Transient Homologous Chromosome Pairing Marks the Onset of X Inactivation

See allHide authors and affiliations

Science  24 Feb 2006:
Vol. 311, Issue 5764, pp. 1149-1152
DOI: 10.1126/science.1122984


Mammalian X inactivation turns off one female X chromosome to enact dosage compensation between XX and XY individuals. X inactivation is known to be regulated in cis by Xite, Tsix, and Xist, but in principle the two Xs must also be regulated in trans to ensure mutually exclusive silencing. Here, we demonstrate that interchromosomal pairing mediates this communication. Pairing occurs transiently at the onset of X inactivation and is specific to the X-inactivation center. Deleting Xite and Tsix perturbs pairing and counting/choice, whereas their autosomal insertion induces de novo X-autosome pairing. Ectopic X-autosome interactions inhibit endogenous X-X pairing and block the initiation of X-chromosome inactivation. Thus, Tsix and Xite function both in cis and in trans. We propose that Tsix and Xite regulate counting and mutually exclusive choice through X-X pairing.

The random form of X-chromosome inactivation (XCI) [reviewed in (1)] is regulated by a “counting” mechanism that enables XCI only when more than one X is present in a diploid nucleus. A “choice” mechanism then stochastically designates one Xa (active X), on which the X-inactivation center (Xic) is blocked from initiating silencing, and one Xi (inactive X), on which the Xic is induced to initiate chromosome-wide silencing. Regulatory elements have been mapped to three noncoding Xic genes, including Xist (24), its antisense partner Tsix (57), and Xite (8). Whereas Xite and Tsix together regulate counting and choice (6, 7, 911), Xist predominantly regulates chromosome-wide silencing (4, 1214). Interestingly, each gene acts in cis, with Xite activating the linked Tsix allele, Tsix repressing the linked Xist allele, and Xist repressing other genes on the same X.

Although cis-acting genes dominate the Xic, Xic function must extend in trans. Notably, the choice of Xa and Xi always occurs in a mutually exclusive manner, so when one X is designated Xa, the other is accordingly designated Xi. The idea of crosstalking is supported by a Tsix–/– knockout, in which choice becomes “chaotic” with the occurrence of 2 Xi, 1 Xi, or 0 Xi per cell (9, 11). Though trans-interaction seems necessary (9, 15), direct evidence has been lacking. In principle, trans-sensing could be accomplished by feedback signaling cascades, diffusible X-linked factors, or direct interchromosomal pairing such as that proposed for T cell differentiation (16).

Because somatic homolog pairing does not generally occur in mammals, we surmised that pairing—should it occur on the X—must take place transiently. Here, we followed the movement of the chromosomes over time using fluorescence in situ hybridization (FISH) in differentiating mouse embryonic stem (ES) cells, a model that recapitulates XCI in culture. We measured the X-X interchromosomal distances for day 0 (pre-XCI), day 2 and day 4 (XCI onset), day 6 ES cells, and mouse embryonic fibroblasts (MEFs) (Fig. 1). By combining two non-overlapping probes, we obtained 99% X detection rates (single probes gave 85 to 90% rates). Only nuclei with two resolvable signals were scored. For each experiment, 150 to 250 nuclei were scored, and similar results were obtained in three independent tests.

Fig. 1.

Evidence for X-X homologous associations. (A) DNA FISH and X-X distribution profiles of wild-type female ES nuclei from day 0 to day 6 of differentiation and of MEFs. Two-probe combination: Xic DNA-green (pSxn-FITC) + Tsx DNA-red (pTsx-Cy3). DAPI (4′,6′-diamidino-2-phenylindole), blue. Each image is a two-dimensional (2D) representation of 3D image stacks of 0.2 μ z-sections. The distributions display the normalized distances, ND = X-X distance/d, where d = 2 × (nuclear area/π)0.5. ND ranges from 0 to 1. Mean distance, open triangle. (B) Cumulative frequency curves for X-X pairs at 0.0 to 0.2 ND. P (KS test) was calculated in pairwise comparison against day 0. Sample sizes for each experiment (n) = 174 to 231. (C) X-X distances <0.05 ND were graphed with standard deviations (SD) from three independent experiments. (D) Proximity pairing is specific to the Xic. X-X distribution profiles for X-linked loci shown in the map. The KS test (P) compared Xic versus flanking loci. n = 166 to 188.

In wild-type XX cells, the X-X distance was highly dynamic during cell differentiation (Fig. 1A). On day 0, the interchromosomal distances approximated a normal distribution, suggesting near-randomness. Interestingly, on day 2, a high proportion of cells began to display close X-X distances, as shown by a left shift in the distribution (Fig. 1A) [Kolmogorov-Smirnov (KS) test, P = 0.01]. This trend continued into day 4 (P < 0.001) and partially returned to baseline on day 6 (P = 0.41). The MEF distribution was completely random, somewhat more so than for day 0 ES cells, perhaps reflecting spontaneous differentiation of some ES cells. Cumulative frequency curves (Fig. 1B) showed that day 2 and day 4 displayed the highest frequency of “promixity pairs,” or pairs with normalized X-X distances (ND) <0.2 (<2.0 μ). Among proximity pairs, one-third displayed 0.2- to 0.5-μ separation (Fig. 1C), a fraction greater by factors of 6 and 16 than in day 0 ES and MEFs, respectively. X painting confirmed the presence of two Xs (fig. S1), thus excluding the possibility of visualizing sister chromatids within XO cells.

Measurement of interautosomal (A-A) distances at 1C [chromosome 1 (Chr1) centromere], Abca2 (Chr2), and chromosome 3 centromere showed normal distributions at all time points (Fig. 1B and fig. S2), demonstrating that proximity pairing was not generally observed. To determine the extent of pairing on the X, we tested four bacterial artificial chromosome (BAC) probes in combination with an Xic probe (Fig. 1D and fig. S3) and found that, whereas Xic movement was constrained by homologous interaction, the flanking regions adopted relatively free positions, with each locus showing near-random distributions across time (fig. S3). Thus, X-X interactions were restricted to the Xic.

The pairing kinetics suggested linkage to XCI, which coincidentally initiates between day 2 and day 4 of differentiation. Because Xist RNA up-regulation is the earliest known cytologic feature of XCI (1), we asked whether pairing could be observed more frequently in Xist+ cells. Indeed, Xist+ cells showed 46% with X-X association (Fig. 2, A and B), indicating that pairing occurs just before or during Xist up-regulation. To pinpoint the time frame, we employed the additional temporal markers, Ezh2 and H3-3meK27, which accumulate on the Xi shortly after Xist up-regulation during the “early Xi maintenance” phase [reviewed in (17)]. On day 2, trans-associations were significantly enriched in Ezh2 cells and in H3-3meK27 cells relative to Ezh2+ and H3-3meK27+ cells (Fig. 2, C and D, and fig. S4). These results restricted trans-interactions to Xist-expressing cells that have not yet recruited Ezh2 and H3-3meK27, thus demonstrating a very early time frame, well before the XCI maintenance phase.

Fig. 2.

Homologous association occurs during the initiation phase of XCI. (A) RNA-DNA FISH for day 2 wild-type XX cells. Xic DNA, green (pSxn-FITC); Xist RNA, red (strand-specific riboprobes-Cy3). (B to D) Cumulative distributions for day 2 wild-type XX cells, comparing Xist+ (n = 74) versus Xist (n = 180) cells (B); Ezh2+ (n = 33) versus Ezh2 cells (n = 178) (C); and H3-3meK27+ (n = 48) versus H3-3meK27 (n = 188) cells (D).

We therefore tested the relation of transinteractions to counting and choice, the two earliest steps of XCI, both of which are regulated by Tsix and Xite. It was previously shown that Tsix+/– mice (XΔTsixX) are disrupted for choice and silence only XΔ (6, 7, 10, 18), whereas XΔTsixXΔTsix mice are disrupted for both counting and choice (9, 11). Xite mutations have similarly affected counting/choice (8, 11). Here, we observed that XΔXiteX cells showed a marked delay in X-X association (Fig. 3, A and B, and fig. S5), implying that losing one Xite allele is sufficient to partially disrupt pairing. This partial effect correlated with aberrant choice in XΔXiteX. However, XΔTsixX cells showed the expected frequency of homologous association, indicating that losing one Tsix allele does not affect pairing. In contrast, XΔTsixXΔTsix cells showed near-random distributions across all time points (Fig. 3B and fig. S5), which supports the argument that deleting both Tsix alleles is required to abolish pairing. Although not statistically significant, day 6 populations showed a slight left shift suggestive of a delayed or weakened attempt to associate. These data demonstrated that Tsix and Xite are required for pairing and implied a tight link between pairing and counting/choice.

Fig. 3.

Tsix and Xite are necessary and sufficient for X-X pairing. (A) Map of the Xic, TsixDCpG and XiteDL, and various transgenes. (B) X-X distributions for Tsix and Xite mutants from day 0 to day 6. n = 181 to 223. KS test compares each curve to the day 0 curve. (C) Tsix alleles and primers (red) used for 3C analysis. BamHI sites, blue arrow. (D) 3C analysis of pairwise interactions in XΔTsix(neo+)X cells and p3.7 females. Primers pairs are indicated to the right of gels. C, positive control ligations. All minus-crosslinking (N) and minus-ligation controls were negative. (E) Relative pairing frequencies (X) on day n (dn) was normalized to β-globin (βg) and to day 0 values, using the equation shown. S, signal intensity quantitated by densitometry. Average and SD from three independent experiments. (F) DNA FISH and X-A distribution curves for transgenic ES cells. The transgene was labeled red by a Neo probe and the X labeled green by a pSx7 probe (for p3.7, pXite, pXist5′, and pTsx cells) or a pTsx probe (for pSx7 cells). The pSx7 partially overlaps the p3.7 and pXite transgenes, but the small overlap makes the signal dim and discernible from the X. For πJL1.4.1, the transgene was labeled green (pSx9 Xist fragment) and the X labeled red (pTsx probe). The KS test compared data sets from day 0 versus day 4. n = 170 to 234.

To learn whether the homologous association represented true physical pairing, we carried out “chromosome conformation capture” (3C) (19), whereby two interacting loci can be detected by crosslinking, intermolecular ligation, and polymerase chain reaction. To obtain necessary polymorphisms for 3C, we used the pairing-competent XΔTsix(neo+)X line, in which one Xic is distinguished by Neo (Fig. 3C) (wild-type could not be used because they lack informative polymorphisms within required restriction fragments). Using three distinct primer pairs [Tsix1-N3 (shown), and TSEN2-N1 and Tsix1-N2 (not shown)], we consistently detected physical contact between the two Tsix loci, whereas no contacts were observed between various Tsix and autosomal controls or the incorrectly oriented Tsix2 primer and N3 (Fig. 3D and fig. S6). The inter-Tsix interaction was strongest on day 4 (Fig. 3E), consistent with FISH analysis. Therefore, inter-Xic pairing indeed underlies homologous association.

To identify sequences that direct pairing, we introduced Xic fragments into ES cells (Fig. 3A and fig. S7) (11) and asked whether autosomal insertions could induce de novo X-autosome (X-A) pairing and affect counting/choice. Intriguingly, autosomal pSx7 led to ectopic X-A pairing in females (Fig. 3F), correlating with aberrant counting and XCI initiation in pSx7 females (11). By contrast, female Xist and Tsx transgenics showed no X-A pairing above background (Fig. 3F), consistent with their normal XCI (11). Furthermore, male pSx7 transgenics did not exhibit X-A pairing (Fig. 3F), consistent with their normal counting and XCI suppression (11).

To dissect specific requirements within pSx7, we tested p3.7, the 3.7 kb Tsix fragment deleted in the pairing-incompetent XΔTsixXΔTsix. p3.7 was remarkably efficient at inducing de novo X-A pairing in XX cells (Fig. 3F), with 3C analysis confirming direct physical interaction between p3.7 and the X (Fig. 3D). The ectopic pairing paralleled the failure of counting/choice and XCI initiation in p3.7 females (11). In contrast, p3.7 males did not induce X-A pairing and accordingly did not manifest a counting defect (11). We also tested pXite (a 5.6-kb fragment deleted in the pairing-compromised XΔXiteX) and found efficient X-A pairing (Fig. 3F), consistent with pXite's profound effect on counting/choice (11). Interestingly, pXite males could also initiate pairing, although they did not exhibit ectopic XCI (11). Because pXite males are thought to lack an X-linked “competence factor” for initiating XCI, we next tested males carrying full-length Xic transgenes (11) to determine whether pairing and XCI could be achieved together. Indeed, πJL1.4.1 males displayed ectopic X-A pairing (Fig. 3F) and, accordingly, initiated counting/choice and silencing (20), further supporting the tight linkage between pairing and XCI initiation. These experiments demonstrated that Tsix and Xite, with sequences as small as 3.7 and 5.6 kb, are sufficient to recapitulate pairing and that, in turn, pairing is required for the earliest steps of XCI.

In transgenic females, we hypothesize that the failure to initiate XCI may be due to a competitive inhibition of X-X interactions by de novo X-A interactions. Indeed, the frequency of X-X interactions was significantly diminished for pSx7, p3.7, and pXite females as compared with wild-type (Fig. 4A versus Fig. 1B). In pSx7 females, X-X pairing rates were less than X-A pairing rates. In p3.7 and pXite females, X-X pairing appeared to be abolished completely (Fig. 4A and fig. S8), with day 2 and day 4 distribution profiles being indistinguishable from day 0 (Fig. 4B) and <2% of nuclei (background) with ND < 0.05 (Fig. 4C). In contrast, X-X pairing remained robust in pTsx and pXist controls (fig. S8). Therefore, ectopic X-A interactions measurably detracted from endogenous X-X interactions. The frequency of X-X pairing directly predicts the frequency of XCI. We propose that the titration of X-X interactions by ectopic Tsix/Xite accounts for the pervasive failure of counting/choice and XCI in transgenic females.

Fig. 4.

De novo X-A pairing inhibits X-X pairing. (A) Disruption of X-X pairing in female transgenic cells. n = 177 to 221. (B) KS test compares data sets from day 0 versus day 2 and from day 0 versus day 4. (C) Average frequency of X-X pairing with standard deviations from three experiments. (D) Model: X-X pairing is required for counting/choice. Allelic crosstalking results in asymmetric chromosome marking (yellow circles, blocked Xic; red circle, induced Xic) and mutually exclusive designation of Xa and Xi. Blue lines, Xist RNA. Ectopic Tsix/Xite transgenes (Tg-Xic) inhibit XCI by titrating away X-X interactions. Loss of pairing in Tsix XΔXΔ causes aberrant counting/choice.

On the basis of this work, we postulate that X-X pairing acts upstream of Xist by mediating counting/choice and providing the necessary crosstalk for mutually exclusive XCI. Pairing interactions clearly do not require Xist expression. In our model (Fig. 4D), two Xs assume random independent positions in pre-XCI cells and then pair homologously at the onset of XCI, with Tsix and Xite acting as nucleation centers. The ensuing crosstalking achieves asymmetric marking of one X to become Xa and the other to become Xi. With counting/choice reflecting the binding of a “blocking factor” to the Xa and the competence factor to the Xi (6, 11), pairing ensures that the two factors bind mutually exclusively.

Remarkably, 3.7 kb of Tsix or 5.6 kb of Xite is sufficient to initiate de novo pairing. Thus, these genes play dual cis-trans roles in XCI by functioning in trans to coordinate pairing/counting/choice and in cis to antagonize Xist. These events may take place simultaneously in time and space. Subtle pairing differences between Tsix and Xite mutants likely reflect length requirements, as indeed XΔXiteX shows weaker pairing than XΔTsixX, and Xite transgenic males pair better than Tsix counterparts. Consistent with this, full-length transgenic πJL1.4.1 males not only pair well but also initiate XCI. Why do X-A interactions generally outnumber X-X interactions? The multicopy transgene nature might increase the avidity of the autosome relative to the X. The ability of X-A pairing to inhibit X-X pairing now provides a mechanism for failed XCI in Tsix/Xite transgenic females: If pairing were required for proper counting/choice, the failure to pair would pose a specific block to XCI. The proposed regulation by interchromosomal pairing creates a new dimension to the problem of gene regulation and is likely to become a recurrent theme in epigenetic phenomena (16, 21).

Supporting Online Material

Materials and Methods

Figs. S1 to S8


References and Notes

View Abstract

Navigate This Article