m(6)A RNA methylation promotes XIST-mediated transcriptional repression

Abstract

The long non-coding RNA X-inactive specific transcript (XIST) mediates the transcriptional silencing of genes on the X chromosome. Here we show that, in human cells, XIST is highly methylated with at least 78 N⁶-methyladenosine (m⁶A) residues-a reversible base modification of unknown function in long non-coding RNAs. We show that m⁶A formation in XIST, as well as in cellular mRNAs, is mediated by RNA-binding motif protein 15 (RBM15) and its paralogue RBM15B, which bind the m⁶A-methylation complex and recruit it to specific sites in RNA. This results in the methylation of adenosine nucleotides in adjacent m⁶A consensus motifs. Furthermore, we show that knockdown of RBM15 and RBM15B, or knockdown of methyltransferase like 3 (METTL3), an m⁶A methyltransferase, impairs XIST-mediated gene silencing. A systematic comparison of m⁶A-binding proteins shows that YTH domain containing 1 (YTHDC1) preferentially recognizes m⁶A residues on XIST and is required for XIST function. Additionally, artificial tethering of YTHDC1 to XIST rescues XIST-mediated silencing upon loss of m⁶A. These data reveal a pathway of m⁶A formation and recognition required for XIST-mediated transcriptional repression.

Extended Data Figure 1. Validation of RBM15 and RBM15B antibodies for iCLIP, construction and comparison of iCLIP library replicates

a, RBM15 and RBM15B exhibit high sequence homology. RBM15 and RBM15B comprise three RRM domains (RRM1, 2 and 3, all in purple) and a C-terminal SPOC domain (green). These domains show high sequence identity between RBM15 and RBM15B (indicated on the shaded areas that connect the compared regions). RRM, RNA recognition motif; SPOC, Spen paralogue and orthologue C-terminal. b, c, Validation of specificity of RBM15 and RBM15B antibodies for iCLIP, performed using immunoprecipitation. In each experiment, we used high (H) and low (L) RNase, as per the iCLIP validation protocol30 (see Methods). The bottom western blots are loading control (GAPDH). To confirm knockdown, RBM15 and RBM15B protein levels are shown. Additionally, we show the amount of protein in the anti-RBM15 or anti-RBM15B pulldowns. These experiments confirm that the RBM15 and RBM15B are knocked down after siRNA transfection. d, e, Autoradiograms of the samples used for the RBM15 and RBM15B iCLIP experiments. Shown are the representative autoradiograms from the nitrocellulose blots of samples used for preparing the RBM15 and RBM15B iCLIP library. The excised portion of the membrane is shown (red square). The red arrow indicates the position of RBM15 and RBM15B protein after high RNase treatment that matches with the size seen in b and c respectively. Both RBM15 and RBM15B show specific RNA–protein conjugates of expected size with a minimal contamination of RNA–protein conjugates of other sizes. f, g, RBM15 and RBM15B iCLIP replicates show reproducible iCLIP tag coverage on the human genome. Three iCLIP library replicates were prepared for RBM15 and RBM15B. We compared the normalized tag counts of replicates in 100 nucleotide bins in the human genome on scatter plots, and estimated the Pearson correlation coefficient (r). Shown are the representative scatter plots (left), and heat maps (right) showing the obtained r value in multiple pairwise replicate comparisons. rep1–rep3, replicate 1–replicate 3 for each protein; RBM15 in f and RBM15B in g. The x and y axes of the scatter plots represent normalized tag counts in uTPM in 100 nucleotide bins on the human genome in rep1 and rep3, respectively. Correlation values are indicated on each tile. From this analysis, RBM15 and RBM15B iCLIP replicates show a similar, highly reproducible iCLIP tag coverage on the human genome. The diagonal dashed line in scatter plots represents reference trend line for a perfect correlation (r = 1, x = y). h, RBM15 and RBM15B show similar binding preferences on XIST. Each of the 30 clusters in the RBM15 data set overlapped with the clusters in the RBM15B data set. We also examined the CITS induced by RBM15 and RBM15B. CITS are single-nucleotide sites that represent direct contacts of these proteins with XIST (Supplementary Tables 3, 4). Most RBM15 CITS (23 out of 37) overlapped with RBM15B CITS (top). This overlap was statistically significant (P < 0.0001) based on a permutation analysis in which we measured the overlap of randomly selected sites on XIST for RBM15 and RBM15B (see Methods). Lastly, a pairwise analysis of iCLIP tag density at each CITS showed that RBM15 and RBM15B binding was highly correlated (bottom).

Extended Data Figure 2. Quantification of X-linked gene silencing upon knockdown of m⁶A readers and writers

a, b, Quantification of Gpc4 spots upon Rbm15 and Rbm15b knockdown (Fig. 1b, c). The number of Gpc4 spots before and after XIST induction (−Dox and +Dox, respectively) (a). Representative RNA-FISH images with DAPI-stained nuclei with Gpc4 spots (green) and XIST staining (pink, last column) are shown (b). The number of Gpc4 spots is indicated on each FISH image. Scale bar, 5 μm. Data in a are mean ± s.e.m. NS, not significant; *****P < 0.0001 relative to Dox-deficient control by unpaired two-sample t-test. c, m⁶A modification is necessary for XIST-mediated gene silencing in female pSM33 cells. Quantification of Gpc4 RNA spots with and without induction of XIST expression (left). Representative RNA-FISH images showing Gpc4 RNA spots (green) with DAPI-stained nuclei (right). Wild-type (WT) cells show a normal XIST-induced silencing whereas Gpc4 spots are partially reduced (24 to 17 spots). Similar to male ES pSM33 cells, female ES cells fail to show XIST-mediated gene silencing upon knockdown of Rbm15/15b or Mettl3. Error bars mean ± s.e.m. for 50 cells per sample. NS, not significant; ****P < 0.0001, relative to no-doxycycline control by unpaired two-sample t-test. d, e, Similar to Fig 3c, d, shown is an siRNA pool that targets a (different) region on Mettl3. The data from Fig. 3c, d for the siRNA pool 1 is also shown here for comparison. In both the siControl and siMETTL3-transfected cells, XIST shows aggregation consistent with its interaction with the X chromosome. Thus, early steps of XIST interaction with the X chromosome may not require m⁶A. Gpc4 counts (d, top) and the change in transcription, as measured by the ratio of Gpc4 +Dox/−Dox. Notably, there is a reduction in Gpc4 and Atrx spots (see Fig. 3d) in siMETTL3-transfected cells, even in the absence of XIST expression. Representative FISH images with DAPI nuclear stain in blue, Gpc4 in green and XIST in pink (e). Following Dox treatment, the number of Gpc4 spots is markedly reduced in the control-transfected cells. However, after knockdown of Mettl3, the number of Gpc4 mRNA spots remain unchanged. Scale bars, 5 μm. Data in c are mean ± s.e.m.across 50 cells. NS, not significant; *****P < 0.0001 relative to no-doxycycline control (top graph) and siControl (bottom graph) by unpaired two-sample t-test. f, g, Similar to d and e, we show a defect in XIST-mediated silencing upon silencing of Ythdc1 as shown in Fig. 4d, e using multiple siRNA pools from different vendors. Targeting a different region of DC1 using a siRNA pool (siDC1-Q) prevents XIST-mediated gene silencing. The data from Fig. 4d, e for the Dharmacon siRNA pool is shown alongside. Data in f are mean ± s.e.m across 50 cells. NS, not significant; ****P < 0.005 relative to no-doxycycline control (top graph) and siControl (bottom graph) by unpaired two-sample t-test. h, DF1, DF2, DF3 and DC2 do not mediate XIST-mediate gene silencing. Quantification of Gpc4 (top left) and Atrx (bottom left) RNA-FISH spots is shown. Representative FISH images with DAPI-stained nuclei (blue) with Gpc4 (green) and Atrx (red) spots are shown (right). The number of detected RNA spots for both the genes are indicated on each FISH image. Scale bars, 5 μm. Data are mean ±s.e.m. across 50 cells from one experiment. ****P < 0.0001 relative to control (−Dox) by unpaired two-sample t-test. i, RBM15/15B and DC1 mediate XIST-mediate gene silencing in differentiating wild-type female ES cells. Quantification of Gpc4 RNA expression was performed in female mouse ES cells in response to retinoic acid-induced (+RA) differentiation by RNA-FISH (left). Representative FISH images showing DAPI-stained nuclei (blue), Gpc4 RNA (green), and XIST (pink) are shown (right). Wild-type cells exhibit normal Gpc4 silencing in response to retinoic acid treatment. Single knockdown of either Rbm15 or Rbm15b also exhibited normal silencing of Gpc4. Double knockdown resulted in no XIST expression (C.-K.C. and M.G., data not shown), reminiscent of the lack of XIST expression in METTL3-deficient ES cells. CRISPR-mediated homozygous knockout of DC1 (Ythdc1^{−/ −}) cells could not be recovered, suggesting that deletion of this gene is lethal. However, heterozygous knockout of DC1 (Ythdc1^−/+) impaired Gpc4 silencing in response to retinoic acid in these cells. These data support the idea that DC1 is required for silencing of X-linked genes during ES cell differentiation. ****P < 0.0001 relative to control by unpaired two-sample t-test. j, qRT–PCR-based validation of effects of RBM15/15B and DC1 on XIST-mediated gene silencing. Gene expression level after XIST induction (+Dox) was normalized to Gapdh before XIST induction (−Dox) in both the siControl and siRbm15/siRbm15b double-knockdown sample. Quantification of the change in gene transcript levels upon expression of XIST is shown for Gpc4 and Atrx. Dox-induced XIST expression led to reduced transcription of both the genes in Control knockdown cells. However, Rbm15 and Rbm15b double knockdown and DC1 knockdown failed to show XIST-induced silencing. **P < 0.01 relative to siControl-transfected cells by unpaired two-sample t-test.

Extended Data Figure 3. Reciprocal co-immunoprecipitation of METTL3–RBM15/15B complex, validation of WTAP, RBM15 and RBM15B knockdown and their lack of effect on XIST levels

a, b, Confirmation of WTAP-dependent METTL3–RBM15/15B interaction by reciprocal co-immunoprecipitation. METTL3 was immunoprecipitated using an antibody against the endogenous protein from nuclear extracts of the siControl- and siWTAP-transfected HEK293T cells under native conditions. Both RBM15 and RBM15B were detected in the METTL3 immunoprecipitates by western blot. The binding of both these proteins was significantly reduced in siWTAP-transfected cells, indicating that METTL3 interacts with RBM15/15B in a WTAP-dependent manner to form a RBM15/15B–WTAP–METTL3 complex. IgG heavy chain signal prevents visualization of WTAP; however, knockdown is seen in the input sample. c, Relative protein band intensities for RBM15/15B–METTL3 co-immunoprecipitation experiments. Shown here are the relative protein band intensities obtained in western blots of RBM15/15B–METTL3 and reciprocal co-immunoprecipitation experiments shown in Fig. 2a and Extended Data Fig. 3a, b, respectively. For METTL3 in RBM15 IP, n = 3; METTL3 in RBM15B IP, n = 3; RBM15 in METTL3 IP, n = 7; and RBM15B in METTL3, n = 3. d, Confirmation of WTAP, RBM15, and RBM15B knockdown. siRNA-transfected HEK293T cell lysates used for assays in Figs 2b, 3b were probed for protein levels using western blot analysis. Knockdown resulted in a significant reduction in the corresponding proteins. None of the siRNA affect METTL3 levels. The antibody for RBM15B recognizes a doublet, but only the lower band is lost after the knockdown. The specificity of this antibody for iCLIP is demonstrated in Extended Data Fig. 1b–e. e, Knockdown of WTAP, RBM15 and RBM15B, as well as double knockdown of RBM15 and RBM15B do not affect XIST RNA levels. Quantification of XIST levels by qRT–PCR from RNA purified from siRNA-transfected cells shows no significant change in XIST RNA levels. f, Validation of the anti-m⁶A antibody approach for pulldown of methylated XIST RNA. To validate XIST quantification used in Fig. 3b, we used a control spike-in RNAs with a single m⁶A, and a eGFP control RNA with no m⁶A residues. Unlike the m⁶A RNA (left), the non-methylated RNA (right) is de-enriched in the immunoprecipitation sample. NS, not significant. g, RBM15/15B bind XIST in m⁶A-independent manner. RBM15/15B binding of XIST in cells deficient in components of the m⁶A methylation machinery (METTL3 and WTAP) is shown. RBM15 and RBM15B were immunoprecipitated and XIST levels were determined by qRT–PCR at three regions (regions 1–3 refer to Fig. 2b, 3a and Extended Data Fig. 4a). XIST binding to RBM15 and RBM15B remains unchanged upon METTL3 and WTAP knockdown at region 1 and 2 where RBM15/15B both show binding. Thus, RBM15 and RBM15B are not binding to XIST in an m⁶A-dependent manner and are not m⁶A readers. At region 3, where both proteins do not show any binding, a basal level of amplification was seen similar to the level detected in IgG control. NS, not significant relative to siControl transfected cells by unpaired two-sample t-test.

Extended Data Figure 4. Zoomed-in views of miCLIP, RBM15 and RBM15B iCLIP tracks on XIST

a, m⁶A residues are broadly distributed along XIST. Shown are m⁶A residues mapped in XIST using miCLIP17;these sites are indicated with red lines. Total RNA at every genomic position are shown in purple. RNA-seq read distribution is shown in grey. Many of the m⁶A sites are clustered in a 2 kb domain surrounding the A-repeat (yellow) region. The zoomed-in region shows m⁶A sites (red lines) and miCLIP tag distribution in a 1-kb region closest to the A-repeat region. Region 1, which contains RBM15/15B-binding sites (see Fig. 2b) is also indicated. b, c, RBM15 and RBM15B bind XIST near m⁶A sites. To determine whether RBM15/15B-binding sites are in proximity to known m⁶A sites, we compared the iCLIP tag clusters with m⁶A sites on XIST. Shown in b are the RBM15 and RBM15B iCLIP, and miCLIP tag distributions on XIST. m⁶A sites are marked with red bars above the XIST gene model. Vertical green shaded boxes mark the regions of miCLIP and RBM15/15B iCLIP tag cluster alignments. A zoomed-in view of a region with high-tag abundance (bottom left) and another with low-tag abundance (bottom right), show examples of m⁶A sites that are in proximity to RBM15B and RBM15B tag clusters. Normalized tags are shown in uTPM. In c, the median distance of RBM15 (left) and RBM15B (right) CITS to the nearest m⁶A site on XIST was determined and compared with a randomly permuted data set of RBM15- and RBM15B-binding sites. RBM15/15B-binding sites show a marked proximity to m⁶A compared to randomly positioned RBM15/15B sites (RBM15, **P = 0.0026, number of permutations, 10,000; RBM15B, ***P = 0.0001, number of permutations, 10,000). This proximity is not due to RBM15 or RBM15B itself binding m⁶A as its binding to XIST was unaffected by METTL3 or WTAP knockdown (Extended Data Fig. 3g). The red dashed line indicates the location of m⁶A sites.

Extended Data Figure 5. RBM15 and RBM15B bind near m⁶A sites on mRNA

a, RBM15/15B binds at-or-near-to m⁶A sites throughout the transcriptome, including at m⁶A sites in XIST and ACTB mRNA. Shown are plots with an average binding-per-base around m⁶A (red curve) or non-m⁶A DRACH (green curve) sites for RBM15 (top left) and RBM15B (top right). The bottom two panels present the tag count per base around m⁶A or non-m⁶A DRACH sites as heat maps. Each row in the heat map is an m⁶A or non-m⁶A site. RBM15 and RBM15B show increased binding at or near m⁶A sites than at non-methylated DRACH sites (~3–4-fold higher). b, RBM15 and RBM15B bind near m⁶A sites on mRNA. Shown is the RNA-seq read (grey), and iCLIP (light blue, RBM15; dark blue, RBM15B) and miCLIP (purple) tag distribution on ACTB mRNA. iCLIP CITS sites are indicated below their respective tracks. miCLIP-identified m⁶A sites are indicated with red bars. Both proteins (light versus dark blue tracks) show a similar binding profile on ACTB mRNA, with considerable overlap of miCLIP tags at various regions along the sequence (vertical green shading). A zoomed-in view of the tag distribution is shown in the bottom panel. The sense DNA sequence of the zoomed-in region is shown above the gene model. A vertical dotted black line running through the middle of the tracks connects the RBM15/15B-binding sites with the DNA sequence that indicates the sequence at the binding site (highlighted yellow). At single-nucleotide resolution RBM15/15B binds a U-rich sequence near m⁶A sites on mRNA also. The binding sites show a clear separation (5 nucleotides) from the putative m⁶A-containing GAC sequence (red bars). RNA-seq reads are shown in absolute read counts, iCLIP and miCLIP tags are shown in uTPM. c, d, Genomic and transcriptomic distribution of RBM15- and RBM15B-RNA binding sites. To determine the types of RNA sequence that contain bound RBM15 and RBM15B, the top 1,000 iCLIP CITS (P < 0.0001) with the highest iCLIP tag coverage (in uTPM) were mapped to different features of the human genome and the overall distribution was determined. Sites mapped to mRNA (blue) represent roughly an equal fraction of all the binding sites of the proteins (~35%). To determine the overall distribution of the RNA-binding sites in mRNA, we further plotted the distribution of all the RBM15- and RBM15B-binding sites on a virtual transcript (shown in d). Metagenes for both RBM15- and RBM15B-binding sites show a similar distribution of the binding sites on the different features of mRNA. Although this metagene shows coverage all along mRNA, as is seen with m⁶A, this distribution does not match the m⁶A metagene. CDS, coding sequence; UTR, untranslated region. e, RBM15 and RBM15B bind U-rich RNA consensus motif. Shown are motifs enriched in both RBM15- and RBM15B-binding sites and the percentage distribution of the sites containing the identified motif is indicated below each motif. U-rich RNA binding motifs (shown as T in this genome-based alignment) were significantly enriched in the sequence at or around the iCLIP-identified RBM15- and RBM15B-binding sites (P < 0.0001). The absence of an m⁶A-like DRACH motif for both the proteins indicates that RBM15/15B does not directly bind m⁶A or DRACH sequences. Notably, the U-rich motif seen with RBM15/15B resembles the uracil-rich HNRNPC-binding motif, which may account for the previously observed proximity between m⁶A and HNRNPC-binding sites. f, g, Knockdown of RBM15B and RBM15B reduced m⁶A levels in cellular mRNA. Schematic diagram of a 2D-TLC (left, f) showing the migration pattern of monophosphate nucleotides after TLC separation. Shown are relative positions of m⁶A (orange dotted circle) and those of adenosine (A), cytosine (C), and uracil (U) (black dotted circles). Arrows indicate the direction of solvent migration in the two dimensions. Middle and right panels show radiochromatograms obtained from 2D-TLC of poly(A) RNA from control and RBM15/RBM15B double-knockdown HEK293T cells. Double knockdown of RBM15 and RBM15B leads to a considerable decrease in m⁶A levels in mRNA (spots marked with black arrow in the middle and right panel). Quantification of m⁶A levels calculated using m⁶A:A + C + U ratio from mononucleotide intensity in two independent biological replicates (g).

Extended Data Figure 6. Validation experiments for iCLIP of YTH proteins: anti-YTH antibodies, and library construction

a, Schematic representation of domain structures of human YTH proteins: DF1, DF2, DF3, DC1 and DC2. The YTH domain (blue) is located internally in DC1, while it is at the C-terminal region in the other proteins. DC1 has a different domain organization to DC2 and the similar DF proteins. The low-complexity and Glu-rich regions are indicated, as are the R3H, DEXDc, ankyrin repeats (ANK), HELICc, HA2 and OB-fold domains. The length of the protein is indicated next to each protein name. b, Validation of DF1, DF2 and DF3 antibody specificity via western blot. Full-length DF1, DF2, and DF3 were expressed as His6-fusion proteins in E. coli. IPTG was used as an inducer of protein expression (-, non-IPTG-treated; +, IPTG-treated). For anti-DF1, His6-DF1 was the major band detected but trace levels of His6-DF2 and His6-DF3 could be detected at longer exposure times. Thus, anti-DF2 and anti-DF3 antibodies are highly specific, while anti-DF1 shows a strong preference towards DF1 over the other DF proteins. c–g, Confirmation of iCLIP antibody pulldown specificity. Autoradiograms of the ³²P-labelled RNA-crosslinked protein conjugates on nitrocellulose membrane (top) for DF1 (c), DF2 (d), DF3 (e), DC1 (f) and DC2 (g) are shown. High (H) and low (L) RNase are used in accordance with the iCLIP validation protocol30 (see Methods). The red arrow indicates the expected size of the YTH protein. In each case, knockdown of the YTH protein mRNA (lanes 3 and 4) abolished RNA pulldown. GAPDH was used as a loading control. To confirm knockdown, YTH protein levels in the input samples and in the anti-YTH pulldown is shown. Antibodies and their antigenic peptide regions on the target proteins are provided in Supplementary Table 1. siRNA and shRNA target sequences in mRNA are listed in Supplementary Table 9. h–l, Autoradiograms from the nitrocellulose blots of samples used for each iCLIP library replicate. For each YTH protein, four biological replicates (rep1–4) were prepared. The red arrow confirms the position of the YTH protein after high RNase treatment and matches the size seen in c–g. Typically, UV crosslinking causes an increase in the intensity of the ³²P signal at the expected size of the YTH proteins (red arrow), indicating the formation of RNA-protein conjugates (lane 1 versus 3 in all panels). In the case of DC1, there is some ³²P signal even in the absence of UV crosslinking (lane 1 versus 3 in k). This type of background signal is due to autophosphorylation activity of the protein or of a co-immunoprecipitating protein kinase that phosphorylates DC1. RNase-sensitive smears were obtained for all of the YTH proteins (compare lanes 4–7 to lane 3 in h–l). Experiments using protein A/G beads that did not include the antibody (lane 2) did not show any signal in the region of interest. Overall, all the replicates of each YTH protein show highly specific RNA–protein conjugates of expected size with a minimal contamination of RNA–protein conjugates of other sizes. The eluted RNA material was used for constructing iCLIP libraries. Shown below the autoradiograms are western-blot loading controls (GAPDH and each YTH protein) as well as the controls that confirm the presence of the YTH protein in the immunoprecipitates.

Extended Data Figure 7. Comparison of transcriptome-wide RNA-binding sites of endogenous YTH proteins by iCLIP

a–e, YTH iCLIP library replicate reproducibility. For each YTH protein (DF1, DF2, DF3, DC1 and DC2), four independent biological replicate iCLIP libraries were constructed. Reproducibility of iCLIP tag coverage was assessed as in Extended Data Fig. 1f, g. Normalized iCLIP tag counts in uTPM from different replicates were compared on a scatter plot and the Pearson correlation coefficient (r) was determined. Scatter plots comparing mean tag counts of rep1 and rep2 (x axis), and rep3 and rep4 (y axis) are shown (left). A similar analysis was carried out for pairwise comparison of all the iCLIP replicates. The obtained correlation coefficients are shown on the heat map (right). The colour of the tiles in the heat map indicates the r value. YTH iCLIP replicates show similar and highly reproducible iCLIP tag coverage. The diagonal dashed line represents reference trend line for a perfect correlation (r = 1, x = y). f, Enriched motifs for each YTH protein based on transcriptome-wide iCLIP binding data. Motif analysis of the binding sites recognized by DF1, DF2, DF3 and DC1 proteins in the iCLIP data showed a DRACH sequence as the most prominent motif, which matches the known consensus motif for m⁶A in the transcriptome. DC2 also showed the DRACH motif, as well as other motifs, which probably reflects its numerous RNA-binding domains (Extended Data Fig. 6a). DC1 predominantly bound DRACH in various transcriptomic and genomic features (Supplementary Table 10). These data suggest YTH proteins bind m⁶A in cells. RNA-binding motifs were identified using MEME analysis (see Methods). The percentage of analysed sites containing the identified motifs is shown in the top right. P values were obtained using the MEME CentriMo tool by a one-tailed binomial test. g, Global comparison of the distribution of YTH-binding sites and m⁶A sites on mRNA. We compared the metagene for each YTH protein binding site to the previously reported metagene of single-nucleotide resolution miCLIP-identified m⁶A sites on mRNA17 (YTH protein, green; miCLIP, orange). Each curve represents a kernel density (y axis) plot of CITS distribution on a virtual transcript (x axis). Transcription start site, 5′ UTR, start codon (AUG), CDS, stop codon, and 3′ UTR are indicated on the virtual transcript. Vertical dashed lines mark UTR-CDS boundaries. Owing to a small number of DC2 sites that map to mRNA, a metagene for DC2 is not shown. h, Pairwise comparison of YTH iCLIP tag coverage at m⁶A sites shows distinct binding preferences for DC1. To determine whether YTH proteins recognize similar m⁶A sites on the basis of iCLIP tag coverage, we estimated the correlation coefficient of iCLIP tag coverage at each of 11,530 m⁶A sites in the transcriptome from a pairwise comparison of two YTH iCLIP libraries at miCLIP-identified m⁶A sites. The Pearson correlation coefficients are shown as a heat map. To identify YTH proteins that show similar binding preferences, libraries were hierarchically clustered based on the obtained correlation coefficients (see dendrogram below the heat map). This indicates that the DF proteins cluster together and show a similar binding pattern, and these proteins target similar m⁶A sites. Both DC1 and DC2 have atypical m⁶A site preferences; DC2 has a weak correlation with known m⁶A sites. Scatter plots used for this comparison are shown in Extended Data Fig. 8a. i, Genomic distribution of RNA-binding sites. To determine the genomic distribution of preferred YTH protein-RNA binding sites, the statistically significant top 1,000 iCLIP CITS (P < 0.0001) with the highest iCLIP tag coverage (in uTPM) were mapped to different features of the human genome. DC1 exhibits prominent binding to ncRNAs (19% of top thousand CITS), while less than 2% of the DF1, DF2 or DF3 CITS lie within annotated ncRNAs, including lncRNAs. Most DF1-, DF2- and DF3-binding sites are located in mRNAs and introns. DC2 had negligible coverage in mRNAs, and predominantly bound to introns and intergenic sequences.

Extended Data Figure 8. DC1 preferentially binds to a subset of m⁶A sites that are primarily localized to XIST and other ncRNAs

a, Pairwise comparison of YTH iCLIP libraries, and identification of DC1 preferred m⁶A sites. Shown are data used to generate the heat map in Extended Data Fig. 7h. In each pairwise analysis, two YTH proteins were compared for their binding to each m⁶A residue using normalized tag counts (see Methods), providing an estimate of the preferred binding partner for each m⁶A site for each YTH protein comparison. Tag counts in a window surrounding each m⁶A genomic coordinate (10 bp upstream and downstream) were determined for each YTH protein. Scatter plots are shown for each pair of indicated YTH proteins. m⁶A sites are plotted as points in which x and y coordinates represent the tag counts in the compared libraries. The DF family of proteins show highly similar binding preferences as indicated by their high Pearson correlation coefficients (r, top right corner of each plot). Hierarchical clustering as shown in Extended Data Fig. 7h supports the overall relatedness of the binding preferences of DF proteins. However, DC1 and DC2 show a pattern different from the DF proteins. DC2 shows low tag coverage on most m⁶A sites, and thus yields low r values. Notably, DC1 shows a global de-enrichment of binding at DF1, DF2 and DF3-preferred sites as seen by the flattened trend line (green). Additionally, DC1 shows enrichment at a unique set of m⁶A sites (the 1% of sites furthest from the trend line is highlighted with a red dashed ellipse in the comparison between DC1 and DF1, DF2 and DF3). b, A Venn diagram showing the number of sites preferred for DC1 over DF1, DF2 and DF3. The vast majority (105, white shaded area) are the same between each comparison, meaning these sites are preferred by DC1 over any DF protein. The rightward projection shows that most of these m⁶A sites are in ncRNA, constituted primarily of XIST m⁶A sites. c, Sequence logo analysis shows that the DC1-preferred m⁶A sites conform to the DRACH-like m⁶A consensus motif seen throughout the transcriptome, not in a novel DC1-specific motif. d, Zoomed-in views of iCLIP tag distribution on XIST for the five YTH proteins on XIST. The miCLIP tag distribution also identifies regions enriched in m⁶A. Only DC1 (blue) exhibits prominent iCLIP tags on XIST, the other YTH proteins do not. Vertical green shading marks the regions of XIST that contain the highest density of m⁶A sites. RNA-seq reads are shown in read counts, iCLIP and miCLIP tags are shown in uTPM. Regions 1 and 2 contain RBM15/15B-binding sites, region 3 does not. These sites are indicated by coloured boxes. DC1 shows a higher number of iCLIP tags at regions 1 and 2, areas containing several m⁶A sites. Although region 3 (grey) shows a putative m⁶A site, DC1 shows poor binding, possibly owing to the structural organization of XIST. e, HNRNPA2B1 does not bind m⁶A sites on XIST. HNRNPA2B1 was previously shown to bind m⁶A sites on primary micro RNA (pri-miRNA) transcripts35. We compared HNRNPA2B1 HITS-CLIP and miCLIP17 tag coverage (±10 bp in uTPM) at 11,530 annotated m⁶A sites, and determined correlation coefficients for m⁶A sites in mRNA (red) and in ncRNA (blue). HNRNPA2B1 does not show any significant binding to m⁶A sites on mRNA and ncRNA. Notably, the miCLIP-identified m⁶A sites17 used in this analysis lacks m⁶A sites from pri-miRNAs.

Extended Data Figure 9. DC1 binds XIST m⁶A in an METTL3-, RBM15-, and RBM15B-dependent manner

a, DC1 interacts with XIST in an RBM15/15B-dependent manner. Quantification of XIST in DC1 immunoprecipitates at regions 1 and 2 (left) by RNA immunoprecipitation followed by qPCR. Western blot analysis of protein from the siRNA-transfected cells (right). Knockdown of METTL3, WTAP, RBM15 and RBM15B leads to a significant decrease in XIST enrichment from DC1 immunoprecipitates with RBM15/RBM15B double knockdown exhibiting the greatest decrease. These data indicate that DC1 binds XIST RNA in a METTL3/RBM15/15B-dependent manner. Region 3 showed no reproducible and detectable amplification, possibly owing to the poor binding of DC1. In Extended Data Fig. 8d, region 3 shows a very low DC1 iCLIP tag coverage. Data are mean ± s.e.m. for three independent experiments. ***P < 0.0001 relative to XIST levels in siControl-transfected cells by unpaired two-sample t-test. b, Validation of DC1 antibody for immunofluorescence. Images of shLacZ- and shDC1-transfected HEK293T cells probed with DC1 antibody. DC1 exhibits a nuclear localization (red). In eGFP-expressing shDC1-transfected cells (arrow), the DC1 antibody signal is substantially lower than in a non-transfected cell in the same field (compare red signal, bottom row). Control knockdown with shLacZ-expressing plasmid shows DC1 staining similar to the non-transfected cells in the same view (red channel, arrow). Nuclei were stained with DAPI. Scale bars, 10 μm. c, d, DC1 preferentially localizes to the XIST subnuclear compartment. 3D-SIM was used to examine the levels of DC1 in the XIST subnuclear compartment compared to an autosomal domain in HEK293T cells following DC1 immunofluoresence labelling and XIST RNA-FISH. HEK293T cells are triploid, and thus exhibit two inactive X chromosomes^,. Left, a representative image showing DC1 (green), XIST (red) and DAPI (nucleus, grey-white) staining. Right, 2× magnification of highlighted regions (squares). DC1 is enriched in the XIST domains over similar dense autosomal compartments (right, top two versus bottom two rows). A distribution analysis of 3D-object counts performed on the DC1 signal in the XIST and autosomal domains also shows a significant enrichment (d, number of nuclei = 5, total XIST domains = 10, total autosomal domains = 10). Regions A and B (yellow squares) highlight two DAPI-stained inactivated X-chromosome territories marked by the presence of XIST (red). Areas C and D (blue squares) mark DAPI-stained autosomal domains. Scale bar, 5 μm. In d, **P = 0.0023 using two-tailed Mann–Whitney test. e, Localization of DC1 in XIST territory is METTL3- and RBM15/15B-dependent. To determine whether DC1 localizes to the XIST subnuclear compartment in an m⁶A-dependent manner, the number of DC1 spots in the XIST domain after METTL3 and RBM15/RBM15B knockdown was assessed by 3D-SIM, followed by image analysis. Knockdown of METTL3 and RBM15/RBM15B led to a significant decrease in the XIST-localized DC1. Box plot shows distribution of percentage of DC1 molecules (green objects) in XIST domain from the different knockdown cells. 10 nuclei per knockdown; ***P = 0.0011, ****P = 0.0147 relative to control knockdown in a two-tailed Mann–Whitney test.

Extended Data Figure 10. Model for the role of m⁶A in XIST-mediated transcriptional silencing and DC1 protein–protein interaction network analysis

a, A model for m⁶A-dependent XIST-mediated gene silencing. RBM15/RBM15B is the portion of the m⁶A methylation complex (that is, RBM15/RBM15B–WTAP–METTL3) that binds XIST. This binding enables methylation of adjacent adenosine residues in DRACH consensus sites. The m⁶A residues act as recruitment sites for DC1, which may facilitate and stabilize the assembly of silencing proteins on XIST. b, Protein–protein interaction (PPI) network analysis identifies a multi-component pathway that might mediate efficient XIST-mediated gene silencing. DC1 has no known protein domain that could directly mediate repression of gene transcription. We mined the PINA2 database28 for the PPI network of DC1, as well as for proteins that interact with DC1-binding proteins and proteins that regulate XIST-mediated gene silencing (SHARP, HDAC3, HNRNPK, HNRNPU, NCOR2 (also known as SMRT), LBR, PRC1 and PRC2). A network of proteins that interact with DC1 is shown, as are the interactions of these proteins (subnetworks). Proteins that are linked to XIST-mediated silencing are indicated in pink (the PRC components) or orange. c–e, Subnetworks showing the presence of proteins involved in transcription repression. Gene Ontology terms were filtered from the main network in b. In c, the DC1-BMI subnetwork is shown. This interaction is based on co-immunoprecipitation of DC1 with BMI1, a component of the PRC complex required to maintain gene repression. BMI1 may recruit SHARP, which directly binds XIST and mediates the recruitment of HDAC3 on the X chromosome. The EMD (emerin) subnetwork shown in d, is significantly enriched in proteins involved in transcription repression (false discovery rate < 0.05, P < 0.05) (Supplementary Table 7). DC1 interacts with EMD, which is linked to proteins that are known to be necessary for XIST-mediated gene silencing (interactions indicated with bold red lines). A separate analysis of DC1 co-immunoprecipitated proteins identified by tandem immunoprecipitation followed by mass spectrometry analysis also shows the presence of SHARP and LBR proteins (interactions indicated with red dotted lines). Protein-binding partners of another DC1-interacting protein, KHDRBS1 (ref. 52) is shown in e. KHDRBS1 (also known as SAM68) is a well-known transcriptional repressor. Here KHDRBS1 is shown to interact with PRC component proteins SUZ12, EZH2, and RNF2. SUZ12 and EZH2 are components of the PRC2/EED–EZH2 complex that mediates histone methylation at K9 and K27 residues, leading to transcriptional repression. KHDRBS1 also interacts with XIST5. RNF2 is a component of PRC1 complex. RNF2 has E3 ubiquitin-protein ligase activity that mediates monoubiquitination of Lys119 of histone H2A (H2AK119Ub). Components of PRC1/2 and are found to be enriched on the inactivated X chromosome4.

Figure 1. RBM15 and RBM15B are necessary for XIST-mediated gene silencing

a, RBM15 and RBM15B show similar binding patterns in XIST. Shown is the distribution of normalized RBM15 and RBM15B iCLIP tags (in unique tags per million, uTPM) and statistically significant CITS. Light blue vertical lines, RBM15; dark blue vertical lines, RBM15B; P < 0.0001. b, c, Knockdown of both Rbm15 and Rbm15b (siRBM15/15B) impair XIST-mediated gene silencing. XIST expression was induced by doxycycline, and the X-linked genes Gpc4 (green) and Atrx (red) were quantified by RNA-FISH (b). Representative FISH images are shown with DAPI nuclear counterstain (blue) (c). The number of detected RNA spots for both genes are indicated on each image. Scale bars, 5 μm. Data are mean ± s.e.m. for 50 cells from one experiment. ***P < 0.001, *****P < 0.0001, relative to siControl by unpaired two-sample t-test. NS, not significant.

Figure 2. RBM15 and RBM15B recruit METTL3 to XIST

a, RBM15 and RBM15B interact with METTL3 in a WTAP-dependent manner. RBM15 (left) and RBM15B (right) were immunoprecipitated from HEK293T nuclear extracts. Co-immunoprecipitation of METTL3 was reduced in siWTAP-transfected cells. The IgG heavy chain (H chain) prevents visualization of WTAP; however, knockdown is seen in the input sample. NE, nuclear extracts. b, Quantification of METTL3-bound XIST upon knockdown of methylation machinery components. XIST was quantified by qRT–PCR using regions selected based on the presence or absence of RBM15- and RBM15B-binding sites (indicated with light blue and blue lines, respectively). Data are mean ± s.e.m. from three independent experiments. **P < 0.001, ***P < 0.0001, relative to siControl by unpaired two-sample t-test.

Figure 3. N⁶-adenosine methylation is necessary for XIST-mediated gene silencing

a, m⁶A residues (red lines) identified via miCLIP are broadly distributed along XIST. Normalized miCLIP17 tags are shown in purple. b, Methylation of XIST requires RBM15 and RBM15B. m⁶A levels in XIST were quantified by m⁶A-RNA immunoprecipitation followed by qRT–PCR of three m⁶A regions of XIST. Data are mean ± s.e.m. from six samples coming from three technical replicates of two biological replicates. ***P < 0.0001, **P < 0.001 relative to siControl by unpaired two-sample t-test. c, d, m⁶A promotes XIST-mediated gene silencing. XIST expression was induced by Dox, and X-linked genes Gpc4 (green) and Atrx (red) were quantified by RNA-FISH (c). Representative FISH images are shown (d). The number of detected RNA spots is indicated on each image. Scale bars, 5 μm. Data are mean ± s.e.m. for 50 cells from one experiment. ****P < 0.005 relative to siControl by an unpaired two-sample t-test.

Figure 4. DC1 binds XIST m⁶A residues and promotes XIST-mediated gene silencing

a, YTH iCLIP tag coverage at 11,530 annotated m⁶A residues. Correlation coefficients for mRNA m⁶A (grey) and non-coding RNA (ncRNA) m⁶A (magenta) are indicated. DF1, DF2 and DF3 show similar correlations between m⁶A abundance and YTH binding for mRNAs (blue line) and ncRNAs (magenta line). DC1 shows preference for ncRNA m⁶A, with the top 1% of DC1-bound m⁶A indicated (dotted ellipse). b, mRNA/ncRNA distribution of the top 1% of DC1-bound m⁶A sites. Most detected ncRNA m⁶A are present on XIST (indicated in green). c, Normalized tag distributions for each YTH protein on XIST shows predominantly DC1 binding. High-density m⁶A regions are indicated by green shading. d, e, Ythdc1 knockdown (siDC1) impairs XIST-mediated gene silencing. XIST was induced by Dox, and X-linked genes Gpc4 (green) and Atrx (red) were quantified by RNA-FISH. Representative FISH images are shown (e). The number of detected RNA spots is indicated on each image. Scale bars, 5 μm. Data are mean ± s.e.m. ****P < 0.005 relative to siControl by an unpaired two-sample t-test.

Figure 5. m⁶A-independent tethering of DC1 to XIST is sufficient to exert XIST-mediated gene silencing

a, Schematic of tethering approach. The 3′ end of XIST was genomically modified with three BoxB sequences (XIST–(BoxB)₃). m⁶A-dependent recruitment of DC1 is blocked in methylation-deficient cells; however, artificial tethering can be achieved with DC1–λN, which binds to the BoxB elements in XIST–(BoxB)₃. b, c, Dox-induced expression of XIST–(BoxB)₃ results in gene silencing in siControl-transfected cells, but not in siMETTL3 or siRBM15 and siRBM15B co-transfected cells. DC1–λN rescued silencing in these cells, suggesting that the primary function of m⁶A in XIST-mediated gene silencing is to recruit DC1 to XIST. Quantification of Gpc4 expression is shown in b. Representative FISH images showing DAPI-stained nuclei (blue), Gpc4 RNA (green), and XIST (pink) are shown in c. Scale bars, 5 μm. Data are mean ± s.e.m. in b for 50 cells from one experiment. *****P < 0.0001 by unpaired two-sample t-test.