OGT prevents DNA demethylation and suppresses the expression of transposable elements in heterochromatin by restraining TET activity genome-wide

Abstract

O-GlcNAc transferase (OGT) interacts robustly with all three mammalian TET methylcytosine dioxygenases. Here we show that deletion of the Ogt gene in mouse embryonic stem (mES) cells results in a widespread increase in the TET product 5-hydroxymethylcytosine in both euchromatic and heterochromatic compartments, with a concomitant reduction in the TET substrate 5-methylcytosine at the same genomic regions. mES cells treated with an OGT inhibitor also displayed increased 5-hydroxymethylcytosine, and attenuating the TET1–OGT interaction in mES cells resulted in a genome-wide decrease of 5-methylcytosine, indicating that OGT restrains TET activity and limits inappropriate DNA demethylation in a manner that requires the TET–OGT interaction and the catalytic activity of OGT. DNA hypomethylation in OGT-deficient cells was accompanied by derepression of transposable elements predominantly located in heterochromatin. We suggest that OGT protects the genome against TET-mediated DNA demethylation and loss of heterochromatin integrity, preventing the aberrant increase in transposable element expression noted in cancer, autoimmune-inflammatory diseases, cellular senescence and aging.

DNA cytosine methylation and demethylation are controlled by DNA methyltransferases (DNMTs) and ten-eleven translocation (TET) methylcytosine dioxygenases, respectively^1-4. The three mammalian members of the TET family (TET1, TET2 and TET3) catalyze the oxidation of 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC) and beyond^5-10. Together, the oxidized methylcytosines are intermediates in both the ‘passive’ replication-dependent and the ‘active’ replication-independent pathways of DNA demethylation (reviewed in refs. 2-4). Tet gene deletion results in the expected increase in DNA methylation at euchromatic regions—gene bodies of expressed genes and active enhancers^3,9. Unexpectedly, in every cell type examined, Tet gene deletion is also reproducibly associated with decreased DNA methylation in heterochromatin, a chromatin compartment defined by replication timing, lamina association or Hi-C data⁸. In mouse embryonic stem (mES) cells, this hypomethylation may in part be due to a limited redistribution of DNMT3A from heterochromatic regions to euchromatic regions previously occupied by TET1 (ref. 8). We asked if TET-interacting partners could modulate the levels and distribution of TET-generated epigenetic marks.

O-GlcNAc transferase (OGT) is an essential X-chromosome-encoded enzyme that modifies the hydroxyl groups of Ser/Thr residues in many intracellular (cytoplasmic and nuclear) proteins with N-acetylglucosamine (GlcNAc)^11-13. OGT is essential for the very earliest stages of embryonic development in mammals¹⁴ and for the proliferation and survival of all primary cells and cell lines examined^11,15. TET proteins recruit OGT to chromatin¹⁶, and all three TET enzymes bind tightly to and are O-GlcNAcylated by OGT^15-22. However, the functional consequences of the TET–OGT interaction are not well understood.

A major function of DNA methylation is the transcriptional repression of transposable elements (TEs), imprinted genes, satellite repeats and other repetitive elements^23-26. TEs include long terminal repeat (LTR)-containing elements—endogenous retroviruses (ERVs) and intracisternal A-type particles (IAPs)—as well as non-LTR retrotransposons (long interspersed elements (LINEs) and short interspersed elements (SINEs))²⁵. The expression of TEs, especially ERVs and all but the youngest LINE-1 (L1) families, is suppressed by the transcriptional co-repressor KAP1 (also known as TRIM28)^25,27, which is recruited to TE sequences by an extensive family of KRAB zinc-finger proteins that coevolved with TEs in an arms race to suppress TE expression²⁸.

To explore the connection between OGT and TET-mediated DNA demethylation in mES cells, we generated male Ogt-floxed (Ogt fl), Cre-ERT2 mES cells that could be inducibly deleted for the Ogt gene by treatment with 4-hydroxytamoxifen (4-OHT)¹⁵. To assess changes in DNA cytosine modification status, we used whole-genome bisulfite sequencing (WGBS), Oxford Nanopore Technologies sequencing (ONT-seq)²⁹, cytosine-5-methylenesulfonate (CMS) immunoprecipitation (CMS-IP)³⁰ and 6-base sequencing³¹, a base-resolution method that simultaneously detects 5hmC, 5mC and all four canonical bases in a single sequencing run. We assigned the observed changes to euchromatic and heterochromatic compartments by Hi-C analysis⁸.

We report that acute deletion of the Ogt gene results in a global increase of 5hmC and a global loss of 5mC that occurs in both euchromatic and heterochromatic compartments by 3–6 days of 4-OHT exposure. mES cells engineered to lack the TET1–OGT interaction¹⁹ showed a similar global loss of 5mC, while treatment of mES cells with the potent and selective OGT inhibitor OSMI-4 (ref. 32) resulted in a rapid genome-wide loss of the O-GlcNAc modification and an increase of 5hmC by 4 days. Loss of 5mC was accompanied by increased expression of TEs (the L1 family and the mouse ERV-L (MERVL) (MT2) and IAP subfamilies); and increased TE expression could be linked in certain cases to increased cis-expression of genes and exons located 3′ of the expressed TE. Thus, OGT, through its catalytic activity and the TET–OGT interaction, restrains TET activity genome-wide, maintaining genome stability by suppressing widespread DNA demethylation and the consequent increase in TE expression.

Results

Global decrease in DNA cytosine modification in Ogt iKO mES cells

To explore whether OGT affected DNA modification pathways and TET activity in mES cells, we used an inducible system to disrupt expression of the Ogt gene¹⁵. We generated male Ogt fl mES cells from Ogt fl Cre-ERT2 Rosa26-YFP^LSL mouse blastocysts and treated them with 4-OHT for 3–6 days; Ogt fl cells not treated with 4-OHT were used as controls (Ctrl)¹⁵ (Fig. 1a,b). We sorted 4-OHT-treated mES cells expressing high levels of YFP (gating strategy in Extended Data Fig. 1a,b) and confirmed, consistent with our previous study¹⁵, that Ogt inducible knock-out (iKO) mES cells resulting from 6 days of 4-OHT treatment displayed an almost total loss of Ogt mRNA (Extended Data Fig. 1c), OGT protein (Fig. 1b, top) and the O-GlcNAc modification (Fig. 1b, bottom). We chose this method for OGT depletion because siRNA- and Cas9/single guide RNA (sgRNA)-based methods yield mixed populations of OGT-high and OGT-low cells, and because cells expressing even low levels of OGT have a selective growth advantage over cells lacking OGT^11,15. Mass spectrometry of whole-cell lysates showed that OGA protein decreased in parallel with OGT protein after Ogt deletion (Extended Data Fig. 3a).

Fig. 1 ∣. Ogt deletion results in reduced DNA methylation genome-wide.

a, A flow chart for generation of Ogt iKO mES cells. Ogt fl, Cre^ERT2, Rosa26-YFP^LSL mES cells were cultured for 6 days with 4-OHT to yield Ogt iKO mES cells. Sorted YFP⁺ Ogt fl mES cells were used for all experiments. Ogt fl mES cells not exposed to 4-OHT were used as control (Ctrl Ogt fl). b, Western blots showing loss of OGT protein (top) and the O-GlcNAc modification (bottom) in whole-cell lysates prepared after 6 days of exposure of Ogt fl mES cells to vehicle or 4-OHT. Actin, loading control. c,d, Violin plots showing global loss of DNA methylation in Ctrl Ogt fl mES cells and Ogt iKO mES cells measured by WGBS (c) and ONT sequencing (d). DNA modification (5mC+5hmC) levels were assessed in 10-kb windows across the genome. Ten biological replicates were measured by WGBS (4) or ONT (6). e, Genome browser view of a portion of Chr. 9, showing 5mC+5hmC in all CpGs with sufficient coverage in Ctrl Ogt fl (blue) and Ogt iKO (orange) mES cells. Overlaid tracks of WGBS (top track) and ONT-seq (second track) are shown. Positive and negative Hi-C PC1 values (third track) identify euchromatin (Hi-C A compartment, green) and heterochromatin (Hi-C B compartment, blue), respectively. RefSeq, NCBI Reference Sequence Database. f, Dot plot of change in DNA methylation (WGBS) in Ctrl Ogt fl and Ogt iKO in euchromatic (Hi-C A) and heterochromatic (Hi-C B) compartments. Each dot represents a 10-kb window containing at least five CpGs covered by at least five reads. g, Violin plots of DNA methylation (ONT sequencing) in euchromatin (left) and heterochromatin (right). Mean values are indicated by horizontal red lines. The majority of 10-kb windows lose DNA modification in Ogt iKO compared with Ctrl Ogt fl mES cells. h, Violin plots of DNA methylation (ONT sequencing) at proximal gene promoters located in euchromatin (left) or heterochromatin (right). Promoters of protein-coding genes (−1,000 to +500 bp relative to the transcription start site) were classified as containing or not containing CGIs. Mean values averaged over the promoter are indicated by horizontal red lines. The numbers of promoters in each category are indicated. In c, d, g and h, box plot elements: red line, median; boxes, first and third quartiles; whiskers, 1.5 × interquartile range.

To determine the global effects of Ogt deletion on DNA cytosine modification, we used WGBS and long-read ONT sequencing²⁹ to map 5mC+5hmC genome-wide, including repetitive regions and TEs (Fig. 1c-h). Both methods revealed a global reduction of 5mC+5hmC in Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 1c,d). The decrease was observed in both heterochromatic (Hi-C B) and euchromatic (Hi-C A) compartments, defined by calculating Hi-C principal component 1 (PC1) values in wild-type (WT) mES cells⁸ (Fig. 1e, tracks 1 and 2; Fig. 1f,g and Extended Data Fig. 1d). DNA methylation (by ONT-seq) at cytosine-guanine dinucleotide (CpG) island (CGI) promoters was uniformly low as expected, whereas non-CGI promoters—which tend to be associated with cell-type-specific genes and located in facultative or constitutive heterochromatin in nonexpressing cell types³³—showed a broad range of DNA methylation that decreased substantially in Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 1h).

OGT iKO mES cells display increased 5hmC

We next asked if 5hmC levels were altered in Ogt iKO compared with Ctrl Ogt fl mES cells. Dot blotting of bisulfite-treated DNA, developed with antibodies to CMS (the product of the reaction of 5hmC with sodium bisulfite⁶), showed a reproducible increase of 5hmC in genomic DNA from Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 2a and Extended Data Fig. 2a). Flow cytometry showed that the gain of 5hmC occurred uniformly in the entire population of OGT-deficient mES cells, not just in a subpopulation of cells (Extended Data Fig. 2b). Notably, the gain of 5hmC occurred despite a decrease in TET1 and TET2 mRNA and protein levels (Extended Data Fig. 3a-c), indicating that OGT suppresses TET enzymatic activity but also maintains TET1 and TET2 expression in WT mES cells. Whole-genome mapping of 5hmC by CMS-IP^30,34-36 showed that the increase in 5hmC occurred to a similar extent in both euchromatic (Hi-C A) and heterochromatic (Hi-C B) compartments (Extended Data Fig. 2c,d), even though 5hmC was predominantly located in euchromatin (positive PC1 values) in Ctrl Ogt fl mES cells as expected⁸. Analysis of CMS-IP data showed that more peaks gained than lost 5hmC (Extended Data Fig. 2e).

Fig. 2 ∣. Ogt deletion results in increased TET activity.

a, DNA dot blot showing increased 5hmC in Ogt iKO compared with Ctrl Ogt fl mES cells. A representative experiment is shown; dot blots of all four replicate experiments are shown in Extended Data Fig. 2a.b, Fractions of 5mC, 5hmC and unmodified cytosines in two biological replicates (r) each of Ctrl Ogt fl and Ogt iKO mES cells. c, Genome browser view of 5hmC (top) and 5mC (bottom) in overlaid tracks of Ogt iKO (orange) and Ctrl Ogt fl (blue) mES cells. Values were averaged across 10-kb windows. Hi-C PC1 values (middle) distinguish euchromatin (green) and heterochromatin (blue). d, Violin plots showing increased 5hmC (top) and decreased 5mC (bottom) in Ogt iKO compared with Ctrl Ogt fl mES cells at whole-genome resolution (left) or by chromatin compartments (right); data from same four samples as in b. Box plot elements: red line, median; boxes, first and third quartiles; whiskers, 1.5 × interquartile range. e, Dot plots showing increased 5hmC (top) and decreased 5mC (bottom) in Ctrl Ogt fl compared with Ogt iKO in euchromatin (Hi-C A) and heterochromatin (Hi-C B) compartments. Averaged DNA modification values in 10-kb windows are shown. log₂FC, log₂ fold change. f, Density plots of 5hmC (top) and 5mC (bottom) in Ctrl Ogt fl plotted against Ogt iKO mES cells in euchromatin (Hi-C A, left) and heterochromatin (Hi-C B, right) compartments. Averaged DNA modification values in 10-kb windows are shown. The large majority of windows in both compartments show increased 5hmC and decreased 5mC. g, Venn diagram showing the overlap between CpGs gaining 5hmC (red) and those losing 5mC (blue) in Ogt iKO mES cells compared with Ctrl Ogt fl mES cells. For the same analysis at the level of 10-kb windows, please see Extended Data Fig. 4b. h, Scatter plots showing the ratio of changes in 5hmC and 5mC in Ogt iKO versus Ctrl Ogt fl mES cells, separated by euchromatin (top, positive PC1 values; Hi-C A) and heterochromatin (bottom, negative PC1 values; Hi-C B) compartments. Averaged DNA modification values in 10-kb windows are shown. The large majority of windows that gain 5hmC in both euchromatin and heterochromatin show loss of 5mC, but a minor subset of windows in euchromatin that gain 5hmC also gain 5mC.

We used 6-base sequencing³¹ to simultaneously detect 5mC, 5hmC and the canonical bases A, C, G and T at base resolution in a single sequencing run (Fig. 2b-h; for quality control data, see Supplementary Fig. 1). This method confirmed both the increase in 5hmC and the decrease in 5mC across the genome of Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 2b,c and Fig. 2d, left). In Ctrl Ogt fl mES cells (blue violin plots), 5hmC is more abundant in euchromatin than in heterochromatin, while 5mC is more abundant in heterochromatin (Fig. 2d, right). When Ogt iKO mES cells (orange violin plots) are compared with Ctrl Ogt fl mES cells (blue violin plots), there is a clear increase in 5hmC and decrease in 5mC in both euchromatic and heterochromatic compartments (Fig. 2d, right, and Fig. 2e,f). Notably, the CpGs that showed a decrease in 5mC overlapped strongly with those displaying an increase in 5hmC (Fig. 2g, 76–80% overlap); the overlap was also striking when decreased 5mC and increased 5hmC were computed for genomic 10-kb windows (containing at least five CpGs with at least five reads covering each CpG; Extended Data Fig. 4a,b). There was an inverse relationship between gain of 5hmC and loss of 5mC in the majority of 10-kb windows in both compartments (Fig. 2h), but a minor subset of 10-kb windows in euchromatin exhibited concomitant increases of both 5hmC and 5mC in Ogt iKO mES cells (Fig. 2h, left, dotted oval). This subset was not observed in heterochromatin, in which the vast majority of windows that gained 5hmC also lost 5mC (Fig. 2h, right).

Decreased DNMT1 and UHRF1 levels in Ogt iKO mES cells

RNA sequencing (RNA-seq), mass spectrometry and western blotting revealed a perceptible decrease in mRNA and protein levels of the maintenance methyltransferase DNMT1, its partner UHRF1 and the two TET enzymes expressed in mES cells, TET1 and TET2 (ref. 37), in Ogt iKO compared with Ctrl Ogt fl mES cells (Extended Data Fig. 3a-c). There was also a minor reduction in KAP1 and DNMT3A protein levels and a small increase in DNMT3B protein levels assessed by western blotting (Extended Data Fig. 3b). Conceivably, the reduction in levels of the DNMT1–UHRF1 maintenance methyltransferase complex could account for the decrease in 5mC in Ogt iKO versus Ctrl Ogt fl mES cells; however, we would not expect to see increased 5hmC under these conditions, especially since TET1 and TET2 mRNA and protein levels were also perceptibly decreased (Extended Data Fig. 3a-c). A plausible alternative explanation was that the reduction of DNMT1–UHRF1 levels observed after Ogt deletion was a secondary phenomenon caused by G1 arrest after Ogt deletion¹⁵. Indeed, studies in normal cells have shown that DNMT1 mRNA and protein are both regulated in a cell-cycle-dependent way^38-41; for instance, G0/G1-arrested human mammary epithelial cells had low levels of DNMT1 protein that were increased by treatment with proteasome inhibitors⁴⁰. DNMT1 degradation requires acetylation by Tip60 histone acetyltransferase and subsequent ubiquitylation by UHRF1; conversely, deacetylation by HDAC1 and deubiquitylation by USP7 (HAUSP) stabilized DNMT1 (ref. 41).

To assess the effect of G1 arrest on DNMT1 and UHRF1 protein levels, we arrested Ctrl Ogt fl mES cells in G1 by treating them with the CDK4/6 inhibitor palbociclib (PD-0332991) for 3 days (Extended Data Fig. 3d). Under these conditions, DNMT1 and UHRF1 protein levels were both substantially reduced (Extended Data Fig. 3e), with no change in OGT or TET activity measured by flow cytometry for O-GlcNAc and 5hmC, respectively (Extended Data Fig. 3f,g). We conclude that G1 arrest of cells at least partly accounts for decreased DNMT1–UHRF1 levels in Ogt iKO cells.

We compared the kinetics of increase in 5hmC at 3 and 6 days after Ogt gene deletion with the kinetics of decrease in DNMT1 and UHRF1 mRNA levels (Extended Data Fig. 5a). We treated Ogt fl mES cells with 4-OHT for 3 and 6 days, sorted YFP⁺ cells and performed flow cytometry for 5hmC (Extended Data Fig. 5b,c). In two independent experiments (Extended Data Fig. 5b,c, left), Ogt iKO mES cells (red histograms) showed a shoulder of low 5hmC staining at 3 days, coinciding with the peak of 5hmC staining in Ctrl Ogt fl cells (blue histograms), but this shoulder disappeared by 6 days when all Ogt iKO cells showed uniformly high 5hmC (Extended Data Fig. 5b,c, right). There were only minor decreases in Tet1, Tet2, Tet3, Dnmt1 and Uhrf1 mRNA levels by quantitative real-time PCR (qRT–PCR) after treatment of Ogt fl mES cells with 4-OHT for 3 days (Extended Data Fig. 5d) compared with the somewhat more pronounced difference at 6 days (Extended Data Fig. 3c). Hence, 5hmC levels increase early in OGT-deficient mES cells, suggesting that DNA demethylation is initiated by OGT deficiency and increased TET activity, although decreased activity of the DNMT1–UHRF1 complex could maintain low levels of 5mC.

OGT enzymatic activity is needed to suppress TET activity

To assess whether OGT catalytic activity was required to suppress TET activity, we used the specific OGT inhibitor OSMI-4 (ref. 32). Ctrl Ogt fl mES cells treated with 20 μM OSMI-4 showed slower proliferation within 2 days (Fig. 3a) and displayed a sharp decrease in O-GlcNAc staining to baseline levels within 12 h that was maintained for 4–6 days (Fig. 3b,c and Extended Data Fig. 6a,b). By 4 days, cells treated with 20 μM OSMI-4 displayed a strong, dose-dependent increase in 5hmC levels, comparable to the increase in 5hmC levels observed in Ogt iKO mES cells (Fig. 3d and Extended Data Fig. 6c). 6-base sequencing revealed that 5hmC levels increased genome-wide, both in heterochromatin and in euchromatin (Fig. 3e,g). Unexpectedly, however, 5mC levels remained high in OSMI-4-treated mES cells (Fig. 3f,g), probably because OSMI-4 treated cells display earlier growth arrest and therefore even lower levels of passive replication-dependent DNA methylation compared with Ogt iKO mES cells. 5hmC interferes with recognition of hemi-methylated CpGs by the DNMT1–UHRF1 complex^42,43, resulting in dilution of DNA methylation at each cycle of DNA replication, but neither 5mC nor 5hmC can be passively diluted in nonreplicating cells.

Fig. 3 ∣. OGT catalytic activity is required to restrain TET activity.

a, Cumulative growth curves of mES cells (Ogt fl) treated with or without the OGT inhibitor OSMI-4 at day 0 and counted at each passage (every 2 days) thereafter until day 6. Data are shown as mean ± s.d. (from three biological replicates). Note the logarithmic scale on the y axis. b, O-GlcNAc levels in OSMI-4-treated and control mES cells assessed by flow cytometry. Quantification of relative mean fluorescence intensity (MFI) of O-GlcNAc staining. For representative histograms of the time course, see Extended Data Fig. 6a. c, Histograms showing reduction of O-GlcNAc levels upon OSMI-4 treatment for 1 or 2 days. For a bar graph of data from multiple experiments, see Extended Data Fig. 6b. d, Increase of 5hmC after OGT inhibition. 5hmC was measured by flow cytometry after 4 days of OSMI-4 treatment. Ogt iKO mES cells are included as reference. The bar graphs represent the mean ± s.d. from three independent experiments. For representative histograms, see Extended Data Fig. 6c. In e–g, 6-base sequencing shows that OGT inhibition is associated with increased TET activity in mES cells. e,f, Left and middle: violin plots showing increased 5hmC levels (e) but unaltered 5mC levels (f) in OSMI-4-treated versus control Ogt fl mES cells. Right: changes in DNA modifications in OSMI-4-treated cells; each dot represents averaged DNA modification values in a 10-kb window, located in euchromatic or heterochromatic compartments (positive and negative Hi-C PC1 values, respectively). In e and f, box plot elements: red line, median; boxes, first and third quartiles; whiskers, 1.5 × interquartile range. g, Genome browser view of 5hmC (top) and 5mC (bottom) in overlaid tracks of OSMI-4-treated (orange) and control (blue) mES cells. Values were averaged across 10-kb windows. Middle: Hi-C PC1 values distinguish euchromatin (green) and heterochromatin (blue). 6-base sequencing data were analyzed from two biological replicates. h, Venn diagram showing the overlap between CpGs gaining 5hmC in Ogt iKO compared with Ctrl Ogt fl mES cells (orange) and those gaining 5hmC in OSMI-4-treated compared with untreated mES cells (‘OGT-inh’, green).

Comparison of 6-base sequencing data from Ogt iKO (Fig. 2g) versus OSMI-4-treated mES cells revealed a substantial overlap of CpGs and 10-kb windows with increased 5hmC in these two cell populations (Fig. 3h and Extended Data Fig. 6d), providing clear evidence that Ogt deletion and OGT inhibition resulted in comparable increases in TET activity at the same CpGs.

Increased TE expression in Ogt iKO mES cells

DNA methylation is known to suppress the expression of many TE families^23-26. Consistent with their genome-wide loss of 5mC, Ogt iKO mES cells displayed substantial upregulation of many TE families, including the youngest L1 families that retain the ability to mobilize (L1 T_F, G_F and A)⁴⁴ (Fig. 4a). We also observed a striking upregulation of ERVs, L1s and IAPEy/IAPEz retrotransposons; the expression of IAPs is normally repressed at least partly by OGT and DNA methylation⁴⁵ (Fig. 4a). By integrating TE expression with ONT-seq data²⁹, we observed that increased expression of MERVL-int and MT2_Mm, the internal and regulatory regions of the MERVL family, respectively (Fig. 4a), correlated with decreased 5mC+5hmC at MT2_Mm, the regulatory region of MERVL (Fig. 4b). Similarly, increased expression of the L1 G_F and L1 A families (Fig. 4a) was associated with decreased 5mC+5hmC at their 5′ untranslated regions (UTRs) (Fig. 4b, violin plots). Finally, although the 5′ UTRs of the youngest mobile L1 T_FI/II subfamilies⁴⁴ were poorly methylated in Ctrl Ogt fl mES cells while the L1 T_FIII subfamily was substantially methylated (Fig. 4b), all three subfamilies showed reduced 5mC+5hmC and increased expression in Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 4a,b).

Fig. 4 ∣. Ogt deletion results in enhanced TE expression.

a, Volcano plot of RNA-seq data showing differential TE expression in Ogt iKO relative to Ctrl Ogt fl mES cells. The number of individual copies analyzed for each TE subfamily is shown in parentheses. The red dots indicate transcripts exhibiting an at least twofold change in expression (log₂FC of ±1) and P-adjusted value <0.05 (using the Benjamini–Hochberg test). b, 5mC+5hmC (ONT-seq) levels of TE subfamilies in Ctrl Ogt fl (blue) and Ogt iKO (orange) mES cells. The violin plots indicate the median, interquartile range and 1.5 × interquartile range. c, RNA-seq data of Ogt iKO versus Ctrl Ogt fl mES cells, analyzed for differential expression of uniquely mapped TEs (log₂FC ±1, P value <0.05). Left: numbers of upregulated (red) or downregulated (gray) TEs in each family. Right: data for the youngest members of each family. d, Violin plots showing changes in 5mC+5hmC (ONT-seq) at regulatory regions of uniquely mapped, differentially expressed TEs in Ogt iKO versus Ctrl Ogt fl mES cells. e, Composite 5mC+5hmC profile (ONT-seq) of a young L1 T_FIII in Ogt iKO and Ctrl Ogt fl mES cells. Top: Ogt iKO mES cells (orange) show decreased average 5mC+5hmC along the length of L1 T_FIII elements compared with Ctrl Ogt fl mES cells (blue), with particularly pronounced loss of 5mC+5hmC in the 5′ UTR regulatory region at left. Middle: the red vertical lines indicate the position of CpGs. Bottom: 5mC+5hmC profiles of ONT long reads fully covering uniquely mapped L1 T_FIII elements. CpGs not confidently called, that is, absolute (log-likelihood ratio) >2.5, were omitted. f, 5hmC levels at 5′ UTRs of indicated L1 families in Ctrl Ogt fl and Ogt iKO mES cells measured by 6-base sequencing; difference of median 5hmC levels between conditions (Ogt iKO - Ctrl Ogt fl) is displayed on top. Only 5′ UTRs containing ≥10 detected CpGs with 20× minimum coverage from two biological replicates are included in the analysis. P values were calculated using a two-tailed paired t-test. g, Genome browser view showing increased 5hmC (CMS-IP), decreased 5mC (ONT-seq) and increased expression at GSAT_MM major satellite regions. Increased expression was confirmed using total ribo-depleted RNA-seq. In the RepliSeq track, E (green) and L (blue) denote early (euchromatic) and late (heterochromatic) replicating domains, respectively; the region shown resides entirely in heterochromatin. h, No change of 5mC+5hmC levels at uniquely mapped MT2/MERVL TEs that show no change in expression (log₂ fold change <0.2 and >−0.2) in Ogt iKO versus Ctrl Ogt fl mES cells. In b, d, f and h, box plot elements: red line, median; boxes, first and third quartiles; whiskers, 1.5 × interquartile range. In b, d and h, ONT-seq data were generated from three biological replicates.

Conventional analysis of TE expression involves aligning RNA-seq reads while permitting multimappers over hundreds or thousands of individual TEs that are very similar in sequence⁴⁶, allowing the mean expression of each TE subfamily to be estimated (Fig. 4a,b). However, this method does not provide specific information about the expression of individual TE loci in the family. To relate the expression of individual uniquely mapped TEs to 5mC and 5hmC, and to ask whether increased TE expression was linked to increased expression of nearby genes, we identified the subgroup of individual TEs within each subfamily that showed increased expression in Ogt iKO compared with Ctrl Ogt fl mES cells (Methods). In each upregulated TE family, the majority of uniquely mapped TEs showed increased expression in Ogt iKO versus Ctrl Ogt fl mES cells, with a much smaller number showing decreased expression (Fig. 4c). This smaller subset of uniquely mapped TEs showed a striking correlation of increased expression with loss of 5mC+5hmC (ONT-seq) in Ogt iKO versus Ctrl Ogt fl mES cells (Fig. 4d). Taken individually, however, the uniquely mapped L1 T_FIII elements showed wide variation in DNA modification, especially at their 5′ UTRs, in both Ctrl Ogt fl and Ogt iKO mES cells (Fig. 4d and Fig. 4e, bottom, compare blue and orange traces). There was no change in DNA modification at uniquely mapped TEs whose expression was not significantly altered (log₂ fold change <0.2 and >−0.2) in Ogt iKO compared with Ctrl Ogt fl mES cells (Fig. 4h).

We used 6-base sequencing to document 5hmC changes at the regulatory regions (5′ UTRs) and transcribed regions of young L1 families and subfamilies (defined as longer than 6 kb and labeled young-L1 in Fig. 4c) in Ogt iKO relative to Ctrl Ogt fl mES cells (Fig. 4f and Extended Data Fig. 4). The 5′ UTRs and transcribed regions of the L1 T_FI and L1 T_FII subfamilies showed decreased 5hmC upon Ogt deletion (Fig. 4f and Extended Data Fig. 4c-e), presumably because their low levels of the TET substrate 5mC (Fig. 4b) promoted rapid loss of both 5mC and 5hmC. By contrast, the 5′ UTRs and transcribed regions of the L1 G_F- and F-type families, and to a lesser extent the 5′ UTRs of the L1 T_FIII subfamily and the L1 A family, gained 5hmC (Fig. 4f and Extended Data Fig. 4c-e) and showed decreased 5mC+5hmC (Fig. 4b). The family of GSAT_MM satellite elements (Fig. 4a), as well as an individual 38-kb element found within the mouse mm10 genome assembly (Fig. 4g) also showed increased 5hmC and decreased 5mC concomitantly with increased expression. We conclude that OGT is required, directly or indirectly, to suppress the expression of many L1, LTR and repetitive elements by preventing the increase of 5hmC in regulatory regions and transcribed units of TEs in normal mES cells.

We confirmed increased TE expression by qRT–PCR (Extended Data Fig. 7a). By mass spectrometry, Ogt iKO cells contained increased levels of peptides derived from L1 ORF-1p and ORF-2p proteins, and from GAG proteins encoded in MERVL-int and IAP elements in Ogt iKO versus Ctrl Ogt fl mES cells (Extended Data Fig. 7b). Expression of a MERVL reporter⁴⁶ stably integrated into Ctrl Ogt fl mES cells was uniformly increased in Ogt iKO mES cells after 4-OHT treatment to delete Ogt (Extended Data Fig. 7c); the rightward shift of the entire population represents an increase in baseline MERVL expression in all Ogt iKO cells, rather than an increase in the small population of MERVL reporter-positive cells (not apparent in this experiment) that cycle in and out of the two-cell (2C) state in WT mES cells⁴⁷. Finally, analysis of data from WT mES cells⁴⁵ showed that the O-GlcNAc modification was enriched at regulatory regions of mouse MERVL, L1 and ERVK TE subfamilies (Extended Data Fig. 7d). Thus, reduced 5mC, increased 5hmC and loss of the O-GlcNAc modification—especially in chromatin proteins that bind TE regulatory regions—are all likely to facilitate increased TE expression after Ogt deletion.

The TE families that were most strikingly upregulated in Ogt iKO relative to Ctrl Ogt fl mES cells—MERVL, IAPs and young L1 T_F, G_F and A elements (Fig. 4a)—were predominantly located in heterochromatic regions of mES cells (Extended Data Fig. 7e). Moreover, TEs whose regulatory regions were enriched for O-GlcNAc (Extended Data Fig. 4d) were more highly enriched in heterochromatin compared with all TEs (Extended Data Fig. 7f).

An IAPEy element that mapped to a unique location in the euchromatic region of the genome was poorly expressed in Ctrl Ogt fl mES cells but showed increased expression in Ogt iKO mES cells (Supplementary Fig. 2a). The 5′ and 3′ LTRs of this element displayed clear evidence of reduced 5mC at specific CpGs by ONT-seq and were occupied by O-GlcNAcylated proteins, OGT, DNMT1, UHRF1 and KAP1 (refs. 45,48-51) (Supplementary Fig. 2a). Similarly, a solo MT2_Mm LTR and a relatively young L1Md_F3 element located adjacent to one another in heterochromatin were suppressed in Ctrl Ogt fl but expressed in Ogt iKO mES cells; the L1Md_F3 5′ UTR coincided with a peak of O-GlcNAc modification and lost 5mC at a subset of CpGs (Supplementary Fig. 2b).

Increased expression of genes and exons 3′ of expressed TEs

TEs are known to influence the expression of nearby genes, in part because their regulatory elements can act as enhancers and promoters in cis^25,26,52-55. Indeed, expression of the Arap2 gene and a MERVL-int/MT2_Mm element located 5′ of Arap2 were both upregulated in Ogt iKO compared with Ctrl Ogt fl mES cells, concomitantly with altered DNA methylation of certain CpGs in the MT2_Mm regulatory (LTR) region just 5′ of the annotated MERVL-int TE (Fig. 5a). Similarly, increased expression of a MERVL element (MERVL-int with two flanking MT2_Mm LTRs) located in an intron of the Colgalt2 gene (Fig. 5b, top) was linked to increased expression of only those coding exons of Colgalt2 that were located 3′ of the MERVL element (Fig. 5b, middle), consistent with cis-regulation of the 3′ but not 5′ exons of Colgalt2 by the expressed MERVL TE. This latter finding was confirmed by analyzing data from previous studies^56,57, showing that early mouse embryos express this MERVL element as well as the 3′ but not 5′ exons of Colgalt2; global MERVL knockdown (KD) in early embryos resulted in reduced expression of both this MERVL element and the 3′ exons of Colgalt2 (Fig. 5b, bottom). In both examples, the MERVL regulatory regions were marked with O-GlcNAc, potentially linking loss of the modification to increased TE and gene or exon expression in Ogt iKO mES cells (Fig. 5). The MERVL elements may behave as alternative promoters, forming fusion transcripts with genes and exons that are located 3′ of the element⁵², or may function as proximal enhancers to activate 3′ genes and exons in cis.

Fig. 5 ∣. Increased TE expression is associated with 3′ transcription.

In all panels, the top four tracks show ONT-seq and RNA-seq data from Ctrl Ogt fl and Ogt iKO mES cells, followed by ChIP–seq for O-GlcNAc, DNMT1 or UHRF1 as indicated in WT mES cells. Bottom track: Hi-C PC1 data indicate euchromatin (green) or heterochromatin (blue). a, Top: genome browser view of a uniquely mapped MERVL-int/MT2 element located 5′ of the Arap2 gene in heterochromatin. ONT tracks: loss of 5mC+5hmC (ONT-seq) in Ogt iKO compared with Ctrl Ogt fl mES cells. RNA-seq tracks: increased expression of the MERVL elements in Ogt iKO compared with Ctrl Ogt fl mES cells. ChIP–seq tracks: O-GlcNAc modification of chromatin-associated proteins occupying the two MT2_Mm regulatory regions. Bottom: increased expression of the MERVL-int/MT2 element correlates with enhanced expression of Arap2. The MERVL TE and the Arap2 gene are encoded on the same strand, suggesting cis regulation of Arap2 by the MERVL TE. b, Top: genome browser view of a uniquely mapped MERVL-int/MT2 element located in an intron of the Colgalt2 gene. RNA-seq tracks: increased expression of a euchromatic MERVL-int/MT2_Mm TE in Ogt iKO versus Ctrl Ogt fl mES cells; ONT tracks: reduced 5mC+5hmC at several CpGs in Ogt iKO mES cells; ChIP–seq tracks: presence of the O-GlcNAc modification on chromatin-associated proteins occupying the two MT2_Mm regulatory regions, and occupancy of the element by Dnmt1 and Uhrf1 (mES cell ChIP–seq data from refs. 45,50). Middle: RNA-seq tracks, only the exons of Colgalt2 located 3′ of this upregulated MERVL-int/MT2_Mm element display enhanced expression in Ogt iKO compared with Ctrl Ogt fl mES cells. The MERVL TE and the Colgalt2 gene are encoded on the same strand, suggesting cis regulation of the 3′ exons of the Colgalt2 gene by the MERVL TE. Bottom: KD of MERVL-int elements is associated with reduced expression of the intronic MERVL-int TE as well as reduced expression of the Colgalt2 gene itself. Data from early embryos with MERVL-int KD as well as its control counterparts⁵⁶ were reanalyzed.

Ogt iKO mES cells showed upregulation of many genes (1,661 up, 749 down), including a large number of 2C-like genes⁴⁷ compared with Ctrl Ogtfl mES cells (Extended Data Fig. 8a). The majority of imprinted genes were also upregulated (Extended Data Fig. 8b), but genes encoding the rapidly evolving KRAB ZFP family²⁸ were downregulated in Ogt iKO compared with Ctrl Ogtfl mES cells (Extended Data Fig. 8c). OGT deletion was also associated with a striking upregulation of interferon-stimulated genes (ISGs) (Extended Data Fig. 8d), with gene set enrichment analysis revealing enrichment for inflammatory response pathways and response to viruses in Ogt iKO compared with Ctrl Ogt fl mES cells (Extended Data Fig. 8e,f). Analysis of our mass spectrometry data¹⁵ revealed a two- to threefold increase in the phosphorylation of IRF3 in Ogt iKO compared with Ctrl Ogt fl mES cells, at several sites (S123, S135, S171 and S379); values were normalized to the changes in the total protein levels of IRF3 (Extended Data Fig. 8g), and data for mTOR and TSC2, proteins found in our previous study to have roles in the growth arrest phenotype of Ogt iKO mES cells¹⁵, were included for reference. The data suggest a potential involvement of OGT in nucleic-acid sensing and antiviral response pathways, as previously noted on the basis of the ability of OGT to modify the RNA sensor MAVS^58,59.

Blocking the TET–OGT interaction results in decreased 5mC

We reanalyzed data from mES cells bearing a single-point mutation, D2018A, in the Tet1 gene¹⁹. The cells showed markedly decreased 5mC+5hmC levels in both euchromatic and heterochromatic compartments by WGBS (Fig. 6a-c), as well as at L1 5′ UTRs, MERVL MT2 and IAP LTRs, and B1 and B2 SINEs (Fig. 6d). Moreover, the reduction of global DNA methylation (5mC+5hmC) occurred in overlapping genomic windows in TET1 D2018A mutant and Ogt iKO mES cells (Fig. 6e and Supplementary Fig. 3). Thus, blocking the TET1–OGT interaction with a single-point mutation in TET1 reduces 5mC+5hmC similarly to OGT deficiency or OSMI-4 treatment (Figs. 1-3).

Fig. 6 ∣. Disrupting the TET1–OGT interaction induces DNA demethylation.

Reanalysis of WGBS data from mES cells bearing a single-point mutation (D2108A) in Tet1 that disrupts the TET1–OGT interaction¹⁹. a, Violin plots of DNA methylation (WGBS, 5mC+5hmC) in the whole genome using 10-kb windows. b, Violin plots of DNA methylation (WGBS, 5mC+5hmC) levels in euchromatin (left) and heterochromatin (right) in WT and Tet1 D2108A (Tet1-mut) mES cells. Mean values are highlighted (red line and white dot). c, Dot plot of changes in DNA methylation (WGBS, 5mC+5hmC) in 10-kb windows in WT and Tet1-mut mES cells in Hi-C A and B compartments (positive and negative PC1 values). d, DNA methylation (5mC+5hmC) levels at indicated TE subfamilies in WT and Tet1-mutant (D2018A) mES cells. e, Venn diagram showing the overlap between CpGs that lose 5mC (red) in Ogt iKO mES cells and those that lose 5mC (blue) in Tet1-mutant (D2018A) mES cells. For a similar diagram showing overlap of 10-kb windows, please see Supplementary Fig. 3. In f and g, TET1 and TET2 proteins bind largely euchromatic regions (positive or low negative PC1 values) in WT mES cells. f, ChIP–seq peaks of TET1 and TET2 from WT mES cells^48,49 were classified by their presence in euchromatin (green) or heterochromatin (blue). RNA polymerase (RNAP) was included as control for euchromatin while UHRF1 and TRIM28 are controls for heterochromatin. g, Top two tracks: genome browser view of the reanalyzed ChIP–seq data for TET1 and TET2 from WT mES cells. Bottom track: Hi-C PC1 data to identify euchromatin and heterochromatin compartments. For additional data, please see Supplementary Fig. 4. In a, b and d, WGBS data from two biological replicates were used; box plot elements: red line, median; boxes, first and third quartiles; whiskers, 1.5 × interquartile range.

Reanalysis of prior chromatin immunoprecipitation followed by sequencing (ChIP–seq) data^48,49 showed that TET1 and TET2 are located primarily in euchromatin (Hi-C A compartment) or regions of low negative PC1 values in mES cells; a relatively small fraction (12%) occupied heterochromatic regions (Hi-C B compartment) (Fig. 6f,g and Supplementary Fig. 4a), with the caveat that the resolution of our Hi-C analysis was ~50 kb whereas ChIP-seq peaks are typically ~250–400 bp. Higher-resolution mapping might show more TET1 and TET2 ChIP-seq peaks to be located in euchromatin (bodies of expressed genes and active enhancers)^9,34-36.

Identification of proteins interacting with endogenous OGT

Numerous OGT-interacting proteins have been identified, often by overexpression in various transformed cell lines²². To identify proteins that interact robustly with endogenous OGT expressed at physiological levels in mES cells, we used homology-directed repair to introduce 3x-Flag or Strep-Tag II epitope tags into the first coding exon of the Ogt gene (Fig. 7a,b). Quantitative mass spectrometry (Methods) identified many known OGT interactors: the NuRD complex, which also interacts with TET1 (refs. 16,60); HCF1, a component of MLL–COMPASS complexes¹¹; PROSER1, an adapter protein that modulates TET2 O-GlcNAcylation^61,62; the ASXL1–BAP1 deubiquitinase complex^63,64; KAP1 (ref. 45); the KAP1-binding KRAB-ZFP proteins ZFP57 and ZFP655 (ref. 65); SUZ12, EED and JARID2, subunits of the PRC2 complex; DNMT1; and USP7 (HAUSP), a deubiquitinase that controls p53/MDM2 stability^66-68 and has been implicated in regulating the stability and function of the DNMT1–UHRF1 complex^68,69 (Fig. 7c,d). Among the listed proteins, only DNMT1, TET1 and TET2 were OGT substrates as judged by O-GlcNAc modification (Fig. 7e).

Fig. 7 ∣. OGT interacts with many heterochromatin-associated proteins.

a, Strategy to generate mES cells with epitope-tagged Ogt. A sequence encoding the 3x-Flag or Strep-Tag epitope tag was knocked-in into the first exon of the Ogt gene in WT mES cells using CRISPR–Cas9-assisted homology-directed repair. b, Western blot analysis of Flag–Ogt and WT mES cells (mESC). Expression of Flag–OGT protein was detected using anti-FLAG antibodies. Actin was used as the loading control. c, Volcano plot highlighting heterochromatin-associated proteins identified by mass spectrometry following co-immunoprecipitation (co-IP) with Flag–OGT but not Strep-Tag–OGT from nuclear lysates (n = 3 biological replicates). P values were calculated using a two-tailed paired t-test. d, Diagram generated through the STRING database showing co-IP of chromatin-associated proteins with OGT. Known interactions identified in human cancer cell lines are shown in gray; new interactions identified in mES cells in this study are shown in red. e, TET1, TET2 and DNMT1 are O-GlcNAcylated (blue dots) and show decreased expression in Ogt iKO compared with Ctrl Ogt fl mES cells. f, Model illustrating the increase of TET activity in both euchromatin and heterochromatin after Ogt deletion. Ogt iKO mES cells lose O-GlcNAcylation and show a substantial decrease of DNMT1 and UHRF1. The genome-wide constraint imposed by OGT on TET activity in WT mES cells is unleashed upon OGT deletion, leading to a rapid genome-wide increase in 5hmC and concomitant decrease of 5mC in both euchromatic and heterochromatic compartments. The decrease of 5mC at the regulatory and transcribed regions of TEs, particularly those in heterochromatin, is accompanied by a striking increase in TE expression in OGT-deficient mES cells.

Reanalysis of published ChIP–seq data^16,45,48-51 showed that OGT, the O-GlcNAc modification and the TET/DNMT proteins TET1, TET2, DNMT1 and DNMT3A co-occupied the 5′ UTR regions of the young silenced LINE-1 subfamilies L1 TF_III and L1_A in WT mES cells (Supplementary Fig. 4b,c), consistent with the possibility that these proteins complex with OGT at TE regulatory regions to control or modulate TE expression. Our data confirm a broad role for OGT in repressing TE expression and sustaining the normal properties and function of heterochromatin.

Discussion

We report here that OGT globally restrains TET enzymatic activity and maintains DNA methylation genome-wide, in a manner that depends on OGT catalytic activity and the TET–OGT interaction (see model in Fig. 7f). The catalytic activity of OGT may be required for O-GlcNAcylation of the TET proteins themselves, other proteins involved in modulating TET activity, or both. TET activity was unleashed upon Ogt gene deletion or OGT inhibition, resulting in ongoing demethylation defined by parallel increases in 5hmC and decreases in 5mC at overlapping CpGs. mES cells engineered to have a single-point mutation in TET1 that abrogated the TET1–OGT interaction also showed a global loss of DNA methylation¹⁹, very similar to that observed in Ogt iKO mES cells.

The combined use of ONT-seq for long-read detection of (5mC+5hmC) and 6-base sequencing for individual assessment of 5mC and 5hmC allowed us to interrogate the precise cytosine modification status—5mC versus 5hmC—of a larger number of CpGs than would have been possible using standard WGBS (Extended Data Figs. 9 and 10). Thus, our finding of increased 5hmC in Ogt-deleted and OSMI-4-treated mES cells substantially extends a previous study showing increased 5hmC levels (by hydroxymethylated DNA immunoprecipitation-qPCR) at TET1-bound genomic regions in OGT-depleted mES cells¹⁶. However, our work is not fully consistent with two previous studies^19,20, in which the authors showed (1) that O-GlcNAcylation of a recombinant TET1 catalytic domain by recombinant OGT was associated with a four- to fivefold enhancement of TET1 activity (5mC-to-5hmC conversion)¹⁹, and (2) that OGT depletion resulted in decreased 5hmC levels at TET1-bound genomic regions in mES cells²⁰. The discrepancy with the first paper¹⁹ could reflect the involvement of modulatory cellular proteins—such as HCF1, ASXL1–BAP1, PROSER1 and others^60-69—that are not represented in assays with recombinant proteins in vitro. In the second paper²⁰, the authors showed that OGT stabilized ectopically expressed TET1 in HEK293T cells through O-GlcNAcylation at Thr535 (ref. 20), implying that decreased 5hmC in OGT-deficient cells might reflect decreased TET1 protein levels (also observed in our study) rather than any direct effect of O-GlcNAcylation on TET1 catalytic activity.

The interpretation of biological outcomes in OGT-depleted cells can be complicated by several factors. Complete loss of OGT protein or activity results in G1 arrest and eventual loss of cell viability^11,15, leading over time to outgrowth of cells with perceptible levels of OGT activity; for this reason, we culture Ogt iKO cells, which show almost complete Ogt deletion and loss of OGT activity¹⁵, in the continuous presence of 4-OHT and purify them regularly by sorting for YFP-expressing cells. Moreover, OGT has numerous substrates whose functions could be altered, conceivably in a cell-type-specific manner, by O-GlcNAcylation or other secondary posttranslational modifications that depend on OGT; as an example, treatment of hepatocyte cell lines with high glucose plus the OGA inhibitor Thiamet G, a regimen expected to increase OGT activity, resulted in a global reduction of DNA methylation (5mC+5hmC) similar to that observed in Ogt iKO cells, apparently due to increased levels of an inhibitory O-GlcNAc modification at serine 878 of DNMT1 (ref. 70). Compensatory mechanisms may come into play: for instance, loss of OGT is associated with a parallel decrease in OGA protein levels (Extended Data Fig. 3a), perhaps arising from the need to maintain balanced O-GlcNAcylation in OGT-depleted cells; similarly, the G1 arrest experienced by OGT-depleted cells sets in motion a series of cell-cycle-associated posttranslational modifications, not necessarily directly dependent on OGT, that diminish DNMT1 and UHRF1 and, perhaps in compensation, TET1 and TET2 mRNA and protein levels. To disentangle the many intertwined connections among OGT and DNA methylation and demethylation pathways, it will be necessary in future studies to explore the kinetic, biochemical and structural aspects of the TET–OGT interaction in detail. It will be important to identify O-GlcNAcylated residues in TETs, DNMTs and OGT, using recombinant proteins and in different cell types, and to conduct follow-up mutational analyses to determine whether and how each of these modifications affects TET or DNMT expression, subcellular localization, protein–protein interactions and catalytic activity.

As expected, the consequences of losing DNA methylation after Ogt deletion were apparent primarily in heterochromatin (Fig. 7f). Decreased DNA or H3K9 methylation in heterochromatin is known to compromise heterochromatin integrity, resulting in derepression of TE expression as well as other manifestations of genome instability⁷¹; and partial OGT depletion with short hairpin RNA in mES cells evoked increased expression of certain TEs as measured by quantitative reverse transcription PCR⁷². Moreover, increased TE expression in Ogt iKO cells by 4–6 days of exposure to 4-OHT was most striking in young L1 T_F and LTR elements (MERVL and IAP/ERVK), whose regulatory regions are bound by O-GlcNAcylated proteins in WT mES cells⁴⁵. Other mechanisms may also be at play: (1) TE repression, especially of evolutionarily young TEs, is also mediated by H3K9 methylation and the HUSH complex⁷³, and proteins in this complex may be modified by OGT; (2) the KAP1–OGT interaction, which requires DNMT1 (ref. 45), could be weakened by reduced heterochromatic 5mC in OGT-deficient cells; and (3) the downregulation of KAP1 and KRAB-ZFP expression that we observe in Ogt iKO mES cells could contribute to the derepression of certain TE families.

Analysis of the gene neighborhood of individual uniquely expressed TEs showed that increased TE expression in OGT-deficient mES cells correlated in certain cases with increased cis-expression of 3′ genes and exons. TEs can provide alternate transcription start sites, and TE regulatory regions can act as enhancers in cis^52-55; either mechanism could underlie the selective increase in transcription of the Arap2 gene and the exons 3′ of the MERVL element in the Colgalt2 gene. TE transcripts can also be exonized (spliced ectopically into existing transcripts)^74,75; there is a strong connection of acute OGT inhibition with changes in the levels and O-GlcNAcylation of splicing factors as well as detained intron retention⁷⁶. Thus, even in the absence of retrotransposition, increased TE expression is probably associated with altered gene expression patterns in cells.

There is emerging evidence for associations among inflammation, cancer, aging, neurodegenerative disorders, TE expression and TET–OGT function. Cancer cells^77-79, cells undergoing replicative senescence^80,81 and cells from aged individuals⁸² display large partially methylated domains, in which DNA hypomethylation occurs across megabase-sized regions corresponding to heterochromatin. Early-onset Alzheimer’s disease and frontotemporal dementia have been associated with TET2 loss-of-function variants in in large European cohorts⁸³. These and other pathological conditions—including BRCA1-mutant cancers⁷⁹, autoimmune/ inflammatory diseases⁸⁴, neurodevelopmental and neurodegenerative disorders⁸⁵ and cellular senescence^81,86—also show increased TE and satellite expression. Treatment of cancer cells with the DNMT inhibitors 5-azacytidine and decitabine—which also results in global reduction of DNA methylation—correlates with increased expression of TEs, increased levels of cytoplasmic nucleic acids and increased ISG expression^87,88. Increased TE and ISG expression, and enrichment of anti-viral response pathways, were also striking consequences of OGT deficiency (this study), and OGT was reported to support anti-viral responses through O-GlcNAcylation and subsequent ubiquitylation of the RNA sensor MAVS^58,59. The increase in IRF3 phosphorylation we observed in OGT-deficient cells is consistent with the hypothesis that the TET–OGT axis suppresses innate immune responses to TE-derived nucleic acids associated with DNA demethylation and increased TE expression, thereby countering ‘sterile’ inflammation in the absence of pathogen infection by maintaining physiological levels of DNA methylation and normal patterns of gene and TE expression across the mES cell genome.

Does the TET–OGT axis have a biochemical role in cancer initiation or progression? Notably, DNMT3A, TET2 and ASXL1 are the three most frequently mutated proteins in clonal hematopoeisis, a premalignant condition that is strongly associated with aging, inflammation and cardiovascular disease⁸⁹; and both TET2 and ASXL1 interact robustly with OGT (refs. 16-18,63,64 and this study). Thus, DNMT3A, TET2 and ASXL1, and possibly OGT itself, are probably involved in cancer initiation (that is, early premalignant stages) rather than being cancer driver proteins that are found mutated or amplified at later stages of malignancy⁹⁰. TET2 mutations occur at the earliest stages of hematopoietic malignancies, and most, if not all, TET2 mutations are loss of function⁶. In solid cancers, all three TET proteins have been reported to be mutated (cBioportal), although the frequency of these mutations is low and most of them are clearly loss of function (nonsense or frameshift mutations located early in the protein or in the catalytic domain). However, a small fraction of all TET1/2/3 mutations reside in conserved residues in the C-terminal 45 amino acids of TET proteins (the C45 peptide). The residue corresponding to D2018A is indeed among these mutations for all three TET proteins, but the variant allele frequency is not high. The OGT gene is an essential gene in DEPmap for all cell lines tested, so there are very few examples of deleterious OGT mutations in cancer, and the mutations that are observed may not necessarily be functional. Sorting out the biological importance of DNMT3A, TET2 and ASXL1 proteins in clonal hematopoiesis and the TET–OGT axis in cancer will require many more biochemical and functional analyses of these WT and mutant proteins.

Methods

Generation and culture of Ogt iKO mES cells

Ogt fl/Y (Ogt fl) male mES cells were previously generated by our group¹⁵ from Ogt^floxed; Ubc-Cre–ERT2^+/KI; Rosa26-LSLYFP^+/KI mouse blastocysts, and cultured with mitomycin-C-treated mouse embryonic fibroblasts as feeder cells using non-2i conditions: knockout DMEM medium (Thermo Fisher Scientific, 10829018) supplemented with leukemia inhibitory factor, 15% KnockOut Serum Replacement (Thermo Fisher Scientific, 10828028), GlutaMAX (Gibco, 35050061), 1× MEM nonessential amino acids (Gibco, 11140-035) and 50 μM β-mercaptoethanol (Gibco, 21985023). Ogt deletion was induced by treating Ogt fl mES cells with 1 μM 4-OHT (TOCRIS, 3412) to generate Ogt iKO mES cells, which were subsequently cultured in media containing 4-OHT to avoid overgrowth by cells that had escaped Ogt deletion. At the time points used in this study (6 days after 4-OHT treatment), Ogt iKO mES cells have undergone G1 arrest but are still fully viable¹⁵.

To generate Ogt iKO mES cells carrying the 2C reporter, we used the MERVL::tdTomato reporter⁴⁷ that was kindly shared by Samuel Pfaff (Addgene no. 40281). A stable reporter cell line was generated by transfection (Lipofectamine 2000, Thermofisher) and drug selection (hygromycin, Thermofisher).

For experiments inhibiting the catalytic activity of OGT with OSMI-4 (ref. 32) (MedChemExpressHY-114361), Ctrl Ogt fl mES cells were cultured in the presence of the inhibitor for the period of times and concentrations indicated in Fig. 3 and Extended Data Fig. 6.

Cell sorting and sample generation

Ogt fl mES cells were cultured for 6 days in the presence of 4-OHT to generate Ogt iKO mES cells. YFP⁺ cells resulting from successful Cre-ERT2-mediated recombination constituted >85% of the entire population (Extended Data Fig. 1); Ctrl Ogt fl mES cells were maintained without 4-OHT. We purified the YFP⁺ population was isolated by fluorescence-activated cell sorting (FACS) using a FACSAria cell sorter (BD Biosciences) with the FACSDiva software to ensure high sample purity and to eliminate the mitomycin-C-treated mouse embryonic fibroblasts used as feeders in our mES cell cultures. The dissociated mES cells were first labeled with the eBioscience Fixable Viability Dye eFluor 780 (Thermofisher, 65-0865-14) to select viable cells and then stained for surface markers with indicated fluorochrome-conjugated antibodies diluted in FACS buffer (1× PBS, 0.5% BSA and 0.01% NaN₃): CD90.2 (BioLegend, 140319) to positively distinguish the mES cells and CD326 (BioLegend, 118227) to negatively exclude mouse embryonic fibroblasts.

RNA-seq

Total RNA was isolated from Ogt iKO and Ctrl Ogt fl mES cell samples using RNeasy plus mini or micro kit (Qiagen). RNA-seq libraries were prepared using the Truseq stranded mRNA kit (Illumina) according to the manufacturer’s protocol. All libraries were assessed using Qubit RNA HS Assay Kit (Thermofisher, Q32855) and TapeStation High Sensitivity RNA ScreenTape Analysis (Agilent, 5067-5579). Libraries were pooled in equal quantity and sequenced using Illumina HiSeq 2500 (Illumina) to produce 20 million paired-end reads (50 × 50 bases) per each RNA-seq library.

WGBS

Genomic DNA was isolated using PureLink Genomic DNA Mini Kit (Thermo Fisher Scientific, K182001) or FlexiGene DNA Kit (Qiagen, 51206). Unmethylated lambda DNA (Promega, D1521) was spiked into the genomic DNA samples at a ratio of 1:200 to monitor the bisulfite conversion efficiency. The samples were then sheared using Covaris S2, purified with Ampure XP beads (Beckman Coulter), processed with NEBNext End Repair and dA-Tailing Modules (NEB) and ligated to methylated Illumina Adaptors using NEBNext Quick Ligation Module (NEB). DNA with ligated adaptors was then treated with sodium bisulfite (MethylCode, Thermo Fisher Scientific) for 4 h and amplified with index primers using KAPA HiFi HotStart Uracil+ Ready Mix (KAPA Biosystems). The samples after PCR amplification were then purified using Ampure XP Beads (Beckman Coulter) and sequenced as 125- or 250-base-pair paired-end reads using Illumina Hiseq 2500 (Illumina).

ONT-seq

After cell sorting, high-molecular-weight genomic DNA was isolated from Ogt iKO and Ctrl Ogt fl mES cells using Nanobind CBB Kit (Circulomics, NB-900-001-01) according to the manufacturer’s instructions. DNA libraries were prepared at the Kinghorn Centre for Clinical Genomics (Australia) using 3 μg input DNA, without shearing, and an SQK-LSK110 ligation sequencing kit. Libraries were each sequenced separately on a PromethION (Oxford Nanopore Technologies) flow cell (FLO-PRO002, R9.4.1 chemistry). Bases were called with guppy 5.0.13 (Oxford Nanopore Technologies).

CMS-IP

CMS-IP to immunoprecipitate 5hmC^6,30,34-36 was performed using 3 μg of genomic DNA as starting material. After library preparation, Illumina sequencing was used to generate 30 million reads from immunoprecipitated CMS-IP samples as well as from input samples. Equal amounts of a PCR amplicon generated to contain 5hmC (5hmC spike-in) were added to all samples for internal calibration.

Simultaneous detection of 5hmC and 5mC by duet multiomics solution evoC (biomodal)

The duet multiomics solution evoC method (6-base sequencing³¹) was performed according to the manufacturer’s instructions, using 80 ng of genomic DNA as starting material from YFP⁺ Ogt iKO cells after tamoxifen treatment, and control YFP-negative Ctrl Ogt fl mES cells sorted without prior treatment with tamoxifen.

Flow cytometry

mES cells were trypsinized into single cells and used for staining or cell sorting. For 5hmC staining, the genomic DNA was denatured by treating the fixed mES cells with 2 N HCl for 30 min. After the removal of HCl, the pH was neutralized with 100 mM Tris–HCl (pH 8.5) for 10 min at room temperature. For staining, we used Active Motif #39791 as the primary antibody (anti-5hmC) and Thermofisher #A32795 as the secondary antibody (IgG).

qRT–PCR

qRT–PCR was performed using Universal SYBR Green Master Mix (Roche) and analyzed using a Step One Plus real-time PCR system (Applied Biosystems) according to the manufacturer’s instructions, and the data were normalized for Gapdh expression. The primers used for qRT–PCR are: MT2/MERVL: forward, CTC AAG GCC CAC CAA TAG TTT C, reverse, CCC ATGT CAA TAA ACT CAG CCT G; LINE/L1: forward, GGA CCA GAA AAG AAA TTC CTC CCG, reverse, CTC TTC TGG CTT TCA TAG TCT CTG G; SINE/B1: forward, GTG GCG CAC GCC TTT AAT C, reverse, GAC AGG GTT TCT CTG TGT AG; GAPDH: forward, GTC GTG GAG TCT ACT GGT GTC, reverse, GAG CCC TTC CAC AAT GCC AAA.

Bioinformatic analyses

For RNA-seq data, reads were aligned using STAR version 2.7.11 (ref. 91) with the parameters --outFilterMultimapNmax 1, --outSAMtype BAM SortedByCoordinate --sjdbOverhang 100 to analyze gene expression. HT-Seq⁹² was used to quantify the gene expression levels using the options htseq-count -s yes, -r pos, -a 10. Normalization and differential expression analyses were performed using DESEq2 (ref. 93), with the parameters fitType parametric, alpha 0.05 and using the Benjamini–Hochberg method. For visualization of the data, we generated tracks using Deeptools⁹⁴ version 3.5.1, with the option bamCoverage. All related plots were made using R-Studio⁹⁵ and Integrative Genome Viewer⁹⁶ version 2.16.0. To measure the expression of TEs, we used the TE-transcript package⁹⁷ version 2.2.3 with the option -multi to use ambiguously mapped reads and then perform the differential expression analysis with DESEQ2 (false discovery rate cutoff of P < 0.05). Genes or TEs with fewer than ten reads total were prefiltered in all comparisons as an initial step. In all our analyses, we used the mm10 genome version, and for analysis of TEs, we specifically used the mm10 repeatMasker version from TE-transcript⁹⁷. For detection of individual TEs we used the STAR aligner with the parameter --outFilterMultimapNmax 1 to then count the amount of uniquely aligned reads using htseq-count against the customized mm10 version from Hammell laboratory as the genome reference.

For ChIP–seq datasets, we used Bowtie⁹⁸ version 2.4.0 for alignments and Deeptools⁹⁴ with the option bamCoverage for generation of the genome tracks. Heatmaps and profiles were generated using computeMatrix with the option scale-regions and plotHeatmap, both from Deeptools.

A Hi-C dataset from WT mES cells⁹⁹ was used for genome compartmentalization. We estimated the relationship between the number of TE copies contained in euchromatin (positive PC1 values) versus heterochromatin (negative PC1 values) using a logarithmic transformation, using the formula log₂((number of elements in euchromatin)/(number of elements in heterochromatin)); the resulting values are plotted on the y axis, while the total number of copies found in the genome (mm10) are shown on the x axis. Positive values on the y axis (upper side of the plot) show TE subfamilies whose copies are enriched in euchromatin, while negative values (bottom side of the plot) show TE subfamilies whose copies are enriched in heterochromatin. Enriched TEs were defined as those having log₂ ratio >0.5 or <0.5.

For WGBS samples, we used Bismark¹⁰⁰ version 0.9.0 and bedGraphToBigWig for track generation. The median coverage for each sample was ~20×.

For WGBS, as well as for ONT and 6-base sequencing data, we used CpGs covered by five or more reads for the analyses. In addition, for analysis using windows of 10 kb, we included only those regions containing more than five CpGs.

For ONT data, reference TE methylation was assessed for mES cell samples aggregated by condition using Methylartist¹⁰¹ version 1.2.482. In brief, CpG methylation calls were generated from ONT reads using nanopolish version 0.13.2118. Using Methylartist commands db-nanopolish, segmeth and segplot with default parameters, methylation statistics were generated for the genome divided into 10-kb bins, protein-coding gene promoters defined the Eukaryotic Promoter Database (−1,000 bp, +500 bp) 119 and reference TEs defined by RepeatMasker coordinates deposited in the TE-transcript package⁹⁷. The median coverage for each sample was ~10×.

For 6-base sequencing, data were processed as in ref. 31 and the median coverage for each sample was ~15×. For CMS-IP including spike-ins^30,102, immunoprecipitation and input samples were aligned against the mm10 genome reference using Bismark¹⁰⁰ version v0.15.0. Reads aligning to a PCR amplicon generated to contain 5hmC (5hmC spike-in) that was equally present across samples were used for internal calibration¹⁰². Peak calling was performed with MACS2, using input DNA to compare against the CMS-IP samples. Peaks from different replicates were merged using bedtools.

Western blots

Whole-cell extracts were prepared by incubating mES cells with radioimmunoprecipitation assay buffer (Thermo Fisher, 89900) supplemented with Benzonase (Sigma-Aldrich, E1014-25KU) and Halt Protease/Phosphatase Inhibitor Cocktail (Thermofisher, 78441) on ice for 1 h. Proteins from whole-cell extracts were resolved using NuPAGE 4–12% bis-tris gel (Thermo Fisher Scientific, NP0321BOX) and transferred onto polyvinylidene difluoride membranes using Wet/Tank Blotting Systems (Bio-Rad). Polyvinylidene difluoride membranes were blocked with 5% nonfat milk in PBST (PBS 1× and 0.05% Tween-20) and incubated with the indicated primary antibodies, followed by secondary antibodies conjugated with horseradish peroxidase (Cell Signaling, 7076 or 7074; dilutions 1:4,000 and 1:5,000, respectively), all of which were diluted in 5% nonfat milk in PBST. Signal was detected with Femto Supersignal ECL substrate (Thermofisher, 34096) and X-ray films (Thermofisher, 34090). The primary antibodies and their dilutions are as follows: anti-Ogt (ab96718; dilution 1:4,000); anti-O-Linked GlcNAc antibody [RL2] (ab2739; dilution 1:4,000); anti-Tet1 (Millipore-Sigma 09-872; 1:4,000); anti-Tet2 (Abcam, ab124297; 1:4,000); anti-Dnmt1 (Abcam, ab19905; 1:4,000); anti-Dnmt3a (Abcam, ab2850; 1:3,000); anti-Dnmt3b (Abcam, ab122932; 1:4,000); anti-Uhrf1 (Santa Cruz, sc-373750; 1:8,000); anti-KAP1 (Abcam, ab10483; 1:8,000); β-Actin (Cell Signaling, 5125; 1:10,000); anti-5hmC (Active Motif, 39769; 1:1,000); anti-Flag M2 (Millipore-Sigma, F3165; 1:3,000).

Generation of endogenously Flag-tagged OGT (Ogt–Flag) mES cells and sample generation

To identify nuclear proteins interacting with endogenous OGT, we generated mES cell lines with 3x-Flag or StrepTag epitope tags introduced by CRISPR–Cas9 homology-directed repair into the first coding exon of the endogenous Ogt gene. Nucleofection with plasmid containing the repair template, an in vitro transcribed sgRNA (EnGen, NEB), and recombinant CAS9 was done using Lonza’s P1 kit, following the manufacturer’s recommendations. Antibiotic selection was performed by culturing the transfected mES cells with 1 mm puromycin (Thermo Fisher, A1113803) for 2 weeks. Stable Ogt-edited mES cells were confirmed by detecting the expression of the OGT-tagged versions by western blot.

Mass spectrometry

We identified proteins that co-immunoprecipitated under native conditions with the anti-Flag antibody from nuclear extracts of the Flag–OGT but not the StrepTag–OGT mES cell line. For sample generation, approximately 30 million epitope-tagged OGT mES stable cell lines were trypsinized and washed twice with cold 1× PBS before placing on ice. Subcellular fractionation into nuclear and cytoplasmic fractions was done using Thermofisher NE-PER Nuclear and Cytoplasmic Extraction Kit (78833), supplemented with protease inhibitor cocktail (Thermo 78429), as described in the manufacturer’s protocol. Protein quantification was done using Bradford colorimetric assay (Bio-Rad 5000006). Anti-FLAG M2 magnetic beads (Sigma M8823) were washed with radioimmunoprecipitation assay buffer before adding 1 μl beads per 1 mg protein to protein samples and undergoing continuous inversion at 4 °C overnight. Proteins were processed and analyzed with minor modifications¹⁰³. Samples were analyzed on an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific).

Mass spectrometry data analysis

Raw data were processed using MaxQuant software¹⁰⁴ version 2.3.1 using the default settings and searched against the mouse UniProt database (12/2017) with common contaminant entries. The settings used for MaxQuant analysis were as follows: enzymes set as LysC/P and Trypsin/P, with maximum of two missed cleavages; fixed modification was carbamidomethyl (Cys); variable modifications were acetyl (protein N-term) and oxidation (Met); label-free quantification (LFQ) was used with a minimum ratio count of 2 and classic normalization; false discovery rate for both protein and peptide identification was 0.01. The ‘match between runs’ feature was enabled. The proteinGroups file from the MaxQuant analysis was used as the input for all downstream analyses. Gene names that were not automatically assigned by MaxQuant were manually added from their Entrez protein IDs. Downstream analysis was performed using Perseus version 2.0.7.0. Among the 2,617 entries in the proteinGroups file, we filtered out 33 entries identified as contaminants. Each of the three replicates within a condition (Flag-tagged or Strep-Tag II-tagged/control) were annotated. LFQ intensities were log₂ transformed and proteins with fewer than two LFQ intensity values across both conditions were filtered out. Missing values were imputed with the minimum detected intensity. A two-tailed t-test with a P-value cutoff of 0.05 and false discovery rate of 0.01 was conducted alongside the calculation of log₂ fold changes for construction of the volcano plot.

Reporting summary

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

Extended Data

Extended Data Fig. 1 ∣. Isolation and 5mC+5hmC quantification for Ogt iKO mESC.

A,B, Flow cytometry gating scheme for analysis of sorted Ctrl Ogt fl (A) and Ogt iKO (B) mESC. The sorting strategy includes successive staining for Live/ Dead; Thy1.2 to eliminate MEFs – mouse embryonic fibroblasts – in the feeder layer; and EpCAM to detect mESC. After 6 days of exposure to 4-OHT, Ogt fl cells yielded >70% YFP⁺ Ogt iKO mESC in which Cre had been activated; these cells, as well as YFP-negative cells from Ctrl Ogt fl mESC that had not been exposed to 4-OHT, were sorted and used for further analyses. C, Left, Genome browser view of RNA-seq data from the Ogt locus showing the considerable decrease of Ogt mRNA in Ctrl Ogt fl compared to Ogt iKO mESC. The ActB locus was used as control (right). D, Density distribution plot of average 5mC+5hmC (WGBS) values in 10 kb windows across the genome, showing loss of 5mC+5hmC in both euchromatin (A-comp, blue) and heterochromatin (B-comp, red) genomic compartments of Ogt iKO (dashed line) compared to Ctrl (solid line) mESC. D stat and p-values were calculated using the two-sample Kolmogorov-Smirnov test, with the ks.test function on R.

Extended Data Fig. 2 ∣. Increased 5hmC levels in Ogt iKO mESC.

A, Left, DNA dot blot assay using antibodies to the 5hmC derivative cytosine-5-methylenesulfonate (CMS) shows increased global 5hmC levels in Ogt iKO compared to Ctrl Ogt fl mESC. Data are representative of four independent experiments. Right, Quantification of relative dot intensity. There is a very reproducible ~1.6-fold increase in 5hmC in Ogt iKO compared to Ctrl Ogt fl mESC, using all three assays described below. B, Left, 5hmC levels in Ogt iKO and Ctrl Ogt fl mESC assessed by flow cytometry with an anti-5hmC antibody. A representative histogram is shown. Right, Quantification of relative mean fluorescence intensity (MFI) of 5hmC staining in 3 independent experiments in Ogt iKO compared to Ctrl Ogt fl mESC. C–E, 5hmC genomic distribution assessed by CMS-IP in Ogt iKO and Ctrl Ogt fl mESC. The CMS-IP samples from two independent experiments included Spike-in controls for internal normalization. C, Left, Genome browser view of 5hmC levels in Ogt iKO and Ctrl Ogt fl mESC. Right, Quantification of normalized reads in CMS-IP peaks. D, Dot plot showing increased 5hmC (quantified as normalized CMS-IP reads) in 5hmC-rich regions in Ctrl Ogt fl and Ogt iKO, plotted in euchromatin (Hi-C A) and heterochromatin (Hi-C B) compartments. Each dot represents a hydroxymethylated region (hmRs). E, Heatmaps of differentially hydroxymethylated regions (DhmRs) in Ogt iKO compared to Ctrl Ogt fl mESC. In A–C, the bar graphs represent the mean +/− SD; p-values were calculated using a two-tailed paired t-test (**<0.01).

Extended Data Fig. 3 ∣. Decreased DNMT1/UHRF1 levels in G1-arrested Ogt iKO mESC.

A–C, Changes in the levels of selected proteins in Ogt iKO compared to Ctrl Ogt fl mESC shown by A, volcano plot of mass spectrometry data²⁹ and B, western blotting (with actin used as loading control). Note the substantial decrease in DNMT1 and UHRF1 caused by the G1-arrest of Ogt iKO mESC²⁹. TET2, DNMT3A1/2, KAP1 and TET1 proteins showed a moderate decrease while DNMT3B protein levels show a substantial increase. C, normalized RNA expression of Dnmt1, Uhrf1, Tet1 and Tet2 in Ogt iKO compared to Ctrl Ogt fl mESC, assessed by RNA-seq. Bar graphs represent the mean +/− SD from two independent biological replicates. D–G, Decreased levels of DNMT1 and UHRF1 in G1-arrested mESC. Ctrl Ogt fl mESC were treated with the indicated concentrations of the CDK4/6 inhibitor Palbociclib for 3 days (D), and protein levels of DNMT1 and UHRF1 in cell lysates were assessed by western blotting (E). O-GlcNAc (F) and 5hmC (G) levels were measured by flow cytometry from two independent replicates. G1 arrest results in DNMT1 and UHRF1 degradation independently of changes in the global levels of O-GlcNAc and 5hmC.

Extended Data Fig. 4 ∣. 6-base sequencing of Ogt iKO and Ctrl Ogt fl mESC.

A,B, The increase of 5hmC and decrease of 5mC in Ogt iKO compared to Ctrl Ogt fl mESC occurs in the same genomic regions. 6-base seq was generated from two independent experiments. A, Genome browser view showing the reciprocal gain of 5hmC (red) and loss of 5mC (blue) in Ogt iKO compared to Ctrl Ogt fl mESC. Plotted are log2 ratios (Ogt iKO/Ctrl Ogt fl) for 5hmC (top) and 5mC (bottom). B, Venn diagram showing the overlap between the 10 kb windows that gain 5hmC in (blue) and those that lose 5mC (red) in Ogt iKO compared to Ctrl Ogt fl mESC. DNA modification values were averaged over all adequately covered CpGs in the 10 kb windows; only windows with 5 or more CpGs were considered. C, Boxplots of differences in 5hmC levels (Ogt iKO – Ctrl) at the 5′ UTR of the indicated LINE-1 families. Box indicates interquartile range with whiskers +/−1.5 times this range. Only 5′ UTRs containing ≥10 detected CpGs with 20x minimum coverage were included in the analysis. D, 5hmC levels in CpG dinucleotides located within the 5′ UTR of the indicated LINE-1 families in Ctrl Ogt fl and Ogt iKO mESC, as measured by 6-base-seq; difference of median 5hmC levels between conditions (Ogt iKO – Ctrl) is displayed on top. Only CpGs with 20x minimum coverage are included in the analysis. p-values were calculated using a two-tailed paired t-test. E, 5hmC levels averaged across the entire transcribed units of individually mapped TEs belonging to the indicated TE subfamilies and showing increased expression in Ogt iKO compared to Ctrl Ogt fl mESC. Like expressed genes which have high 5hmC in their gene bodies, expressed TEs are enriched for 5hmC in their transcribed regions as well as in their regulatory regions (Fig. 4f). In C–E, Box indicates the median, interquartile range with whiskers +/−1.5 times this range.

Extended Data Fig. 5 ∣. Time-dependent increase of 5hmC after Ogt deletion in mESC.

A, Flowchart of experiments. B, 5hmC levels (assessed by flow cytometry) increase in a large fraction of Ogt iKO cells by day 3 (see shoulder in red histograms); by day 6, the entire population of Ogt iKO cells (red) shows high 5hmC levels compared to Ctrl Ogt fl cells (blue). C, quantification of data from the two independent experiments in B. D, 3 days after Ogt deletion, Tet1, Tet2, Dnmt1 & Uhrf1 mRNA levels show only a slight decrease. In C and D, bar graphs represent the mean +/− SD from two independent experiments.

Extended Data Fig. 6 ∣. OGT inhibition results in increased 5hmC.

A, O-GlcNAc levels in mESC treated with or without the OGT inhibitor, OSMI-4, were assessed by flow cytometry. Histograms show the reduction of O-GlcNAc with time. The corresponding bar graph is shown in Fig. 3b. B, Quantification of relative mean fluorescence intensity (MFI) of O-GlcNAc staining after 1 and 2 days of OSMI-4 treatment. For a representative histogram see Fig. 3c. Bar graphs represent the mean +/− SD from three independent experiments. C, Histograms showing the increase of 5hmC observed after treatment with different doses of OSMI-4 (light and dark blue) for 4 days. Data from Ogt iKO mESC were included for reference. D, Venn diagram of 6-base sequencing data showing the overlap of averaged DNA modification values in 10 kb windows gaining 5hmC in Ogt iKO compared to Ctrl Ogt fl mESC (orange) and those gaining 5hmC in OSMI-4-treated compared to untreated mESC (‘OGT-inh’; green).

Extended Data Fig. 7 ∣. TEs regulated by OGT are enriched at heterochromatic regions.

A, Increased expression of MT2/MERVL and LINE1/L1 in Ogt iKO compared to Ctrl Ogt fl mESC, assessed by RT-qPCR. B, Increased detection of peptides from LINE L1, MERVL-int and IAP elements in Ogt iKO compared to Ctrl Ogt fl mESC. Mass spectrometry samples of whole-cell lysates from three independent experiments were analyzed against a reference dataset of proteins encoded in TEs (see Methods). C, Increased tdTomato fluorescence in Ogt iKO mESC transfected with a tdTomato reporter driven by the MERVL-MT2 LTR. Upper left, schematic of the MERVL::tdTomato reporter; lower left, representative flow cytometry histograms, Right, quantification of mean fluorescence intensity (MFI) in three independent experiments. The MERVL LTR reports on the 2C-like state in mESC. D, O-GlcNAc enrichment at chromatin-associated proteins located within TEs, based on O-GlcNAc ChIP-seq⁴⁵. O-GlcNAcylated TEs were defined as more than 50% of enrichment (IP/Input) and p-Adjusted value < 0.05 (using the Benjamini-Hochberg test) from three independent replicates. E, Presence of TE subfamilies in euchromatin or heterochromatin domains in WT mESC. Top, For each subfamily of TEs (represented as a single dot), we counted the number of individual TE copies contained in euchromatin (positive Hi-C PC1 values) or in heterochromatin (negative Hi-C PC1 values). The log2 ratio of the number of TE copies contained in euchromatin versus heterochromatin are plotted on the y-axis while the total number of copies found in the genome (mm10) are shown on the x-axis. Enriched TEs, log2 ratio > or < +/− 0.5. Color codes, TE classification by subfamilies (shown at top). Gray dots, TE-subfamilies with no preferential presence in euchromatin or heterochromatin. Bottom, total number of TE subfamilies enriched in euchromatin or heterochromatin by classes of TEs (LINEs, LTRs, SINEs, DNA-transposons, and Satellites). F, TEs containing high levels of the O-GlcNAc modification are enriched in heterochromatin. Ratios of TEs in heterochromatin versus euchromatin are shown for the whole mouse genome (black bar), for all TEs (white bar) and for those TEs that are enriched in O-GlcNAc (as shown in Fig. 5a and Supplementary Fig. 8; purple tracks). In A–C, bar graphs represent the mean +/− SD from three independent experiments; p-values were calculated using a two-tailed paired t-test (**<0.01).

Extended Data Fig. 8 ∣. Transcriptional signatures of Ogt iKO mESC.

A–D, Scatter plots comparing expression of A, genes related to the 2C-like state; B, imprinted genes; C, Krüppel-associated box domain (KRAB) zinc finger (ZNF) genes; and D, Interferon-stimulated genes (ISGs) in Ogt iKO versus Ctrl Ogt fl mESC. E,F, Gene Set Enrichment Analysis (GSEA) comparing Ogt iKO versus Ctrl Ogt fl mESC. Enriched pathways in Ogt iKO mESC include E, Inflammatory response and F, Response to virus. Normalized enrichment scores (NES) and Nominal p-values were calculated for each group¹⁰⁵. G, IRF3 phosphorylation in Ogt iKO mESC. Phosphorylation ratios for the indicated proteins at the indicated residues were extracted from our earlier mass spectrometry analysis of Ogt iKO mESC¹⁵. Phospho-signals were normalized to changes in total protein levels (see associated Table).

Extended Data Fig. 9 ∣. Comparison of ONT, WGBS and 6-base sequencing datasets.

A, Percentage and number (in parenthesis) of CpGs identified in the Hi-C-defined euchromatin and heterochromatin compartments in mouse embryonic stem cells using the mm10 genome annotation (1^st pie chart); CpGs identified by Oxford Nanopore Technologies (ONT) long-read sequencing (2^nd pie chart), by whole-genome bisulphite sequencing (WGBS, 3^rd pie chart), and by biomodal evoC 6-base sequencing (4^th pie chart). While all technologies identify CpGs with a similar euchromatin-heterochromatin proportion, ONT and evoC 6-base seq identify a larger number of CpGs than WGBS. B, Percentage of CpGs identified at 1x minimum coverage in gene regulatory features such as CpG islands, promoters, and distal and proximal enhancers in the mouse genome mm10, by ONT long-read sequencing, by WGBS, and by biomodal evoC 6-base sequencing. While ONT and evoC 6-base seq are able to detect most CpGs within these regions, our observations of poor coverage at CpG islands and promoters by WGBS are consistent with previous reports of WGBS displaying an A/T positive-bias and underrepresentation of C/G-rich regions¹⁰⁶. C, Percentage of CpGs identified at 1x minimum coverage in the 5′ UTR regulatory regions of LINE-1 transposable elements in the mouse genome mm10, by ONT long-read sequencing, by WGBS, and by biomodal evoC 6-base sequencing. While ONT long-read sequencing is able to detect most CpGs within these repetitive elements over short-read sequencing-based methods (that is, WGBS and evoC 6-base seq), evoC 6-base seq outperforms WGBS, and displays a better detection performance at older elements (such as L1MdF) than in younger elements (such as L1MdTFI and L1MdTFII)⁴⁴, presumably since older elements have accumulated mutations that facilitate short-read mapping.

Extended Data Fig. 10 ∣. Table comparing ONT, WGBS and 6-base sequencing datasets.

Numbers and percentages of CpGs detected at 1x minimum coverage by Oxford Nanopore Technologies (ONT) sequencing, by whole-genome bisulphite sequencing (WGBS), and by biomodal evoC 6-base sequencing, in the Ctrl Ogt fl and Ogt iKO mES cells datasets generated in this study.

Data availability

All RNA-seq, WGBS and CMS-IP data have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE252760. 6-base sequencing data were deposited under the GEO accession code GSE247534. ONT sequencing data were deposited in the European Nucleotide Archive (ENA) under identifier PRJEB67735. Previously published WGBS data from Tet1D2018A mutant cell line used for comparisons are available via GEO under accession code GSE119666. ChIP–seq and Hi-C datasets analyzed in this study are available via GEO under accession codes GSE39154, GSE93539, GSE57700, GSE115972, GSE158460, GSE118785 and GSE96107. Source data are provided with this paper.