Orchestrating Asymmetric Expression: Mechanisms behind Xist Regulation

Abstract

Compensation for the gene dosage disequilibrium between sex chromosomes in mammals is achieved in female cells by repressing one of its X chromosomes through a process called X chromosome inactivation (XCI), exemplifying the control of gene expression by epigenetic mechanisms. A critical player in this mechanism is Xist, a long, non-coding RNA upregulated from a single X chromosome during early embryonic development in female cells. Over the past few decades, many factors involved at different levels in the regulation of Xist have been discovered. In this review, we hierarchically describe and analyze the different layers of Xist regulation operating concurrently and intricately interacting with each other to achieve asymmetric and monoallelic upregulation of Xist in murine female cells. We categorize these into five different classes: DNA elements, transcription factors, other regulatory proteins, long non-coding RNAs, and the chromatin and topological landscape surrounding Xist.

1. Introduction

Dosage compensation of gene expression between chromosomes is essential for survival. Female mammalian cells carry two X chromosomes, while male cells carry an X and a Y chromosome, generating an X-linked gene dosage imbalance between the sexes. To achieve dosage compensation, female diploid epiblast cells inactivate a single X chromosome very early during embryonic development via a complex process known as X chromosome inactivation (XCI). This process results in the epigenetic silencing of one randomly selected X chromosome (Xi), indicating that female and male cells must sense how many X chromosomes they carry.

In mice, two waves of epigenetic silencing occur. Firstly, imprinted XCI (iXCI) leads to the inactivation of the paternal X chromosome following fertilization. This paternal Xi is later re-activated in the inner cell mass of the embryo, and a second, random XCI (rXCI) wave occurs, where a single X is randomly chosen for inactivation in the early epiblast. Most in vitro studies of XCI are performed in mouse embryonic stem cells (ESCs), which carry two active X chromosomes (Xa) and initiate rXCI upon exit of pluripotency and differentiation, recapitulating events in the epiblast of embryos. However, not all genes on the Xi are silenced; they are called escapees or escaping genes. Even though XCI happens in all mammals, different species show different patterns of it. While mice display both types of XCI, rabbits, monkeys, and humans, for instance, only show rXCI, and marsupials display iXCI only. This illustrates the variety of XCI mechanisms present within the mammalian class. New data are emerging on different mechanisms in other mammals, such as rabbits, monkeys and especially humans (reviewed in [1]). However, since most XCI research has been performed in mice and very exciting new data are still being generated nowadays, we focus this review on mouse rXCI.

1.1. Kicking off X Chromosome Inactivation: The Stochastic Model

In 1971, Mary Lyon proposed a model where a diffusible X-linked factor would guarantee XCI of a single X [2]. This model has been further developed into a stochastic model, where female cells sense X chromosome dosage, leading to the inactivation of a single X chromosome while preventing inactivation of the single X chromosome in male cells. The stochastic model proposes that XCI is achieved by a tightly controlled balance between X-encoded activators and autosomally encoded repressors [3]. Many autosomally encoded repressors are pluripotency factors or are linked to the pluripotent state and therefore prevent the inactivation of an X chromosome in the inner cell mass of the embryo (ICM) or in ESCs. ESCs are derived from the ICM, which will give rise to the embryo proper. Upon differentiation, the downregulation of repressors of XCI and upregulation of activators of XCI would tilt the balance towards XCI. Thus, female exclusive activation of XCI is mediated by the double dose of X-encoded XCI activators that are required to overcome the threshold set by the XCI repressors. Initiation of XCI is stochastic, and a negative feedback loop involving rapid silencing of some XCI activators prevents inactivation of the second X chromosome. The application of novel machine learning and mathematical modeling techniques indicates that XCI dynamics can be recapitulated based on the principles of the stochastic model with a limited number of activators and repressors regulating XCI [4,5].

1.2. The X Inactivation Center and the Tsix/Xist Tandem

Research on truncated human X chromosomes and balanced X-autosome translocations led to the discovery of the X inactivation center (Xic), a locus on the X chromosome essential for XCI [6,7]. The mouse Xic spans many long non-coding RNAs and several protein-coding genes, covering at least 800kb (reviewed in [8]). Later, transgenic studies showed the long non-coding RNA (lncRNA) gene Xist, embedded within the Xic, to be required for XCI to happen. Xist is critical for both iXCI and rXCI in mice. Endogenous Xist deletions prevent XCI from happening on the X chromosome carrying the deletion, while ectopic insertions on autosomes result in their silencing [9,10]. The mouse Xist gene is a 22 kb-long gene with no conserved open reading frame (ORF) and is transcribed, polyadenylated, and alternatively spliced into a 15 kb-long lncRNA [11,12,13]. Xist is transcribed antisense to another lncRNA named Tsix that spans Xist entirely and negatively regulates its expression during development [14]. Biallelic expression of Tsix in the pluripotent state thus maintains the silent state of Xist in female ESCs. Exit of pluripotency triggers a break from symmetric Xist expression to an asymmetric Tsix/Xist expression state, where Tsix expression is maintained on the Xa while Xist expression is greatly upregulated on the selected Xi. Finally, Xist expression is locked on the Xi, and the eventual downregulation of Tsix from the Xa happens once XCI is established (Figure 1). Many of these types of antisense gene expression regulation have been described in animals and plants, such as imprinted genes or the Flc-COOLAIR gene-tandem involved in plant flowering [15].

In this review, we hierarchically discuss the several levels of Xist regulation that play a role in both the maintenance and breakage of Xic symmetry in mouse ESCs during rXCI. We have categorized the levels that contribute to the Tsix/Xist tandem expression pattern into five different classes: DNA elements, transcription factors, other proteins, lncRNAs, and the chromatin and topological landscape surrounding Xist.

2. DNA Elements That Impact Xist Expression

2.1. The Promoter Region of Xist

The discovery of the Xist gene was a significant breakthrough in understanding the mechanism of XCI. Its lncRNA product is now accepted as the master regulator behind the epigenetic silencing of one of the two X chromosomes in female mammals. Xist was first identified and isolated from a mouse cDNA library in 1991 [7]. It was discovered that Xist RNA spreads in cis, interacting with the chromatin of the X chromosome, leading to the spread of heterochromatinization and subsequent inactivation of the chromosome [12,16]. In an effort to understand the transcriptional regulation of Xist and thus the mechanisms kicking off XCI, follow-up studies centered on the characterization of the Xist promoter.

Early studies identified the mouse Xist gene’s minimal promoter region to be approximately 0.4 kb in size and located right upstream of the major transcriptional start site (TSS). This minimal promoter is a weak constitutive TATA-like promoter with a possible initiator element spanning the TSS [17]. Xist promoter sequences have been compared between several mammals, including humans, mice, rabbits, and horses, showing a high degree of conservation between these species [18]. Follow-up studies identified multiple Xist promoters, termed P1, P2, and P0, with P1 being the first identified region just upstream of Xist’s TSS. The P0 promoter of Xist was mapped 6.5 kb upstream of the TSS and gives rise to unstable Xist transcripts. The P2 promoter was mapped 1.5 kb downstream of the TSS and thus within Xist exon 1, giving rise to a stable Xist transcript (Figure 2A,B). Promoter switching from P0 to P1/P2 was then proposed as a regulatory mechanism controlling Xist expression with the exit of pluripotency [19]. However, recent chromatin immunoprecipitation assays followed by sequencing (ChIP-seq) of the P2 promoter of Xist seem to indicate this region might act as an internal regulatory element, or enhancer, rather than a promoter [20,21].

2.2. Distant Regulatory Regions

While the promoter regions of Xist drive its expression, distant DNA regulatory elements (RE) seem to play a role in titrating its expression. A regulatory region located ∼10 kb downstream of Xist TSS, and thus within Xist intron 1, was shown to bind several pluripotency factors (see Section 3.2) and play a role in Xist repression in the pluripotent state (Figure 2A) [22]. A subsequent study confirmed these findings through genetic deletions of Xist intron 1. Transgenic male ESCs carrying an Xist intron 1 deletion show moderately upregulated Xist expression in the pluripotent state, which is exacerbated upon Tsix co-removal [23]. However, in two contrasting studies, deletion of Xist intron 1 in female ES cells does not seem to impact Xist expression both in vivo and in vitro but rather skews the future Xi choice towards the mutated allele [24,25].

Identification of REs involved in Xist expression has proved to be a challenging task due to numerous factors influencing Xist expression within the Xic. Distinguishing direct from indirect effects on Xist expression via genetic perturbations within the Xic is difficult. However, a recent publication has successfully addressed this challenge. Using an elegant screening approach taking advantage of dCas9 fused to the repressor KRAB (CRISPRi) and detecting Xist levels by fluorescently activated cell sorting, several candidate regions were epigenetically and systematically silenced [26]. As expected, epigenetic inactivation of the previously discussed Xist promoter regions, as well as of known Xist regulators (discussed later in this review), impairs Xist upregulation upon exit of pluripotency, serving as a validation of the screening approach. A novel regulatory cluster located ~150 kb upstream of the Xist TSS, termed RE93-97, was also identified. Targeting this cluster with CRISPRi revealed a gradual decrease in Xist expression, pointing to RE93-97’s role as a Xist enhancer cluster. Furthermore, capture Hi-C analysis revealed RE93-97 interacts with Xist’s P2 promoter upon exit of pluripotency, suggesting this regulatory cluster responds to differentiation cues (see Section 3.3).

In a follow-up study by the same group, another pivotal RE responsible for driving Xist expression was identified [27]. Employing a CRISPR activation screening approach, several protein factors that result in Xist expression upon their overexpression were identified. A comprehensive analysis of potential candidates and their binding sites revealed that the GATA family of transcription factors (see Section 3.3) binds not only RE93-97 but also a newly discovered regulatory element located ~100 kb upstream of the Xist TSS termed RE79. Genetic deletions of RE97, as well as RE79, validated previous discoveries regarding the significance of RE93-97 in rXCI while elucidating the role of RE79 in driving Xist upregulation expression in iXCI.

Gene transcription regulation is achieved not only by many different protein factors binding to promoters and REs, such as RNA Pol II, transcription factors (TFs), chromatin remodellers, etc., but also through complex chromatin architecture. In this next section, we will analyze which transcription factors are important for Xist regulation.

3. Transcription Factors That Impact Xist Upregulation

Transcription factors regulating Xist expression can either be autosomally or X-encoded. In this review, we call TFs or other proteins that are autosomally encoded and regulate Tsix or Xist expression as Xist activators or repressors, while proteins that are X-encoded and ultimately impact Xist expression as XCI activators. There are currently no known X-linked Xist repressors.

3.1. X-Linked TFs: XCI Activators

A limited number of X-linked TFs are involved in Xist regulation. Recent data have shown that X-linked TF GATA1, a widely expressed TF in vertebrates, acts as an XCI activator [27]. Using a pooled CRISPR library to upregulate target genes in a Xist-sensitized male ESC line, the authors were able to discover GATA1, 4, and 6 as potential activators of Xist. GATA1 is an X-linked gene whose overexpression leads to increased Xist expression in ESCs, probably mediated by its indirect effect on GATA6’s overexpression, which binds the distal Xist enhancer elements RE79 and RE93-97 (see Section 2.2). Although GATA factors have been shown to induce differentiation in ESCs [28], GATA1 overexpression does not lead to an obvious differentiation phenotype. Moreover, its expression is very low in ESCs, which could explain why its knockdown (KD) does not influence rXCI kinetics. As the authors pointed out, however, its role in Xist overexpression might happen during iXCI, when its expression peaks during preimplantation development. Indeed, triple knockout (KO) of GATA1, 4, and 6 abrogates Xist expression at the 8-cell stage.

3.2. Autosomally Encoded TFs: Xist Repressors

Several TFs have been described in the last few decades as playing a role in Xist repression. They are all autosomally encoded, and the majority of them belong to the pluripotency factor network. Female mouse ESCs are powerful model systems to study rXCI, harboring two Xas and initiating rXCI upon exit of pluripotency and differentiation. Many studies have delved into how the pluripotency factors OCT4, NANOG, SOX2, and REX1 mechanistically repress Xist in ESCs. ChIP experiments revealed that OCT4, NANOG, and SOX2 directly bind to Xist intron 1 (Figure 2A) [22]. Removal of OCT4 in male ESCs, which normally should not upregulate Xist upon differentiation, leads to a loss of NANOG and SOX2 binding to intron 1 and a concomitant increase in Xist expression, indicating that these factors act as repressors of Xist in the pluripotent state. This elegant model specifies that ESCs, in their pluripotent state, show very little Xist expression, while upon differentiation, downregulation of the pluripotency factor network results in the derepression and release of Xist. Confirming these results, as indicated above (see Section 2.2), removal of intron 1 in male ESCs leads to a slight upregulation of Xist [23]. It was shown later that deletion of intron 1 does not overtly abrogate Xist repression in male and female undifferentiated ESCs but seems involved in Xist induction and in preferential inactivation of the mutated allele upon differentiation [24,25]. Although minimal, OCT4 and SOX2 also bind Tsix at its DxPas34 minisatellite region, and Xite, a gene encoding another long non-coding RNA that acts as a Tsix enhancer (see Section 5.1) and is involved in Tsix upregulation and concomitant Xist downregulation [29]. Finally, in ESCs, OCT4, NANOG, and SOX2 also downregulate the expression of Rnf12, an important X-linked XCI activator ([30,31]; see Section 4.1).

Additional pluripotency factors have also been implicated in the regulation of XCI initiation. REX1 clearly binds to the DxPas34 region of Tsix in female cells and to the Xist promoter and its promoter distal region (Figure 2A) [32,33]. Rex1 KD leads to reduced RNA Pol II recruitment to DxPas34 and the Tsix 3′ regions as well as reduced H3K36me3 deposition, but no reduction at the Tsix TSS, indicating that REX1 is involved in Tsix elongation [32]. In addition, luciferase studies of the Xist promoter in conjunction with REX1 overexpression also indicated the direct role of REX1 in Xist repression [33]. REX1 overexpression studies in female ESCs led to the conclusion that REX1 is an activator of Tsix and a repressor of Xist. Moreover, overexpressing REX1 in Tsix-Cherry Xist-GFP ESCs, where the regular relationship between Tsix and Xist is absent, also showed REX1’s role in regulating Xist and Tsix independently [34]. Indeed, Rex1 KO cells exhibit an increased number of Xist clouds and double clouds at the ESC stage and during differentiation [35]. Nevertheless, Rex1 KO mice are fertile and viable, pointing to the robustness of XCI when perturbed during embryonic development. Finally, the pluripotency factors KLF4 and C-MYC also bind the 5′ region of Tsix [32].

The transcriptional regulator PRDM14, expressed in ESCs and primordial germ cells, and involved in reprogramming, has been reported to regulate Xist expression [36,37,38]. PRDM14 binds Xist intron 1 DNA, while Prdm14 KD leads to upregulation of Xist in female ESCs. However, its mechanism of action might be indirect, by repressing Rnf12 in ESCs through decreased binding of PRC2 and concomitant decreased H3K27me3 deposition around the Rnf12 promoter region. However, Prdm14 KO was reported not to affect Xist expression in another study [38].

Another autosomally encoded TF repressor of Xist is CTCF, an 11-Zn finger protein involved in a wide variety of gene regulatory mechanisms, such as insulation, gene activation and repression, and in setting up and maintaining the higher-order chromatin architecture, such as the boundary formation of topologically associating domains (TADs) (reviewed in [39] and explained further in Section 6.3). In silico analysis of the Tsix DxPas34 and Xist promoter regions showed that these regions contain many CTCF consensus binding sequences, which bind CTCF in vitro and in vivo (Figure 2A) [20,40,41]. Furthermore, while Ctcf KD leads to Tsix downregulation and Xist upregulation, its overexpression leads to Xist downregulation, all pointing to CTCF acting as a repressor of Xist [40]. Its mechanism of action seems to be counteracted by the lncRNA Jpx/Enox (see Section 5.2), an activator of Xist, which titrates CTCF down from the Xist locus, leading to Xist upregulation upon differentiation [41]. Finally, deletion of a strong CTCF binding site in a region called RS14 at the Tsix/Xist junction leads to the absence of XCI from the mutant X in heterozygous differentiating female ESCs [42].

Finally, the MSL complex regulates sex determination and dosage compensation in Drosophila and is composed of several subunits, among which are MOF, MSL1, and MSL2, the latter binding DNA directly (reviewed in [43]). Dosage compensation in Drosophila is achieved by the MSL complex being recruited to the single male X chromosome, depositing H4K16ac through the histone acetyltransferase activity of MOF and concomitant X-linked gene upregulation. Analysis of the binding patterns of MSL1/2 and MOF in mouse ESC cells showed their prominent recruitment to the Tsix DxPas34 region [44]. Msl2 KD results in decreased H4K16ac in this region, followed by decreased Tsix expression and increased Xist expression. The authors then show that MSL2 facilitates proper recruitment of REX1 to DxPas34 and the general transcription factor YY1 (see next section) to Tsix’s promoter. MSL2 removal also leads to chaotic XCI, with an increase in female differentiating ESCs bearing two Xist clouds. The authors propose a mechanism where the MSL complex binds Tsix, enhancing its expression and thus repressing Xist indirectly. However, several other factors involved in direct or indirect regulation of Xist also seem to be misregulated in Msl2 KD experiments, making the exact role of the MSL complex in Xist regulation somewhat complex.

Altogether, these findings suggest that a set of autosomally encoded transcription factors act in conjunction to keep Xist repressed through direct and indirect mechanisms. While the pluripotency network is intricately associated with the repression of Xist at the ESC stage, its disappearance upon differentiation releases Xist’s repression, permitting its upregulation and XCI.

3.3. Autosomally Encoded TFs: Xist Activators

One of the best-studied autosomally encoded activators of Xist expression is YY1, a general transcription factor involved in context-dependent gene activation and repression (reviewed in [45]). Interestingly, REX1 is a paralog of YY1, where REX1 arose through a retrotransposition event in mammals, and both proteins share a highly similar consensus binding motif [46]. This might also explain why YY1 binds the Tsix DxPas34 region and downstream of the Xist TSS as REX1 does (Figure 2B) [20,32]. Making use of a Tsix-STOP construct that displays Xist upregulation from the single X chromosome in male cells, a study showed that YY1 binds downstream of the Xist TSS both before and during XCI [20]. In contrast, YY1 recruitment disappears in differentiated male control ESCs carrying a silenced Xist locus on the single Xa. YY1 deposition is DNA methylation-dependent. This suggests a role for YY1 as an activator of Xist on the Xi, which was confirmed with Yy1 KD studies in differentiating female ESCs [20] and by another study [47]. In line with this, Yy1 KD studies in female MEFs revealed that YY1 is important for Xist maintenance expression or localization [20,48]. This study described an elegant mechanism for the role of YY1 vs. REX1 in Xist expression, where both proteins compete for the same binding sequences around the Xist TSS, with YY1 acting as an activator and REX1 acting as a repressor. REX1 downregulation upon exit of pluripotency leads to increased YY1 binding, resulting in Xist upregulation.

As stated above, several autosomally encoded members of the GATA TF family (GATA2-6) have been shown to regulate Xist expression [27]. Upregulation of GATA members 1 to 6 by dCRISPR-VP64 leads to increased Xist expression in mouse ESCs, potentially explained by extensive cross-activation of the members. For instance, independent overexpression of all GATA members consistently results in upregulation of Gata4 and Gata6. In an independent GATA6-inducible overexpression system, Xist was upregulated in a dose-dependent manner within a few hours. CUT&TAG experiments indicated that GATA 2, 3, 4, and 6 bind separate distal Xist enhancers (RE79 and RE97) in extraembryonic cells. Since mouse ESCs show negligible GATA TF expression, the authors overexpressed GATA6 and detected its binding to RE79 only. This study also shows that the removal of RE79 and RE97 abrogates Xist expression during iXCI in vivo, and given that they are bound by several of the GATA factors (GATA2 and 3 in trophectoderm; GATA4 and 6 in primitive endoderm), the latter might be important for Xist sustained expression in a wide variety of differentiated tissues where these factors are expressed. In conclusion, GATA TF family members are thus potent activators of Xist in iXCI in extraembryonic tissues and, when overexpressed, in mouse ESCs [27].

4. Non-DNA-Binding Factors That Impact Xist Upregulation

4.1. X-Linked Protein Regulators: XCI Activators

The stochastic model for X chromosome counting postulates the presence of X-linked XCI activators whose workings are counterbalanced by autosomally encoded repressors. Studies during the last couple of decades by two independent groups have shown that the E3 ubiquitin ligase RNF12/RLIM acts as an X-linked activator of XCI. Ectopic insertion of BACs harboring Rnf12 in male and female ESCs revealed how RNF12 overexpression leads to dosage-dependent upregulation of Xist [30]. Female Rnf12 heterozygous cells, phenotypically identical to male cells in terms of Rnf12 expression, manage to inactivate an X chromosome, albeit at a reduced rate, indicating the presence of additional XCI activators [24]. However, contradicting data were provided as to whether Rnf12 KO ESCs fail to inactivate an X chromosome in vitro [24,35,49]. Truncation of Rnf12 by a Neo cassette or complete removal of the ORF of Rnf12 results in failure to upregulate Xist in vitro [24,35], whereas another study shows that RNF12 is dispensable for XCI in the epiblast [49]. These discrepancies might be explained by the different backgrounds of the mice used or by differences in developmental and in vitro differentiation conditions, such as RA-induced differentiation, EB differentiation, or low O2 levels that may affect the robustness of XCI [50]. From these results, it is clear that in vivo XCI is much more robust than in vitro XCI. Both groups agree, however, that imprinted Xist expression is highly dependent on Rnf12 expression in vivo [35,51]. The mechanism through which RNF12 is implicated in Xist expression was clarified by conducting immunoprecipitation followed by mass spectrometry. This analysis revealed that REX1 is the primary TF co-purifying with RNF12 in ESCs [33]. Subsequent experiments revealed that RNF12 ubiquitinates REX1, targeting it for proteasomal degradation, in line with the increased REX1 stabilization found in Rnf12 KO ESCs [33,35]. Moreover, the pluripotency factors OCT4 and NANOG have also been shown to repress Rnf12 in ESCs (Figure 2A) [24,31], leading to a slight upregulation of Rnf12 upon differentiation in female cells. Altogether, these results have led to the proposal of the existence of an RNF12/REX1 axis that regulates Xist expression. Increased levels of RNF12 lead to decreased REX1 levels, which release Xist repression, resulting in increased Xist levels and the initiation of XCI. In fact, removal of both Rnf12 and Rex1 partially rescues the Rnf12 phenotype [35], further suggesting the presence of additional X-encoded activators of XCI.

Indeed, two recent studies have implicated the role of two X-linked but XCI-escaping histone demethylases, KDM5C/JARID1C and KDM6A/UTX, implicated in sexual differentiation (see review [52]) with opposing roles in enhancing Xist expression [21,53]. While KDM5C is an H3K4 demethylase, KDM6A is an H3K27 demethylase [54,55]. Kdmc5 KO mice show female-specific lethality, and E5.5 epiblast cells display fewer or smaller Xist RNA clouds, suggesting a defect in Xist RNA expression [21]. In vitro differentiation of Kdm5c KO ESCs to epiblast-like cells (EpiLCs) also results in reduced Xist upregulation, while ectopic KDMC5 expression in Tsix KO sensitized male cells leads to increased Xist expression. In addition, KDM5C is strongly recruited to the Xist P2 promoter, leading to H3K4me2/3 demethylation into H3K4me1, along with an increase in H3K27ac, which are common histone modifications of active enhancers (Figure 2B). The authors propose that the Xist P2 promoter acts as an enhancer-like region of Xist. Finally, Kdm5c KO female epiblasts and EpiLCs show residual Xist expression, indicating again that several mechanisms relying on different X-encoded dose-dependent XCI activators are at play.

A recent preprint shows how another escaping gene, Kdm6a, might be involved in Xist expression regulation. Kdm6a KO ESCs exhibit female-specific impaired differentiation in vitro, decreased Xist expression levels with a concomitant reduction in Xist clouds, and impaired silencing of X-linked genes [53,56]. The authors propose a model where a double dosage of KDM6A in ESCs leads to H3K27me3 removal from the Xist promoter region, allowing for its upregulation. However, there is little KDM6A recruitment to the Xist promoter at days 2 and 7 of differentiation, which is when Xist expression is already affected (Figure 2B). This result and how KDM6A affects other players of XCI in the Xic will need to be further addressed.

KDM5C and KDM6A thus seem to be XCI activators through different mechanisms, and their being X-linked is in accordance with the stochastic model of Xist activation, inducing persistent Xist expression in female cells upon differentiation. However, since both genes escape XCI, they cannot participate in the feedback loop where decreasing X-linked activator expression upon monoallelic inactivation prevents upregulation of the other Xist allele.

4.2. Autosomally Encoded Proteins: Xist Repressors and Activators

Groundbreaking work by several groups described the large multidomain SPEN/SHARP as a critical factor in Xist’s silencing of the X chromosome [57,58,59,60,61]. While SPEN interacts with several co-repressors, such as the NCoR2/SMRT complex and chromatin remodelers and deacetylases such as NURD [58,62], it interacts through its SPOC domain with Xist’s RNA A repeat region to activate HDAC3 on the X, leading to silencing of gene expression ([63]; reviewed in [64]). Interestingly, Spen KO female ESCs fail to upregulate Xist expression upon differentiation [65]. Because SPEN binds the Tsix promoter region (Figure 2B) [63], the authors postulated a mechanism whereby SPEN silences Tsix, allowing for Xist upregulation upon differentiation. This was also predicted in a previous systems biology analysis of XCI [66]. Indeed, Spen KO differentiating ESCs show increased Tsix expression compared to WT cells, and co-removal of Tsix and Spen leads to the rescue of Xist expression and the formation of Xist clouds [65]. In contrast, a subsequent study showed that rapid SPEN removal by AID-Auxin led to increased Xist expression [67]. The discrepancy in results might be due to the differentiation conditions, different approaches, and/or timing of SPEN removal.

Chromatin remodelers play crucial roles in modulating the compaction and relaxation of chromatin during differentiation and development, facilitating or impeding access of TFs to chromatin and their binding targets. Chromatin remodeler CHD8 binds to the Xist promoter region in undifferentiated ESCs, and its removal by KD or KO in the pluripotent state leads to Xist expression reduction due to reduced chromatin accessibility, suggesting a role for CHD8 in activating Xist [47]. However, its role in Xist expression during differentiation seems more complex since Cdh8 KO cells show a reverse phenotype, i.e., increased Xist expression, seemingly due to increased chromatin accessibility and YY1 binding to the Xist promoter (Figure 2A,B). Which other factors see their recruitment to the Xist promoter region change upon CHD8 removal? is an interesting question that needs further research. All in all, CHD8 seems to fine-tune Xist expression in the pluripotency stage and during differentiation.

Another group of interesting factors involved in Xist expression are RIF1 and KAP1/TRIM28. RIF1 and TRIM28 are versatile proteins: RIF1 is involved in wide-ranging mechanisms, such as replication timing, gene expression, chromatin contacts, and G-quadruplexes [68,69,70], while KAP1/TRIM28 is an epigenetic co-repressor involved in genomic integrity, DNA repair, imprinting, and other processes (reviewed in [71]). Dynamic interplay and recruitment of these factors to the Xist P2 region tilts the balance towards Xist upregulation on one allele or the other. Dynamic RIF1 recruitment is asymmetrically broken by KAP1 binding at the P2 promoter of one allele, leading to higher Tsix steady-state levels in cis and further removal of RIF1 from that allele, preventing Xist from upregulating (Figure 2A,B) [72]. This does not seem to be Xist P2 H3K9me-mediated. On the other allele, however, because KAP1 does not bind to it, Tsix is not upregulated, leading to further stable binding of RIF1 and Xist upregulation. The authors describe data supporting this complex bookmarking model. However, the exact mechanism of KAP1 recruitment to the Xic and its modulation of Tsix levels are still open questions. The concurrent interplay between RIF1, KAP1, Xist, and Tsix leads to a break in the symmetry of Tsix and Xist expression and Xa/Xi choice.

5. Long Non-Coding RNAs That Impact Xist Expression

5.1. lncRNA Xist Repressors

The Xic harbors several lncRNAs that act as modulators of Xist expression. These are spatially segregated into two TADs known as the Tsix TAD and the Xist TAD, harboring Xist activators and repressors, respectively (Figure 3, see Section 6.3). LncRNAs mostly regulate Xist in cis, either through transcription, enhancer-promoter and promoter-promoter interactions, or chromatin modifications. The most studied and known lncRNA that regulates Xist is Tsix, its antisense gene, with which it forms a binary switch. In ESCs, both genes are expressed biallelically, with high and low Tsix and Xist levels, respectively. However, at XCI onset, Tsix is downregulated and switched off from one of the alleles concurrently with Xist upregulation; this chromosome will become the Xi (Figure 1 and Figure 3B). Conversely, on the other chromosome, Tsix stays on, preventing upregulation of Xist; this chromosome will become the Xa. The role of Tsix in Xist expression was elucidated by means of targeted deletions of the Tsix promoter region or insertions to truncate its transcriptional unit, skewing that chromosome for inactivation by upregulation of Xist upon differentiation [14,73,74]. Interestingly, Tsix exerts its role on Xist not by its lncRNA molecule itself but by the act of transcription running through the Xist promoter [75]. Transcription of Tsix was later found to establish a transient repressive chromatin state on Xist, since Tsix truncation leads to increased H3K4me2/3 and H3K9ac deposition around the Xist promoter region, with decreased H3K9me, H3K27me3, and DNA methylation (see Section 6.1) [76,77,78,79,80]. Interestingly, introducing a polyadenylation signal in Tsix just 1 kb downstream of the Xist promoter leads to the same effect, indicating that it is antisense transcription running through the Xist promoter that establishes these repressive chromatin marks, impeding Xist upregulation [80]. On the other hand, forced upregulation of Tsix on one chromosome forces this X chromosome to stay active, while the WT chromosome is always inactivated [74]. The transcriptional interference mechanism was further confirmed and simulated in silico in a later study [4]. Finally, conformational switching between a more compact or relaxed Tsix TAD (see Section 6.3) results in breakage of the Tsix/Xist binary switch, with differential Tsix and Xist expression between the chromosomes [81]. A more compact TAD is associated with higher Tsix expression levels, probably due to more cis interactions with distal regulatory elements. It is now well established that Tsix acts as a repressor of Xist in cis.

Another lncRNA called Xite, located upstream of Tsix, acts as a Tsix-specific enhancer, regulating XCI choice [82]. Deletion of Xite, but not its truncation via a splice acceptor and polyadenylation signal, results in lowered Tsix expression in cis upon exit of pluripotency, forcing skewed XCI of the mutated chromosome, indicating that the promoter of Xite acts as an enhancer of Tsix (Figure 3A). It seems, however, that it does not regulate Tsix at the ESC stage, probably because Tsix’s expression is high and is involved in Tsix expression persistence upon differentiation. Moreover, as one would expect from an enhancer, its role in Xist repression is relatively smaller compared to Tsix since heterozygous deletions of the Xite promoter region skew XCI somewhat towards the X chromosome with the mutation, while Tsix mutants show full skewing.

Deletion of a 245 kb region upstream of Tsix results in decreased Tsix expression and increased Xist expression [83]. Also, deletion of that region in Tsix-Cherry Xist-GFP ESCs where the regular relationship between Tsix and Xist is absent also leads to Xist upregulation, suggesting the presence of a Xist repressor that is Tsix independent in the Tsix TAD. Subsequent analysis of the lncRNA Linx located within that region showed that the Linx promoter (LinxP), not its transcripts or transcription, represses Xist and thus the choice of which X is inactivated (Figure 3A) [83,84]. Given that no long-range contacts were detected between LinxP and the Xist promoter, how this is achieved mechanistically remains unclear. Interestingly, when a 2 kb construct of LinxP was inserted upstream of Xist at different distances, Xist expression was upregulated [83]. It seems thus that LinxP has a dual role as a silencer or enhancer, which is genomic context- or TAD-dependent. Another recent study found a similar, if not identical, lncRNA in the same region, Lppnx, whose promoter deletion led to increased Xist expression from that allele [85]. The authors link this lncRNA to the X controlling element, Xce [86], a region upstream of Xite and Tsix. Depending on the background origin of the X chromosomes in female mice, XCI is not fully random. Some alleles are preferentially inactivated with respect to other alleles. The authors show that OCT4 is required for Lppnx expression and that deletion of Xist intron 1 results in rescue of the phenotype, suggesting that Lppnx acts through Xist’s intron 1. Indeed, deletion of Lppnx leads to decreased recruitment of OCT4 and REX1 to Xist’s intron 1 and Tsix’s DxPas34. Since the promoters of Linx and Lppnx are very close and multiple alternative splice transcripts are generated from that location, it is unclear whether these two transcripts involved in Xist repression and/or Xce are the same or different lncRNAs [87,88]. These results warrant additional research.

Finally, another known gene located between Xite and the protein coding gene Chic1 is Tsx, a lncRNA expressed in many tissues; however, it seems to code for the TSX protein in testis (Figure 3A) [89]. Its deletion in ESCs leads to a mild increase in female and male cells with double and single Xist clouds, respectively, while also showing lower Tsix levels in female ESCs and upon differentiation. However, not much else is known about its mechanism of action.

To conclude, several lncRNAs present in the Tsix TAD act in unison to repress Xist directly or via Tsix. Asymmetries born from their expression and/or interactions with Tsix and Xist lead to Xist upregulation or skewing of XCI upon differentiation of ESCs.

5.2. lncRNA Xist Activators

Several other lncRNAs lie upstream of Xist, contained within the Xist TAD. One of the first ones from this set to be involved in Xist expression is Jpx/Enox, a lncRNA that lies ~10 kb upstream of Xist (Figure 3B). Jpx’s expression increases 10–20-fold in both male and female differentiating ESCs, following a similar pattern of expression as Xist and also escaping XCI [90,91]. RNA FISH experiments indicated that when expressed from the Xi, it follows a similar nuclear localization pattern as genes escaping XCI, with its nascent RNA detected adjacent to Xist accumulations or clouds. Heterozygous deletions of Jpx, however, seem to decrease Jpx expression from the WT allele, indicating a feedback mechanism in trans, while Xist cloud formation was very much abrogated compared to WT cells, suggesting a role for Jpx in activating Xist in trans too during differentiation. Knockdown of Jpx at the posttranscriptional level also led to reduced Xist expression, indicating an RNA-mediated mechanism of action and not through the act of transcription or as an enhancer [91]. Moreover, random transgene insertion of a Jpx-Xist construct in female cells resulted in an increased number of cells with 2–3 clouds, suggesting again a trans mechanism [92]. However, a different study cast doubt on its mechanism working solely in trans [93]. Indeed, a heterozygous deletion of a large piece of DNA upstream of Xist from Jpx to Rnf12 does not compromise Xist upregulation from the WT chromosome, although its levels are reduced. This reduction of Xist expression in the Jpx-Rnf12 heterozygous deletion is not rescued by introducing a Jpx transgene, suggesting Jpx’s effect might be in cis and not in trans. On the other hand, another study revisiting these results concluded that heterozygous deletion of the Xic spanning from Tsx to Ftx (see next paragraph), and thus including Jpx, abrogates XCI [94]. Finally, another group recently reported that Xist expression in post-XCI cells is Jpx-dependent, but nascent Xist transcription is unmodified, suggesting that Jpx acts on Xist at the posttranscriptional level [95]. Why such results contradict each other remains an open question, though these differences might be due to the different background strains for each chromosome, the parental origin of the mutated chromosomes, and/or the differentiation conditions. They concur that Jpx acts as an activator of Xist, but its exact mechanism of action is still debated.

The discovery of another gene upstream of Xist, called Ftx, with no ORF and whose RNA is predominantly nuclear, suggested it might be a lncRNA [96]. Its expression increases during the differentiation of female ESCs, mirroring Xist, while staying mostly unchanged in differentiating male cells. As Jpx, it partially escapes XCI. Ftx removal by knocking out its first five exons and alternative promoters in male cells results in decreased H3K4me2 deposition and increased CpG methylation around the Xist promoter, both of which result in decreased Xist expression [96]. Ftx heterozygous and homozygous deletions in female cells lead to impaired Xist upregulation [97]. Analysis of the 3D conformation of the region shows strong interactions between the Ftx and Xist promoters in female undifferentiated ESCs. Directing a dCAS9 fused to KRAB to the Ftx promoter also results in Xist downregulation, indicating that Ftx transcription or its lncRNA is important in cis for Xist expression (Figure 3B). Similar results were obtained in vivo in a separate study [98]. Deletion of Ftx led to compromised XCI in vivo, with many genes being expressed from the Xi, probably due to diminished Xist expression in many tissues analyzed. It thus seems that promoter-promoter interactions between the Ftx promoter and the Xist promoter in cis drive Xist expression in ESCs and differentiation, in vitro and in vivo [97,98].

As stated above (see Section 2.2), several regulatory elements around 75–150 kb upstream of the Xist TSS have recently been found to positively affect Xist expression [26]. Several relatively short lncRNAs are expressed from that region, newly named Xert (Figure 3B). The main TSS is enriched with typical enhancer chromatin marks, such as H3K27ac and H3K4me1. Xert is upregulated along with Xist, Jpx, and Ftx at the onset of differentiation. This expression profile suggests co-regulation of Xist as Jpx and Ftx in cis. Transient silencing of Xert using CRISPRi indeed leads to decreased Xist expression in differentiating female cells, while its overexpression in male cells results in increased Xist expression. By deleting the Xert promoter region, the authors conclude that its effect on Xist expression is in cis [26]. In addition, increased interactions between Xist and Xert were reported during differentiation, further supporting its role as a Xist enhancer. Although these findings suggest that Xert exerts its effect on Xist upregulation through its enhancer capabilities, a potential role in Xist regulation for the Xert RNA or the act of transcription through the locus remains to be studied.

It is thus clear that several mechanisms are at play that regulate Xist and Tsix expression during the onset of XCI, such as the lncRNA product, the promoter of the lncRNA acting as an enhancer, or transcription running through the locus. These studies exemplify how difficult it is to identify common principles of action among these players. To overcome this challenge, conducting ectopic expression experiments for these lncRNAs, knocking them down, introducing termination signals, etc. are necessary to elucidate their specific roles in Xist regulation.

6. The Chromatin Landscape at the Onset of Random XCI

6.1. Chromatin Landscape at the Tsix/Xist Tandem

As previously discussed, the regulation of Xist expression involves multiple factors that orchestrate changes in the chromatin landscape within the Xic. It is now evident that several histone post-transcriptional modifications (PTMs) contribute to the maintenance of symmetric Tsix expression while repressing Xist expression in the pluripotent state. Upon exit of pluripotency, localized changes in these PTMs within the future Xi assist in the downregulation of Tsix and concomitant upregulation of Xist, ultimately breaking symmetric Tsix expression.

The clearest evidence for the role of histone PTMs in maintaining asymmetric expression came from ChIP studies of the chromatin landscape at Tsix and Xist. Early studies in female trophoblast stem cells (TSCs), which express Xist from the paternal allele, revealed an enrichment of H3K4me2 at the P1 promoter of Xist on the paternal allele. Concomitantly, H3K4me2 is absent at the P1 promoter on the maternal allele [77]. In line with this, a reversed pattern was observed at the Tsix TSS, whereby H3K4me2 is enriched at the TSS of Tsix on the maternal allele while absent from the paternal allele. The same study then explored the role of Tsix in the establishment of H3K4me2. Male mouse ESCs carrying a stop signal near the TSS of Tsix, thus truncating its expression, show a gain of H3K4me2 at the Xist P1 promoter. Moreover, a reduction of H3K4me2 across the Tsix/Xist tandem was also observed, highlighting the role of Tsix in modifying the histone PTM landscape at the tandem (Figure 3A,B). Similar results were shown in a separate study that same year in male MEFs, which do not express Xist, where the introduction of a stop signal near the TSS of Tsix leads to a gain in H3K4me2 at both the P1 and P2 promoters of Xist [79].

Follow-up investigations into Tsix’s mechanism for modifying the chromatin landscape elucidated that it is the transcriptional activity of Tsix across the Xist’s promoters rather than its lncRNA product that instigates these modifications. In accordance with previous findings and as stated in Section 5.1, the insertion of a stop signal near the Xist P2 promoter in male mouse ESCs, thus allowing for 93% transcription of Tsix, shows a gain in H3K4me2 at both the P1 and P2 promoters of Xist [80]. Moreover, a loss in H3K9me2 at both Xist promoters was also observed in the transgenic male lines alongside diminished DNA methylation, pointing towards the maintenance of a complex balance between euchromatic and heterochromatic PTMs by the transcription of Tsix across Xist’s promoter regions in the pluripotent state (Figure 3A).

Further exploration of histone PTMs induced by Tsix expression later revealed a role for H3K36me3 and H3K27me3 in repressing Xist expression. H3K36me3 is enriched across Tsix and highly enriched over the Xist promoter in male mouse ESCs in the pluripotent state. When Tsix is genetically deleted, enrichment of H3K36me3 is lost at the Xist promoters [99]. Interestingly, the same study finds enrichment of H3K27me3 at the Xist promoters in the transgenic lines, perhaps due to the expansion of the H3K27me3 hotspot upstream of Xist (see next section), indicating that Tsix transcription may have a bivalent role at the Xist promoters in maintaining both a euchromatic state by preventing H3K27me3 spread and a heterochromatic state by depositing H3K36me3 (Figure 3A). It is now known that H3K27me3 plays a critical role during iXCI. Early during oogenesis, an H3K27me3 domain is established, spanning Xist and upstream sequences up to Rnf12. This mark is maintained throughout the first cell divisions after fertilization, preventing the upregulation of Xist from the maternal X chromosome [100]. Indeed, its loss results in aberrant maternal XCI in pre-implantation development. The imprint that keeps the maternally inherited Xist allele inactive during mouse iXCI is thus polycomb-dependent H3K27me3 deposition at Xist and its upstream sequences. Finally, a comprehensive ChIP-seq analysis of histone PTM induced by Tsix has confirmed the complex interplay between euchromatic and heterochromatic marks governing the Xist promoter. Ectopic Tsix upregulation with a doxycycline-inducible system in EpiSCs, which have already initiated XCI, leads to diminished enrichment of H3K4me2 and increased enrichment of H3K36me3 at both promoters of Xist [101], confirming previous results. Additionally, this study further expanded on the changes in histone PTMs at the Xist promoters by Tsix. At endogenous Xist expression, the euchromatic marks H3K4me3 and H3K27ac are particularly enriched at the P2 promoter of Xist. Concomitantly, Tsix induction by doxycycline diminishes the presence of the euchromatic marks, resulting in the enrichment of the heterochromatic marks H3K9me2, H3K9me3, H3K27me1, and H4K20me2. It is worth noting that other dynamic chromatic changes may also play a role in the regulation of Tsix. Despite not being the primary focus of these studies, both forced Xist upregulation and exit of pluripotency show a slight gain of H3K27me3 and diminished H3K27ac at the Tsix TSS, most likely via SPEN [26,65,102], indicating a transition towards a heterochromatic state (Figure 3B).

An important mechanism in the epigenetic control of promoters is DNA methylation. The Xist promoter in ESCs is decorated partially with DNA methylation [10]. As with histone marks, Tsix blocks euchromatinization of the Xist promoter in ESCs by triggering DNA methylation in the ESC state. Indeed, the removal of Tsix in male ESCs and in vivo leads to decreased or loss of CpG methylation at the Xist promoter region [78,79,80]. Upon differentiation and XCI, Tsix triggers DNA methylation at the Xist promoter on the future Xa, mediated by the de novo methyltransferase DNTM3A, maintaining it silenced [76]. As stated earlier, Ftx also plays a role in CpG methylation at the Xist promoter, since its deletion results in increased methylation in the ESC state [96]. In iXCI, however, the maternal Xist allele is kept silent in a DNA-methylation-independent mechanism, since deletion of Dnmt3a/b does not result in Xist or Tsix misexpression in extraembryonic tissues [103].

Taken together, these studies suggest both Xist P1 and P2 promoters are decorated with both heterochromatic and euchromatic PTMs, maintaining them in a poised state. At the onset of XCI, the removal of heterochromatic PTMs at these promoters occurs on the future Xi, aiding in the upregulation of Xist. On the other hand, the TSS of Tsix is decorated with euchromatic PTMs in the pluripotent state, aiding in symmetric Tsix expression from both X chromosomes. The onset of XCI then leads to the removal of euchromatic PTMs and later the deposition of heterochromatic PTMs at the Tsix TSS, resulting in Tsix downregulation in the future Xi.

6.2. Chromatin Landscape at Xist Regulators

Although studied to a lesser degree, alterations in histone PTMs outside the Tsix/Xist tandem within the Xic locus may also contribute to the regulation of Xist expression. Early ChIP studies identified an H3K27me3/H3K9me3 hotspot extending ~350 kb upstream of Xist in female mouse ESCs in the pluripotent state [104]. Initially, this discovery was intriguing since heterochromatic marks were conventionally associated with the Xi. The characterization of this H3K27me3 hotspot was further elucidated by follow-up studies in female mouse ESCs with skewed XCI upon exit of pluripotency. ChIP analysis of H3K27me3 female mouse ESCs showed enrichment of H3K27me3 from the 5′ end of the Xist gene up to the 5′ end of Ftx in the pluripotent state, further confirming the previous findings (Figure 3A) [105]. More importantly, the appearance and disappearance of the H3K27me3 hotspot correlate with pluripotency status, whereby exit of pluripotency shows ablation of the H3K27me3 hotspot, hinting at its plausible role in silencing Xist cis activators within the Xic in the pluripotent state.

Upon exit of pluripotency, loss of H3K27me3 at Xist cis activators within the Xic is accompanied by gain of euchromatic marks. Notably, Jpx, Ftx, and Xert (see Section 5.2) show increased H3K27ac and H3K4me3 at their TSSs [26,105]. Moreover, the RE93-97 enhancer region also gains H3K27ac and H3K4me1 (see Section 2.2), whereas another enhancer region, RE79, characterized by H3K27ac enrichment in TSCs, does not show H3K27ac accumulation during rXCI (Figure 3B) [27]. On the other hand, Tsix cis activators within the Xic exhibit a mirrored pattern, undergoing a loss of euchromatic and gain of heterochromatic marks upon exit of pluripotency. In the pluripotent state, the euchromatic marks H3K27ac and H3K4me3 are enriched at the TSS of Linx and across Xite, while Tsx does not show enrichment of these marks at its TSS but rather at an upstream region (Figure 3A) [26,83,102]. Exit of pluripotency shows depletion of these euchromatic marks on the Xi and concomitant gain of H3K27me3, most likely indicating their downregulation upon XCI kickoff (Figure 3B).

Taken together, these studies suggest that Xist cis activators within the Xic are decorated with a heterochromatic H3K27me hotspot in the pluripotent state, which is removed upon onset of XCI, while Tsix cis activators within the Xic show an inverse trend with respect to their chromatin state. Allele-specific re-analysis of publicly available ChIP-seq data in female mouse ESCs both during and upon exit of pluripotency can help us further elucidate the chromatin changes of Xic regulators upon onset of XCI.

6.3. Higher Chromatin Structure of the Xic

Advancements in chromosome conformation capture technologies have enabled the comprehensive assessment of chromosomal interactions, both locally and on a genome-wide scale (reviewed in [106]). Early insights on the higher chromatin organization of the Xic came from a 3C study conducted in mouse ESCs. This study demonstrated that Xist establishes contacts with the Jpx locus in the future Xi, while Tsix establishes contacts with the Xite locus in the future Xa upon exit of pluripotency [107]. It was then proposed that both Xist and Tsix segregate into two distinct regulatory hubs, facilitating the observed asymmetric switch of these lncRNAs during XCI.

This initial notion of a higher-order chromatin separation between Xist and Tsix was later confirmed by 5C and high-resolution microscopy studies [84]. We now know that both Xist and Tsix form two distinct TADs, facilitating their spatial segregation (Figure 3A). TADs are typically ~1 Mb in size and contain loop domains corresponding to recurrent or stable contacts between loci 5 to 500 kb apart [106]. The Xist TAD is ~550 kb in size and spans from the 3′ end of Xist to a region upstream of Rnf12, while the Tsix TAD is ~300 kb in size and spans from a region downstream of Nap1L2 to the 3′ end of Xist. These TADs are separated by RS14 (see Section 3.2), which binds CTCF and forms the TAD boundary [42,84]. Gene expression within each TAD is highly correlated in female mouse ESCs throughout differentiation, indicating that this physical clustering coordinates gene expression patterns throughout development. Deletion of a 58 kb region containing the TAD boundary results in uncoordinated gene expression within each TAD and increased ectopic contacts between them. Interestingly, this deletion does not result in the merging of these two neighboring domains but rather seems to attenuate them, indicating that other elements contribute to defining them [84].

The relevance of the bipartite TAD organization of the Xic in the transition from symmetric to asymmetric gene expression of Xist and Tsix upon onset of XCI was later revealed from subsequent studies in both male and female mouse ESCs. Taking advantage of CRISPR-Cas9, a ~40 kb region was inverted within the Xic spanning from Xist P1 to the promoter of Tsix, thereby relocating both Xist and Tsix alongside their respective promoters into the opposite TAD [108]. This inversion does not seem to attenuate either’s TAD structure or alter their boundary. However, while transgenic lines do not exhibit altered Tsix expression in the pluripotent state, both male and female transgenic lines exhibit ectopic Xist upregulation. Moreover, prolonged Tsix expression throughout differentiation was observed in female transgenic lines. Consequently, these findings collectively suggest that inversion of the Tsix/Xist locus results in a Xist regulatory pattern resembling that of Tsix and vice versa, partly overriding the previously discussed Xist regulation rules such as inhibition by pluripotency and double gene dosage. This highlights the impact of the cis-regulatory landscape on the proper timing of gene expression patterns.

A recent transgenic study has evaluated the importance of the Tsix TAD topology on Xist regulation [109]. A 245 kb region was inverted, spanning from the downstream region of Nap1L2 to the downstream region of Tsx in both male and female mouse ESCs, thus preserving the integrity of the Xite and Tsix loci. Comprehensive 5C analysis of the altered Xic points towards the formation of new but similar loop interactions within the Tsix TAD, altering its topology. Intriguingly, the new Tsix TAD topology leads to a gain of insulation across the Tsix/Xist TAD boundary. Inverted lines show Tsix downregulation in the pluripotent state, highlighting the relevance of the TAD topology on maintaining Tsix expression, although this effect was lost upon exit of pluripotency. Interestingly, both male and female inverted lines display ectopic Xist upregulation upon exit of pluripotency, but not in the pluripotent state, suggesting that changes in one TAD structure can influence the expression patterns of a neighboring TAD, proposing the existence of inter-TAD communication. However, it is likely that Xist’s observed misregulation upon exit of pluripotency is tightly linked to the downregulation of Tsix or other factors from the pluripotent state.

Taken together, these studies underscore the role of the higher chromatin organization within the Xic at the onset of XCI. TADs ensure correct temporal gene expression patterns both within and between neighboring TADs upon exit of pluripotency, guaranteeing Xist upregulation at the appropriate developmental stage.

7. Concluding Remarks

How female cells manage to count the number of X chromosomes per diploid genome and initiate inactivation has puzzled scientists for decades. Increasing amounts of data now support a stochastic model that breaks symmetric Tsix/Xist expression from a single X chromosome in female murine ESCs upon differentiation. Over the last decades, research has disentangled the role of autosomally encoded factors that repress or activate Xist expression and the role of X-linked XCI activators. A tightly regulated balance of trans and cis factors thus ensures a counting mechanism whereby a single X chromosome in diploid female cells upregulates Xist monoallelically, leading to a cascade of events that will eventually silence most genes along that single X. This then leads to the establishment of a negative feedback loop whereby the X-linked XCI activators are silenced on the Xi in cis, preventing inactivation of the remaining active X chromosome in trans. Several factors contribute to this cascade of events to kick off Xist upregulation monoallelically, and thus XCI. We believe these mechanisms work at different hierarchical levels: DNA elements, transcription factors, other regulatory proteins, lncRNAs, and the chromatin and topological landscape. See Supplementary Table S1 for a summary of all factors involved in Xist expression in this review. Disentangling the hierarchical relevance of each element is difficult, as they may cooperate synergistically or redundantly, but also in a timely manner.

Since cells in the pluripotent ESC state have two active X chromosomes, where Xist expression is kept biallelically low or repressed, the connection between the pluripotency factor network and Xist repression was self-evident. Evidence now indicates that the pluripotency factor network and other transcription factors and proteins aid in maintaining biallelic Tsix expression and thus indirectly inhibiting Xist expression in ESCs, while also directly inhibiting Xist. Downregulation of the pluripotency factor network seems to be one of the first cues for XCI kickoff. It is thus essential for the cell to integrate the pluripotency factor network, differentiation cues, and X dosage to establish the correct Xist expression pattern. DNA elements would consequently sense these cues in the form of protein factors binding to them, allowing for subsequent changes in the chromatin landscape and transcriptional outcomes, thus facilitating and locking in an asymmetric Tsix/Xist expression pattern upon differentiation. Moreover, although Tsix has a very distinct role in Xist inhibition, Xic lncRNAs generally seem to have a more modulating function in symmetry breaking between Xist and Tsix, with variability in their mechanism of action, be it the lncRNA molecule self, transcription, or promoter and enhancer sharing. The local topology of the Xic indicates its importance in the interplay between these different levels. Finally, further levels of Xist post-transcriptional control that fall outside the scope of this review are alternative splicing and Xist RNA molecular stability. Alternative splicing of Xist leads to a long and short isoform, both functional in XCI during the differentiation of ESCs [13], while Xist RNA stability modulation plays an important role in the downstream cascade of XCI [17,110].

While this review focuses on mouse rXCI, it is worth noting that differences have been described between mechanisms regulating XCI in different species. For example, while Tsix has a clear role in murine rXCI, this is not the case in humans, possibly due to the fact that the TSIX gene body does not fully overlap with that of XIST, and thus its transcription does not run through the XIST promoter. Moreover, female human cells in pre-implantation embryos display two clouds of XIST in vivo, while in vitro-grown human pluripotent stem cells display one or two XIST clouds depending on derivation procedures and culture conditions (reviewed in [1]). However, as with other mammals, human differentiated cells show a single inactive X chromosome with a single XIST cloud. Additionally, it has been shown that human FTX does not seem to play a role in XIST regulation, while JPX does regulate XIST through the act of transcription, possibly involving promoter or enhancer co-activation, and does not involve lncRNA-mediated regulation of XIST as described for mice [95]. Altogether, this highlights the similarities and differences between mouse and human XCI. It is also worth noting how, even within rodents, patterns of XCI and Xist expression may differ between species [111], illustrating the variety of mechanisms that lead to dosage compensation in mammals.

While more than three decades have passed since its discovery, understanding how Xist is regulated at the different levels of gene regulation explained in this review is still an ongoing task. Current open questions that warrant further research remain, such as the actual role of Jpx in Xist expression modulation and the differences and similarities between Linx and Lppnx. Are there also additional Xist and XCI activators? More of them will probably come to light in the future. It is also interesting to hypothesize whether any of the lncRNAs described in this review contain small ORFs that might be functional [112]. In vivo studies seem to point to Xist regulation as a highly redundant process. In other words, while in vitro deletions of cis elements or trans regulators misregulate Xist expression in mouse ESCs, in vivo deletions of several of these factors seem primarily redundant, suggesting the presence of very robust overlapping mechanisms. Given how essential XCI is for cell survival during female development, the presence of several redundant, “fool proof” mechanisms seems evolutionarily sensible.