Journal of Evolutionary Biology

Selective abortion and the evolution of genomic imprinting


Jason B. Wolf, Department of Biology and Biochemistry, University of Bath, Claverton Down, Bath BA2 7AY, UK. Tel./fax: +44 122 538 5012; e-mail:


Mothers can determine which genotypes of offspring they will produce through selective abortion or selective implantation. This process can, at some loci, favour matching between maternal and offspring genotype whereas at other loci mismatching may be favoured (e.g. MHC, HLA). Genomic imprinting generally renders gene expression monoallelic and could thus be adaptive at loci where matching or mismatching is beneficial. This hypothesis, however, remains unexplored despite evidence that loci known to play a role in genetic compatibility may be imprinted. We develop a simple model demonstrating that, when matching is beneficial, imprinting with maternal expression is adaptive because the incompatible paternal allele is not detected, protecting offspring from selective abortion. Conversely, when mismatching is beneficial, imprinting with paternal expression is adaptive because the maternal genotype is more able to identify the presence of a foreign allele in offspring. Thus, imprinting may act as a genomic ‘cloaking device’ during critical periods in development when selective abortion is possible.


In sexually reproducing organisms, offspring genotypes result from combinations of paternal and maternal genotypes and thus offspring essentially represent an alien organism in relation to maternal genotype, i.e. an allograft (Clark et al., 1987, 1999; Ober et al., 1998; Richman & Naftolin, 2006). This dissimilarity causes a fundamental problem in mammals (in which young are produced inside the mother’s body) whereby the offspring’s dissimilar genotype produces antigens that would trigger a maternal immune response (Medawar, 1953; Richman & Naftolin, 2006; e.g. in rhesus factors). On the other hand, genetic similarity at specific loci such as the major histocompatibility complex (MHC; which governs immune responses by presenting antigens at the cell surface for inspection of T-cell antigen receptors) is often selected against resulting in a high abortion rate of MHC homozygous embryos. For example, significant foetal loss was observed among couples matching for the entire HLA (human leukocyte antigen; the human equivalent to the MHC) haplotype in humans (Hedrick, 1988; Ober et al., 1998; Li et al., 2004). Likewise, Knapp et al. (1996) reported increased foetal loss in captive pig-tailed macaque pairs matching for MHC I.

The different selection pressures resulting from the requirements of offspring and maternal genotype to be similar or dissimilar at specific loci result in a variety of mechanisms by which compatibility can be achieved (e.g. Bainbridge, 2000). For example, in humans suppression of paternal HLA I in placental trophoblast cells occurs so that the cells sequestering the embryo from the mother are prevented from expressing paternal HLA antigens (e.g. Clark et al., 1987, 1999; Zeh & Zeh, 1996).

One of the best known mechanisms that can result in apparent similarity or dissimilarity between offspring and maternal genotype is genomic imprinting as it renders gene expression effectively hemizygous and parent-of-origin specific (Engel, 1997; Wolf & Hager, 2006). The principal property of genomic imprinting is that generally only one of the two alleles is expressed at a locus depending on their parent-of-origin, either the paternal or the maternal copy, although expression of the two parental alleles to different degrees has also been reported (Hayward et al., 1998; Wolf et al., 2008). Such uniparental gene expression is caused by different mechanisms, chiefly among which are differential methylation, histone modification or antisense RNA (e.g. Reik & Walter, 2001; Morison et al., 2005; Wood & Oakey, 2006). With only one allele being expressed, heterozygous offspring will either be similar or dissimilar to the maternal genotype. Thus, it seems a compelling argument for a role of genomic imprinting in genetic incompatibility can be posited (e.g. Lucifero et al., 2004).

Despite evidence for imprinting in the MHC (Kanbour-Shakir et al., 1990, 1993; Yuan et al., 1994; Kunz et al., 1996), which is closely involved in genetic compatibility (Roy Choudbury & Knapp, 2001; Penn, 2002), no study has formally explored whether genomic imprinting would indeed be selected for as a mechanism in the above scenarios and what possible testable predictions can be developed. Such an investigation will not only provide novel information about the possibly important role of epigenetics in human diseases associated with defects in the imprinting mechanism but also yields directions for empirical research into mammalian pregnancy and development research. Using a one-locus model of offspring and maternal genotype, we investigate whether genomic imprinting could be selected for to increase or reduce genetic similarity and generate testable predictions about the expression pattern at loci involved in genetic compatibility.

The model and results

We develop a simple model to examine whether systems of genetic incompatibility, where the maternal genotype selectively aborts offspring based on their genotype at the same locus, favours genomic imprinting. Two scenarios are considered: (i), selective abortion is increased by genetic similarity (incompatibility scenario) and (ii) selective abortion is decreased by genetic similarity (compatibility scenario).

We consider a locus, A, with two alleles A1 and A2 and frequencies p1 and p2 respectively. Mating is assumed to be random such that the frequencies of the genotypes in parents, their offspring and those of parent–offspring genotype combinations conform to Hardy–Weinberg expectations. Thus, genotype frequencies are approximately the same in parents and offspring, implying that selection is sufficiently weak such that changes in genotype frequencies within a generation caused by selection can be ignored. Because selection is based on maternal–offspring genotype combinations, rather than on individual genotypes, no major effect on genotype frequencies within a generation are expected. A derivation of this model assuming assortative mating shows that, although assortative mating affects mean fitness, it does not alter the conditions under which imprinting is favoured (assuming that the population is not 100% inbred such that there would be no heterozygotes).

In the first scenario, we assume that maternal genotypes have an increased probability of selectively aborting genotypes that are genetically similar to themselves. Such a system might evolve as a means of increasing heterozygosity in a population. In this incompatibility scenario, the probability of aborting is increased by sharing and reduced by nonsharing of alleles. This is measured as a change in the probability of survival (i.e. fitness), denoted s, where the probability of survival is reduced by the sharing (matching) of an allele by a value of −½s and is increased for nonshared (mismatched) alleles by a value of ½s (therefore s denotes the strength of selection). For example, homozygous offspring that have homozygous mothers of the same genotype have a fitness value of 1 – s because they necessarily share two alleles. Homozygous offspring of homozygous mothers that do not share alleles cannot occur under Mendelian inheritance, but could potentially be constructed experimentally and would have a fitness value of 1 + s (c.f. Table 1). Homozygous offspring of heterozygous mothers have a fitness value of 1 because they share one maternal allele but do not share the other.

Table 1.   Fitness values for the (a) incompatibility and (b) compatibility scenario.
 Offspring genotype
  1. A description of the parameters is given in the text. Note that several maternal–offspring genotype combinations cannot occur under Mendelian inheritance, but could be constructed experimentally. These combinations appear in parentheses.

(a) Maternal genotype
 A1A11 − s1 + is(1 − is)(1 + s)
A2A2(1 + s)(1 − is)1 + is1 − s
(b) Maternal genotype
 A1A11 + s1 − is(1 + is)(1 − s)
A2A2(1 − s)(1 + is)1 − is1 + s

We assume that the fitness of heterozygous offspring is altered by genomic imprinting and allelic expression in offspring is determined by the parameter i. Complete silencing of the maternally inherited allele (and expression of the paternal allele) occurs when = 1 and complete silencing of the paternally inherited allele (expression of the maternal allele) when = −1. The fitness of heterozygotes is determined by which allele they express, under the assumption that hemizygous expression of an allele results in a fitness value equal to that expected for homozygous expression of that same allele. For example, imprinted expression in heterozygous offspring of homozygous mothers results in a fitness value of 1 + is because offspring express the paternally inherited allele that is necessarily not shared by the mother. Heterozygous offspring that inherited an allele from their father matching their mother’s genotype cannot occur under Mendelian inheritance but, again, could be generated experimentally and would have the fitness of 1 − is. Heterozygous offspring of heterozygous mothers have a fitness value of 1, regardless of the pattern of imprinting, as each of their alleles is shared with one maternal allele and not shared with the other and so the allelic matching and mismatching cancel out. The fitness values for the incompatibility scenario are given in Table 1a. Note that, although hemizygous expression of a locus may not result in the exact same fitness as homozygous expression of the locus, the qualitative results are not affected by this assumption. The most likely violation of this assumption would be that there is a fitness cost to imprinting as a result of hemizygous expression (presumably due to the expression of deleterious recessive mutations). Such a cost imposes a straightforward limit to the evolution of imprinting, where the pattern of selection generated by selective abortion must outweigh any fitness cost associated with imprinting (see also Mochizuki et al., 1996).

In the second scenario, we assume that mothers have an increased probability of selectively aborting offspring genotypes that are genetically dissimilar to themselves. Such a system might occur in mammals as a result of maternal immune response, where offspring with a nonmatching genotype are more likely to be recognized as foreign and aborted either during implantation or placental development. We refer to this scenario as the compatibility scenario where the probability of aborting is decreased by the sharing of alleles and increased by nonsharing. The structure of this model is analogous to the incompatibility scenario, but with the pattern of fitness reversed. The fitness values for the matching scenario are given in Table 1b.

Population mean fitness (inline image) is calculated from the product of the frequencies of maternal–offspring genotype combinations and their expected fitness values (Table 1). In the incompatibility scenario, mean fitness is given by:


whereas in the compatibility scenario mean fitness is equal to:


Consequently, in the incompatibility scenario, the effect of imprinting on mean fitness, as given by the partial derivative of mean fitness with respect to the value of the imprinting parameter i (inline image), is


which demonstrates that this scenario favours paternal expression (positive value of i) in proportion to the amount of allelic variation (measured by p1p2) and the effect of selective abortion on the probability of survival (s).

Conversely, in the incompatibility scenario the effect of imprinting on mean fitness is:


demonstrating that maternal expression is favoured in this scenario (negative value of i) under the same conditions as in eqn (3).


We have presented two scenarios, one where selective abortion favours genotypic mismatching of mothers and their offspring (incompatibility scenario) and a second that favours matching (compatibility scenario). The compatibility model is built on the assumption that, for some loci, the maternal genotype may recognize offspring as a foreign tissue to the degree to which they mismatch the maternal genotype and thereby, mismatching results in an increased likelihood of spontaneous abortion by the mother (e.g. Clark et al., 1999; Richman & Naftolin, 2006). In this scenario, selection is predicted to favour genomic imprinting with maternal expression (i.e. silencing of the paternally inherited allele) because it hides the paternally inherited allele from the maternal genotype, and as a result, the ‘foreign’ paternally inherited allele is not detected and the offspring is protected from an increased probability of being aborted. Thus, imprinting can be viewed as a genomic ‘cloaking device’, which could also be adaptive in other scenarios (e.g. Roulin & Hager, 2003).

In contrast, the incompatibility model is based on the assumption that for some loci the maternal genotype may favour offspring that are genetically dissimilar to the mother as a means of increasing genetic variation at specific loci in the offspring (e.g. MHC; Knapp et al., 1996). Indeed, this scenario leads to an increase in the heterozygosity of offspring and maintains allelic variation in a population. In the case of genetic incompatibility, selection favours genomic imprinting with paternal expression because it enables the maternal genotype to identify the presence of a foreign allele in the offspring and, consequently, to reduce the probability of abortion of these offspring (which will have, on average, higher heterozygosity). Here, by hiding the maternal allele imprinting would act as a genomic cloaking device of the maternal allele, which may facilitate the identification and assessment of the paternal allele.

Our model makes clear predictions about the expected patterns of imprinting and expression that can be tested empirically. When the pattern of selective abortion favours matching of the offspring genotype to the maternal genotype, maternal expression of the locus is advantageous. Conversely, paternal expression is advantageous when allelic mismatching is favoured by the pattern of selective abortion. Although some maternal–offspring genotype combinations cannot exist under Mendelian inheritance (see Table 1), this does not affect the ability to test the predictions of the model. Because natural populations do not include these combinations, they play no role in the evolution of imprinting and, therefore, the naturally occurring maternal–offspring genotype combinations should provide the necessary evidence for the form of selection that leads to the evolution of imprinting.

The pattern of selective abortion assumed in the mismatching scenario is supported by data showing that spontaneous abortion may be associated with sharing of alleles at loci involved in histocompatibility (HLA, MHC; Hedrick, 1988; Knapp et al., 1996; Ober et al., 1998; Li et al., 2004). Although evidence for the presence of imprinting at any of the histocompatibility loci remains mixed and of limited taxonomic scope, the reported paternal expression of MHC class I antigens in the rat matches the pattern expected under the model (Kanbour-Shakir et al., 1990, 1993; Kunz et al., 1996). However, no imprinting of MHC has been detected in the mouse or horse (Donaldson et al., 1994; Drezen et al., 1994). It is important to bear in mind that, in most cases, it remains unknown whether the relevant loci show imprinted expression in the relevant tissues during the critical period in development when selective abortion occurs. Indeed, as imprinted gene expression can both be very flexible temporally and tissue specifical (Reik, 2007; Cheverud et al., 2008), relevant genes previously thought to show biallelic expression might still show imprinted expression during this critical period. We therefore emphasize that the predictions derived from our model apply only to the period during development when such selection/abortion would occur (e.g. during implantation or early placental development). For example, examination of placentas from later developmental stages (e.g. full term) may not provide insights into whether certain genes are imprinted during the developmental phase, which is presumably early in development, when selective abortion tends to occur, and thus existing data may not provide direct tests of our model predictions. Finally, our model suggests that, for loci where the pattern of imprinting has been selected for as discussed here, defects in the imprinting mechanisms may contribute to spontaneous abortion.


This work was supported by a grant from the BBSRC to JBW and a NERC Fellowship to RH.