Glycogen synthase kinase-3 (GSK-3) activity regulates mRNA methylation in mouse embryonic stem cells

Glycogen synthase kinase-3 (GSK-3) activity regulates multiple signal transduction pathways and is also a key component of the network responsible for maintaining stem cell pluripotency. Genetic deletion of Gsk-3α and Gsk-3β or inhibition of GSK-3 activity via small molecules promotes stem cell pluripotency, yet the mechanism underlying the role for GSK-3 in this process remains ambiguous. Another cellular process that has been shown to affect stem cell pluripotency is mRNA methylation (m6A). Here, we describe an intersection between these components, the regulation of m6A by GSK-3. We find that protein levels for the RNA demethylase, FTO (fat mass and obesity–associated protein), are elevated in Gsk-3α;Gsk-3β-deficient mouse embryonic stem cells (ESCs). FTO is normally phosphorylated by GSK-3, and MS identified the sites on FTO that are phosphorylated in a GSK-3–dependent fashion. GSK-3 phosphorylation of FTO leads to polyubiquitination, but in Gsk-3 knockout ESCs, that process is impaired, resulting in elevated levels of FTO protein. As a consequence of altered FTO protein levels, mRNAs in Gsk-3 knockout ESCs have 50% less m6A than WT ESCs, and m6A-Seq analysis reveals the specific mRNAs that have reduced m6A modifications. Taken together, we provide the first evidence for how m6A demethylation is regulated in mammalian cells and identify a putative novel mechanism by which GSK-3 activity regulates stem cell pluripotency.

GSK-3 activity is the sum of two largely redundant proteins, GSK-3␣ and GSK-3␤, and in general, GSK-3 is a negative regulator of cellular signaling (2). Rare among kinases, GSK-3 is active at a basal state, whereas pathway activation from upstream signaling cascades results in the inhibition of GSK-3 activity (2). GSK-3␣ and GSK-3␤ together regulate signal transduction pathways such as Wnt, protein kinase A, Hedgehog, transforming growth factor-␤, nuclear factor of activated T-cells, and phosphatidylinositol 3-kinase (PI3K)-dependent insulin signaling in a variety of biological settings (3)(4)(5).
Whereas GSK-3 activity clearly plays an important role in stem cell pluripotency, other biological processes are also important for pluripotency. Recent work has demonstrated an important role for the methylation of adenosines (m 6 A) in mRNA in the regulation of stem cell pluripotency (36 -38). m 6 A is the most common internal modification to mRNA in mammalian cells (39 -41), yet little was known about the biology of this modification until a recent flurry of research led to the identification of the enzymes that add and remove m 6 A (METTL3/METTL14 and FTO/ALKBH5, respectively) (42)(43)(44), as well as enrichment-based sequencing experiments to identify mRNAs modified by m 6 A (45,46). With respect to stem cell pluripotency, it has been shown that reduced m 6 A levels promote pluripotency (36 -38). One explanation that has been proposed for how m 6 A levels affect stem cell pluripotency is that mRNAs for several components of the core pluripotency circuitry in mouse ESCs have reduced m 6 A, thereby increasing the half-life of these factors (36).
Here we describe the regulation of m 6 A mRNA by GSK-3 in mouse ESCs. We find that FTO (fat mass and obesityassociated) protein levels are elevated in Gsk-3 DKO ESCs compared with WT ESCs, resulting in a concomitant decrease in m 6 A levels. The mechanism underlying this effect is shown to be the phosphorylation of FTO by GSK-3 and subsequent ubiquitination, a process that is impaired in Gsk-3 DKO ESCs.

FTO protein levels are elevated in Gsk-3 DKO ESCs
Reduced GSK-3 activity and decreased levels of m 6 A are both associated with promotion of pluripotency. We therefore investigated whether there was a direct connection between GSK-3 and m 6 A. Because stem cell pluripotency is enhanced by reduced m 6 A mRNA levels, an increase in FTO protein would be one route to achieving decreased m 6 A mRNA levels. Because the stereotypical role for GSK-3 is to negatively regulate target substrates, our hypothesis was that FTO protein levels are kept low by GSK-3 activity, and consequently, we should observe an increase in FTO protein levels in Gsk-3 DKO ESCs. We initiated our study by examining the levels of FTO protein in WT and Gsk-3 DKO ESCs grown under standard conditions (supplemented with LIF). We observed no difference in FTO protein between these cells (Fig. 1A). We then realized that, by using culture conditions containing LIF, both WT and Gsk-3 DKO ESCs would be pluripotent. Thus, adding LIF would not help us to understand how loss of GSK-3 was promoting pluripotency. It had been shown previously that Gsk-3 DKO ESCs would remain pluripotent when grown in the absence of LIF for as long as 14 days (12). Under these same conditions, WT ESCs progressively differentiate. By growing ESCs without LIF, we were able to obtain conditions in which we could now assess how loss of GSK-3 promotes pluripotency. Persistence of pluripotency in Gsk-3 DKO ESCs in the absence of LIF was verified by examining mRNA levels of the pluripotency-associated genes Nanog and estrogen receptor-␤ (Esrrb) by qPCR. As seen in Fig. 1B, relative levels of Nanog and Esrrb mRNA remain elevated in Gsk-3 DKO ESCs compared with WT ESCs, demonstrating that even without supplementation with LIF, Gsk-3 DKO ESCs remain pluripotent. We then isolated proteins from WT and Gsk-3 DKO ESCs, separated by SDS-PAGE, and examined FTO protein levels by Western blotting. We observed substantially higher FTO protein levels in Gsk-3 DKO ESCs (Fig.  1C). We then asked whether higher FTO protein levels was due to increased levels of mRNA. qPCR analysis of FTO mRNA revealed no increase in Gsk-3 DKO ESCs, and in fact, the mRNA levels were decreased (Fig. 1D). These data suggest that the elevated levels of FTO protein seen in Gsk-3 DKO ESCs are probably due to post-transcriptional effects of GSK-3.

m 6 A quantification
Given that FTO is an RNA demethylase and FTO protein is elevated in Gsk-3 DKO ESCs relative to WT ESCs, we sought to quantitatively measure the amounts of m 6 A mRNA in ESCs by UHPLC-MS. Total RNA was isolated from WT and Gsk-3 DKO ESCs grown in the absence of LIF for 14 days. We then enriched for mRNA, which was then digested into single nucleotides. Targeted UHPLC-MS detection of individual nucleotides allowed us to calculate the relative amount of m 6 A, which was normalized against the total adenosine pool to correct for any differences in the amount of RNA analyzed in each experiment. Analysis revealed a 49% reduction in m 6 A nucleotides in Gsk-3 DKO ESCs compared with WT ESCs (Fig. 1E), consistent with the higher levels of FTO protein observed in Gsk-3 DKO ESCs.

Small-molecule inhibitors of GSK-3 increase FTO protein
The elevation of FTO protein was not limited to Gsk-3 DKO ESCs. We extended our initial observation to WT ESCs treated with small-molecule inhibitors of GSK-3. WT ESCs were grown in the absence of LIF for 14 days and then treated with increasing doses of the GSK-3 inhibitors SB-415,286 or lithium chloride for 48 h. Examination of FTO protein by Western blotting revealed a dose-dependent increase in FTO protein (Fig. 2,  A and B). These results strongly support the notion that GSK-3 inhibition, even for a short time, is sufficient to result in elevated levels of FTO protein in ESCs.

FTO phosphorylation
Because FTO protein levels are modulated by GSK-3 activity, we hypothesized that FTO is phosphorylated by GSK-3, and a repeated GSK-3 consensus motif was found within mouse FTO (amino acids 245-257) (Fig. 3A). To test whether FTO phosphorylation occurs in mouse ESCs, we overexpressed murine V5-tagged FTO in both WT and Gsk-3 DKO ESCs. Phosphorylated proteins were enriched using a Pro-Q Diamond kit, and proteins were analyzed by Western blotting for the presence of FTO. Examination of total protein lysates revealed that FTO-V5 was expressed in both WT and Gsk-3 DKO ESCs, indicating successful transfection of cells. Analysis of phospho-enriched proteins revealed FTO-V5 in WT ESCs, but not in Gsk-3 DKO ESCs (Fig. 3B). From this experiment, we conclude that FTO is normally phosphorylated in WT ESCs, and this phosphorylation is impaired in Gsk-3 DKO ESCs.

Mapping of FTO phosphorylation in vivo
We next sought to identify the precise amino acids within FTO that are phosphorylated in WT ESCs and not in Gsk-3 DKO ESCs. ESCs were transfected with murine FLAG-tagged FTO and then immunoprecipitated, trypsinized, and analyzed by LC-MS/MS. These results definitively showed the presence of phosphorylation in the region of FTO containing the GSK-3 consensus sequences, specifically on Ser-249 and Ser-253. Collision-induced dissociation of the singly phosphorylated peptide and carbamidomethylated (M ϩ 3H) 3ϩ ion (1380.2174 m/z, sequence SAVAVYSYSCEGSEDESEDESSFEGRDPDT-WHVGFK) generates fragment ions that suggest a mixed population of isobaric peptides that harbor phosphorylated species on at least two residues (Fig. 3C). The observed fragment ions primarily suggest a mix of phosphorylation sites on both serine residues found in the sequence . . . EG(SEDES)ED . . . (where the parentheses denote the amino acid motif that was found to be phosphorylated, where both serines were phosphorylated). No doubly or triply phosphorylated forms of this sequence were observed.   Blots were then stripped and reprobed with ␣-tubulin antibody to ensure equal loading of protein lysates.

GSK-3 regulation of m 6 A
To confirm that phosphorylation of FTO within the GSK-3 consensus motif accounts for the phosphorylation identified by phospho-enrichment, we created murine FTO in which serines 249 and 253, found to be phosphorylated by LC-MS/MS, are mutated to alanines (pCAGEN-FLAG-FTO-S249A/S253A). When WT FLAG-FTO and FLAG-FTO S249A/S253A were transfected into WT ESCs and phosphorylated proteins were enriched using the Pro-Q Diamond kit, we recovered less of the serine mutant FTO compared with WT FTO (Fig. 3D). This suggests that changing the GSK-3 consensus motif in FTO is sufficient to reduce its phosphorylation in vivo, further strengthening the argument that GSK-3 is the kinase responsible for FTO phosphorylation.

FTO is a substrate for GSK-3␤
Because the role for GSK-3 in controlling FTO protein levels could be indirect, we sought to ask whether GSK-3 could directly phosphorylate FTO. We performed in vitro kinase assays using recombinant GSK-3␤ and recombinant mouse FTO protein. Using conditions that resulted in the phosphorylation of recombinant 2N4R Tau (47) by GSK-3␤ (data not shown), incubation of FTO with recombinant GSK-3␤ in the presence of fresh ATP resulted in phosphorylation of FTO (Fig.  3E). From this experiment, we conclude that FTO is probably a direct substrate for GSK-3␤. In addition, the relatively weak phosphorylation of FTO by GSK-3␤ points to the likely requirement of a priming kinase for optimal FTO phosphorylation by GSK-3␤.

FTO ubiquitination
Because FTO protein levels are increased in Gsk-3 DKO ESCs compared with WT ESCs, we speculated that this could be due to a reduction in FTO ubiquitination and subsequent degradation. To assess this possibility, we examined the status of FTO ubiquitination in WT and Gsk-3 DKO ESCs. Cells were co-transfected with FLAG-FTO and HA-ubiquitin, then

GSK-3 regulation of m 6 A
treated with lactacystin to inhibit the proteasome, allowing for the accumulation of ubiquitinated proteins. We then immunoprecipitated FTO using anti-FLAG antibody and performed Western blotting for ubiquitin using an anti-HA antibody. We found that FTO is modified by ubiquitin in WT ESCs, but this modification is significantly impaired in Gsk-3 DKO ESCs (Fig. 4A).
Next, we examined whether preventing FTO phosphorylation by GSK-3 was sufficient to affect FTO ubiquitination. WT ESCs were transfected with HA-ubiquitin and either FLAG-FTO or FLAG-FTO S249A/S253A. Using the same procedure described above for Fig. 4A, we observed that FTO modification by ubiquitin is reduced in the FTO S249A/S253A mutant compared with WT FTO (Fig. 4B). Therefore, polyubiquitination associated with proteasome-mediated protein degradation is impaired with respect to FTO in Gsk-3 DKO ESCs, providing a likely explanation for the increased FTO protein levels seen in Gsk-3 DKO ESCs.

GSK-3␤ rescues m 6 A levels and pluripotency gene expression in Gsk-3 DKO ESCs
Reducing GSK-3 activity via genetic deletion or small-molecule inhibitors promotes stem cell pluripotency, as well as regulating m 6 A mRNA levels. Re-expressing GSK-3␤ in Gsk-3 DKO ESCs would therefore be predicted to reverse these effects. Gsk-3 DKO ESCs stably expressing GSK-3␤ (DKO: GSK-3␤) were created (Fig. 5A) and analyzed. WT, Gsk-3 DKO, and DKO:GSK-3␤ ESCs were grown in the absence of LIF for 14 days and then analyzed for gene expression and m 6 A levels. Compared with Gsk-3 DKO ESCs, DKO:GSK-3␤ ESCs had significantly reduced levels of expression of both pluripotencyassociated genes Nanog and Esrrb, as assessed by qPCR (Fig.  5B). Furthermore, the reduced levels of m 6 A seen in Gsk-3 DKO ESCs were almost completely restored to WT levels in DKO: GSK-3␤ ESCs (Fig. 5C). These data, along with the in vitro kinase assay results, help to reinforce the notion that GSK-3 is indeed regulating FTO in vivo, and this regulation is intimately connected to controlling stem cell pluripotency.

m 6 A-Seq
Whereas m 6 A mRNA levels in Gsk-3 DKO ESCs were approximately half of that detected in WT ESCs, the LC-MS/MS analysis gave no information as to which specific mRNAs had reduced m 6 A modifications. To obtain this information, we performed m 6 A-Seq, making modifications to the approach taken by Meyer et al. (45) and Dominissini et al. (46). We first isolated total RNA from WT and Gsk-3 DKO ESCs, enriched for mRNA using oligo(dT) beads, and then immunoprecipitated m 6 A mRNA using an anti-m 6 A antibody coupled to magnetic beads. The resulting m 6 A mRNA was then chemically fragmented, cDNA was created using random primers, adaptors were ligated to the ends, and next-generation sequencing was performed. m 6 A-Seq analysis was performed to identify transcripts that had at least 2-fold fewer m 6 A modifications in Gsk-3 DKO ESCs compared with WT ESCs and relative to input reads for WT and Gsk-3 DKO ESCs. 899 mRNAs were identified that met these criteria (Fig. S1). To verify the results obtained from the m 6 A-Seq experiment, we performed meRIP-qPCR on the mRNA from WT and Gsk-3 DKO ESCs that was immunoprecipitated with the m 6 A antibody, specifically examining mRNAs that were found to be hypomethylated in Gsk-3 DKO ESCs. We found that m 6 A levels for the pluripotency-associated mRNAs Esrrb (Fig. 6A) and c-myc (Fig. 6B) were both reduced by ϳ40% with respect to input mRNA in Gsk-3 DKO ESCs, providing independent validation of our m 6 A-Seq results.

GSK-3 regulation of m 6 A
Our approach for detecting m 6 A relied on an antibody against m 6 A, but this antibody was recently shown to detect both m 6 A and N 6 ,2Ј-O-dimethyladenosine (m 6 A m ) (48), making it impossible for us to discriminate between these two RNA modifications using this technique (49). However, because m 6 A m is only found directly adjacent to the 7-methylguanosine cap on mRNA (50), we are confident that most of the changes we observe are m 6 A, but we cannot discount the possibility of changes in m 6 A m as well.
Because we were interested in whether there is a direct connection between mRNAs with GSK-3-dependent reduction and mouse ESC pluripotency, we wanted to examine whether any of the core pluripotency factors described by Young (51) had reduced m 6 A modifications in Gsk-3 DKO ESCs. We found that only a handful of the mRNAs for these pluripotency factors (Esrrb, c-myc, Sox2, and Smad1) had reductions in m 6 A. Furthermore, because Mettl3 Ϫ/Ϫ ESCs also have reduced m 6 A mRNA levels as well as enhanced pluripotency (36), we looked for concordance between mRNAs that show reduced m 6 A modifications in both Gsk-3 DKO ESCs and Mettl3 Ϫ/Ϫ ESCs. With respect to the mRNAs described as being part of the pluripotency circuit, we found reduced m 6 A levels on Lin28, c-myc, and Max in Gsk-3 DKO ESCs. Therefore, whereas there are modest effects on known regulators of pluripotency, the overall effect on these regulators could help to explain the persistent pluripotency seen in Gsk-3 DKO ESCs.

Discussion
The epitranscriptome landscape in eukaryotic cells has recently been revealed in detail. Besides m 6 A-Seq studies that have precisely identified the location of methylated adenosines in RNA molecules (45,46), details are also emerging regarding the reversible nature of epigenetic marks on RNA (52). This reversibility implies that RNA modifications are regulated, but the nature of this regulation is just beginning to be understood. Here, we describe a compelling role for GSK-3 in the regulation of m 6 A levels via FTO (Fig. 7).
We provide evidence that GSK-3 plays a role in modulating mRNA adenosine methylation in mouse embryonic stem cells through controlling the protein levels of the RNA demethylase FTO. Whereas GSK-3 inhibition or deletion has been known to regulate embryonic stem cell pluripotency, the precise role for GSK-3 has been difficult to pin down. Several different aspects of GSK-3 function have been investigated, but a great deal of focus has been on examining the GSK-3-dependent signaling pathways that promote pluripotency. Most of these studies have focused on Wnt and PI3K signaling because activation of each pathway results in inhibition of GSK-3, but the conclusions have not been clear-cut. For example, with respect to Wnt signaling, it has been proposed that pluripotency via GSK-3 inhibition works through both Tcf3-dependent (26, 53, 54) and Tcf3-independent (24) mechanisms. In addition, several lines of evidence also support a role for PI3K in promoting stem cell pluripotency (55). Detailed mechanistic studies have determined that GSK-3 activity was regulated in part by PI3K signaling, yet the precise mechanism by which inhibition of GSK-3 was promoting stem cell pluripotency remained elusive (8,29,34,56). Our findings raise the possibility that Wnt signaling and/or PI3K signaling via GSK-3 could be regulating mRNA methylation, thereby controlling stem cell pluripotency. Further studies will be required to determine whether specific signaling pathways have an effect on FTO phosphorylation and mRNA methylation.
The phosphorylation of proteins has been shown to be a common prerequisite for subsequent polyubiquitination (57), and GSK-3 activity has been shown to precede substrate ubiquitination (58). A prime example is ␤-catenin, the main effector

GSK-3 regulation of m 6 A
of canonical Wnt signaling. ␤-Catenin is ubiquitinated and degraded only after phosphorylation by GSK-3 on N-terminal serines and threonines (59). Activation of Wnt signaling by the addition of Wnt ligand (60), genetic deletion of GSK-3 (12), and small-molecule inhibition of GSK-3 (10, 11, 61) all result in the accumulation of ␤-catenin protein. We observe a similar paradigm for GSK-3 with respect to FTO. Although we provide evidence that FTO is a direct target of GSK-3, the in vitro kinase assay for FTO lacked the robustness of phosphorylation seen in the positive control (Tau). This result probably reflects that GSK-3 often phosphorylates substrates that have been primed by another kinase (62); Tau does not have a requirement for priming with respect to GSK-3 (63). Therefore, it is quite possible that phosphorylation of FTO by another kinase would result in enhanced phosphorylation by GSK-3. Because our in vitro kinase assay only examined the effect of GSK-3 on FTO phosphorylation, the next reasonable step would be to identify kinases that work in tandem with GSK-3 in the phosphorylation of FTO.
With our data describing a novel role for GSK-3 activity in modulating post-transcriptional modifications of mRNA, one could envision how signaling pathways that converge on GSK-3 via distinct routes (e.g. Wnt and PI3K) could synergize with respect to gene expression. For example, whereas one GSK-3dependent pathway could promote the transcription of a specific mRNA, a different GSK-3-dependent pathway could elevate FTO protein levels, thereby reducing methylation of the same mRNA. An exciting possibility raised by our work is that transcription and RNA modifications are occurring concurrently, resulting in a fine-tuning of the net transcriptional response to signaling events.
These data may also be relevant in further understanding how signaling pathways have the capacity to directly alter the m 6 A modifications of specific mRNAs, as well as providing insights into the mechanisms of GSK-3-mediated signaling in other cell types and disease states. FTO has been shown to be expressed at high levels in the brain (64 -66), implicating an important role in regulating mRNA methylation. Although details on the relationship between RNA modifications and disease are just beginning to be examined in depth (67), it is worth noting that GSK-3 is inhibited by the mood-stabilizing drug lithium (9,10). A recent study showed that genes involved in m 6 A modification are associated with major depressive disorder in a Han Chinese population (68). The data presented here

GSK-3 regulation of m 6 A
raise the possibility that toggling the levels of methylated adenosine on mRNA in the brain could play a role in mood stabilization.
The FTO locus in humans has been shown unequivocally to be associated with predisposition to obesity (69 -72), and genetic deletion of FTO in mice was shown to protect animals from becoming obese. Despite the connection between FTO and obesity, the precise role that FTO plays in this process is not well-understood, although regulation of the expression of nearby genes has been proposed (73). However, this explanation is independent of the biological function of the FTO protein as a 2-oxoglutarate-dependent RNA demethylase (64,74). GSK-3 is a major regulator of cellular metabolism and has been linked to obesity via its role in regulating the insulin/PI3K pathway. Interestingly, it was recently found that transgenic mice expressing a form of GSK-3 that is insensitive to inhibition by PI3K signaling were protected from developing metabolic syndrome, which includes obesity (75). Although it is not yet known if the role of FTO in obesity is connected to GSK-3 activity and mRNA methylation levels, it is a tantalizing possibility that should be explored further.
Although we identify a new mechanism that results in changes in mRNA methylation, the biological effects of these changes are likely to be complicated. Since the discovery that m 6 A tags on mRNA are reversible (43), it has become widely accepted that this modification results in decreased mRNA half-life. This is due to the recognition of m 6 A mRNA by the reader protein YTH domain family 2 (YTHDF2) (76). Yet numerous studies have shown that the m 6 A modification of mRNA affects nearly every step in its life cycle, and the resulting effect on m 6 A-modified mRNA depends on the specific YTH domain family reader protein that binds to the mRNA (77). Further complicating the fate of m 6 A-modified mRNA is the recent discovery that insulin-like growth factor-2 mRNAbinding proteins (IGFBP1-3) also bind to m 6 A-containing transcripts and have the opposite effect as YTHDF2, resulting in stabilization of mRNA (78). Taken together, this means that we are unable to speculate what effect reducing m 6 A marks has on the mRNAs affected in Gsk-3 DKO ESCs.
While this manuscript was in preparation, data from Jaffrey and colleagues (48) were published showing that FTO does not specifically demethylate m 6 A RNA, but instead prefers to demethylate m 6 A m . Our approach for detecting m 6 A relied on an antibody against m 6 A, but this antibody was shown to detect both m 6 A and m 6 A m (49). In addition, whereas the approach we used for m 6 A-Seq did not provide us with the precise locations of demethylated adenosines in RNA from Gsk-3 DKO ESCs, we nonetheless observed that hundreds of mRNAs had regions largely devoid of methylated adenosines. This strongly suggests thatdemethylationofm 6 A m byFTOisnottheonlyRNAdemethylation occurring in Gsk-3 DKO ESCs. Mauer et al. (48) also provided evidence that m 6 A demethylation is mediated by Alkbh5. Our m 6 A-Seq data raise the possibility that GSK-3 activity could be directly or indirectly regulating Alkbh5, in addition to FTO. We do not observe an increase in Alkbh5 protein or mRNA in Gsk-3 DKO ESCs (data not shown), but it is possible that either GSK-3 regulates Alkbh5 activity or that there is coordination between the demethylation of m 6 A m and m 6 A. Further investigation will be required to determine whether either of these possibilities is indeed true.
In summary, we have uncovered the first example of a mechanism by which mRNA demethylation can be regulated. This finding dovetails nicely with the recent discovery from Vallier and colleagues (79) that Activin/Nodal signaling via Smad2/ Smad3 regulates the activity of the Mettl3 mRNA methyltransferase complex. Interestingly, it has been shown that simultaneous inhibition of GSK-3 and Activin/Nodal signaling in mouse ESCs promotes pluripotency (80), yet at the time of publication, neither of these molecules were known to regulate mRNA methylation. In light of both the findings from Bertero et al. (79) and those described here by our group on the regulation of mRNA methylation, an intriguing possibility is that these experiments could be reinterpreted as attaining pluripotency via the simultaneous inhibition of mRNA methylation and promotion of mRNA demethylation. Further experiments will be able to directly test this hypothesis. Taken together, we provide strong evidence demonstrating that the effects of GSK-3 are not limited to canonical signaling pathways but can affect epigenetic processes as well.

GSK-3 regulation of m 6 A
protease inhibitor mixture (1:100; Sigma) and lysed by vigorous vortexing. For V5 immunoprecipitations, 10% of the lysate was set aside, whereas the remaining protein lysate was added to V5-agarose beads (Bethyl Laboratories). Lysates were incubated with beads for 2 h at room temperature while rotating and then centrifuged at 3000 ϫ g for 1 min at 4°C to pellet beads. Pellet was washed four times with lysis buffer containing protease inhibitor mixture (1:100; Sigma). Protein was eluted from the beads by adding 40 l of 1ϫ Tris/Tricine sample buffer and incubating for 5 min at 100°C. Beads were pelleted by centrifugation at 14,000 rpm for 1 min, and the supernatant was transferred to a fresh microcentrifuge tube and stored at Ϫ20°C until use. For FLAG immunoprecipitations, 10% of the lysate was set aside, whereas the remaining protein lysate was added to FLAG-magnetic agarose beads (Sigma). Lysates were incubated with beads for 4 h at room temperature while rotating, placed on the magnet, and supernatant was saved. Beads were washed three times in lysis buffer and then stored at 4°C.

RNA isolation, cDNA synthesis, and quantitative PCR
1 ϫ 10 6 ESCs/well were plated on a gelatin-coated 6-well plate. RNA was isolated using TRIzol reagent (Thermo Fisher Scientific) and extracted using Direct-zol RNA Miniprep columns (Zymo) following the manufacturer's protocol. RNA was quantified using a Nanodrop 2000 (Thermo Fisher Scientific).
2 g of total RNA was used to synthesize cDNA. cDNA was synthesized using the High Capacity Reverse Transcriptase kit

Enrichment of phosphoproteins
Total protein was isolated from WT and Gsk-3 DKO mESCs that were transfected with pCAGEN-FTO-V5, pCAGEN-FTO-FLAG, or pCAGEN-FTO-S249A/S253A-FLAG. Samples were enriched for phosphoproteins using the Pro-Q Diamond phosphoprotein enrichment kit (Molecular Probes) according to the manufacturer's protocol for nondenatured protein lysates. Lysates were then separated on 7.5% Tris/Tricine acrylamide gels, and Western blotting for V5 or FLAG was performed as described above.

In vitro kinase assay
Recombinant human GSK-3␤ (Sigma-Aldrich) was incubated with recombinant mouse FTO (Sigma-Aldrich) or recombinant 2N4R His-tagged Tau in the presence of fresh ATP for 1 h at 30°C. Reactions were then run on 7.5% Tris/ Tricine acrylamide gels, and then the gels were stained with Pro-Q Diamond for 1 h followed by destaining for 1 h to detect phosphorylated proteins. Gels were imaged using a Gel Doc XRSϩ (Bio-Rad) using the 560DF50 filter.

Protein mass spectrometry
Immunoprecipitated FLAG-FTO from WT and Gsk-3 DKO ESCs washed with 100 l of 100 mM ammonium bicarbonate (pH 8.0) three times while retained on FLAG magnetic beads. Cysteine residues were reduced and alkylated at 22°C using 10 mM DTT (Sigma-Aldrich) for 1.5 h followed by 25 mM iodoacetamide (Sigma-Aldrich) for 45 min in the dark. Proteins were digested on beads (1:20 enzyme/protein) with either trypsin or endoproteinase LysC (Promega) for 6 h at room temperature on a shaker. Peptides were removed with the supernatant and acidified to pH 2.5 using formic acid (Fisher).
Peptides were loaded directly onto a Waters nanoAcquity UPLC for separation on a 75 m ϫ 250-mm C18 BEH UPLC column (1.7-m particle size; Waters) coupled to a Thermo LTQ-Orbitrap Velos mass spectrometer. A linear 70 min, 0.3 l/min flow UPLC gradient was performed using 0.1% formic acid in water (solvent A) and 0.1% formic acid in acetonitrile (solvent B) with the following conditions: 2% solvent B hold for 17 min, 2-7% solvent B in 3 min, 7-32% solvent B in 25 min, 32-95% solvent B in 5 min, 95% solvent B hold for 4 min, 95 to 2% B in 1 min, 15-min hold at 2% B. MS scans were acquired at 60,000 resolution in the Orbitrap, and a top 10 data-dependent CID tandem MS method was used to acquire MS/MS scans in the ion trap. All data were searched against a Uniprot Mus musculus protein database using Mascot (Matrix Science) with the following parameters: 50-ppm peptide mass tolerance; 0.5-Da fragment ion tolerance; variable modifications including Gln to pyro-Glu conversion on N-terminal Gln, oxidation of Met, and phosphorylation of Ser/Thr/Tyr; fixed modification carbamidomethyl Cys; 1% false discovery rate; and two missed cleavages with trypsin enzyme specificity. All expected proteolyzed peptides were investigated by hand for the presence of any singly or doubly phosphorylated forms in the MS and MS/MS spectra. Phosphorylated peptides were sequenced by hand and confirmed by manual inspection of the MS/MS spectra. An additional LC-MS/MS experiment was conducted using the same parameters as above, plus a targeted MS/MS scan event at

Detection of ubiquitinated FTO
For detection of ubiquitinated FTO, we plated 6 ϫ 10 6 WT and Gsk-3 DKO ESCs per 10-cm plate. Cells on each plate were transfected with 5.5 g of pCAGEN-FLAG-FTO, 5.5 g of HAubiquitin, and 1 g of pMax-GFP using the PEI method. 24 h later, medium was changed, and cells were treated with 10 M lactacystin for 6 h. Cells were then collected in lysis buffer containing deubiquitination enzyme inhibitors (N-ethylmaleimide and TPEN (N,N,NЈ,NЈ-tetrakis(2-pyridinylmethyl)-1,2-ethanediamine)) to ensure recovery of ubiquitinated proteins. Lysates were separated on 7.5% Tris/Tricine acrylamide gels, and Western blotting for HA and FLAG was performed as described above.

m 6 A quantification
To quantify the amount m 6 A-modified mRNA, we first enriched for mRNA by isolating the population of poly(A) ϩ RNA from total RNA. 5 g of total RNA was isolated from WT and Gsk-3 DKO ESCs as described above and then incubated with oligo(dT) 25 magnetic beads (New England Biolabs) on a rotator overnight at 4°C. Tubes were then placed on a magnetic plate for 1 min, and supernatant was saved as "oligo(dT) other" for further analysis. Beads were washed with 100 l of RNasefree H 2 O three times. To elute RNA from beads, 20 l of cold 10 mM Tris-HCl, pH 7.5, was added to beads and incubated at 80°C for 2 min. Tubes were returned to the magnet, and eluted RNA was immediately transferred to a new tube and placed on ice. Poly(A) ϩ RNA was quantified using the Qubit HS RNA assay kit (Thermo Fisher Scientific). For m 6 A quantification, we used the method described by Jia et al. (43). 1 g of poly(A) ϩ mRNA was digested by 4 l of nuclease P1 (2 units) in 40 l of LC-MS/MS buffer (25 mM NaCl and 2.5 mM ZnCl 2 ) for 1 h at 37°C, followed by the addition of 1 M NH 4 HCO 3 and 0.5 units of alkaline phosphatase and incubation at 37°C for another 1 h. To confirm digestion of RNA, samples were run on a 1% agarose gel, with an undigested RNA sample as a control. Samples were then analyzed by UHPLC-MS (Vanquish UHPLC coupled online with a Q Exactive quadrupole-Orbitrap mass spectrometer, operated in positive-ion mode; Thermo Fisher Scientific) at the University of Colorado School of Medicine Proteomics Core Facility, and the ratios of methyladenosine to adenosine were reported. Retention times and linearity ranges were determined against inhouse libraries of light and deuterated standards available in the core for adenosine and methyladenosine (Sigma-Aldrich). m 6 A levels were also quantified using the EpiQuik m 6 A RNA methylation quantification kit (Epigentek) following the manufacturer's instructions. Briefly, 200 ng of total RNA from WT, Gsk-3 DKO, and Gsk-3 DKO/Gsk-3␤ rescue ESCs was bound to wells in a clear 96-well plate via RNA high-binding solution and then incubated with m 6 A capture and detection antibodies, followed by colorimetric detection of signal (A 450 nm ) using an EnSpire plate reader (PerkinElmer Life Sciences). The readings were then used to calculate the relative quantification using a formula provided by the manufacturer.

m 6 A-Seq
RNA was isolated from WT and Gsk-3 DKO ESCs using TRIzol (Thermo Fisher Scientific) as described above. 15 g of total RNA was used for mRNA isolation with oligo(dT) 25 magnetic beads (New England Biolabs). 7 g of poly(A) ϩ RNA from each sample was then incubated with Protein G magnetic beads (New England Biolabs) coupled with anti-m 6 A antibody (New England Biolabs, catalog no. E1611A, lot 0011602) rotating overnight at 4°C. Beads were washed at room temperature, two times with reaction buffer (150 mM NaCl, 10 mM Tris HCl, pH 7.5, and 0.1% Nonidet P-40), two times with low-salt reaction buffer (50 mM NaCl, 10 mM Tris HCl, pH 7.5, and 0.1% Nonidet P-40), and two times with high-salt reaction buffer (500 mM NaCl, 10 mM Tris HCl, pH 7.5, and 0.1% Nonidet P-40). Beads were then resuspended in Buffer RLT (Qiagen) and placed on a magnet, and then m 6 A mRNA-containing solution was removed and placed in new tubes. Samples were concentrated further using MyOne Silane Dynabeads (Thermo Fisher Scientific). m 6 A mRNA values obtained using Qubit were: as follows 122 ng (WT) and 163 ng (DKO). m 6 A mRNA was chemically fragmented by incubating in 100 mM Tris-HCl, pH 7.0, and 100 mM ZnCl 2 at 94°C for 5 min, and the reaction was stopped with 0.5 M EDTA (82). Fragmented RNA was cleaned and concentrated using a commercial kit (Zymo). m 6 A-Seq was performed using the IonProton sequencing platform (Thermo Fisher Scientific) at PrimBio (Exton, PA).

meRIP-qPCR
50 ng of WT and Gsk-3 DKO mRNA from cells grown without LIF before and after immunoprecipitation with anti-m 6 A antibody was used for meRIP-qPCR. Briefly, SuperScript IV VILO cDNA kit (Thermo Fisher Scientific) was used to synthesize cDNA using the manufacturer's protocol. cDNA was diluted to 0.5 ng/l for use in qPCR. Reactions were run on a StepOne real-time PCR system (Applied Biosystems) using PrimeTime gene expression master mix (IDT) and PrimeTime qPCR assays (IDT) for c-myc (Mm.PT.58.13590978) and Esrrb (Mm.PT.58.14246772). All Ct values were normalized to a mouse Gapdh endogenous control (Mm.PT.39a.1) (IDT), and relative quantification was calculated from the median Ct value.