O-GlcNAcylation of co-activator-associated arginine methyltransferase 1 regulates its protein substrate specificity

Co-activator-associated arginine methyltransferase 1 (CARM1), also known as protein arginine methyltransferase4 (PRMT4), is a type I PRMT that asymmetrically dimethylates arginine on target proteins. CARM1 was first identified as an interacting partner of the glucocorticoid receptor interacting protein 1 (GRIP1) co-activator protein that enhanced transcriptional activity of several steroid hormone receptors [1]. Subsequent studies showed that CARM1 activates many cancer-relevant transcription factors including PPARγ (peroxisome proliferator-activated receptor γ), p53, NF-κB (nuclear factor kappa-light-chain-enhancer of activated B-cells), E2F1 (E2F transcription factor 1) and β-catenin [1–5]. The methyltransferase activity of CARM1 has been extensively studied. CARM1 methylates and modulates the functions of a plethora of proteins. The first CARM1 substrate identified was histone H3 [1]. CARM1-specific histone H3 methylation at Arg¹⁷ has been linked to gene activation and is considered part of the ‘histone code’ [6]. Other substrates of CARM1 include BAF155 (BRG1-associated factor 155), SmB (small nuclear ribonucleoprotein polypeptide B), SAP49 (splicing factor 3B subunit 4), U1C (U1 small nuclear RNP specific C), CA150 (co-activator of 150 kDa), HuR (Hu antigen R), HuD (Hu antigen D), TARPP (thymocyte cAMP-regulated phosphoprotein) and PABP1 [poly(A)-binding protein 1], many of which are involved in mRNA processing and transcription elongation [7,8]. The ability of CARM1 to activate many transcription factors and methylate diversified substrates allows it to orchestrate a broad spectrum of biological processes. However, little is known about how the methyltransferase activity and substrate specificity of CARM1 are regulated.

Proteins in the PRMT family share a common domain organization. The core region contains a methyltransferase catalytic site, methyl donor S-adenosyl methionine (SAM)-binding pocket and dimerization arm. Although the catalytic cores are highly conserved among PRMTs, CARM1 possesses a unique C-terminus [9]. This C-terminal region is not essential for CARM1's methyltransferase activity in vitro; however, it exhibits an autonomous transcriptional activation function whose mechanism remains unclear. The C-terminal region can also mediate protein–protein interaction. For example, TIF1α (transcription intermediary factor 1 α)/TRIM24 (tripartite motif-containing 24) was reported to interact with this domain [10].

O-GlcNAcylation (O-linked-β-N-acetylglucosaminidation), found on >1000 nuclear and cytoplasmic proteins, is characterized by the addition of N-acetyl-D-glucosamine to serine and threonine residues of target proteins [11–14]. O-GlcNAcylation is a reversible post-translational modification (PTM) governed by two enzymes: O-GlcNAc (O-linked-β-N-acetylglucosamine) transferase (OGT) [15,16] and O-GlcNAcase (OGA) [17] (reviewed in [18–21]). UDP–GlcNAc, the GlcNAc donor molecule, is generated in the hexosamine biosynthesis pathway. Protein O-GlcNAcylation has been shown to regulate diverse protein functions including stability, localization, protein–protein interaction, transcriptional activity, enzymatic activity and substrate specificity (reviewed in [18,22–24]).

Using top-down MS, we found that nearly 100% of CARM1 is automethylated in vivo and that 50% of this automethylated CARM1 exhibits a 203 Da mass shift, indicative of mono-O-GlcNAcylation. We mapped Ser⁵⁹⁵, Ser⁵⁹⁸, Thr⁶⁰¹ and Thr⁶⁰³ as major O-GlcNAcylation sites located in the C-terminus of CARM1 protein, implying that CARM1 could exist as a mixture of mono-O-GlcNAcylated forms. O-GlcNAcylation of CARM1 did not appear to alter stability, nuclear–cytoplasmic distribution, dimerization capability and co-activator activity of CARM1 on a few tested transcription factors. However, O-GlcNAcylation of CARM1 affects its substrate specificity. Our findings reveal that O-GlcNAcylation of the CARM1 C-terminus regulates substrate specificity and thus may affect a variety of CARM1 functions depending on substrate methylation.

Full-length mouse CARM1 cDNA was cloned into Halo–tag vector, pFC14K (Promega), as described previously [25]. CARM1^S595A, CARM1^S598A, CARM1^T601A, CARM1^T603A and CARM1^QM (CARM1 quadruple mutant) were generated by site-directed mutagenesis using pFC14K–CARM1 construct as a template. CARM1^ΔCTD (C-terminal domain deleted CARM1) was generated by PCR cloning of mouse CARM1 encoding amino acid 1–553 into a pFC14K Flexi vector. The expression and purification of Halo–tagged recombinant CARM1 were performed as previously described [25].

The high-resolution FT-ICR (Fourier transform-ion cyclotron resonance) MS analysis of Halo–tag-purified CARM1 was performed as described previously [26].

Purified CARM1 protein was digested with endoprotease Lys-C (Promega) for FT-ICR MS analysis. Purified recombinant CARM1 (20 μg) was incubated with 200 ng of Lys-C for 1 h at 37°C in 50 mM NH₄HCO₃ and 10% acetonitrile solution at pH 8. Subsequently, the mixture was desalted using a 10 kDa NMWL Amicon Ultra (Millipore). Samples were injected using nano-ESI in vehicle containing 50% methanol and 1% acetic acid through a Triversa Nanomate injector (Advion). For fragmentation analysis, a single charged state was isolated and dissociated by CAD (collisionally-activated dissociation) using 12–16% collision energy or ECD (electron-capture dissociation) at 3% activation energy for 50 ms.

A total of 45 ng of purified OGA (PRO-E0255, lot 2012–0255, Prozomix) was used per 5 μg of purified CARM1. The reaction was incubated at 37°C for 4 h.

Recombinant CARM1 (200 ng) was incubated with substrates (500 ng of purified substrates or 10 μg of cell lysate) in buffer containing 5 mM MgCl₂, 20 mM HEPES, pH 7.9, 1 mM EDTA, 1 mM DTT, 10% glycerol containing 2 μl of [³H] S-adenosylmethionine (10 Ci/mmol, Perkin Elmer) for 1–6 h. The reaction was then resolved by SDS/PAGE and stained with Coomassie Brilliant Blue [0.05 (w/v):50:10:40; Coomassie Brilliant Blue R-250/methanol/acetic acid/H₂O] overnight followed by destaining with dye-free buffer for 2–4 h. Gels were then incubated with Amplify scintillation fluid (GE Healthcare Life Sciences) for 20 min prior to drying on a Whatman paper and exposing to an X-ray film. For 2D electrophoresis, 2.5 μg of purified CARM1 was incubated with 500 μg of Cal51^CARM1−/− cell lysates. The reaction was resolved using isoelectric focusing pH 4–8. 2D electrophoresis was performed by Kendrick Labs.

Enrichment of O-GlcNAcylated CARM1 was adapted from the protocol described by Zachara et al. [27]. Purified CARM1 (20 μg) was incubated with 100 μl of WGA (wheat germ agglutinin) agarose bead (Vector Labs) for 1 h on ice. After centrifugation, the flow-through fraction was collected. The column was then washed with WGA wash buffer (25 mM Tris, 300 mM NaCl, 5 mM CaCl₂ and 1 mM MgCl₂) followed by elution with WGA elution buffer (0.2% v/v Nonidet P40 and 1 M N-acetyl-D-glucosamine).

CARM1-null human embryonic kidney (HEK293T), MCF7 (ERa positive breast cancer cell line) and MDA-MB-231 (triple-negative breast cancer cell line) cell lines were described in [28]. CARM1-null Cal51 cell lines were generated using the same method as described in [28]. Flag–tagged mouse CARM1^WT (CARM1 wild-type) or CARM1^QM was cloned into pCDH-lentiviral vector and pBABE retroviral vector. pCDH viral vector was used to transduce recombinant CARM1 into MDA-MB − 231^CARM1−/− cells, whereas pBABE viral vector was used to transduce CARM1 into MCF7^CARM1−/− cells. Cells were then selected in medium containing 2 μg/ml puromycin for 1 week and pooled cells were used in the experiments.

For CARM1-mediated activation of Sertad, NFκBIB (nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, beta) and p53, HEK293T^CARM1−/− cells, seeded on a 48-well plate, were transiently transfected with GAL4–Luc (luciferase reporter vector driven by a GAL4 responsive element) reporter plasmid (20 ng), Renilla luciferase plasmid (20 ng), GAL4-fused designated transcription factor in pCMX vector (20 ng) and Flag–tagged CARM1^WT or Flag–tagged CARM1^QM in pCMX vector.

For autonomous activation activity of full-length CARM1, HEK293T^CARM1−/− cells, seeded on a 24-well plate, were transiently transfected with GAL4–Luc reporter plasmid (50 ng), Renilla luciferase plasmid (50 ng) and pM vector encoding Gal4-fused CARM1^WT or Gal4-fused CARM1^QM (200 ng).

Empty vectors of the respected constructs were used to adjust total amounts of plasmids. Proteins were allowed to express for 2 days. Cells were then lysed with buffer containing 100 mM potassium phosphate and 0.2% Triton X-100. All experiments were performed in triplicate. Luciferase signal was normalized to Renilla luciferase internal control.

CARM1^WT or CARM1^QM stably expressed MCF7^CARM1−/− was maintained in Phenol Red-free Dulbecco's modified Eagle's medium (DMEM) supplemented with 5% charcoal-stripped FBS for 3 days. Cells were then treated with DMSO or 10 nM oestradiol (E2) for 4 or 12 h. RNA was then extracted using EZNA kit (Omega Bio-Tek) according to manufacturer's protocol. RNA (2 μg) was added to Superscript II (Life Technologies) reverse transcriptase with random primers. Reversed transcribed reactions were diluted 100-fold and used as template for qPCR analyses. Reactions were performed in triplicate. Cycle threshold (C_T) values were compared with standard curves prepared from serially diluted cDNA of the same cell type, then normalized to internal control RPL13A.

Purified proteins or cell lysates were resolved by SDS/PAGE and transferred to nitrocellulose membrane. Membranes were blocked with 5% milk in PBS with 0.1% Tween 20 for conventional Western blot or Odyssey Blocking Buffer (Licor) for quantitative Western blot analysis. Membranes were probed with mouse anti-CARM1 (ab110024, Abcam) at 1:2000 dilution, mouse anti-O-GlcNAc RL2 (ab2739, Abcam) at 1:100 dilu-tion, mouse anti-p21 (05-345, Millipore) at 1:1000 dilution, rabbit anti-R551me2a CARM1 at 1:2000 dilution [29] or mouse anti-β-actin (A5441, Sigma–Aldrich) at 1:10000 dilution as primary antibodies. Horseradish peroxidase-conjugated goat anti-mouse, horseradish peroxidase-conjugated goat anti-rabbit or IRDye 680 LT goat anti-mouse (Licor) were used as secondary antibody at 1:10000 dilution.

For the detection of O-GlcNAcylation on endogeneous CARM1 in MCF7, MCF7 cells were maintained in DMEM supplemented with 10% FBS. Cells were lysed with Triton lysis buffer (50 mM Tris, 150 mM NaCl, 10% glycerol and 0.5% Triton X-100). Cell lysates were incubated with rabbit polyclonal CARM1 antibody or IgG overnight at 4°C. The mixture was then incubated with Dynabead® protein A (Life Technologies) for 1 h at room temperature. The immunoprecipitated proteins were dissolved in SDS sample buffer heated at 95°C for 5 min.

For the effect of loss of O-GlcNAcylation on CARM1 dimerization, HEK293T^CARM1−/− cells were co-transfected with CARM1^WT and CARM1^QM containing either Halo–tag or Flag–tag. After 2 days, cells were harvested and lysed with Triton lysis buffer. Cell lysates were incubated with Anti-FLAG® M2 Beads (Sigma–Aldrich) overnight at 4°C. Immunoprecipitated proteins were then dissolved in SDS sample buffer.

Cells were seeded on a coverslip in a six-well plate. At 24 h after seeding, cells were washed with PBS then fixed with 4% paraformaldehyde. Fixed cells were incubated with 0.2% Triton X-100 in PBS for 3 min for permeabilization, then blocked with 3% BSA in PBST (phosphate-buffered saline with Tween 20) for 1 h at room temperature. Cells were then washed with PBST followed by incubation with mouse anti-CARM1 antibody (ab110024, Abcam) diluted 1:500 in 3% BSA for 1 h at room temperature. Subsequently, cells were washed with PBST, then incubated with Alexa Fluor 594 conjugated goat-anti-mouse antibody (Life Technologies) diluted 1:1000 in 3% BSA for 1 h at room temperature. Slides were prepared with Prolong gold mounting reagent with DAPI (Life Technologies), enabling visualization of cell nuclei. All images were collected using a Nikon Ti Eclipse inverted fluorescence microscope using a ×60 objective.

Images of >50 cells under each condition were processed using NIS Elements software, with DAPI and CARM1 intensity profiles determined across the nuclear and cytoplasmic compartments using a transect method. Briefly, the DAPI profile was used to define the nuclear–cytoplasmic boundary, subsequent to determining the average CARM1 intensity in the nucleus compared with the cytoplasm.

A study by Cheung et al. [30] showed that CARM1 expressed in murine neuroblastoma cell lines was O-GlcNAcylated; however, the O-GlcNAcylation site(s) of CARM1 was unknown. To examine whether CARM1 is O-GlcNAcylated in breast cancer cells, we immunoprecipitated CARM1 from MCF7 cells and probed with an anti-O-GlcNAc antibody. In accordance with the previous report [30], endogenous CARM1 was O-GlcNAcylated as detected by Western blot (Figure 1A). We next purified CARM1 from HEK293T cells using Halo–tag technology, as previously described [25]. Halo–tag was cleaved off, yielding highly pure CARM1 protein (Figure 1B) which was subjected to top-down MS. Nearly 100% of CARM1 was found to be di-methylated, as we previously reported [31]. Approximately 50% of this automethylated CARM1 contains a 203 Da mass shift, indicative of a single O-linked β-N-acetylglucosamine (O-GlcNAc) moiety (Figure 1C).

To identify O-GlcNAcylation site(s), we performed MS with a limited Lys-C digestion (i.e. middle-down MS) of purified CARM1. Peaks matching the expected mass of a peptide corresponding to CARM1 residues 552–608 were observed at two charge states: 5+ (A, experimental mass 7128.28 Da) and 4+ charge state (B, experimental mass 7128.33 Da) (Figure 2A). A second peak 203 Da larger than the expected mass of the unmodified CARM1 fragment was observed for both charge states (A', experimental mass 7331.36 Da and B', experimental mass 7331.37 Da for 5+ and 4+ charge state respectively). This mass shift is consistent with the addition of an O-GlcNAc moiety to a subpopulation of the C-terminal fragment of CARM1. Amino acid sequence analysis using MS/MS coupled with CAD of peaks A, A', B and B' confirmed that amino acid sequence of these four peaks match residues 552–608 of CARM1 (Supplementary Figure S1).

To further refine O-GlcNAcylation region of CARM1, the A' peak which exhibited 203 Da increase in mass was isolated for MS/MS using ECD, a low-energy dissociation method that should retain O-GlcNAc modifications. In the present study, peptide fragments consistent with O-GlcNAcylation of residues 594–606 of the CARM1 C-terminus were detected. This region of CARM1 contains four possible O-GlcNAcylation sites: Ser⁵⁹⁵, Ser⁵⁹⁸, Thr⁶⁰¹ and Thr⁶⁰³ (Figure 2B).

The C-terminus of CARM1 was predicted to be disordered [32]. Using PONDR-FIT algorithm [33], we also predict that the putative O-GlcNAcylation region of CARM1 (amino acids) 595–603) resides in a disordered region (0.7≤P≤1). This echoes the previous report that the likelihood of O-GlcNAcylation to occur in the disordered region is 6-fold higher than the structured area [14]. Next we mutated O-GlcNAcylation sites to alanine individually or in combination and expressed mutant CARM1 proteins in HEK293T cells using the Halo–tag system (Figure 2C). The O-GlcNAcylation levels of the purified recombinant CARM1 proteins were analysed by quantitative Western blot (Figure 2D), where recombinant CARM1 purified from Escherichia coli served as a negative control because O-GlcNAcylation is absent from E. coli. Consistent with our MS data, O-GlcNAcylation of a CARM1 variant lacking amino acids 554–608 (CARM1^ΔCTD) was undetectable. CARM1 variants lacking one of the putative O-GlcNAc sites (CARM1^S595A, CARM1^S598A, CARM1^T601A andCARM1^T603A) exhibited a decrease in O-GlcNAcylation level to various degrees. When all four sites were mutated to alanine (CARM1^QM), a drastic decrease in O-GlcNAcylation was observed to a level comparable with the negative control, implying that these four sites are the major sites of CARM1 O-GlcNAcylation. Moreover, alignment of CARM1 sequence across species revealed that the four O-GlcNAcylation sites were conserved among vertebrates (Figure 2E), suggesting that CARM1 O-GlcNAcylation sites, which were preserved through the course of evolution, might possess significant biological functions.

Next, we sought to investigate whether O-GlcNAcylation affects the methyltransferase activity of CARM1. Our MS analysis indicated that the recombinant protein expressed from HEK293 cells is an approximately 1:1 mixture of CARM1 with and without the O-GlcNAc moiety (Figure 1C). In order to enrich O-GlcNAcylated CARM1, we incubated recombinant CARM1 with WGA, a plant-derived lectin that had a high affinity for N-acetyl-D-glucosamine. CARM1 in the flow-through, later referred to as the O-GlcNAc-depleted fraction, was largely unmodified. The O-GlcNAcylated CARM1 was concentrated on the beads, washed and eluted using a saturated solution of N-acetyl-D-glucosamine (referred to as the O-GlcNAc enriched fraction). A diagram illustrating the O-GlcNAcylated protein enrichment procedure using WGA is shown in Figure 3(A). Additionally, we incubated CARM1^input with the purified OGA to achieve partial removal of the O-GlcNAc moiety on CARM1. The O-GlcNAc level of CARM1^input treated with OGA, O-GlcNAc enriched and depleted fractions were validated using the anti-O-GlcNAc antibody by Western blot (Figure 3B).

Next we investigated the effect of O-GlcNAcylation on the methyltransferase activity of CARM1 using an in vitro methylation assay. Total cell lysates of Cal51^CARM1−/− was used as a source of protein substrates. Since CARM1 is not expressed in this cell line, all the methylation sites on CARM1 substrates are available for incorporation of [³H] CH₃ by recombinant CARM1 in vitro. With short exposure, only two major bands were shown to be strongly methylated by CARM1. CARM1^input preferentially methylated a substrate at ∼50 kDa. Depletion of CARM1 O-GlcNAcylation, either by WGA resin or by OGA-mediated O-GlcNAc removal, shifted the substrate specificity of CARM1 toward a substrate at ∼60 kDa, whereas the O-GlcNAc-enriched CARM1 preferentially methylated a substrate at ∼50 kDa (Figure 3C). Further, the methylation pattern of cell lysate by CARM1^QM resembled that of the O-GlcNAc-depleted CARM1, implying that CARM1^QM carries the biochemical property of the non-O-GlcNAcylated CARM1. For subsequent biochemical studies, we substituted O-GlcNAc-depleted CARM1 with CARM1^QM. We next examined whether methylation of histone H3 and PABP1, two CARM1 substrates [8], was affected by CARM1 O-GlcNAcylation. Our results showed that O-GlcNAc enriched and O-GlcNAc-depleted CARM1 methylated histone H3 and PABP1 to a similar extent, suggesting that O-GlcNAcylation did not affect the ability of CARM1 to methylate these two substrates (Figure 3D).

Although the result of our 1D analyses implies that substrates of CARM1 can be classified into three classes, e.g. those that are preferentially methylated by O-GlcNAcylated CARM1, those that are preferentially methylated by non-GlcNAcylated CARM1 and those that are methylated regardless of O-GlcNAc status of CARM1, band isolation and subsequent protein identification by MS are likely to yield many false-positive hits, due to the limited resolution of 1D SDS/PAGE to resolve proteins within a complex cell lysate. To increase resolution and to determine if O-GlcNAcylation of CARM1 affects methylation of other cellular substrates, we performed a large-scale in vitro methylation assay using lysates obtained from Cal51^CARM1−/− cells as a source of methylation substrates and resolved the proteins by 2D gel (Figure 3E). Some substrates were found preferentially methylated by O-GlcNAc-depleted CARM1 (Figure 3E, white arrows), whereas others were found preferentially methylated by O-GlcNAc-enriched CARM1 (Figure 3E, black arrows). There were proteins methylated similarly by both forms of CARM1 (Figure 3E, red arrows).

Spots which exhibited a difference in autoradiographic pattern were excised. Proteins were retrieved by in-gel digestion and subjected to protein identification by MS. As a result, 840 proteins were identified (Supplementary Table S1). Further substrate validation is needed to determine which protein is a bona fide substrate for CARM1 and which protein is differentially methylated by O-GlcNAcylated and non-O-GlcNAcylated CARM1.

We previously reported automethylation of CARM1 [31]. Since automethylation and O-GlcNAcylation both occur at the C-terminus, we ask whether the two PTMs exhibit cross-talk. In the present study, O-GlcNAc and automethylation level of CARM1 mutants defective of O-GlcNAcylation (CARM1^QM) and automethylation (CARM1^Arg551K) were determined using anti-O-GlcNAc and site-specific anti-asymmetrically dimethylated Arg⁵⁵¹ (anti-R551me2a CARM1) [29] antibodies by Western blot. Purified CARM1^WT served as a positive control. We found that automethylation and O-GlcNAcylation of CARM1, although both occur at the C-terminus, are probably independent events (Supplementary Figure S2).

We have previously reported CARM1 knockout MDA-MB-231 (MDA-MB − 231^CARM1−/−) and MCF7 (MCF7^CARM1−/−) cell lines generated by zinc finger nuclease (ZFN) technology [28]. Flag–tagged CARM1^WT and CARM1^QM were restored in these cell lines, which allow direct comparison with CARM1^WT and CARM1^QM with no interference from endogenous CARM1 in these cells.

To examine whether the subcellular localization of CARM1 is affected by O-GlcNAcylation, we performed immunofluorescence using the MDA-MB-231 parental cell line and MDA-MB − 231^CARM1−/− restored with CARM1^WT, CARM1^QM or empty vector control. Western blotting results shows that the stable cell lines express CARM1^WT and CARM^QM at comparable levels (Figure 4A). In the MDA-MB-231 parental cells, endogeneous CARM1 is localized primarily in the cytoplasm. CARM1^WT and CARM1^QM are indistinguishable from that of endogeneous CARM1 (Figure 4B). Quantification of CARM1 immunofluorescence intensity revealed no detectable difference in nuclear–cytoplasmic distribution of CARM1^WT and CARM1^QM (Figure 4C), indicating that loss of O-GlcNAcylation does not affect cellular localization of CARM1.

Next we examined the effect of O-GlcNAcylation on CARM1 protein stability. MDA-MB-231 parental cells and CARM1^WT or CARM1^QM expressing MCF7^CARM1−/−MDA-MB − 231^CARM1−/− were treated with cycloheximide (CHX) to inhibit protein synthesis. CARM1 protein in MDA-MB-231 cells appeared to be very stable, as even after 24 h treatment with CHX CARM1 protein level remained unchanged, whereas p21 was completely undetectable after 6 h CHX treatment (Supplementary Figure S3A). The stability of CARM1 protein in CARM1^WT- and CARM1^QM-restored MDA-MB − 231^CARM1−/− cells was similar to that of the parental cells. No discernible difference was observed in the protein stability of CARM1^WT and CARM1^QM (Supplementary Figure S3B), indicating that O-GlcNAcylation does not affect CARM1 protein stability.

Next we asked whether O-GlcNAcylation could affect dimerization of CARM1. In the present study, Halo–tag and Flag–tag CARM1 (WT and QM) were expressed in HEK293T cells with CARM1 knocked out by ZFN (HEK293T^CARM1−/−) [28]. Immunoprecipitation was performed using Anti-FLAG® M2 Beads. We found that CARM1^QM could form dimers with both CARM1^WT and CARM1^QM (Supplementary Figure S3C), indicating that dimerization of CARM1 was not obviously affected by O-GlcNAcylation under these conditions. These results were consistent with the previous observation that the dimerization domain resides in the central catalytic region [32,34].

Because both the O-GlcNAcylation sites of CARM1 and the co-activator activity reside in the C-terminus [9], we asked whether CARM1 O-GlcNAcylation affects co-activator activity. We have previously shown that CARM1 increases the transcriptional activity of Sertad, NFκBIB and p53 [31]. Using a mammalian one-hybrid approach, we examined the co-activator activity of CARM1^QM with these transcription factors. Plasmids encoding Sertad, NFκBIB and p53 fused to the DNA-binding domain of GAL4 were co-transfected with CARM1^WT or CARM1^QM and a luciferase reporter driven by a GAL4-responsive element (GAL4–Luc). CARM1^WT and CARM1^QM exhibited similar activities to activate these transcription factors (Figure 5A), suggesting that loss of O-GlcNAcylation does not compromise the ability of CARM1 to activate these transcription factors. Moreover, the activities of GAL4–CARM1^WT or GAL4–CARM1^QM to activate a GAL4-response element fused luciferase reporter [9] were found to be indistinguishable (Figure 5B), further supporting the notion that the co-activator function of CARM1 is not regulated by O-GlcNAcylation.

The C-terminus of CARM1 mediates its interaction with the co-activator protein TIF1α [10]. Since O-GlcNAcylation sites of CARM1 reside at the C-terminus, we determined if O-GlcNAcylation affects the interaction of these two proteins in a mammalian two-hybrid assay. CARM1^WT and CARM1^QM were fused to the GAL4 DNA-binding domain whereas TIF1α was fused to herpes simplex virus (VP16). Our result showed that CARM1^WT and CARM1^QM interacted with TIF1α similarly, indicating that O-GlcNAcylation does not affect the interaction between CARM1 and TIF1α (Supplementary Figure S4A).

CARM1 was previously shown to form a ternary complex with GRIP1 and TIF1α [10]. Together, these three cofactor proteins synergistically activated transcription of a reporter gene [10]. To investigate whether O-GlcNAc affects the synergistic activity of this complex, we performed reporter assays by co-transfecting GAL4–GRIP1, TIF1α and CARM1^WT or CARM1^QM with GAL4–Luc. In agreement with the previous report, we observed a synergistic increase in reporter activity when GAL4–GRIP1, TIF1α and CARM1 were transfected with the GAL4–Luc reporter (Supplementary Figure S4B) [10]. A dose-dependent increase in reporter activity was observed as the amount of transfected CARM1 construct increased. However, CARM1^WT and CARM1^QM displayed a similar ability to activate the GAL4–Luc reporter, indicating that O-GlcNAcylation of CARM1 does not affect the synergistic interactions of these co-activators.

Finally, we investigated whether O-GlcNAc regulates the activity of CARM1 on oestrogen receptor α (ERα) target genes. Either CARM1^WT or CARM1^QM was restored in MCF7^CARM1−/− cells to comparable levels (Figure 5C) and qPCR of previously reported ERα–CARM1 co-regulated genes including pS2, c-Myc, PTGES (prostaglandin E synthase), EGR3 (early growth response factor 3) and IGFBP (insulin-like growth factor-binding protein) [31,35] was performed. As expected, the expression levels of these genes are E2-responsive. However, no statistical difference was observed between their expression levels in MCF7–CARM1^WT and MCF7–CARM1^QM cells, indicating that O-GlcNAcylation status of CARM1 did not influence the expression levels of these genes (Figure 5D).

Although CARM1 was previously reported to be O-GlcNAcylated [30], neither the extent nor the sites of O-GlcNAcylation were known. Using high-resolution top-down MS, we have discovered that ∼50% of recombinant CARM1 expressed from HEK293T cells is mono-O-GlcNAcylated. We did not detect higher glycoforms (i.e. di- or tri-glycosylated) of CARM1, suggesting that they may not exist in the condition we investigated or are present at a low level that is under the detection limit by MS. Furthermore, we identified four adjacent putative O-GlcNAcylation sites at the C-terminus of CARM1, indicating that mono-O-GlcNAcylated CARM1 is probably comprised of a mixture of forms with modification at Ser⁵⁹⁵, Ser⁵⁹⁸, Thr⁶⁰¹ or Thr⁶⁰³. Multiple O-GlcNAcylation sites located proximal to each other on a protein is a well-known phenomenon, having been observed on over 30 proteins [36,37]. For example, DRP1 (dynamin-related protein 1), a protein involving in mitochondria fission, is O-GlcNAcylated at Thr⁵⁸⁵ and Thr⁵⁸⁶ [38]. The C-terminal domain of RNA polymerase II is O-GlcNAcylated at Ser⁵ and Ser⁷ [39]. Transcription factor C/EBPβ (CCAAT/enhancer binding protein (C/EBP), beta) is O-GlcNAcylated at Ser¹⁸⁰ and Ser¹⁸¹ [40]. Thus far, no consensus sequence has been identified for protein O-GlcNAcylation [41], although unstructured regions are preferred for O-GlcNAcylation [14]. It is likely that these disordered regions are transiently stabilized upon binding to OGT, require stabilization provided by other binding partners or both.

The C-terminus of CARM1 was shown to have autonomous activation function [9]. O-GlcNAcylation has been reported to affect diverse functions of protein, including enzymatic activity, protein–protein interaction, DNA binding and transcriptional activity of transcription factors, subcellular localization and protein stability (reviewed in [18,22–24]). For example, the stability of Snail1, p53 and plakoglobin protein is regulated by O-GlcNAcylation [42–44]. Subcellular localization of actin-binding protein cofilin, transcription factor Sp1 and neuroD1 is altered by O-GlcNAcylation [45–47]. Subcellular localization of CARM1 was reported to be cell-type dependent. For example, CARM1 is primarily localized in the cytoplasm in HEK293 cells, but is found in the nucleus in MCF7 and HeLa cells [48]. Our result indicated that CARM1 localizes primarily in the cytoplasm in MDA-MB-231 cells. Further, our results showed that O-GlcNAcylation of CARM1 does not alter the co-activator activity, stability, localization of CARM1 or the growth of breast cancer cell lines (result not shown), yet it does regulate the substrate specificity of CARM1. In line with this finding, O-GlcNAcylation of CK2 also regulates its substrate specificity [49]. Future work is warranted to identify substrates differently methylated by O-GlcNAcylated and non-GlcNAcylated CARM1 in order to better understand the functional significance of CARM1 O-GlcNAcylation. The fact that O-GlcNAcylation does not affect many aspects of CARM1's function implies that O-GlcNAcylation is likely to be a mechanism for fine-tuning CARM1 function. Among the three phosphorylation events on CARM1, two occur in the central catalytic domain, mutation of which abrogates ERα-mediated transcription [50,51]. Phosphorylation at the third site Ser⁴⁴⁸ enables CARM1 to directly interact with un-liganded ERα [52]. This direct interaction is important for ligand-independent activation of ERα and possibly contributes to tamoxifen resistance. Finally, automethylation of CARM1 at Arg⁵⁵¹ modulates CARM1-mediated exon skipping [31]. Taken together, functions of CARM1 appear to be regulated by multiple PTMs, the majority of which are localized to the C-terminus. Competition of O-GlcNAcylation and phosphorylation on serine or threonine residues has been reported on some proteins. For example, Thr⁵⁸ of c-Myc [53], Ser¹⁶ of ERβ [54], Ser⁴⁵² of heat shock protein (Hsp90β) [55] and Ser⁷³³ of IκB kinase (IKKβ) [56] can be either O-GlcNAcylated or phosphorylated. However, even with the use of non-ergodic dissociation methods such as ECD in MS, we were unable to detect phosphorylation at these O-GlcNAcylation sites. Although we observed the labile O-GlcNAc modification on CARM1, no H₃PO₄ (98 Da) shift was observed even in the presence of phosphatase inhibitor cocktails, suggesting that phosphorylated CARM1 is below the limit of detection (<5%) by top-down MS. This result agrees with previous reports that CARM1 phosphorylation occurs mainly in mitosis [50,51,57] or in response to specific stimuli [52] in the central domain and implies that phosphorylation is unlikely to play a significant role in modulating CARM1 O-GlcNAcylation.

CARM1 O-GlcNAcylation sites are mapped to the C-terminal domain which possesses an autonomous activation activity and is essential for full co-activator activity of CARM1 [9]. Mutation of O-GlcNAcylation sites did not alter the ability of CARM1 to activate a few transcription factors or change the expression levels of endogenous ERα target genes. We could not exclude the possibility that O-GlcNAcylation of CARM1 alters the regulation of other genes and other transcription factors that were not tested in our study. Although we found that O-GlcNAcylation of CARM1 at its C-terminus altered substrate specificity (Figures 3C and 3E), mutation of O-GlcNAcylation residues did not result in changes in methylation of histone H3 and PABP1 (Figure 3D). One possible explanation is that, for some CARM1 substrates, the C-terminus provides an essential docking platform for efficient methylation. O-GlcNAcylation may alter the affinity of this interaction, thus shifting substrate specificity of CARM1 towards a subpopulation of substrates. In support of this notion, O-GlcNAcylation was shown to alter protein–protein interactions between many proteins, such as Sp1–Oct1 [58], Sp1–Elf1 [59] and Stat5–CBP [60]. For more potent CARM1 substrates, such as histone H3 and PABP1, the C-terminus of CARM1 may be dispensable for their methylation. For example, we have shown that even when CARM1 was reduced to 10% of endogenous level in MCF7 cells, CARM1-mediated methylation of PABP1 was not strikingly affected [61]. Co-crystal structures of full-length CARM1 with some substrates will help to elucidate whether the role of the C-terminus in regulating CARM1 substrate binding is substrate specific.

Cellular O-GlcNAcylation is known to be sensitive to various stresses including glucose availability, inflammatory stimuli, insulin and oxidative stress, among others (reviewed in [18,19,62]). UDP–GlcNAc, situated at the nexus of the glucose, lipid, nucleotide and amino acid biosynthesis pathways, is believed to be a sensor for cellular metabolites. Moreover, diseases with impaired metabolism such as neurodegenerative disease, diabetes, cardiovascular disease and cancer often exhibit altered levels of cellular O-GlcNAcylation (reviewed in [12,21,22,62–65]).

A few studies have implicated CARM1 as a regulator of cellular metabolism [5,66–69]. CARM1 was shown to be a co-activator of PPARγ [5], a key transcription factor regulating glucose and lipid metabolism. The expression level of CARM1 in cattle liver [66] and subcutaneous adipose tissue [67] is sensitive to types of fatty acid in diet. Moreover, CARM1 regulates expression of key enzymes in gluconeogenesis [68] and glycogen metabolism [69]. We speculate CARM1 O-GlcNAcylation status, like ‘histone codes’, functionally define CARM1 function in cellular stress and metabolism processes (e.g. CARM1 O-GlcNAcylation level may respond to cell glucose levels and be part of the central hub of cell growth signals). Determining CARM1 O-GlcNAcylation levels in response to different cellular stimuli may provide insights into this mechanism.

CAD

collisionally-activated dissociation

CARM1

co-activator-associated arginine methyltransferase 1

CARM1^QM

CARM1 quadruple mutant

CARM1^WT

CARM1 wild-type

CHX

cycloheximide

DMEM

Dulbecco's modified Eagle's medium

E2

oestradiol

ECD

electron-capture dissociation

ERα

oestrogen receptor α

GRIP1

glucocorticoid receptor interacting protein 1

HEK

human embryonic kidney

OGA

O-GlcNAcase

O-GlcNAc

O-linked-β-N-acetylglucosamine

O-GlcNAcylation

O-linked-β-N-acetylglucosaminidation

OGT

O-GlcNAc transferase

PABP1

poly(A)-binding protein 1

PBST

phosphate-buffered saline with Tween 20

PPARγ

peroxisome proliferator-activated receptor γ

PRMT

protein arginine methyltransferase

PTM

post-translational modification

TIF1α

transcription intermediary factor 1 α

WGA

wheat germ agglutinin

ZFN

zinc finger nuclease

Purin Charoensuksai and Wei Xu designed the experiments. Purin Charoensuksai performed the experiments. Peter Kuhn performed MS. Lu Wang provided CARM1-knockout cells. Nathan Sherer supervised confocal immunofluorescent analysis. Purin Charoensuksai, Peter Kuhn and Wei Xu wrote the manuscript.

We thank Dr Michael R. Stallcup for providing the pM–CARM1, pM–GRIP1 and pSG5–TIF1α plasmids and Dr Richard R. Burgess for critical reading of the manuscript.

This work was supported by the HOPE Scholar Award [grant number W81XWYH-11-1-0237 (to W.X.)]; and the Royal Thai Government Scholarship (to P.C.).