Mammalian Growth and Parent-of-Origin Effects

Causes of Parent-of-Origin Effects

A number of mechanisms can cause parent-of-origin effects. It may be difficult to distinguish between these effects without detailed pedigree or molecular analyses (Fig. 2). One potential underlying cause of parent-of-origin effects is variation in mitochondrial DNA (mtDNA). The vast majority of zygotic mtDNA is derived from the oocyte (Hutchison et al. 1974). Thus, any phenotypes caused by mtDNA variation will show a matrilineal inheritance pattern; males will not pass the trait.

Another class of genes, termed “maternal effect” genes, are those in which offspring phenotype is dependent on maternal genotype. Mature oocytes, in contrast to sperm, contain many gene products critical for development. Because of the haploid nature of germ cells, alleles that would otherwise be recessive may exhibit phenotypes when they are expressed in oocytes. Thus, a female could be either heterozygous or homozygous for the allele for her offspring to show the effect. The offspring genotype at the gene in question, heterozygous or homozygous, does not affect the phenotype in this case.

Maternal effect loci in mammals may even produce phenotypes in offspring that have no copies of the causative allele. Genes that affect uterine environment, lactation, or postnatal care all have this potential. That is, allelic variation in these genes could affect the offspring without being passed on to them (e.g., if the allele is dominant and the mother is heterozygous; Fig. 2B).

Sex chromosome–linked genes may also underlie parent-of-origin effects. Most mammalian species have a Y chromosome found only in males. Although there are some known exceptions (e.g., the XY females of the transcaucasian mole vole [Ellobius lutescens]), most Y-linked genes will affect only males and be passed only to other males (Just et al. 2002). The parental effects on mammalian X chromosomes are more subtle. The best known are those in which recessive X-linked phenotypes are disproportionately seen in males. This sex disparity in X chromosome phenotypes seems obvious given the single copy in males versus 2 in females. However, females silence 1 of their 2 X chromosomes such that gene expression is roughly equivalent between the sexes (Lyon 1961).

Female mammals are typically thought to randomly inactivate 1 X chromosome. For some time, however, it has been known that Mus undergo preferential inactivation of the paternal X in the placenta and other extraembryonic tissues (Takagi and Sasaki 1975). Paternal X inactivation also has been observed in the placentas of deer mice (Peromyscus maniculatus; Vrana et al. 2000) and domestic cows (Bos taurus; Xue et al. 2002). Female marsupials of several genera have been shown to inactivate the paternal X chromosome in both extraembryonic and somatic tissues (Graves 1987). Examination of metatherian data suggests that marking of the paternal chromosome, rather than random choice, was the ancestral therian mechanism of X inactivation.

This hypothesis has been strengthened by recent developmental studies in Mus suggesting that the paternal X is initially marked for silencing in embryonic tissues as well (Sado and Ferguson-Smith 2005). Early in development, however, the paternal X is “reset” in cells destined to form the embryo itself, and the inactivation then becomes random. That is, the paternal X is inactivated in 50% of cells and the maternal X is silenced in the other 50%. This mosaicism of X inactivation is thought to protect against deleterious recessive mutations: a female that is heterozygous for such a mutation will still have a proportion of cells that have a functional copy of the gene.

Once one of the X chromosomes is inactivated, it stays silent in daughter cells produced by cellular division. The continuing silence of the same X allele in descendant cells is an example of epigenetics. Epigenetic effects are those in which heritable changes in gene activity are produced without a change in DNA sequence. Although epigenetic effects are typically thought of as occurring within an organism, they also may involve transmission to offspring. Such Lamarckian effects have been demonstrated in plants (Adams and Wendel 2005; Takeda and Paszkowski 2005).

Genomic imprinting is an epigenetic phenomenon similar to X inactivation, but occurs on autosomes. Genomic imprinting may be the most common cause of parent-of-origin effects. So-called imprinted genes are expressed from only 1 parental allele (Tilghman 1999). That is, there are genes that are only expressed from the maternally inherited copy, and those that are only expressed from the paternally inherited copy (Fig. 3A).

A deleterious or maladaptive mutation in an imprinted gene will be exposed if it is on the expressed parental chromosome, despite the presence of a wild-type copy. Imprinted genes are reset every generation, such that a paternal allele passed from father to daughter becomes a maternal allele in her offspring (Fig. 3B). Thus, the hallmark of imprinted inheritance is dependence of the phenotype on which parent contributed the allele, rather than which sex inherits it. The Igf2 mutations noted only cause the growth retardation phenotypes in young whose father passed the mutation; this is because the Igf2 gene is exclusively expressed from the paternal allele. Growth-retarded females cannot produce growth-retarded offspring if mated to a wild-type male. In contrast, grand-offspring produced from sons of this female may exhibit the growth phenotype (Fig. 2A).

The gene Igf2 was the 1st gene reported to be imprinted (DeChiara et al. 1990), because of the unexpected inheritance pattern of the mutant phenotype. Almost simultaneously, a report came that the Igf2 receptor (Igf2r) was expressed only from the maternal allele (Barlow et al. 1991). What is intriguing is that the Igf2r protein actually acts as a sink for Igf2 by targeting it for degradation. Igf2r has the curious property of being expressed from the opposite parental allele as that of Igf2, and its product acting antagonistically to the Igf2 gene product. The 3rd imprinted gene discovered also proved to be maternally expressed, and also to be involved in Igf2 function, albeit indirectly (Bartolomei et al. 1991). This gene, termed H19, does not encode a protein; rather only an RNA is produced (Brannan et al. 1990). Whether this RNA itself is functional is open to debate; however, the DNA sequence itself is well conserved. Other imprinted genes were discovered slowly through the 1990s both by serendipity and designed screens. Sequencing of the Mus genome allowed more rapid discovery through identification and testing of genes located near known imprinted loci. Today the number of imprinted transcripts stands at greater than 80 (Morison et al. 2005), although an exact count is difficult.

Regardless of the total, a number of generalizations can be made regarding imprinted genes:

One annoying facet of the genomic imprinting literature has been the inconsistency of terms used to describe a gene's parental expression pattern. Generally, the silenced allele is the one referred to as “imprinted.” An ongoing dilemma is whether to refer to a gene by its silenced (imprinted) or active (expressed) allele. For example, Igf2 may be termed a “paternally expressed gene” or a “maternally imprinted gene.” Proponents of the former point out that it is the silencing that is atypical and hence being studied. For those interested in allelic variation or population biology, however, it is more logical to refer to the active allele.

Mammalian genes subject to genomic imprinting encode products that fall into many families—secreted ligands, receptors, transcription factors, splicing factors, intracellular signaling molecules, diverse enzyme groups, and noncoding RNAs. Despite this diverse array of gene products, involvement in both pre- and postnatal growth, metabolism, or behavior emerge as common themes (Reik et al. 2003). Strengthening the case made by transgenic mouse models, a number of human diseases have been identified in which altered expression of imprinted genes is the primary defect. Many of these imprinting diseases are characterized by abnormal growth, including Beckwith–Wiedemann syndrome, Silver–Russell syndrome, and Prader–Willi syndrome. Prader–Willi syndrome is notable in that afflicted individuals are obese because of abnormalities of the central nervous system that lead to overeating (Cassidy 1984). Recent evidence also has strengthened the case for imprinted genes playing a role in 2 common maternal–fetal conditions, pre-eclampsia and gestational diabetes (Oudejans et al. 2004).

Some imprinted genes appear to directly function in both growth and behavior; an interesting example is that of the paternally expressed gene Pw1/Peg3. Pw1/Peg3 was 1st identified in a screen for genes involved in muscle cell determination and later found to be imprinted (Relaix et al. 1998). The Pw1/Peg3 gene encodes a large (∼1,500 amino acid) protein with 12 zinc finger domains, and is expressed in numerous cell types, particularly in central nervous system and myogenic lineages. Mice that inherit an induced Pw1/Peg3 mutation from their father are growth retarded, similar to those produced by the Igf2 mutants (Li et al. 1999). These young also suckle poorly and show a later onset of puberty. Female young that inherit the mutation are fertile, but do not eat sufficiently during pregnancy and exhibit poor lactation and reduced maternal care (Curley et al. 2004). The young of mutant females also show growth retardation, even if they received a functional allele from their father, showing that Pw1/Peg3 also has a maternal effect phenotype (e.g., Fig. 2B).

Molecular Mechanisms of Genomic Imprinting

Although imprinted domains tend to be relatively discreet, it is not entirely clear what defines them. All imprinted chromosomal regions examined contain a relatively small region of DNA where one parent's allele exhibits greater amounts of DNA methylation than that derived from the other parent. DNA methylation in mammals is primarily found on cytosine residues in the sequence CG (Bird 1986). Regions of the genome rich in DNA methylation tend to have little or no gene transcription activity. The enzymes that add these methyl groups, the DNA methyltransferases (Dnmts), are active in the germlines of both sexes (Mertineit et al. 1998). Indeed, mature sperm and oocytes exhibit very different patterns of DNA methylation. These gametic DNA methylation marks are typically thought to be the actual imprint or mark that specifies 1 allele be silenced.

The DNA methylation levels are not the only marks distinguishing the expressed and silenced alleles of imprinted genes. Methylation and other modifications of the histone proteins that act as chromosome scaffolds also are found in imprinted and other silenced regions. For example, a methylated version of histone 3 is associated with inactive genes. Reciprocally, an acetylated form of histone 3 is typically associated with active regions of chromatin, whereas removal of those acetyl groups signals silencing (Jones et al. 1998). Collectively, the state of modifications of these proteins has been termed the “histone code.” The silenced allele of an imprinted gene thus has a different code than the active allele (Hu et al. 2000).

Although the histone code clearly plays a role in genomic imprinting, the general thought is that it cannot be the primary mark or imprint: during mammalian spermatogenesis, histones are replaced by a class of structural proteins called protamines (Govin et al. 2004). Protamines are more efficient at packing DNA into the sperm nucleus. A small number of protamines may persist for a time after fertilization, and thus could play a role in conveying paternal marks (Meistrich et al. 2003).

The emerging picture indicates that the histone, DNA, and possibly other chromatin modifications act in a combinatorial fashion to insure that a gene is either silenced or activated. For example, the Dnmt enzymes recruit histone deacetylases—both these states (methylated DNA and histones lacking acetyl groups) are associated with silenced genes. These modifications are similar to those seen in silent regions of the genome (heterochromatin) and the silenced X chromosome.

Multiple modifications are likely necessary to maintain silencing of a region. For example, evidence to date suggests that metatherian mammals have overall lower levels of DNA methylation than eutherian mammals (Loebel and Johnston 1996). Correspondingly, the silenced paternal X chromosome still shows some expression in female marsupials (Graves 1987). Similarly, eutherian placentas exhibit lower levels of DNA methylation than the embryo proper as well as a greater tendency toward abnormal expression of imprinted genes. For example, a subset of imprinted genes located near the Lit1 gene on Mus chromosome 7 appear to use only a trimethylated form of histone 3 to maintain silencing. That is, the germline DNA methylation in this region has been lost. Strikingly, all these genes are primarily or exclusively expressed in the placenta (Lewis et al. 2004).

When imprinted gene transcription is perturbed such that there is a switch from monoallelic to biallelic expression, loss of imprinting is said to have occurred. Loss of imprinting can be induced by a number of factors, including genetic variation as well as induced mutations. For example, deletion of any one of a number of the Dnmt genes, whose products regulate DNA methylation, induces loss of imprinting. Many of these induced deletions of the Dnmts have maternal effects on imprinting (Bourchis et al. 2001; Howell et al. 2001; Kaneda et al. 2004). That is, females that initially inherit the mutation are normal, but their offspring exhibit loss of imprinting when mated to a wild-type male. The Dnmt maternal effects are specific to the oocyte, as demonstrated by embryo transfer experiments in which Dnmt mutant fertilized oocytes are moved to the uterus of a wild-type female, and vice versa.

As suggested by these data, Dnmt gene products have been found in mature oocytes before fertilization. Gene products in the DNA methyltransferase family have several functions. Maintenance Dnmts methylate newly synthesized partner strands of already methylated regions (e.g., Dnmt1—Li et al. 1993), de novo methyltransferases add methylated cytosines to a region based on other cues (e.g., Dnmt3a and Dnmt 3b—Bestor 2000; Bestor and Verdine 1994), and Dnmt3L lacks the ability to actually methylate DNA, but appears to interact with other Dnmts to target specific regions of the genome, particularly imprinted domains (Aapola et al. 2000). Dnmt3L also appears to directly activate histone modifying enzymes (Deplus et al. 2002).

The differentially methylated regions (DMRs) produced by Dnmt activity are thought to be key in regulating the entire cluster of imprinted genes in which they lie. One type of DMR (primary) originates in the germline of a parent and may mark an allele from the beginning of development. Other DMRs (secondary) are not present initially, but arise during development. Deletion or mutation of primary DMRs typically result in loss of imprinting of most genes within the associated cluster (Verona et al. 2003). In such cases, these DMRs also are referred to as imprint control regions or elements. Surprisingly, the actual DNA sequence at these regions is often very divergent between mammalian species. One common feature is that most primary DMRs have repeated sequences that are not found elsewhere in the genome. Interestingly, plants have a mechanism that senses repeated sequences and silences them via DNA methylation; this system is thought to be a viral defense mechanism. Whether the same repeat-induced silencing mechanism also exists in mammals is unclear. However, forceful arguments have been made that mammalian DNA methylation was originally selected for as a retroviral defense (McDonald et al. 2005).

Surprisingly little is known about mechanisms of DNA demethylation in mammals. Profiles of DNA methylation levels in house mice show a rapid drop during preimplantation development. The demethylation of paternal DNA appears to be an active process, whereas maternal DNA methylation is lost through lack of maintenance (i.e., newly synthesized strands are not methylated—Oswald et al. 2000). Most cytosine methylation across the genome is lost; however, a small amount at primary DMRs survives. The genome then begins a remethylation process, with germ cells, somatic tissues, and extraembryonic tissues differing in final density of methylated cytosines (Santos et al. 2002).

A recent study of domestic sheep challenges the assumption that DNA methylation dynamics are similar across mammals (Beaujean et al. 2004). Fertilized sheep oocytes do not actively demethylate the paternal genome. A sheep paternal pronucleus transplanted to a mouse oocyte is demethylated, suggesting no intrinsic protection against demethylation. The reciprocal experiment yielded more surprising results: sheep oocytes were able to demethylate transplanted mouse paternal pronuclei, despite being unable to demethylate conspecific male DNA.

The differences between genes, cluster regulation, species, and nomenclature underscore confusion in the field: no one model of imprinting control or gene organization fits all the known clusters of imprinted genes, let alone those yet to be discovered. Further, most of the data are from either laboratory Mus or humans. The former do not represent natural populations, the latter are not tractable for many experiments.

Imprinted domains usually contain at least 1 of each type of DMR, typically 1 primary and multiple secondary DMRs. The relationship of the imprinted genes to these DMRs and to each other varies greatly (Fig. 4). The heavily methylated regions overlie the areas where gene transcription begins in some cases, such as for Pw1/Peg3 and neighboring Usp29 gene (Fig. 4A; Li et al. 2000). In many cases noncoding RNAs (ncRNAs) are found in imprinted clusters. In particular, the silenced X chromosome specifically expresses a large ncRNA termed Xist (inactive X chromosome–specific transcript). The Xist locus is found in the region genetically known to mediate this process, the X-inactivation center (Brockdorff et al. 1991). The Xist RNA spreads from the X-inactivation center to coat the entire inactivated chromosome (Clemson et al. 1996). Whether the spreading Xist RNA directly or indirectly (i.e., through recruiting other factors) silences transcription is not understood.

Noncoding RNAs may play a direct role in autosomal imprinted gene regulation in certain clusters. The enormous (50,000+ bases) Lit1 RNA is the only paternally expressed transcript in its cluster of imprinted genes (Mitsuya et al. 1999). The spreading Lit1 RNA is speculated to interfere with transcription of other genes on the paternal allele (Fig. 4B; Mancini-Dinardo et al. 2006). Lit1 is an example of a so-called antisense transcript, as are a number of ncRNAs. That is, these transcripts arise from the noncoding strand of DNA directly opposite a strand encoding a gene. The Lit1 transcript begins in an intron of the Kcnq1 potassium channel–encoding gene and crosses several others. Such overlapping transcripts render both counts of imprinted genes and elucidating the role of ncRNAs in regulation difficult.

The gene pair for which imprinted gene regulation has been best characterized is the Igf2–H19 pair. The primary DMR lies between the 2 genes in this pair (Fig. 4C; Tremblay et al. 1995). The unmethylated maternally derived DNA sequence at the Igf2–H19 primary DMR specifically binds the CTCF (CCCTC-binding factor) transcription factor (Hark et al. 2000). The bound CTCF protein on the maternal allele is thought to prevent the Igf2 gene from interacting with a set of transcriptional-enhancing elements that lie on the other side (i.e., away from Igf2) of H19 (Fig. 4C). In contrast, the CTCF protein is unable to bind to the methylated paternal allele. The absence of CTCF binding allows the Igf2 gene to access the enhancer elements on the paternal allele. The paternal DNA methylation at the CTCF binding region extends into the H19 gene, insuring its silence on this allele.

The Igf2 secondary DMRs are involved in formation of DNA loops that facilitate the interaction with the enhancer elements (Murrell et al. 2004). On the paternal allele, the primary DMR associates with the secondary DMR farthest from the start site of Igf2 transcription. This arrangement brings the enhancer elements in close proximity to the start site of Igf2. On the maternal allele, the primary DMR interacts with a secondary DMR very close to the Igf2 start site; this arrangement isolates the H19 gene and enhancer elements.

At least 1 other pair of imprinted genes resembles the Igf2–H19 configuration, the Dlk1–Gtl2 pair: Dlk1 is a protein coding gene involved in growth pathways, whereas Gtl2 produces ncRNA transcripts of various lengths (Schmidt et al. 2000). A primary DMR lies between the 2 genes. A mutation in this regulatory domain, rather than the genes themselves, appears to be responsible for the hypertrophic buttock musculature of the domestic sheep callipyge phenotype (Charlier et al. 2001). However, the emerging picture is that imprinted domains and local methods of regulation are quite heterogeneous. Knockouts of several potential genome-wide imprinting regulation genes have had effects only on a subset of imprinted genes and clusters. Even responses to widespread changes in histone modification and DNA methylation are not uniform among imprinted genes. Although there are commonalities among these regions, the variation suggests that different genes and clusters have become imprinted at different points in history.

Phylogenetic Distribution of Genomic Imprinting in Mammals

The similar mechanistic and phylogenetic distribution between autosomal genomic imprinting and X chromosome inactivation has strengthened the hypothesis that the 2 phenomena share an evolutionary origin in mammals. The lack of evidence for either autosomal or X imprinting in monotremes suggests that both processes evolved after the divergence of this lineage from therian mammals (Fig. 5). Imprinting variation among eutherian mammals suggests that there have been multiple steps in the evolution of this phenomenon, and that it is ongoing.

Comparative genomic studies suggest that the clusters of genes found to be imprinted in Mus are generally well conserved across eutherians (complete mammalians genomes, available at  http://www.ensembl.org). However, general gene content conservation does not prove identical imprinting status of those genes or conservation of the boundaries of the domain. Directly assessing the imprinting status of a gene requires the ability to distinguish the 2 parental alleles (Vrana 2006). One must identify heterozygous individuals as well as ascertaining 1 or preferably both parental genotypes. Genera of mammals for which genomic imprinting has been directly demonstrated are listed in Table 1.

The presence of DMRs has been used as a proxy for imprinting status: if a gene has a DMR present, then it is thought to be imprinted. Typically the differential allelic DNA methylation patterns are present even in nonexpressing tissues. A counterexample to this inference is the Igf2r gene. The gene is not imprinted in humans despite the presence of a DMR (Kalscheuer et al. 1993; Riesewijk et al. 1996; Smrzka et al. 1995). Igf2r also is not imprinted in prosimian ring-tailed lemurs (Lemur catta), common tree shrews (Tupaia glis; order Scandentia), and Philippine flying lemurs (Cynocephalus volans; order Dermoptera—Killian et al. 2001). Igf2r imprinting has been demonstrated in the metatherian genus Didelphis, the rodent genera Peromyscus and Rattus, as well as in 3 artiodactyl species (domestic sheep, pigs, and cows). The interpretation of these data is that Igf2r imprinting was present in the common ancestor of therian mammals, but has been lost in Primates, Dermoptera, and Scandentia. These 3 orders have been grouped together in the superorder Euarchonta, suggesting a single loss of imprinting status in this gene (Fig. 5).

Interspecific discrepancies in imprinting status have now been observed in a number of other genes (approximately 32 between humans and house mice as of this writing [Appendix I]). For example, Vrana et al. (2001) discovered a placental lactogen gene in Peromyscus showing paternal expression. Muroid placental lactogen genes are duplications of the prolactin gene, encoding the well-known pregnancy hormone. In contrast, primates have no duplications of prolactin; the analogous loci termed placental lactogens in Primates are instead duplications of the growth hormone gene (Soares 2004). Human and house mouse orthologs are reciprocally imprinted (i.e., expressed from the opposite parental allele) in at least 2 cases (Morison et al. 2005).

Humans are the only species that has been examined for intraspecific imprinting variation. One study revealed variation in the degree of parental bias in imprinted multiple loci within the relatively homogenous Japanese population (Sakatani et al. 2001). A number of inherited diseases also show significant phenotypic differences dependent on which parent contributed the allele.

Recently, a class of human hydatidiform moles was uncovered that has a normal 1:1 ratio of parental genomes rather than an excess of paternal DNA (Judson et al. 2002). An allele that apparently predisposes women to these placental overgrowths has been mapped to the imprinted domain containing the Pw1/Peg3 gene. These biparental hydatidiform moles are associated with changes in DNA methylation at imprinted domains. An emerging view is that the majority of cases in which this syndrome recurs are due to such susceptibility alleles; it will be particularly interesting to determine if these alleles are deleterious in all genetic backgrounds or environments. The changes in DNA methylation at DMRs suggest that changes in imprinted gene expression are directly responsible for the hydatidiform mole phenotypes, and reinforce the notion that placental tissues are particularly sensitive to these changes.

Rationale for Genomic Imprinting

Understanding the biological role of genomic imprinting is particularly interesting in that it is a counterintuitive process: Why give up the advantages of diploidy at a locus by silencing 1 allele and potentially exposing deleterious recessive mutations? What selective pressures have given rise to this seemingly bizarre form of gene transcriptional regulation? Are these the same forces that maintain the process?

Chromatin modifications such as DNA methylation and histone modifications that underlie monoallelic expression already were present before the origin of mammalian imprinting (Reik and Lewis 2005). The raw material for genomic imprinting must be provided by differences in such modifications that arise during oogenesis and spermatogenesis. Imprinting selection hypotheses must explain why only certain genes retain evidence of these gametic differences.

One clear consequence of imprinting is that it precludes parthenogenesis; this was 1st noted when the reciprocal pronuclear transplantations were performed. This hypothesis assumes that new allelic combinations generated by sexual reproduction are so important that the processes cannot be obviated. The antiparthenogenesis hypothesis predicts that imprinting will not be found in groups where parthenogenesis exists. However, this prediction has not yet been tested, although the opportunity exits within amniotes (e.g., a proportion of wild turkeys [Meleagris gallopavo] are parthenogenetic, as are several populations of the lizard genus Cnemidophorus—Harada and Buss 1981; Kearney and Shine 2004; Olsen and Marsden 1954).

Genomic imprinting also could function as a means of preventing inappropriate development of oocytes still in the ovary (Varmuza and Mann 1994). Varmuza and Mann (1994) noted that invasive trophoblast tissue was one of the major tissues impaired in parthenogenetic embryos, thus rendering most such growths benign. In contrast, risk of metastasizing tumorigenic trophoblast is much greater in hydatidiform mole pregnancies. This “ovarian time bomb” hypothesis described a consequence of imprinting, but does not explain maternally expressed genes, nor predict that they should also be involved in the same growth pathways as paternally expressed genes. The hypothesis does predict that there should be no imprinting outside of eutherian mammals, that is, before the innovation of the trophoblast lineage. The finding that Igf2 (Killian et al. 2000, 2001; O'Neill et al. 2000) is imprinted in didelphid marsupials and the small macropodid Macropus eugenii (tammar wallaby) argues strongly against this hypothesis.

The most compelling and controversial hypothesis to explain the existence of genomic imprinting is based on kin selection (Hamilton 1964). Trivers (1974) noted that parents and their offspring might have differing genetic interests. Because only 50% of an embryo's alleles are necessarily shared with its mother, nonshared alleles may be selected to extract more maternal resources than would be advantageous for the mother's other current or future offspring.

Haig (1996) particularly explored potential “parental antagonism” in mammalian pregnancies, positing the placenta as a major “battleground.” This kinship–parental antagonism hypothesis (Haig 1996; Moore and Haig 1991) proposes that in any nonmonogamous, outbred population, parental interests will differ: males benefit most when their offspring extract the greatest amount of maternal resources possible during gestation and postnatal care. This selection on paternally transmitted alleles is particularly strong if a female's other offspring are likely to be sired by other males that are not closely related to the 1st male.

This antagonism can be depicted as being realized through different parental optima for expression levels of a given gene. For example, paternal interests may benefit from higher expression levels of a prenatal growth factor such as Igf2. A small increase in Igf2 levels would result in greater birth weight of the offspring, and hence a greater rate of survival. These larger neonates then impose a greater cost on the mother. Total expression level of such a gene is predicted to gravitate toward the higher of 2 parental optima: once 1 parent's allele has been completely silenced, it can no longer affect overall levels of gene expression. Wilkins and Haig (2003) termed this phenomenon the “loudest voice prevails” effect.

The other parent may counter such a strategy by modulating expression levels of a gene with antagonistic effects. The maternal expression of Igf2r, which lowers Igf2 protein levels, is an example of this effect. A striking example of reciprocally imprinted antagonistic genes lies within the Gnas locus. The Gnas gene complex encodes “G protein” subunits, which are key components of intracellular signaling from hormone receptors (Peters et al. 1999). Deletion of a shared exon between 2 variants of the protein results in reciprocal metabolic phenotypes depending on which parent contributes the mutant allele; inheriting a maternal deletion results in obese offspring, whereas inheriting a paternal deletion yields very lean offspring (Chen et al. 2005).

The kin selection hypothesis predicts that imprinted genes should affect growth control and behavior; different genes may acquire imprinting at different times, resulting in somewhat heterogeneous mechanisms; no selection for imprinting should occur in oviparous taxa that also lack parental care; and selection should be reduced in monogamous or highly inbred taxa. Many, but not all, imprinted genes have met a simple interpretation of the hypothesis: that paternally expressed genes should promote growth (maternal allele silent, as per the “loudest voice prevails” effect), and maternally expressed genes should inhibit growth (Haig 2004).

Zeh and Zeh (2000) proposed that parental antagonism plays a major role in differing speciation patterns across taxa. They noted that mammals appear to have much more rapid reproductive isolation than nonviviparous vertebrate groups. The role of viviparity in rapid reproductive isolation is supported by the great diversity in eutherian placentation, including morphology, degree of maternal invasion, and gene expression (Carter and Enders 2004). Placental abnormalities are common in both interspecific hybrids and embryo transfer experiments (Allen et al. 1993). The numerous associations between altered imprinted gene expression and placental phenotypes suggest a role for imprinted genes in mammalian reproductive isolation.

Further study of monotremes and marsupials should help clarify the role of selection on mammalian imprinting. Both groups would appear to be subject to parental antagonism because of lactation. However, monotremes are oviparous. If the viviparity hypothesis is correct, genomic imprinting should play a more significant role in marsupials than in monotremes. As noted, studies have confirmed both autosomal and X chromosome imprinting in the former, whereas there is no evidence for genomic imprinting in monotremes. The number of imprinted genes or severity of parental bias in expression also should be less in metatherian mammals than that observed in eutherian mammals. Relatively weaker selection for imprinting of genes during prenatal development is predicted in marsupials because of their typically shorter gestation times, simpler and more transitory placenta, and the greater amount of yolk in metatherian than eutherian ova.

Hurst (1997) and others have criticized the kin selection hypothesis of genomic imprinting on grounds that it is not rigorously testable or sufficiently quantitative, and that it is too accommodating of seemingly contradictory data. For example, the kin selection hypothesis predicts rapid evolution of imprinted genes (McVean and Hurst 1998), but the gene regions examined to refute this hypothesis were limited to expressed portions. Given that genomic imprinting is essentially a phenomenon of gene transcription, regulatory changes seem to be more likely.

Burt and Trivers (1998) noted that genes expected to be under selection in parental antagonism involving genomic imprinting could include the imprinted genes themselves, genes that encode products that affect imprinting status of 1 or more imprinted genes, and genes that themselves or whose products are affected by imprinted genes. This scenario obviously presents a large and complex set of targets to examine for selection.

The prevalence of mammalian imprinted genes involved in growth pathways is such that they cannot be ignored by any hypothesis that seeks to explain the phenomenon. A straightforward alternative to the kin selection hypothesis is that mammalian growth requires very tight regulation, and silencing 1 allele helps accomplish this. That mammalian physiology requires such regulation is not obvious, except that unlike many other vertebrate groups, mammals show determinate growth. The higher metabolic rates associated with homeothermy also might result in greater free-radical production and hence DNA damage, presumably responsible for the high rate of mammalian cancer. IGF2 is one of the most common transcripts found in human tumors (Kim et al. 1998; Rainier et al. 1993). Perhaps without the epigenetic regulation of imprinting, growth control genes would show greater amounts of ectopic expression, even during normal development. Mammalian growth requires a control mechanism such that it is tightly regulated, but can evolve rapidly enough to allow for the huge variation observed between species. Differential imprinting of growth-related genes could provide this sort of flexibility. This mammalian growth sensitivity model does not predict that paternally expressed loci should be growth promoting and maternally expressed loci growth retarding. If this trend among imprinted genes holds, then the parental antagonism hypothesis has greater explanatory power.

The monoallelic expression of imprinted genes suggests a role in reproductive isolation under the Dobzhanksy–Muller speciation model (Muller 1942; Orr 1993). This model posits 2 or more interacting loci in which an ancestral population is homogenous (AABB). Population isolation leads to derived alleles: population 1 = AAbb and population 2 = aaBB. Both A–b and a–B combinations are either selectively advantageous or neutral relative to the ancestral A–B interaction. However, a–b interactions may produce a deleterious phenotype and thus limit hybridization between the 2 populations. Imprinting of 1 or more of the loci involved in this model may increase the chances of hybrid dysgenesis. If the new alleles are recessive, heterozygous hybrids (AaBb, Aabb, and aaBb) would not necessarily be at a disadvantage. Should allele A or B be silenced, however, the chances of an a–b (deleterious) interaction increase. A similar model involving X-linked genes has been invoked by several authors to explain Haldane's rule, the disproportionate effects of hybridization on the heterogametic sex (Haldane 1922). Our work on hybrid dysgenesis in the genus Peromyscus is consistent with such a model: we have implicated an X-linked locus and another locus mapping to an imprinted domain in the effects (described below). At the very least it is clear that placentation is a major barrier to mammalian interspecies hybridization. Imprinted genes clearly play a significant role in this process.

Genomic Imprinting in the Peromyscus maniculatus Species Complex

One of the best documented cases of reproductive isolation involving parent-of-origin effects involves the cricetid rodent genus Peromyscus (Steppan et al. 2004). The deer mouse (P. maniculatus) is the most widespread of this most common group of North American mammals. P. maniculatus extends from Alaska to Mexico City, from Pacific to Atlantic coasts, and from sea level to approximately 14,000 feet (Dewey and Dawson 2001). Notably absent from this range is the southeastern United States, a gap filled by a recently diverged species, the oldfield mouse (P. polionotus). Ongoing colonies of randomly bred individuals from a single locale were established in 1948 for P. maniculatus (descended from 40 animals of the subspecies bairdii from Washtenaw County, Michigan) and in 1952 for P. polionotus (from 21 animals of the subspecies subgriseus from Ocala National Forest, Florida). These colonies have been used for genetic, developmental, physiological and behavioral studies (Dewey and Dawson 2001).

Early studies suggested that P. maniculatus × P. polionotus hybrids could occasionally be produced. Watson (1942) 1st demonstrated that hybrids were more readily obtained when crossing female P. maniculatus with male P. polionotus than the reciprocal combination. Liu (1953) demonstrated that this asymmetry was due to fetal mortality in the latter. Dawson (1965) 1st quantified the hybrid phenotypes and genetics, showing that female P. maniculatus (strain BW, subspecies P. m. bairdii, from Washtenaw County, Michigan) mated with male P. polionotus (strain PO, subspecies P. p. subgriseus, from Ocala National Forest, Florida) resulted in offspring significantly smaller than either parental strain. These BW × PO hybrids were not only growth retarded at birth, but remained smaller throughout life (Fig. 6C). Maintaining this undersized phenotype is atypical, because most mammals that are growth retarded at birth exhibit faster or prolonged growth, presumably to reach a set selected size. Both sexes of the BW × PO hybrids are fertile. The reciprocal cross, female PO crossed with male BW, results in disproportionately large offspring, few of which survive past birth. The rare postnatal survivors were almost entirely female, fertile, and remained larger than either parental strain (Dawson et al. 1993).

Rogers and Dawson (1970) showed that the hybrid phenotypes are particularly pronounced in the placenta: PO × BW placentas average 5–6 times greater in mass than those of the BW × PO cross. The latter hybrids show reduced amounts of the spongiotrophoblast layer of the muroid placenta (Fig. 6A). The spongiotrophoblast contains endocrine cells as well as gives rise to cell lineages that invade the maternal uterine wall (Cross et al. 2003). Additional studies by multiple groups have demonstrated the repeatability of these phenotypes (Duselis et al. 2005; Maddock and Chang 1979).

One potential explanation of such reciprocal hybrid growth phenotypes is interactions between species-specific mitochondrial and nuclear genes. Dawson et al. (1993) tested this hypothesis using selective backcrosses to obtain animals whose genomic content was 1 strain, while their mtDNA was derived from the other. Contrary to the mtDNA hypothesis, these crosses resulted in lessened growth effects, suggesting that interactions between nuclear genes with parent-of-origin effects are responsible for the hybrid dysgenesis. As an alternative hypothesis, they proposed that mispairing of imprinted chromosomal domains might be the cause of the phenotypes.

My colleagues and I began our investigations of the Peromyscus growth phenotypes by verifying the phenotypes. Consistent with earlier reports, we found growth phenotypes that were particularly pronounced in the placenta. Indeed, in the female PO × male BW cross, we frequently observed conceptuses consisting only of a large, dysmorphic placental mass, reminiscent of hydatidiform mole disease. Further, we noted that when embryos were present in the PO × BW cross, they were not only larger, but always exhibited abnormal morphologies. Many of these phenotypes are similar to diseases in humans and transgenic Mus caused by altered expression of imprinted genes. As suggested by earlier studies, our current research shows that fewer than 50% of pregnancies from PO × BW crosses survive past midgestation (A. Duselis and P. Vrana, in litt.).

We then tested whether the phenotypes might be a consequence of P. polionotus having lost genomic imprinting. P. polionotus, unlike P. maniculatus, has been described as largely monogamous (Foltz 1981). Reciprocal growth observations could be explained if imprinting does not occur, or control is weakened by lack of selection in the monogamous species, as suggested by the kin selection hypothesis. In contrast, we found that imprinting has been maintained in P. polionotus at all 3 loci tested (Vrana et al. 1998). However, because, the 2 species are thought to have diverged in the Pleistocene (Blair 1950), there likely has been insufficient time for a complete loss of genomic imprinting.

We next tested the parental expression status of known imprinted genes in these hybrids. We found only minor perturbations of imprinting in the undersized hybrids, but the more lethal PO × BW oversized hybrids showed complete loss of imprinting of a number of unlinked loci. For at least 2 genes (H19 and Pw1/Peg3), this loss of imprinting was accompanied by reduced DNA methylation on the silenced allele. We do not yet know if the alterations in DNA methylation are limited to imprinted regions; however, analysis with methylation-sensitive restriction enzymes suggest the changes are not genome-wide. Examination of these data strongly suggests that at least some aspects of imprinting control can evolve rapidly.

Our efforts to elucidate the genetics of the Peromyscus hybrid dysgenesis have primarily focused on the overgrowth. We have shown that 2 distinct genetic mechanisms contribute to the PO × BW phenotypes. First, we identified 2 regions of the genome that contribute to the overgrowth. We mapped these loci by performing 2 backcrosses with the undersized hybrids: 1 in which the maternal genome was varied (female F₁ × male BW), and 1 in which the paternal genome is varied (female PO × male F₁). The former revealed an association of overgrowth with the distal PO X chromosome, whereas the latter cross showed linkage to the BW Pw1/Peg3 gene (Vrana et al. 2000). Analysis of intercross animals (F₁ × F₁) suggested that the PO X locus must be inherited maternally, and the BW Pw1/Peg3 linked locus inherited paternally to manifest the overgrowth (Vrana et al. 2000). Another unmapped locus associated with the platinum coat-color mutation also affects growth in backcross animals and shows dependence on parental origin (W. D. Dawson et al., in litt.)

Our current refining of the areas of linkage has revealed candidate loci in both domains: within this region of the X chromosome is the homeobox containing gene Esx1 (Loschiavo et al., in press). Based on a gene knockout, Esx1 directly impacts placental structure and growth and indirectly affects embryo growth (Li and Behringer 1998). Surprisingly, the gene shows little mouse–human conservation–there is only 65% amino acid identity in the DNA binding homeodomain (Fohn and Behringer 2001). Outside that key motif, the 2 sequences are essentially not alignable.

In contrast to the genomic region surrounding Esx1, the Pw1/Peg3 imprinted domain is very gene dense. The Pw1/Peg3 gene knockout suggests a role in placental and embryonic growth as well as behavior. Further, we have found a number of amino acid substitutions between the PO and BW alleles. However, comparative genomic analyses suggest that the entire domain is rapidly evolving: domestic cows, humans, and mice show differences in both gene content and imprinting status. For example, Zim2 is another paternally expressed zinc-finger gene that overlaps Pw1/Peg3 (sharing some exons) in humans (Kim et al. 2000). Zim2 is present in Mus, but does not share exons with Pw1/Peg3, and is maternally expressed. The Mus genome has another maternally expressed zinc-finger gene, Zim1, that lies between Zim2 and Pw1/Peg3. Both the human and domestic cow genomes lack Zim1; the latter has an unrelated, nonimprinted gene termed Ast1 in the equivalent position (Kim et al. 2004).

However, these 2 mapped loci do not explain all the PO × BW phenotypes. Both backcrosses can contribute PO alleles maternally and BW alleles paternally; both yielded overgrowth. However, the female F₁ × male BW offspring did not exhibit the loss of imprinting, aberrant morphologies, or the degree of placental overgrowth observed in the PO × BW F₁ cross (Table 2). In contrast, the female PO × male F₁ cross produced litters that exhibited all the attributes of the PO × BW cross, including maternal lethality due to inability to pass term fetuses through the birth canal.

The difference between the 2 crosses was suggested by another cross: we 1st bred the small hybrid animals together in nonsibling matings for 2 generations to create a 3rd generation (G₃) intercross. We then performed a series of crosses in which G₃ females were mated with BW males. We expected this cross to yield the same phenotypic range as the F₁ × BW cross: in both cases the maternal genome is an equivalent mix of the 2 strains, whereas the paternal genome is BW. The 3rd generation intercross animals differ in 1 important aspect from F₁ animals: whereas both types would contain roughly 50% BW and 50% PO alleles, all F₁ mice would be uniformly heterozygous at all autosomal loci (Fig. 7). In contrast, individual G₃ animals would either be heterozygous, or homozygous for the BW or the PO allele of each gene. The female G₃ × male BW crosses yielded a number of litters that exhibited more extreme overgrowth, developmental abnormalities, and loss of imprinting. The differences between the F₁ × BW and G₃ × BW crosses suggested that maternal homozygosity of PO alleles might underlie the loss of imprinting and associated defects (Fig. 7).

Comparing an F₁ × F₁ (using the small hybrids) cross with a G₃ × G₃ cross revealed a similar effect. Again, both females are an equal mix of PO and BW alleles, as are the males. Both crosses produce a spectrum of sizes from undergrowth to overgrowth. However, more severe overgrowth appears in the G₃ × G₃ cross, but not the F₁ × F₁ cross (Table 2). Examination of a number of crosses has supported our maternal effect hypothesis (Duselis et al. 2005). The maternal effect appears to be due to a small number of loci, and not to be associated with the Dnmt genes. The PO maternal effect gene product(s) are predicted to be essential to normal imprinting, but cannot recognize or maintain imprints in the presence of incoming foreign BW paternal DNA or other sperm factor.

Our preliminary analysis of the undergrowth phenotypes also indicated involvement of an X chromosome locus. The Esx1 gene appears to be proportionately overexpressed in the BW × PO placentas. However, we have detected no linkage to Pw1/Peg3 or nearby imprinted genes. Ongoing analysis of the BW × PO animals suggests some changes in imprinted gene expression levels without formal loss of imprinting.

One major question is whether the Peromyscus hybrid effects are limited to the populations from which the BW and PO strains were derived. We have tested crosses with BW and the P. polionotus subspecies leucocephalus (strain LS), originating from individuals on Santa Rosa Island, Florida. We have not detected any differences between PO × BW hybrids and LS × BW hybrids. Further, PO × LS hybrids do not exhibit any abnormal prenatal growth phenotypes, or any loss of imprinting.

Several subspecies of P. maniculatus limited to the Pacific Northwest have been designated a separate species, P. keeni (Chirhart et al. 2005). Several preliminary studies have suggested that P. keeni crosses with neighboring subspecies of P. maniculatus yield nonviable offspring and exhibit parent-of-origin effects (Dice 1949; Liu 1954). Several other former subspecies of P. maniculatus also have been proposed to have species-level status (Chirhart et al. 2005). It would be particularly interesting to assess gene flow in these Peromyscus populations at the X-linked and imprinted regions we have already implicated in the hybrid dysgenesis. I predict that these same regions will stand out in showing disjunct allele flow between populations. It may also be informative to assess patterns of DNA methylation at specific differentially methylated regions across the P. maniculatus complex. Very few data are available on population-level epigenetic variation in any species.

To further assess the generality of the P. maniculatus bairdii × P. polionotus subgriseus hybrid effects, several experiments appear paramount. First would be to assess imprinted gene status and growth effects in hybrid fetuses. An easier assay is to test whether DNA methylation or other epigenetic marks are altered in hybrid animals. Mating patterns also should be assessed to determine if they differ between the 2 populations.

Environmental Effects on Epigenetic Gene Regulation

Epigenetics is a still-emerging field that has relevance for many areas of biology. One of the most important but least understood aspects of epigenetic gene regulation is the role of the environment. Examination of data to date suggests that environmental effects on epigenetics are significant. Domestic cow embryos that have been produced by in vitro fertilization often exhibit overgrowth and corresponding misexpression of imprinted genes (Allen and Reardon 2005; Blondin et al. 2000). Human children derived from in vitro fertilization exhibit a much higher incidence of Beckwith–Wiedemann and other syndromes associated with misregulation of imprinted genes (Allen and Reardon 2005; Niemitz and Feinberg 2004). Studies with Mus embryos have shown that even small changes in culture media may affect epigenetic marks and the expression levels of imprinted genes (Doherty et al. 2000). Again, extraembryonic tissues seem particularly sensitive to these effects (Mann et al. 2004).

Although the cultured embryo studies are suggestive, they involve unnatural environments. The “Fetal Programming of Adult Disease” hypothesis posits that even small alterations in fetal growth due to changes in maternal diet or environment result in increased susceptibility to cardiovascular disease, certain cancers, and other ailments decades later (Barker 1992, 1993, 1994). Although we do not know how these signals are transduced, perhaps maternal signals influence fetal epigenetic marks. One example of such an effect involves the hypoxia-inducible factor proteins in the developing placenta. The hypoxia-inducible factor proteins sense O₂ levels, in this case those in the maternal uterine blood vessels. Interestingly, a class of enzymes that modulate epigenetic marks, histone deacetylases, have been shown to be reliant on hypoxia-inducible factor activity in developing placental cells (Maltepe et al. 2005). In turn, histone deacetylase activity affects the proportions of cell types, presumably via gene expression. Hypoxia-inducible factor activity has also been implicated in activating Igf2, further linking imprinted genes and pathways in these critical events (Feldser et al. 1999).

More striking is a recent study of human monozygotic twins that strongly suggests that environmental effects on epigenetic marks are inevitable during postnatal life (Fraga et al. 2005). Global patterns of DNA methylation and histone acetylation are nearly identical in most 3-year-old monozygotic twins. In contrast, 50-year-old twins inevitably exhibited significant differences in these profiles. Further, differences in level of gene expression were correlated with the degree of variation in epigenetic marks.

Nearly entirely unexplored are potential situations where altered gene expression via epigenetic change is an adaptive response to an environmental change. Clearly, the potential benefits of such changes are enormous. Haig (2004) has noted that under the kin selection hypothesis, it would be advantageous to modulate genic imprinting status based on such factors as interrelatedness and potential length of relationship. Such changes could potentially be mediated by pheromones. A recent study shows that male and female Mus avoid urinary odors of their maternal, but not paternal, origin (Isles et al. 2002). Thus imprinted gene–environmental interactions also may play a role in dispersal patterns. I suggest that epigenetic changes underlie the complex interplay between genetics and environment.

Imprinted genes can be regulated by these same epigenetic changes, and have been shown to be involved in mammalian growth, metabolism, and behavior. Differential growth underlies not only overall size differences, but within specific tissues may underlie other morphological changes as well. I predict that changes in control of imprinted gene expression or imprinted gene interactions will be shown to be a major force underlying the spectacular diversity seen in mammalian size, morphology, and behavior.