Physiological functions of imprinted genes



Genomic imprinting in gametogenesis marks a subset of mammalian genes for parent-of-origin-dependent monoallelic expression in the offspring. Embryological and classical genetic experiments in mice that uncovered the existence of genomic imprinting nearly two decades ago produced abnormalities of growth or behavior, without severe developmental malformations. Since then, the identification and manipulation of individual imprinted genes has continued to suggest that the diverse products of these genes are largely devoted to controlling pre- and post-natal growth, as well as brain function and behavior. Here, we review this evidence, and link our discussion to a website ( containing a comprehensive database of imprinted genes. Ultimately, these data will answer the long-debated question of whether there is a coherent biological rationale for imprinting. © 2002 Wiley-Liss, Inc.


Articles published in the mid 1980s first alerted biologists to the existence of genomic imprinting in mammals (McGrath and Solter, 1984; Surani et al., 1984; Cattanach and Kirk, 1985). These studies used two different approaches—replacement of male or female pronuclei in fertilized oocytes, or breeding of mice carrying Robertsonian chromosomal translocations, to create conceptuses whose genomes were either completely derived from one parent, or contained specific chromosomes that were uniparental in origin. With the first approach, the conceptuses never developed to term, although they contained a normal amount of genomic DNA. The bi-maternal (gynogenetic) embryos developed to the early somite stage before dying. These embryos were growth-retarded, and their extraembryonic tissues (placenta and yolk sac) were extremely small. Androgenetic conceptuses (bi-paternal DNA complement) showed a very different phenotype, with the embryos arresting in development at an earlier stage, and showing overgrown extraembryonic tissues. The mice that were bred to have uniparental disomies (UPDs) for individual chromosomes also carried a normal amount of DNA, but they also showed abnormal phenotypes. In the initial findings, and in subsequent studies, these phenotypes often including overgrowth or growth retardation, as well as abnormal behavior, which depended on which chromosome or chromosomal region had been made uniparental. Not all chromosomes produced abnormal phenotypes when present as UPDs, but those that did gave striking phenotypic differences with paternal vs. maternal origin of the UPD (Beechey, 1999).

None of these observations were compatible with Mendelian genetics, which would predict that gene expression, and hence development, should not be affected by the parental origin of the DNA. A way out of this dilemma was provided by postulating that the DNA is somehow modified during gametogenesis, with the pattern of modifications differing in male vs. female gametes. The maternally transmitted chromosomes would, therefore, differ functionally from those contributed by the father, so a bi-parental complement of DNA would be required for normal development. Two questions arose: first, what type of modification of the DNA and/or chromatin could account for imprinting and second, what could be the identities of the imprinted genes? These two questions remain relevant today, and they underlie the two major objectives of the imprinting field—to define the mechanism of imprinting and to understand the functions of imprinted genes.

There is good evidence for DNA methylation as a mechanism of imprinting. Methylation at cytosines in CpG dinucleotides satisfies a number of criteria for an imprinting mechanism: it can silence genes when it occurs in their promoters, it can be propagated to daughter cells, thereby acting as an epigenetic mark, and yet it can also be erased under special conditions, for example, when DNA replication occurs in the absence of a maintenance methyltransferase, or perhaps in the presence of an active demethylase. This susceptibility to erasure is critical, since the imprint must be “reset” during gametogenesis in each successive generation. Many imprinted loci show allele-specific DNA methylation, and in the absence of the maintenance methyltransferase, or in the presence of methyltransferase inhibitors, the imprinting of several genes is erased (Li et al., 1993; Dao et al., 1999). Imprinted genes are sometimes found clustered together in megabase-scale ‘chromosomal domains’ (Maher and Reik, 2000). Since the roles of DNA methylation and chromosomal domains in imprinting have been extensively reviewed, we will not deal with these further. The remainder of our discussion will focus on the physiological roles of imprinted genes, and consider how these roles may reveal a biological rationale for imprinting.


That there may be such a rationale is suggested by the fact that as the list of known imprinted genes has grown, there has been a striking concordance between the findings in humans and mice. In these two mammals, whose lineages diverged in evolution about 90 million years ago, imprinting is conserved, both in terms of its presence or absence and in its parental “direction” (i.e., the parental origin of the active vs. silenced allele), for nearly every imprinted gene. For loci where the cross-species comparison can be done, there are only three exceptions. These are the Igf2r gene, which is imprinted strongly in mice, but weakly or not at all in humans (Kalscheuer et al., 1993), and two other genes, U2af1-rs1 and Impact, which are also imprinted in mice, but not in humans (Pearsall et al., 1996; Okamura et al., 2000). Since the list of bona fide imprinted genes now has more than 50 members, these three exceptions “prove the rule.” Imprinting is also conserved in other mammals, including sheep (Charlier et al., 2001a,b) and at least one marsupial, the opossum (Killian et al., 2000). So, while certain other classes of vertebrates may lack imprinting (O'Neill et al., 2000), there is evidently natural selection for the maintenance of this phenomenon in placental mammals.

There have been a number of different proposals to explain the conservation of imprinting, and we will consider one model that seems to be holding up well to experimental data. This is the intergenomic conflict hypothesis, which was proposed 10 years ago by (Haig and Graham (1991) and Moore and Haig. (1991). This theory was motivated by data from mice with germline knockouts (KOs) of two prototypical imprinted genes, Igf2 and Igf2r. The Igf2 gene, which encodes insulin-like growth factor-II, is imprinted such that the paternal allele is expressed and the maternal allele silenced, while the Igf2r gene, which encodes a clearance receptor for the circulating Igf2 peptide, is imprinted in the opposite direction (maternal allele expressed; paternal allele silenced). The two deletions produced opposite effects on growth, with growth retardation in the Igf2-null conceptuses, and overgrowth in the Igf2r-null conceptuses.

The conflict hypothesis seeks to explain these observations by positing opposite maternal vs. paternal “drives” to control allocation of maternal resources to each conceptus. According to this hypothesis, in settings of multiple paternity, the father will propagate his genome most efficiently, if his germline imprints genes in a pattern that promotes the growth of his offspring, both in utero and in the post-natal period. The mother, by contrast, is postulated to propagate her genome more successfully by imprinting genes to prevent undue metabolic demands on her resources by any single conceptus, or by any single pregnancy. The conflict hypothesis was supported by mathematical models for efficiencies of the spread of maternal vs. paternal alleles in populations, given the realistic premises of promiscuous mating and maternal (but not paternal) nurturing of the offspring.

This model makes a strong prediction that the function of an imprinted gene should correlate with its direction of imprinting. Imprinted genes that are paternally expressed/maternally silenced are predicted to promote growth of the offspring, either in utero or in the perinatal period, or in some other way increase demands on maternal resources, while imprinted genes with the opposite direction of imprinting should have the opposite effect. (In this review, we use the term “maternally expressed” as a short-hand, but “expressed from the maternal allele” should be understood.) It may be important that, in principle, behavioral traits, such as food seeking, can also affect the demands placed by offspring on maternal resources. Each report of a novel imprinted gene has provoked discussion of how that gene may fit in with the conflict model, but it has taken a decade to accumulate enough data so that a definitive answer is in sight. The hypothesis is by no means proven, and there are reasonable competing ideas (Pardo-Manuel de Villena et al., 2000), but some variant of the conflict model appears increasingly likely to be correct.

However, it is important to consider the biases, which may lead to false conclusions about the function of imprinted genes. The emphasis on growth-altering functions of imprinting arose from the early documentation of genomic imprinting of Igf2 and its regulating receptor (Igfr). Furthermore, discovery of imprinted loci from the use of mice with Robertsonian translocations would bias toward viable and gross phenotypes, which may result in overrepresentation of growth-related abnormalities. Lastly, a disproportion number of the imprinted genes have been discovered because of their association with the growth disorder Beckwith-Wiedemann syndrome (BWS) or the neurological disorders Prader–Willi syndrome (PWS) and Angelman syndrome (AS), perhaps further biasing our perception of the processes regulated by imprinting. These biases can only be eliminated by more data from genome-wide scans for imprinted genes.

Here, we focus our discussion on those imprinted genes for which there is reasonable in vivo evidence of function. This evidence is mainly from KOs in mice, and the counterpart of these experiments in the form of natural deletions or mutations in humans, and for one imprinted chromosomal region, in sheep. In the sections below, we discuss all imprinted genes for which such data are available. In the text, we have grouped these genes by their direction of imprinting (paternal vs. maternal), rather than by chromosomal location. However, Figure 1 shows these loci in the context of the human chromosomes.

Figure 1.

Imprinted genes for which in vivo experimental data are available. The gene names are colored according to functional category, with bold type indicating strong evidence and plain type suggestive evidence. The genes are shown next to the chromosome homologue from which they are expressed. M, maternal; P, paternal.


Igf2 (mouse distal Chr7, human Chr11p15.5)

The prototype paternally expressed gene is Igf2 (mouse loci will be indicated conventionally in lower case and human loci in upper case). This was the first endogenous gene proven to be imprinted in mammals (DeChiara et al., 1991). This gene, together with the oppositely imprinted gene Igf2r, was one of the examples that originally motivated the conflict hypothesis for imprinting. Igf2 encodes the precursor peptide for insulin-like growth factor II (Baker et al., 1993), and it is expressed in a wide range of fetal tissues. The maternal allele of Igf2 is silenced; paternal allele active in many fetal tissues of mice and humans (DeChiara et al., 1991). Igf2 is imprinted oppositely from the closely linked H19 gene (see below), and these two genes are inversely regulated in cis, with expression controlled by a differentially methylated ‘imprinting control region’ upstream of H19 (Leighton et al., 1995a; Bell and Felsenfeld, 2000; Hark et al., 2000).

Insulin like growth factor II is a mitogen for fetal and placental cells, and it also has an anti-apoptotic function (Baker et al., 1993; Christofori et al., 1994; Morison and Reeve, 1998). A subset of individuals with the BWS of somatic overgrowth show abnormal biallelic expression of IGF2 (“loss of imprinting”) in some tissues (Weksberg et al., 1993; Morison et al., 1996; Morison and Reeve, 1998), while other cases are due to mutations or silencing of a different imprinted gene, p57KIP2 (see below). The tissues that are most affected in BWS coincide with the normal sites of high IGF2 expression (Hedborg et al., 1994). A mouse model for BWS was constructed by breeding loss of imprinting of Igf2 (an H19 deletion allele) into a p57Kip2-null background (Caspary et al., 1999). IGF2 loss of imprinting is also a frequent phenomenon in human cancers, including Wilms' tumors and other malignancies associated with BWS (Ogawa et al., 1993; Rainier et al., 1993; Feinberg, 1999). In Wilms' tumors, this loss of imprinting is invariably linked to silencing and hypermethylation of H19 (Moulton et al., 1994; Steenman et al., 1994; Taniguchi et al., 1995). Germline deletion of Igf2 in mice causes generalized growth retardation, and overexpression of Igf2 causes tissue hyperplasia and tumor susceptibility (Sun et al., 1997; Pravtcheva and Wise, 1998). So, Igf2 matches the predictions of the conflict hypothesis.

Ins1 (mouse Chr19) and Ins2/INS (mouse distal Chr7, human Chr11p15.5)

Mice have two insulin genes, Ins1 and Ins2, with Ins2 orthologous to the single human INS gene. The maternal allele of Ins2/INS is repressed; paternal allele active in yolk sacs (and variably in thymus glands) of mice and humans (Giddings et al., 1994; Deltour et al., 1995; Duvillie et al., 1998), and this is also true of Ins1 in mice. However, there is biallelic expression in other tissues, including pancreas (Giddings et al., 1994). Insulin is best known as a hormonal regulator of glucose and energy metabolism, but it is also a growth factor. Consistent with this, hyperinsulinemia in humans is a major cause of high birth weight infants (Weintrob et al., 1996). Similarly, the deletion of Ins2, together with its homologue Ins1, in mice had the expected consequences of growth retardation and metabolic abnormalities (Duvillie et al., 1997). But no evidence has yet been shown for a parent-of-origin effect in this phenotype, so there is still no proof that Ins1/Ins2 conform to the predictions of the conflict model.

Kcnq1ot1, aka Kvlqt1-as, Lit1 (mouse distal Chr7, human Chr11p15.5)

The maternal allele of Kcnq1ot1 is silent; paternal allele active in humans and mice (Lee et al., 1999; Smilinich et al., 1999). The LIT1/KvLQT1-antisense RNA arises from the KvDMR1 CpG island, located in an intron of the human KvLQT1/KCNQ1 gene, and an orthologous transcript arises from an intronic CpG island of the murine Kcnq1 locus (Engemann et al., 2000). LIT1 encodes a non-translated RNA. Expressed sequence tags (ESTs) define the LIT1 RNA, and this RNA may be a long primary transcript collinear with the DNA sequence (Smilinich et al., 1999; Engemann et al., 2000). There is no evidence for splicing to a discrete size. Based on a deletion engineered by homologous recombination in somatic cells, the KvDMR1 CpG island functions as an imprinting control element (Horike et al., 2000). Evidence for this also comes from studies of BWS, in which a subset of cases show loss of DNA methylation at KvDMR1 (Smilinich et al., 1999). This is linked to downregulation of p57KIP2 mRNA and, less frequently, with loss of imprinting of IGF2 (Lee et al., 1999; Smilinich et al., 1999; Horike et al., 2000). KvDMR1, therefore, is a second ‘imprinting center’ on human Chr11p15.5/mouse distal Chr7, analogous to the differentially methylated region upstream of H19. As a paternally expressed gene that antagonizes expression of the growth-inhibitor p57KIP2 (see below) in cis, Kcnq1ot1 conforms to the predictions of the conflict hypothesis.

Dlk1 (mouse distal Chr12, human Chr14q32, sheep Chr18q)

The maternal allele of Dlk1 is silent; paternal allele expressed in mice (Schmidt et al., 2000; Takada et al., 2000), humans (Wylie et al., 2000), and sheep (Charlier et al., 2001a,b). Like Igf2 and H19, Dlk1 and a gene called Meg3/Gtl2 (see below) are closely linked and expressed from opposite alleles. Dlk1 (Delta-like) encodes a transmembrane protein with multiple EGF-repeats. This structure is similar to ligands for Notch signaling. Dlk1 mRNA is detected in many fetal and adult tissues, including endocrine organs (Tornehave et al., 1996; Schmidt et al., 2000; Takada et al., 2000), and it is abundant in skeletal muscle in sheep (Charlier et al., 2001a,b). Dlk protein inhibits adipocyte differentiation and alters hematopoietic cell differentiation in culture (Garces et al., 1999; Ohno et al., 2001). Dlk1 and other imprinted genes, including Meg3/Gtl2, map to the Callipyge locus of sheep. The Callipyge trait of skeletal muscle hypertrophy appears to result from dysregulation of multiple genes in this imprinted domain (Charlier et al., 2001a,b). Although the mutation accounting for Callipyge has not yet been identified, the DLK1 gene is over-expressed in the skeletal muscle of these sheep (Charlier et al., 2001a,b). So, while it is not clear whether the Callipyge trait has an impact on allocation of maternal resources, and although a gene-specific KO of Dlk1 is pending, the Dlk1 gene is likely to fit with the general prediction of the conflict hypothesis that paternally expressed imprinted genes should promote growth.

Sgce (mouse proximal Chr6, human Chr7q21-q22)

The maternal allele of Sgce is silenced; paternal allele active (Piras et al., 2000; El Kharroubi et al., 2001). This gene encodes sarcoglycan-epsilon; one of five members of the sarcoglycan family, which are components of the dystrophin–sarcoglycan complex in muscle and in other tissues (Bonnemann et al., 1996; Ettinger et al., 1997). Sgce protein is found in smooth muscle of blood vessels (Straub et al., 1999) and in brain cells (Zimprich et al., 2001). Sarcoglycans are thought to functionally link the cytoskeleton to the extracellular matrix (Bonnemann et al., 1996). Loss of Sgce may account for embryonic lethality associated with maternal UPD of mouse proximal Chr6, but this remains unproven. SGCE was also considered a candidate for Silver-Russell syndrome (SRS; human Chr7q), characterized by intrauterine and post-natal growth retardation, and relative macrocephaly associated with mild facial anomalies. However, heterozygous loss-of-function mutations in the SGCE gene were recently found to cause a different disorder, myoclonus-dystonia syndrome (Zimprich et al., 2001). A parent-of-origin dependence of the penetrance of that genetic disease is consistent with imprinting of human SGCE. It is not yet clear whether the phenotype of myoclonus-dystonia (SGCE-deficiency) is informative regarding the conflict hypothesis, but behavioral changes associated with this syndrome (obsessive-compulsive disorder and panic attacks) may prove to be relevant.

RasGrf1, aka Grf1 (mouse distal Chr9, human Chr15q23-q24)

The maternal allele of Grf1 is silenced, paternal allele active in multiple mouse tissues, including brain (Plass et al., 1996; Brambilla et al., 1997; Pearsall et al., 1999). This gene encodes a guanine nucleotide exchange factor (GEF), which stimulates Ras signaling by promoting the GTP-bound state (Overbeck et al., 1995). Grf1 is highly expressed in the brain (Ferrari et al., 1994; Finkbeiner and Dalva, 1998). RasGrf1 KO mice are viable and − mat/+ pat heterozygotes lack RasGrf protein in brain, as predicted. The mice showed memory or learning deficits, suggesting involvement of RasGrf protein in synaptic remodeling (Brambilla et al., 1997). RasGrf1-null mice are up to 15% smaller than littermate controls, with the maximal growth difference seen in the adults; the effect is small in magnitude, but it has been reproduced in two independent laboratories (Itier et al., 1998; Orban et al., 1999). This phenotype has been attributed to reduced growth hormone secretion; probably due to a hypothalamic neuronal deficit (Itier et al., 1998). Thus, Grf1 promotes growth, but apparently only in the post-natal period. It, therefore, falls into the group of paternally expressed imprinted genes that at least superficially adhere to the predictions of the conflict hypothesis in terms of growth, but that do not have obvious effects on allocation of maternal resources. Future studies of behavioral aspects of the Grf1-null phenotype in the perinatal period might shed more light on this.

Mest, aka Peg1 (mouse proximal Chr6, human Chr7q32)

The maternal allele of Mest/Peg1 is repressed; paternal allele active in mice, sheep, and humans (Kaneko-Ishino et al., 1995; Kobayashi et al., 1997; Riesewijk et al., 1997; Feil et al., 1998). There is strong expression and functional imprinting in many fetal tissues; the imprint persists in adult tissues, but expression declines. This gene encodes a putative hydrolase enzyme (Kaneko-Ishino et al., 1995), with sequence similarity to epoxide hydrolases of bacteria and plants, but there are no direct biochemical studies showing enzymatic activity (Kaneko-Ishino et al., 1995). Paternal transmission of a knockout allele in mice caused embryonic and placental growth retardation, associated with reduced post-natal growth and reduced survival in the mutant progeny (Lefebvre et al., 1998). Mest-deficient females showed abnormal maternal behavior (deficient nest building and pup retrieval, and impaired placentophagia) (Lefebvre et al., 1998). The embryonic and placental phenotypes are consistent with the predictions of the conflict hypothesis. As has been pointed out, the maternal phenotype is neutral in terms of conflict per se, but is nonetheless a critical function of the Mest gene that is likely to have preserved the expression of this gene during evolution (Keverne, 2001).

Peg3, aka Pw1 (mouse proximal Chr7, human Chr19q13.4)

The maternal allele of Peg3/Pw1 is silenced; paternal allele active in multiple fetal tissues, and in adult brain and ovary of mice, as well as in several fetal and adult human tissues (Kuroiwa et al., 1996; Li et al., 2000; Hiby et al., 2001; Murphy et al., 2001). Peg3 encodes a large Kruppel-type zinc finger transcription factor with a proline-rich domain, which is expressed widely during fetal development of mice, and strongly in adult neurons and skeletal muscle (Kuroiwa et al., 1996; Relaix et al., 1996). Human PEG3 has strongest expression in the placenta and ovary, but PEG3 mRNA is also detected in other tissues (Kim et al., 1997). In cell lines, Peg3/Pw1 protein associates with TRAF2, a downstream effector of tumor necrosis factor (TNF) signaling, but not with other TRAF family members (Relaix et al., 1998). Peg3 expression activated NF-κB via IκB–NF-κB dissociation, and acted synergistically with TRAF2. Peg3/Pw1 is induced during p53-mediated cell death (Deng and Wu, 2000). Data from Peg3-knockout mice indicate that like Peg1, this gene regulates both offspring growth and maternal behavior (Li et al., 1999). The Peg3-null offspring had small placentas and were about 20% smaller than their wild-type littermates at birth. These newborns grew poorly after birth, probably due to reduced suckling. There was also markedly reduced nursing behavior and a reduced number of oxytocin-positive neurons in the hypothalamus of Peg3-null adult females. However, the TNF pathway appeared normal in fibroblasts from the Peg3 knockout mice (Ledgerwood et al., 2000). The effects of the Peg3 KO on placental and fetal growth support the conflict hypothesis.

Nnat (mouse distal Chr2, human Chr20q11.2-q12)

The maternal allele of Nnat is silenced; paternal allele active in mice and humans (Kagitani et al., 1997; Kikyo et al., 1997; Evans et al., 2001). Neuronatin is a transmembrane protein that is a member of the proteolipid-family. It is a neuronal protein characteristic of post-mitotic cells, with strongest expression in the perinatal period of CNS development (Joseph et al., 1994, 1995; Wijnholds et al., 1995). It is downregulated on differentiation of PC12 cells into neurons, and in adult brain (Joseph et al., 1996). Nnat is also expressed in proliferating mesodermal derivatives, including the limb buds, condensing mesenchyme of lung and other structures. Human NNAT is highly expressed in embryonal cancers, including medulloblastomas and Wilms' tumors (Li et al., 2002). The neuronatin protein shows some similarity to PMP1 and phospholamban, which function as regulatory subunits of ion channels. There is preliminary evidence from sub-regional UPDs in mice for the involvement of Nnat in cerebellar folding and post-natal viability. Furthermore, while more data are needed, the embryos lacking Nnat expression (those with maternal UPD for a sub-region of distal Chr2) appear to be growth retarded (Kikyo et al., 1997; Williamson et al., 1998). So, while a gene-specific KO, and the creation of transgenic mice overexpressing this gene, will be required to confirm this, Nnat may turn out to meet the predictions of the conflict hypothesis.

Ndn (mouse central Chr7, human Chr15q11-q12, PWS region)

The maternal allele of Ndn is silenced in multiple fetal and adult tissues; paternal allele active (Jay et al., 1997; MacDonald and Wevrick, 1997; Sutcliffe et al., 1997; Watrin et al., 1997). Ndn encodes a nuclear protein, necdin, which shows neuron-specific expression in adults (Maruyama et al., 1991; Aizawa et al., 1992; Uetsuki et al., 1996). Necdin is anti-proliferative in cell transfections (Hayashi et al., 1995). It was reported to bind p53 and other cell-cycle related proteins such as E2F1; possibly allowing post-mitotic neuronal survival by preventing p53-mediated apoptosis (Taniura et al., 1998). Early post-natal lethality was observed in mice with deletion of the paternal allele (Gerard et al., 1999), but viability was seen in another similar knockout strain (Tsai et al., 1999). A third study found variable early post-natal lethality, a small but significant decrease in the number of hypothalamic neurons and abnormal behavior (Muscatelli et al., 2000). PWS maps to the imprinted domain that contains NDN, and expression of NDN is absent from PWS brains, as well as from tissues of a mouse model for PWS imprinting mutations (MacDonald and Wevrick, 1997; Bielinska et al., 2000). Based on the data from mouse models, it is thought that NDN may directly contribute to some aspects of the PWS phenotype (Nicholls, 1999). The behavioral aspects of PWS include failure-to-thrive (hypotonia, poor suckling) in the early post-natal period, followed by obsessive food-seeking behavior later in life. The neonatal phenotype of poor food-acquiring behavior fits the predictions of the conflict hypothesis, although the later hyperphagia does not. Further analysis of these phenotypes and studies of the hypothalamic deficits in humans and in mouse models will be of great interest.

Snrpn (mouse central Chr7, human Chr15q12, PWS region)

The maternal allele of Snrpn is silent; paternal allele active in many fetal and adult tissues, both in mice and humans (Cattanach et al., 1992; Leff et al., 1992; Ozcelik et al., 1992; Glenn et al., 1993). Snrpn encodes an Sm protein, which is a splicing factor component, expressed at highest levels in neurons (Grimaldi et al., 1993). Like the other Sm proteins, the product of the Snrpn gene is a component of the U snRNPs. SmN is thought to substitute for the SmB protein in the Sm heptameric ring that forms the core of snRNPs (Pannone and Wolin, 2000). Whether this confers any novel (neuron-specific?) properties to the ring is not yet known. Snrpn knockout mice, in which only the coding-region has been deleted, lack a strong phenotype (Yang et al., 1998; Tsai et al., 1999). As predicted from its localization to the PWS deleted region and its direction of imprinting, SNRPN is not expressed in PWS tissues (Glenn et al., 1993). Micro-deletions in the 5′ DNA of the SNRPN gene can cause PWS by acting in cis to abrogate the expression of multiple genes in the Chr15q11-q13 imprinted domain (Sutcliffe et al., 1994; Buiting et al., 1995; Saitoh et al., 1996). This ‘domain-effect’ has been mimicked, both molecularly and in terms of phenotypic consequences, in knockout mouse models (Yang et al., 1998; Bielinska et al., 2000). However, in view of the negative data from the Snrpn-specific coding-region KOs, SNRPN itself is not considered a strong candidate for directly contributing to the PWS phenotype (Yang et al., 1998). The available data for Snrpn are, therefore, uninformative with regard to the conflict hypothesis. A second coding unit, Snurf (“Snrpn upstream reading frame”) encodes a small acidic protein, which is translated from an upstream reading frame of the bicistronic Snurf-Snrpn mRNA (Gray et al., 1999). The function of Snurf is not yet known.

Pwcr1, aka MBII, HBII (mouse central Chr7, human Chr15q11.2, PWS region)

The maternal allele of the Pwcr1/MbII locus is silent; paternal allele active in mice and humans (Cavaille et al., 2000; de los Santos et al., 2000; Meguro et al., 2001). This locus, which is included in a large (∼400 kb) primary transcript that emanates from the 5′ region of the Snrpn gene (Runte et al., 2001), consists of multiple tandemly repeated intronless genes, with a core sequence encoding a small nucleolar RNA (snoRNA) (Cavaille et al., 2000; de los Santos et al., 2000; Meguro et al., 2001). Many of the genes in this locus are expressed exclusively or predominantly in brain (Cavaille et al., 2000). In general, snoRNAs function within nucleoli as guidance RNAs in the post-transcriptional 2′-O-ribose-methylation of ribosomal RNA and other small nuclear RNAs. However, the brain-specific C/D box snoRNA HBII-52 has an 18-nt conserved complementarity to a segment of serotonin 2C receptor mRNA, leading to speculation that it may affect the processing of this mRNA (Cavaille et al., 2000). Paternal deletions of this region (Snrpn-Pwcr1-Ube3a) in mice show failure to thrive and lethality in the early post-natal period (Tsai et al., 1999). Since Snrpn and Ube3a deletions are individually not lethal, deficiency of Pwcr1 may account for the lethality (Cavaille et al., 2000). PWCR1/HB11, which lies within the PWS deleted region, is therefore, considered a candidate gene for directly contributing to the PWS phenotype. As mentioned above (see Ndn gene), the failure-to-thrive phenotype of PWS is compatible with the conflict hypothesis.

Ipw (mouse central Chr7, human Chr15q11q12, PWS region)

The maternal allele of Ipw is silenced; paternal allele active in mice and humans (Wevrick et al., 1994, 1996; Wevrick and Francke, 1997; Yang et al., 1998). Ipw (Imprinted in Prader-Willi syndrome) encodes an abundant, spliced, and polyadenylated RNA, which lacks conserved open reading frames. Although a discrete spliced Ipw RNA can be seen on Northern blots, the Ipw exons appear to be included in a very large primary transcript that emanates from the ‘imprinting center’ near the 5′ end of Snrpn (Runte et al., 2001). Ipw RNA accumulates in the cytoplasm, and shows high expression in brain, as well as other tissues (Wevrick et al., 1994). There is evolutionary conservation of its gene structure in mammals (Wevrick et al., 1994, 1996; Wevrick and Francke, 1997). Mutant mice with a deletion encompassing Ipw appear normal after both maternal and paternal transmission of the deletion (Johnson et al., 1995), suggesting lack of a strong Ipw-null phenotype. As predicted from the localization of Ipw to the major PWS deletion, and its direction of imprinting, Ipw RNA is absent from PWS tissues and from tissues of a mouse model for PWS (Wevrick et al., 1994, 1996; Yang et al., 1998). These negative data are not informative with regard to the conflict hypothesis.

Xist (and Tsix) (mouse X, human X)

The maternal allele of Xist is repressed; paternal allele active, in placenta (Kay et al., 1994; Norris et al., 1994; Ariel et al., 1995; Zuccotti and Monk, 1995). The inactivation of the entire X-chromosome shows a marked parent-of-origin dependence in the trophectoderm of mice, such that the paternal X is preferentially inactivated. In contrast, X-inactivation is (usually) a random process in human placenta and in other tissues. Xist encodes a non-translated RNA, which is expressed from and binds to (“coats”) the inactive-X chromosome (Jaenisch et al., 1998). Knockout experiments indicate that Xist is necessary for X-inactivation in ES cells and in mice (Penny et al., 1996; Marahrens et al., 1997). In addition, Xist transgenes can mediate changes at autosomal integration sites that resemble X-inactivation (Lee et al., 1996). So imprinted X-inactivation in mouse placenta may be a consequence of the imprinting of Xist, and/or its oppositely imprinted antisense transcript, Tsix (Lee, 2000; Sado et al., 2001), in trophectoderm cells. The data for Xist are uninformative regarding the conflict hypothesis.


Igf2r (mouse proximal Chr17, human Chr6q26)

The Igf2r gene is strongly imprinted in fetal mouse tissues, with the paternal allele silenced and maternal allele active (Barlow et al., 1991), but IGF2R is either weakly imprinted (polymorphic among individuals) or non-imprinted in human tissues (Kalscheuer et al., 1993; Xu et al., 1993; Oudejans et al., 2001). An antisense transcript, designated as Igf2r-as/Air, originates from an intronic CpG island and is oppositely imprinted from Igf2r. This transcript, and/or its promoter, is necessary for Igf2r imprinting (Wutz et al., 1997). The Igf2r protein binds Igf2 and clears it from the circulation, but does not transmit a growth signal (Baker et al., 1993). An independent function in intracellular trafficking is to target lysosomal enzymes, which are marked by the mannose-6-phosphate modification, to their destination in the lysosome (Kornfeld, 1987). Igf2r knockout mice show fetal overgrowth and lethality in late gestation (Lau et al., 1994; Wang et al., 1994). Conversely, biallelic expression of Igf2r, engineered by deleting a repressive DNA element from the paternal allele, caused fetal growth retardation (Wutz et al., 2001). Data from Igf2r, together with Igf2, originally motivated the conflict hypothesis, and Igf2r certainly meets the predictions of this model.

Meg3, aka Gtl2 (mouse distal Chr12, human Chr14q32, sheep Chr18q)

The paternal allele of Meg3 is silent; maternal allele active in mice (Schmidt et al., 2000; Takada et al., 2000), humans (Wylie et al., 2000), and sheep (Charlier et al., 2001a,b). Meg3 encodes an abundant non-translated RNA that is spliced and polyadenylated. This RNA is found in yolk sac endoderm, and in many fetal and adult tissues (Schmidt et al., 2000; Takada et al., 2000). It is strongly expressed in adult skeletal muscle (Charlier et al., 2001a,b). In the Callipyge trait of skeletal muscle hypertrophy in sheep, which maps to this locus and which presumably affects a cis-acting regulatory element, there is downmodulation of Meg3 expression, as well as altered expression of other genes in this imprinted domain, including increased expression of Dlk1 (see above). Because Meg3 and Dlk1 are closely linked and oppositely imprinted in a pattern reminiscent of Igf2/H19, it is presumed that one function of Meg3 expression is to control Dlk1 expression in cis. Gene-specific KOs will be necessary to confirm this, but for both Meg3 and Dlk1, the Callipyge phenotype at least superficially supports the predictions of the conflict hypothesis. Also potentially consistent with the conflict model, growth retardation is seen in humans with maternal UPD for Chr14, but the imprinted gene(s) responsible for this phenotype are not yet known (Sutton and Shaffer, 2000).

Gnas, aka GNAS1, Gs-alpha (mouse distal Chr2, human Chr20q13.2-q13.3)

The murine Gnas locus is imprinted, with the maternal allele expressed (Yu et al., 1998, 2000, 2001). GNAS1 is variably imprinted in humans (Hayward et al., 1998a,b); and other transcripts in the complex GNAS1 locus (Nesp, Gnasxl) are also imprinted. The expressivity of human GNAS1 mutations, which are associated with the unfortunately named syndromes pseudo-hypoparathyroidism and pseudo-pseudo hypoparathyroidism, depends on parent-of-origin (reviewed in Lalande, 2001). Gnas encodes the alpha subunit of a major heterotrimeric Gs signaling protein. The Gs-alpha subunit is a component of the heterotrimeric stimulatory G-protein that couples multiple hormone receptors (parathyroid hormone receptor and others) to adenylyl cyclase. Gnas knockout mice show an imprinted phenotype, with increased insulin-sensitivity and adipocyte hypertrophy in the null animals (Yu et al., 1998, 2000, 2001). Given these anabolic effects in the KO mice, Gnas fits the general prediction of the conflict hypothesis that maternally expressed imprinted genes should restrain growth.

Ube3a, aka E6AP (mouse central Chr7, human Chr15q11-q13, AS region)

The paternal allele of Ube3a is silenced in hippocampal and cerebellar neurons; there is biallelic expression in many other tissues (Albrecht et al., 1997; Rougeulle et al., 1997; Vu and Hoffman, 1997). The imprinting of Ube3a may be controlled by anti-sense transcripts from an “imprinting center” near the 5′ region of the Snrpn gene (Rougeulle et al., 1998; Runte et al., 2001). Ube3a encodes a ubiquitin conjugating enzyme. Proteins are conjugated to ubiquitin in a series of reactions catalyzed by activating (E1), conjugating (E2), and ligase (E3) enzymes. The Ube3a gene encodes the E6AP ubiquitin-protein ligase (E3), which was first identified by its ability to mediate the degradation of the p53 tumor suppressor protein, acting in conjunction with the human papillomavirus E6 protein (Huibregtse et al., 1991; Scheffner et al., 1993). A separate line of research, based on positional cloning, led to the finding that this protein is mutated in AS, a human neurodevelopmental disorder associated with severe mental retardation and hyperactivity. This condition was originally assigned to the Chr15q11-q13 imprinted domain by evidence from UPDs, chromosomal deletions, and genetic linkage (Nicholls et al., 1989; Wagstaff et al., 1993; Lalande et al., 1999). Point mutations in UBE3A cause some cases of AS, while other cases (paternal UPD and Chr15q11-q13 deletion) are presumed due to loss of expression of this gene (Kishino et al., 1997; Matsuura et al., 1997; Malzac et al., 1998). KO mice lacking expression of Ube3a show a subtle neurological deficit, including reduced learning and deficient long-term potentiation of synaptic transmission (Jiang et al., 1998a). Increased levels of p53 protein were observed in the hippocampal and cerebellar neurons of these mice. These mice, therefore, model some aspects of AS in humans. Neonatal hyperactivity, which is characteristic of AS, is weakly supportive of the conflict hypothesis, particularly in view of the opposite phenotype of the closely linked and reciprocally imprinted syndrome PWS (see Ndn and Pwcr1 above).

H19 (mouse distal Chr7, human Chr11p15.5)

The paternal allele of H19 is silenced; maternal allele active in a wide array of mesenchymal and epithelial tissues, both in mice and humans (Bartolomei et al., 1991; Zhang and Tycko, 1992). H19 encodes an abundant spliced and polyadenylated RNA that has a conserved exon–intron structure among mammals, but that lacks conserved open reading frames. H19 RNA accumulates in the cytoplasm (Brannan et al., 1990; Li et al., 1998). H19 and Igf2 (see above) are inversely regulated in cis through shared insulator and enhancer sequences (Leighton et al., 1995a; Bell and Felsenfeld, 2000; Hark et al., 2000). Accordingly, net expression of H19 generally parallels that of Igf2 (Pachnis et al., 1988; Ohlsson et al., 1994; Leighton et al., 1995b; Drewell et al., 2000), but the two genes are expressed from opposite alleles. Somatic overgrowth was observed in knockout mice lacking the H19 gene and its immediate upstream sequences (Leighton et al., 1995b). Igf2 showed loss of imprinting (i.e., activation of the maternal allele) in these mice and crossing of H19-minus females with Igf2-minus males abrogated the overgrowth of the conceptuses (Leighton et al., 1995). This implied that the biological function of the H19 locus, with its 5′ flanking DNA, may be restricted to controlling Igf2 in cis. However, in another study, deletion of a smaller sequence element upstream of H19 led to activation of the paternal H19 allele, and these mice were growth-retarded, without a measurable effect on Igf2 expression (Drewell et al., 2000). Expression of H19 RNA also inhibited soft agar growth and tumorigenicity in some, but not all, transfected cancer cells (Hao et al., 1993; Isfort et al., 1997), so a possible trans-acting function of the RNA deserves future scrutiny. H19 is silenced, either by de novo DNA hypermethylation of the maternal allele and immediate upstream DNA, or by loss of this allele, in most Wilms' tumors and in several other embryonal neoplasms (Moulton et al., 1994; Steenman et al., 1994; Liu et al., 1995; Taniguchi et al., 1995; Casola et al., 1997; Fukuzawa et al., 1999). Biallelic methylation of H19, associated with relaxation of IGF2 imprinting, is also observed in a specific subset of cases of classical BWS (Catchpoole et al., 1997) with a high predisposition to Wilms' tumor (Engel et al., 2000; Weksberg et al., 2001; DeBaun et al., 2002), and in a related BWS-like overgrowth disorder, also with a predisposition to Wilms' tumor (Morison et al., 1996). In summary, while the physiological role of H19 RNA is not yet known, by several criteria, the H19 locus satisfies the predictions of the conflict hypothesis.

Ascl2, aka Mash2, HASH2 (mouse distal Chr7, human Chr11p15.5)

The paternal allele of Ascl2 is silenced; maternal allele active in mice and, probably, in humans (Guillemot et al., 1995; Alders et al., 1997; Caspary et al., 1998; Tanaka et al., 1999). Expression and functional imprinting of Ascl2 is only observed in placenta (Guillemot et al., 1994, 1995). This gene encodes a transcription factor in the basic helix-loop-helix (bHLH) family (Janatpour et al., 1999). In the placenta, Ascl2 acts in a cell-autonomous manner to inhibit the differentiation of proliferating spongiotrophoblast into giant cells (Guillemot et al., 1994; Tanaka et al., 1997). Placentas of Ascl2-knockout mice have a relative increase in giant cells and a deficiency of spongiotrophoblast, and the conceptuses fail at mid-gestation (Guillemot et al., 1994; Tanaka et al., 1997). These findings indicate an important role for this gene in placental development, but are difficult to interpret in terms of the conflict hypothesis.

Cd81 aka Tapa1 (mouse distal Chr7, human Chr11p15.5)

The paternal allele of Cd81 is weakly repressed; maternal allele active, only in extraembryonic tissues (Caspary et al., 1998; Dao et al., 1999). This gene is expressed biallelically in other tissues. Tapa1/Cd81 encodes a tetraspan membrane protein, expressed by many different tissues (Maecker et al., 1997). There are negative effects on cellular proliferation after antibody-mediated Cd81 crosslinking on the surface of lymphocytes. Knockout mice lacking Cd81 are viable, and also show alterations in lymphocyte proliferation in response to mitogens (Maecker and Levy, 1997; Miyazaki et al., 1997). There is no evidence of an imprinted phenotype in these mice. The data for Cd81 are, therefore, uninformative with regard to the conflict hypothesis.

Kcnq1, aka Kvlqt1 (mouse distal Chr7, human Chr11p15.5)

The paternal allele of Kcnq1 is repressed; maternal allele active in multiple fetal tissues of mice and humans (Lee et al., 1997a; Gould and Pfeifer, 1998; Jiang et al., 1998b); but there is biallelic expression in adult tissues, including heart (Gould and Pfeifer, 1998). Kcnq1 encodes a voltage-sensitive potassium channel. When KvLQT1/KCNQ1 protein is co-expressed with KCNE1 or related proteins, the two subunits heterodimerize to form a functional potassium channel (reviewed in Robbins, 2001). Missense mutations in the KCNQ1 gene cause the long QT syndrome (LQT1) of cardiac arrhythmias, which because of lack of imprinting of this gene in heart is transmitted as a dominant trait, without a parent-of-origin effect. In a recessive variant of this syndrome, JLNS, there is also deafness due to deficient endolymphatic circulation, reviewed in (Wang et al., 1998). In two studies, knockout mice lacking Kcnq1 exhibited deafness and loss of balance, with disruption of the inner ear anatomy due to insufficient endolymph (Lee et al., 2000; Casimiro et al., 2001). This was the most obvious abnormality, but other features were observed. In one of the two studies, homozygous mice also displayed threefold enlargement by weight of the stomach resulting from mucous neck cell hyperplasia, but electrocardiographic changes were not observed (Lee et al., 2000). The other study did not describe the gastric changes, but showed non-lethal electrocardiographic abnormalities in the mutant mice (Casimiro et al., 2001). How the data for KvLQT1/KCNQ1 pertain to the conflict hypothesis is not clear. Although the gastric hyperplasia conforms to the prediction that a maternally-expressed imprinted gene should inhibit tissue growth, no evidence was given for a parent-of-origin effect on this phenotype.

Cdkn1c aka p57Kip2 (mouse distal Chr7, human Chr11p15.5)

The paternal allele of Cdkn1c is repressed; maternal allele active in multiple fetal tissues of mice and humans, but the imprint is “leaky” (partial repression) in human tissues (Hatada and Mukai, 1995; Chung et al., 1996; Hatada et al., 1996a; Matsuoka et al., 1996). The p57Kip2 gene encodes a cyclin-dependent kinase inhibitor, with tissue-specific variations in its expression (Matsuoka et al., 1995). The Chr11p15.5 imprinted domain coincides with two human disease loci: the somatic overgrowth and cancer predisposition disorder BWS and a Wilms' tumor locus (WT2). The CDKN1C gene has, therefore, been evaluated as a candidate for both phenotypes. Maternally-transmitted mutations in CDKN1C have been found in a small subset (∼10%) of cases of BWS, indicating that this gene can cause this syndrome (Hatada et al., 1996b, 1997; Lee et al., 1997b; O'Keefe et al., 1997; Lam et al., 1999). However, such mutations are rare or absent in Wilms' tumors and other BWS-associated malignancies such as hepatoblastoma (O'Keefe et al., 1997; Hartmann et al., 2000). Wilms' tumors show reduced, but not absent, expression of p57Kip2 mRNA relative to normal fetal kidneys (Orlow et al., 1996; Overall et al., 1996; Taniguchi et al., 1997). Consistent with all of these observations, a genotype–phenotype correlation exists in BWS, such that individuals with CDKN1C mutations do not have a high risk of Wilms' tumor (Engel et al., 2000; Weksberg et al., 2001; DeBaun et al., 2002). Expression of p57 protein in transfected tumor cells blocks cell growth (O'Keefe et al., 1997; Tsugu et al., 2000). Developmental abnormalities, paradoxically without net overgrowth, are seen in p57Kip2 knockout mouse embryos (Yan et al., 1997; Zhang et al., 1997). The lack of overgrowth is accounted for by an apoptotic cellular response to loss of a cell cycle checkpoint (Yan et al., 1997; Takahashi et al., 2000). However, mouse conceptuses with a double knockout of p57Kip2 and H19, with loss of imprinting of Igf2, showed striking placental overgrowth (Caspary et al., 1999). Overall, the findings for p57Kip2 with respect to cell proliferation in vivo and in vitro are compatible with the predictions of the conflict model.

Tssc3, aka Ipl (mouse distal Chr7, human Chr11p15.5)

The paternal allele of Tssc3 is strongly repressed; maternal allele active in placenta and yolk sac of mice and in human placenta. There is a weaker allelic bias in other fetal and adult organs, although fetal liver shows a strong functional imprint (Qian et al., 1997; Lee and Feinberg, 1998). Tssc3/Ipl encodes a small cytoplasmic protein containing a pleckstrin-homology domain (PH-domain) (Frank et al., 1999). In general, PH-domain-containing proteins mediate cellular processes downstream of phosphatidyl-inositol polyphosphate second messengers (Ferguson et al., 2000). The pathway in which the Ipl protein functions is not yet known, but placental overgrowth, without a corresponding fetal overgrowth is observed in Ipl KO mice (Frank et al., 2002). These data support the conflict hypothesis from the standpoint of maternal alleles, i.e., Ipl functions to restrain placental growth, and thereby, moderates demands on maternal resources. But the evidence seems to be neutral with regard to the paternal allele (no fetal overgrowth in the Ipl KO, hence no obvious benefit to paternal alleles from silencing Ipl). A similar situation pertains for the maternally expressed gene Esx1 (see below).

Esx1, ESX1L (mouse X, human X)

The paternal allele of Esx1 is repressed; maternal allele active, in mouse placenta (Li and Behringer, 1998). This is a special type of imprinting, unique to X-chromosome genes. As mentioned above (see Xist gene), the inactivation of the entire X-chromosome shows a parent-of-origin dependence in the trophectoderm of mice, such that the paternal X is preferentially inactivated in placenta. The imprinting of Esx1 most likely results from this phenomenon. Esx1 encodes a paired-type homeobox protein, which is highly expressed only in placenta and male germ cells (Li et al., 1997). An Esx-1 orthologue, ESX1L, is also X-linked, but because of the lack of a parent-of-origin effect on X-inactivation in humans, it is not likely to be imprinted (Fohn and Behringer, 2001). Esx1 exerts a negative influence on growth of the placenta at early stages of development (placentas of Esx1-knockout conceptuses are larger). But this gene has a net positive effect on fetal growth at later stages, in the sense that fetal size reduced in the knockout, probably due to placental pathology (Li and Behringer, 1998). So, the data for Esx1 support the conflict hypothesis from the standpoint of maternal, but not paternal, alleles (see also, Tssc3/Ipl, above).


More than half the members of the growing set of known imprinted genes have yet to be critically examined for their functions in living mice and/or humans. Imprinted genes for which these in vivo data are pending are listed comprehensively at the Genomic Imprinting Website (


The characteristics of the imprinted genes that we have covered here, together with information from additional genes annotated at the Genomic Imprinting Website (, make it clear that regulation of gene dosage by imprinting affects a diverse array of biochemical pathways. These range from protein ubiquitination to growth factor clearance to transcriptional modulation, as well as the production of several classes of non-translated RNAs. But progress in assessing the functions of these genes in vivo has also highlighted some unifying themes. As shown in Figure 1, there is now suggestive evidence to support a predominant involvement of this class of genes in controlling growth and neurobehavioral traits. By considering the effects of offspring behavior on maternal provision of nutrition and other forms of care, these two general functions can, in principle, be unified by the hypothesis of intergenomic conflict. Given the complexities of interpreting the relevant phenotypes, a definitive proof of this model, or disproof by counter-example, may be elusive (Solter, 1998). But the intriguing data obtained to this point argue that systematic attempts to catalogue the biological functions of all imprinted genes will be well rewarded. The Genomic Imprinting Website ( is a dynamic information base that will be valuable in following the success or failure of the conflict theory as additional data become available.


This work was supported by grants from the NIH and from the Human Frontiers Science Project to B.T., and from the Cancer Society of New Zealand to I.M.M. The authors also acknowledge valuable insights from two additional Web-based resources concerned with genomic imprinting and, and to point interested readers to these two excellent sites.