X-inactivation: an epigenetic change that is heritable from cell to daughter cell, but not from parent to child
In humans, females normally have two X chromosomes (46,XX) while males have one X and one Y (46,XY). The fact that normal healthy people can have different numbers of X and Y chromosomes requires explanation. Animals, including humans, do not readily tolerate having wrong numbers of chromosomes. Chromosomal aneuploidies (extra or missing chromosomes, as in Down syndrome) have severe, usually lethal, consequences. Nevertheless, in organisms with an XX/XY sex determination system, males and females must be able to develop normally despite having different sex chromosomes. For the human Y chromosome, the solution is to carry very few genes. Some of these have counterparts on the X chromosome so that both XX and XY individuals have two copies, while many are related to male sexual function and so are dispensable in females.
The human X chromosome, on the other hand, carries many essential genes. Conceptuses that lack an X chromosome cannot survive, while the many X-linked dis eases show the importance of individual X-linked genes. Different organisms solve the problem of coping with either XX or XY chromosome constitutions in different ways. In male Drosophila flies, genes on the single X chromosome are transcribed at double the rate of those on the X chromosomes of females. Mammals, including humans, take a different approach: they use X-inactivation (sometimes called lyonization after its dis coverer, Dr Mary Lyon).
Early in embryogenesis each cell somehow counts its number of X chromosomes, and then permanently inactivates all X chromosomes except one in each somatic cell. At very early stages in development, both X chromosomes are active, but X-inactivation is initiated at the late blastula stage as cells begin to differentiate. Inactivated X chromosomes are still physically present, and on a standard mitotic karyotype they look entirely nor mal, but the inactive X fails to decondense after mitosis. It remains condensed throughout the cell cycle and most genes on the chromosome are permanently silenced in somatic cells. In interphase cells the inactive X may sometimes be seen under the micro scope as a Barr body or sex chromatin (Figure 1). Regardless of the karyotype, each somatic cell retains a single active X:
• XY males keep their single X active (no Barr body);
• XX females inactivate one X in each cell (one Barr body);
• Females with Turner syndrome (45,X) do not inactivate their X (no Barr body);
• Males with Klinefelter syndrome (47,XXY) inactivate one X (one Barr body);
• 47,XXX females inactivate two X chromosomes (two Barr bodies).
Fig1. Barr bodies. (A) A cell from a 46,XX female has one inactivated X chromosome and shows a single Barr body (arrow). (B) A cell from a rare 49,XXXXY male has three inactivated X chromosomes and shows three Barr bodies. (Courtesy of Malcolm Ferguson-Smith, University of Cambridge.)
A 46,XX cell may inactivate the maternal or the paternal X chromosome—it is a random choice, made independently by each cell—but whichever is chosen for inactivation, that same one is inactivated in all daughter cells (Figure 2A). The body of an adult female is thus a mosaic of cell clones, each clone retaining the pattern of X-inactivation that was established in its progenitor cell early in embryonic life. This can have implications for women who are heterozygous for an X-linked pathogenic loss-of-function mutation. A woman carrier of hemophilia A, for example, has one X chromosome with an intact Factor VIII gene and one with a nonfunctional copy. On average around half her cells will have the intact copy active. Factor VIII is a circulating protein and there is an averaging effect: she will have around half the normal level of the clotting fac tor, but that is normally sufficient for her to avoid clinical consequences. On the other hand, a woman heterozygous for the cell-autonomous condition X-linked ectodermal dysplasia (OMIM #305100) has some patches of skin with normal sweat glands, clonal progeny of an embryonic cell that inactivated the mutation-bearing X, and patches lacking sweat glands, derived from a cell that inactivated the normal X. Figure 2B shows a somewhat analogous example from the calico cat.
Fig2. X-inactivation. (A) In human embryos a randomly chosen X chromosome is inactivated in each cell of a 46,XX embryo, but once made, the choice is transmitted through all subsequent rounds of mitosis. (B) The calico (tortoiseshell and white) cat is heterozygous at an X-linked coat color locus. One allele specifies black coat color, the other orange. The different color patches reflect clones in which different X chromosomes are inactivated. The white patches are the result of an unrelated coat color gene. Calico cats are always female, apart from occasional XXY males. (Adapted from Migeon BR [1994] Trends Genet 10:230–235; PMID 8091502. With permission from Elsevier.)
X-linked severe combined immunodeficiency (X-SCID; OMIM #300400) shows another effect. X-SCID is caused by loss of function of the IL2RG gene. This X-linked gene is required for development of both B and T lymphocytes. Affected males have a lethal immunodeficiency. In heterozygous women, the descendants of any precursor cell that inactivated the normal X are unable to give rise to lymphocytes, and so all the lymphocytes she does have are progeny of cells that inactivated the mutation-bearing X. Thus X-inactivation in lymphocytes is 100% skewed, although other tissues show the normal random X-inactivation. Highly skewed X-inactivation in any tissue can be a pointer to heterozygosity for an X-linked condition. X-inactivation can affect a woman carrying a chromosomal translocation between one X chromosome and an autosome.
X-inactivation is an epigenetic process: the DNA sequence is unaltered by inactivation. Inactivation is stable through mitosis but not across the generations. During oogenesis the inactive X is reactivated and all memory of X-inactivation status is erased. The previously inactive X has the same chance as the previously active X of being passed on to any child, and in a daughter it has the same chance as the paternal X of being inactivated in any particular cell.
Initiating X-inactivation: the role of XIST
Inactivation is initiated at the 1 Mb X-inactivation center (XIC) at Xq13. The earliest observed event in XX cells is a transient pairing of the two XIC sequences. This is prob ably the mechanism by which the X chromosomes are counted, because counting is disrupted if the XIC sequences are deleted, duplicated, or translocated. XIC encodes a large noncoding RNA, XIST (X-inactivation-specific transcript), which is expressed only from the inactive X chromosome. The primary transcript undergoes splicing and polyadenylation to generate a 19 kb mature noncoding RNA. XIST is required to establish X-inactivation, but not to maintain it: in differentiated cells that have already undergone X-inactivation, loss of XIST does not cause reactivation.
Most detailed studies of the mechanism have been performed in the mouse, looking at the onset of random X-inactivation as pluripotent embryonic stem cells (ESCs) start to differentiate. This has revealed a complex set of noncoding RNAs that regulate Xist expression. However, the timing and initial mechanism of X-inactivation are very different in mice and humans (and, confusingly, papers do not always make clear whether they are talking about mice or humans). Little of the organization of the mouse Xic is conserved in humans. In the mouse, expression of Xist from the active X chromosome is silenced by an antisense transcript (Tsix) that alters the chromatin configuration at the Xist locus. There is a human TSIX gene, but compared to mouse Tsix it is truncated at the 5′ end so that it does not cover the XIST promoter and does not have the CpG island that is essential for X-inactivation function in the mouse. The papers by Chang & Brown (2010) and by Migeon (2016) (PMID 20211024 and 26805440, respectively; see Further Reading) give more detail.
One way or another, in both species the XIST/Xist RNA comes to coat the whole inactive X. Spreading depends on physical continuity of the chromosome. On an X chromosome that is split in two by an X-autosome translocation, inactivation is limited to the segment that includes XIC. It cannot jump to the detached portion. Exactly how this causes silencing is not clear. Numerous interacting partners of the XIST RNA have been identified but their roles, if any, are largely undefined (reviewed by Moindrot & Brockdorff [2016], PMID 26816113; see Further Reading). Directly or indirectly, XIST recruits repressive proteins including the PRC1 and PRC2 Polycomb complexes that organize the chromatin into a closed, transcriptionally inactive conformation. The chromatin of the inactive X comes to carry modifications typical of heterochromatin (H3K9me3, H3K27me3, unmethylated H3K4). In addition, many nucleosomes carry macro-H2A, a variant of histone H2A. A major stabilizer of inactivation is methylation of the normally unmethylated CpG islands at promoters of genes. However, this cannot be the sole force maintaining inactivation because around 40% of the genes have low-CpG promoters that lack the islands.
The action of the maintenance DNA methyltransferase DNMT1, perhaps together with some of the histone modifications, ensures that whichever X (maternal or paternal) is initially inactivated in a cell of the inner cell mass, that same X is inactivated in all daughter cells derived from it. As described above, females are thus mosaic for clones of cells expressing either the maternal or the paternal X chromosome.
Escaping X-inactivation
X-inactivation is not a blanket inactivation of the entire chromosome. The two pseudoautosomal regions escape inactivation, but even outside these regions X-inactivation is patchy. Around 15% of human X-linked genes escape inactivation, at least in some tissues and some individuals. X-inactivation is tighter in the mouse, where only 3–6% of genes consistently escape inactivation in cell lines. In one study, different hybrid cells containing independent copies of a human inactive X were used to investigate the transcription of 612 X-linked genes: 458 of the genes were inactivated in all or most of the cell lines, but 94 were expressed. The remaining 60 genes showed a variable pattern of expression in different cell lines (Figure 3). Despite the general requirement for physical continuity, the spreading X-inactivation is evidently able to jump over these escape genes and continue its progress. Some of the genes that escape inactivation are genes that have a functional counterpart on the Y chromosome, for which there would be no need for dosage compensation. Many, however, have no such counterpart.
Fig3. Genes that escape X-inactivation. Columns in the rectangle show results of systematic reverse transcription-PCR tests for expression of X-linked genes in nine independent somatic cell hybrids. Each hybrid contained a single inactive human X chromosome. Blue bars identify genes that were expressed from the inactive X chromosome in a particular hybrid, and yellow bars mark genes that were not expressed. Many genes, scattered all along the X chromosome, escape inactivation in one or more of the hybrids. Some genes, such as XIST, escape inactivation in all nine hybrids, whereas others show a more patchy pattern of inactivation. cen, centromere. (Reprinted from Carrel L & Willard HF [2005] Nature 434:400–404; PMID 15772666. With permission from Springer Nature. Copyright © 2005.)
At imprinted loci, expression depends on the parental origin
The whole of Mendelian pedigree interpretation is based on the premise that the parental origin of a gene is irrelevant: a heterozygous person is the same, regardless whether the mutated allele came from their father or their mother. However, for about 100 human genes this is not true. These genes retain a memory—an imprint—of their parental origin. The imprint may persist through all the mitoses that generate the whole adult body (although with some genes only certain tissues retain the imprint). In all cases the imprint must be erased in the germ line and replaced by one appropriate to the sex of the person. A man may receive a gene with a maternal imprint from his mother, but if he passes it on to his child it must then carry a paternal imprint (Figure 4). Thus imprinting is a classic reversible epigenetic process.
Fig4. Imprinting is epigenetic and reversible. A man might receive a gene carrying a maternal imprint from his mother. The imprint persists in his somatic cells, but if he passes that gene on to a child, it will then carry a paternal imprint.
Early evidence that parental genomes are not wholly equivalent came from the observation that zygotes with two maternal or two paternal genomes develop abnormally, as ovarian teratomas or hydatidiform moles, respectively. Hydatidiform moles are abnormal conceptuses that lack an embryo and consist of just hypertrophic extra-embryonic membranes; for teratomas. In mice, chromosomal manipulations make it possible to produce animals that have the correct number of chromosomes, but where both members of one chosen pair come from the same parent (uniparental disomy, UPD). Systematic exploration showed that for some chromosomes, although not others, UPD is pathogenic. The abnormal phenotypes are sometimes complementary for different parental origins; for example, overgrowth in paternal UPD and growth retardation in maternal UPD. For some chromosomes, UPD is lethal. These effects are caused by UPD for small numbers of specific genes on the relevant chromosome, rather than by the whole chromosome. Further work has identified about 100 imprinted loci on 11 of the mouse chromosomes.
In humans, a similar systematic investigation is not possible, but genotyping occasionally reveals UPD in patients or in healthy individuals, thus highlighting chromosomal regions where UPD matters, and others where it does not. The mechanism that produces most cases of human UPD, trisomy rescue. Some microdeletion or microduplication syndromes show parent-of-origin effects, for example Prader–Willi and Angelman syndromes with a microdeletion of 15q11q13, and Beckwith–Wiedemann syndrome (BWS) with various abnormalities of 11p15 (see below). Many of the mouse imprinted genes are now known also to be imprinted in humans, although the correspondence is not perfect. The Otago University catalog of parent-of origin effects (http://igc.otago.ac.nz/home.html) gives a comprehensive list of definitely or possibly imprinted genes in humans and other species.
To confirm imprinting of a gene, it is necessary to identify an individual who is heterozygous for a sequence variant present in the mature mRNA. Messenger RNA from different tissues can then be checked for monoallelic or biallelic expression, and the origin of each allele can be determined by typing the parents. Random monoallelic expression is surprisingly common, so for a claim of imprinting to be convincing it should be based on consistent data from many independent samples. In many cases imprinting is partial, with one parental allele expressed at a higher level than the other; often it is confined to certain tissues or to certain stages of development. In humans, imprinted genes have been identified on chromosomes 6q24–q26, 7q21–q22, 7q32 (and maybe 7p12), 11p13, 11p15, 14q32, 15q11–q13, 19q13, and 20q13.
Mechanisms underlying imprinting
Examining imprinted regions, one finds short sequences where the DNA is differentially methylated on the paternal and maternal chromosomes, and often noncoding antisense transcripts (Table 1). The imprint is not a special mark present in gametes; rather it is a failure to remove methylation from specific sequences in the wave of genome-wide demethylation in the early zygote (Figure 5). In most cases the signal is methylation of a gene-associated CpG on the maternal chromosome; the few examples of paternal specific methylation affect CpGs in intergenic DNA.
Table1. EXAMPLES OF HUMAN IMPRINTED CLUSTERS SHOWING GENES AND NONCODING RNAS
Fig5. Changes in DNA methylation during mammalian development. Drastic and often tissue specific changes in overall methylation accompany gametogenesis and early embryonic development. Note the breaks (slashes) in the developmental lines. PGCs, primordial germ cells.
One mechanism by which differential methylation can affect gene expression is illustrated by the IGF2/H19 cluster. As previously mentioned, H19 is a noncoding RNA that also includes the gene for microRNA miR675. IGF2 (insulin-like growth factor 2) is an important fetal growth factor. The basic mechanism influencing gene expression is competition for an enhancer, controlled by a differentially methylated imprinting control region (ICR; Figure 6). The DLK1/MEG3 cluster at 14q32 functions in a similar way.
Fig6. Paternal methylation of an intergenic insulator governs competition for an enhancer. Binding of the CTCF insulator protein to a differentially methylated region determines the outcome of the competition. On the maternal chromosome, the imprinting control region (ICR) is unmethylated, allowing it to bind CTCF (beige oval). This prevents the IGF2 gene from accessing the enhancers, and so the enhancers drive H19 expression. On the paternal chromosome, methylation of the imprinting control region prevents binding of CTCF, thereby allowing IGF2 to outcompete H19 for access to the enhancers. (Adapted from Wallace JA & Felsenfeld G [2007] Curr Opin Genet Dev 17:400–407; PMID 17913488. With permission from Elsevier.)
An alternative mechanism is exemplified by the second imprinted domain on 11p15 and by the Prader–Willi/Angelman imprinted region on 15q11 (Figure 7). Again there is differential methylation, but here maternal-specific methylation of a CpG island prevents expression of a long noncoding antisense RNA. The paternal-specific RNA shuts off expression of nearby genes. At least for some of those genes, the effect seems to be due simply to the fact of transcription of the noncoding RNA, not to any property of the transcript. Probably there are other interactions at work as well, because in both cases there is imprinted expression of neighboring genes that do not overlap the antisense RNA and there is a great deal of alternative splicing.
Fig7. Maternally methylated promoters prevent transcription of long noncoding antisense RNAs. E Maternally expressed genes are shown in red, paternally expressed in blue, and biallelically expressed in dark gray (there are some reports that ATP10A is only maternally expressed). Nonexpressed genes are shown in pale gray. Colored boxes with pink infill are imprinted only in extra embryonic tissue. Blue wavy arrows show long noncoding RNAs. Asterisks show the differentially methylated imprinting control centers. tel, telomere; cen, centromere. (A, adapted from Pauler FM et al. [2012] Curr Opin Genet Dev 22:283–289; PMID 22386265. With permission from Elsevier.)
Clinically, abnormalities in these regions manifest as developmental syndromes with parent-of-origin effects. On 15q11, lack of a maternal UBE3A product causes Angelman syndrome (OMIM #105830). The lack can be due to a microdeletion, pater nal UPD, or a point mutation in UBE3A. Lack of the paternal SNHG14 RNA because of a microdeletion or maternal UPD causes Prader–Willi syndrome (OMIM #176270). The immediate cause is probably lack of the SNORD116 snoRNA cluster encoded in an intron of the SNHG14 RNA. With both syndromes, cases due to microdeletions and cases due to UPD are indistinguishable, suggesting that overexpression of the relevant gene has no adverse effects. This is different from the situation on 11p15. Overexpression of IGF2 or underexpression of CDKN1C cause BWS (OMIM #130650), an overgrowth condition, while the reciprocal underexpression of IGF2 or overexpression of CDKN1C cause Silver–Russell syndrome (SRS, OMIM #180860), in which there is growth retardation. A confusing variety of molecular events including UPD, deletions, duplications, or defective methylation can cause these effects. With all these syndromes we have given a rather simplified description of causes here; interested readers should consult OMIM for more details and references.
The regions described here show features common to several imprinted regions:
• Imprinting is controlled by one or more small regions where the DNA is differentially methylated on the maternal and paternal chromosomes;
• Imprinted regions are often complex, with clusters of genes, some paternally imprinted and others maternally imprinted. There may also be nonimprinted (biallelically expressed) genes in the cluster. Imprinting of some genes can be tissue-specific;
• There are often overlapping and oppositely imprinted sense and antisense transcripts, only one of which can be expressed at any one time. This points to some sort of flip-flop mechanism, in which transcription from one DNA strand prevents transcription from the opposite strand. The actual antisense transcript often seems to have no function of its own; what matters is the fact of transcription.
When a pathogenic variant in an imprinted gene is compatible with survival and fertility, the resulting pedigrees show unusual features (Figure 8). The variant may be inherited from a parent of either sex, and may affect persons of either sex, but the resulting phenotype will be apparent only when it is inherited from one sex of parent. So, if an imprinted gene is expressed only from the paternal chromosome, persons inheriting a pathogenic variant of the gene would show the resulting phenotype only if they inherited it from their father. A man who inherited it from his mother would be phenotypically normal, but would be at risk of having affected children.
Fig8. Pedigrees of conditions with imprinted gene expression. (A) In this family, autosomal dominant glomus tumors (OMIM #168000) manifest only when the gene is inherited from the father. (B) In this family, autosomal dominant Beckwith-Wiedemann syndrome (OMIM #130650) manifests only when the gene is inherited from the mother. (A, family reported in Heutink P et al. [1992] Hum Mol Genet 1:7–10; PMID 1301144. With permission from Oxford University Press; B, family reported in Viljoen D & Ramesar R [1992] J Med Genet 29:221–225; PMID 1583639. With permission from the BMJ Publishing Group Ltd.)
One might reasonably ask what is the function of such complicated mechanisms? One popular theory is based on a conflict of evolutionary interest between fathers and mothers. Selfish gene theory suggests that paternal genes might be best propagated by ensuring that offspring are born as robust as possible, even at the expense of the mother—a man can father children by many different women. Maternal genes, in contrast, are best propagated if the mother remains capable of further pregnancies. So, the theory goes, paternal genes program the fetus to extract nutrients at the greatest possible rate from the mother through the placenta, whereas maternal genes act to limit the depredations of a parasitic fetus. This does fit many imprinted loci, for example those at 11p15 described above, or those responsible for the abnormal development of conceptuses with two paternal or two maternal genomes. Hydatidiform moles, with two paternal genomes, are a mass of extra-embryonic membranes with no embryo, while ovarian teratomas, with two maternal genomes, are a disordered mass of fetal tissues with no membranes. It is less clear why imprinting should often be variable between individuals or tissues, or only involve mild allelic imbalances. Perhaps such variability suggests that one should not always try too hard to find a clear function in every case. Gene expression in single cells is often rather chaotic, with random monoallelic expression; maybe some of the weaker or more variable imprinting effects reflect a similar lack of tight control. It is amazing that cells work at all; natural selection ensures that they work well enough, but that does not require perfection and absolute precision in every detail.
Transgenerational epigenetic memory is a controversial and poorly understood subject
Epigenetic modifications are remembered through mitosis, by definition, but not normally through meiosis—consider X-inactivation for example. Reported exceptions need careful analysis: there are many possible reasons for parent-offspring resemblances. No fancy mechanism is required to explain why ten generations of a family all speak Chinese. Intrauterine environment can be important. Rats nurtured by a stressed mother are more likely themselves to be stressed because of intrauterine hormonal effects. Epigenetic marks made by a fetus in response to its environment affect metabolism and health in later life so, for example, underweight babies born to starved mothers after the 1944–1945 Dutch “hunger winter” developed into adults with increased adiposity. Given that a fetus already has the primordial germ cells that will produce the second generation, even maternal effects on grandchildren’s health have many possible interpretations.
Paternal effects are more interesting. For example, in the Överkalix region of north ern Sweden, the risk of cardiovascular and diabetes-related death of individuals could be related to increased food supply during the prepubertal growth period of their grand fathers. These and similar studies are reviewed by Pembrey et al. (2014) (PMID 25062846; see Further Reading). Mice provide a tractable experimental system for investigating such effects, particularly since the involvement of the fathers can be limited to pro viding sperm for in vitro fertilization. For example, Chen and colleagues (2016) (PMID 26721680; see Further Reading) fed male mice on either a normal diet (ND) or a high-fat diet (HFD). As expected, the HFD mice became obese, glucose intolerant, and insulin resistant. Sperm heads of ND and HFD mice were injected into normal mouse oocytes and the embryos transferred into surrogate mothers. All male offspring from both groups were fed a normal diet. Although there were no obvious differences in body weight over 16 weeks, offspring produced by the HFD-group sperm had developed severely impaired glucose tolerance and insulin resistance by 15 weeks. Other groups have reported simi lar findings and several studies have demonstrated second-generation paternal effects (summarized by Patti [2013], PMID 23435955; see Further Reading).
The causative factor in the experiment of Chen and colleagues must have been in the sperm heads, so they set out to identify it. They found they could replicate the glucose intolerance, though not the insulin resistance, by injecting normal zygotes with purified RNA from sperm heads of HFD, but not ND, mice. Further fractionation showed the effect was due to small (30–34 nt) RNA fragments derived from transfer RNAs. Work by other groups has confirmed the role of RNA fragments in similar effects.
A second area where RNA is involved concerns the strange phenomenon of paramutation. This describes the situation when an organism shows the phenotype of a mutation that was present in an ancestor, despite not having inherited any of the mutant DNA. A French group studied mice with a loss-of-function allele at the Kit locus. Heterozygotes have white spotting. When these heterozygotes were intercrossed or backcrossed with wild-type homozygotes, a proportion of the homozygous wild type offspring showed the white spotting characteristic of heterozygotes, despite having the wild-type genotype (Figure 9). The effect persisted for a few generations, but with diminishing intensity. It could be replicated by microinjecting sperm RNA from a heterozygous mouse into the pronuclei of normal fertilized eggs. Surprisingly, expression of the paramutation phenotype required the DNMT2 RNA methyltransferase. This and some related transgenerational effects depend in some way on transposons, as only alleles with transposon insertions show the effects. It is intriguing that transposons are silenced by an RNA-dependent mechanism (piRNAs).
In summary, transgenerational epigenetic effects are a confusing topic. Such effects undoubtedly occur, but whether they are isolated oddities or part of a whole major, underappreciated substratum of genetics is hard to say. There are many individual reports, but they do not fit easily into a single narrative. The wide-ranging review by Miska & Ferguson-Smith (2016) (PMID 27846492; see Further Reading) attempts to provide a conceptual framework.
Fig9. Paramutation at the Kit locus in mice. Mice heterozygous for a Kit mutation (+/−) show white spotting (seen here on the tail). When heterozygous mice were crossed, most (24 out of 27) of the genotypically wild-type (+/+) offspring nevertheless showed the white spotting characteristic of heterozygous animals. The wild-type allele inherited from the heterozygous parent has been somehow changed (paramutated, asterisk). The change is unstable: when +/* mice were crossed with wild-type animals, fewer than the predicted 50% of offspring showed the expected +/* phenotype. (Data from Rassoulzadegan M et al. [2006] Nature 441:469–474; PMID 16724059.)
