Assisted reproduction: the epigenetic perspective

Abstract

Developmental pathways in humans and other organisms are buffered against changes in genotype and environment. Therefore, it should not come as a surprise that most of the children conceived by assisted reproduction technology (ART) are healthy, although ART bypasses a lot of biological filters and subjects the gametes and the early embryo to environmental stress. If, however, the buffer breaks down, the development of certain tissues or organs may follow abnormal trajectories. We argue that both normal and abnormal development in children conceived by ART can be explained by epigenetic mechanisms, which control the establishment and maintenance of gene expression patterns in the placenta and fetus. Imprinted genes are of special importance in this respect. There is increasing evidence that genetic factors in infertile couples as well as environmental factors (hormones and culture media) can have adverse effects on epigenetic processes controlling implantation, placentation, organ formation and fetal growth. In addition, loss of epigenetic control may expose hidden genetic variation.

Introduction—epigenetics and development

During specific stages of development, only specific sets of genes are active. The accessibility of transcription factor binding sites on the DNA is determined by specific patterns of chromatin modifications and DNA methylation. To an approximation, chromatin exists in a transcriptionally competent or a transcriptionally silent state. Acetylation of histone H3 at lysine 9 (H3K9), for example, is typically found in transcriptionally competent chromatin, whereas H3K9 methylation is a hallmark of the transcriptionally silent state. In mammals, the silent state of certain regions is also marked by the methylation of cytosine residues located in CpG dinucleotides at the 5′ end of genes. When genes are switched on or off by a transcriptional activator or repressor at a specific time point during development, the activity states often persist for several rounds of cell divisions, even if the primary signal has faded away. The persistence of gene activity states is controlled by the epigenotype. It is cell heritable, but potentially reversible. In general, it is erased during germ cell development, but there are examples of germ line transmissions (Chong and Whitelaw, 2004)

Although epigenetic states are relatively stable, it has been estimated that the loss of epigenetic control (epimutation) may be one or two orders of magnitude greater than that of somatic DNA mutation (Bennett-Baker et al., 2003). This suggests that the contribution of epimutations to human disease is probably underestimated. Recent data have shown that epimutations not only lead to inappropriate expression or repression of the affected gene, but can also expose hidden genetic variation (Sollars et al., 2003). As proposed by C.H. Waddington in his famous canalization model, developmental pathways are buffered against changes in genotype and environment (Waddington, 1959). Buffering allows the accumulation of genetic variation within a population without its manifestation at the phenotypic level (Rutherford and Henikoff, 2003). If, however, under certain circumstances the buffer fails, hidden genetic variation can be phenotypically expressed. Rutherford and Lindquist (1998), for example, found that reductions in Hsp90 (heat shock 90-kDa protein 1α) chaperone activity in Drosophila melanogaster caused phenotypic variation of seven adult structures of the fly, providing a plausible mechanism by which environmental stress can produce heritable morphological variants. D. Ruden and coworkers have recently expanded these findings by showing that Hsp90 affects development by altering the chromatin and that the disruption of chromatin-remodelling pathways can have similar consequences (Sollars et al., 2003). Interestingly, this was observed to be a predominant maternal effect. Thus, the structure of chromatin, i.e. the epigenotype before the onset of zygotic transcription, seems to affect transcriptional events occurring much later during development (Rutherford and Henikoff, 2003).

Given the canalization, or robustness of development, it is not surprising that the vast majority of children conceived by assisted reproduction technology (ART) are healthy, although ART bypasses a lot of biological filters such as selective gamete resorption, selective sperm uptake, sperm competition and selective syngamy, and subjects the gamete and early embryo to environmental stress (hormones, culture media and physical stress). On the other hand, developmental plasticity provides organisms with the ability to develop a certain range of phenotypes in response to environmental cues. Such cues probably affect gene expression by inducing epigenetic changes. We think that low birth weight (Schieve et al., 2002), congenital malformations (Hansen et al., 2002; Katalinic et al., 2004) and imprinting disorders (Gosden et al., 2003) in some ART children can, at least in parts, be attributed to epigenetic variation. This is not to say that these problems are solely or mainly caused by the technique. It is possible that some subfertile couples have a genetic predisposition to epigenetic instability, which makes their offspring more susceptible to epigenetic changes—independently of whether or not they are conceived by ART.

Genomic imprinting

Imprinting mechanisms

Genomic imprinting is an epigenetic process by which the male and the female germ line of therian mammals confer a sex-specific mark (imprint) on certain chromosomal regions. As a consequence, the paternal and the maternal genome are functionally nonequivalent and both are required for normal embryonic development (McGrath and Solter, 1984; Surani et al., 1984). The parental copies of imprinted regions differ with respect to DNA methylation, histone modification and consequently gene expression. Despite the identification of parent-of-origin specific differences of imprinted chromosome regions, the nature of the primary imprint is still a matter of debate.

Genomic imprints are erased in primordial germ cells, newly established during later stages of germ cell development, and stably inherited through somatic cell divisions during postzygotic development (Figure 1A). They survive the global waves of DNA demethylation and remethylation during early embryonic development (Mayer et al., 2000), although it is unclear what protects them. In somatic cells, the imprint is read by the transcription machinery and used to regulate parent-of-origin specific gene expression so that only the paternal or the maternal allele of a susceptible gene is active.

To date, approximately 80 imprinted genes have been identified. Many imprinted genes are involved in regulating resource acquisition of the embryo and fetus. In fact, it has been proposed that imprinting coevolved with the placenta. In therian mammals, the fetus grows at the expense of the mother. As proposed by the genetic conflict theory (Wilkins and Haig, 2003), the paternal genome is ‘interested’ in extracting as many resources from the mother as possible. By contrast, maternally inherited genes protect the mother from being exhausted by the fetus (Constancia et al., 2004).

Imprinted genes are not randomly distributed in the genome, but tend to occur in clusters. This suggests that the primary control of imprinting is not at the single gene level, but at the chromosome domain level. Indeed, several clusters have been found to contain a cis-acting imprinting centre (IC) which controls imprint establishment and imprint maintenance (Buiting et al., 1995).

Imprinting defects: the role of genetic factors

IC mutations lead to aberrant imprinting of the affected chromosome domain and cause disease. However, imprinting defects can also occur in the absence of an IC mutation. In fact, in most patients with a recognizable syndrome and an imprinting defect, the aberrant imprint is a primary epimutation that occurred spontaneously without any DNA sequence alteration (Buiting et al., 2003). Imprinting defects have been found in approximately 1% of patients with Prader–Willi syndrome, 3% of patients with Angelman syndrome (AS), 50% of patients with Beckwith–Wiedemann syndrome (BWS) and 50% of patients with transient neonatal diabetes mellitus .

Figure 2 illustrates imprinting defects in AS and BWS. AS is a rare neurogenetic syndrome characterized by severe mental retardation, lack of speech, jerky movements and a happy disposition (incidence 1/15 000 newborns). It is caused by the loss of function of the UBE3A gene, which encodes an enzyme involved in targeted protein degradation. In the brain, the gene is active on the maternal chromosome only. In contrast to many other imprinted genes, monoallelic expression of UBE3A is not associated with differential DNA methylation of the promoter/exon 1 region. There is some tentative evidence that the paternal allele is silenced by an antisense RNA, which originates at the neighbouring SNRPN locus (Rougeulle et al., 1998; Runte et al., 2001). In normal individuals, SNRPN is methylated on the maternal chromosome and expressed from the paternal chromosome (Ozcelik et al., 1992; Zeschnigk et al., 1997). In AS patients with an imprinting defect, the maternal SNRPN allele is unmethylated and expressed, and the maternal UBE3A allele is silenced. Likewise, in the majority of BWS patients with an imprinting defect, the maternal KCNQ1OT1 antisense allele is unmethylated and expressed, and the maternal CDKN1C allele is silenced (Diaz-Meyer et al., 2003). BWS is an overgrowth syndrome characterized by high birth weight, hypoglycemia, macroglossia, exomphalos and increased risk of Wilms’ tumour (estimated incidence, 1/20 000 newborns).

Epimutations affecting imprints can arise during imprint erasure, imprint establishment or imprint maintenance (Figure 1B–D). Primary imprinting mutations represent stochastic events which occur at a certain probability, but there is also suggestive evidence that the error rate is influenced by genetic and environmental factors. It is possible, for example, that apparently neutral sequence variants of imprinting control regions are associated with an increased or decreased error rate (Murrell et al., 2004). The same may be true for sequence variants of imprinting factors that act in trans, such as DNA methyltransferases, but this has not yet been investigated.

As shown in animal studies, loss-of-function mutations of DNA methyltransferases affect all imprinted domains as well as other chromosomal regions. Mutations of the de-novo methylase Dnmt3a and of the maintenance methylase Dnmt1, for example, lead to loss of imprinting and embryonic lethality (Jaenisch and Bird, 2003). Dnmt3L does not have enzymatic activity, but is important for de-novo methylation and imprint establishment in the gametes. Targeted disruption of Dnmt3L in mice caused azoospermia in homozygous males, and heterozygous progeny of homozygous females died before midgestation. As shown by bisulfite genomic sequencing of DNA from oocytes and embryos, removal of Dnmt3L prevented methylation of sequences that are normally maternally methylated. The defect was specific to imprinted regions and caused biallelic expression of genes that are normally expressed only from the allele of paternal origin (Bourc’his et al., 2001; Bourc’his and Bestor, 2004). In humans, DNMT3L is only detected after fertilization (Huntriss et al., 2004). Therefore, the mechanism and/or timing of imprint establishment in humans, at least at some loci, may be different, and human embryogenesis may be more vulnerable to environmental cues than murine embryogenesis.

Imprinting defects: the role of environmental factors

Both DNA methyltransferases and histone methyltransferases use S-adenosyl-methionine (SAM) as a methyl donor. SAM levels are dependent on folic acid, and the enzymes involved in one-carbon metabolism use vitamin B cofactors. As folic acid and vitamin B are provided by nutrition, it should not be surprising that epigenetic states can be influenced by the diet. Changes in DNA methylation by folate have been observed in various types of cancers as well as in animal models (Garfinkel and Ruden, 2004). However, there is little evidence that genomic imprints can be affected by nutrition. After folate washout in patients with hyperhomocysteinaemia, Ingrosso et al. (2003) observed biallelic expression of H19, which normally is expressed from the maternal allele. After folate treatment, they observed a shift back to monoallelic expression. It should be noted, however, that the authors did not study the genomic imprints, i.e. the methylation patterns directly, and that expression changes can occur in the absence of imprint changes.

There is some evidence that imprints can be impaired by hormone treatment and cell culture, which are essential components of ART. As ART does not deal with primordial germ cells, but only with gametes and the early embryo, it can not affect the process of imprint erasure, but may interfere with imprint establishment or imprint maintenance. In this respect, it is interesting to note that different loci acquire their imprints at different stages of gamete development (Lucifero et al., 2004). Although there is some controversy about the exact time at which the SNRPN imprint, for example, is established (El-Maarri et al., 2001; Geuns et al., 2003), it is probably one of the latest imprints. On the other hand, the SNRPN imprint is rather stable in cell culture (unpublished data). Thus, it is tempting to speculate that SNRPN may be more vulnerable to hormone treatment, which is used for superovulation, than to embryo culture, whereas other loci may be more vulnerable to cell culture than to hormone treatment.

As the effects of embryo culture on gene expression and mammalian development has been reviewed in this journal before (Khosla et al., 2001), we will not focus on this topic in this review. However, one recent study needs to be discussed. Ecker and colleagues have recently reported that the culture of preimplantation mouse embryos affects the behaviour of adult mice (Ecker et al., 2004). To exclude a potential effect of different maternal environments, which is a confounding factor in many such studies, the authors have used an experimental design in which genetically marked embryos from different culture conditions were transferred at the blastocyst stage to the same foster mother. They did not find any effect of culture on the incidence of development to term and on sensory and motor development, but on the behavioural phenotype of adult mice such as anxiety, locomotor activity and spatial memory. ‘Cultured’ mice were less anxious, but had a poorer spatial memory. It is difficult to extrapolate these findings to humans, but they call for more behavioural studies in IVF children and adults. So far, however, none of the studies in this field, not even prospectively controlled studies, have found a difference in the behaviour of children born after IVF or ICSI compared to those conceived spontaneously (Cederblad et al., 1996; Golombok et al., 1996, 2001; Levy-Shiff et al., 1998; Sutcliffe et al., 2003; Barnes et al., 2004).

Although we do not know of any study which has directly investigated the effect of hormone treatment on imprinted gene expression, there is indirect evidence for such an effect in mouse, cow and humans. Gonadotrophins are commonly used for superovulation to obtain sufficient numbers of oocytes for experimental studies in mice and assisted reproduction in humans. However, superovulation with gonadotrophins appears to have adverse effects on implantation and fetal development. It has been reported that hormone treatment increases the risk of implantation failure, hypertension, bleeding, placenta praevia and low birth weight (Ertzeid and Storeng, 1992; Tanbo et al., 1995; Maman et al., 1998). In order to determine whether the adverse effects are caused by impaired oocyte competence and/or changes in uterine milieu, Ertzeid and Storeng (2001) have performed a carefully designed study based on an embryo donation model. Embryos from superovulated and non-stimulated females were transferred to separate uterine horns within the same superovulated or non-stimulated pseudopregnant recipient mice. They found that ovarian stimulation impairs both oocyte quality as well as uterine milieu. As compared to embryos from non-stimulated mice, embryos derived from superovulated donor mice developed less frequently to the blastocyst stage.

The latter finding was confirmed and expanded by Shi and Haaf (2002), who investigated the methylation pattern at the two-cell stage of mice. The fertilized mouse egg actively demethylates the paternal genome within 4 h after fertilization, whereas the maternal genome is only passively demethylated by a replication-dependent mechanism after the two-cell stage. Genomic imprints are protected against demethylation, but aberrations in the genome-wide methylation pattern can serve as an indicator of epigenetic reprogramming errors. Shi and Haaf found that 10% of embryos derived from non-stimulated mice had an abnormal methylation pattern as determined by immunofluorescence staining with an antibody against 5-methyl-cytosine, compared to 20% of embryos derived from superovulated mice. The authors also found that certain media used for embryo culture increased the rate of abnormal methylation patterns.

In livestock breeding programs, the occurrence of large offspring syndrome (LOS) is quite alarming (Ceelen and Vermeiden, 2001). LOS is reminiscent of the BWS in humans. Oocyte maturation and culture conditions have been found to change the methylation and expression pattern of imprinted genes (Young et al., 2001). In cows, oocyte maturation with high levels of gonadotrophins is an integral component of IVF. Until there is more scientific data, oocyte maturation in humans should be regarded as experimental and potentially risky with regard to imprinting defects. In our mind, in vitro maturation of oocytes should be limited to prospective studies with close follow-up of pregnancies and children born. This is even more important, since the clinical experience with pregnancies and birth after in vitro maturation in human is still limited (Chian et al., 2004).

Imprinting defects in children conceived by ART

Cox et al. (2002) suggested that ART might be associated with an increased risk of imprinting defects. The authors described two children with AS and an imprinting defect who were conceived by ICSI. In both cases, an IC deletion was excluded. Thus, the imprinting defects were primary epimutations that had occurred spontaneously. One child had somatic mosaicism, which indicates that the imprinting defect had occurred after fertilization, i.e. as the result of an imprint maintenance failure in an early embryonic cell. In 2003, another ICSI child with AS and an imprinting defect was reported (Orstavik et al., 2003). Previously, single IVF/ICSI children with BWS, but without analysis of the genetic defect had been described (for review see Gosden et al., 2003). It is of interest to note that imprinting defects in AS and BWS are characterized by hypomethylation of imprinting control regions on the maternal chromosome (Figure 2)

So far, follow-up studies of cohorts of children conceived by ART have not revealed an increased risk for imprinting disorders. However, imprinting disorders are so rare that a slight to moderate increase in incidence after ART can not be detected in single centre or nationwide follow-up studies, which typically involve less than 10 000 children. In order to address the question, whether ART is associated with an increased risk of imprinting disorders, international multi-centre studies on more than 100 000 children would be needed. Alternatively, the prevalence of ART in patients with an imprinting disorder or the prevalence of an imprinting defect in patients born to a couple that underwent ART can be determined. Such patient-based studies are more feasible than large follow-up studies, but require either disease registers or parents support groups that can provide the necessary information.

Using such a study design, three groups reported a 3–6-fold increased prevalence of ART in children with BWS (DeBaun et al., 2003; Gicquel et al., 2003; Maher et al., 2003). These authors compared the proportion of ART births in patients with BWS to that in the general population. Statistical dependence on the proportion of ART births, however, is a problem, because there might have been a reporting bias of patients conceived by ART, and the frequency of ART births in the general population might have been higher than estimated.

A large case-control study has recently been reported by an Australian group (Halliday et al., 2004). Among 1 316 500 live births in Victoria between 1983 and 2003, the authors identified 37 cases of BWS. Record linkage of these cases and 148 matched controls identified ART as the method of conception in four BWS cases and one control, giving an odds ratio (OR) of 17.8 [95% confidence interval (CI) 1.8–432.9], and Fisher’s-exact-test two-sided P = 0.006. Three of three ART patients studied had an imprinting defect. The authors estimate the risk of BWS in their IVF population as 1/4 000, or nine times greater than in the general population. Despite this highly increased risk, the absolute risk of conceiving a child with BWS after ART remains very low (approximately 1/5 000 newborns). Therefore, prenatal testing for BWS or AS after ART is usually not indicated.

In view of the experimental data in animals and the epidemiological evidence in humans, it is likely that ART is associated with an increased risk of imprinting disorders. Assuming the association were genuine, it is unclear whether the risk can be attributed to infertility itself and/or the technique. With regard to the technique, it is unclear whether hormonal stimulation, gamete manipulation or the culture conditions are a problem. In a recent case series on patients with BWS, the only consistent finding was that all women had received ovarian stimulation (Chang et al., 2005). To shed more light on these questions, we have conducted a cohort study on patients with AS (Ludwig et al., 2005). We have found an increased prevalence of imprinting defects in patients with AS born to subfertile couples (defined as having had a time to pregnancy >2 years and/or infertility treatment; relative risk, 6.25; 95% CI 1.68–16.00). Interestingly, the relative risk was the same in untreated couples with time to pregnancy >2 years and in couples treated by ICSI or hormonal stimulation alone, although the increase did not reach statistical significance, possibly because of small numbers. The relative risk was highest in couples with time to pregnany >2 years and infertility treatment (relative risk, 12.50; 95 CI 1.40–45.13). Our findings suggest that imprinting defects and subfertility can have a common, possibly genetic cause, and that superovulation rather than ICSI may further increase the risk of conceiving a child with an imprinting defect. Based on this study and the study by Chang et al. (2005), it is tempting to speculate that superovulation leads to the maturation of epigenetically imperfect oocytes that would not have been ovulated without treatment or disturbs the process of DNA methylation in the oocyte. Evidence for such an effect comes from studies in mice (Shi and Haaf, 2002; Alexandridis et al., 2005).

As our cohort of patients did not contain children conceived by conventional IVF without ICSI, we do not know whether culture conditions may also increase the risk. Based on the finding that the SNRPN methylation imprint is established rather late in oocyte development, but relatively stable under culture conditions (see above), it is tempting to speculate that genetic predisposition and hormone stimulation is a greater risk factor for AS imprinting defects than the culture conditions are and that the reverse may be true for BWS.

The epigenetic control of implantation and placentation

Imprinted genes play a pivotal role in the development and function of the placenta. They regulate nutrient supply to the fetus by affecting overall growth of the placenta or of particular structures (such as the labyrinthine trophoblast) or by affecting specific transporters and channels (Reik et al., 2003). It is therefore no surprise that imprinting errors can lead to spontaneous abortions. As mentioned above, heterozygous progeny of Dnmt3L^–/– mice die before midgestation, because they lack maternal methylation imprints (Bourc’his et al., 2001). These mice have a generalized imprinting defect. As some genes are preferentially or exclusively imprinted in the placenta, it is also conceivable that single locus defects increase the risk of placental dysfunction and abortion. Of course, abortions do not only result from epigenetic defects, but epigenetic defects may account for a significant proportion of abortions which are not associated with chromosomal aberrations of the placenta and the embryo.

It is well documented that the risk of spontaneous abortions is increased in women who underwent infertility treatment. Based on this observation, several textbook authors have concluded that ART increases the risk of spontaneous abortions. However, recent studies suggest that not only the way of conception, but also problems residing in the patients themselves increase the risk (Pezeshki et al., 2000; Wang et al., 2004). Within the ART cohort in one of these studies (Wang et al., 2004), the risk of spontaneous abortion was significantly increased in patients with a higher level stimulation as compared to those with low level of stimulation (P < 0.01). For the IVF group, the risk was higher as compared to the gamete intrafallopian transfer (GIFT) group, even after adjustment for age and other factors (GIFT versus. IVF: relative risk (RR) 0.74 [CI 0.34–0.99]). The results may point out that ovarian stimulation as well as the prolonged in vitro culture might increase the risk of spontaneous abortion, since GIFT does not involve long-term in vitro culture as it is done in IVF but only the retrieval of oocytes with the direct subsequent transfer into the fallopian tube.

Genetic and epigenetic factors also appear to play a role in other pregnancy complications. Pandian et al. (2001) performed a retrospective study over a 10-year period and identified patients with idiopathic infertility who had more than 1 year time-to-pregnancy (n = 877). Interestingly, independently of whether the singleton pregnancies in the study group of subfertile couples were established by infertility treatment or spontaneously, there was an increased risk of abruptio placentae (RR 3.05 [95% CI 1.4–6.2]) and pre-eclampsia (RR 5.61 [95% CI 3.3–9.3]) even after adjusting for patients’ age and other factors. The association between infertility and an increased risk of pre-eclampsia was confirmed by others. Basso et al. (2003) reported data from the Danish National Birth Cohort including 44 732 pregnancies with a full data set. The risk for pre-eclampsia was increased after a time-to-pregnancy exceeding 1 year. Data were adjusted for irregular cycles, age at delivery, body mass index (BMI) and smoking.

Another study from Denmark showed an increased risk of preterm delivery to be associated with time-to-pregnancy (Henriksen et al., 1997). Two cohorts of patients with more than 12 000 women were included in the analysis. The risk of preterm delivery was 1.6–1.8-fold increased in patients with time-to-pregnancy exceeding 1 year, irrespective of whether the patients conceived spontaneously or after infertility treatment. Other data, again from the Danish National Birth Cohort, were analysed for the same question (Basso and Baird, 2003). The authors described an increased risk for premature birth of 1.36 [95% CI 1.08–1.71] for primiparas and 1.57 [1.20–2.05] for multiparas for those patients who conceived without treatment after more than 12 months time-to-pregnancy. The estimates were adjusted for mother’s age, BMI, smoking, social status, sex of the baby, age at menarche, cycle regularity and length. This analysis included more then 55 000 women.

Finally, a study on perinatal mortality from the UK should be mentioned (Draper et al., 1999). In this case-control study, a history of infertility was found to increase the chance for perinatal death by an OR of 2.9 [95% CI 1.8–4.5] irrespective of treatment. Women with untreated infertility had a comparable increased risk like those with treated infertility with an OR of 3.3 [95% CI 1.6–6.8] and 2.7 [95% CI 1.5–4.7], respectively. The risk in the treated group was unrelated to the kind of treatment (hormonal stimulation only or ART). Infertility in that study was defined as any mention in the medical records of delayed conception or investigation or treatment for delayed conception. Interestingly, the risk for perinatal mortality in the treated group was reduced to 1.5 [95% CI 0.8–2.9] when multiple pregnancies were taken into account. The risk in the untreated group with infertility did not change, since multiple pregnancies were not a confounder. Mainly prematurity was the risk leading to perinatal mortality, confirming the results of the studies mentioned above (Henriksen et al., 1997; Pandian et al., 2001; Basso et al., 2003).

It is a matter of debate, whether some of the pregnancy complications are associated with loss of epigenetic control. Mice with paternal uniparental disomy 12, for example, show severe deficits in placentar vascularity and are a good model not only for implantation problems, but also for an increased risk of abnormal placentation, spontaneous abortions, pre-eclampsia and premature birth. (Georgiades et al., 2000). Another recently published study in 24 families with pre-eclampsia identified a susceptibility locus on chromosome 10q22.1 (Oudejans et al., 2004). The linkage and expression data in that study suggest that pre-eclampsia involves maternally expressed imprinted genes that operate in the first trimester placenta.

It is tempting to speculate that differences in placental function—because of the changes in epigenetic control—may also explain the well-known differences in placental hormone synthesis in IVF pregnancies. These abnormalities lead to a higher rate of conspicuous results in biochemical first trimester screening (Ribbert et al., 1996; Wald et al., 1999; Liao et al., 2001; Wojdemann et al., 2001; Maymon and Shulman, 2002; Ghisoni et al., 2003; Hui et al., 2003; Muller et al., 2003). Data for other kinds of infertility treatment, however, are not known.

The epigenetic control of birth weight

As shown in several studies, the use of ART accounts for a disproportionate number of infants with low birth weight and very low birth weight. This is in part due to an increased number of multiple gestations, but also to higher rates of low birth weight in singleton infants (Schieve et al., 2002; Katalinic et al., 2004). An increased risk of low birth weight was also seen in infants conceived with oocytes from apparently fertile persons and in infants from pregnancies carried by women who were unlikely to have an underlying uterine or other infertility-related disease (Schieve et al., 2002). By contrast, the birth weight of children born by gestational carriers was not significantly different from that of spontaneously conceived children, but the numbers were very small. Schieve et al. (2002) conclude from their study that low birth weight is directly related to ART, although the mechanism underlying this apparent association is unknown.

Recent data suggest that ovarian stimulation may be one factor. As shown by Keizer et al. (2004) singletons born after conventional stimulation IVF have a lower birth weight compared to singletons born after natural cycle IVF with minimal stimulation using gonadotrophins. Likewise, Mitwally et al. (2004) demonstrated a negative correlation between estradiol levels achieved by ovarian stimulation and subsequent birth weight. Both studies have not yet been published in detail, and a selection bias explaining these observations can not be excluded. This is also mentioned by one of these groups (Keizer et al., 2004).

Why should ovarian stimulation decrease birth weight? Schieve et al. (2002) refer to the work of Johnson et al. (1995) who showed that the use of gonadotrophins increases the level of insulin-like growth factor-binding protein 1 (IGFBP-1), which has been linked to intrauterine growth restriction. The problem with this explanation is that gonadotrophins are used for superovulation, i.e. before fertilization. Thus, they should have no direct effect on the levels of IGFBP-1 during intrauterine growth. For superovulation to have an effect on embryonic and fetal gene expression, one has to assume that gonadotrophins interfere with the establishment of the epigenetic pattern of the oocyte or that superovulation increases the rate of aberrantly programmed oocytes. Although these are likely explanations for an increased risk of maternal imprinting defects (see above), the negative effect of superovulation on fetal growth is more difficult to understand.

Resource acquisition of the embryo and fetus is tightly controlled by imprinted genes (Miozzo and Simoni, 2002). These genes operate in the placenta, the fetus, or both. Paternally expressed genes such as PLAGL1, PEG1/MEST, IGF2 and PEG3 enhance fetal growth, whereas maternally expressed genes such as IGF2R, GRAB10, H19, CDKN1C, TSSC3, MASH2 and MEG3 restrict fetal growth. Rare overgrowth syndromes in humans and animals are caused by the loss of maternal gene expression (CDKN1C in BWS and IGF2R in LOS; see paragraph on Imprinting defects). Low birth weight may be caused by the loss of paternal expression of a growth promoting gene or by the aberrant activation of a normally silent paternal allele of a growth-restricting gene. In both cases, the problem appears to reside on the paternal genome, and thus, low birth weight may be linked to a paternal-specific factor.

Marques et al. (2004) have recently described that some oligozoospermic men have sperm with defective methylation of the CTCF binding site between IGF2 and H19. These two genes share two enhancers located downstream of H19. In normal cells, the CTCF binding site is unmethylated on the maternal chromosome. The binding of CTCF to this site prevents the enhancers from activating IGF2, and H19 is active. On the paternal chromosome, the site is methylated and unable to bind CTCF. As a consequence, IGF2 is active and H19 is silent (Hark et al., 2000). The prediction would be that transmission of an unmethylated paternal allele (with the help of ART) leads to the presence of two inactive IGF2 alleles and two active H19 alleles in the embryo and hence growth retardation. If this scenario were correct, we should see an increased rate of low birth weight after ICSI, which is mainly performed because of male-factor sterility, but not after IVF without ICSI or after hormone treatment only. However, this is not the case.

There are at least two other possible explanations. One is that there is an imprinting defect on the paternal genome, but that it was induced by the oocyte. As mentioned in the paragraph on Imprinting defects, the oocyte cytoplasm demethylates the paternal genome within the first few hours after fertilization, but spares imprints (Mayer et al., 2000). It is possible that the demethylation machinery in superovulated oocytes is somehow compromised and more error prone than in normal oocytes. As a consequence, some paternal methylation imprints may be removed. The second possibility is that non-imprinted genes involved in fetal growth control carry an epimutation.

E. Maher has pointed out correctly that low birth weight in children conceived by ART may have important lifelong implications for health (Maher et al., 2003). This is because epidemiological studies have suggested that there may be a link between low birth weight and adult insulin insensitivity and cardiovascular disease (Barker et al., 1993; Gluckman and Hanson, 2004). As suggested by Gluckman and Hanson (2004), there are several mechanisms, which operate at different levels, that may account for this association. First, epigenetic changes may alter long-term gene expression programs. Second, tissue differentiation may be altered. For example, effects on the allocation of blastocyst stem cells to inner cell mass or trophectoderm lineages appear to influence the relative growth trajectories of the placenta and fetus. Lastly, changes in homeostatic control mechanisms may be induced. The various prenatal changes may alter the organism’s response to the post-natal environment. As most of the children conceived by ART are below the age of 20, we will have to wait for many more years to see whether these individuals are at an increased risk to develop adult-onset chronic disease.

The epigenetic control of pattern formation and morphogenesis

Differential gene expression is one of the most important mechanisms underlying development. Spatially and temporally restricted expression of the homeobox (HOX) genes along the main and appendicular axes, for example, is essential for correct patterning of the early vertebrate embryo. After the HOX gene expression domains have been established, the Polycomb and trithorax group proteins perpetuate the activity state of the HOX genes within and outside the expression domains (Deschamps et al., 1999). The Polycomb and trithorax group proteins are epigenetic regulators involved in forming repressive and active chromatin, respectively. It is obvious that defects in establishing and maintaining gene expression patterns lead to malformations.

There are several reports that children born after ART may have a 1.4–2.0-fold increased risk of malformations (Hansen et al., 2002; Katalinic et al., 2004), although other studies have not found such an effect. Two meta-analyses on this question have estimated the risk to be about 1.3-fold (Rimm et al., 2004; Hansen et al., 2005). Especially prominent are cardiovascular, urogenital and musculoskeletal defects, but there is no specific pattern. Factors that may increase the risk of birth defects include the advanced age of the infertile couples and the underlying cause of their infertility. For example, chromosomal aberrations in an individual can affect his/her fertility as well as the development of the offspring (Meschede et al., 1998; Scholtes et al., 1998; Peschka et al., 1999).

For some time, it was suggested that ICSI or the male factor may be responsible for the increased risk of major malformations. However, since there is no difference between children born after conventional IVF and ICSI (Bonduelle et al., 2002; Hansen et al., 2002; Rimm et al., 2004; Lie et al., 2005), but an increased risk for major malformations for conventional IVF as well as for ICSI as compared to spontaneously conceived children (Hansen et al., 2002; Katalinic et al., 2004; Hansen et al., 2005), the story is not so simple. The severeness of the male factor or the origin of sperm—ejaculate, epididymis, testicle—also did not influence the risk of major malformations significantly in the largest prospective controlled study to date (Ludwig and Katalinic, 2003). It would be helpful to have valid data for larger cohorts of children born after hormonal stimulation only or intrauterine insemination. These data would help to prove whether the in vitro culture or hormonal stimulation or the underlying reduced fertility are responsible for the observed increased risk. We would like to suggest that part of the problem is of epigenetic nature.

Although malformation syndromes have not yet been linked to epimutations, it is not unreasonable to assume that epimutations may cause dysgenesis or dysplasia of certain tissues or organs. It is possible, for example, that a master developmental gene is silenced by an epimutation, similar to the silencing of a tumor suppressor gene in the development of hyperplasia and cancer (Greger et al., 1989). However, in contrast to the latter two conditions (for example hyperplasia of the tongue in patients with BWS), dysgenesis and dysplasia, which underlie most of the malformations, are much more difficult to study at the molecular level. Owing to the reduced size or absence of the affected structure, it is difficult or impossible to obtain a tissue sample. The availability of such a sample, however, is necessary, because epimutations often occur in a mosaic form and can be found only in the affected tissue. Epimutations of master developmental genes may occur on the basis of a hereditary predisposition to epigenetic instability and/or as the result of environmental influences.

Another epigenetic explanation for the apparently increased occurrence of malformations in children conceived by ART is the exposure of hidden genetic variation. This explanation is speculative, but not unreasonable. As mentioned in the Introduction, developmental pathways are buffered against changes in genotype and environment. Under stress, the buffer may fail and heritable morphological variants may occur. It is tempting to speculate that genetic factors associated with infertility or that ART-induced stress may expose genetic variants that cause malformations.

Conclusions

The adverse health outcome in some ART children may in part be due to the loss of epigenetic control during development (summarized in Figure 3). Although there are several studies on the influence of hormones and in vitro culture on methylation patterns and imprinted gene expression, next to nothing is known about possible genetic factors that predispose to the loss of epigenetic control. The loss of epigenetic control, triggered by environmental and genetic factors, may directly change developmental trajectories and/or expose hidden genetic variance. Studies on these topics are difficult to perform in humans, but likely to yield important results, both for basic research as well as for medical treatment.

Acknowledgements

We thank Dr Karin Buiting for helpful discussions. Part of the work performed by the authors was supported by the Deutsche Forschungsgemeinschaft and the Gesellschaft zur Förderung der Endokrinologischen Forschung (Endokrinologikum Hamburg).

References

Author notes

¹Institut für Humangenetik, Universitätsklinikum Essen, Essen and ²Endokrinologikum Hamburg, Hamburg, Germany