Distinct RNA N-demethylation pathways catalyzed by nonheme iron ALKBH5 and FTO enzymes enable regulation of formaldehyde release rates

Significance

Nonheme iron enzymes FTO and ALKBH5 play central roles in epigenetic RNA regulation by catalyzing the oxidation of N6-methyladenosine (m6A) to produce N6-hydroxymethyladenosine (hm6A) and adenosine (A), respectively. Here, we provide a mechanistic rationale for these distinct biochemical outcomes by identifying that ALKBH5 performs m6A demethylation via an unprecedented covalent-based mechanism with concomitant and rapid release of A and formaldehyde (FA), whereas FTO liberates hm6A to release A and FA over longer timescales. This work reveals foundational biochemical differences between these closely related but nonredundant epigenetic enzymes and identifies ALKBH5 as an endogenous source of rapid formaldehyde generation in cells.

Abstract

The AlkB family of nonheme Fe(II)/2-oxoglutarate–dependent oxygenases are essential regulators of RNA epigenetics by serving as erasers of one-carbon marks on RNA with release of formaldehyde (FA). Two major human AlkB family members, FTO and ALKBH5, both act as oxidative demethylases of N6-methyladenosine (m6A) but furnish different major products, N6-hydroxymethyladenosine (hm6A) and adenosine (A), respectively. Here we identify foundational mechanistic differences between FTO and ALKBH5 that promote these distinct biochemical outcomes. In contrast to FTO, which follows a traditional oxidative N-demethylation pathway to catalyze conversion of m6A to hm6A with subsequent slow release of A and FA, we find that ALKBH5 catalyzes a direct m6A-to-A transformation with rapid FA release. We identify a catalytic R130/K132/Y139 triad within ALKBH5 that facilitates release of FA via an unprecedented covalent-based demethylation mechanism with direct detection of a covalent intermediate. Importantly, a K132Q mutant furnishes an ALKBH5 enzyme with an m6A demethylation profile that resembles that of FTO, establishing the importance of this residue in the proposed covalent mechanism. Finally, we show that ALKBH5 is an endogenous source of FA in the cell by activity-based sensing of FA fluxes perturbed via ALKBH5 knockdown. This work provides a fundamental biochemical rationale for nonredundant roles of these RNA demethylases beyond different substrate preferences and cellular localization, where m6A demethylation by ALKBH5 versus FTO results in release of FA, an endogenous one-carbon unit but potential genotoxin, at different rates in living systems.

The one-carbon (1-C) cycle is a universal hub in all living organisms that regulates processes spanning metabolism to epigenetics (1–6). Formaldehyde (FA) is a ubiquitous 1-C unit in this context with broad contributions to health and disease. Indeed, chronic exposure to elevated levels of FA is associated with various cancers and several neurological, diabetic, and cardiovascular diseases (6). However, FA is also endogenously produced by a diverse array of biochemical processes, such that substantial levels of FA are detectable in cells and animals (7–15), where it is metabolized by the ADH5 detoxification system (6). Included are pathways in folate-driven 1-C metabolism through serine hydromethyltransferase enzymes 1 and 2 (SHMT1 and SHMT2) to balance serine/glycine pools (2, 6, 16) as well as enzymes for epigenetic DNA and RNA regulation, such as nonheme Fe(II)/2-oxoglutarate (2OG)–dependent oxygenases (17–20) that serve as oxidative N-demethylases to erase 1-C marks from nucleic acid substrates.

The JmjC N^ε-methyl-lysine demethylases (KDMS) and the AlkB family represent two major types of epigenetic enzymes (20, 21), where N-methyl oxidation of nucleic acid substrates results in FA release over a range of timescales depending on the stability of the resulting N-hydroxymethyl product (22). In particular, N6-hydroxymethyladenosine (hm6A) formed by monooxidation of m6A (23) is relatively stable compared to other hydroxymethyl-containing nucleobases and has been reported to decay to the free adenosine (A) base over 10 h (22). m6A is the most prominent modification of messenger RNA (mRNA) (24, 25) and is installed by the S-adenosylmethionine–dependent METTL3/14-WTAP writer complex (26) and removed by two human AlkB eraser enzymes, fat mass and obesity-associated protein (FTO) (4) and AlkB family member 5 (ALKBH5) (27). Internal m6A modifications control the fate of mRNA (28–32) through translation, splicing, localization, stability, and decay and are connected to cancer progression, immune responses, and metabolic states (27, 29, 32–36).

Despite their global similarities, the FTO and ALKBH5 RNA demethylases are nonredundant and exhibit key differences in RNA substrate preferences and cellular localization (37, 38). As such we became interested in deciphering other foundational features that define these epigenetic regulatory enzymes, particularly in the context of FA as both a genotoxin and endogenous mediator of the 1-C cycle. In this study, we now report the identification of distinct mechanistic pathways in FTO and ALKBH5 that promote differing timescales of FA release. Oxidative demethylation of m6A by FTO produces hm6A as the major product and we identify a role for E234 in stabilizing the nascent hm6A product, which slowly decomposes to A and FA, by analysis of the reported structure of the FTO:N6-methyldeoxyadenosine (m6dA) complex (Protein Data Bank [PDB] ID code 5ZMD) (39). In contrast, we find that ALKBH5 catalyzes a direct m6A-to-A conversion with concomitant rapid release of FA, undergoing a single nonheme Fe(II)-2OG–dependent oxidation and requiring one 2OG cofactor to produce one succinate molecule per m6A demethylation event. A key R130/K132/Y139 triad in the nucleotide binding region is responsible for the unique capacity of ALKBH5 to catalyze FA release, with direct detection of an unprecedented protein:substrate covalent intermediate that relies on K132 and Y139. Interestingly, a K132Q ALKBH5 mutant abolishes the covalent-based mechanism and alters the m6A demethylation profile to resemble that of FTO. Finally, we show that ALKBH5 is an endogenous source of FA in cells by selective small interfering RNA (siRNA) knockdown and activity-based sensing of this 1-C unit. Taken together, our findings reveal a molecular basis for the divergent biochemical pathways of N-demethylation and FA production catalyzed by FTO and ALKBH5, providing a starting point for further investigations into how and why epigenetic regulation is connected to 1-C metabolism through this dynamic 1-C unit.

Results and Discussion

FTO and ALKBH5 Exhibit Distinct m6A Demethylation Profiles In Vitro.

We first performed the characterization and quantification of the intermediates and products formed upon m6A demethylation catalyzed by FTO and ALKBH5. Quantification was performed by monitoring the unique mass fragment transition of respective nucleoside standards (SI Appendix, Fig. S1 A–E), modified from reported methods (23, 40) using UltraPerformance LC (UPLC)–triple quadrupole LC tandem mass spectrometry (MS/MS). A and m6A nucleosides were commercially obtained (SI Appendix, Fig. S1 A and B). hm6A was accessed by incubating A with 50× molar excess FA overnight at 37 °C. hm6A was obtained by subtracting unreacted A from known starting concentration of A standards (SI Appendix, Fig. S1C). f6A standard was synthesized in-house by modification of a reported procedure (23, 40) (SI Appendix, Fig. S1D). The enzymatic reactions were monitored at 1, 5, 10, 30, and 60 min. FTO (5 µM) gave ca. 72% hm6A, 6% f6A, and 18% A after 10 min of incubation with 10 µM 5′-UUGCGG(m6A)CUCCUGUUG-3′ (hereafter known as m6A oligo) at 37 °C (Fig. 1A). Indeed, hm6A persisted (half-life [t_1/2] ∼3 h; SI Appendix, Fig. S2A) as the major product and f6A (t_1/2 ∼7 h, SI Appendix, Fig. S2B) as a minor product of FTO m6A demethylation over a 60-min time course. In contrast, ALKBH5 (3.5 µM) was incubated with 10 µM m6A oligo at room temperature and >98% demethylation was observed after 10 min (Fig. 1B). Negligible amounts of hm6A and f6A were detected during the monitored time points. We observe a stark contrast between the m6A demethylation profiles of FTO and ALKBH5, consistent with previous reports (23, 41). FTO behaves like a classical nonheme Fe(II)-2OG–dependent dioxygenase by performing stepwise oxidation, reminiscent of the TET family of enzymes (42), whereas ALKBH5 catalyzes a unique direct m6A-to-A conversion.

Fig. 1.

Time course profile of m6A demethylation by (A) FTO and (B) ALKBH5. Products formed at various time points (0, 1, 5, 10, 30, and 60 min) where (A) 5 µM FTO and (B) 3.5 µM ALKBH5 were incubated with 10 µM of m6A oligo at 37 °C and room temperature, respectively. The individual concentrations of A, m6A, hm6A, and f6A quantified at each time point were expressed as a percentage (y axis), where the sum of the concentrations of A, m6A, hm6A, and f6A at each time points adds up to 100%. Error bars indicate the mean ± SD (n = 3 biological replicates).

ALKBH5 m6A Demethylation Requires One Canonical Iron- and 2OG-Dependent Oxidation and Directly Liberates FA.

With the observation of direct production of A by ALKBH5, we first sought to decipher how many nonheme Fe(II)- and 2OG-dependent oxidations occur per m6A demethylation by ALKBH5. Because the formation of one nonheme Fe(IV) oxo species consumes one 2OG and produces one succinate molecule, we quantified the amount of succinate converted from 2OG and the corresponding amount of m6A and A produced at 2, 4, and 6 min in vitro. Importantly, we did not detect significant amounts of hm6A or f6A under these experimental conditions [1.3 µM ALKBH5, 40 µM Fe(II), and 40 µM 2OG with 10 µM m6A oligo substrate]. AlkB enzymes are known to perform nonspecific 2OG turnover in the presence of Fe(II) and 2OG in the absence of substrate, and excess succinate is also known to compete with 2OG for cofactor binding. Various enzymatic conditions were evaluated to ensure that the in vitro reaction was not saturated (e.g., exhibited a linear trend), with less than 25% of m6A consumed over the course of the experiment in order to maximize ALKBH5–m6A engagement and to minimize nonspecific 2OG turnover. Succinate was quantified by multiple reaction monitoring (MRM) of the unique mass transition 117 → 73.3 (43) and retention time (∼2 min) (SI Appendix, Fig. S3 A and B). A succinate calibration curve showing linearity within the concentration range of succinate generated in vitro was also prepared alongside the same experimental sample to ensure accurate quantification. The corresponding concentrations of A and m6A were then quantified using the same set of samples. During the course of this reaction (SI Appendix, Fig. S3C), the rate of succinate production (0.49 µM/min) was found to be nearly equal to the rate of m6A demethylation (0.45 µM/min) and A production (0.48 µM/min) (Fig. 2A). These data establish that ALKBH5 undergoes one canonical nonheme Fe(II)-2OG–dependent oxidation per m6A demethylation event.

Fig. 2.

Characterization of stoichiometry of ALKBH5-catalyzed m6A demethylation. (A) Quantification of succinate and A produced per m6A demethylation event at 2, 4, and 6 min of reaction. The y axis (concentration in micromolar) represents the changes in concentrations of m6A, A, and succinate across 2-, 4-, and 6-min time points. Over the course of 2 to 6 min in vitro reaction time points monitored, the concentration of m6A decreased in a linear manner by ∼1.8 µM, and both concentrations of succinate and A increased in a linear manner by ∼1.9 µM. Reaction conditions: 1.3 µM ALKBH5 was incubated with 40 µM Fe(II), 40 µM 2OG, and 10 µM m6A oligo at room temperature. The rate of succinate produced (∼0.487 µM/min) is nearly equal to the rate of m6A demethylation (∼0.451 µM/min) and A production (∼0.478 µM/min). Error bars indicate the mean ± SD (n = 4, 2 biological replicates × 2 technical replicates). (B) In vitro time-course tracking of FA produced over the monitored time points (0, 1,5, 10, and 15 min). The amount of FA is quantified and presented as a percentage (y axis) of the sum of the concentrations of A, m6A, hm6A, and f6A at each time point. Error bars indicate the mean ± SD (n = 3 biological replicates).

As the stoichiometry for this ALKBH5-catalyzed m6A demethylation is expected to generate one equivalent of FA per turnover to A, we measured the production of FA with respect to m6A demethylation and A formation at 0, 1, 5, 10, and 15 min. Trans-1,2-diaminocyclohexane was used to derivatize FA for quantification via monitoring MRM of the unique mass transition 126.9 → 98 and retention time (∼1.3 min) (SI Appendix, Fig. S4 A–C). A calibration curve was prepared with known concentrations of FA and linearity was ensured within the range of measurement. A control experiment was performed in parallel using control substrate, 5′-UUG​CGG​ACU​CCU​GUU​G-3′ (hereafter known as A oligo), to account for background noise (SI Appendix, Fig. S4 D and E). Using this method, we were able to account for ca. 90% of the FA produced with respect to m6A consumed and A formed at each time point after 1 min (Fig. 2B). Notably, the amount of FA detected closely mirrored the formation of A. As FA is a highly reactive carbonyl species, we speculated that the small fraction of unaccounted FA may be consumed by side reactions with the cosubstrates and/or buffer components. Nevertheless, these experiments establish that FA is released rapidly and concomitantly upon m6A demethylation by ALKBH5. In contrast, the majority of the single-carbon unit for FTO-catalyzed m6A demethylation is trapped in the form of hm6A. In this case, the rate of FA release by FTO m6A demethylation is an inverse function of the stability of hm6A.

To rule out other possibilities such as a second oxidation of m6A via the intermediacy of hm6A to form f6A, which may liberate formate after hydrolysis, we sought to detect this 1-C unit. Formate was first derivatized as reported with slight modification (44) (SI Appendix, Fig. S5 A and B), quantified using calibration curves prepared with known concentrations of formate, and linearity was ensured within the concentration range of measurement. We performed the experiments using ALKBH5 and a catalytically dead ALKBH5 H204A mutant and measured the amount of formate after 60 min (SI Appendix, Fig. S5 C and D). Ten micromolar m6A oligo was 98% consumed by ALKBH5 while a negligible amount of m6A was demethylated by the catalytically dead ALKBH5 H204A mutant, but no detectable amount of formate was measured in either experiment. This negative result is consistent with the aforementioned observations that support a stoichiometry of one nonheme Fe(II)-2OG–dependent oxidation with one equivalent of 2OG cofactor consumed and one equivalent of FA and A formed per m6A oxidative demethylation event catalyzed by ALKBH5.

Active-Site Analysis of FTO and ALKBH5 Provides a Molecular Rationale for the Observed Differences in Product Profiles upon Oxidative m6A Demethylation.

Inspection of the previously reported crystal structure of FTO:m6dA (PDB ID code 5ZMD) (39) provides a molecular rationale for why hm6A rather than A is produced as the major product of m6A oxidation catalyzed by FTO. Analysis of the structure shows that the m6dA nucleobase forms hydrogen bonds between R96 and the N1 endocyclic nitrogen and between E234 and the N7 endocyclic nitrogen (Fig. 3A). These hydrogen bonds would imbue the nucleobase with a partial positive charge, inductively and mesomerically lowering the highest-occupied molecular orbital of the N6 lone pair and thereby disfavoring Schiff base formation (Fig. 3C), which we posit as a key step in the demethylation mechanism of ALKBH5 (discussed below). ALKBH5, in contrast, has a proline residue (P207) where FTO has E234. Once m6A is oxidized, hm6A may either dissociate from the FTO nucleotide binding region or undergo a second oxidation to produce f6A as a minor product, but Schiff base formation is unlikely (Fig. 3C). As attempts to obtain a structure of ALKBH5 complexed to a m6A substrate have yet to prove fruitful, we superimposed the double-stranded beta helical (DSBH) domain of the FTO:m6dA complex with the crystal structure of ALKBH5 (PDB ID code 4NRP) (41) (SI Appendix, Fig. S6) to probe how ALKBH5 interacts with the m6A nucleobase. The DSBH domains, the conserved catalytic triad (HXD…H motif, where X can be any amino acid), metal ions, and N-oxalylglycines (NOG) of ALKBH5 and FTO generally align well (rmsd = 4.84 Å, by PyMOL) (SI Appendix, Fig. S6). In the FTO:m6dA complex, V94, R96, and E234 lie in plane with the m6dA nucleobase (Fig. 3A and SI Appendix, Fig. S6). The corresponding residues in ALKBH5 are R130, K132, and P207, respectively (Fig. 3B and SI Appendix, Fig. S6), and these three residues are in plane with the aligned m6dA nucleobase. In addition, V94 and R96 of FTO and R130 and K132 of ALKBH5 are part of the nucleotide recognition loop 1 (NRL1). Together with nucleotide recognition loop 2 (NRL2), these loops are characteristic structural features located at the N-terminal of human AlkB DSBH domain and are known to play a role in substrate recognition (19, 45, 46). As a hydrophobic residue, the P207 of ALKBH5 is unable to have any polar interactions with the m6dA nucleobase. Instead, we suggest that m6A recognition by ALKBH5 is facilitated by hydrogen bond interactions between R130 and the N3 endocyclic nitrogen and K132 and the N1 endocyclic nitrogen (Fig. 3B and SI Appendix, Fig. S6). In addition, we also noticed that Y108 of FTO (Fig. 3A) and Y139 of ALKBH5 (Fig. 3B), both residues within the NRL2, are proximal to the N6-methyl group. This observation led us to propose that the unique m6A demethylation activity of ALKBH5 arises from the interplay of R130, K132, and Y139 with an hm6A intermediate.

Fig. 3.

Structural analysis of FTO and ALKBH5 provides insight into m6A demethylation biochemistry. (A) Nucleobase recognition region of FTO:m6dA cocrystal structure (PDB ID code 5ZMD, purple). Key residues which facilitate understanding of the FTO m6A demethylation mechanism are in yellow sticks. (B) Superimposition of ALKBH5 (PDB ID code 4NRP, pale orange) with FTO:m6dA cocrystal structure, with FTO removed for clarity. Key residues which facilitate understanding of the unique m6A demethylation mechanism of ALKBH5 are in pink sticks. (C) A graphical illustration of the FTO nucleobase recognition region interacting with hm6A. An additional hydrogen bond with E234 should disfavor Schiff base formation and explains why hm6A is the major product of FTO. (D) A graphical illustration of how K132 and Y139 of ALKBH5 in the nucleobase recognition region facilitate formation of a covalent intermediate via elimination of water.

Identification of a Key R130/K132/Y139 Catalytic Triad for ALKBH5-Catalyzed m6A Demethylation.

These structural analyses led us to hypothesize that R130, K132, and Y139 play key roles in the mechanism by which ALKBH5 catalyzes m6A demethylation. As a component of the NRL1, R130 has a clear role in m6A nucleobase recognition. In view of its orientation and proximity, we hypothesized that the proximal basic K132 residue could facilitate dehydration of the N6-hydroxymethyl group formed upon m6A oxidation by serving as a hydrogen bond donor for the hydroxyl leaving group (Fig. 3D). The nascent Schiff base may then be attacked by the now deprotonated, nucleophilic K132 or by Y139, both of which sit nearby, to form a covalent protein:substrate intermediate. After proton transfer, cleavage of this nascent aminal (for K132) or mixed aminal (for Y139) intermediate to form an iminium or oxocarbenium ion, followed by hydrolysis, would liberate FA.

In order to probe this mechanistic hypothesis, we systematically mutated R130, K132, and Y139 residues and measured their effects on enzymatic activity and product distribution. The integrity of the mutants’ DSBH domain, which houses the catalytic iron center (45), was probed using a thermal shift assay (SI Appendix, Fig. S7 A–D). For the ALKBH5 wild-type protein and mutants, the thermal stability increased (curve shifted right) upon incubation with the generic inhibitor NOG relative to the apo form, indicating that these mutations only perturb the nucleotide recognition domain and not the catalytic iron center. Indeed, the catalytically dead ALKBH5 H204A mutant, which is unable to chelate iron, did not display a thermal shift when incubated with NOG, reflecting NOG specificity in probing the integrity of the catalytic iron center (SI Appendix, Fig. S7E).

We incubated 5 µM and 10 µM ALKBH5 (R130Q) with 10 µM of m6A oligo for 1 h at 37 °C and observed no enzymatic activity (SI Appendix, Fig. S8). In light of its position within the NRL1, we reasoned that the positively charged R130 is crucial for m6A nucleobase recognition through binding via an electrostatic and/or hydrogen bonding interaction with the N3 endocyclic nitrogen lone pair. Although Q130 can hydrogen-bond to the N3 endocyclic nitrogen, its charge neutral and shorter side chain may attenuate the strength of a hydrogen-bond interaction.

Next, we tested the enzymatic activity and product distribution of ALKBH5 (K132Q). Mutation of K132 to Q132 removes the availability of a relatively acidic ammonium proton near the N6-methyl group but still retains the hydrogen bond interaction capacity, although perhaps weakening it due to a shorter side chain. Interestingly, 5 µM ALKBH5 (K132Q) incubated with 10 µM m6A oligo for 1 h at 37 °C resulted in the accumulation of 5′-UUGCGG(hm6A)CUCCUGUUG-3′ (hereafter known as hm6A oligo) as the major product (Fig. 4A), where ca. 56% of the m6A substrate was oxidized with accumulation of ca. 40% hm6A, 15% A, and <1% f6A at the 10-min time point. The amount of hm6A peaked at the 30-min time point at ca. 56%. At the 60-min point, ca. 92% of m6A substrate had been oxidized with accumulation of ca. 55% hm6A, 35% A, and <2% f6A. The activity of ALKBH5 (K132Q) mutant is more sluggish than its wild-type counterpart and its product profile is similar to that of FTO, with hm6A as the major product. The slower kinetics of K132Q could be attributed to the lysine residue also playing a role in m6A nucleobase recognition and/or substrate positioning for methyl oxidation by affording an electrostatic interaction with the N1 endocyclic nitrogen lone pair (Fig. 3D). Neutral glutamine has a shorter side chain and affords a weaker hydrogen bonding interaction. These data highlight the importance of K132 in the ALKBH5 direct m6A demethylation to A.

Fig. 4.

Identification and characterization of a covalent intermediate in m6A demethylation catalyzed by ALKBH5. (A) Time-course profile of 10 µM m6A oligo incubated with 5 µM ALKBH5 (K132Q) at 37 °C. Products formed at various time points (0, 1, 5, 10, 30, and 60 min) were monitored. The individual concentrations of A, m6A, hm6A, and f6A quantified at each time point were expressed as a percentage (y axis) where the sum of the concentrations of A, m6A, hm6A, and f6A at each time point adds up to 100%. Error bars indicate the mean ± SD (n = 3 biological replicates). (B) Native PAGE gel assay to visualize the covalent intermediate via multiplex fluorescent imaging. Representative runs included ALKBH5 and control proteins (FTO and three ALKBH5 mutants). RNA oligomers are 3′ biotinylated for tracking and imaging. ALKBH5 was imaged using primary anti-ALKBH5 (rabbit) with a secondary antibody Alexa Fluor 546. The 3′-biotinylated RNA oligos were read out using a streptavidin conjugated to Alexa Fluor 647.

Finally, we tested the activity of the ALKBH5 (Y139F) and found that it exhibited low reactivity but produced A as the major product. Negligible amounts of hm6A and f6A were detected. The extent of m6A oxidation was stoichiometric with respect to the input of ALKBH5 (Y139F) after 1 h at 37 °C (SI Appendix, Fig. S9 A and B). The hydroxyl group of tyrosine presumably accelerates one or more elementary steps in the demethylation sequence, including the rate-limiting step, since the product profile does not change but the rate of the reaction decreases dramatically with the Y139F mutation. We speculate that Y139 may also affect substrate recognition and/or binding.

Direct Observation of a Covalent Intermediate in ALKBH5-Catalyzed m6A Demethylation.

We next posited that the direct demethylation of m6A by ALKBH5 via oxidation of m6A to hm6A might proceed through a covalent intermediate owing to the well-positioned R130/K132/Y139 catalytic triad. In this case, cleavage of the N6 carbon unit by Fe(II)/2-OG–catalyzed oxidation could result via a covalent acetal/aminal adduct with participation of K132 and/or Y139, with subsequent hydrolysis liberating the single N6 carbon of hm6A as FA. This rapid and coordinated direct demethylation mechanism makes it challenging to trap a significant amount of a putative covalent intermediate for detection and visualization. We hypothesized that we might observe a persistent covalent intermediate if we charged active ALKBH5 with hm6A instead of m6A, reasoning that using this substrate at the next higher oxidation level would result in further oxidation of hm6A to f6A with reformation of Fe(II) in situ, but that carbon removal from f6A in the presence of Fe(II) would be slower relative to carbon removal from hm6A in the presence of Fe(II), as f6A could form more stable orthoaminal intermediates, with unsaturated or hydrated intermediates also possible due to K132 and/or Y139 interactions (SI Appendix, Fig. S10). Therefore, we incubated active ALKBH5 [loaded with Fe(II) and 2OG to facilitate formation of the active Fe(IV)=O oxidant in situ] directly with hm6A, anticipating formation of Fe(II) and f6A and formation of a covalent adduct with K132 and/or Y139.

We thus sought to probe for a covalent adduct at the ALKBH5 active site upon incubation with hm6A and designed a native polyacrylamide gel electrophoresis (PAGE) gel-based assay to detect a potential covalent intermediate (SI Appendix, Fig. S10A). To prepare hm6A, we incubated 5 µM FTO with ∼10 µM m6A oligo-3′-biotin for 10 min at 37 °C to afford ∼6 µM hm6A oligo-3′-biotin. We then quenched the reaction by snap-freezing it in cryogenic nitrogen and applied freeze–thaw cycles (i.e., warming up to room temperature and then snap-freezing) three times, in order to denature the FTO. Without further purification of the reaction mixture, we incubated 2 µM ALKBH5 with an estimated final concentration of 3 µM hm6A oligo-3′-biotin at room temperature for 1 h. The reaction mixture was then subjected to native PAGE gel electrophoresis at 4 °C. As a negative control, we used A oligo-3′-biotin subjected to the same experimental workflow and conditions. The protein and the nascent covalent intermediate were transferred to a polyvinylidene difluoride membrane and visualized using multiplex fluorescence imaging.

We expected the covalent intermediate to migrate much faster than its apo ALKBH5 counterpart in a native gel environment due to the higher negative charge-to-mass ratio imbued by the multinegatively charged phosphate backbone of the 16-mer RNA oligonucleotide. Indeed, we observed a supershifted ALKBH5 band in the sample incubated with hm6A but not with the control sample. The fluorescence bands of the supershifted ALKBH5 and the 3′-biotinylated RNA superimposed perfectly (Fig. 4B). This result confirmed that ALKBH5 forms a covalent adduct with the hm6A oligo-3′-biotin substrate. Moreover, control experiments performed with FTO as well as ALKBH5 R130Q, ALKBH5 K132Q, and ALKBH5 Y139F mutants in the same manner did not give any supershifted bands in the hm6A and control samples, indicating that these three key residues are required to form the detected covalent intermediate. The faint supershifted 3′-biotin RNA bands that are present in all lanes are likely experimental artifacts in light of the weak RNA binding affinity of FTO (40).

We performed intact MS experiments on ALKBH5 to further characterize covalent adduct formation. We incubated 2 µM ALKBH5 with ∼3 µM final concentration of 5′-UUGCGG(hm6A)CUCCAGAUG-3′ (hereafter known as hm6A oligo-2A) and hm6A oligo, respectively. Both hm6A oligos were generated as described above using FTO and the corresponding m6A counterpart. Along with an intense signal for apo ALKBH5 at 28,703 Da, we identified an intense signal of 33,857 Da for hm6A oligo-2A and 33,812 Da for hm6A oligo (SI Appendix, Fig. S11 A, B, and D), and a difference of 45 Da was measured experimentally. The expected difference between the most intense signals of the two covalent adducts is 46.1 Da, based on the difference in molecular weight between two adenosines and two uridines. The experimental value of 45 Da is within a ±1-Da error range of the theoretical value and within the accuracy range of the MS. Furthermore, the deconvoluted peaks of the covalent adducts are separated by 15 to 16 Da (SI Appendix, Fig. S11 C and E). These additional masses are consistent with incorporation of oxygen atoms into the molecule and may arise via nonspecific oxidations of proximal residues by the nonheme Fe(II) center (47). Fe(II) and 2OG were introduced together with hm6A substrate to facilitate ALKBH5 oxidation of the putative covalent intermediate. Together, these experiments are consistent with detection of a covalent adduct of ALKBH5 and f6A, arising from hm6A oxidation (SI Appendix, Fig. S10B).

Proposed Mechanism for Direct m6A-to-A Conversion Catalyzed by ALKBH5.

These observations are consistent with a mechanistic model in which K132 promotes Schiff base formation on hm6A, which may then undergo subsequent nucleophilic attack by K132 or Y139 (SI Appendix, Fig. S18). The absence of a supershifted band for ALKBH5 Y139F mutant and its proximity to the nascent iminium upon K132 catalyzed dehydration could suggest participation in nucleophilic attack, but Y139 may alternatively play a role in nucleobase recognition via hydrogen bonding to the N6 nitrogen (Fig. 3D) such that the absence of Y139 disfavors optimal orientation for nucleophilic attack by K132. The absence of a supershifted band for the K132Q ALKBH5 mutant may be attributed to the inability of this mutant to form a Schiff base and is consistent with the observation of hm6A as the major enzymatic product of m6A oxidation. Although it is clear that the interplay between these three residues is required for covalent intermediate formation, we cannot definitively assign which nucleophilic residue initiates the attack on the Schiff base (Fig. 3D). Experiments to probe which residue is the initial nucleophile are complicated by the possibility that a mixed aminal reversibly forms between K132 and Y139 after C–C bond cleavage and prior to release of A (Fig. 5 and SI Appendix, Fig. S18), but note that the K132Q mutant has an altered product distribution whereas the Y139F mutant maintains the same product distribution as wild type is consistent with K132 participating in Schiff base formation.

Fig. 5.

Proposed mechanistic pathways that lead to fast release of FA catalyzed by ALKBH5. Route I (orange), where K132 attacks the Schiff base. Route II (blue), where Y139 attacks the Schiff base. Elimination of A via Route I yields an iminium while via Route II yields an oxocarbenium. Intermediate formation of a methylene bridge is probable, as is transfer of the single carbon unit (red) to Y139 for eventual FA release. Note: All steps are chemically reversible, but reactions are depicted in the forward direction for simplicity.

If K132 were to first perform the nucleophilic attack at the Schiff base, the mechanism would proceed through route I (Fig. 5 and SI Appendix, Fig. S18) for A elimination. In the alternative scenario, if Y139 were to first perform the nucleophilic attack at the Schiff base, the mechanism would proceed through route II (Fig. 5 and SI Appendix, Fig. S18). Since both intermediates are labile and K132 and Y139 are adjacent to each other in space, formation of a methylene bridge between K132 and Y139 is a probable intermediate prior to hydrolysis and may facilitate release of A. In an effort to further probe putative K132 iminium and/or Y139 oxocarbenium intermediates (cf. Fig. 5), we treated ALKBH5 with sodium borohydride (NaBH₄) midreaction with the m6A oligo substrate to trap these intermediates as stable methyl adducts. Indeed, we observed a methyl group on K132 upon digestion and analysis by tandem LC-MS/MS (SI Appendix, Fig. S19). This observation is consistent with covalent bond formation at K132 but cannot exclude the transient involvement of Y139. Based on our analysis (discussed below) on the position of Y139 on NRL2, the single-carbon unit may end up on Y139 prior to release of FA. Although the oxocarbenium arising from Y139 and FA condensation would be less stable, it is also more reactive than the iminium arising from K132, so the Curtin–Hammett principle may govern the kinetics of FA release. The position of Y139 in NRL2 may also facilitate liberation of FA. In ALKBH5, K132 lies within a rigid beta strand of NRL1 that forms part of the major beta sheet of the DSBH motif and Y139 lies within the flexible NRL2 region (Fig. 3B and SI Appendix, Fig. S12). Both residues are proximal (∼6.5 Å radius) to the catalytic nonheme Fe(II) center. This assertion is supported by the observation that the NRL2 (amino acids 141 to 148, colored pink) of ALKBH5 (PDB ID code 4NRP) is disordered in the crystal structure (SI Appendix, Fig. S13), but NRL2s (colored pink) in the crystal structures of AlkB homologs (Escherichia coli AlkB, ALKBH1-3, and FTO) (39, 48–51) (SI Appendix, Fig. S13) are not disordered and the amino acid residues are resolved. Furthermore, disordered residue prediction of full-length AlkB homologs using IUPred2A server (52), which is based on biophysical models of molecular interactions, supports that the NRL2 region of ALKBH5 has the highest probability of being disordered as compared to other AlkB homologs (SI Appendix, Fig. S14 A–F). Therefore, it is possible that the flexible NRL2 could “swing” outward to expose a Y139 oxocarbenium ion/hemiacetal intermediate to the aqueous environment to facilitate hydrolysis and liberation of FA.

ALKBH5 Is an Endogenous Source of FA in Live Cells.

Finally, we inquired whether direct conversion of m6A to A and FA catalyzed by ALKBH5 serves as a dynamic source of free FA at the cellular level via activity-based sensing of FA fluxes in live cells with genetic perturbation of ALKBH5. Using FAP573 (14), a fluorescent activity-based sensing FA probe developed previously by our laboratory, we compared FA levels in cells with siRNA-mediated knockdown of ALKBH5 (ALKBH5-KD) versus negative control siRNA. The specificity of siALKBH5 was confirmed using Western blot analysis (SI Appendix, Fig. S15 A and B) and the suitability of anti-ALKBH5 for the immunostaining experiment was verified using a HEK293 (parental line) stably expressing N-terminal myc-tagged ALKBH5. Signal from myc immunostaining and ALKBH5 immunostaining colocalized with DAPI nuclear marker, reflecting ALKBH5 nuclear localization and anti-ALKBH5 specificity (SI Appendix, Fig. S16). HEK293 cells were transfected with 10 nM negative control siRNA and siALKBH5, respectively, and at ∼48 h posttransfection detection of FA with the FAP573 probe was conducted. As anticipated, FA-dependent FAP573 fluorescence signal intensities (Fig. 6 A and B) and levels of ALKBH5 expression (Fig. 6 C and D) in the ALKBH5-KD cells were decreased relative to the control cells. Increasing siRNA doses to 30 nM showed even larger decreases in FAP573 fluorescence signal intensity (Fig. 6 E and F) and ALKBH5 knockdown (Fig. 6 G and H). Analogous experiments were performed in the MCF-7 breast cancer cell line since ALKBH5 is known to play a role in stem cell proliferation in breast cancer (27). MCF-7 cells were transfected with 30 nM of negative scrambled control siRNA or siALKBH5, and as was observed in the HEK cell models, FA-dependent FAP573 fluorescence signal intensity (Fig. 6 I and J) and ALKBH5-dependent immunostaining (Fig. 6 K and L) levels were lower in ALKBH5-KD cells relative to negative control siRNA. Taken together, these experiments establish that ALKBH5 expression, which decreases m6A levels in cells by erasing 1-C RNA marks (33), also affects the amount of free FA production as this nonheme iron enzyme produces one equivalent of FA per demethylation event.

Fig. 6.

Activity-based sensing with FAP573 shows ALKBH5-dependent FA production in live cells. (A–D) ALKBH5 KD in HEK293 (parental line) transfected with 10 nM siRNA. (E–H) ALKBH5 KD in HEK293 (parental line) transfected with 30 nM siRNA. (I–L) ALKBH5 KD in MCF-7 cell line transfected with 30 nM siRNA. (A, E, and I) Live-cell imaging of FAP573 fluorescence was performed ∼48 h posttransfection of negative control siRNA (−ve siRNA, top row) and siALKBH5 (bottom row). FAP573 fluorescence (Left) and corresponding bright-field images (Right). (B and F) Data of fluorescence intensities from 4 wells × 2 regions of interest (ROIs) per well (n = 8). Horizontal bar is the average of the eight data points. For data presentation, FAP573 fluorescence data points for −ve siRNA are set to 0 and FAP573 fluorescence data points for siALKBH5 are presented as an offset from 0 (*P < 0.0315, ***P < 0.0005). (C, G, and K) Live cells used for FAP573 experiment were fixed for ALKBH5 immunostaining. −ve siRNA (top row) and siALKBH5 (bottom row). ALKBH5 immunofluorescence (Left) and corresponding DAPI + bright-field images (Right). (D and H) Data of fluorescence intensities from 4 wells × 2 ROIs per well (n = 8). Horizontal bar is the average of the eight data points. For data presentation, ALKBH5 immunofluorescence data points for −ve siRNA are set to 0 and ALKBH5 immunofluorescence data points for siALKBH5 are presented as an offset from 0 (****P < 0.0001). (J) Data of fluorescence intensities from 4 wells × 3 ROIs per well (n = 12). Horizontal bar is the average of the 12 data points. For data presentation, FAP573 fluorescence data points for −ve siRNA are set to 0 and FAP573 fluorescence data points for siALKBH5 are presented as an offset from 0 (**P < 0.0045). (L) Data of fluorescence intensities from 4 wells × 3 ROIs per well (n = 12). Horizontal bar is the average of the 12 data points. For data presentation, ALKBH5 immunofluorescence data points −ve siRNA are set to 0 and ALKBH5 immunofluorescence data points for siALKBH5 are presented as an offset from 0 (****P < 0.0001). (Scale bars, 50 µm in all images.)

Concluding Remarks

We have identified a molecular basis for distinct pathways for oxidative N-demethylation of RNA catalyzed by the nonheme iron enzymes FTO and ALKBH5 involved in epigenetic regulation that promote varying release rates of FA, a potent genotoxin that is endogenously produced in significant amounts in cells. We discovered that ALKBH5 directly demethylates m6A to A with concomitant and rapid FA formation, whereas FTO gives hm6A as a major product (Fig. 1A) and f6A as a minor product. The t_1/2 of hm6A (∼3 h) produced by FTO leads to FA release over longer timescales. The collective data are consistent with a model in which ALKBH5 catalyzes m6A-to-A conversion by an unprecedented covalent intermediate mediated by a key R130/K132/Y139 catalytic triad, which is conserved among human, mouse, and zebrafish (SI Appendix, Fig. S17), with rapid formation of both A and FA.

This work provides further foundational evidence for nonredundant roles of these major RNA demethylase eraser enzymes that have broader biological implications. Indeed, emerging data show that FTO and ALKBH5 exhibit distinct substrate preferences, including different sites on mRNA, transfer RNA, and small nuclear RNAs (4, 27, 40, 53, 54), with FTO having a broader substrate preference (m6A, hm6A, N1-methyladenosine [m1A] and N6,2′-O-dimethyladenosine [m6Am]) whereas ALKBH5 is more specific toward m6A in mRNA. Such differences have spawned a field of RNA epigenetics. Our findings point to differences in FA production as another factor that distinguishes these classes of enzymes, as the rate of FA released into the cellular environment depends on the stability of the N-hydroxymethylated intermediate (hm6A). The distinct m6A demethylation mechanisms of FTO and ALKBH5 can affect the spatiotemporal dynamics of FA availability in cells. In the case of ALKBH5, the single-carbon unit in m6A can be discharged rapidly as FA to the immediate environment. For FTO, FA is discharged/released slowly and at a distance proportional to diffusion of hm6A away from the location of m6A demethylation, since the 1-C unit is released in the form of stable hm6A product. Taking into account N-hydroxymethylated nucleobases more broadly, it has been shown that the stability can range from hours (hm6A) to seconds (N3-hydroxymethylthymidine [hm3T] and N3-hydroxylmethyluridine [hm3U]) under physiological conditions (22), which can in turn be potential cellular sources of FA. Moreover, unlike lysine demethylation in histones catalyzed by lysine-specific histone demethylase 1 (LSD1), where FA is sequestered by the cosubstrate tetrahydrofolate upon release (55), it is tempting to speculate that ALKBH5 has evolved to purposefully release FA for possible biological regulatory purposes upon m6A demethylation. Indeed, the cellular imaging data (Fig. 6) show that free FA levels are modulated in an ALKBH5-dependent manner. We hypothesize that the spatiotemporal availability of FA, a potent 1-C electrophile, can influence proximal protein and nucleic acid targets and give rise to a continuum between 1-C metabolism and 1-C signaling.

Materials and Methods

Details of reagents and experimental procedures for complementary DNA sequences, cloning, bacteria and mammalian cells expression and purification of wild-type FTO, ALKBH5, and ALKBH5 mutants, f6A synthesis, concise methods for m6A, hm6A, f6A and A quantification, in vitro enzymatic assays, FA quantification, formate quantification, succinate quantification, native gel cross-linking assay, intact MS assays, thermal shift assays, cell culture, stable cell lines generation, Western blot, siRNA knockdown, immunofluorescence and live cell imaging experiments, and statistical analysis are described in SI Appendix.

Data Availability.

All data are available in the paper or SI Appendix.