Our results identify H3K27m3 and H4K20m1 as specific modifications that mark the Xist
-expressing chromosome in undifferentiated ES cells and contribute to the epigenetic histone code of the Xi (). We did not observe an enrichment of H3K9m2 or H3K9m3 signals on the Xist
-expressing chromosome, which has been reported by other studies. This could be a shortcoming of our transgenic system, but we also did not detect the H3K9m2 or H3K9m3 signals in female mouse primary embryonic fibroblasts (less than 2% of the cells). We attribute the different observations in other studies to the various antisera used. We supply peptide blot analysis for our antisera that suggest that the antibodies are highly specific (see Figure S1
). This is also supported by the specific staining patterns in immunofluorescence experiments. The lysine 9 methylation signal observed in other studies could potentially be a result of cross reactivity with H3K27m3, a fact we can exclude for our H3K9 antibodies based on the staining pattern and peptide blots. Alternatively, our antibody might not recognise the H3K9m2 modification in the context of the chromosome. However, this is unlikely since the H3K9m3 signals for the pericentric regions and the Y chromosome are clearly identified. The H3K9m2 antiserum has been successfully used in ChIP analysis of the minor centromeric repeats (Yan et al. 2003
) and reacts with these repeats in immunofluorescence, but does not show cross reactivity to H3K27m3. This suggests that our reagent is able to detect the modification in both ChIP and immunofluo-rescence experiments. Using highly specific antisera, we failed to see a strong signal for H3K9m2 in either ChIP or immunofluorescence experiments (see Figures and S2E
). In our ChIP analysis two chromosomal loci showed an increase for H3K9m2 upon Xist
expression in differentiated ES cells, suggesting some enrichment for H3K9m2. We take these data to indicate that H3K9m2 is not a prominent mark of X inactivation but might be enriched locally to some degree upon differentiation.
Using Xist alleles that express a mutated version of Xist, which has a deletion of repeat A sequences and is unable to cause silencing, we showed that both H3K27m3 and H4K20m1 were established in the absence of transcriptional repression. This demonstrates that neither modification is sufficient to trigger silencing.
expression led to rapid H3K27m3, which was complete after 1 to 2 d of Xist
expression in both ES cells and differentiated cells (see Figure S3A
and , columns 1 and 2). This kinetics follows the localisation of Xist
RNA, which accumulates between 4 and 12 h after doxycycline addition in ES cells (Wutz and Jaenisch 2000
), suggesting that H3K27m3 is an immediate effect. We have further shown that in undifferentiated ES cells no progressive accumulation of the histone modifications occurs over time by comparing the percentage of cells showing H3K27m3, H4K20m1, and Ezh2 staining after 3 and 10 d expressing either full-length Xist
RNA or a silencing-deficient mutant lacking repeat A (see Figure S3C
). We have shown that H3K27m3 is a reversible modification throughout ES cell differentiation and depends at all stages on Xist
expression. In undifferentiated ES cells H3K27m3 disappeared 48 h after Xist
expression was turned off, corresponding to about two cell divisions. The kinetics would be consistent with the idea that replication is involved in the replacement of methylated histones, albeit our data do not rule out an active enzymatic process of demethylation. Importantly, we have observed nearly unchanged methylation levels 24 h after Xist
expression has been turned off (see Figure S3B
). This could reflect the intrinsic stability of the trimethylation mark or the persistence of the Eed/Ezh2 complex, which can stably associate with metaphase chromosomes from which Xist
RNA is displaced (see C; Mak et al. 2002
). The transient maintenance of H3K27m3 might be significant for the mechanism of X inactivation. It could explain our observation that the inactive state will be “locked in” roughly 24 h after Xist
loses its ability to initiate silencing, it will be locked in at 72 h of ES cell differentiation (Wutz and Jaenisch 2000
Efficient methylation is established only when Xist
expression is induced early in ES cell differentiation. The window in which Xist
causes efficient methylation overlaps precisely with the initiation window, in which transcriptional silencing can be initiated. Yet methylation is independent of initiation of silencing. This would be consistent with the notion that H3K27m3 is necessary but not sufficient for silencing. However, this is unlikely, as a previous report has shown that in Eed mutant embryos, initiation of silencing is normal, but a defect in the maintenance of the inactive state leads to reactivation at later stages (Wang et al. 2001
). Lower levels of Ezh2 and Eed could explain the restriction on the ability of Xist
to induce H3K27m3 efficiently in differentiated ES cells (Silva et al. 2003
). We do not favour this interpretation, as this restriction is observed at day 2 in differentiation, when Ezh2 and Eed protein levels are still high (see C). Our data further show that the ability to efficiently methylate a chromosome late in ES cell differentiation is a feature of the chromosome and not a function of the protein levels of Eed and Ezh2. This is also in line with our observation that chromosome-wide H3K27m3 in clone 36 ES cells, in which Eed messenger RNA was reduced to 10%–15% of wild-type levels by stable RNAi, was still detected in 45%–60% of cells compared to 80% in control clone 36 cells (data not shown). Therefore, less abundant levels of Eed are sufficient to achieve efficient methylation. Xist
induction later in ES cell differentiation or in cells of embryonic origin establishes H3K27m3 in only a small percentage of cells. The significance of H3K27m3 in this small number of cells is unclear at present.
The restriction of efficient methylation to early ES cell differentiation and the finding that methylation is reversible logically require that a chromosomal memory exists that enables H3K27m3 maintenance during differentiation. Previous models have suggested that a lock-in of X inactivation is based on chromosomal silencing, arguing that self-maintaining heterochromatin structures establish the principal form of memory. Our data clearly demonstrate that H3K27m3 is maintained in the absence of transcriptional repression, suggesting a chromosomal memory independent of silencing on the Xi. Using the inducible Xist
expression system we have directly demonstrated the chromosomal memory (see ). A chromosome that had been exposed to Xist
and been H3-K27 trimethylated early could be remethylated later in differentiation, after a period where Xist
was turned off and methylation decayed, with significantly greater efficiency than a chromosome that had not expressed Xist
early (see ). We have further determined the time point in ES cell differentiation when the chromosomal memory is established and found that it overlaps with the transition from Xist
-dependent and reversible silencing to irreversible silencing. These data place the establishment of the memory in a critical phase of X inactivation. We note that the establishment of efficient H3K27m3 in the initiation window and the implementation of the memory are separated by a gap of approximately one cell division in ES cell differentiation. This parallels the gap between initiation of silencing and the maintenance of the silenced state independent of Xist.
Our kinetic measurements indicate that H3K27m3 would decay from the Xist
-expressing chromosome after two cell divisions; therefore, H3K27m3 could bridge the gap (critical window). We suggest that Xist
expression and H3K27m3 might be the signal to recruit a chromosomal memory mediating the lock-in of X inactivation (). In this model, silencing would be specified by separate signals depending on repeat A of Xist,
which we predict would interact with the memory at the transition from reversible to irreversible and Xist
-independent repression. In this regard we note that silencing or repeat A sequences enhance the efficiency of H3K27m3 in undifferentiated ES cells (see B). However, there is no requirement for repeat A when ES cells are induced to differentiate (see and S3C
). This could point to interactions between the silencing machinery and the Ezh2/Eed methylation complex specifically in ES cells.
Model for the Transition from Initiation to Maintenance of X Inactivation
The molecular basis for the chromosomal memory is presently unknown. Our data rule out the possibility that continuous Xist
RNA expression or silencing is required for maintenance of the chromosomal memory and suggest that H3K27m3 is also not involved. The latter interpretation has to be treated cautiously, as it depends on the sensitivity of our assay to detect H3K27m3. Formally it is conceivable that low levels of H3K27m3 undetected by our assay could remain on the chromosome. Presently, it is also unclear what the role of H4K20m1 is and to what extent it interacts with H3K27m3. A H4-K20–specific histone methyltransferase has been identified (Fang et al. 2002
; Nishioka et al. 2002
; Rice et al. 2002
), and we have performed in vitro functional analysis of the mouse Pr-Set7 protein (Figure S5
; Protocol S1
). Our results indicate that Pr-Set7 is a monomethylase for H4-K20. Its involvement in X inactivation and the function of H4K20m1 remain unclear at present. Future work is needed to identify the components of the memory configuration and to determine its precise function in X inactivation.