1. Introduction
Genomic imprinting is the name of the form of non-Mendelian gene expression in which the two copies of a gene at a locus have different levels of expression. The archetypal case is that of insulin-like growth factor 2 (IGF-2) in which the maternal copy is silent in most fetal tissues, with only the paternally inherited copy being transcribed. Some 60 or so mammalian loci are currently known to be imprinted (Morison et al., 2001), but there is little consensus about the proportion of the genome subject to imprinting. By silencing (or at least downregulating) one copy of a gene, imprinting negates (or reduces) what is considered to be the major advantage of diploidy in mammals, namely, the ability to mask recessive deleterious mutations. Thus, the evolution of imprinting from an ancestral state of standard Mendelian (i.e., biallelic) expression appears paradoxical, apparently reducing an individual’s fitness. Several hypotheses seeking to resolve this paradox have been proposed by a number of authors. Here we examine a number of the more plausible ideas, highlighting their various strengths and weaknesses. In brief, however, no single hypothesis appears able to explain all the observations.
2. Prevention of parthenogenesis and the ovarian time-bomb hypothesis
The oldest ideas note that if different loci are oppositely inactivated (i.e., the maternal copy is transcribed at one locus and the paternal copy at another), imprinting at essential loci would require both paternal and maternal contributions to the developing zygote, thereby preventing parthenogenesis. Indeed, recent experiments with mice show that if appropriate expression at the normally imprinted H19 and Igf2 loci occurs, at least some parthenogenetic embryos can survive to adulthood and reproduce (Kono etal., 2004). The observed absence of any parthenogenetic mammalian species consequently led some authors to argue that imprinting may have evolved for this purpose. Parthenogenesis is considered to be disadvantageous because it stops the genetic recombination that occurs as a consequence of sexual reproduction. Parthenogenetic lineages are inclined to be evolutionary dead ends, lacking the ability to respond to novel selection pressures.
The main problem with this argument, however, is that it is “group-selectionist”: the purported advantage of avoiding parthenogenesis accrues to the species, not to an individual. Indeed, an individual with parthenogenetic abilities might have a selective advantage, able to reproduce even in the absence of suitable mates. This selection for parthenogenesis at the level of the individual would subvert selection against it at the level of the species. In most cases of conflict between selection pressures in opposite directions at different levels, selection at the lower level – for instance, individuals rather than species – prevails. Thus, prevention of parthenogenesis for the good of the species is not considered a likely cause of the evolution of imprinting in mammals.
Nevertheless, an individual-level advantage for avoiding parthenogenesis has been suggested by Varmuza and Mann (1994). They argued that a haploid egg spontaneously developing in an ovary would amount to ovarian cancer, and imprinting may have evolved to prevent such a scenario. Inactivating the maternal copy of a growth-enhancing gene would defuse this “ovarian time bomb”. Iwasa (1998) pointed out that the same protection is afforded by upregulating the maternal copy of a growth-inhibiting gene. Moreover, the level of expression in the diploid developing zygote can be maintained by concomitantly downregulating the paternal copy of this same gene, possibly to the point of silencing it. Thus, the ovarian time-bomb hypothesis predicts that growth-affecting genes active in the early stages of embryogenesis are likely candidates for imprinting and that growth enhancers should be maternally inactivated, whereas growth inhibitors should be paternally silenced. And indeed, this prediction is often met: the growth-enhancing Igf2 is maternally inactivated in fetal tissues in all mammalian species examined so far, and the growth-inhibiting Igf2r is paternally inactivated in mice and rats (but not in humans).
Mathematical modeling of the ovarian time-bomb hypothesis (Weisstein etal., 2002) implies that the verbal hypothesis is plausible: the selection pressures envisaged could lead to the evolution of imprinting. The modeling shows also that the selection pressure required to lead to the evolution of imprinting need not be very strong, contradicting the objection that ovarian cancer was too rare to be worth the cost of the loss of functional diploidy. Nevertheless, it is less clear why so many loci should be imprinted: surely, the imprinting of one or two critical loci would provide sufficient protection. The flip side of this problem is that the hypothesis at least offers a weak explanation for why not all growth-affecting genes important in early development (e.g., IGF1) are not imprinted. Finally, the hypothesis offers no explanation for genes that are not involved in early development.
3. The genetic-conflict hypothesis
Perhaps the best-known explanation for the evolution of imprinting is that invoking the different, conflicting genetic interests of mothers, fathers, and their offspring (Haig and Graham, 1991; Haig, 1992). A mammalian mother is equally related to all the offspring in a single (and subsequent) pregnancies, so her genetic contribution to the next generation is maximized by ensuring the survival of as many of these offspring as possible. To at least the first level of approximation, these genetic interests are best served by equally dividing the nutrients and care she provides among these offspring. One way to accomplish this goal would be to turn off any growth-enhancing genes in her offspring, so she can control the transfer of her various resources. A father’s genetic perspective is quite different, however, because most mammals have some degree of multiple paternity. A mammalian father has no assurance that all the offspring born to a female with which he has mated will be his. Consequently, his genetic success is greater if somehow his offspring obtain more maternal resources, maybe at the expense of any half-sibs or even the mother herself. Inactivating the paternal copy of a growth-inhibiting gene would achieve that end.
Thus, the genetic-conflict hypothesis makes the same predictions as the ovarian time-bomb hypothesis about the sorts of loci likely to be imprinted (i.e., growth-affecting genes important in fetal development) and the direction of imprinting (i.e., maternal inactivation of growth enhancers and paternal silencing of growth inhibitors). Mathematical modeling by various groups confirms the basic plausibility of the hypothesis (Spencer etal., 1999). Some modeling (Spencer et al., 1998) predicts that under certain circumstances, a locus can be polymorphic in imprinting status, with some individuals having two active copies of the genes and others just one. Indeed, two loci, the Wilm’s tumor suppressor, WT1 (Jinno etal., 1994), and the serotonin-2A (5-HT2a) receptor, HTR2A (Bunzel etal., 1998), appear to fulfill this prediction. Importantly, this prediction differs from that derived from modeling of the ovarian time-bomb hypothesis, so these observations lend important support to genetic conflict over the ovarian time bomb. The genetic-conflict hypothesis can also apply in arenas other than fetal development, for example in postnatal care, and so the range of loci likely to be imprinted by this mechanism is greater than under the ovarian time bomb. This prediction appears largely, but not completely, fulfilled (Tycko and Morison, 2002). Nevertheless, the genetic-conflict hypothesis is less able to explain why other growth-affecting genes such as Igf1 are not imprinted.
4. Differential selection on males and females
We have only a limited number of observations on imprinting at sex-linked loci. Indeed, the best known simply infer the presence of imprinting from observations of chromosomal abnormalities, especially X-chromosome monosomy in mice (see Article 46, UPD in human and mouse and role in identification of imprinted loci, Volume 1). The parental origin of the single X in these XO mice has significant growth effects: if it is paternally derived, the mice are developmentally retarded (Jamieson et al., 1998), implying the presence of a paternally inactivated growth enhancer. Thus, the direction of imprinting at X-linked loci appears to be opposite to that predicted by both the genetic-conflict and ovarian time-bomb hypotheses. Observations like these led Iwasa and Pomiankowski (1999) to propose that selection for different phenotypes in males and females – especially different sizes – could lead to imprinting.
These authors noted that changing the expression level of a paternally derived gene on the X chromosome would affect only female offspring. In contrast, alteration of expression at a maternally derived locus subject to dosage compensation (see Article 15, Human X chromosome inactivation, Volume 1) would have greater effects on male offspring. Hence, greater male size, common in mammalian species, could be achieved by maternally inactivating an X-linked growth inhibitor and/or paternally silencing a growth enhancer. These predictions are the opposite of those made by the genetic-conflict hypothesis (Spencer et al., 2004). Moreover, the sorts of loci that might be subject to imprinting under Iwasa and Pomiankowski’s hypothesis is considerably greater: any loci affecting characters for which optimum male and female phenotypes differ could be imprinted. The paucity of clear examples of imprinted X-linked genes, therefore, could be seen as evidence against this suggestion.
Spencer et al. (2004) argued that the above ideas could be extended to autosomal loci underlying characters for which being more similar to one parent than the other is advantageous. For example, given that male mammals usually disperse further than females, genes important in local adaptation in a heterogeneous habitat might be preferentially expressed from the better-adapted maternal copies and hence subject to paternal inactivation. There are, however, no current examples of imprinting that support these latest ideas.
In summary, several hypotheses have been proposed to explain the paradox of imprinting – the apparently disadvantageous functional haploidy at imprinted loci – but not one explains all the observations. Some suggestions not discussed here (e.g., better control of gene expression) have little support, either empirical or theoretical, but three hypotheses – ovarian time bomb, genetic conflict, and differential selection on males and females – appear to do far better in plausibly explaining many known observations.
Related articles
Article 15, Human X chromosome inactivation, Volume 1; Article 26, Imprinting and epigenetic inheritance in human disease, Volume 1; Article 28, Imprinting and epigenetics in mouse models and embryogenesis: understanding the requirement for both parental genomes, Volume 1; Article 29, Imprinting in Prader-Willi and Angelman syndromes, Volume 1; Article 30, Beckwith-Wiedemann syndrome, Volume 1; Article 31, Imprinting at the GNAS locus and endocrine disease, Volume 1; Article 32, DNA methylation in epige-netics, development, and imprinting, Volume 1; Article 33, Epigenetic reprogramming in germ cells and preimplantation embryos, Volume 1; Article 36, Variable expressivity and epigenetics, Volume 1; Article 38, Rapidly evolving imprinted loci, Volume 1; Article 39, Imprinting and behavior, Volume 1; Article 41, Initiation of X-chromosome inactivation, Volume 1; Article 45, Bioin-formatics and the identification of imprinted genes in mammals, Volume 1; Article 46, UPD in human and mouse and role in identification of imprinted loci, Volume 1
