Intersection of the RNA Interference and X-Inactivation Pathways

See allHide authors and affiliations

Science  06 Jun 2008:
Vol. 320, Issue 5881, pp. 1336-1341
DOI: 10.1126/science.1157676


In mammals, dosage compensation is achieved by X-chromosome inactivation (XCI) in the female. The noncoding Xist gene initiates silencing of the X chromosome, whereas its antisense partner Tsix blocks silencing. The complementarity of Xist and Tsix RNAs has long suggested a role for RNA interference (RNAi). Here, we report that murine Xist and Tsix form duplexes in vivo. During XCI, the duplexes are processed to small RNAs (sRNAs), most likely on the active X (Xa) in a Dicer-dependent manner. Deleting Dicer compromises sRNA production and derepresses Xist. Furthermore, without Dicer, Xist RNA cannot accumulate and histone 3 lysine 27 trimethylation is blocked on the inactive X (Xi). The defects are partially rescued by truncating Tsix. Thus, XCI and RNAi intersect, down-regulating Xist on Xa and spreading silencing on Xi.

X-chromosome inactivation (XCI) (1) balances X-chromosome dosages between XX and XY individuals. XCI is initiated by Xist (2, 3) and opposed by Tsix (4). How Xist induces XCI on inactive X (Xi) and how Tsix stably silences Xist on active X (Xa) remain two unanswered questions. A role for RNA interference (RNAi) has long been speculated. RNAi refers to the repressive influence of double-stranded RNA (dsRNA) on gene transcription and transcript stability (5, 6). Numerous similarities, including the involvement of noncoding RNAs, can be found between XCI and RNAi silencing of constitutive heterochromatin. However, a deficiency of Dicer (Dcr) has no obvious effect on maintaining Xi in T cells (7) and, although Xist and Tsix RNAs are perfectly complementary, dsRNAs had never been observed in vivo.

Here, we formally explore a role for RNAi in XCI. To search for small RNAs (sRNAs) within Xist/Tsix, we performed Northern analysis in mouse embryonic stem (ES) cells, a model that recapitulates XCI during cell differentiation ex vivo; and in mouse embryonic fibroblasts (MEFs), post-XCI cells that faithfully maintain one Xi. At repeat A, a region of Xist required for silencing (8), we observed sRNAs at ∼30 nucleotides (nt) and ∼37 nt in the Tsix orientation and at ∼25 and ∼35 nt in the Xist orientation (Fig. 1A). At Xist exon 7, sRNAs occurred between 24 and 42 nt on the Tsix strand and at ∼25 and ∼35 nt on the Xist strand (Fig. 1B). At the promoter, robust quantities of Tsix-strand sRNAs were observed (Fig. 1C). sRNAs were also seen on the Xist strand, implying that low-level sense transcription must occur at the promoter. The integrity of all Northern blots was confirmed by micro RNA 292-as (miRNA292-as) and tRNA controls (Fig. 1 and fig. S1). The sRNAs were developmentally regulated, being unmeasurable in the pre-XCI [day 0 (d0)] and post-XCI (MEF) states and detectable only during XCI (d4 and d10). Furthermore, sRNAs occurred in both XX and XY cells. For discussion purposes, we call them xiRNA for their X-inactivation center origin, distinct from the smaller small interfering RNA (siRNAs) and miRNAs.

Fig. 1.

sRNAs derived from Tsix/Xist. (A) xiRNAs from Xist repeat A (XA) region (map) detected by Northern analysis. Sense (s) and antisense (as) riboprobes detected Tsix and Xist, respectively. miR292-as controls are shown on the same blots. M, male; F, female. (B) Northern analysis of xiRNAs from Xist exon 7 (XEx7). (C) Northern analysis of Xist promoter (XP) region. (D) Northern analysis of mutant cells. WT lanes are identical to those in (A) (concurrent analysis).

To determine whether xiRNA production depends on antisense expression, we investigated ES cells in which Tsix was deleted (TsixΔCpG) (4) and the Tsix regulator Xite (XiteΔL) (9). Deleting Tsix resulted in a dramatic reduction in antisense-strand xiRNA (Fig. 1D). A residual level of xiRNA was still detectable, consistent with cryptic promoter activity in TsixΔCpG (4). Deleting Xite likewise reduced antisense xiRNA levels, consistent with a requirement for Xite in transactivating Tsix (9). In the sense orientation, both deletions also compromised xiRNA production. Thus, sRNAs are indeed generated from Xist/Tsix and depend on Tsix and Xite expression.

The presence of xiRNAs implied that Tsix and Xist must exist as long duplex precursors. However, the developmental timing of xiRNA appearance is paradoxical: Although Tsix and Xist are biallelically expressed on d0, they become monoallelically expressed on opposite Xs during XCI (4). On d0, three to five copies per chromosome of Xist RNA are present, whereas Tsix occurs at >10-fold molar excess (1012). Upon XCI, Tsix is down-regulated on Xi as Xist up-regulates >30-fold. On Xa, Tsix persists as Xist is down-regulated. How would dsRNA form when Tsix and Xist, both cis-limited, are on opposite chromosomes during XCI?

To determine whether Tsix and Xist formed dsRNA, we devised an in vivo ribonuclease (RNase) protection assay based on differential susceptibility of single-stranded RNA (ssRNA) and dsRNAs to RNase A/T1. We permeabilized replicate preparations of d0 ES cells in a non-denaturing buffer containing deoxyribonuclease I (DNase I) and RNase, and performed strand-specific reverse transcription polymerase chain reaction (RT-PCR) on the protected RNAs. To confirm assay sensitivity, a positive control into which one copy per cell of in vitro–transcribed and annealed Tsix:Xist dsRNA was spiked could readily be detected with this protocol (Fig. 2, A and B). The abundant single-stranded ribonucleotide reductase M2 (Rrm2) and glyceraldehyde-3-phosphate dehydrogenase (Gapdh) transcripts were not amplified, indicating that our assay was specific for dsRNA. We consistently observed RNase-protected Xist and Tsix RNA strands in XX and XY ES cells, suggesting the presence of dsRNA (Fig. 2B). Real-time RT-PCR quantitation showed that ∼16% of Xist and ∼13% of Tsix strands were protected (Fig. 2C). As expected of duplexes, approximately equal stoichiometric ratios of the two strands were present in the RNase-protected fraction. Kinetic analysis revealed decreasing amounts of dsRNA during differentiation in XX and XY cells (Fig. 2D). Thus, steady-state quantities of both dsRNA and xiRNA are developmentally regulated, but in an opposite manner.

Fig. 2.

Tsix and Xist RNA form long duplexes in vivo. (A) Map of Tsix/Xist and primer pairs (asterisks). (B) In vivo RNAse protection assay. (C) Relative quantities of Xist and Tsix in duplexes measured at position 2 (base pairs 1206 to 1337 of Xist) by strand-specific real-time RT-PCR of protected RNA (RNase+) as compared with total levels (RNAse). Quantities are standardized to Xist in the Xist-Tsix duplex (for Xist, RNase+ = 1). Error bars indicate 1 SD in triplicate reactions. (D) Quantities of protected Tsix or Xist RNAs (RNase+) relative to total Tsix or Xist (RNase) for in vivo RNase protection assays. Error bars indicate 1 SD in triplicate reactions. (E) In vivo RNase protection assays to test allelic origin of dsRNA using strand-specific, allele-specific real-time RT-PCR with SNP-based primers for Xcas or Xmus alleles (top). PCR of control genomic DNA shows high specificity (98% for mus and >99.99% for cas). Error bars indicate 1 SD in triplicate reactions. For test samples, the mus and cas fractions were amplified separately, normalized to genomic DNA (Xcas:Xmus = 1:1), and plotted as a function of time. Percent of Cas = [Xcas RNA/(Xcas RNA + Xmus RNA)] × 100. Because the bars show relative allelic fractions, quantities are only comparable within a single time point.

The inverse correlation over time raised the possibility that dsRNA is processed to xiRNA. To address potential allelic differences, we performed allele-specific RNase protection assays using two genetically marked female ES cell lines: wild-type (WT) 16.7, which carries Xs from Mus castaneus (Xcas) and M. musculus (Xmus) and undergoes random XCI [with a natural 30:70 specific-specific bias (13)], and TsixΔCpG/+ mutants, which harbor a Tsix deletion on Xmus (4) and therefore always inactivate Xmus in the 16.7 background. Total Tsix RNA (with no RNase treatment) decreased >10-fold over time, but a low residual level could still be detected at d4 and d10 as expected (4, 9, 14). From this residual pool, using single-nucleotide polymorphism (SNP)–based allele-specific primers at position 3, we unexpectedly found that duplexed Tsix (RNase-protected) in TsixΔCpG/+ cells predominantly originated from Xi (Xmus) (Fig. 2E), the X on which the major Tsix promoter is deleted. Likewise, the Xist strand found in duplex form originated from Xi. Thus, Tsix:Xist duplexes are detected primarily from Xi.

Duplexes may form only on Xi, or they may form on both Xs but be stable only on Xi. The latter possibility is notable, considering the inverse kinetic relationship between the appearance of long dsRNA versus xiRNA. Could dsRNA be processed to xiRNA on Xa? Several observations favored this idea. First, dsRNA was selectively lost from Xa. Second, xiRNA production depended on Tsix, a gene expressed from Xa from d4 to d10. Finally, despite lacking Xi, XY cells produced xiRNAs.

Because dsRNAs are substrates for Dcr, we tested Dcr's role by deleting Dcr's RNAseIII domain in female ES cells (15) (fig. S2). Because Dcr-deficient (Dcr–/–) cells are known to grow poorly (15), we introduced a Dcr transgene expressed at ≪5% of WT levels (Fig. 3, A and B) and improved the growth of Dcr–/– clones (henceforth referred to as Dcr–/–). Northern analysis revealed diminished xiRNA levels (Fig. 3, C and D), suggesting that xiRNA production depends on Dcr. All tested Dcr-deficient clones behaved similarly. Xist expression prematurely increased 5- to 10-fold in pre-XCI cells (Fig. 3B), implying increased Xist transcription or greater RNA stability. Male Dcr–/– clones likewise showed significant Xist derepression on d4 (fig. S3). Thus, Dcr regulates xiRNA levels and antagonizes Xist upregulation in ES cells.

Fig. 3.

Dcr deficiency impairs xiRNA production and XCI. (A) Quantitative real-time RT-PCR of indicated transcripts normalized to β-actin. (B) Xist quantitation plotted on a log scale. (C and D) Northern analyses of miRNA292-as control (C) and xiRNAs (D) in mutant cells. There is an accumulation of miRNA292-as precursors (asterisk) in Dcr–/– cells. (E) Immuno-RNA FISH for Xist and H3-3meK27 on d10. 4′,6′-diamidino-2-phenylindole (DAPI), blue. (F) Phase contrast images of d10 embryoid bodies (EB).

RNA immunofluorescence in situ hybridization (immunoFISH) analysis showed that Dcr has additional XCI effects. Despite elevated Xist levels, Xist could not “coat” the X nor induce heterochromatic changes (Fig. 3E). On d10, Xist RNA accumulation occurred in only 0.4% of cells (n = 278 cells) and histone 3 (H3) lysine 27 (K27) trimethylation (H3-3meK27) in 0.7% of cells (n = 278 cells). By contrast, in Dcr 2lox/ [heterozygous Dcr knockout carrying one conditional allele (2loxP sites) and one deleted allele] controls, Xist accumulated in 56.8% and H3-3meK27 in 83.1% of cells (n = 148 cells). Moreover, the X-linked Mecp2 gene failed to be dosage-compensated in Dcr–/– cells, whereas it appropriately decreased 1.5- to twofold in controls (Fig. 3A). Therefore, in addition to local effects on Xist, Dcr also affected Xi globally, because the formation of Xist and H3-3meK27 domains was compromised without Dcr.

Because XCI and cell differentiation are linked (16, 17), the XCI defects might be explained by Dcr's pleiotropic effects on differentiation (15, 18) rather than specific effects on XCI. Indeed, Dcr–/– clones differentiated poorly and continued to express Oct4 and Nanog pluripotency factors on d10 (Fig. 3, A and F, fig. S3). To determine whether Dcr specifically affects XCI, we truncated Tsix by inserting a polyadenylate cassette in Dcr–/– cells (fig. S4), reasoning that disabling Tsix, which negatively regulates Xist, might overcome the failure to accumulate Xist RNA. As expected, Dcr–/–Tsix–/+ double mutants (Dcr-TST) and Tsix–/+ controls (TST) showed truncated Tsix expression from Xmus and highly skewed XCI patterns (Fig. 4A). Although Dcr-TST cells continued to differentiate poorly (fig. S5), total Xist levels were restored to nearly WT levels during differentiation (Fig. 4B). Furthermore, disabling Tsix partially restored Xist localization to Xi (Fig. 4C). Therefore, Dcr's effect on XCI can be genetically separated from its effect on cell differentiation.

Fig. 4.

Tsix genetically interacts with Dcr. (A) Allele-specific RT-PCR analysis. All RT reactions were negative (not shown in figure). Xist d0 samples (asterisks) were 10-fold overloaded to visualize low expression. trpA, triple poly-A cassette; Puro, puromycin; IRES, internal ribosome site; SA, splice acceptor. (B) Real-time RT-PCR of indicated transcripts, each normalized to β-actin. (C) Immuno-RNA FISH for Xist and H3-3meK27 domains (arrowheads) on d10. n, sample size. (D) Frequency of aberrant H3-3meK27 enrichment in the Xist+ subpopulation of indicated cell lines. n = 100 to 150 cells. (E) Model of the intersection of RNAi and XCI. (F) Methylation-sensitive restriction analysis of the Xist promoter. Genomic DNA was digested with EcoRV or Eco RV+Hpa II. The percent of uncut (methylated) DNA at Hpa II is plotted.

Additionally, to the extent that Xist levels and localization were restored in Dcr-TST cells, H3-3meK27 was only partly rescued in Xist+ cells (Fig. 4D). In WT and TST controls, Xist accumulation was almost always accompanied by robust H3-3meK27. In contrast, 30 to 40% of Xist+ Dcr-TST cells displayed weak or no H3-3meK27 foci, implying that H3-3meK27 also depends on Dcr. These data showed that Xist accumulation and H3-K27 methylation are genetically separable. We conclude that Dcr intersects XCI in several ways. Locally, Dcr controls xiRNA and Xist expression. Globally, it regulates Xist accumulation and H3-3meK27 on Xi.

In aggregate, our data suggest specific effects of RNAi on XCI (Fig. 4E). Dcr and Tsix/Xist genetically interact, and a second-site mutation in Tsix partially suppresses the Dcr–/– effect on Xist. We propose that Tsix:Xist duplexes initially form on both Xs. During XCI, continued expression of Tsix on Xa would lead to dsRNA processing to xiRNAs, which would locally repress Xist in cis, an idea reminiscent of transcriptional gene silencing (TGS) (6, 1921). Consistent with allele-specific TGS at Xist, RNA-directed DNA methylation by Tsix has been proposed (10). Here we found that abrogating Dcr and/or Tsix resulted in decreased methylation at the 5′ end of Xist (Fig. 4F). By our model, extremely low Tsix and Xist expression might be sufficient to maintain TGS on Xa in post-XCI cells (19). On Xi, chromosome-wide accumulation of Xist RNA and H3-3meK27 depends on Dcr. These ideas support the emerging concept of nuclear RNAi processes in mammals (20, 21). Because Dcr is not known to cleave RNAs to 25 to 42 nt, the observed effects on XCI may be partially indirect. Alternatively, Dcr may have properties yet to be discovered in mammals. XCI now provides a new system in which to investigate RNAi processes.

Supporting Online Material

Materials and Methods

Figs. S1 to S5


References and Notes

View Abstract

Navigate This Article