Studies have shown considerable individual variability in the level of expression of genes (e.g. 12). In general, however, humans cannot tolerate imbalances for expression of substantial numbers of genes, as demonstrated by the lethality of the majority of chromosomal aneuploidies. Aneuploidy for the sex chromosomes is better tolerated, being observed in approximately 1/500 births 3, presumably because all but one X chromosome is inactivated. X chromosome inactivation ensures the dosage equivalence of X-linked genes between females who have two X chromosomes and males who have a single X chromosome and the sex-determining Y chromosome 4. However, more than 15% of human X-linked genes escape inactivation, being expressed from both the active and inactive X chromosome 5. While such an escape from inactivation may maintain dosage equivalence for X-linked genes with Y homologs, the majority of human genes that escape inactivation no longer have functional Y equivalents, and thus may show relative overexpression in females (reviewed in 6). Substantially fewer genes have been shown to escape inactivation in mice. Although this species difference in expression could reflect less extensive murine expression surveys, a reduced number of genes escaping inactivation is supported by the less drastic phenotype caused by monosomy of the X chromosome in mice (reviewed in 7). In humans, the over or under-expression of genes that escape inactivation is a major contributor to the phenotypes associated with X chromosome aneuploidies, but may also contribute to expression differences between chromosomally normal males and females (e.g. 8).

The study of genes that escape X inactivation can provide insight into such phenotypes, as well as contributing to our understanding of epigenetic silencing mechanisms. Inactivation occurs early in mammalian development, and the stable silencing of the X chromosome involves the acquisition of many features of heterochromatin. It is not known if escape from inactivation is a resistance to the initial silencing event, or rather reflects a high frequency of reactivation of an initially silenced gene, as data appear to support both possibilities. Many genes that escape inactivation in humans are clustered together, which may be indicative of regions that are resistant to the initial signal (e.g. 9). However, analysis of Smcx, one of the few mouse genes expressed from the murine inactive X, has shown reactivation of the gene during early development 10. Surprisingly, recent results have shown that Smcx has a histone modification pattern suggested to demarcate biallelically rather than monoallelically-expressed (imprinted and other X-linked) genes 11, suggesting that the gene is committed to escape inactivation prior to undergoing inactivation.

In addition to the genes that are subject to, or escape from, inactivation, there are some human genes that show heterogeneous X chromosome inactivation, being expressed from the inactive X in some females, but silenced on the inactive X in others 51213. Such genes provide an opportunity to study the same region when silent or active on an inactive X chromosome; and can thus provide insights into the features allowing expression from the inactive X chromosome. The human inactive X chromosome maintains its silent status when isolated in a mouse/human somatic cell hybrid, providing a model system to study the inactive X chromosome apart from its active counterpart. The largest survey of gene expression from the inactive X chromosome 5 analysed expression in a panel of nine inactive-X containing hybrids. That study defined heterogeneous inactivation as expression in three to six of the nine hybrids, which was observed for 60 of the 624 X-linked genes analysed. This variable expression is not restricted to the hybrid system, but has also been demonstrated in cells from females 51213.

We now report the further characterization of one of these genes, the X-linked tissue inhibitor of metalloproteinases, TIMP1, located in Xp11.23. Our previous studies have demonstrated that TIMP1 is variably expressed from the inactive X in both somatic cell hybrids and human females. When TIMP1 is expressed from the inactive X, flanking genes (including ARAF1 and ELK1 which lie ~20 and 55 kb from TIMP1, respectively) are not expressed, suggesting that expression is being controlled in a gene-specific rather than regional fashion 12. Quantitative RNase protection assays showed substantial variability in expression levels from the active X, precluding using expression levels to determine inactive X expression in females. Further studies in hybrids demonstrated that TIMP1-expressing clones were unstable and that methylation does not appear to be the principle controlling feature allowing variable expression of TIMP1 from the inactive X chromosome 14. In this study we report the analysis of other features characteristic of an inactive X to determine which features might predispose TIMP1 to expression from the inactive X chromosome in a subset of females.

The inactive X acquires many of the general features of heterochromatin (reviewed in 15), and we have now examined replication timing, nuclease sensitivity, and histone modifications for TIMP1. Late replication of the inactive X at the chromosome level is observed after Giemsa staining following bromodeoxyuridine incorporation 16. Regions such as the distal and proximal short arm that contain a large proportion of genes that escape inactivation are not delayed in their replication, supporting a regional basis to escape from inactivation. Replication of individual X-linked genes has been analysed by fluorescent in situ hybridization (FISH) or amplification of BuDR incorporated DNA after flow cytometry. Although correspondence is not complete between different techniques 17, such methods have generally shown that genes that escape inactivation are early replicating (reviewed in 18). DNase sensitivity is a general feature of active chromatin, and promoters of genes subject to inactivation have been seen to be less available for digestion by nucleases (e.g. 19). Modifications to the histones associated with the inactive X are reflective of both general heterochromatic changes and ones specific to the facultative heterochromatin of the inactive X. Using antibodies to acetylated histones the inactive X chromosome stains very palely 20, while antibodies to histone H3 methylated at lysine (K) 9 (H3mK9) are generally associated with heterochromatin and those to methylated lysine 27 (H3mK27) specifically mark the inactive X chromosome 2122. Chromatin immunoprecipitation (ChIP) has revealed that genes subject to inactivation show limited acetylation and elevated histone methylation (H3mK9/27) at their promoters relative to genes escaping inactivation 2324.

The facultative heterochromatin of the inactive X chromosome is a fascinating system to study the epigenetic modifications associated with silent chromatin as both an active and inactive version of most X-linked genes exist in female cells. However, not all genes on the 'inactive' X chromosome are subject to silencing 26, and these genes that escape inactivation must somehow maintain an active state on an otherwise silent chromosome. TIMP1 is subject to inactivation in many females, and we now show that when subject to inactivation the anticipated features of inactive chromatin are assembled – promoter DNA hypermethylation, nuclease insensitivity and histone methylation and hypoacetylation. In other females, however, we have previously demonstrated that TIMP1 continues to be expressed from the inactive X chromosome, and in this study we have exploited such chromosomes to analyze the epigenetic features that are associated with expression from the inactive X. By comparing expression of TIMP1 that occurs naturally with that induced by the demethylating agent 5-azacytidine we are also able to propose which feature may predispose to expression of TIMP1 in otherwise silent chromatin. We observed that expression of TIMP1 was associated with an active chromatin structure, despite the presence of the gene on the inactive X chromosome, except in three situations.

First, in female cell lines with or without TIMP1 expression from the inactive X there was a very similar extent of replication asynchrony, suggesting that the expression of TIMP1 was ocurring from a late-replicating region of the X chromosome. This is not surprising, as the genes flanking TIMP1, which are not variable in their inactivation, are located within 55 kb, so it is likely that at least one of ARAF1 or ELK1 shares a replication origin with TIMP1. We assessed replication timing by FISH, and detected a frequency of replication asynchrony for the HPRT locus in close agreement with the frequencies previously reported for this gene 27. It has been suggested that the 'double dot' versus 'single dot' pattern may reflect chromatid association in addition or instead of replication asynchrony 17. Regardless of the underlying cause, the results were highly reproducible and consistently showed less asynchrony for the TIMP1 locus relative to the HPRT locus, although still in the range that is generally considered asynchronous 28. This reduced asynchrony at TIMP1 could reflect, or contribute to, weaker epigenetic silencing being established in this region of the X chromosome. There was a slight, but not statistically significant, trend for the female showing expression of TIMP1 from the inactive X chromosome to show even less replication asynchrony for TIMP1, however it seems unlikely that such a small difference could contribute substantially to TIMP1 expression from the inactive X, and thus other factors must be involved in permitting the expression of TIMP1 from the inactive X.

The other two situations in which expression and active chromatin features were not concordant were found in the somatic cell hybrids, and these results are summarized in Figure 5. As shown in the grey box, DNA methylation and additional features of silent chromatin were detected along with expression in the Xi/+/M (t86, t81) cells. We attribute this to heterogeneity in the population of cells, as subcloning yielded both methylated/silent and unmethylated/expressing clones. In addition to DNA methylation, these clones were observed to be insensitive to DNase and have methylated H3K9 residues near the TIMP1 promoter. As the assays for these features relied on PCR, a positive signal could be obtained from a subpopulation of cells with a silent chromatin structure, while another population of cells could be positive for expression and active chromatin modifications (histone acetylation). We also suggest that heterogeneity accounts for expression in the presence of DNA methylation in several of the demethylated clones, including the t11-az-10-7 clone. However, unlike the mixed population of subclones obtained with all other Xi/+/M cells, eight of eight subclones of t11-az-10-7 were methylated and expressing, and six of these subclones examined by RPA showed a consistently reduced level of expression relative to their unmethylated counterparts. While it was surprising that subcloning did not isolate distinct subpopulations, the presence of an unmethylated subpopulation was detectable by bisulphite sequencing. Thus heterogeneity again seems the most likely explanation for DNA methylation, DNase insensitivity and histone H3K9 methylation in these expressing cells. However, without single cell assays it is not possible to rule out that there are methylated cells that express TIMP1 at reduced levels and show additional features of silent chromatin. These two classes of clones were the only exceptions to concordance between nuclease sensitivity and gene expression at the promoters of the TIMP1 and ARAF1 genes; and previous studies of the nucleosomal organization of the HPRT1 gene promoter have shown that methylation does not directly affect the differential positioning of nucleosomes on active and inactive X-linked promoters 19. Thus we believe that the methylated/expressing/nuclease insensitive clones reflect the presence of a subpopulation of silent cells. This heterogeneity demonstrates the unstable nature of silencing for TIMP1 in these cells.

The third exception is the most interesting, and is highlighted in Figure 5 with a dark grey fill. Acetylation of histones is generally seen for active genes, and acetylated histone H4 showed complete concordance with expression for all genes. However, the Xi/-/M clones (t86-6P and 1 U as well as t81-4) showed ongoing acetylation at H3 despite having lost TIMP1 expression from the inactive X chromosome. The Xi/-/M subclones were derived from two different unstable, naturally expressing inactive X chromosomes (in t86 and t81), however expression of TIMP1 in these hybrids was now stably silenced, and thus this result was not a reflection of a mixed population of cells. Furthermore, acetylation cannot simply reflect a failure of this region to reset chromatin structure, as the demethylated t11-az-9-10 hybrid which had lost expression had also lost acetylation. Methylation of H3 K9 was observed in these silent clones, so the Xi/-/M cells, despite being a homogeneous population, appear to show both methylation and acetylation of histone H3. This may reflect the lack of specificity of the antibodies used for these experiments, as the acetylated H3 antibody recognizes acetylation at residues 9 and 14, while the antibody to methylated K9 can cross-react with methylation of K27. Thus immunoprecipitation by both antibodies may reflect modification of specific lysines. It is also possible that only a particular set of nucleosomes show the acetylation mark, and as the sonicated fragments immunoprecipitated in the assay averaged ~600 bp, modifications on several different nucleosomes flanking the primers used could result in immunoprecipitation. Regardless of the specific site of modification, acetylation of histone H3 was the one feature that was consistently associated with X chromosomes that were naturally predisposed to expression from the inactive X chromosome, regardless of expression status. Thus we propose that histone H3 acetylation differs at the TIMP1 genes in females, predisposing some females to expression from the inactive X. Unfortunately this hypothesis is difficult to test, as females generally show acetylation at TIMP1, due to the presence of the active X chromosome. It would be necessary to study clonal populations of cells from a female with a polymorphism close enough to the promoter to be analysed by ChIP, and currently no such polymorphisms are known. Our previous work did not show an association between expression of TIMP1 and a downstream polymorphism in exon 5 12.

A different predisposing epigenetic mark for expression from the inactive X has previously been reported in mice. Methylation of H3 at K4 was restricted to the promoter in undifferentiated embryonic stem cells for genes that would subsequently be expressed monoallelically (i.e. genes subject to X inactivation or imprinting), while autosomal genes or genes that escape inactivation had H3 K4 methylation in the gene body as well as the promoter 11. Such methylation was observed in the body of the Smcx gene which is initially silenced, and then reactivates, reminiscent of the H3 acetylation mark and instability of silencing that we observe for TIMP1. While many genes escape inactivation in humans, TIMP1 provides us with a unique opportunity to examine the gene in its expressed and silent state in hybrid somatic cells. Thus, it will be interesting to study additional chromatin modifications for TIMP1, including H3 K4 methylation, as well as additional modifications recently associated with the inactive X chromosome such as H4 K20 methylation 29 or H2A ubiquitinylation 3031. Even more interesting would be to determine if the changes seen are observed for additional X-linked genes that are variable in their inactivation status, if suitable model systems could be developed.

Previous examinations of genes that escape inactivation have shown an active chromatin structure (reviewed in 7), and recent results further demonstrate that epigenetic modifications seem to be heterogeneous in their distribution along the X 32. These 'flavours' of inactive X chromatin may correspond to clusters of genes that escape inactivation. While genes that escape inactivation tend to be clustered in blocks and enriched on the short arm of the X chromosome, genes such as TIMP1 that are variable in their inactivation are more evenly distributed along the X chromosome (5, reviewed in 6), and thus may reflect different mechanisms leading to expression from the inactive X. The CTCF boundary factor has recently been shown to be found between the domain of escape and that of inactivation 33. It is not known whether such boundaries flank the variably inactivated genes. The five clones that showed reactivation of ELK1 were all TIMP1+, suggesting a co-ordinate reactivation between the two genes. Only one ARAF1 reactivant was identified, and this clone was also TIMP1+. However, this co-ordinate reactivation of TIMP1 with flanking gene(s) is different from the TIMP1-specific expression seen for the inactive X naturally. The ZNF157/41 genes did not seem to show co-ordinate reactivation with the TIMP1 region, however all three ZNF41 positive clones were ZNF157 positive, which suggests they may be in a separately controlled domain.

While association with a CpG island does not differ between genes subject to or escaping from inactivation, genes heterogenous in their inactivation are more likely to lack an island 5, consistent with the TIMP1 promoter being the least CpG dense of the region. Although TIMP1 lacks a CpG island, our analysis suggests that methylation of the few CpGs present near the promoter generally correlates well with silencing. Demethylation resulted in at least transient expression of all genes examined, confirming the importance of DNA methylation in stable maintenance of X chromosome silencing. Examination of the ELK1, TIMP1 and ARAF1 promoters after demethylation showed the presence of DNA methylation, even when the clones were expressing the genes. The majority of these clones subsequently resilenced these genes, as has been observed for other genes following demethylation 34, perhaps reflecting only partial initial demethylation or the spread of silencing from adjacent silent regions that retained DNA methylation or other epigenetic marks of silencing.

It has been proposed that the evolutionarily recent addition of the X short arm may predispose genes located there to escape from silencing 35. Interestingly, TIMP1 is very close to the evolutionary breakpoint between the region added to the X after marsupial divergence. ARAF1 is found on the marsupial X while TIMP1 and its surrounding gene SYN1 are autosomal in marsupials, suggesting that they are part of the eutherian addition to the human X 36. Since genes closely flanking TIMP1 are normally subject to X inactivation, even when TIMP1 is expressed, evolutionary history alone cannot explain the escape from inactivation, although it may result in a different genomic context that contributes to expression from the inactive X. However, inspection of genomic features in the region surrounding TIMP1 does not show a substantial difference in the frequency of repetitive elements near TIMP1 relative to the flanking ARAF1 and ELK1 genes (Figure 2A).

Several factors, including reduced replication asynchrony, lower CpG density, and more recent evolutionary addition to the X, may contribute to less stringently controlled inactivation for TIMP1. However, we propose that there is a difference between females for a feature unique to TIMP1 that predisposes some females to expression of the gene. This mark appears to be at least reflected in a difference in acetylation of histone H3 on the inactive X in females predisposed to expression of TIMP1 from the inactive X chromosome. While histone acetylation appears to be a predisposing mark for expression of TIMP1, there may be a different hierarchy of epigenetic modifications permitting expression of other genes from the inactive X. Elucidation of these mechanisms is important not only as a model for epigenetic gene regulation but because genes that escape inactivation contribute to the phenotype of X chromosome anueploidies, and may also result in differential male/female expression levels and disease susceptibilities.

H3 – histone H3; H4 – histone H4; K – lysine; ChIP – chromatin immunoprecipitation; X – X chromosome; Y – Y chromosome; Xi – inactive X; Xa – active X; M – methylated; U – unmethylated; + expressing; - not expressing; S – single signal; D – double signal.

CA carried out the molecular genetic studies and data analysis and drafted the manuscript. CB conceived of the study, participated in the experimental design and data analysis and wrote the final manuscript. Both authors read and approved the final manuscript.