Regulation of X chromosome inactivation by the X inactivation centre

Transcription

1 Regulation of X chromosome inactivation by the X inactivation centre Sandrine Augui*, Elphège P. Nora* and Edith Heard Abstract X-chromosome inactivation (XCI) ensures dosage compensation in mammals and is a paradigm for allele-specific gene expression on a chromosome-wide scale. Important insights have been made into the developmental dynamics of this process. Recent studies have identified several cis- and trans-acting factors that regulate the initiation of XCI via the X inactivation centre. Such studies have shed light on the relationship between XCI and pluripotency. They have also revealed the existence of dosage-dependent activators that trigger XCI when more than one X chromosome is present, as well as possible mechanisms underlying the monoallelic regulation of this process. The recent discovery of the plasticity of the inactive state during early development, or during cloning, and induced pluripotency have also contributed to the X chromosome becoming a gold standard in reprogramming studies. Homogametic and heterogametic sexes In species with sexual dimorphism, the sex that can produce two different types of gametes (X and Y or Z and W) is called heterogametic, whereas the sex that can produce only one type of gamete (X or Z) is called homogametic. Imprinted Epigenetic marking of a locus on the basis of its parental origin, which can result in differential expression of the paternal and maternal alleles in specific tissues or developmental stages. Mammalian Developmental Epigenetics Group, Unit of Genetics and Developmental Biology, Institut Curie, CNRS UMR3215, INSERM U934, Paris F 75248, France. *These authors contributed equally to this work. Correspondence to E.H. Edith.Heard@curie.fr doi: /nrg2987 Sex chromosome dimorphism leads to a genetic imbalance between the homogametic and heterogametic sexes, which mammals compensate for by inactivating one of the two X chromosomes during female development. Although this chromosome-wide silencing process was originally described more than 50 years ago (TIMELINE), the underlying molecular mechanisms remain poorly understood. One of the most intriguing aspects of X chromosome inactivation (XCI) is that two homologous X chromosomes are differently treated within the same nucleus. How the inactive state is set up and faithfully transmitted through cell division remains a central question for which answers are only now beginning to emerge. This Review will focus on the recent progress that has been made in our understanding of the initiation of XCI, as well as the reversibility of the inactive state during specific stages of development and in the context of reprogramming experiments. In mice, which have been the favoured model for XCI studies, there are two waves of XCI, the first being imprinted (paternal XCI) and the second random. Imprinted inactivation of the paternal X chromosome (Xp) is initiated shortly after fertilization. This silent state is maintained in extra-embryonic tissues but lost in the inner cell mass (ICM), which gives rise to the embryo proper. Shortly after this, random inactivation of either the maternal X chromosome (Xm) or the Xp is initiated in the cells of the ICM. In vitro differentiation of mouse embryonic stem cells (ESCs), which are derived from the ICM and have two active X chromosomes, is accompanied by random XCI and has been extensively used to dissect the early events underlying this process. As such, the regulation of random XCI is more thoroughly understood and is the main focus of the Review, although imprinted XCI is also discussed. A growing number of new molecular players have been implicated in XCI over recent years. How they function together to control XCI and how this fits in with or challenges the original views of the process remains largely unclear. The aim of this Review is to examine the role of these recently identified molecular players in the context of the initial historical notions underlying the process of XCI; that is, the concepts of counting, choice and sensing/competence (BOX 1). We will focus primarily on the X inactivation centre (Xic) and the key non-coding X-inactivation specific transcript (Xist) it produces, which represents the trigger for chromosome-wide silencing. We first explain briefly how the Xic was functionally and physically identified. We then describe how Xist underlies some, but not all, of the functions attributed to the Xic and review our current knowledge on the increasingly complex regulatory network controlling Xist expression. Finally, we discuss random and imprinted XCI in the context of mouse development and the recent insights that XCI has brought into reprogramming processes. NATURE REVIEWS GENETICS VOLUME 12 JUNE

2 Timeline Landmarks in our understanding of the initiation of random XCI Discovery of a dense structure in female somatic nuclei called the Barr body 140 Based on phenotypic variegation in the coat colours of heterozygous female mice, Lyon proposed that one of the two X chromosomes is stably inactivated in female cells 142 Identification of an X controlling element (Xce), which induces a skew in choice of the Xi 40 Discovery of the Xist/XIST gene as a candidate for the Xic 6 8 Demonstration that large single-copy Xist transgenes are insufficient for full Xic functions during random XCI 34 ( ) Discovery of numerous Xist molecular regulators (see main text) The Barr body is proposed to be an inactive X chromosome (Xi) 141 ( ) Lyon, Russell and Grumbach propose that inactivation spreads from a unique locus (the X inactivation centre, (Xic)) located on the X chromosome ,143 ( ) Definition of the Xic and its functions 1,2 ( ) Demonstration that Xist is essential for initiation of XCI in mice 10,11 and that multicopy Xist transgenes can induce XCI to some extent ,144 Identification of the Xist antisense unit, Tsix 56,145,146 Demonstration that Xist RNA is sufficient to initiate cis-inactivation 12 Polycomb group proteins (PcG proteins). A class of proteins originally described in Drosophila melanogaster that form large complexes and maintain the stable and heritable repression of several genes throughout development. Trithorax group proteins (TrxG proteins). A class of proteins originally described in Drosophila melanogaster that form large complexes and maintain the stable and heritable expression of several genes throughout development. The X inactivation centre Early studies of XCI patterns in mouse embryos or embryonic cells that carried translocated or truncated X chromosomes revealed the existence of a single X linked locus, the Xic, that needs to be physically linked to a chromosome to trigger its inactivation 1 (FIG. 1). Random XCI is only triggered in cells with at least two Xic-bearing chromosomes 2, suggesting that the two copies of the Xic are able to potentiate each other in trans, a phenomenon that has been referred to as competence, or sensing 3 5 (BOX 1). In XX cells, either one of the two X chromosomes will be inactivated, a process known as choice (BOX 1). The autosomal ploidy of a cell (the number of sets of autosomes that is contains) also seems to affect the number of X chromosomes that will be inactivated, a phenomenon known as counting (BOX 1). The precise mechanisms underlying these processes are only now being unravelled and recent data suggest that they are highly interconnected, both genetically and molecularly. Xist RNA triggers cis-inactivation The Xic harbours the Xist gene 6 8 (FIG. 1B), which produces a non-coding RNA (ncrna) that is retained in the nucleus and that, in its spliced form, can coat the chromosome from which it is expressed 9. It is devoid of any significant ORF and is only expressed from the inactive X chromosome (Xi) in somatic cells. During both female mouse development and in vitro differentiation of female mouse ESCs, Xist is monoallelically upregulated. This upregulation is tightly correlated with the onset of XCI and precedes the initiation of silencing (FIG. 2). Deletions of Xist have demonstrated that it is necessary in cis to induce chromosome-wide silencing 10,11. Furthermore, inducible expression of Xist cdna transgenes on autosomes demonstrated that Xist RNA is sufficient to trigger cis-inactivation of the chromosome from which it is expressed during an early developmental time window 12. How exactly Xist RNA induces gene silencing still remains a mystery, but the highly conserved A repeat region of Xist is crucial for its silencing function, whereas other parts of the RNA ensure its cis-coating capacity 13,14. Expression of an Xist cdna lacking the A repeat region in differentiating mouse ESCs has revealed that the transcript can induce several chromatin modifications on the chromosome that it associates with, independently of transcriptional repression. These modifications include recruitment of Polycomb group proteins (PcG proteins), the histone variant macroh2a, the Trithorax group protein (TrxG protein) ASH2 like (ASH2L) and heterogeneous nuclear ribonucleoprotein U (hnrnpu; also known as SAFA) 12, Wild-type Xist RNA has also been shown to induce the spatial reorganization of the X chromosome, creating a repressed nuclear compartment that is depleted of the transcription machinery and into which genes are recruited when they are silenced Based on the above evidence, Xist activation clearly triggers the establishment of chromosome-wide silencing. Therefore, much of the research into the mechanisms of XCI initiation has focused on regulation of this particular gene and the ncrna it produces. However, an important observation from studies of Xist knockouts is that heterozygous Xist mutants are still able to initiate XCI from the wild-type X chromosome 10,11. Thus, Xist sequences alone cannot account for the competence function of the Xic, which means other elements must be responsible for female-specific (XX) Xist activation and XCI initiation. As discussed below, it is now clear that Xist s unique expression pattern is controlled by a complex interplay of long-range cis-acting elements and trans-acting factors. Xist regulation during random XCI How is female-specific, monoallelic Xist upregulation achieved and why does it only occur within a precise time window during development and differentiation? In the following sections we describe what is known about the different levels of control acting on Xist during random XCI (see FIG. 3A for a summary). Xist is expressed at very low levels in undifferentiated male and female ESCs, but becomes upregulated on one X chromosome upon differentiation of female cells. Although it is now clear that Xist is controlled mainly at the transcriptional 430 JUNE 2011 VOLUME 12

3 Box 1 Key concepts in X chromosome inactivation Before the discovery of the many molecular actors in X-chromosome inactivation (XCI), some key concepts relating to the steps necessary for inactivation to occur were proposed. Although theoretical, these notions became, to an extent, dogmatic over the years. However, these concepts are now being revised in the face of new molecular insights. Counting This refers to the process by which a cell determines its X/autosome (X/A) ratio in order to maintain only a single active X chromosome per diploid autosome set. It was first proposed by Lyon and Grumbach based on humans with abnormal numbers of X chromosomes 118,119. A normal XY male, or an XO female, shows no inactivation of the unique X chromosome, whereas XXX and even XXXX individuals display one active X chromosome and inactivation of all supernumerary X chromosomes Choice Refers to how one of the two X chromosomes is selected for inactivation. During random XCI, the probability that the paternal or the maternal X chromosome will be chosen for XCI is equal, unless mutations or polymorphisms are present within the X-inactivation centre (Xic) 41. The selection of one X chromosome for inactivation must somehow preclude the initiation of XCI of the other X chromosome and is thus a part of the trans-function of the Xic. Sensing/competence This describes a permissive state for XCI that occurs only when there is more than one X chromosome present in a cell. It must be noted that sensing/competence is implicit in the original concept of counting (as defined above) and involves both XX-recognition, as well as assessment of the X/A ratio. However, investigation of phenotypes of different Xic mutants has led to a distinction being made between the two concepts 4,5,47. Pluripotency factors A class of proteins that maintain pluripotency the capacity to give rise to a wide range of, but not all, cell lineages of stem cells. level 23, post-transcriptional maturation events may also participate. For example, recent studies have shown that deletion of the A repeat region of Xist prevents accumulation of the spliced form of Xist RNA during differentiation 24 and somehow disrupts the gene s correct upregulation during development 25. Repression of Xist in undifferentiated ESCs What accounts for the low expression level of Xist in undifferentiated ESCs? Several circumstantial lines of evidence have pointed to pluripotency factors as negative regulators of the XCI process (FIG. 3Ba). In mouse ESCs, inducible knockout of Nanog or Oct4 (also known as Pou5f1) leads to ectopic Xist upregulation and chromosome coating in a fraction of differentiating male ESCs 26. Another study reported that knockdown of Oct4 leads to Xist RNA accumulation on both X chromosomes in a fraction of differentiating female ESCs 27. The binding of OCT4, NANOG, SOX2, transcription factor 3 (TCF3; also known as TCFE2A) and the PR domain containing protein PRDM14 within the first intron of Xist 26,28,29 had led to the proposal that such factors might repress Xist expression, via this region, in undifferentiated ESCs. However, deletion of this intronic region of Xist was recently shown to have no impact on Xist repression in undifferentiated ESCs 30, although the chromosome with the deleted allele is mildly favoured for XCI upon differentiation. Furthermore, it has recently been shown, using reporter assays, that a construct containing just the Xist core promoter can be activated during female mouse ESC differentiation 23, and that OCT4, NANOG and SOX2 do not bind the Xist promoter 26,28,31. Therefore, these pluripotency factors probably control Xist activity indirectly via intermediate regulators. As we discuss later, both upregulation of RING finger protein 12 (RNF12; also known as RLIM) and Xic Xic homologous pairing events during differentiation may represent such intermediates 27,32. Female-specific activation of Xist What is the mechanism underlying the specific upregulation of Xist in cells with more than one X chromosome? Several lines of evidence point to the existence of long-range regulatory elements that are required for Xist s XX specific upregulation. First, female cells carrying a 58 kb deletion including Xist on one X chromosome can still initiate XCI on the wildtype X chromosome 33, implying that they can still sense their XX status and are still competent for XCI. Second, large 460 kb single-copy Xist transgenes in male ESCs are unable to trigger Xist during differentiation, either from the transgene or from the endogenous Xic 34. This implies that critical Xic sequences that are needed to render cells competent for XCI must be missing from these large DNA fragments (BOX 2). Thus, the sequences underlying XX specific Xist activation must lie some distance from the gene itself. In a quest to identify these sequences, investigation of the genomic neighbourhood of Xist led to the identification of at least three X linked loci that are possibly implicated in the activation of Xist during random XCI in female mouse ESCs. One is the X pairing region (Xpr), which lies kb 5 to Xist. Xpr is able to mediate homologous trans-interactions between the two Xic loci (known as pairing ) in female mouse ESCs before Xist activation. This ability, which is also shown by Xpr single-copy transgenes, was proposed to participate in female-specific Xist expression, as Xpr Xpr interactions do not normally occur in male cells 5. A recent report describing the unusual genomic instability of the Xpr region when present as a transgene in male cells 35 could be indicative of recombination pathways being involved in Xpr pairing. However, the mechanisms underlying Xpr pairing in female cells and the impact of this on Xist transactivation remain to be elucidated. A second locus that has clearly been shown to have a role in XX specific Xist activation is Rnf12, which lies approximately 500 kb 5 to Xist (FIG. 3Bb). This gene produces a trans-acting factor, RNF12, which has a ubiquitin ligase activity. Overexpression of RNF12 can induce Xist RNA coating of the single X chromosome in differentiating male mouse ESCs and of both X chromosomes in differentiating female mouse ESCs 32. Based on such observations, it has been proposed that RNF12 can activate Xist when present above a certain threshold. In mouse ESCs, this threshold is proposed to be reached only when two X chromosomes are active. How RNF12 activates Xist remains to be determined, but one possibility is that its ubiquitin ligase activity acts to degrade a repressor of Xist. Recently, RNF12 has been shown to be capable of activating the core promoter of Xist 30. Importantly, heterozygous deletion of Rnf12 delays, but does not prevent, XCI in female mouse ESCs 32,36, implying that additional Xist activation mechanisms, present in XX but not XY cells, must exist. NATURE REVIEWS GENETICS VOLUME 12 JUNE

4 Figure 1 The X inactivation centre. A The X-inactivation centre (Xic) has been defined as the minimum region both necessary and sufficient to trigger X-chromosome inactivation (XCI) 2. The existence of a unique locus controlling the initiation of XCI was first proposed in 1964, based on studies of individuals or cell lines with balanced X autosome translocations. Aa In normal female cells, there is random XCI such that there is an equal probability of either X chromosome undergoing inactivation. Ab In studies of the reciprocal translocation T(X;16)16H (also known as T16H, Searle s translocation) only one of the two translocation products was found to be inactivated, suggesting the existence of an X linked region (the Xic) that is required in cis for XCI to occur Note that 16 X is not found to be inactivated, which is due to secondary counter selection. Ac Surprisingly, when the same translocation is unbalanced there is no inactivation process at all, suggesting that in the absence of two Xics, a cell does not detect the presence of two X chromosomes 132. Ad Subsequent studies involving female embryonic cells where one of the two X chromosomes was truncated 2,132 confirmed this hypothesis. It was revealed that neither the truncated X chromosome (HD2 truncation) nor the intact X chromosome showed any sign of XCI based on cytological staining. This indicates that at least two Xics are required for a cell to initiate XCI. Ae By contrast, for a truncation that does not remove Xic, random XCI still takes place. B In addition to providing a functional definition of the Xic, these chromosomal rearrangements define the physical boundaries of the locus. In mice (shown), the minimum candidate region for the Xic has been defined, based on studies in developing mouse embryos or differentiating embryo-derived (EK) cells 133. The Xic lies between the T16H breakpoint 134,135 and the HD3 breakpoint 1,2, a region spanning 8 cm (10 20 Mb). Here, only the elements around Xist are shown. Some of these elements, such as the Xist antisense gene (Tsix) or RING finger protein 12 (Rnf12; also known as Rlim) gene are now known to be involved in Xist regulation. Xist and its antisense Tsix, as well as regulators of Tsix Xite (X-inactivation intergenic transcription element) and DXPas34 are shown at higher resolution under the Xic map. In humans (not shown), the XIC has been proposed to map between the T(X:14) and rea(x) breakpoints, a region spanning 700 kb 136,137. However, the human XIC has been defined through the analysis of X-inactivation status in somatic cells of patients with X chromosomal deletions or translocations, rather than in embryonic cells where XCI is actually initiated. Thus, it cannot be excluded that some of these rearrangements could have arisen after initiation of XCI. Cdx4, caudal X linked gene 4; Chic1, cysteine rich hydrophobic 1; Cnbp2, cellular nucleic acid binding protein 2; Ftx, five prime to Xist; Jpx, also known as Enox (expressed neighbour of Xist); Nap1l2, nucleosome assembly protein 1 like 2; Tsx, testis specific X linked; Xpct, X linked PEST-containing transporter. 432 JUNE 2011 VOLUME 12

5 Figure 2 The cycle of XCI in female mouse embryos and ESCs. a In mice, X-chromosome inactivation (XCI) begins at the four cell stage see the middle of the top part of this panel. Inactivation is initially imprinted, with preferential inactivation of the paternal X chromosome (Xp). Studies of parthenogenetic and gynogenetic embryos suggest that this imprint is maternal. This is because, in the presence of two maternal X chromosomes (Xm), there is no XCI until blastocyst formation, implying that the Xm cannot be inactivated prior to this, despite the presence of two X chromosomes 89,138. Furthermore, experiments performed with non-growing oocytes showed that an X chromosome from an immature oocyte (prior to establishment of imprinting) behaves like an Xp during embryogenesis and can undergo early X-inactivation 104. Once established, the Xp remains inactive in extra-embryonic tissues (trophectoderm and placenta) but is reactivated in the inner cell mass (ICM) of the blastocyst in pre-epiblast cells, which gives rise to the embryo. A second wave of inactivation then occurs in the ICM and randomly affects either the Xp or the Xm. The inactive state is then stably maintained and transmitted through cell divisions in the soma but the inactive X chromosome (Xi) is reactivated during the formation of the female germ line. Imprinted and random inactivation are both Xist-dependent and both seem to involve RING finger protein 12 (RNF12; also known as RLIM). A maternal pool of RNF12 may be required for initiation of imprinted Xp inactivation and two copies of Rnf12 may be required to activate Xist in female embryos during random XCI (see main text for discussion). b RNA fluorescent in situ hybridization (FISH) in mouse embryonic stem cells (ESCs). In undifferentiated cells, the two X chromosomes are active, as shown here by biallelic expression of α-thalassaemia/mental retardation syndrome X-linked gene (AtrX). In these cells, Xist is expressed at a low level and is hardly detectable. During differentiation, one of the two Xist alleles is upregulated. Xist RNA coats the X chromosome from which it is produced and triggers X inactivation, which leads to the monoallelic expression of X linked genes such as AtrX in differentiated cells. Tsix, Xist antisense gene. NATURE REVIEWS GENETICS VOLUME 12 JUNE

6 Figure 3 Summary of Xist regulation at the onset of XCI. Numerous factors are implicated in Xist regulation. A Network of genetic interactions. Note that here arrows do not necessarily imply direct regulation. Repression of Xist by SOX2, although not formally assessed, is to be expected, given that it shares the vast majority of its targets with OCT4 and NANOG. B Possible molecular mechanisms involved in regulating Xist. Ba Binding sites of pluripotencyassociated transcription factors within elements of the network. It is still unclear whether binding to these sites actually mediates control of Xist and Tsix (Xist antisense gene) expression see main text for details. Bb The activation of Xist also requires X linked activators such as RING finger protein 12 (RNF12; also known as RLIM) in a dose-dependent manner. The upregulation of RNF12 during differentiation is thought to activate Xist, thereby triggering cis-inactivation and ultimately lowering RNF12 levels. This feedback loop ensures that one X chromosome remains active. Bc Different modes of cis-regulation operate on each chromosome. On the future active X chromosome, Tsix expression is stimulated by the X-inactivation intergenic transcription element (Xite) and DXPas34. On the future inactive X chromosome, the A-repeat region is required for the accumulation of the spliced form of Xist, which mediates silencing. However, an effect of this region on the Xist promoter in XX-differentiating cells still cannot be excluded (dashed arrow). Bd The regulatory activity of the A-repeat region has been proposed to involve the expression of a short RNA, RepA. RepA has been proposed to bind Polycomb repressive complex 2 (PRC2) and recruit it to the Xist promoter, somehow resulting in the activation of Xist. This binding of RepA to PRC2 would be antagonized by Tsix on the future active X chromosome. CTCF, CCCTC-binding factor; Jpx, also known as Enox (expressed neighbour of Xist); OCT4, also known as POU5F1; PcG, Polycomb group; PRDM14, PR domain zinc finger protein 14; REX1, reduced expression protein 1 (also known as ZFP42); YY1, transcriptional repressor protein Yin and Yang JUNE 2011 VOLUME 12

7 Box 2 Transgenesis studies of the X-inactivation centre Numerous experiments involving transgenesis have attempted to identify the minimum region necessary to recapitulate the functions of the X-inactivation centre (Xic). Single-copy Xist transgenes of up to 460 kb are unable to trigger Xist upregulation during differentiation of male embryonic stem cells (ESCs), either from the transgene or from the endogenous Xic 34. This shows that crucial Xic sequences required to render cells competent for X-chromosome inactivation (XCI) are missing from these large DNA fragments. Importantly, such single-copy transgenes can trigger imprinted XCI when paternally inherited (see the main text), implying that Xic sequence requirements are different between the two forms of XCI 34,96,125. Surprisingly, the lack of random XCI functions for single-copy transgenes can be bypassed at least partially using multicopy arrays, which can initiate inactivation in cis in male and female cells 126,127. However, their capacity to trigger Xist expression from the endogenous X chromosome is limited, and inactivation of the endogenous X chromosome or the transgene is neither random nor exclusive in such lines 32,34,128. Importantly, female mouse ESCs were also reported to be unable to trigger Xist expression from these single-copy transgenes, even though they are competent to trigger random XCI of their endogenous X chromosomes 34,125. Thus, not only do such ectopic single-copy Xic fragments lack sequences to efficiently trigger the endogenous Xist allele(s) in trans, they also cannot respond to trans-activating competence signals, such as RING finger protein 12 (RNF12; also known as RLIM) 32, originating from the endogenous X chromosomes in XX cells. Most of these missing elements still remain to be identified. Intriguingly, contrasting outcomes have been reported in two different studies concerning the effects of homozygous deletion of Rnf12 in ESCs. In one case, complete abrogation of XCI is reported 30, whereas in the other only a slightly reduced efficiency of XCI is observed 36. Whatever the cause of these differences, RNF12 is clearly a key activator of Xist. It also has an essential role during imprinted XCI, as will be discussed later. However, additional XCI activation mechanisms that can act in a partially redundant fashion during random XCI must exist and remain to be identified. Another element that has recently been implicated in Xist activation in XX cells is the Jpx locus (also known as Enox (expressed neighbour of Xist)) which is situated immediately 5 to Xist 37. This locus lies within a 120 kb region that is hyperacetylated in undifferentiated female (but not male) mouse ESCs 38. Like many other loci within the Xic, Jpx produces an ncrna and has been proposed to enable female-specific Xist activation, as its heterozygous deletion impedes XCI initiation. However, unlike Rnf12 transgenes, Jpx transgenes that are unlinked to Xist do not activate endogenous Xist expression in male ESCs 32. How exactly Jpx, or its surrounding chromatin environment, influences Xist activation in female cells remains to be determined. A recent study reported that another long ncrna Ftx (five prime to Xist) can control the expression levels of Xist, Tsix and Jpx in male cells. However, the role of Ftx in XCI remains to be investigated in female cells 39. In conclusion, it is becoming increasingly evident that the competence of XX cells for Xist upregulation is not mediated by one, but by several loci that act at the DNA level and/or at the level of the proteins or RNAs that they produce, with partially redundant activity. It should be emphasized that, to date, no Xic sequence introduced as a single extra copy in a male ESC has yet been reported to induce XCI in a fashion that is reminiscent of normal XX ESCs (BOX 2). Identifying the remaining unknown Xic elements and factors, and understanding how their interplay controls Xist expression, is an exciting challenge for the future. Monoallelic regulation of Xist during random XCI During random XCI in mice, expression of Xist is restricted to a single allele. This is the result of both counting and choice (BOX 1) and is influenced by regulatory elements within the Xic. Some of these elements are intimately linked to the regulation of XCI by pluripotency factors and the XX competence-regulation mechanisms described above. The Xce locus. In female inbred mice, the two X chromosomes have an equal chance of being chosen for XCI. However, XCI choice can be biased by alleles at the X linked X controlling element (Xce), which maps within Xic and causes non-random XCI in heterozygotes At least three natural alleles of Xce have been identified, based on skewed XCI patterns in Xce heterozygotes. Although Xce has been genetically mapped to the region 3 to Xist 43, its exact nature, location and mechanism of action are still not known. Furthermore, given the complexity of the Xist regulatory landscape, Xce alleles may correspond not just to one, but to several polymorphic controlling elements within this subregion of the Xic Tsix-mediated repression of Xist. Analysis of targeted deletions and studies with transgenes have revealed that the region lying immediately 3 to Xist is essential for correct monoallelic regulation of this gene. Heterozygous deletion of a 65 kb region 3 to Xist in female mouse ESCs results in nonrandom Xist upregulation and inactivation of the mutated X chromosome 47. Subsequent sequence replacement 48,49 and reinsertion 50,51 strategies have shown that the major Xist repressor in this region is its antisense transcription unit, Tsix (FIG. 1). Indeed, disruption of Tsix transcription by a 3.7 kb deletion encompassing its promoter recapitulates the skewing observed with the 65 kb deletion, although additional elements regulating Xist may lie within this larger region 48,50,51. Although deletion of the Tsix promoter and enhancers leads to skewed XCI of the mutated copy of the X chromosome, this does not enhance the overall expression level of Xist 50. Thus, antisense transcription appears to control the binary decision of whether to upregulate Xist during differentiation. In fact, the ratio of sense/antisense transcription across Xist seems to be crucial in determining which allele will be upregulated. When antisense transcription is artificially driven across Xist, this prevents its upregulation in cis 52,53. Conversely, enforced Xist transcription is sufficient to induce preferential inactivation of the mutated chromosome, without altering Tsix transcript levels 54,55 (FIG. 3Bc). However, Xist repression in undifferentiated cells does not rely on Tsix alone. This is because Tsix deletion does not lead to high-frequency ectopic Xist activation before differentiation; rather, it does so only after differentiation is induced, when pluripotency-factor downregulation begins 47,48,52, NATURE REVIEWS GENETICS VOLUME 12 JUNE

8 Dicer An RNase III family endonuclease that processes dsrna and precursor micrornas into small interfering RNAs and micrornas, respectively. CCCTC-binding factor (CTCF). A highly conserved DNA-binding protein with 11 zinc fingers that, in mammalian genomes, binds to regulatory elements such as insulators. How does Tsix prevent Xist upregulation? Several reports have now clearly established that antisense transcription across the Xist promoter is accompanied by modifications of its chromatin structure 23,57,59 63, although the precise function of these chromatin changes is unclear. Whether it is the act of transcription or the Tsix RNA itself that participates in these chromatin changes is also still an open question. The possible existence of Xist Tsix RNA duplexes and small RNAs (~25 42 nucleotides long) that match the Xist promoter and the A repeat region has been reported in differentiating mouse ESCs 64, suggesting the potential involvement of an RNAi-like mechanism. However, although a Dicer mutation was reported to increase Xist levels in this study, more recent analyses have shown that this is likely to be due to demethylation of the Xist promoter 65. This demethylation is due to the downregulation of micrornas that regulate the DNA methyltransferase 3A (DNMT3A) when the microrna machinery is impaired 65,66. The repressive action of Tsix has also been proposed to rely on competition for Xist-activating factors. In particular, the Xist A repeat region, which is essential for Xist RNA accumulation as well as for Xist-mediated gene silencing in cis at the level of the Xist RNA has also been reported to produce a 1.6 kb non-coding transcript, RepA, independently of the main Xist promoter 67. It has been proposed that this RepA transcript enhances Xist expression via recruitment of the Polycomb repressive complex 2 (PRC2) and that Tsix RNA prevents Xist activation by somehow competing with RepA RNA for PRC2 recruitment 23,67 (FIG. 3Bd). However, the exact role that PRC2 serves in the activation of Xist is still unclear. Indeed, it had been suggested that PRC2 also participates in Xist repression in male cells, in synergy with Tsix 62. Additionally, PRC2 does not seem to be required for Xist upregulation, as a lack of EED protein a key component of PRC2 does not prevent XCI initiation in female embryos, nor in male Tsix mutants 62,68. Thus, the functional relevance of the connections among RepA, PRC2 and the initiation of XCI remains unclear. Although the exact molecular mechanisms underlying the role of Tsix in regulating Xist are still not fully understood, the genetic evidence suggests that antisense transcription has a key role in the cis-regulation of Xist. Importantly, however, Tsix heterozygous mutants do not simply induce Xist expression more frequently from the mutated chromosome. Somehow, inactivation of the Xist wild-type X chromosome also seems to be prevented, or bypassed, owing to accelerated XCI on the deleted X chromosome, which results in rapidly reduced RNF12 levels 33. This finding highlights the fact that there must be some mechanism to coordinate XCI between the two chromosomes. Indeed, this coordination seems to be lost in Tsix homozygous mutants, as an increased number of cells activate Xist from both X chromosomes 4. The general picture that is emerging from such studies is that Tsix plays a central part in the accurate monoallelic expression of Xist during random XCI initiation. This, of course, begs the question of how Tsix itself is regulated. Regulation of Tsix expression. Perhaps surprisingly, the Tsix core promoter is dispensable for basal transcription of this RNA and its deletion does not skew choice 69. Conversely, several critical regulatory elements of Tsix have been defined. For example, the DXPas34 minisatellite, which was initially identified based on differential methylation patterns of the active X chromosome versus the Xi 44,70, acts as an enhancer of the Tsix promoter in reporter assays 71. Furthermore, DXPas34 removal results in loss of Tsix transcription and nonrandom Xist upregulation in mouse ESCs and mice 56,69,72. This suggests that loss of this regulatory element is responsible for the phenotypes observed with the larger deletions of the Tsix promoter region that encompass the DXPas34 minisatellite 47,48. Interestingly, transcription can be initiated at DXPas34 (REF. 69) and the 5 ends of some Tsix isoforms map to the minisatellite 73. In addition, several of the transcription factors that have been shown to regulate Tsix expression, such as CCCTC-binding factor (CTCF) and its PcG co-factor YY1, as well as the stem-cell factor reduced expression protein 1 (REX1; also known as ZPF42), bind to DXPas34 (REFS 31,74). Another enhancer of Tsix, Xite (X-inactivation intergenic transcription element), lies kb upstream of the Tsix start site. Deletion of Xite results in mildly skewed XCI and accelerated downregulation of Tsix upon differentiation 75. Several pluripotency factors have been found to target Xite and their binding sites are necessary for Xite reporter constructs to be transactivated 27. These pluripotency factors include SOX2 (REFS 27,28), OCT4 (REFS 27,28) and NANOG 28. In summary, Tsix is a regulator of Xist and is itself regulated by several long-range cis-acting elements, as well as by pluripotency factors (FIG. 3A). Although mutations in Tsix or Xite on one X chromosome can clearly skew the choice of X chromosome to be inactivated, the question remains as to how asymmetry in Xist expression patterns is established when two genetically identical Tsix alleles are present. Ensuring monoallelic expression of Xist. After Xist has been upregulated and XCI triggered on one X chromosome in female cells, repression of the second allele of Xist (on the active X chromosome) is maintained by DNA methylation of its promoter. This is supported by the fact that impairment of both Dnmt3a and Dnmt3b leads to ectopic Xist activation at late developmental stages in both males in females 76. However, the mechanisms ensuring that only one Xist allele is expressed at the outset of XCI are still not fully understood. One pathway that has been proposed to explain the asymmetric expression of Xist is a negative feedback loop that involves Rnf12 (and other possible X linked Xist activators) 32 and is triggered by Xist expression itself 30 (for review, see REF. 77). Upon initiation of XCI on one X chromosome, rapid Xist RNA-mediated silencing of Rnf12 would result in downregulation of the protein, thereby diminishing the activating effect of RNF12 on Xist transcription (FIGS 3Bb,4a). This feedback model is based on the hypothesis that Rnf12 (and other potential X linked XCI-promoting factors) must be rapidly 436 JUNE 2011 VOLUME 12

9 Figure 4 Models for monoallelic regulation of Xist. Monoallelic Xist expression may be achieved through several (not mutually exclusive) mechanisms. a The feedback model proposes that each X chromosome produces an X-chromosome inactivation (XCI) promoting factor (or factors), which will activate Xist in a dose-dependent fashion. Initiation of XCI will lead to the downregulation of such a factor on one allele, bringing its concentration below the threshold required to activate the second Xist allele 33,77. It has been proposed that RING finger protein 12 (RNF12; also known as RLIM) is one such factor 32. To serve as a robust feedback mechanism, the downregulation of the XCI-promoting factors would need to occur very rapidly, before Xist activation on the second allele. b The X-inactivation centre (Xic) pairing model proposes that physical interactions between the two X-pairing regions (Xpr) may render the two X chromosomes competent for Xist expression. Subsequent trans-interactions between homologous Tsix (Xist antisense gene) and Xite (X-inactivation intergenic transcription element) regions could enable monoallelic Xist upregulation 5,78,79 by promoting a symmetry-breaking event between the two alleles 139. This model was recently supported by live-cell imaging of Tsix-pairing events followed by Xist/Tsix RNA fluorescent in situ hybridization (FISH) 81. c The stochastic and secondary selection-based model proposes that each X chromosome has a low and intrinsic probability of activating Xist. Because only cells with one active X chromosome per diploid set of autosomes can survive, this leads to counter-selection against cells with two active or two inactive X chromosomes. Counter-selection may either involve cell death or the ability to shift to an XCI pattern that is compatible with cell proliferation 33. Such a model requires that cells remain competent for Xist activation for numerous cell cycles 85, a property that has not yet been examined in vivo (in peri-implantation mouse embryos). d Pre-emptive states corresponding to different propensities for Xist activation have been proposed to exist prior to XCI. In this model each chromosome can alternate between these states, which have been proposed to involve alternative structural configurations at the level of sister chromatid cohesion between the X chromosomes 82. downregulated by Xist RNA to avoid activation of the second Xist allele (FIG. 4a). Transient homologous Xic pairing has also been proposed to play a part in the coordination and asymmetric treatment of the two Xics during initiation of XCI (FIG. 4b). Indeed, initial Xic Xic pairing events mediated by Xpr are followed by pairing at Tsix/Xite region in differentiating XX ESCs at the moment of Xist upregulation 5,78,79. Although pairing at Tsix/Xite is clearly not necessary for Xist activation 78,79, several observations indicate these events may be linked to the monoallelic regulation of Xist. First, Tsix/Xite trans-interactions occur at the time of Xist activation 78,79. Second, Tsix deletions, which have been shown to skew choice, also abrogate pairing at the Tsix/Xite region 78,79. Third, single-copy transgenes of the region containing Xist, Tsix and Xite that are unable to activate Xist either in cis or in trans are also unable to associate with the endogenous Xic 78. Last, ectopically provided Tsix and Xite multicopy arrays can mediate efficient pairing with the endogenous Xic regions and inhibit XCI in females 4,79,80. Insights into the events immediately downstream of pairing have recently been obtained using live cell imaging of tagged Tsix loci 81. Tsix expression was found to become transiently monoallelic after separation of the loci, thus providing a window of opportunity for Xist upregulation in cis to the silent Tsix allele. Depletion of CTCF and OCT4 by knockdown, as well as transcriptional inhibition, has been shown to disrupt Tsix/Xite pairing and perturb Xist expression. However, it is unclear whether the Xist deregulation observed is directly due to the disruption of these chromosomal interactions or due to other effects 27,80. The mechanisms that drive pairing, and its exact role (or roles) in XCI, remain to be precisely elucidated. Another mechanism that has been proposed to account for monoallelic Xist upregulation is that Xist activation would occur stochastically at a low frequency and that this would be followed by a counter-selection of NATURE REVIEWS GENETICS VOLUME 12 JUNE

10 Eutherians Mammals in which the development of progeny takes place in the mother s body thanks to the placenta, a fetal membrane that facilitates nutrient and waste exchange between the fetus and the mother. Meiotic sex chromosome inactivation (MSCI). Silencing and heterochromatinization of sex chromosomes in the male germ line during meiosis. Androgenetic Androgenetic embryos are produced by the fusion of two haploid paternal genomes. Parthenogenote A uniparental embryo produced by the development of an unfertilized egg. Gynogenote An embryo produced by the fusion of two haploid maternal genomes. cells that have not established XCI correctly during differentiation. This means that they have either inactivated both X chromosomes or nei ther X chromosome, both of which would be deleterious situations with aberrant X chromosome dosage (FIG. 4c). However, although some degree of selection following inaccurate XCI may occur during in vitro differentiation of mouse ESCs 4,33, it rarely seems to occur in vivo in mice (C. Corbel, I. Okamoto and E.H., unpublished observations). A final proposed model is that, before random XCI, the two Xic loci are differently poised for Xist activation (FIG. 4d). Indeed, in undifferentiated mouse ESCs, the two X chromosomes exist in alternative and alternating structural states at the level of their sister chromatid cohesion 82. These states appear to be anticorrelated between the two X chromosomes in the same cell. Furthermore, chromosome-wide patterns of asynchronous sister chromatid cohesion states are altered by mutations within Xist or Tsix. However, the basis for these coordinated, alternating states and their possible role in random monoallelic choice during XCI remain to be determined. In conclusion, multiple models have been proposed for the asymmetric expression and monoallelic regulation of Xist. These are not mutually exclusive; in fact, they may be exploited at different levels and to varying extents, in order to accomplish appropriate Xist and XCI patterns during development. Indeed, in different species, some of these routes to achieve monoallelic XCI may be exploited more than others, as will be discussed later. Impact of the X/autosome ratio on XCI Experiments involving triploid and tetraploid embryos demonstrated that the number of inactive X chromosomes seems to depend on autosomal ploidy, with the majority of cells retaining one active X chromosome per diploid set of autosomes 83,84 (see Counting in BOX 1). More recently, kinetic measurements revealed that, for the same number of X chromosomes, Xist upregulation happens more rapidly in differentiating mouse ESCs with a high X/autosome (X/A) ratio than in cells with a low X/A ratio 85. This suggests that autosomal factors directly regulate the probability of Xist activation; secondary selection is, however, clearly measurable at later stages in these cell populations. What could the nature of these autosomal counting factors be? Given that most pluripotency factors are autosomal, it is tempting to speculate that they might be part of this X/A counting mechanism. The exact nature of counting still remains mysterious and it should be noted that all of the Xic mutations so far proposed to affect counting have only been investigated in diploid cells. Therefore, they have not been tested for their sensitivity to autosomal dosage, which is how defects in X/A counting should be assessed. The regulation of imprinted XCI Although random XCI is believed to be the norm in eutherians, in mice XCI is initially subject to imprinting during pre-implantation development, with exclusive inactivation of the Xp (FIG. 2). At the time of zygotic genome activation (ZGA), both X chromosomes are active but the Xp rapidly initiates XCI following imprinted Xist expression from the 2 4 cell stage onwards XCI seems to be complete by the blastocyst stage. Imprinted inactivation of the Xp is maintained in the extra-embryonic tissues but is reversed in the ICM, where random XCI subsequently takes place How is this imprinted form of XCI controlled? A robust maternal imprint that prevents inactivation of the Xm exists in mice, given that XCI does not occur in Xm disomies, leading to early lethality owing to defects in extra-embryonic development Xist is clearly essential for imprinted XCI, as a deletion of the paternal Xist allele also leads to early lethality 8. However, it has been proposed that the Xp may also be predisposed to silencing due to its heterochromatinization in the XY body during meiotic sex chromosome inactivation (MSCI) in the male germ line 92,93. In support of this, a recent study suggested that some genes on the Xp may be silenced independently of Xist during cleavage stages 94. However, a subsequent study came to the conclusion that Xist is in fact required for initiation of XCI of X linked genes 95, although silencing of repetitive elements on the Xp may persist independently of Xist from the male germ line into the zygote on the Xp. The demonstration that Xist is sufficient for the initiation of imprinted XCI in mice came from a study 96 showing that autosomal Xist transgenes can initiate imprinted cis-inactivation independently of MSCI when they are paternally transmitted. How then is Xist regulated during early mouse embryogenesis? Contrary to the situation for random XCI, Xist expression is strictly dependent on parental origin immediately after fertilization. Whereas Xist is exclusively transcribed from the paternal allele, the maternal allele of Xist is never expressed during early pre-implantation development Importantly, paternal Xist expression occurs regardless of X chromosome number, unlike the situation during random XCI. For example, in androgenetic XX embryos, Xist RNA coating of both X chromosomes is seen, although this is resolved to monoallelic Xist expression by the blastocyst stage 101. Furthermore, XO androgenotes also initiate Xist RNA coating, but this later disappears. Conversely, parthenogenotes (XmXm) show a complete absence of Xist expression up to the morula stage, after which some Xist upregulation is observed 102,103. Taken together, these data suggest that only the paternal but not the maternal Xist allele can respond to the transcription factor environment present in cleavage-stage mouse embryos. Recently, the maternal pool of RNF12 was shown to be essential for paternal Xist expression, as imprinted XCI is not initiated in Rnf12 +/ female embryos derived from Rnf12 deficient oocytes 36. What prevents Xist expression from the Xm in cleavagestage mouse embryos? Evidence that a repressive imprint is deposited during egg maturation came from the observation that an X chromosome derived from non-growing, rather than fully grown, oocytes can be inactivated in gynogenotes 104. However, the nature of this imprint is still unknown. Unlike many autosomal imprinted loci, Xist imprinting does not seem to rely on differential 438 JUNE 2011 VOLUME 12