Summary

Introduction

X chromosome inactivation (XCI) is a process by which mammals reduce the number of active X chromosomes to one per diploid set of autosomes, thereby allowing for dosage equality between the sexes. Normal female cells, with two X chromosomes, are where XCI is most commonly observed. However, it can take place in any cell, male or female, that has more than one X chromosome. During mouse embryogenesis, female cells undergo two separate XCI events. The first, imprinted XCI, is characterized by the inactivation of the paternal X chromosome (Xp) in all cells. In the late blastocyst, cells within the inner cell mass (ICM) reactivate the Xp before a second round of inactivation occurs, called random XCI (Mak et al., 2004; Okamoto et al., 2004). In random XCI, either the Xp or the maternal X chromosome (Xm) is subject to inactivation, and this stochastic choice appears to be made independently in each cell (Lyon, 1961). Once complete, the inactive chromosome is maintained as such throughout the cell lineage (Lyon, 1961). Thus, cells derived from progenitors with an inactive Xp will have an inactive Xp, and those derived from a cell that has an inactive Xm will maintain an inactive Xm. Because gene silencing that occurs from XCI is so stable, it is used as a paradigm for epigenetic gene regulation.

Random XCI can be thought of as occurring in four stages: initiation, spreading, maintenance and reactivation. During initiation, a cell determines how many X chromosomes need to be inactivated to achieve a ratio of one active X chromosome per diploid set of autosomes and identifies which specific chromosomes to inactivate. These processes are called counting and choice (see Glossary in Box 1 for more information). Once a chromosome is designated for inactivation, transcriptional silencing spreads, reducing the expression of almost all of its genes. The inactive status of an X chromosome is then maintained throughout the cell lineage with the exception of the primordial germ cells (PGCs), where it is reactivated by embryonic day (E) 12.5 (Chuva de Sousa Lopes et al., 2008; Kratzer and Chapman, 1981; Monk and McLaren, 1981). In these cells, the epigenetic marks established during XCI are erased and formerly inactivated genes are re-expressed.

Since random XCI was first postulated in 1961, researchers have struggled to identify the mechanisms for counting and choice. The earliest information came from observations of X aneuploid cells and polyploid cells. Because cells with different autosomal ploidies had different numbers of active X chromosomes, the autosomes have been implicated in the counting process. For example, diploid cells almost always have a single active X chromosome, whereas tetraploid cells maintain two active X chromosomes. This led to the notion that a cell maintains one active X chromosome per diploid set of autosomes.

More recent information about how cells count and choose active X chromosomes comes from studies in which sequences within the X inactivation center (XIC) were modified (see Glossary in Box 1). Most of the modifications affected expression of the non-coding RNAs Xist (Borsani et al., 1991; Brockdorff et al., 1991; Brown et al., 1991a) and Tsix (see Box 2 for more information) (Lee et al., 1999a). The Xist and Tsix genes are antisense to each other and are transcribed at low levels prior to XCI. However, during the initiation stage of XCI, Xist and Tsix assume opposite fates on the X chromosomes. Xist is upregulated and its RNA transcripts coat the entire inactive X (Xi) chromosome (Panning et al., 1997; Sheardown et al., 1997), while Tsix is repressed (Lee et al., 1999a). By contrast, increased levels of Tsix transcription repress Xist on the active X (Xa) chromosome (Lee et al., 1999a; Luikenhuis et al., 2001; Shibata and Lee, 2004).

Mutations within the XIC, in the form of large deletions or directed modifications to Xist or Tsix, can pre-determine which X chromosome is to be inactivated or can prevent XCI entirely. Thus, these mutations either directly affect counting and choice mechanisms or override them. Most experimental mutations are performed using mouse embryonic stem (ES) cells, which, when induced to differentiate, are thought to recreate random XCI, as seen in the developing embryo. Mutations in the XIC that either disrupt or bypass counting cause the X chromosome in male cells to be inactivated or prevent XCI in otherwise normal female cells. Mutations that only disrupt or bypass choice do not cause ectopic inactivation in male cells but, in cells with multiple X chromosomes, they ensure the inactivation of either the mutated X chromosome in all cells or the wild-type X chromosome in all cells.

In this Hypothesis article, we review the observations and experiments that shed light on the underlying mechanisms for counting and choice. We then describe the models of XCI that have been developed to interpret these data (Figs ​(Figs1, 1, ​,2)2) and discuss experimental results that they do not account for. Finally, we present our own model, which aims to also account for a number of published experimental results that are not incorporated into these other models. It is our hope that this new model will inspire a re-interpretation of XCI data, as well as future experiments and additional models.

Open in a separate window

Fig. 1.

Three models of X chromosome inactivation. (A) The blocking factor model. (a) Diploid autosomes work together to create a single blocking factor (blue shape), which can bind to only one X chromosome, preventing it from inactivating. (b) Nicodemi and Prisco used computer simulations to show that if the autosomes produce a swarm of blocking factors (blue dots) that can bind to each other and to the X chromosome, then all of the blocking factor molecules will accumulate on a single X. (B) The two factor model. Autosomes produce blocking factors and X chromosomes produce transacting competence factors (red shape). Blocking factors bind to competence factors with a two to one stoichiometry and then bind to one X chromosome; the remaining competence factor binds to the other X, which inactivates. (C) The sensing and choice model. After cells start to differentiate, the two Xpr regions (yellow) pair and upregulate Xist (red) on both X chromosmes. The Tsix/Xite region (blue) pairs and chooses which chromosome to inactivate, and represses Xist on the other. The X chromosome that represses Xist remains active (green), and the other becomes inactive (red).

Open in a separate window

Fig. 2.

Stochastic model of X chromosome inactivation. (A) The various components of the stochastic model. (B,a) Prior to cell differentiation, the stochastic model proposes that the autosomes produce a trans-acting factor (purple stars) that induces Tsix expression on both X chromosomes (blue shape). (b) Tsix, when bound by autosomal trans-acting factors, represses Xist transcription (red shape). (c) After differentiation, an X-linked locus (yellow triangle) produces a trans-acting factor (red circles) that attempts to upregulate Xist. (d) Competition between Tsix- and Xist-promoting factors creates a probability for each X chromosome to inactivate. Cells that do not inactivate either X chromosome will continue to produce the factor that promotes Xist upregulation in subsequent cell cycles. Cells that inactivate both X chromosomes will either die, or reactivate one. The percentages shown here reflect the proportions of each XCI configuration observed 7 days after differentiation. Xa, active X chromosome; Xi, inactive X chromosome.

What we know about counting, choice and the loci involved

Early evidence of the occurrence of counting during random XCI indicated that autosomes play a role in this process (Lyon, 1972). This conclusion came from the number of Xa chromosomes that are observed in cells with different autosomal ploidies (see Table 1). In general, the more sets of autosomes that are present in a cell, the more Xa chromosomes it will contain. For example, diploid mouse cells maintain one active X chromosome after random XCI, whereas tetraploid cells maintain two active X chromosomes. The difference in the number of Xa chromosomes between diploid, triploid and tetraploid cells indicates that the autosomal ploidy affects counting. Human diploid cells can inactivate up to four X chromosomes in order to create a ratio of one active X chromosome per diploid set of autosomes (Gartler et al., 2006; Grumbach et al., 1963). Although it is not known how autosomes influence counting, one common idea is that they produce trans-acting signals that interact with the X chromosomes.

When considering the data presented in Table 1, it is important to understand that secondary cell selection can affect these results and their interpretation. For example, it is possible that a subpopulation of cells in the embryo fail to execute XCI properly and inactivate too many or too few X chromosomes. These cells would be selected against, making them difficult to detect. Takagi and colleagues have shown that cell selection is a significant force in the long-term phenotype of mouse embryos following random XCI (Takagi et al., 2002). Data from Monkhorst et al. also support this conclusion (Monkhorst et al., 2008). They examined diploid XX mouse ES cells 3, 5 and 7 days after the induction of differentiation, and observed all possible XCI combinations, from no Xa chromosome to two Xa chromosomes being present, at all time points. However, the percentage of cells with unconventional XCI patterns decreased over time, demonstrating that these cells were selected against or had modified their inactivation pattern.

It is generally accepted that random XCI initiates from a single region on the X chromosome rather than from multiple regions. Intuitively, this was thought to be the case because XCI is not heterogeneous in the sense of one X chromosome having a mosaic pattern of active and inactive regions that are complementary to the other X chromosome (Gartler and Riggs, 1983). This was later confirmed by translocation studies in both mice and humans (Brown et al., 1991b; Rastan, 1983). The single region on the X chromosome that is responsible for initiating random XCI is called the X chromosome inactivation center (XIC; see Glossary, Box 1). Since the identification of the XIC, there have been many deletion and sequence replacement studies that have aimed to identify the internal elements that directly affect counting or choice (Fig. 3; Tables ​Tables2, 2, ​,33).

Open in a separate window

Fig. 3.

Part of the mouse X chromosome inactivation center and sequence replacements and deletions within it. Sequences used to replace portions of the XIC are shown above the diagram, which shows their relative locations. The effects of these replacements are listed in Table 3. Below the diagram, the dotted lines show XIC regions that have been deleted. The effects of these deletions are listed in Table 2.

In one of the earliest X chromosome deletions, X^Δ65kb (del-pBA X) (Clerc and Avner, 1998; Morey et al., 2004), a region from the 3′ Xist exons to the 3′ end of Chic1, was removed (see Fig. 3). All X chromosomes with this deletion, including those in X^Δ65kb:X, X^Δ65kb:0 (X0; see Glossary, Box 1), and X^Δ65kb:Y ES cells, become inactivated after differentiation. The inactivation of the X^Δ65kb chromosome in X^Δ65kb:X and X^Δ65kb:Y ES cells was assessed by the appearance of Xist coating and by the failure for RNA-FISH (fluorescent in situ hybridization) probes to bind to Mecp2 (methyl CpG binding protein 2) and Chic1 (cysteine-rich hydrophobic domain 1) transcripts, both X-linked genes. This indicates that the X^Δ65kb deletion disrupts or bypasses counting, as the single X chromosome was inactivated in X^Δ65kb:0 and X^Δ65kb:Y cells, and disrupts choice, because X^Δ65kb:X cells always inactivate the mutant chromosome. The results from the X^Δ65kb:0 and X^Δ65kb:Y cells suggest that this region contains elements that prevent or repress ectopic XCI.

To refine the XIC region that affects counting and choice, researchers systematically re-introduced genomic fragments into the 65-kb deletion. The first modification, X^Δ65kb+16kb, returned 16 kb, including the 3′ Xist exons and Tsix, but not Xite, to the X^Δ65kb deletion (see Fig. 3) (Morey et al., 2001). X^Δ65kb+16kb:X ES cells preferentially inactivate the mutant chromosomes, indicating that choice remains disrupted in these cells. Because only X^Δ65kb+16kb:X cells were examined, it is not clear whether this insertion restores counting. However, once 37 kb was added back to the original deletion, X^Δ65kb+37kb (Morey et al., 2004), returning both Tsix and Xite to the mutant X chromosome (see Fig. 3), normal counting was observed in X^Δ65kb+37kb:0 ES cells. These results indicate that the 37-kb region between the 3′ Xist exons and the start of the Xite gene contains elements that prevent ectopic XCI.

There is controversy over whether eliminating antisense Tsix transcription through the Xist locus influences or bypasses counting. For example, one mutation, X^{Tsix stop} (Ma2L), which inserts a transcriptional stop signal into Tsix before it overlaps with Xist (see Fig. 3), has been reported to cause both low (Luikenhuis et al., 2001) and high (Vigneau et al., 2006) levels of ectopic XCI in male ES cells. Furthermore, Vigneau and colleagues published results from two additional XIC modifications, X^ΔDXPas34 (Δ34) and X^TsixΔmajor (ΔAV) (see Fig. 3), which reduce or eliminate Tsix transcription before the RNA polymerase enters the 3′ Xist exon on the antisense strand. The X^ΔDXPas34 deletion removes the DXPas34 fragment, a 1.6-kb CG-rich tandem repeat within Tsix. The X^TsixΔmajor deletion removes the DXPas34 fragment and the major promoter of Tsix. Vigneau et al. reported that mutant X chromosomes that carry either deletion in male ES cells inactivated after differentiation (Vigneau et al., 2006). In contrast to these experiments, there are two separate Tsix mutations that prevent or severely reduce transcription across Xist, but do not affect counting. Neither the X^{Tsix IRESβgeo} mutant (Tsix^AA2Δ1.7, ΔTsix) (Ohhata et al., 2006; Sado et al., 2001), which replaces the second exon of Tsix after the major promoter with an IRESβgeo cassette (see Fig. 3), nor the X^TsixΔCpG mutant (Lee, 2005; Lee and Lu, 1999), which replaces the major promoter of Tsix, the following exon, and the DXPas34 region with a Pgk-neo cassette (see Fig. 3), caused inactivation in male embryos. The results of Ohhata et al. are particularly compelling because they were derived from embryos that carried the Tsix mutation. However, more work needs to be done to resolve this issue.

Some of the most interesting data has come from Jeanie Lee's homozygous X^TsixΔCpG ES cells (see Fig. 3) (Lee, 2005). After differentiation, X^TsixΔCpG:X^TsixΔCpG cells appear to assume one of three XCI patterns. Some cells execute XCI normally and have one active X chromosome; the remaining cells either inactivate both X chromosomes or have two active chromosomes. The cells with one inactive X chromosome appear to undergo random XCI with unbiased choice.

There have also been several mutations within the XIC that affect Xist transcription. Penny and colleagues created X^{XistΔpromoter} mutant chromosomes (see Fig. 3) by replacing the Xist promoter sequence with a Pgk-neo gene (Penny et al., 1996), which prevented all Xist transcription. In X^{XistΔpromoter}:X ES and chimeric embryonic cells only the wild-type X chromosome is inactivated. In order to compare the role of the Xist promoter sequence with that of Xist transcription in promoting gene silencing, Marahrens et al. created X^XistΔ1-5, in which the first five Xist exons were replaced with a neo gene while the promoter sequence was left intact and functional (see Fig. 3) (Marahrens et al., 1997). In X^XistΔ1-5:X mice, only the wild-type X chromosome inactivates, and in X^XistΔ1-5:Y and X^XistΔ1-5:0 mice, ectopic XCI is not observed. Most recently, Monkhorst and colleagues removed Xist, Tsix and Xite from the X chromosome, producing the X^ΔXTX mutation (see Fig. 3) (Monkhorst et al., 2008). They then examined the influence of this mutation in X^ΔXTX:X ES cells, as well as in X^ΔXTX:X and X^ΔXTX:Y mice. In all cells examined, the mutant remained active and the wild-type X chromosome became inactive after random XCI, showing that choice is either disrupted or bypassed by this deletion.

In addition to the XIC mutations that aim to eliminate Xist, Tsix and Xite transcription, there have also been experiments that have increased the expression or dosage of these genes. Mutations that increase Xist transcription cause the X chromosomes with these mutations to be preferentially inactivated (Nesterova et al., 2003). By contrast, mutations that increase Tsix transcription prevent XCI on the mutant chromosome (Luikenhuis et al., 2001). Using a multicopy transgene approach, Lee performed experiments to increase the dosage of Tsix and of Xite. The 5′ end of Tsix, or a large Xite fragment, when present on an autosome in multiple copies could prevent XCI (Lee, 2005); in female ES cells, both X chromosomes remained active after differentiation. Because these transgenes were inserted into autosomes, it is possible that they titrate trans-acting XCI signals.

Recent publications have described that pairing between X chromosomes takes place before XCI (Augui et al., 2007; Bacher et al., 2006; Xu et al., 2006). Pairing between the Tsix/Xite region was observed first (Bacher et al., 2006; Xu et al., 2006), followed by the identification of a region that is 200-kb upstream from Xist, called Xpr (see Box 2), which pairs slightly earlier, after ES cells begin to differentiate (Augui et al., 2007). X chromosome pairing at the Tsix/Xite region is not observed in X^Δ65kb:X or in X^TsixΔCpG:X^TsixΔCpG ES cells (Bacher et al., 2006; Xu et al., 2006). Interestingly, X^TsixΔCpG:X and X^ΔXite:X ES cells show pairing at the Tsix/Xite region (Xu et al., 2006), as do X^Δ65kb+16kb:X ES cells. Although it is not known how X^TsixΔCpG:X pair when X^TsixΔCpG:X^TsixΔCpG do not pair, the differences in pairing amongst X^TsixΔCpG:X^TsixΔCpG, X^Δ65kb+16kb:X and X^Δ65kb:X cells suggest pairing elements exist within and upstream of Tsix. Xu et al. suggest that Ctcf (CCCTC-binding factor) binding to sites within Tsix and Xite mediate pairing by attracting the Ctcf DNA-binding protein, and additional unknown binding proteins, to the region. The Xpr region has been shown recently to pair as an ectopic, single-copy transgene (Augui et al., 2007). Xu et al. and Augui et al. speculate that pairing at the Tsix/Xite region is part of the mechanism for choice. Augui et al. have also hypothesized that Xpr pairing is involved in determining the total number of X chromosomes in the cell.

Current models for XCI counting and choice

The two-factor model

The two-factor model (Fig. 1B) was originally proposed by Gartler and Riggs (Gartler and Riggs, 1983) and later adopted by Lee to interpret XCI experiments (Lee, 2005; Lee and Lu, 1999). This model relies on a BF that prevents XCI, as well as on a competence factor, C, that initiates XCI. Cells start with two competence factors, produced by the X chromosomes, that function in trans. One C, however, is neutralized by the BF. According to this model, each X chromosome binds either BF, to remain active, or C, to initiate XCI, in a mutually exclusive way (Lee, 2005). Mutually exclusive binding is proposed to be mediated by pairing in the Tsix/Xite region, and mutations that disrupt this pairing are thought to allow both BF and C to bind to the same chromosome (Lee, 2005; Xu et al., 2007; Xu et al., 2006).

One problem with this model is that it does not account for the X^TsixΔCpG:X^TsixΔCpG ES cell data in which antisense Tsix transcription across Xist is severely reduced on both X chromosomes. X^TsixΔCpG:X^TsixΔCpG mutants produce three different cell populations: one with normal XCI, a second in which both X chromosomes remain active, and a third in which both X chromosomes become inactive. The two-factor model, however, can only predict two of the three outcomes. In an X^TsixΔCpG:X^TsixΔCpG cell, one possibility is that BF binds to one X chromosome and C binds to the other, just like in a wild-type cell. This would result in normal XCI, and this is seen in a fraction of the mutant ES cells after differentiation. The other possibility is that both BF and C bind to the same X chromosome and nothing binds to the other. In this case, both chromosomes may remain active, or both may inactivate, but the model cannot account for the two different outcomes occurring simultaneously, as is seen in Fig. 4.

Open in a separate window

Fig. 4.

A problem with the two factor model. After XCI, X^TsixΔCpG:X^TsixΔCpG ES cells make three cell populations: one with a single active X (Xa) chromosome and a single inactive X (Xi) chromosome; one with two Xa chromosomes; and one with two Xi chromosomes. (A) Cells with normal XCI result from blocking factor (BF) binding to one X chromosome and the competence factor (C) binding to the other. (B) Cells with two Xa chromosomes and two Xi chromosomes result from BF and C binding to the same X chromosome. However, it is not clear how having both BF and C bound to a single X chromosome results in these two different populations of cells.

The two-factor model needs to provide more details about how BF and C bind to X chromosomes. Without knowing how BF and C bind to X chromosomes, it is unclear how this model accounts for the X^TsixΔCpG:X ES cell data. In these cells, the chromosomes pair and the mutant X chromosome is always chosen for inactivation (Lee and Lu, 1999; Xu et al., 2006). With pairing, the two-factor model predicts that the BF will bind to one X chromosome and C will to bind to the other, but it is currently unknown why C would never bind to and inactivate the wild-type X chromosome, even though it contains all of the necessary binding sites.

The two-factor model also needs to explain how BF and C are produced in quantities that prevent stochastic variations in their levels from resulting in ectopic XCI in XY and X0 cells, or a lack of XCI in XX cells. Small and biologically plausible fluctuations in either BF or C concentrations could cause XY and X0 cells to have more C than BF, increasing the likelihood of ectopic XCI occurring, similar to that seen in X^Δ65kb:0 and X^Δ65kb:Y ES cells. Likewise, XX cells might produce only one C molecule, which is then neutralized by BF, and then fail to undergo XCI entirely. The two-factor model also needs to explain why, in XY and X0 cells, C is always neutralized by BF before it binds to the X chromosome. One would expect the transformation of C into BF to fail from time to time, resulting in ectopic XCI.

A novel feedback model

We have developed a novel model, called the feedback model, which combines aspects of older models and takes into account implications of recent XCI data. Our intention is to develop a model that can account for the results of experiments that have studied XCI in X^ΔXTX:X, X^Δ65kb:X, X^TsixΔCpG:X^TsixΔCpG, X^TsixΔCpG:Y, X^TsixΔCpG:0 and X^{Tsix IRESβgeo}:Y ES cells and mice, because no one model has accomplished this as yet. The feedback model consists of three major components (Fig. 5): a signaling feedback loop, a mechanism for choice, and a description of X-linked loci that directly affect XCI initiation.

Open in a separate window

Fig. 5.

The feedback model of X chromosome inactivation. (A) A description of the components required for the model. (B,a) The active X chromosomes produce a trans-acting signal, A, that saturates specific sites on the autosomes. (b) Once saturated, the autosomes produce a swarm of inactivation signals, I. These signals bind to each other and to XCI inhibitors on the X chromosomes. Once all of the XCI inhibitors on an X chromosome are sufficiently bound by I, the XCI initiator induces inactivation. (c) With only one active X chromosome producing A, the autosomes are no longer saturated with A and stop producing I.

X-linked loci that directly influence XCI initiation

The third component of the feedback model consists of specifying, in general terms, at least one locus within the XIC that initiates X inactivation, XCI_init, and several loci that repress the activity of the XCI_init locus, XCI_repress (Fig. 5A). One candidate for the XCI_init locus is Xist because it is expressed early during XCI initiation and coats the entire X chromosome that becomes inactivated. Reasonable candidate XCI_repress loci are both Tsix and Xite (Lee, 2005; Lee et al., 1999a; Lee and Lu, 1999; Ogawa and Lee, 2003; Sado et al., 2001). In XY ES cells and male mice, deleting Tsix (Lee and Lu, 1999) or Xite (Ogawa and Lee, 2003) does not result in XCI; however, the single X chromosome undergoes XCI when both Tsix and Xite are removed (Clerc and Avner, 1998). Furthermore, the overexpression of Tsix (Luikenhuis et al., 2001; Stavropoulos et al., 2005) inhibits XCI, as do multicopy transgenes that contain Tsix and Xite (Lee, 2005).

In the feedback model, an X chromosome inactivates when the effects of all the XCI_repress loci over the XCI_init locus are removed (Fig. 6). This can be due to XCI_repress loci being bound by the I signal, or to deletions that remove XCI_repress loci or other mutations that prevent XCI_repress from functioning normally. Because the model assumes that there are multiple XCI_repress loci in each XIC, the deletion or mutation of only one will predispose that X chromosome to inactivation, as it would require less I signal to inhibit the remaining XCI_repress loci, but would not cause ectopic XCI. The deletion or mutation of all XCI_repress loci in a single XIC would ensure the inactivation of that X chromosome, even in cells with only one X chromosome.

Open in a separate window

Fig. 6.

Three different paths to X inactivation. (A) In a normal X chromosome, all of the XCI_repress loci must be turned off by I. (B) In a mutant X chromosome where some of the XCI_repress loci have been removed, only a small amount of I is required to initiate XCI. Because these chromosomes have a lower threshold to overcome before initiating XCI, there is a higher probability that they will inactivate. (C) In a mutant X chromosome where all of the XCI_repress loci have been removed, XCI will initiate without any I.

The feedback loop between X chromosomes and autosomes can reproduce the counting that takes place in normal cells, and in cells with unusual numbers of X chromosomes and autosome ploidy. Male cells, for example, with only one X chromosome cannot create enough of the initial A signal to invoke inactivation. In diploid cells with more than two X chromosomes, the signaling feedback loop inactivates all but one X chromosome because the autosomes would continue to be saturated with A until only one X chromosome remained active. In polyploid cells, the additional autosomes would require more A before producing I in sufficient quantities to inactivate an X chromosome, thereby allowing more Xs to remain active. Thus, the feedback model can reproduce the counting that takes place in normal, X chromosome aneuploid and polyploid cells.

The feedback model explains the results from the X^Δ65kb, X^Δ65kb+16kb, and X^Δ65kb+37kb deletions. The X^Δ65kb deletion, which removes both Tsix and Xite and causes the mutant X chromosome to inactivate in all cells, potentially removes all of the XCI_repress loci. Thus, one would expect the XCI_init locus to act in an uninhibited way and to inactivate the X chromosome, regardless of the number of other active or inactive X chromosomes present in the cell (Fig. 6). The X^Δ65kb+16kb deletion, which removes only one XCI_repress locus, Xite, but not Tsix, does not cause a failure in counting, but does cause the mutant X chromosome to inactivate in X^Δ65kb+16kb:X ES cells. Under the feedback model, removing only one XCI_repress locus from the mutant X chromosome would only skew choice towards inactivating the mutant, it would leave counting unchanged (Fig. 6). This is because at least one other XCI_repress locus would inhibit ectopic XCI, allowing for proper counting, but the reduction of overall repression would limit the amount of the I signal required to inactivate the mutant X chromosome. The X^Δ65kb+37kb deletion, which does not remove either Tsix or Xite from the mutant X chromosome, does not disrupt counting or choice. The feedback model would interpret these results as suggesting that the X^Δ65kb+37kb deletion leaves all XCI_repress loci intact.

The feedback model can also explain the chaotic counting that is observed in X^TsixΔCpG:X^TsixΔCpG ES cells. In mutant cells in which both X chromosomes remain active, neither X is bound by sufficient quantities of the I signal to inactivate because the X chromosomes did not pair. Cells with one or two inactive X chromosomes result from the reduction of Tsix transcription across Xist. Because Tsix transcription can repress Xist transcription, it is one of several potential XCI_repress loci. With one fewer XCI_repress locus, X^TsixΔCpG chromosomes would require less I to initiate inactivation than would wild-type X chromosomes. In these cases, smaller clusters of I could bind to one or both Xs, even in conditions that prevented the formation of a single, large macromolecule.

The results from the X^ΔXTX deletion are also easily understood with the feedback model. By removing Xist, and thus the XCI_init locus, from an X chromosome, the mutant would not inactivate, but would still be able to produce the A signal. Thus, normal counting should take place in heterozygous cells, but choice should be skewed; the mutant X chromosome should remain active in all cells.

It is possible that the Xpr pairing is the signal to synthesize A, and that the gene that encodes A is in this region. However, it is more likely that A is the product of more than one locus. Although the locus in the Xpr may be relatively dominant, because a single ectopic copy is sufficient to drive XCI in XY cells, transgenes that include Xist (Herzing et al., 1997; Lee et al., 1999b; Lee et al., 1996) but that do not include the Xpr can also induce ectopic XCI, but multiple copies need to be present. Perhaps these transgenes contain weaker A-producing loci that need to be present in sufficient numbers in order to drive XCI without an additional Xpr region. Thus, given a large number of Xist-containing transgenes, an XY cell might generate enough of the A signal to generate the I signal. The cell's response to the I signal, however, would depend on the structure of the transgene. Transgenes that also contain XCI_repress loci would probably bind the I signal because of the large number of binding sites they would provide. This idea is supported by Lee's transgene experiments with a multicopy transgene that contains Tsix and Xist, which prevented XCI in XX cells (Lee, 2005). Transgenes that contain Xist but not XCI_repress loci might not bind the I signal, thereby allowing the I signal to bind to the wild-type X chromosome at levels that depend upon the presence or absence of Tsix/Xist pairing: with pairing, one would expect a high level of binding; without pairing, one would expect a relatively low level of binding.

Conclusions

Since random XCI was first proposed in 1961 as the means for ensuring that only one X chromosome is active per diploid set of autosomes, scientists have used models to try to understand their experimental results. Here, we present the feedback model, which was designed to maximize the number of experimental results that it could account for. Because it incorporates Lyon's inter-chromosomal feedback signaling, we hope that there will be a renewed interest in searching for trans-acting signals that might originate from the X chromosomes and that might be received by autosomes prior to the initiation of XCI. Our model also suggests that multiple loci are involved in repressing the initiation of XCI, and not a single locus. This concept allows the model to reconcile data from different laboratories that would otherwise be contradictory. It is our hope that the feedback model provides a clearer picture of how X chromosome mutation and deletion data should be interpreted, and that it will encourage researchers to develop hypotheses for future experiments.

Notes

The authors would like to thank A. Fedoriw and M. Calabrese for thoughtful feedback and suggestions. T.M. is supported by the NIH and J.S. is supported by the NIEHS. Deposited in PMC for release after 12 months.

References