We report the first comprehensive analysis of differences in DNA methylation at a large number of CpG sites, including sites in all imprinted genes, between children conceived using ART and a control group who had natural conception. One goal of our analysis was to determine whether there was any relationship between CpG methylation levels at specific sites and mode of conception.
Our methylation results demonstrated that embryonic and extraembryonic tissue types are separated easily on the basis of methylation levels at specific CpG sites, as expected if CpG methylation plays a significant role in the epigenetic identity of cell and tissue type.
Because the site-specific CpG methylation levels showed no obvious clustering of in vivo
or in vitro
conceptions within tissues, we performed two-way ANOVA at each CpG site, testing for differences between the two groups. We observed an overall lower level of specific CpG site methylation in placenta and higher level in cord blood. If these differences are characteristic of differences in embryonic versus extraembryonic tissues, it may suggest differences in the way ‘outer’ and ‘inner’ blastomeres of pre-implantation embryos (29
) respond to in vitro
culture. Methylation of normally unmethylated CpGs at imprinted genes has been reported to occur sporadically in mouse embryos that have been cultured in vitro
). However, we are unaware of any other reports of widespread increases in CpG methylation (18% of all probes on the array) in embryonic tissues from human or animal embryos cultured in vitro
. The overall higher level of specific CpG site methylation observed in cord blood and lower level observed in placenta does not appear to be a property of only the CpGs that differ significantly but appears to be a characteristic of the probe set, as a whole. The overall lower level of methylation we observed in placenta in the in vitro
group is consistent with observations for several imprinted genes in placentae of mouse embryos subjected to prior culture in vitro
). The most likely alternative explanation is that the increases in CpG methylation in cord blood reflect a bias in the modest number of genes present on the array with respect to the potential for increases in methylation versus decreases in methylation; i.e. if the ‘normal’ state of most CpGs on the array in cord blood DNA is unmethylated/undermethylated, as appears to be the case by inspection of Fig. , then the only change possible to detect is to become more methylated.
Overall, our results indicate that conception in vitro is associated with small, but statistically significant, differences in methylation of CpG sites compared with individuals conceived in vivo. Whether this association is a result of the ART process itself, or a characteristic of the patient population served by ART, cannot be determined at this time.
We also investigated whether the CpG methylation differences observed predict differences in gene expression between the two groups. Our results suggest that a fraction of the observed differences in methylation are associated with differences in transcription of adjacent genes, however, neither the magnitude nor the direction of the change between groups can be predicted from the difference in CpG methylation at many of the sites examined. The magnitude of the effect on transcription appears to be relatively small for most genes, although a fraction of in vitro conceived individuals have transcript levels at individual genes that are outside the range found in the in vivo conceived population. Of note, in the cases for which a significant between-group difference is observed, is that the entire ART population appears to be shifted with respect to the control population in both methylation (Fig. ) and steady state transcript levels of individual genes (Fig. ). This relative shift in the population mean suggests that any effect of ART is relatively constant, and in the same direction for each individual.
We note that 6 of the 12 genes tested for differences in steady-state transcript level between the two groups (Table ) are transcriptionally imprinted (32
) and one (EGFR
) may be transcribed predominately from one allele (24
). One might have predicted, a priori, that imprinted genes are more likely to be influenced by environmental factors because only one allele need be activated or silenced to have a significant effect on transcription. In fact, a large fraction of the imprinted genes that have been tested in the extraembryonic tissues of mouse embryos subject to in vitro
culture do show aberrant expression of the normally silenced allele (21
). Although we observe that two of the six imprinted genes examined show significant differences in steady state transcript level, we do not observe enrichment for imprinted genes among those with significant differences because three of the six genes that are not known to be imprinted in Tables and also show differences in mean transcript levels in placenta. Whether this circumstance indicates that the genome of extraembryonic cells is especially sensitive to disruption by environmental factors or biological factors that occur in conjunction with infertility is a question that bears further investigation. In this regard, it is noteworthy that significant methylation differences were observed at only two of the six loci that showed significant transcript level differences in placenta (Tables and ). It is possible that this discrepancy is the result of meso-scale intra-individual placental mosaicism in the fraction of cells that contain a methylated CpG site or that express a particular allele (33
). All of the placental biopsies used in this study were taken from directly behind the umbilical cord insertion but adjacent samples could differ in some epigenetic parameters (35
). It is also likely that our failure to detect methylation differences at some loci reflects our inability to discriminate true methylation differences between the groups because of the small number of individuals assayed using the array. In addition, transcript levels of some genes may be more or less sensitive to gene methylation differences in placenta than in cord blood.
At this juncture, the differences observed in transcript levels are of unclear phenotypic significance. However, it is notable that one of the two genes exhibiting a significant difference in mean transcript level in blood (CEBPA
) and four of the six genes with significant differences in placenta (CEBPA
) have been linked to adipocyte development and differentiation (36
), insulin signaling and/or obesity (38
). Whether these gene regulatory abnormalities observed at birth might be connected to birth weight, obesity, type II diabetes or hypertension later in life is a significant concern. Such differences have the potential to affect many aspects of embryonic development and fetal growth, as well as influence long-term patterns of gene expression that might be associated with increased risk of many human diseases.
In conclusion, we have shown that in vitro conception is associated with quantitative differences in DNA methylation and that some of these differences may have a significant effect on gene expression. There seem to be only two likely explanations for these differences between children conceived by ART and children conceived in vivo; either infertile couples, themselves, are more likely to have gametes that bear abnormal epigenetic marks or some aspect of the ART process causes the epigenetic differences we observe. Determination of which of these possibilities is most likely is an important question that may be approached by comparing the offspring of couples who are infertile as a result of physical barriers to fertilization, such as tubal ligation, with those who are infertile for other reasons. If ART is causal of the methylation and gene expression differences we observe, ART procedures may be modified to minimize these effects after testing in a suitable animal model.