RNA m⁵C oxidation by TET2 regulates chromatin state and leukaemogenesis

TET methylcytosine dioxygenases (TET1, TET2 and TET3) mediate oxidation of DNA 5-methylcytosine (5mC) to regulate gene expression in a wide range of different biological systems^6,7,8. Among them, TET2 is unique in that it distinctly exhibits high mutation ratios in myeloid malignancies (Extended Data Fig. 1a), with frequent IDH mutations observed in human cancers also thought to mainly act through TET2 inhibition^9,10,11. TET2 deficiency led to genomic DNA hypomethylation⁹, suggesting that functional outcomes caused by TET2 deficiency might not primarily associate with its DNA oxidation activity. TET2 is also unique among TET enzymes in that it is not covalently linked to the zinc-finger CXXC ___domain protein CXXC4 or CXXC5^12,13,14; the interaction between TET2 and CXXC4/CXXC5 is critical for DNA binding by TET2¹⁵ (Extended Data Fig. 1b). In mouse embryonic stem (mES) cells, it was shown that TET2 binds to PSPC1, an RNA-binding protein, to mediate RNA 5-methylcytosine (m⁵C) oxidation¹⁶. Other studies also reported RNA m⁵C oxidation by TET2 or Drosophila TET homologue^17,18,19,20. We and others have recently reported chromatin regulation through reversible N⁶-methyladenosine modification on chromatin-associated RNA (caRNA)^{21,22,23,24,25}. These advances prompted us to examine potential chromatin regulation through TET2-mediated caRNA m⁵C oxidation.

Tet2-knockout (KO) mES cells exhibited more open chromatin (Fig. 1a) and elevated global transcription (Fig. 1b) compared with the wild type (WT). Transcription rates of protein-coding genes were also higher in Tet2-KO mES cells when compared to those in WT (Extended Data Fig. 1c–e). The more-open chromatin state agrees well with the previously reported global DNA hypomethylation caused by TET2 deficiency²⁶ (Extended Data Fig. 1f) but is inconsistent with a predominant DNA 5mC oxidation function.

a,b, Representative immunofluorescence images and quantification characterizing chromatin accessibility (a) and nascent RNA synthesis rate (b) of WT and Tet2-KO mES cells. a.u., arbitrary units; ATAC–see, assay of transposase-accessible chromatin with visualization; EU, 5-ethynyl uridine. c, Representative immunofluorescence images characterizing the nascent RNA synthesis rate in Tet2-KO (left) or Pspc1-KO (right) mES cells, overexpressing either an EV control, WT mouse Tet2 (WT) or catalytically dead mouse Tet2 (HxD). For a–c, six images were taken for each condition. For a–c, scale bars, 40 μm. d, GSEA enrichment analysis between genes upregulated (upDEGs) after Pspc1 depletion and upregulated DEGs after Tet2 KO in mES cells. Left, whole-cell RNA-seq (Gene Expression Omnibus (GEO): GSE103269 and GSE48518). Right, caRNA-seq (this study). e, Spike-in-calibrated overall caRNA levels in WT mES cells, and Tet2-KO mES cells overexpressing either EV control, WT or HxD. Data are mean. For a–c, the box plots show the median (centre line), upper and lower quartiles (box limits) and 1–99% (whiskers). P values were calculated using two-tailed Wilcoxon–Mann–Whitney tests (a and b), two-tailed unpaired t-tests with Welch’s correction (c) and two-tailed permutation tests (d). n = 3 biological replicates (a–c and e). The depicted genome-wide data represent an integration of three biological replicates.

Source Data

Full size image

To examine the functional outcomes of RNA oxidation by TET2, we studied Pspc1-KO mES cells. Spike-in-calibrated assay for transposase-accessible chromatin using sequencing (ATAC–seq) revealed globally increased chromatin accessibility in both Pspc1-KO and Tet2-KO mES cells (Extended Data Fig. 1g), with the more-opened chromatin loci notably overlapping and correlating with each other (Extended Data Fig. 1h,i). The DNA 5mC levels did not significantly change, while an increase in caRNA m⁵C levels was detected after Pspc1 KO using ultra-high-performance liquid chromatography–tandem mass spectrometry (UHPLC–MS/MS) measurements (Extended Data Fig. 1j). The same chromatin openness and global transcription increases were observed in Pspc1-KO mES cells (Extended Data Fig. 1k,l). Overexpression of an RNA-binding-null PSPC1 mutant did not rescue these chromatin changes (Extended Data Fig. 1m). Moreover, overexpression of WT Tet2, but not its catalytic dead mutant (Tet2^HxD, hereafter HxD)²⁷, decreased the transcription rate and chromatin accessibility in Tet2 KO, but not in Pspc1-KO mES cells (Fig. 1c). Global chromatin accessibility profiling also demonstrated a similar trend (Extended Data Fig. 1n,o). These results indicate that the TET2-mediated chromatin compaction and transcription repression are dependent on its enzymatic activity on RNA.

Consistent with the role of TET2 in suppressing transcription, the upregulation of caRNAs in Pspc1-KO mES cells, rather than whole-cell mRNAs, exhibited a stronger correlation with the changes observed after Tet2 KO (Fig. 1d and Extended Data Fig. 1p). The elevated caRNA expression correlates with DNA hypomethylated regions instead of hypermethylated regions (Extended Data Fig. 1q). Overexpression of WT Tet2, but not HxD, was able to restore normal caRNA expression levels in Tet2-KO mES cells (Fig. 1e). Thus, while TET2 can act on either DNA or RNA, in mES cells, the TET2-mediated gene repression changes appear to be associated with its RNA-targeting activity.

We next investigated whether m⁵C on caRNA is a substrate of TET2. UHPLC–MS/MS identified m⁵C in ribosomal RNA (rRNA)-depleted caRNAs (Fig. 2a). Tet2 KO led to a notable increase in caRNA m⁵C level, accompanied by a decrease in the levels of its oxidation product 5-hydroxymethylcytosine (hm⁵C) (Fig. 2a and Extended Data Fig. 2a).

a, The RNA m⁵C levels and 5-hydroxylmethylcytosine levels in the chromatin-associated fraction. m⁵C/C and hm⁵C/C values were obtained by normalizing absolute concentrations of m⁵C and hm⁵C to C. b, The average profile and heat map of the m⁵C level in WT mES cells, along with the corresponding ATAC–seq signals in WT and Tet2-KO mES cells on repeat RNA. The left colour bar shows meRIP signal and the right colour bar shows ATAC–seq signal. c, The m⁵C enrichment at various repeat RNA families. The size of each dot corresponds to the number of loci (subfamilies in ERVK or L1 shown) that were m⁵C methylated in WT mES cells. Exact loci numbers are indicated in parentheses. SINE, short interspersed nuclear element. d, Cumulative curve showing the log₂-transformed fold change in repeat RNA expression after Tet2 KO. Repeat RNAs were grouped on the basis of their m⁵C enrichment quantified by log₂[IP/input] using m⁵C meRIP-seq data. No m⁵C, log₂[IP/input] < 1; low, 1 ≤ log₂[IP/input] < 2; medium: 2 ≤ log₂[IP/input] < 3; high, log₂[IP/input] ≥ 3. e, Local RNA m⁵C and DNA 5mC changes at IAP loci at different timepoints after DOX-induced dCas13–TET2-CD tethering. P values were determined by comparing values at the corresponding timepoints with values at 0 h, individually. qPCR, quantitative PCR. For d, the box plots show the median (centre line), the upper and lower quartiles (box limits) and 1.5× the interquartile range (whiskers). For a and e, data are mean ± s.d. P values were calculated using two-tailed unpaired t-tests with Welch’s correction (a and e). NS, P > 0.05. n = 3 biological replicates (a and e). The depicted genome-wide data represent an integration of three biological replicates.

Source Data

Full size image

We profiled caRNA m⁵C by methylated RNA immunoprecipitation followed by sequencing (meRIP–seq; Extended Data Fig. 2b,c). Most of the caRNA m⁵C peaks are in repeat RNA (Extended Data Fig. 2d). These m⁵C-marked repeat RNAs are associated with increased local chromatin accessibility (Fig. 2b and Extended Data Fig. 2e), with the long terminal repeat (LTR) and long interspersed nuclear element (LINE) families mostly enriched (Fig. 2c). We also performed quantitative amplicon sequencing analysis of selected amplicons among these repeat RNAs after ultrafast bisulfite treatment²⁸. We observed more caRNA m⁵C hypermethylation in these amplicons from Tet2-KO mES cells (Extended Data Fig. 2f–i). ERVK and L1 families bear the most hypermethylation sites (Extended Data Fig. 2h). We examined ERVK and L1 with IAPEz-int/RLTR10 and L1MdA_I/II subfamilies as their respective representatives because they are the top-ranked subfamilies observed in the m⁵C meRIP enrichment. Compared with empty vector (EV) control or HxD, overexpressing WT Tet2 caused notable decreases in chromatin accessibility in these repeat regions (Extended Data Fig. 2j). While intracisternal A particle (IAP) displayed an increase in local chromatin accessibility after either Tet2 or Pspc1 KO (Extended Data Fig. 2k), LINE1-associated chromatin accessibility increased only after Tet2 KO but not Pspc1 deletion (Extended Data Fig. 2k), suggesting that LINE1-associated chromatin changes may depend on TET2 in a PSPC1-independent manner.

We further analysed the correlation between caRNA expression changes with ATAC signals after Tet2 KO. Tet2 KO resulted in a global increase in chromatin accessibility (Extended Data Fig. 3a). While caRNA transcription only occurred at 46% of chromatin regions with identified ATAC peaks (Extended Data Fig. 3b), 60% of those with a more opened chromatin state also exhibited an increase in caRNA level after Tet2 KO (Extended Data Fig. 3c). Changes in chromatin accessibility correlate well with caRNA increases in these regions (Extended Data Fig. 3d). m⁵C-marked caRNA abundance showed even greater increases after Tet2 KO (Fig. 2d). These collectively suggest that chromatin-associated regulatory RNA (carRNA) m⁵C methylation regulates local chromatin accessibility.

NOP2/Sun RNA methyltransferase 2 (NSUN2) and DNA methyltransferase 2 (TRDMT1) are likely candidates that may install caRNA m⁵C, as both are known to localize in the cell nucleus and mediate RNA m⁵C methylation^29,30. Only Nsun2 depletion caused an approximately 70% decrease in caRNA m⁵C abundance (Extended Data Fig. 3e) without affecting the DNA 5mC levels (Extended Data Fig. 3f). Nsun2 knockdown (KD) also led to a more-closed chromatin state (Extended Data Fig. 3g,h). These changed regions largely overlap and negatively correlate with the more opened chromatin loci in Tet2-KO mES cells (Extended Data Fig. 3i,j). Moreover, transcriptome-wide alterations caused by Nsun2 depletion exhibited patterns that contrast the gene expression changes caused by Tet2 depletion (Extended Data Fig. 3k–n).

Purified TET2 protein is known to mediate oxidation of RNA m⁵C to hm⁵C¹⁸. We also identified an increase in caRNA m⁵C in Tet2-KO mES cells (Fig. 2a), with IAP RNA from the LTR family as a notable example (Fig. 2c). IAP also had most hypermethylated m⁵C sites in response to Tet2 KO (Extended Data Fig. 2g–i). Tet2 depletion led to increased caRNA IAP m⁵C levels (Extended Data Fig. 2f–i), increased local chromatin accessibility (Extended Data Fig. 2j,k) and accelerated transcription of its target RNAs (Extended Data Fig. 4a), suggesting that IAP RNA is a main downstream regulator of the TET2-mediated chromatin regulation in mES cells.

We next blocked IAP RNA methylation using an anti-sense oligo (ASO) targeting its main m⁵C site (Extended Data Fig. 4b). This ASO selectively blocked IAP RNA m⁵C installation (Extended Data Fig. 4b), with IAP RNA levels remaining almost unchanged (Extended Data Fig. 4c). Administration of this ASO led to more closed local chromatin at IAP loci (Extended Data Fig. 4d). Thus, m⁵C methylation on chromatin-associated IAP RNA could regulate the local chromatin state.

To establish the key causal relationship, we constructed the locus-specific RNA-targeting system (Extended Data Fig. 4b) by fusing dCas13 with the TET2 catalytic ___domain (TET2-CD). The dCas13–TET2-CD fusion protein was stably expressed with the guide RNA under the control of a doxycycline (DOX)-responsive Tet operator (Extended Data Fig. 4e). Acute expression of the guide RNA caused rapid dCas13–TET2-CD recruitment and reduction of RNA m⁵C methylation on IAP transcripts, followed by increased local DNA 5mC methylation (Fig. 2e and Extended Data Fig. 4e). By contrast, tethering of HxD did not alter DNA or RNA methylation, demonstrating that this effect is dependent on its oxidation activities but not the protein scaffolding effect³¹. The increased DNA 5mC methylation caused by TET2 targeting to RNA agrees with the widespread DNA hypomethylation or chromatin opening after TET2 inactivation frequently observed in embryonic stem cells, haematopoietic stem cells and cancer cells⁵. We therefore conclude that the global chromatin and transcriptional regulation effects of TET2 are most likely mediated through RNA m⁵C oxidation.

To further confirm this enzymatic-activity-dependent regulation, we treated mES cells with an inhibitor against TET enzymes³². Time-lapse tracking after treatment revealed early chromatin opening (Extended Data Fig. 4f,g) and caRNA m⁵C increases (Extended Data Fig. 4h (bottom)), followed by genomic DNA 5mC increases (Extended Data Fig. 4h (top)). These changes in chromatin state were most likely due to the RNA oxidation activity of TET2. By contrast, tethering TET2-CD to IAP loci using DNA-binding dCas9 led to expected DNA hypomethylation and increased chromatin accessibility (Extended Data Fig. 4i–l).

DNA hypermethylated and hypomethylated regions were both found in Tet2-KO mES cells (Extended Data Fig. 1f), accompanied by changes in chromatin accessibility in the opposite directions (Extended Data Fig. 5a–c). The increased caRNA expression was observed only in DNA hypomethylated regions after Tet2 KO in mES cells (Extended Data Fig. 1q). To separate effects on DNA versus RNA, we further analysed DNA 5mC changes caused by Tet2 KO at different genomic regions, including enhancers, promoters and repeats (Extended Data Fig. 5d). We observed a negative correlation between changes in enhancer transcription and DNA methylation resulting from Tet2 KO in mES cells, which contrasts with the pattern observed for repeat RNA (Extended Data Fig. 5d). Furthermore, these DNA hypermethylated regions resulting from Tet2 KO were enriched at enhancer and CXXC5-bound regions while being depleted in repeats (Fig. 3a), agreeing with the notion that DNA 5mC oxidation by TET2 in these enhancer regions leads to local transcription activation.

a,b, Feature enrichment in hypermethylated differentially methylated regions (DMRs) compared with hypomethylated DMRs (hyperspecific; a), and in hypomethylated DMRs compared with hypermethylated DMRs (hypospecific; b) using the odds ratio. c, Overlapping ratios of histone modifications with DNA hypomethylated regions for Tet2-KO versus WT mES cells. d, The H2AK119ub levels at IAP loci after TET2-CD (catalytic ___domain; WT or catalytically dead HxD) tethering by dCas13. e, Representative image showing nucleic acids cross-linked to MBD6. f, Spike-in-calibrated ATAC–seq signals or H2AK119ub signals in WT (siNC WT) and Tet2-KO (siNC Tet2 KO) mES cells, as well as in Mbd6-KD Tet2-KO (siMbd6 Tet2 KO) mES cells. g, Changes in H2AK119ub on m⁵C hypomethylated (hypo.), unchanged (invar.) and hypermethylated (hyper.) repeat RNA after Tet2 KO versus WT. h, H2AK119ub changes in m⁵C hypermethylated repeat RNA in Tet2-KO versus WT compared with Mbd6-KD Tet2-KO versus Tet2-KO mES cells. i, The correlation between ATAC signal fold changes in Tet2-KO versus WT mES cells, and comparing Mbd6 KD with control Tet2-KO mES cells. PCC, Pearson correlation coefficient. j, The correlation between changes in ATAC and H2AK119ub signals, comparing Mbd6 KD with control Tet2-KO mES cells. k, The H2AK119ub levels and ATAC signal at IAP RNA after dCas13–MBD6-MBD tethering. l, Schematics of the proposed pathway of MBD6 regulating chromatin state through caRNA m⁵C binding. Data are mean ± s.d. (d and k) and mean ± s.e.m. (f). For g and h, the box plots show the median (centre line), the upper and lower quartiles (box limits) and 1.5× the interquartile range (whiskers). P values were calculated using two-tailed unpaired t-tests with Welch’s correction (d and k), two-tailed Wilcoxon–Mann–Whitney tests (g and h), two-tailed t-distribution with n − 2 d.f. (i and j). n = 3 biological replicates (d–f and k). The depicted genome-wide data represent an integration of three biological replicates.

Source Data

Full size image

In contrast to enhancer regions, repeat loci were enriched in DNA hypomethylated regions after Tet2 depletion (Fig. 3b). Moreover, these DNA hypomethylated regions were more accessible after Tet2 or Pspc1 KO (Extended Data Fig. 5c). These suggest that the elevated chromatin accessibility resulting from TET2 depletion cannot be attributed to its 5mC oxidation activity in DNA but, rather, its oxidation activity on RNA m⁵C. We conclude that TET2 can mediate either DNA 5mC or RNA m⁵C oxidation by engaging different protein partners (Extended Data Fig. 5e). It is the repeat RNA (for example, LTR RNA) m⁵C oxidation that causes chromatin compaction, which dominates chromatin and transcription regulation in mES cells.

We computationally analysed histone modifications that correlate best with the m⁵C oxidation by TET2 on RNA. We found that H2AK119ub, a major chromatin repressive mark installed by PRC1³³, ranked as the top (Fig. 3c and Extended Data Fig. 5f). H2AK119ub can be erased by the polycomb repressive deubiquitylase (PR-DUB) complexes^33,34,35,36, and we also identified BAP1 (core component of PR-DUB) binding as one top enriched genomic feature (Fig. 3b). Consistently, tethering of WT dCas13–TET2-CD, but not HxD, increased H2AK119ub at the IAP loci (Fig. 3d). Time-lapse tracking of H2AK119ub and H3K27me3 marks³⁷ at IAP and control LINE loci after acute TET inhibition also showed an early response in H2AK119ub (Extended Data Fig. 5g). We next profiled H2AK119ub changes with and without Tet2 deletion. Approximately 37% of ATAC–seq peaks were marked with H2AK119ub (Extended Data Fig. 5h), among which H2AK119ub was downregulated in around 89% of these genomic regions (Extended Data Fig. 5i,j). Moreover, approximately 85% of genomic regions with higher ATAC signals overlapped well with H2AK119ub loss (Extended Data Fig. 5k,l). Thus, most of the opened chromatin regions caused by Tet2 KO display decreased H2AK119ub, which may initiate changes in local chromatin accessibility and transcription.

Previous studies have identified MBD5 and MBD6 as partner proteins of PR-DUB, and their localization to heterochromatin appeared to be independent of DNA 5mC^38,39. MBD5 and MBD6 both possess a conserved but structurally distinct methyl-binding ___domain (MBD) but do not bind to DNA³⁸ (Extended Data Fig. 6a). We speculated that these two proteins might bind to RNA m⁵C, which may then recruit PR-DUB to mediate H2AK119ub deubiquitination at the m⁵C-methylated LTR loci for transcription activation.

We examined nucleic acids bound by MBD5 or MBD6 in mES cells (Extended Data Fig. 6b). We observed cross-linked RNA but not DNA, as an RNase treatment almost completely abolished nucleic acid signals, while the effect of a DNase treatment was minor (Fig. 3e and Extended Data Fig. 6c). The purified MBD ___domain of MBD6 preferentially bound to a single-stranded oligonucleotide probe containing m⁵C over unmethylated or hm⁵C-modified probe (Extended Data Fig. 6d), and this binding was not affected by the binding of MBD6 to ASXL1 (Extended Data Fig. 6e), a protein that bridges MBD6 with PR-DUB⁴⁰. Cellular MBD5 and MBD6 also enrich m⁵C-containing RNA (Extended Data Fig. 6f). Thus, our data revealed that MBD5 and MBD6 are RNA-binding proteins that preferentially recognize RNA m⁵C.

While the RNA-binding targets of MBD5 and MBD6 substantially overlap with each other (Extended Data Fig. 6g), MBD6 appears to dominate the regulation of H2AK119ub and repeat RNA expression in mES cells, as KD of Mbd6 was sufficient to reverse elevated expression of LTRs caused by Tet2 KO, whereas Mbd5 KD did not do so (Extended Data Fig. 6h). The global H2AK119ub levels also significantly increased only after Mbd6 KD (Extended Data Fig. 6i). Mbd6 KD caused a global decrease in the caRNA m⁵C levels (Extended Data Fig. 6j). Consistent with this, IAP RNAs were stabilized in Tet2-KO mES cells and were destabilized after Mbd6 KD (Extended Data Fig. 6k). m⁵C methylation appears to stabilize LTR RNAs and this effect is mediated largely through MBD6. We therefore focused on MBD6 in subsequent studies, although MBD5 may have important roles in other cell types.

Functionally, Mbd6 KD was able to rescue the genome-wide increased chromatin accessibility (Fig. 3f (left)) and decreased the H2AK119ub levels (Fig. 3f (right)) in Tet2-KO mES cells. Consistently, we observed decreased H2AK119ub at caRNA m⁵C hypermethylated sites after Tet2 deletion (Fig. 3g), and this change could be largely reversed by Mbd6 KD (Fig. 3h). Mbd6 or Nsun2 KD (Extended Data Fig. 6l) both decreased chromatin openness caused by Tet2 KO in mES cells (Extended Data Fig. 6m), with consistent global decreases in chromatin accessibility (Extended Data Fig. 6n–r). Moreover, genomic regions with altered chromatin openness overlap and correlate well between different groups (Fig. 3i and Extended Data Fig. 6s,t; Tet2 KO versus WT, small interfering RNA against Mbd6 (siMbd6) versus siNC in Tet2 KO, and siNsun2 versus siNC in Tet2 KO). Local chromatin accessibility changes in response to Mbd6 KD exhibit a negative correlation with the resulting H2AK119ub increases (Fig. 3j). Acute recruitment of dCas13 fused to the methyl-CpG binding ___domain of MBD6 protein (dCas13–MBD6-MBD) was also sufficient to reduce local H2AK119ub and induce open chromatin at IAP loci (Fig. 3k).

Together, our results reveal that TET2-mediated caRNA m⁵C oxidation reduces both MBD6 binding and local histone H2AK119ub deubiquitination, leading to closed chromatin and transcription suppression (Fig. 3l). Whether the oxidation product hm⁵C may further promote caRNA degradation remains to be investigated in the future.

TET2 deficiency in haematopoietic stem and progenitor cells (HSPCs) is well known to cause open chromatin and genome instability, finally leading to myeloid malignancy⁴¹. The most important feature of TET2-deficient (Tet2^−/−) HSPCs (Lin⁻KIT⁺ cells, capturing HSPCs) is an enhanced self-renewal capacity of haematopoietic stem cells and skewed differentiation towards granulocytic/monocytic lineages in vitro⁴. Consistent with our observations in mES cells, we observed a global increase in chromatin accessibility in TET2-deficient HSPCs (Extended Data Fig. 7a). We designed a chimera assay to study effects of Mbd6 KD in vivo (Extended Data Fig. 7b), and performed competitive transplantation assays using WT + control short hairpin RNA (shNC), WT + shMbd6, Tet2^−/− + shNC or Tet2^−/− + shMbd6 CD45.2⁺ HSPCs versus CD45.1 competitor bone marrow (BM) cells (ratio of HSPCs at around 1:9; Fig. 4a). The donor cell (CD45.2⁺) chimerism in the recipients transplanted with Tet2^−/− + shNC HSPCs steadily increased, reaching around 50% at 6 months after transplantation (Fig. 4b), while the donor cell population in the peripheral blood (PB) of mice receiving Tet2 KO + shMbd6 HSPCs remained comparable to those receiving WT/shNC or WT/shMbd6 HSPCs at a low level around 5% (Fig. 4b). Similar trends for the donor cell (CD45.2⁺) chimerism were observed in cells from the BM or spleen at 6 months after transplantation (Fig. 4c). Consistently, the recipients transplanted with Tet2^−/− + shNC HSPCs exhibited mild splenomegaly, while the spleen sizes from animals receiving WT + shNC, WT + shMbd6 or Tet2^−/− + shMbd6 HSPCs were normal (Fig. 4d,e), demonstrating almost full rescue of the defects caused by Tet2 KO with additional Mbd6 KD.

a, Schematics of the mixed chimera transplantation assay. b, Quantification of CD45.2⁺ cells in the PB of recipient mice at different timepoints after transplantation. c, Quantification of CD45.2⁺ cells in the BM (left) or spleen (right) of recipient mice. d,e, Representative image (d) and quantification (e) of spleen size isolated from recipient mice at 24 weeks after transplantation. f,g, Colony formation analysis of the serial replating assay (f) and flow cytometry analyses of suspension cultures (g) of WT and Tet2^−/− HSPCs with (shMbd6) or without (shNC) Mbd6 KD. h, The correlation between changes in ATAC–seq signal and H2AK119ub signal in response to Mbd6 KD in Tet2-KO HSPCs. i, Integrative Genome Viewer visualization of the H2AK119ub and ATAC signal around the Socs3 or Nfkbia genes in WT, Tet2 KO and shMbd6 after Tet2 KO in HSPCs. The values in parentheses represent the scale of signal in each track. Data are mean ± s.d. (b, c and e–g). P values were calculated using two-tailed unpaired t-tests with Welch’s correction (b, c and e–g) and two-tailed t-distribution with n − 2 d.f. (h). n = 3 biological replicates (b–g). The depicted genome-wide data represent an integration of three biological replicates.

Source Data

Full size image

KD of Mbd6 significantly reduced the replating potential of Tet2-KO HSPCs in vitro (Fig. 4f and Extended Data Fig. 7c). Mbd6 KD disrupted the TET2-loss-induced prolonged maintenance of stem/progenitor cells and promoted differentiation of HSPCs towards myeloid lineages in vitro (Fig. 4g and Extended Data Fig. 7d). Consistently, Mbd6 KD rescued global chromatin opening (Extended Data Fig. 7e–g) and H2AK119ub loss (Extended Data Fig. 7h–j) caused by Tet2 KO, with the increased H2AK119ub negatively correlating with decreased chromatin accessibility in HSPCs (Fig. 4h). Nsun2 KD partially rescued the prolonged maintenance and differentiation phenotypes of Tet2-KO HSPCs (Extended Data Fig. 7k,l). KD of other potential RNA m⁵C writer proteins (Nsun5 and Trdmt1) did not alter these processes (Extended Data Fig. 7k,l). Similarly to that in mES cells, the IAP RNA lifetime was consistently elevated in Tet2-KO HSPCs, and these changes were dependent on the enzymatic activity of TET2 (Extended Data Fig. 7m). We further used a Tet2 mutant that stalls TET2-mediated oxidation at the hm⁵C stage⁴². The IAP half-life profile of HSPCs from this mutant is similar to that of WT HSPCs, suggesting that potential hm⁵C oxidation by TET2 may not further contribute to IAP destabilization.

Increased IAP abundance in Tet2-KO HSPC caRNA was also confirmed (Extended Data Fig. 8a). We proceeded to examine the functional outcomes of targeted IAP RNA m⁵C oxidation using the dCas13–TET2-CD fusion construct in HSPCs (Extended Data Fig. 8b). Targeted IAP m⁵C oxidation also partially rescued the enhanced replating potential of HSPCs by TET2 loss (Extended Data Fig. 8c). The expression level of stem/progenitor markers (Lin⁻KIT⁺; Extended Data Fig. 8d) and the myeloid lineage marker CD11b (Extended Data Fig. 8e) could be restored by targeted oxidation of IAP RNA m⁵C by ectopic expression of a dCas13–TET2-CD fusion protein, but not by its catalytic dead mutant in Tet2-KO HSPCs.

Similarly, steric blockade of m⁵C sites of IAP RNA partially rescued the enhanced replating potential of HSPCs by TET2 loss (Extended Data Fig. 8f,g). IAP blockade disrupted the prolonged maintenance of stem/progenitor cells and enhanced differentiation of HSPCs towards myeloid lineages (CD11b⁺) in vitro (Extended Data Fig. 8h,i).

To reveal the underlying mechanism, we conducted RNA-sequencing (RNA-seq) analysis of the effects of IAP ASO treatment. Genes downregulated, but not upregulated, by IAP ASO treatment exhibited a higher overlap with the genes upregulated by Tet2 KO (Extended Data Fig. 8j,k). Further functional analysis of genes that were upregulated by Tet2 KO, downregulated by IAP ASO treatment, and with nearby m⁵C-marked caRNA revealed enrichments in pathways including osteoclast differentiation and the NOD-like receptor signalling pathway (Extended Data Fig. 8l). Among these, SOCS3 acts as a potent inhibitor of HSPC differentiation⁴³, while Nfkbia encodes a member of the NF-κB inhibitor family, therefore promoting HSPC proliferation^44,45. Correspondingly, the chromatin accessibility at their genomic regions increased in Tet2-KO HSPCs but decreased after Mbd6 KD (Fig. 4i).

After we demonstrated that the m⁵C–TET2–LTR–MBD6 axis is important for HSPC function, we next studied its role in leukaemia fitness. While modest inhibition of proliferation was observed in TET2 WT cell lines, almost complete proliferation blockade was observed for SKM-1 cells, an acute myeloid leukaemia (AML) cell line bearing a TET2 mutation, after MBD6 KD (Extended Data Fig. 9a). To further confirm this synergistic lethal effect, we compared the proliferation of WT, TET2-KO K-562 and TET2-KO THP-1 cells with MBD6 depletion (Extended Data Fig. 9b,c). MBD6 KD markedly attenuated proliferation of TET2-KO cells (Fig. 5a), along with an increased global H2AK119ub levels (Fig. 5b). The attenuated growth of TET2-KO K-562 and THP-1 cells by MBD6 KD could be rescued by MBD6 overexpression (Extended Data Fig. 9d,e). We also observed synergistic inhibition of proliferation when knocking down NSUN2 in TET2-KO K-562 and THP-1 cells (Extended Data Fig. 9f,g).

a, Proliferation of WT or TET2-KO K-562 or THP-1 cells with (shMBD6) or without (shNC) MBD6 KD. n = 4. b, The K-562 H2AK119ub level changes after MBD6 KD. n = 3. c, NSG mice were transplanted with K-562 (left) or THP-1 (right) cells and their overall survival is shown as the Kaplan–Meier estimator. n = 5 mice. d, Heat map illustrating the spike-in-calibrated ATAC–seq signals on ATAC–seq peak regions in WT (siNC WT) and TET2-KO (siNC TET2 KO) K-562 cells, as well as in NSUN2 KD (siNSUN2 TET2 KO) and MBD6-KD (siMBD6 TET2 KO) TET2-KO K-562 cells. n = 3. Data are row-normalized using z scores. e, The correlation of changes in ATAC signals between TET2-KO versus WT K-562 cells, and comparing NSUN2 KD or MBD6 KD with control in TET2-KO K-562 cells. f, m⁵C methylation level changes of K-562 cells in different repeat RNA families after TET2 KO. The size of the circle represents the number of loci methylated by m⁵C. g, The H2AK119ub signals around HERVH-int genomic loci. The colour bar shows H2AK119ub signal. Data are mean ± s.d. P values were calculated using two-tailed unpaired t-tests with Welch’s correction (a and b), log-rank Mentel–Cox tests (c) and two-tailed t-distribution with n − 2 d.f. (e). The depicted genome-wide data represent an integration of three biological replicates.

Source Data

Full size image

To test whether MBD6 loss affects leukemogenesis in vivo, especially in the absence of TET2, we transplanted WT + shNC, TET2 KO + shNC, WT + shMBD6 or TET2-KO + shMBD6 K-562 cells into adult NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice (Extended Data Fig. 9h). Mice receiving shMBD6 cells exhibited substantially decelerated leukemogenesis compared with shNC cells; those that were transplanted with TET2 KO + shMBD6 cells survived significantly longer (125–135 days or 163–220 days, respectively) (Fig. 5c). Consistently, shNC-recipient mice showed markedly higher human chimerism in the BM and PB than the shMBD6-recipient mice (Extended Data Fig. 9i). Similar results were also observed in an in vivo xenotransplantation study with THP-1 cells under the same experimental settings (Fig. 5c). MBD6 KD prolonged survival significantly longer in mice receiving TET2 KO + shMBD6 cells than WT + shMBD6 cells. WT or TET2-KO recipient mice showed substantially higher human CD33⁺CD45⁺ cell chimerism in the BM and PB compared with WT + shMBD6 or TET2 KO + shMBD6 recipient mice (Extended Data Fig. 9j). Thus, MBD6 KD markedly attenuated leukaemic progression in vivo, specifically in the absence of TET2.

MBD6 KD could also suppress transplanted TET2 WT human cells in vivo (Extended Data Fig. 9i,j). We hypothesized that, in TET2 WT cells, MBD6 maintains the expression of pro-proliferating genes, with a larger portion suppressed by TET2. TET2 depletion led to the activation of these genes, with malignant cells becoming addictive to these pathways for proliferation. Supporting our hypothesis, we found that genes that were specifically downregulated by MBD6 KD in TET2-KO cells, rather than in WT controls, were enriched with cell proliferation (Extended Data Fig. 9k,l).

MBD6 KD reversed the excessive caRNA expression after TET2 KO, but not for whole-cell RNA in K-562 cells (Extended Data Fig. 10a,b). Consistent with what we observed in mES cells, MBD6 KD was able to rescue the more-open chromatin state (Extended Data Fig. 10c,d) and decreased H2AK119ub (Extended Data Fig. 10e,f) caused by TET2 KO. The increases in H2AK119ub correlate well with the decreases in ATAC signals (Extended Data Fig. 10g). Examination of chromatin changes caused by NSUN2 KD corroborated the dependence of this regulation on RNA m⁵C (Fig. 5d,e and Extended Data Fig. 10h).

LTR, particularly the ERV1 family with HERVH-int as representative subfamilies, exhibited higher m⁵C methylation changes in TET2-KO K-562 cells (Fig. 5f). MBD6 and NSUN2 also similarly regulate the H2AK119ub levels on HERVH-int elements (Fig. 5g). Finally, we investigated the downstream signalling pathways that are involved in this m⁵C–TET2–MBD6–BAP1 axis in leukaemia cells. We found a notable overlap between the genes upregulated by TET2 KO and those downregulated by MBD6 KD in TET2-depleted K-562 cells (Extended Data Fig. 10i). The enriched terms of these overlapped genes are highly conserved, similar to what we observed in Tet2 KO and IAP ASO treatment in mouse HSPCs (Extended Data Fig. 10j,k).

carRNAs offer a platform for dynamic chromatin regulation. Recent studies have shown that caRNA N⁶-methyladenosine modification has essential roles in global and local chromatin state regulation during mouse early embryo development and in the progression of cancer^{21,22,23,24,25,46}, with the methyltransferase METTL3 functioning as a writer, the binding proteins such as YTHDC1 and RBFOX2 functioning as readers, and FTO functioning as an eraser to reversibly control transcription. We speculated that other RNA modifications may be present on carRNAs and affect chromatin regulation in a similar manner.

TET2 can mediate DNA 5mC oxidation and is an established tumour suppressor for myeloid malignancies. However, TET2 mutations are known to cause global DNA hypomethylation instead of hypermethylation (if it functions as a DNA demethylase), a puzzle that lacks mechanistic explanation. By analysing genomic features associated with hypomethylated or hypermethylated genomic regions, we observed that DNA hypermethylation occurs to enhancers but hypomethylation occurs to repetitive elements, with hypermethylated regions overlapping well with CXXC5-bound genomic loci while hypomethylated regions are enriched in BAP1 binding. We further identified that TET2 mediates RNA m⁵C methylation on carRNA, in particular the LTR repeat RNA, to regulate chromatin state and transcription. A very recent report described caRNA m⁵C oxidation in glioma⁴⁷, but the connection between m⁵C and its oxidation with chromatin state change was not established. We found that MBD6 preferentially recognizes m⁵C on the repeat RNA, which recruits the BAP1 complex to mediate H2AK119ub deubiquitination and gene activation. The m⁵C oxidation by TET2 on these LTR RNAs antagonizes gene activation through the m⁵C–MBD6–BAP1 deubiquitination axis. Loss of TET2 leads to caRNA m⁵C hypermethylation, more-open chromatin and widespread DNA hypomethylation that activates genes critical for leukaemogenesis, explaining the accelerated myeloid malignancy induced by TET2 inactivation (Fig. 4i and Extended Data Fig. 8l).

Our studies also suggest a bimodal function of TET2—when TET2 binds to CXXC4/CXXC5, it mediates DNA 5mC oxidation at the enhancer; however, when recruited by RNA-binding proteins such as PSPC1, TET2 mediates chromatin-associated repeat RNA m⁵C oxidation; this RNA m⁵C oxidation activity by TET2 dictates the global chromatin regulation in mES cells, HPSCs and leukaemia cells. This study therefore reveals a NSUN2–TET2–MBD6–BAP1 axis in chromatin and transcription regulation through repeat RNA m⁵C. Practically, it provides targets for future targeted therapies against TET2 mutant malignancies.

Animals and tissues

Tet2^−/− mice were generated as described⁴¹. These mice used in this study were backcrossed for more than six generations with C57BL/6 mice. WT C57BL/6 and Tet2^−/− mice (aged 6–8 weeks), including both male and female, were used throughout this study and maintained under standard laboratory housing conditions with food and water ad libitum. All the mice were randomly assigned to experimental groups and data analyses were blindly performed by two lab members independently. All animal studies were performed with the approval from the Institutional Animal Care and Use Committee (IACUC), protocol number 30979/20190086AR at The University of Texas Health Science Center at San Antonio (UTHSCSA) and conducted in accordance with the institutional and national guidelines and regulations.

Xenotransplantation of human leukaemia cells

For in vivo xenotransplantation study procedures, 1 × 10⁶ K-562 cells were injected intravenously via the tail vein into adult NSG mice (aged 6–8 weeks) pretreated with 250 cGy whole body irradiation. At 28–39 days after transplantation, PB was collected from the submandibular vein, and the BM was isolated from the tibias and femurs. Human CD33⁺ chimerism in BM and PB cells were analysed by BD FACSCelesta flow cytometer (BD Biosciences).

2 × 10⁴ THP-1 cells were injected intravenously through the tail vein into adult NSG mice (6–8 weeks old) pretreated with 250 cGy whole-body irradiation. At 20–22 days after transplantation, human CD33⁺CD45⁺ chimerism in BM and PB cells were analysed using the BD FACSCelesta flow cytometer.

A cohort of mice from each transplantation group was monitored until they became moribund or died.

Competitive repopulation assay

The competitive repopulation assay was performed to assess the effect of TET2 and/or MBD6 KD on the repopulating potential of HSPCs in vivo. In total, 2 × 10⁴ Lin⁻KIT⁺ cells isolated from the BM cells of 8-week-old WT or Tet2-KO mice (CD45.2⁺) were lentivirally transduced with short hairpin RNA (shRNA) plasmid pLKO.1-shC002 (MilliporeSigma, SHC002: shNC) or pLKO.1-shMbd6 (Millipore-Sigma, TRCN0000178563) and incubated in suspension culture containing 20% FBS in complete RPMI-1640 medium supplemented with 100 ng ml⁻¹ mSCF, 10 ng ml⁻¹ mIL-3, 10 ng ml⁻¹ IL-6 and 20 ng ml⁻¹ mFlt3. Then, 48 h after transduction, Lin⁻KIT⁺ cells from each transduction were transplanted along with 1 × 10⁶ 8-week-old BoyJ (CD45.1⁺) BM competitor cells into lethally irradiated (800 cGy) BoyJ recipients through the tail-vein injection. The CD45.2/CD45.1 chimeras in the PB were monitored monthly for 6 months. Recipients were euthanized 6 months after transplantation to analyse the CD45.2/CD45.1 chimeras in the BM and spleen.

Haematopoietic stem and progenitor cell sorting, colony assay and in vitro differentiation assay

For haematopoietic stem and progenitor Lin⁻KIT⁺ cell selection, magnetic-activated cell sorting was applied with autoMACS Pro Separator (Miltenyi Biotec). In brief, the lineage-positive cells (Lin⁺) were depleted from total BM cells of 6–8-week-old mice using the Direct Lineage Cell Depletion Kit (Miltenyi Biotec, 130-110-470), and the Lin⁻ cells were then sorted with KIT (CD117) MicroBeads (Miltenyi Biotec, 130-091-224). The purity of selected cells was analysed by flow cytometry.

For colony assay, HSPCs were plated in triplicate in methylcellulose medium (MethoCult, M3134) supplemented with mouse stem cell factor (mSCF; 100 ng ml⁻¹), interleukin-3 (mIL-3; 10 ng ml⁻¹), thrombopoietin (mTPO; 50 ng ml⁻¹), granulocyte-macrophage colony-stimulating factor (mGM-CSF; 10 ng ml⁻¹), human erythropoietin (hEPO; 4 U ml⁻¹) and interleukin-6 (hIL-6; 50 ng ml⁻¹, PeproTech). The colonies were imaged using STEMvision (StemCell Technologies) and scored on day 7, and these colonies were then sequentially replated every 7 days for replating assay. Colony cells were also collected and analysed for expression of stem and progenitor markers and myeloid linage markers by flow cytometry.

The HSPCs were also incubated in suspension culture containing 30% FBS and 2% BSA in complete RPMI-1640 medium supplemented with 100 ng ml⁻¹ mSCF, 10 ng ml⁻¹ mIL-3, 50 ng ml⁻¹ mTPO and 10 ng ml⁻¹ mGM-CSF. Cells were collected and analysed for expression of stem/progenitor markers at day 7 and myeloid lineage markers at day 14 by flow cytometry.

Flow cytometry analysis

Cells were stained with PerCP-Cy5.5 mouse lineage antibody cocktail (BD Biosciences, 561317) and PE rat anti-mouse CD117 (BD Biosciences, 553869) antibody for haematopoietic stem and progenitor cells analysis. Brilliant Violet 421 (BV421) anti-mouse/human CD11b (Mac-1) (BioLegend, 101236) was used to analyse myeloid lineage. PerCP-Cy5.5 mouse anti-mouse CD45.2 (BD Biosciences, 552950) and FITC mouse anti-mouse CD45.1 (BD Biosciences, 553775) antibodies were used for analysing CD45.2/CD45.1 chimeras in a competitive repopulation assay.

Human CD33 chimerism was analysed with PE mouse anti-human CD33 (BD Biosciences, 561816) and PE-Cy7 rat anti-mouse CD45 (BD Biosciences, 552848) in PB and BM cells from NSG mice that were xenotransplanted with K-562 cells. Human CD33/CD45 chimerism was analysed with PE mouse anti-human CD33 (BD Biosciences, 561816) and APC mouse anti-human CD45 (BD Biosciences, 555485) in PB and BM cells from NSG mice that were xenotransplanted with THP-1 cells. All flow cytometry data were analysed using FlowJo-V10 software (TreeStar). Examples of the gating strategies are provided in Supplementary Figs. 2 and 3.

Cell culture

WT and Tet2^−/− mES cells were gifts from the B. Ren laboratory^26,48. The control and KO mES cells have been shown to be pluripotent by chimera formation assay. All mES cells were kept in DMEM (Gibco, 11995065) supplemented with 15% heat-inactivated stem-cell-qualified fetal bovine serum (Gemini Bio Products, 100-525), 1× l-glutamine (Gibco, 25030081), NEAA (Gibco, 25030081), LIF (Millipore-Sigma, ESG1107), 1× β-mercaptoethanol (Gibco, 21985023), 3 μM CHIR99021 (StemCell Technologies, 72052) and 1 μM PD0325901 (StemCell Technologies, 72182) at 37 °C and 5% CO₂. For stable TET2 overexpression mES cells, empty vector, WT Tet2 or Tet2 HxD mutant bearing piggyBac plasmids were constructed and transfected into Tet2-KO or Pspc1-KO mES cells using Lipofectamine 3000 Transfection Reagent (Invitrogen, L3000001) according to the standard protocol. Stable expression clone selection was performed using 0.1 mg ml⁻¹ hygromycin B (Gibco, 10687-010) for 2 weeks. The medium was replaced every 24 h. ES cells were passaged on gelatin-coated plates twice to clear feeder cells before experiments.

WT THP-1, K-562 and TF-1 cells were obtained from the American Type Culture Collection (ATCC). The SKM-1 cell line was obtained from DSMZ (German Collection of Microorganisms and Cell Cultures). WT OCI-AML3 cell was a gift from L. Godley. WT and TET2^−/− K-562 and THP-1 cells were gifts from B. K. Jha as previously generated⁴⁹. THP-1, K-562, SKM-1 and OCI-AML3 cells were kept in RPMI-1640 (Gibco, 61870036) with 10% fetal bovine serum (FBS, Gibco 26140079) at 37 °C under 5% CO₂. TF-1 was kept in RPMI-1640 (Gibco, 61870036) with 10% FBS (Gibco 26140079) and 2 ng ml⁻¹ recombinant GM-CSF (Peprotech, 300-03) at 37 °C under 5% CO₂. U-87 MG (HTB-14), LN-229 (CRL-2611), Hep G2 (HB-8065), HeLa (CCL-2), HCT 116 (CCL-247), A549 (CCL-185) and A-375 (CRL-1619) cells were obtained from the American Type Culture Collection (ATCC). U-87 MG and LN-229 were kept in ATCC-formulated Eagle’s minimum essential medium (ATCC, 30-2003) supplemented with 10% FBS (Gibco, 26140079) and 5% FBS (Gibco, 26140079), respectively. Hep G2, HeLa, HCT 116, A549 and A-375 cells were kept in DMEM (Gibco, 11995065) supplemented with 10% FBS (Gibco, 26140079). All cell types were kept at 37 °C and 5% CO₂.

shNC and shMBD6 THP-1 and K-562 cell lines were constructed by lentivirus transduction with TransDux MAX Lentivirus Transduction Reagent (System Biosciences, LV860A-1). Lentiviral particles were prepared by using HEK293T cells and lentiviral packaging plasmids pCMV-VSV-G and pCMV-dR8.2 (pCMV-VSV-G and pCMV-dR8.2 were gifts from B. Weinberg (Addgene plasmid, 8454; and Addgene plasmid, 8455)) and shRNA plasmid pLKO.1-shC002 (Millipore-Sigma, SHC002) or pLKO.1-shMBD6 (Millipore-Sigma, TRCN000038787). Then, 48 h after transfection, lentiviral particles were precipitated using the PEG-it Virus Precipitation Solution (System Biosciences, LV810-1). shNC and shMBD6 THP-1 and K-562 cell lines were kept in RPMI-1640 (Gibco, 61870036) with 10% fetal bovine serum (FBS, Gibco) and 1 μg ml⁻¹ puromycin (Gibco, A1113803) at 37 °C under 5% CO₂. Small interfering RNA (siRNA) or gene overexpression plasmids transfection in K-562 and THP-1 cells were performed according to the manufacturer’s instructions for SF Cell Line 4D-Nucleofector X Kit (Lonza Biosciences, V4XC-2032, FF-120 for K-562) or SG Cell Line 4D-Nucleofector X Kit (Lonza Biosciences, V4XC-3024, FF-100 for THP-1)

TET2-KO THP-1 cell line for PDX model was generated using CRISPR–Cas9 system. Single-guide RNAs were designed using the CRISPick tool (https://portals.broadinstitute.org/gppx/crispick/public) and then cloned into LentiCRISPR V2-GFP vector by Synbio Technologies. THP-1 cells were infected by lentiviral particles for 72 h and followed by GFP-positive cell selection using the BD FACSMelody Cell Sorter (BD Biosciences). KO efficiency was verified by western blotting.

shNC, shMBD6 (Millipore-Sigma, TRCN0000178563), shNsun2 (Millipore-Sigma, TRCN0000325347), shNsun5 (Millipore-Sigma, TRCN0000097512) or shTrdmt1 (Millipore-Sigma, TRCN0000328293) Lin⁻KIT⁺ HSPCs were constructed by electroporation with the P3 Primary Cell 4D-Nucleofector X Kit S (Lonza Bioscience, V4XP-3032) by program CV-137.

siRNA and plasmid transfection

Two or three individual siRNAs, or a pool of four siRNAs targeting different regions of the same transcript (Dharmacon siRNA) were used for KD of human or mouse transcripts. siRNA transfections in mES cells and other adherent cell lines were performed using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen, 13778075) according to the manufacturer’s instructions. Transfections in human leukaemia cells (THP-1, TF-1, OCI-AML3, SKM-1) were performed by electroporation using the SG Cell Line 4D-Nucleofector X Kit L (Lonza Bioscience, V4XC-3024) with program FF-100. Transfections in K-562 cells were performed with the SF Cell Line 4D-Nucleofector X Kit L (Lonza Bioscience, V4XC-2012) with program FF-120.

Plasmid transfections in mES cells or HEK293T cells were performed using the Lipofectamine 3000 Transfection Reagent (Invitrogen, L3000015) according to the manufacturer’s instructions.

Cell proliferation assay

The cell proliferation assays for adherent and suspension cells were performed similarly. Cells were seeded into 96-well plates before assaying in 100 μl settings with CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega, G3582) according to the manufacturer’s instructions. Then, 2,000–10,000 cells were seeded per well at day 0 and the cell proliferation was monitored every 24 h by incubating the cell suspension with MTS reagent at 37 °C for 1 h.

DNase I–TUNEL assay

For cell line samples, mES cells were reseeded to 10 cm cell culture dishes 12 h before siRNA transfection. The DNase I–TUNEL assay was performed using DeadEnd Fluorometric TUNEL System (Promega, G3250) according to the manufacturer’s instructions after cell fixation with paraformaldehyde and permeabilization with Triton X-100. Two independent experiments were performed. Cells were treated with 1 U ml⁻¹ of DNase I (Thermo Fisher Scientific, EN0521) for 5 min at 37 °C before rTdT labelling. Flow cytometry was performed on a BD Fortessa (BD Biosciences), and data were analysed using Flowjo (TreeStar).

Nascent RNA imaging assay

mES cells were reseeded in Nunc Lab-Tek II Chambered Coverglass (Thermo Fisher Scientific, 155409) 12 h before treatment. The nascent RNA synthesis assay was performed using Click-iT RNA Alexa Fluor 488 Imaging Kit (Invitrogen, C10329) according to the manufacturer’s instructions. 5-Ethynyl uridine incubation was performed for 1 h before washing away by cell medium. Cell nucleus was counterstained with Hoechst 33342 (Abcam, ab228551). The samples were imaged on a Leica SP8 laser scanning confocal microscope at University of Chicago. The fluorescence intensity across different samples were quantified with Cellprofiler v.3.0 with a custom workflow. The total RNA synthesis rate was obtained by multiplying the average intensity in each cell by the area of each cell.

ATAC–see analysis

Assay of transposase-accessible chromatin with visualization (ATAC–see) of mES cells was performed as described in the original report⁵⁰. ATTO-590-labelled imaging oligos were purchased from Integrated DNA Technologies (IDT) and the oligonucleotide sequences are as follows: Tn5MErev, 5′-[phos]CTGTCTCTTATACACATCT-3′; Tn5ME-A-ATTO590, 5′-/5ATTO590/TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG-3′; Tn5ME-B-ATTO590: 5′-/ATTO590/GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG-3′. The oligos were assembled with recombinant Tn5 transposase (Active motif, 81286) to produce the Tn5 transposome. Cell fixation, permeabilization and labelling were performed as described in the original report⁵⁰.

Recombinant protein purification

Standard molecular cloning strategies were used to generate C-terminally MBP–6×His-tagged MBD ___domain of MBD6 (residues 1–100). The human MBD6 coding sequence was obtained from Origene (Origene, SC324058). The full-length coding sequence was cloned using PrimeSTAR GXL DNA Polymerase (TaKaRa Bio, R050B). Recombinant proteins were expressed in E. coli BL21 (DE3) grown to an optical density at 600 nm of 0.6 in LB medium. The expression was induced with 0.6 mM IPTG at 16 °C for 20 h and cells were collected by centrifugation.

For purification of MBP tagged MBD ___domain of MBD6, bacterial pellet was resuspended in a lysis buffer containing 25 mM Tris-HCl (pH 7.5), 500 mM NaCl, 20 mM imidazole, 10 mM β-mercaptoethanol (β-ME) and protease inhibitors (ethylenediaminetetraacetic-acid-free protease inhibitor cocktail tablet, Millipore-Sigma 4693132001) and disrupted by sonication for 3 min. The cell lysates were clarified by centrifugation at 26,000g for 30 min and the supernatant was applied to Ni²⁺-NTA resin (Thermo Fisher Scientific, 88221) and washed with lysis buffer, and the bound proteins were eluted with lysis buffer supplemented with 250 mM imidazole. The eluted protein was bound back to amylose resin (NEB, E8021S) before washing with lysis buffer. The bound protein was eluted with 1% maltose in lysis buffer. The eluted protein was analysed by SDS–PAGE and concentrated by centrifugal filtration (Amicon Ultra-15). Final concentrated protein was aliquoted, flash-frozen and stored at −80 °C for future use.

RT–qPCR

To quantify expression levels of transcripts, total RNA was reverse transcribed using the PrimeScript RT Master Mix (TaKaRa Bio, RR0361) with oligo dT primer and random hexamers as primers. The cDNA was then subjected to quantitative PCR (qPCR; LightCycler 96 system, Roche) using FastStart Essential DNA Green Master (Roche, 06402712001) with gene-specific primers. The relative changes in expression were calculated using the ΔΔC_t method.

Western blot analysis

Protein samples were prepared from respective cells by lysis in RIPA buffer (Thermo Fisher Scientific, 89900) containing 1× Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific 78441). Protein concentration was measured by NanoDrop 8000 Spectrophotometer (Thermo Fisher Scientific). Lysates of equal total protein concentration were heated at 90 °C in 1× loading buffer (Bio-Rad, 1610747) for 10 min. Denatured protein was loaded into 4–12% NuPAGE Bis-Tris gels (Invitrogen, NP0335BOX) and transferred to PVDF membranes (Thermo Fisher Scientific, 88585). Membranes were blocked in Tris-buffered saline, 0.1% Tween-20 (TBST) with 3% BSA (Millipore-Sigma, A7030) for 30 min at room temperature, incubated in a diluted primary antibody solution at 4 °C overnight, then washed and incubated in a dilution of secondary antibody conjugated to HRP for 1 h at room temperature. Protein bands were detected using SuperSignal West Dura Extended Duration Substrate kit (Thermo Fisher Scientific, 34075) with a FluroChem R (Proteinsimple). Blot intensities were quantified with Fiji (ImageJ) Analyse-Gel module. Uncropped gels with size marker indications are provided in Supplementary Fig. 1.

Dot blot

Oligonucleotide probes end-labelled with Alexa Fluor 488 dye was spotted on a positively charged Nylon membrane (Roche, 11209299001). The membrane was dried at room temperature for 5 min before UV cross-linking at 254 nm with a Stratalinker (Stratagene) for two times to achieve a 4,500 J m⁻² UV flux. The membrane was then blocked in Tris-buffered saline, 0.1% Tween-20 (TBST) with 3% BSA (Millipore-Sigma, A7030) for 30 min at room temperature. Primary antibodies were diluted according to the manufacturer’s instructions and incubated with the membrane for 60 min at room temperature. The membrane was washed and incubated in a dilution of secondary antibody conjugated to HRP for 60 min at room temperature. The final membrane was detected using SuperSignal West Dura Extended Duration Substrate kit (Thermo Fisher Scientific, 34075) with the iBright 1500 system (Invitrogen, A44241).

Cell fractionation

Fractionation of mES cells, K-562 or THP-1 cells was performed according to the published protocol⁵¹ with the optimized concentration of NP-40 (MilliporeSigma, 492018) for each cell line. In brief, 5 × 10⁶ to 1 × 10⁷ cells were collected and washed with 1 ml cold PBS/1 mM EDTA buffer, then centrifuged at 4 °C and 500g to collect the cell pellet. Then, 200 μl ice-cold lysis buffer (10 mM Tris-HCl, pH 7.4, 0.05% NP-40, 150 mM NaCl) were added to the cell pellet and incubated on ice for 5 min, then gently pipetted up the cell lysate over 2.5 volumes of chilled sucrose cushion (24% RNase-free sucrose in lysis buffer) and centrifuged at 4 °C and 15,000g for 10 min. All the supernatant was collected as cytoplasmic fraction and the nuclei pellet was washed once by gently adding 200 μl ice-cold PBS/1 mM EDTA to the nuclei pellet without dislodging the pellet. The nuclei pellet was resuspended in 200 μl prechilled glycerol buffer (20 mM Tris-HCl, pH 7.4, 75 mM NaCl, 0.5 mM EDTA, 0.85 mM DTT, 0.125 mM PMSF, 50% glycerol) with gentle flicking of the tube. An equal volume of cold nucleus lysis buffer (10 mM HEPES, pH 7.6, 1 mM DTT, 7.5 mM MgCl₂, 0.2 mM EDTA, 0.3 M NaCl, 1 M urea, 1% NP-40) was then added, followed by vigorous vertexing for 5 s twice. The nuclei pellet mixtures were incubated for 2 min on ice, then centrifuged at 4 °C and 15,000g for 2 min. The supernatant was collected as the soluble nuclear fraction (nucleoplasm). The pellet was gently rinsed with cold PBS/1 mM EDTA without dislodging and was then collected as the chromosome-associated fraction.

Fractionation of HSPCs was performed similar to ES cells with minor modifications. In brief, HSPCs were cultured in vitro for 2 h after sorting on the autoMACS Pro Separator, and then ice-cold lysis buffer (10 mM Tris-HCl, pH 7.4, 0.15% IGEPAL CA-630, 75 mM NaCl) was used to separate the cytoplasmic fraction. The procedures for isolating the nuclear fraction and chromosome-associated fraction were the same as that of ES cells.

Quantitative analysis of modified base levels using UHPLC–MS/MS

The nucleic acid digestion step for RNA was as follows: 75 ng ribo-depleted RNA was digested by nuclease P1 (MilliporeSigma, N8630) in 20 μl buffer containing 20 mM ammonium acetate at pH 5.3 for 2 h at 42 °C. Then, 1 U of FastAP thermosensitive alkaline phosphatase (Thermo Fisher Scientific, EF0651) was added to the reaction and FastAP buffer was added to a 1× final concentration before incubation for 2 h at 37 °C. For DNA, genomic DNA was purified from cells according to the standard protocol of the Monarch Genomic DNA Purification Kit (NEB, T3010S). An additional RNase A (Thermo Fisher Scientific, EN0531) digestion step was performed on the purified DNA and the reaction was recovered with DNA Clean & Concentrator-5 (Zymo Research, D4014). Then, 200 ng DNA was digested with Nucleoside Digestion Mix (NEB, M0649S) at 37 °C for 2 h.

The samples were diluted and filtered (0.22 μm, Millipore) and injected into a C18 reversed-phase column coupled online to the Agilent 6460 LC–MS/MS spectrometer in positive electrospray ionization mode. The nucleosides were quantified using retention time and the nucleoside to base ion mass transitions (for RNA: 268 to 136 for A; 284 to 152 for G; 258 to 126 for m⁵C and 274 to 142 for hm⁵C; for DNA: 228 to 112 for dC, 242 to 126 for 5mdC, 258 to 142 for 5hmdC). Quantification was performed by comparing with the standard curve obtained from pure nucleoside standards running with the same batch of samples.

Chromatin-associated RNA-seq

Chromatin-associated RNA-seq analyses of mES cells, K-562 and HSPCs were performed similarly. After caRNA isolation, ERCC RNA spike-in mix (Invitrogen, 4456740) was added to purified total caRNA according to the ratio recommended by the standard protocol. Ribosomal RNA was depleted from isolated chromatin-associated RNA with RiboMinus Eukaryote System v2 (Invitrogen, A15026) followed by size-selection using the standard protocol of RNA Clean & Concentrator-5 (RCC-5, Zymo Research, R1013). RNA libraries were constructed with SMARTer Stranded Total RNA-Seq Kit v2 - Pico Input Mammalian (TaKaRa Bio, 634411) according to the manufacturer’s instructions. Three replicates were performed for each condition. Libraries were sequenced on the NovaSeq 6000 sequencer.

ATAC–seq analysis

ATAC–seq was performed using the ATAC–seq kit (Active Motif, 53150) according to the manufacturer’s instructions. In brief, 50,000 to 100,000 cells were aliquoted for each replicate and mixed with equal amounts of Drosophila spike-in (Active Motif, 53154). Cells were then permeabilized with buffer containing 0.1% Tween-20 and 0.01% Digitonin, both supplied by the original kit. Accessible chromatin regions were tagged with pre-assembled Tn5 transposome. Tagged genomic DNA was extracted from cells and DNA libraries were obtained by PCR amplification. Pooled libraries were sequenced on the NovaSeq 6000 sequencer. For ATAC–qPCR, tagged genomic DNA was extracted and amplified by PCR for 8 cycles using the indexing primers from the original kit. Amplified DNAs were subjected to qPCR analysis using individual primer sets.

m⁵C methylated RNA immunoprecipitation with spike-in

m⁵C modified or unmodified mRNA spike-ins were in vitro transcribed from firefly luciferase or Renilla luciferase coding sequences with mMESSAGE mMACHINE T7 Transcription Kit (Invitrogen, AM1344) and manually reconstituted dNTP mixes with 20% m⁵CTP/CTP ratio. 5-methylcytidine-5-triphosphate was obtained from TriLink Biotechnologies (N-101405). Yielded RNA was purified by using the standard protocol of RNA Clean & Concentrator-5 (Zymo Research, R1013). The spike-in RNA mixes were then applied to RNA before fragmentation.

Total RNAs from whole cell or the chromatin-associated fractions were randomly fragmented by incubation at 94 °C for 4 min using 1× fragmentation buffer (NEB, E6186A). Fragmentation was stopped by adding 1× stop solution. Spike-in RNAs were added to each sample. Then, 4 μg anti-m⁵C antibody (Diagenode, MAb-081-100) was conjugated with 30 μl of protein G beads (Invitrogen, 1003D) in 300 μl IP buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.05% Triton X-100 (v/v), 1 mM spermidine) for 2 h at 4 °C on a rotating wheel. The same procedure was performed for a control reaction using mouse IgG isotype control (Abcam, ab37355). Bead–antibody complexes were washed three times with IP buffer and finally brought to 250 μl with IP buffer. After heat denaturation and quick chill on ice, 10 μg samples of RNA were added to the bead–antibody complexes and incubated with 1 μl SUPERase•In RNase Inhibitor (Invitrogen, AM2694) overnight at 4 °C on a rotating wheel. After several washes with IP buffer, RNA was incubated in 100 μl elution buffer (5 mM Tris-HCl pH 7.5, 1 mM EDTA, 0.05% SDS, and 200 μg proteinase K (Invitrogen, 25530049)) for 1 h at 50 °C. Beads were removed by centrifugation in a microcentrifuge, and the supernatant was purified with RCC-5 without size selection. Immunoprecipitated RNAs were eluted in water and then analysed using RT–qPCR. For next-generation sequencing, the immunoprecipitated RNAs were used as inputs for library constructions with the SMARTer Stranded Total RNA-Seq Kit v2—Pico Input Mammalian (TaKaRa Bio, 634411) according to the manufacturer’s instructions. Libraries were sequenced on the NovaSeq 6000 sequencer.

For analysing the effects of GC ratio and m⁵C modification levels, we designed three different in vitro transcription templates to get 70%, 50% or 30% GC ratio RNA products based on firefly luciferase mRNA (Supplementary Table 2). DNA oligos were purchased from Integrative DNA Technologies and annealed with a complementary DNA oligo (T7; Supplementary Table 2) to enable T7 DNA polymerase binding. In vitro transcription was performed using the mMESSAGE mMACHINE T7 Transcription Kit (Invitrogen, AM1344) and manually reconstituted dNTP mixes with a 0%, 0.2%, 2% or 20% m⁵CTP/CTP ratio. 5-methylcytidine-5-triphosphate was obtained from TriLink Biotechnologies (N-101405). Yielded RNA was purified using the standard protocol of the RNA Clean & Concentrator-5 (Zymo Research, R1013) kit. meRIP–qPCR experiments were performed according to the protocol mentioned above, and yeast tRNA (Invitrogen, AM7119) was mixed with RNA probes as a carrier.

RNA amplicon bisulfite sequencing

caRNAs were isolated from Tet2 WT or Tet2-KO mES cells as aforementioned. Ultrafast bisulfite (UBS) conversion was performed according to the published protocol²⁸. Reverse transcription was then performed with SuperScript III Reverse Transcriptase (Invitrogen, 18080093) using individual RT primers (Supplementary Table 2). The resulting cDNA was amplified for 10 cycles using NEBNext Ultra II Q5 Master Mix (NEB, M0544S) according to the standard protocol except that the T_m was set to 50 °C. Amplified DNA was quantified using the universal p5 primer (Supplementary Table 2) and p7 primer from NEBNext Multiplex Oligos for Illumina (NEB, E7500S). cDNAs amplified from different amplicons were then pooled together based on qPCR quantifications to achieve equal sequencing depth in the final DNA library. A final amplification was performed using the two primers (universal p5 primer and p7 primer from NEBNext Multiplex Oligos for Illumina) for 15 cycles using NEBNext Ultra II Q5 Master Mix (NEB, M0544S). PCR products were recovered using 1.0 volume of AMPure XP beads (Beckman Coulter, A63882) and subjected to sequencing on a NovaSEQ-X sequencer.

meDIP analysis

For methyl-DNA immunoprecipitation (meDIP) analysis, genomic DNA was extracted from cultured cells using the Monarch Genomic DNA Purification Kit (New England Biolabs, T3010S). Unmethylated lambda DNA (Promega, D1521) was spiked at a 0.5% ratio for quality control of the immunoprecipitation. DNAs were then fragmented to 200–1,000 bp by incubation for 22 min with NEBNext dsDNA Fragmentase (New England Biolabs, M0348S). The fragmented DNA was then denatured at 95 °C for 5 min and immediately cooled on ice for another 5 min. The input samples were removed and saved on ice for later use. The reaction was conducted in IP buffer (150 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40) at 4 °C overnight. The beads were then washed three times with IP buffer, followed by three washes by high-salt wash buffer (500 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40). Immunoprecipitated DNA was extracted by proteinase K digestion (Invitrogen, 25530049) before qPCR analysis. High-throughput sequencing libraries were constructed using xGen Methyl-Seq Lib Prep kits (IDT, 10009860) and sequenced on the NovaSEQ-X sequencer.

RNA synthesis rate assay

The RNA synthesis rate was measured with a procedure modified from the protocol Click-iT Nascent RNA Capture Kit, for gene expression analysis (Invitrogen, C10365). mES cells were seeded to 6 cm dishes at the same density in three replicates. After 42 h, cells were treated with 1 mM 5-ethynyl uridine for 10 min, 20 min and 40 min before RNA collection using TRIzol Reagent (Invitrogen, 15596026). Ribosomal RNA was depleted from total RNA preps before the click reaction with biotin azide (PEG4 carboxamide-6-azidohexanyl biotin). Biotinylated RNA was enriched using Dynabeads MyOne Streptavidin T1 (Invitrogen, 65601). ERCC RNA spike-in mix (Invitrogen, 4456740) was added to the eluted RNA with the amount proportional to the total RNA of each sample before rRNA depletion. Spiked RNAs were used as an input for RNA-seq library construction using the SMARTer Stranded Total RNA-Seq Kit v2—Pico Input Mammalian (TaKaRa Bio, 634411) according to the manufacturer’s instructions. Libraries were sequenced on the NovaSeq 6000 sequencer.

CUT&Tag analysis

Cleavage under targets and tagmentation (CUT&Tag) analysis was performed using the CUT&Tag-IT Assay Kit (Active motif, 53160) according to the manufacturer’s instructions. In brief, 0.2 million cells were used as an input for one replicate and washed with 1× wash buffer. Washed cells were conjugated to concanavalin A beads and permeabilized with Digitonin-containing buffer before incubation with primary antibodies (anti-H3K27me3, anti-H2AK119ub or normal rabbit IgG). Preassembled protein A-Tn5 transposome-enabled DNA tagmentation was performed after secondary antibody conjugation. Equal amounts of Drosophila spike-in chromatin preps (Active Motif, 53083) were added to each samples and subjected to the Tn5 tagmentation reaction. Tagged DNA was extracted by proteinase K digestion and amplified by PCR with indexed primers to yield DNA libraries. DNA libraries were subjected to qPCR analysis with gene-specific primers or high-throughput sequencing on the NovaSeq 6000 sequencer.

Construction of induced tethering mES cell lines

Cell lines stably expressing dCas13 protein fusion with catalytic ___domain of mouse TET2 (TET2-CD) or catalytic dead mutants were constructed first from WT mES cells. The coding sequence of dCas13 was cloned from plasmid pCMV-dCas13-M3nls, which was a gift from D. Liu (Addgene plasmid, 155366). The coding sequence of TET2-CD was cloned from the plasmid pcDNA3-FLAG-mTET2 (CD), which was a gift from Y. Xiong (Addgene plasmid, 89736), and the catalytic-dead mutant was cloned from the plasmid pcDNA3-Flag-Tet2 CD Mut, which was a gift from Y. Zhang (Addgene plasmid, 72220). pLR5-CBh-dCas9-hEzh2-IRES-Hyg was a gift from H. Ochiai (Addgene plasmid, 122375). The coding sequences of TET2-CD (or mutant) and dCas13 or dCas9 were fused. The fusion protein was delivered to mES cells with the piggyBac transposon system using the pLR5 vector and selected with hygromycin B (Gibco, 10687010). Sequences expressing guide RNA for dCas13 were cloned into a plasmid expressing a Tet operator controlled H1 operator (H1-2O2)⁵². This tet-pLKO-sgRNA-puro plasmid was a gift from N. Gray (Addgene plasmid, 104321). The guide-RNA expression plasmid was delivered into the TET2-CD-fusion protein-expressing mES cells by lentivirus. The resulting cell lines were selected with puromycin (Gibco, A1113803).

ASO and plasmid transfection in HSPCs

The steric-blocking antisense oligonucleotides (ASOs) (Integrated DNA Technologies) targeted to the hypermethylated motifs were fully modified with 2′-O-methoxyethyl (2′MOE) bases and phosphorothioate bonds, which were also incorporated with a fluorescent dye Cy5 at the 3′ end to monitor transfection efficiency. The NC5 ASO was used as a negative control that was not targeted to the human or mouse genome.

IAPEz-int 2′MOE: AGTTGAATCCTTCTTAACAGTCTGCTTTACGGGAAC

Sequence: /52MOErA/*/i2MOErG/*/i2MOErT/*/i2MOErT/*/i2MOErG/*/i2MOErA/*/i2MOErA/*/i2MOErT/*/i2MOErC/*/i2MOErC/*/i2MOErT/*/i2MOErT/*/i2MOErC/*/i2MOErT/*/i2MOErT/*/i2MOErA/*/i2MOErA/*/i2MOErC/*/i2MOErA/*/i2MOErG/*/i2MOErT/*/i2MOErC/*/i2MOErT/*/i2MOErG/*/i2MOErC/*/i2MOErT/*/i2MOErT/*/i2MOErT/*/i2MOErA/*/i2MOErC/*/i2MOErG/*/i2MOErG/*/i2MOErG/*/i2MOErA/*/i2MOErA/*/i2MOErC//3Cy5Sp/

MERVL 2′MOE: ACCATTACTGGGTATGTTAT

Sequence: /52MOErA/*/i2MOErC/*/i2MOErC/*/i2MOErA/*/i2MOErT/*/i2MOErT/*/i2MOErA/*/i2MOErC/*/i2MOErT/*/i2MOErG/*/i2MOErG/*/i2MOErG/*/i2MOErT/*/i2MOErA/*/i2MOErT/*/i2MOErG/*/i2MOErT/*/i2MOErT/* /i2MOErA/*/i2MOErT//3Cy5Sp/

NC5 2′MOE: GCGACTATACGCGCAATATG

Sequence: /52MOErG/*/i2MOErC/*/i2MOErG/*/i2MOErA/*/i2MOErC/*/i2MOErT/*/i2MOErA/*/i2MOErT/*/i2MOErA/*/i2MOErC/*/i2MOErG/*/i2MOErC/*/i2MOErG/*/i2MOErC/*/i2MOErA/*/i2MOErA/*/i2MOErT/*/i2MOErA/* /i2MOErT/*/i2MOErG//3Cy5Sp/

The crRNA targeting the primary m⁵C sites on IAPEz sequence based on our RNA bisulfite sequencing results was custom-synthesized and cloned into the pLentiRNAGuide_002-hU6-RfxCas13d-DR-BsmBI-EFS-EGFP:P2A:Puro-WPRE vector. The catalytic ___domain of mouse TET2 (mTET2-CD) or a catalytically dead mutant TET2(H1304Y/D1306A) (mTET2CDHxDCD) was cloned into the pLV[Exp]-[EF-1sc>[NLS-RfxCas13d]:[Linker]:P2A:mCherry(ns):T2A:Bsd vector. All of these plasmids were synthesized, constructed and confirmed by VectorBuilder.

All of the ASOs and plasmids were transfected into HSPCs using electroporation with the P3 Primary Cell 4D-Nucleofector X Kit S (Lonza Bioscience, V4XP-3032) with the program CV-137.

ASO transfections

We designed ASOs targeting the primary m⁵C sites on IAPEz or MERVL sequences based on our RNA m⁵C sequencing results. ASO transfections in mES cells were performed using the Lipofectamine RNAiMAX Transfection Reagent (Invitrogen, 13778075) according to the manufacturer’s instructions.

Cross-linking and immunoprecipitation and PAR-CLIP

Cultured mES cells or human leukaemia cells (SKM-1, WT and TET2^−/− THP-1 and K-562) were UV cross-linked at 254 nm with a Stratalinker (Stratagene) twice to achieve a 4,500 J m⁻² UV flux and then flash-frozen in liquid nitrogen. For photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP), 4-thiouridine was added to the cell culture medium 14 h before UVA irradiation (365 nm) three times, 1,500 J m⁻² each. The pellets were thawed on ice and resuspended in 3 volumes of ice-cold CLIP lysis buffer (50 mM HEPES pH 7.5, 150 mM KCl, 2 mM EDTA, 0.5% (v/v) NP-40, 0.5 mM DTT, 1 × Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78442), 1 × RNaseOUT recombinant ribonuclease inhibitor (Invitrogen, 10777019)). The pellets were lysed by rotating at 4 °C for 15 min after passing through a 26 G needle (BD Biosciences). Embryo suspensions were sonicated on the Bioruptor system (Diagenode) with 30 s on/30 s off for 5 cycles. Lysates were cleared by centrifugation at 21,000g for 15 min at 4 °C on a benchtop centrifuge. The supernatants were applied to Flag-antibody-conjugated (Abcam, ab205606) protein A beads (Invitrogen, 1001D) and left overnight at 4 °C on an end-to-end rotor. The beads were washed extensively with 1 ml wash buffer (50 mM HEPES pH 7.5, 300 mM KCl, 0.05% (v/v) NP-40, 1 × Halt protease and phosphatase inhibitor cocktail, 1 × RNaseOUT recombinant ribonuclease inhibitor) at 4 °C five times. Protein–RNA complex conjugated to the beads was treated with 8 U μl⁻¹ RNase T1 (Thermo Fisher Scientific, EN0541) at 22 °C for 10 min with shaking. The input samples were digested in parallel. Then, input and IP samples were separated on an SDS–PAGE gel and gel slices at corresponding size ranges were treated by proteinase K (Invitrogen, 25530049) elution. RNA was recovered with TRIZol reagent (Invitrogen, 15596026). T4 PNK (Thermo Fisher Scientific, EK0031) end repair was then performed with purified RNA before library construction with the NEBNext Small RNA Library Prep Set for Illumina (NEB, E7330S). Libraries were pooled and sequenced on the NovaSeq 6000 sequencer.

Electrophoretic mobility shift assay

Recombinant MBD6-MBD–MBP–His protein was purified from Escherichia coli BL21 (DE3). Different concentrations of proteins were mixed with 100 nM FAM-labelled oligo probes in 1 × binding buffer (20 mM HEPES pH 7.5, 40 mM KCl, 10 mM MgCl₂, 0.1% Triton X-100, 10% glycerol and 1 × RNaseOUT Recombinant Ribonuclease Inhibitor (Invitrogen, 10777019)). The probe–protein mixture was incubated on ice for 30 min. The mixtures were loaded to a 10% Novex TBE Gel (Invitrogen, EC62755BOX). After gel running at 4 °C in 0.5× TBE for 2 h, the gel was washed twice in 0.5× TBE for 5 min. Washed gel was imaged with the GelDoc imaging system (Bio-Rad) with channel ‘FAM’. Individual K_D values were determined from a regression equation Y = [P]/(K_D + [P]), where Y is the fraction of probe bound at each protein concentration. The fraction bound is determined from the background-subtracted signal intensities using the expression: bound/(bound + unbound). [P] is protein concentration in each sample.

Quantitative analysis of RNA modification levels of CLIP RNA

Cultured mES cells were washed twice with DPBS before UV cross-linking at 254 nm with a Stratalinker (Stratagene) and flash-frozen in liquid nitrogen. The pellets were thawed on ice and resuspended in 3 volumes of ice-cold CLIP lysis buffer (50 mM HEPES pH 7.5, 150 mM KCl, 2 mM EDTA, 0.5% (v/v) NP-40, 0.5 mM DTT, 1 × Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78442), 1 × RNaseOUT recombinant ribonuclease inhibitor (Invitrogen, 10777019)). The pellets were lysed by rotating at 4 °C for 15 min after passing through a 26 G needle (BD Biosciences). The cell suspensions were sonicated on the Bioruptor system (Diagenode) with 30 s on/30 s off for 5 cycles. Lysates were cleared by centrifugation at 21,000g for 15 min at 4 °C on a benchtop centrifuge. The supernatants were applied to Flag-antibody (Abcam, ab205606) conjugated protein A beads (Invitrogen, 1001D) and left overnight at 4 °C on an end-to-end rotor. Beads were washed extensively with 1 ml wash buffer (50 mM HEPES pH 7.5, 300 mM KCl, 0.05% (v/v) NP-40, 1 × Halt Protease and Phosphatase Inhibitor Cocktail, 1 × RNaseOUT Recombinant Ribonuclease Inhibitor) at 4 °C five times. Then, the input and IP samples were treated by proteinase K (Invitrogen, 25530049) to release cross-linked RNA. RNA was recovered with TRIZol reagent (Invitrogen, 15596026). Ribosomal RNA was then removed using the RiboMinus Eukaryote System v2 (Invitrogen, A15026) with purification and size-selection using the RNA Clean & Concentrator-5 (Zymo Research, R1013) kit. Recovered RNAs were subjected to digestion and MS/MS analysis.

Biotinylation of immunoprecipitated RNAs

Biotin labelling of immunoprecipitated RNA was performed according to a published protocol⁵³.

Fluorescence microscopy

For immunolabelling, cells were fixed with 4% PFA in DPBS at 37 °C for 5 min, permeabilized with methanol at −20 °C for 8 min, dried at room temperature for 10 min and then washed three times with DPBS at room temperature. The chambers were blocked in blocking buffer (DPBS, 0.5% BSA, 0.05% Triton X-100, 1:100 SUPERase·In (Invitrogen, AM2694)) for 1 h at room temperature and primary antibodies were diluted in blocking solution according to the suggested fold from the manufacturer’s and incubate at room temperature for 1 h. Chambers were washed three times with 0.05% Triton X-100 in DPBS, then 1:1,000 diluted goat anti rabbit IgG-AF568 conjugate (Invitrogen, A-11011) in blocking solution was added to each well and the chambers were incubated at room temperature for 1 h. The chambers were then washed three times with 0.05% Triton X-100 in DPBS and fixed with 4% PFA in DPBS for 30 min at room temperature and washed three times with DPBS. Nuclei were counterstained with 2 µg ml⁻¹ Hoechst 33342 (Abcam, ab145597) in DPBS at room temperature for 20 min, wash with DPBS three times. The chambers were stored at 4 °C before proceeding to imaging on a Leica SP8 laser-scanning confocal microscope at University of Chicago.

Lifetime profiling

Transcription inhibitor actinomycin D (Act D, Abcam ab141058) was applied to a final concentration of 2.5 μM in mES cell medium to cultured mES cells or cultured Lin⁻KIT⁺ mouse HSPCs. Actinomycin D treatment started at 48 h after siRNA transfection (if any). RNAs were extracted from cells at different timepoints after actinomycin D treatment (10 min, 3 h and 6 h). Custom spike-in RNA (in vitro transcribed from firefly luciferase coding sequence) was added proportional to the yield of total RNA for different samples for RNA quantifications. RNA abundance was normalized to the value at 10 min for each condition.

DNA-seq data analysis

Raw reads were trimmed with Trimmomatic (v.0.39)⁵⁴ and then mapped to mouse genome (mm10) or human genome (hg38), together with Drosophila melanogaster chromatin (spike-in chromatin), using bowtie2 (v.2.4.1)⁵⁵ using the default mode, where multiple alignments are searched and the best one is reported. Mapped reads were deduplicated using the Picard tool MarkDuplicates (v.2.26.2; http://broadinstitute.github.io/picard/).

For ATAC–seq, reads that mapped to the mitochondrial genome were discarded before deduplication. Peaks were identified using MACS2⁵⁶ with the default mode, except for the parameters ‘--shift −75 --extsize 150 --nomodel --call-summits’. For CUT&Tag–seq, peaks were called using MACS2 with the default mode, except for the parameters ‘--broad --broad-cutoff 0.01’. For both ATAC–seq and CUT&Tag-seq, peaks that appeared in at least two biological replicates were retained for subsequent downstream analyses. The chromatin accessibility (ATAC) and H2AK119ub levels (CUT&Tag) were normalized by considering both sequencing depth and spike-in Drosophila melanogaster chromatin.

For meDIP–seq, differentially methylated regions were identified using MEDIPS⁵⁷ with the following settings: diff.method = ‘edgeR’, p.adj = ‘bonferroni’, MeDIP = True, CNV = False, minRowSum = 10. Regions with an adjusted P value of less than 0.1 were defined as significantly differentially methylated regions.

Nascent RNA-seq data analysis

Raw reads were trimmed with Trimmomatic (v.0.39)⁵⁴, and then aligned to mouse genome and transcriptome (mm10, version M19) as well as external RNA Control Consortium (ERCC) RNA spike-in control (Thermo Fisher Scientific) using HISAT2 (v.2.2.1)⁵⁸. Annotation files (version M19 for mouse) were obtained from GENCODE database (https://www.gencodegenes.org/)⁵⁹. Reads on each GENCODE annotated gene were counted using HTSeq (v.0.12.4)⁶⁰ and then normalized to counts per million (CPM) using edgeR packages in R⁶¹. CPM was converted to attomole by linear fitting of the RNA ERCC spike-in. The RNA level and EU adding time were fitted using a linear mathematical model, and the slope was estimated as transcription rate of RNA.

CLIP–seq data analysis

Low-quality reads were filtered using ‘fastq_quality_filter’, and adapters were clipped using ‘fastx_clipper’, then adapter-free reads were collapsed to remove PCR duplicates using ‘fastx_collapser’ and, finally, reads longer than 15 nucleotides were retained for further analysis (http://hannonlab.cshl.edu/fastx_toolkit/). Reads from rRNA were removed. The preprocessed reads were mapped to mouse genome (mm10) using bowtie (v.1.0.0)⁶² with ‘-v 3 -m 10 -k 1 --best --strata’ parameters. Mapped reads were separated by strands with samtools (v.1.16.1)⁶³ and peaks on each strand were called using MACS2 (v.2)⁵⁶ with parameter ‘-nomodel, --keep-dup 5, -g 1.3e8, -extsize 150’ separately. Significant peaks with q < 0.01 identified by MACS2 were considered. Peaks identified in at least two biological replicates were merged using bedtools (v.2.31.0)⁶³ and were used in the following analyses.

RNA-seq data analysis

Raw reads were trimmed with Trimmomatic (v.0.39)⁵⁴, then aligned to mouse (mm10) or human (hg38) genome and their corresponding transcriptome, together with external RNA Control Consortium (ERCC) RNA spike-in control (Thermo Fisher Scientific) when applicable, using HISAT2 (v.2.2.1)⁵⁸. Annotation files (version M19 for mouse, and version v29 for human in gtf format) were obtained from GENCODE database (https://www.gencodegenes.org/)⁵⁹. Reads were counted for each GENCODE annotated gene using HTSeq (v.0.12.4)⁶⁰ and for caRNAs using featureCounts⁶⁴, and then differentially expressed genes were called using DESeq2 package in R⁶⁵ with P < 0.05. In this step, the spike-in normalization factor was calculated by dividing the number of reads mapped to ERCC spike-ins by the total number of mapped transcriptomic reads. This factor was then included in the size factor calculation for DESeq2.

m⁵C meRIP–seq data analysis

Raw reads were trimmed with Trimmomatic (v.0.39)⁵⁴, then aligned to mouse (mm10) or human (hg38) genome and transcriptome, together with m⁵C modified or unmodified mRNA spike-ins (see the ‘m⁵C methylated RNA immunoprecipitation with spike-in’ section for details), using HISAT2 (v.2.1.0)⁵⁸. Annotation files (version M19 for mouse, and version v29 for human in gtf format) were downloaded from the GENCODE database (https://www.gencodegenes.org/)⁵⁹. Mapped reads were deduplicated using a Picard tool ‘MarkDuplicates’ (v.2.26.2) (http://broadinstitute.github.io/picard/). The remaining reads were separated by strands with samtools (v.1.16.1)⁶³ and peaks on each strand were called using MACS2 (v.2)⁵⁶ with the parameters ‘--nomodel --extsize 150’. Genome-specific parameters ‘-g hs’ for human and ‘-g mm’ for mouse were separately applied. We required significant peaks (q < 0.01) to appear in all biological samples to be considered validated. Peaks within the same conditions were then merged using bedtools (v.2.31.0)⁶³ for subsequent analysis. To quantify m⁵C methylation levels, we initially compared reads mapped to m⁵C-methylated spike-ins with those mapped to their unmethylated counterparts to confirm satisfactory pull-down efficiency. The m⁵C methylation levels were determined by calculating the log₂-transformed fold changes between immunoprecipitation) and input samples. The normalization factor was calculated by dividing the number of reads mapped to the m⁵C-methylated spike-in by the total number of transcriptomic reads. This approach enabled us to quantify the global changes in m⁵C levels under different conditions.

Chromatin-associated RNA UBS amplicon-seq analysis

Adapter sequences and low-quality reads were trimmed using cutadapt (v.4.0). Only properly paired reads with a length less than 20 nucleotides were retained. The 7 nucleotides of the UMI at the 5′ end of the insert fragments (R2) were extracted. Clean reads were then mapped to the mouse genome sequence (mm10) using the HISAT-3n tool⁶⁶ with the ‘--base-change C,T’ argument. To leverage the strand-specific property of the library, the ‘--directional-mapping’ parameter was applied. To increase the accuracy of site identification, only properly paired reads without soft clipping were retained. To eliminate unconverted clusters, reads containing more than three unconverted C sites, or where more than one-third of the total C sites were unconverted, were discarded. A binomial model was used to calculate a P value for each site, and sites with a P value less than 0.01 were classified as m⁵C sites.

Antibodies

The antibodies used in this study are summarized below: rabbit monoclonal anti-H2AK119ub antibody (Cell Signaling Technology, 8240S, 1:1,000 for western blot, 1:50 for CUT&Tag); rabbit monoclonal anti-H3 antibody (Cell Signaling Technology, 4499S, 1:1,000); mouse monoclonal anti-TET2 antibody (MilliporeSigma, MABE462, 1:500); rabbit monoclonal anti-GAPDH antibody, HRP conjugate (Cell Signaling Technology, 8884S, 1:1,000); rabbit monoclonal anti-DDDDK tag antibody (Abcam, ab205606, 1:1,000 for western blot, 1:50 for immunoprecipitation); rabbit polyclonal anti-SNRP70/U1-70K antibody (Abcam, ab83306, 1:1,000); mouse monoclonal anti-5-methylcytosine antibody (Diagenode, C15200081-100, 1:1,000 for dot blot, 1:50 for meRIP); mouse monoclonal anti-hm⁵C antibody (Diagenode, C15200200-100, clone Mab-31HMC, 1:1,000 for dot blot); rabbit monoclonal anti-H3K27me3 antibody (Cell Signaling Technology, 9733S, only for CUT&Tag experiments, 1:50); mouse monoclonal anti-BAP1 antibody (Santa Cruz, sc-28383, 1:50 for CUT&Tag). Goat anti-rabbit IgG, HRP conjugated antibody (Cell Signaling Technology, 7074S, 1:2,000) and horse anti-mouse IgG, HRP conjugated antibody (Cell Signaling Technology, 7076S, 1:2,000) were used as secondary antibodies. Mouse IgG-isotype control (Abcam, ab37355, 1:50 for immunoprecipitation) and rabbit IgG-isotype control (Abcam, ab37415, 1:50 for immunoprecipitation) were used as normal IgG controls. PerCP-Cy5.5 mouse lineage antibody cocktail (BD Biosciences, 561317, 1:100); PE rat anti-mouse CD117 (BD Biosciences, 553869, 1:100); Brilliant Violet 421 (BV421, 1:100) anti-mouse/human CD11b (Mac-1) (BioLegend, 101236, 1:100); APC mouse anti-human CD45 (BD Biosciences, 555485, 1:100); PE mouse anti-human CD33 (BD Biosciences, 561816, 1:100); PE-Cy7 rat anti-mouse CD45 (BD Biosciences, 552848, 1:100); PerCP-Cy5.5 mouse anti-mouse CD45.2 (BD Biosciences, 552950, 1:100) and FITC mouse anti-mouse CD45.1 (BD Biosciences, 553775, 1:100). All antibodies were applied at a dilution fold according to the manufacturer’s suggestions for specific use unless specified elsewhere in the Methods.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.