21 in the transcription start sites of GET1/WRB were associated with ART. The gene has been associated with congenital heart defects in Down syndrome and its methylation with imprinting disorders (54). Although we observed these 2 maternally methylated CpGs to be approximately 5% lower, no DMR was identified. Placental methylation in WRB has been found to be associated with ART (25). In the sensitivity analyses, after removing the few cases of hypospadias and Down syndrome (<5) in our cohort (Supplemental Table 4), lower methylation at these CpGs remained associated with ART at birth.

Studies regarding methylation of the remaining CpGs were also limited. The CpG (cg24413339) identified in SPRN is also on chromosome 10, about 130kb from SYCE1. Although SPRN is produced in the gonads of male and female animals, suggesting a role for reproduction, its role has primarily been studied with regards to prion diseases (55). cg27073142 is located in the transcription start site of SORD, which codes for sorbitol dehydrogenase, and variants have been identified in relation to diabetic neuropathy (56). A whole genome bisulfite sequencing study identified a DMR in SORD associated with preeclampsia (n = 13) (57), but a large consortium of studies using microarrays have not replicated this finding for hypertensive disorders in pregnancy (58). cg13403462 and cg14921437 are located in variants of NE-CAB3 and ACTL3. NECAB3 codes for a neuronal calcium binding protein, which has been examined in schizophrenia (59). ACTL3 has only been investigated in relation to cancer (60). cg08681856 is in KLK3 and its methylation was associated with maternal plasma folate levels (61) but has otherwise only been examined in cancer. Lastly, cg02289287 is in a long noncoding RNA and cg01061626 is in a predicted gene (KIAA2013), both of which have unknown function.

Perhaps of greater interest are the regions identified to differ in maternally imprinted genes (IGF1R, PPIEL, SVOPL GNAS, L3MBTL, BLCAP, HYMAI/PLAGL1, SNU13, MEST). Of these, GNAS remained hypomethylated at 8–10 years of age and, as previously mentioned, was identified in previous studies (20, 22, 26, 62). As an imprinting gene, its methylation has been examined across a spectrum of disorders (63). With respect to reproduction, methylation of GNAS in spermatozoa was associated with idiopathic male infertility (64). Although previous studies have also observed hypomethylation at this gene with respect to ART, the impact, given small levels of hypomethylation were observed, is unclear. Evidence suggests its methylation could be associated with fetal growth, specifically with increased growth associated with exposure to gestational diabetes (65), but when we ran additional analyses adjusting for birthweight, the GNAS region remained associated.

Our analysis of a large contemporary cohort with longitudinal samples remains unique. We were able to include a comparison group receiving non-ART fertility treatment (OI/IUI) and measurements in mid-childhood. Nevertheless, we were limited by the number of ART cases to fully tease apart techniques such as frozen embryo transfer. Also, as the participants were spread out throughout New York State, we were unable to follow-up all children at clinics. Hence, although many differences were found at birth, it is difficult to conclude, as a previous study had (26), that differences “largely resolved” in later age because of the difference in sample size and a small overlap of children who were measured at both time points. Methylation “effects” of ART could arguably be more evident in newborns than in adults for whom postnatal exposures might have some confounding effects, but with an unequal sample size, it is difficult to compare. As with other studies, use of leukocyte DNA has unclear relevance for specific conditions because of the cell type specificity of methylation patterns. For the birth cohort, we were careful to use a recent cell reference panel that accounts for nucleated red blood cells with a larger group of newborn data than that in previous studies (40). Residual cell type differences may remain. This study including predominantly white participants from the geographical location of New York may not be generalizable to all of the United States but is more broadly inclusive than studies following children from a single clinic. Lastly, compared with the original cohort of children, those consenting to analysis had mothers with higher education, who were more likely white and were less likely to have smoked during pregnancy (44). As these factors are also associated with fertility treatment, we observed a larger proportion of the sample as conceived by fertility treatment compared with the proportion in the original cohort (40% vs. 30%). However, if anything, this helped make the comparison group more similar to the fertility treatment group. Furthermore, we did not see any differences in the proportion with congenital conditions for the nonincluded children compared with the current sample (data not shown).

CONCLUSION

We observed significant differences in DNA methylation of newborns conceived with ART but not with OI/IUI when compared with that in newborns of fertile controls. The techniques related to ART and the underlying causes of infertility as contributors to DNA methylation differences remain difficult to tease apart. Unfortunately, heterogeneity of practices across clinics do not lend well to gathering information on factors such as culture media from participants. However, evidence across studies suggested that methylation differences exist although different CpGs have been identified. Research both to understand the etiology of infertility and to pool together results across studies using the same analytic methods is needed to decipher the mechanisms further.