Structures of Human ALKBH5 Demethylase Reveal a Unique Binding Mode for Specific Single-stranded N6-Methyladenosine RNA Demethylation*

Background: ALKBH5 catalyzes demethylation of m6A single-stranded RNA (ssRNA). Results: ALKBH5 structures reveal the structural basis of its substrate selectivity and inhibition by citrate. Conclusion: ALKBH5 specifically binds to and demethylates m6A ssDNA/ssRNA. Citrate is a modest inhibitor of ALKBH5. Significance: This study provides insights into the molecular mechanism of ALKBH5 as an m6A ssRNA demethylase and will facilitate the design of selective inhibitors. N6-Methyladenosine (m6A) is the most prevalent internal RNA modification in eukaryotes. ALKBH5 belongs to the AlkB family of dioxygenases and has been shown to specifically demethylate m6A in single-stranded RNA. Here we report crystal structures of ALKBH5 in the presence of either its cofactors or the ALKBH5 inhibitor citrate. Catalytic assays demonstrate that the ALKBH5 catalytic domain can demethylate both single-stranded RNA and single-stranded DNA. We identify the TCA cycle intermediate citrate as a modest inhibitor of ALKHB5 (IC50, ∼488 μm). The structural analysis reveals that a loop region of ALKBH5 is immobilized by a disulfide bond that apparently excludes the binding of dsDNA to ALKBH5. We identify the m6A binding pocket of ALKBH5 and the key residues involved in m6A recognition using mutagenesis and ITC binding experiments.

In higher eukaryotic organisms, more than 100 distinct modifications have been identified in cellular mRNA, tRNA, and rRNA (1,2). Among these modifications, N 6 -methylated adenosine (m 6 A) 3 is the most prevalent internal modification in mRNA; m 6 A has been identified in many eukaryotes and viruses (3,4). There is increasing evidence that m 6 A is associated with mRNA metabolism by affecting, for example, the stability of nascent mRNA, the rate of transcription, and mRNA splicing (5). Based on a recently developed high throughput sequence analysis, m 6 A is not distributed randomly but is enriched near stop codons and in coding sequences (6).
Emerging studies on regulatory roles of mRNA modifications have identified m 6 A "writers" (m 6 A methyltranferases), m 6 A "erasers" (m 6 A demethylases), and m 6 A "readers" (m 6 A specific binding domains). METTL3 was the first identified functional m 6 A RNA methyltransferase and contains both a S-adenosylmethionine binding motif and the key catalytic residues for methylation (7). In 2011 and 2013, He and co-workers (8,9) reported that FTO and ALKBH5 can act as specific mRNA m 6 A demethylases, suggesting that the m 6 A modification is dynamic in vivo. They also demonstrated that METTL14 displays m 6 A methylation activity and forms a stable complex with METTL3 to methylate mammalian nuclear RNA (10). Very recently, the YTH domain containing proteins YTHDF1-3 were shown to specifically recognize m 6 A containing singlestranded RNA (6,11).
Both the FTO and ALKBH5 RNA demethylases belong to the AlkB subfamily of the Fe(II)/2-oxoglutarate (2OG) dioxygenase superfamily. Members from the 2OG dioxygenase superfamily act on diverse substrates involved in the regulation of protein biosynthesis. For example, some members of the Jumonji C (JmjC) subfamily are histone demethylases (12,13), and the TET subfamily of dioxygenases catalyze 5-methylcytosine oxidation (14,15). Although various 2OG oxygenases act on different substrates, they share a common distorted doublestranded ␤-helix (DSBH) fold and conserved, though not identical, 2OG and ferrous ion binding sites, suggesting a common evolutionary origin (16 -19).
The human AlkB family comprises nine members: ALKBH1-8 and FTO; the family is named after their Escherichia coli ortholog, AlkB (20). Although the E. coli AlkB and the human AlkB family members share the stereotypical 2OG oxygenase DSBH fold and have related 2OG and iron binding sites, their substrate selectivities differ. E. coli AlkB exhibits activity toward m 1 A and m 3 C containing DNA/RNA (21,22); ALKBH1 and ALKBH3 strongly prefer m 1 A-or m 3 C-containing ssDNA/RNA (23)(24)(25); ALKBH2 preferentially demethylates m 1 A or m 3 C containing double-stranded DNA (dsDNA) (24,25); ALKBH4 regulates the demethylation of actin (26); ALKBH7 is involved with the programmed cell necrosis, although a nucleic acid demethylase activity for it has yet to be reported (27); ALKBH8 was demonstrated to hydroxylate 5-methoxycarbonylmethyluridine in tRNA (28); and FTO prefers both m 3 T ssDNA and m 6 A ssRNA (9,29). ALKBH5 is reported to catalyze N-demethylation of m 6 A ssRNA or ssDNA, but not other tested modified nucleotides (8,30).
Despite the important and diverse biological functions of the AlkB family, only a few human protein structures in the AlkB family have been characterized structurally, including ALKBH2 in complex with dsDNA (31), apo-ALKBH3 (32), ALKBH8 RRM-AlkB double domain (33), and FTO in complex with 3meT (34) and in complex with various inhibitors (35). As the first identified eukaryotic RNA demethylase, FTO catalyzes demethylation of 3meU in ssRNAs and 3meT in ssDNAs (9,29) and has been implicated in obesity (36,37). ALKBH5 is expressed in different tissues compared with FTO (8) and apparently exhibits a stricter substrate preference by only catalyzing demethylation of m 6 A containing ssRNAs and favoring the sequence of (Pu[GϾA] m 6 AC [A/C/U]) over random sequences (8). Although a preferred substrate of ALKBH5 has been identified, the molecular basis of its selectivity for m 6 A, as well as for its discrimination between ssRNA and dsRNA/ dsDNA, has been unclear.
Here we present two crystal structures of the ALKBH5 catalytic domain: one in a 2OG-bound form and the other in an inhibitor-bound form. Our enzymatic assay results show that the ALKBH5 catalytic domain can demethylate m 6 A containing ssRNA and ssDNA and that citrate is a weak ALKBH5 inhibitor. The crystal structures provide insights into how ALKBH5 specifically acts on single-stranded RNA containing m 6 A. By molecular modeling of the ALKBH5 substrate complex structure, we identify potential m 6 A binding residues, the identities of which are validated by mutagenesis and ITC binding experiments. During revision of this manuscript, two other manuscripts on human ALKBH5 structures were reported (38,39); their structural data and conclusions are consistent with our structural results, which are presented here.

Cloning, Expression, and Purification of Recombinant ALKBH5
Catalytic Domain-The human ALKBH5 catalytic AlkB domain (residues 74 -294) was subcloned into pET28a-MHL vector and expressed in a method similar to previously described (40). The cloned vector was transformed into E. coli BL21-V2R-pRARE2. Recombinant protein was produced at 16°C as an N-terminal His-tagged protein after induction using 0.8 M isopropyl ␤-D-thiogalactopyranoside (final concentration) at an A 600 of 1.0. The recombinant protein was purified by HiTrap nickel column, and the His tag was cleaved overnight by the pTEV protease. The cleaved protein was applied to a second HiTrap Ni column and further purified by Superdex 75 gel filtration (GE Healthcare) and ion exchange chromatography (HiTrap TM ). The desired protein was pooled and concentrated to 25 mg/ml in a buffer containing 20 mM MES, pH 6.5, 150 mM NaCl.
Isothermal Titration Calorimetry-Isothermal titration calorimetry (ITC) measurements were recorded at 25°C using a VP-ITC microcalorimeter (MicroCal Inc.) or MicroCal ITC200 (GE Healthcare). For 2OG and Mn 2ϩ binding to ALKBH5 (74 -294) and the mutants, 25 injections were performed on VP-ITC by injecting 10 l of 2 mM 2OG and Mn 2ϩ into a sample cell containing 40 M of ALKBH5 protein. For citrate binding to ALKBH5 (74 -294), 18 injections were performed on ITC200 by injecting 2 l of 10 mM ammonium citrate into a sample cell containing 220 M of ALKBH5. For ssDNA or ssRNA binding to ALKBH5 (74 -294) and its mutants, 15 injections were performed on ITC200 by injecting 2 l of 600 -700 M nucleic acids into a sample cell containing 40 -55 M of ALKBH5. All ITC binding experiments were performed in a buffer containing 20 mM MES, pH 6.5, 150 mM NaCl. The 2OG-Mn 2ϩ and citrate were dissolved into the same buffer. Proteins and nucleic acids were dialyzed into the same buffer before binding experiments were performed. All modified DNAs or RNAs were purchased from Thermo Fisher Scientific Inc. The concentration of the proteins and DNAs/RNAs were estimated by absorbance spectroscopy using the extinction coefficients, A 280 and A 260 , respectively. Binding isotherms were plotted, analyzed, and fitted in a one-site binding model by Origin Software (MicroCal Inc.) after subtraction of respective cofactors or inhibitor only control.
Protein Crystallization-Purified ALKBH5 stock was diluted to 15 mg/ml, mixed with 5 mM 2OG and 5 mM MnSO 4 , and then crystallized using the hanging drop vapor diffusion method at 18°C. The ALKBH5-2OG-Mn 2ϩ was crystallized in a buffer containing 0.2 M ammonium dihydrogen phosphate, 20% PEG

Data collection and refinement statistics
The data were compiled using PDB_EXTRACT (57), PHENIX (58), and IOTBX (59) software. The highest resolution shell is shown in parentheses.

The Crystal Structures of ALKBH5 RNA Demethylase
3350. The ALKBH5 was crystallized with citrate in a buffer containing ammonium citrate, 20% PEG 3350. Before flashfreezing crystals in liquid nitrogen, crystals were soaked in a cryoprotectant consisting mother liquor plus 12% glycerol. Structure Determination-Diffraction data were collected using an FR-E rotating copper anode with an R-axis-IV-HTC detector (FR-E, Rigaku Corp.) under cooling (Cryostream, Oxford Cryosystems) at 100 K. Diffraction images were reduced to intensities with XDS (41) and merged with AIMLESS (42). Merged intensities were converted to structure factor amplitudes with TRUNCATE (43). Free reflections were selected in thin resolution shells with the program SFTOOLS (B. Hazes). A crystal structure of the citrate complex was solved using the program PHASER (44) with a search model identified by the BALBES server (45) and based on coordinates from PDB entry 3H8R (46). Map improvement with ARP/WARP (47) and PARROT (48) was followed by iterations of automated model building in BUCCANEER (49) and further phase improvement with PARROT. Additional automated model building was performed with ARP/WARP (50). The structure of the cofactor complex was solved by direct placement of preliminary protein coordinates from the citrate complex structure into the nearly isomorphous unit cell of the cofactor complex, followed by rigid body refinement of the protein chains. Both complex atomic models were further refined using iterations of manual rebuilding with COOT (51), restrained refinement with REFMAC (52), and model geometry validation with MOLPROBITY (53). Electron density at the metal binding site of the cofactor complex was interpreted homologously to PDB entry 2IUW. Automatically generated local noncrystallographic symmetry restraints were imposed during refinement of the complex. The two structures of ALKBH5 have been deposited into the Protein Data Bank with the accession codes 4OCT and 4O61, respectively. Molecular representations were prepared with PyMOL molecular graphics system, version 1.7.0.1 (by Schrödinger).
In Silico Modeling of ALKBH5 Ribonucleotide Complex-The modeled structure of ALKBH5 in complex with iron, 2OG, and m 6 A were constructed by using the Hex docking server with the complex structure of ALKBH2 (PDB entry 3H8O) as the reference structure.
Activity Assays for Wild Type and Mutant ALKBH5 74 -294 -A 25-l reaction mixture containing a final concentration of 5 M ALKBH5 74 -294 (wild type/Y139A mutant/E153A mutant/ R130A mutant), 20 M 5-mer ssRNA with the sequence 5Ј-GGm 6 ACU-3Ј (ELLA Biotech, Munich, Germany), 300 M 2OG, 2 mM L-ascorbate, 150 M diammonium Fe(II) sulfate complex, and 25 mM Tris-HCl, pH 7.5, was set up in triplicate. The reaction was incubated at room temperature for 7 min before being quenched by an equivalent volume of 20% formic acid. The quenched reactions were analyzed by MALDI-ToF mass spectrometry as previously described (38). The same assay conditions were used to investigate the enzymatic activity of wild type ALKBH5 74 -294 with a 5-mer ssDNA with the sequence 5Ј-dGdG (m 6 A)dCdT-3Ј (m 6 A is a ribonucleotide) (Thermo Fisher Scientific Inc.).
IC 50 Determination-Inhibition assays were performed in triplicate in 25-l reaction mixtures (final volume), each containing 700 nM ALKBH5 66 -292 (38), 3 M 5-mer ssRNA with the sequence 5Ј-GG(m 6 A)CU-3Ј (ELLA Biotech, Munich, Germany), 3 M 2OG, 1 mM L-ascorbate, 100 M diammonium Fe(II) sulfate complex, different sodium citrate concentrations ranging from 1 M to 1 mM, and 25 mM Tris-HCl, pH 7.5. Controls in triplicate without inhibitors were carried out. All solutions were made up freshly, immediately prior to use. Reaction mixtures were incubated at room temperature and quenched after 6 min with an equivalent volume of 20% (v/v) aqueous formic acid and analyzed using MALDI-TOF mass spectrometry (38).

RESULTS
Overall Crystal Structure of ALKBH5-A crystal structure of the catalytic domain of human ALKBH5 (amino acids 74 -294), with 2OG and manganese, was determined to 2.3 Å resolution (see Table 1 for statistics of data collection and refinement). The asymmetric unit contains two ALKBH5 molecules with a backbone root mean square deviation of 0.1 Å. The dimer interface likely is a crystallization artifact, because the protein behaves as a monomer in solution as shown by gel filtration experiments (data not shown). The core of the ALKBH5 cata-lytic domain contains the double-stranded ␤-helix (DSBH or jelly-roll) fold that is stereotypical of the 2OG oxygenases and consists of a total of 11 ␤ strands (␤1-␤11) and 5 ␣ helices (␣1-␣5) (Fig. 1). In addition to some N-or C-terminal residues, residues 141-147, located in a likely flexible loop between ␤3 and ␤4, are also not apparent in the crystal structure. The ␤ strands fold into two sheets; one four-stranded antiparallel ␤ sheet (␤5-␤10-␤7-␤8) packs against the other seven-stranded antiparallel ␤ sheet (␤2-␤3-␤4-␤11-␤6-␤9-␤1) and forms the core of the fold (Fig. 1, B and C). The helix ␣1 is proximal to ␣5 and flanks one side of the core. ␣3 and ␣4 form the helix-kinkhelix motif and pack against the long helix ␣2; together, these three helices buttress the seven-stranded ␤-sheet (Fig. 1B). On the basis of the search results from the Dali Server, the fold of ALKBH5 is most similar to those of ALKBH2, ALKBH3, and ALKBH8, with z scores of 18.3, 17.3, and 17.8, respectively. For the highest score hit, ALKBH2, the root mean square deviation between ALKBH5 and ALKBH2 is 2.5 Å, although the sequence identity between the catalytic domains of ALKBH5 and ALKBH2 is only 19%. Structure-based sequence alignment of ALKBH1-8, FTO, and E. coli AlkB suggests that the DSBH core fold (corresponding to ␤4 -␤11 of ALKBH5) is well conserved (Fig. 1A).
ALKBH5 Binds to 2OG and Mn 2ϩ -Like with other ALKBH proteins, ALKBH5 requires binding of the cosubstrate 2OG and ferrous iron for catalysis ( Fig. 2A) (8). In our crystallization trial, we replaced the Fe(II) ion with Mn(II) to obtain a catalytically inactive form of ALKBH5 (Fig. 2B) (20). In the 2OG-Mn(II) complex ALKBH5 structure, the Mn(II) ion is complexed by NE2 of His-204, a carboxylate oxygen of Asp-206, NE2 of His-266, and the C1-carboxylate and C2-carbonyl groups of 2OG, respectively (Fig. 2C). Based on the sequence alignment of the AlkB family members, His-204, Asp-206, and His-266 are absolutely conserved across the family (Fig. 1A). Binding of 2OG is also apparently stabilized by electrostatic and hydrogen bonds with the ALKBH5 active site. One oxygen of the C1-carboxylate in 2OG is positioned to interact electrostatically with the side chain of Arg-283 in addition to chelating Mn, whereas the other oxygen of the 2OG C-1 carboxylate is recognized by the side chain of Asn-193 via electrostatic interaction. The 2OG C-5 carboxylate forms a salt bridge with Arg-277 and a hydrogen bond with the side chain of Tyr-195 (Fig.  2C). Both Arg-277 and Arg-283 are well conserved in the AlkB family (Fig. 1A). Hence, ALKBH5, like other AlkB family members, binds to 2OG and the metal ion in a conserved manner.
ALKBH5 Can Catalyze the Demethylation of m 6 A ssRNA, as Well as ssDNA-To investigate whether the crystallized ALKBH5 catalytic domain is active, we performed catalytic assays; we found that in the presence of Fe(II) and 2OG, the ALKBH5 construct demethylates m 6 A in both ssRNA and ssDNA (Fig. 3, A and B). In contrast, ALKBH5 was not observed   3). B, ALKBH5 also catalyzes the ssDNA: dGdG(m 6 A)dCdT. The activity of ALKBH5 toward ssRNA was set to 100% for comparison. The activity of ALKBH5 toward ssDNA was 168 Ϯ 390%. No activity was detected for ALKBH5 when Fe(II) was replaced by Mn(II). C, K d values of ALKBH5 (wild type or variant) binding to modified ssDNA/ssRNA as determined by ITC.
to display any activity toward the same substrates when Fe(II) was replaced by Mn(II), consistent with a previous report (Fig.  3B) (20). ALKBH5 displayed ϳ1.5-fold higher activity with modified ssDNA compared with modified ssRNA under our standard assays conditions (Fig. 3B). ITC binding assays of ALKBH5 to GG(m 6 A)CU RNA and dGdG(m 6 A)dCdT DNA indicated that the m 6 A ssDNA binds to ALKBH5 with similar strength to ssRNA (33 M versus 37 M, Fig. 3C). Taken together, the results reveal that the ALKBH5 catalytic domain (residues 74 -294) is active and can demethylate ssDNA and ssRNA with similar activity; note, however, that m 6 A ssDNA may not be a physiologically relevant ALKBH5 substrate.
Crystal Structure of ALKBH5 in Complex with the Inhibitor Citrate-We also crystallized the ALKBH5 catalytic domain in a crystallization condition containing 0.2 M ammonium citrate and determined the structure at a 1.9 Å resolution. Interest-ingly, a citrate molecule, instead of 2OG and Mn 2ϩ , was observed in the active site of ALKBH5, apparently competing out both cofactors under the crystallization conditions. The C5-carboxylate of the citrate molecule is positioned to hydrogen bond with the side chains of Lys-132 and Asn-193; the C-1 and C-6 carboxylates of the citrate are positioned to hydrogen bond with Ser-217 and Arg-283, respectively; and the citrate C-3 carbonyl oxygen is positioned to hydrogen bond with N ⑀2 of His-266 (Fig. 4A). Most of the residues involved in the citrate binding are involved in binding 2OG and Mn(II) as observed in the ALKBH5-2OG/Mn(II) structure (Fig. 2C).
Recently, Aik et al. (35) reported that citrate can act as a modest inhibitor of the human ALKBH enzyme FTO. In the FTO-citrate complex structure, the citrate molecule binds to FTO in a different manner, i.e. the citrate molecule replaces 2OG, but active site metal ion is still present (Fig. 4B). We per- formed the ITC experiments to measure the binding affinity of ALKBH5 to citrate (K d , 50 M), which was found to be ϳ5-fold weaker than that of the 2OG/Mn 2 binding to ALKBH5 (9 M) (Fig. 4C). The IC 50 of citrate for ALKBH5 was measured at 488 M, which is comparable to that for human FTO (300 M) under standard assay conditions (Fig. 4D) (35). Therefore, although citrate is observed to adopt different binding modes in the FTO and ALKBH5 crystal structures, it can act as an inhibitor for both human m 6 A demethylases.
Structural Elements of ALKBH5 Determining Its Specificity toward Single-stranded m 6 A-It has been reported that ALKBH5 specifically demethylates m 6 A single-stranded RNA. In contrast, ALKBH2 is known to just modify double-stranded DNA. To investigate the substrate specificity of ALKBH5, we superimposed the catalytic domains of ALKBH5 and ALKBH2. Despite the structural similarities between the two 2OG oxygenases, significant differences exist in the corresponding nucleic acid binding regions of ALKBH2 and ALKBH5 (Fig. 5). First, the positively charged motif of ALKBH2, Arg-241 to Lys-243, is not conserved in ALKBH5 (Fig. 1A). The RKK motif of ALKBH2, located in a loop corresponding to the loop linking ␤10 and ␤11 (Fig. 1A), is critical in recognizing the backbone of the complementary DNA strand in the ALKBH2-dsDNA complex (Fig. 5B). Second, in ALKBH2 two short ␤ strands of (␤3-␤4) contact the major groove of dsDNA, and Phe-102 in the loop between ␤3 and ␤4 points to the major groove and stacks with the A-T base pair; These structural features are missing in the corresponding region of ALKBH5 (Fig. 5C). Last but not least, a long loop connecting ␤7 and ␤8 in ALKBH2, which contacts the backbones of dsDNA, deviates from the corresponding loop (amino acids 229 -243) of ALKBH5. The loop of ALKBH5 (amino acids 229 -243) would cause a steric clash with the complementary strand of dsDNA. Furthermore, the conformation of the analogous loop in ALKBH5 is apparently "anchored" to ␤10 because of the presence of a disulfide bond between Cys-230 (loop) and Cys-267 (␤10) (Fig. 5D). These two cysteine residues are well conserved across all ALKBH5 orthologs, but not in other members of the AlkB family (Figs. 1A and 6). Collectively, these structural features differentiate ALKBH2 and ALKBH5 and rationalize the preference of ALKBH5 for single-stranded over double-stranded nucleic acids.
Structural Modeling Indicates an m 6 A-specific Binding Pocket-Three enzyme-substrate complex structures of AlkB family members have been solved, i.e. for the E. coli AlkB protein in complex with m 1 A ssDNA (20,31), ALKBH2 in complex with m 1 A dsDNA (31), and FTO in complex with m 3 T (34). In all three structures, the substrates occupy a conserved binding pocket close to the cofactors and share some common features. The methyl groups of the modified bases are positioned close to the active site metal and 2OG C-1 carboxylate binding sites (and thus presumably the dioxygen binding site). The modified nucleobases are sandwiched between the rings of the first of the iron-binding histidines (corresponding to His-204 of ALKBH5) and one of the active site assigned "aromatic cage" residues (Trp, Phe, or Tyr, corresponding to Tyr-141 of ALBKH5). In addition, the modified bases are also specifically recognized by corresponding residues via hydrogen bonds; for example, N 6 of m 1 A forms hydrogen bond with Asp-135 of AlkB, N 6 of m 1 A forms two hydrogen bonds with Tyr-122 and Glu-175 of AlkB, and O 2 of m 3 T forms two hydrogen bonds with Arg-96 of FTO (Fig. 7, A-C).
On the basis of these complex structures, we modeled an ALKBH5 substrate complex structure with the metal, 2OG, and m 6 A (prime substrate binding succeeds that of 2OG in the consensus mechanism for 2OG catalysis, with oxygen binding last). The model predicts that ALKBH5 can accommodate m 6 A in a manner similar to that in which other AlkB family proteins are observed to bind modified nucleic acids. Namely, m 6 A is positioned in a positively charged pocket formed by Arg-130, Tyr-139, His-204, and 2OG (Fig. 7, D and E). Specifically, the methyl group of m 6 A is positioned close to the C-1 carboxylate of 2OG, and the side chain of Tyr-139. In contrast to the three ALKBH/ FTO substrate complexes mentioned above, m 6 A in the modeled ALKBH5 structure is not predicted to be sandwiched between two aromatic residues. Instead, Arg-130 may stack with the m 6 A base via cation-interactions or interact with the RNA backbone via electrostatic interactions. Furthermore, m 6 A packs against the ring of His-204 (Fig. 7E). Of note, Arg-130 and Tyr-139 are conserved in ALKBH5 orthologs in other organisms (Fig. 6). In both ALKBH5 structures, we observed that the backbone of another active site aromatic residue, Tyr-141, is invisible in current substrate-free structures. However, we could not exclude the possibility that Tyr-141 may become ordered when binding to m6A, and it is possible that this aromatic residue may also stack with m 6 A.
To evaluate the roles of the conserved residues in the putative m 6 A binding pocket, we made three variants, R130A, Y139A, and E153A. Glu-153 is not predicted to be directly involved in m 6 A binding and was used as a control. The enzyme assay results showed that neither R130A nor Y139A displays detectable activity toward m 6 A RNA. In contrast, E153A display activity similar to wild type ALKBH5 (Fig. 3A). In support of the enzyme assay results, ITC binding results reveal that neither R130A nor Y139A binds to the 5-mer m 6 A RNA within the limits of detections (Fig. 3C). On the other hand, we found that both these variants bind to 2OG and Mn(II), albeit a little more weakly than does the wild type ALKBH5 (Ͻ2-fold; Fig.  2D). Hence, both enzymatic assay and the ITC binding data broadly support the modeled structure and proposed m 6 A pocket.

DISCUSSION
Two recently identified m 6 A RNA demethylases, FTO and ALKBH5, dynamically regulate the profile of m 6 A in vivo (8,34). Our results define a crystal structure of human ALKBH5 in the presence of 2OG and Mn 2ϩ . ALKBH5 adopts an extended DSBH fold similar to other AlkB family members; cofactor/ cosubstrate binding residues are conserved not only in its orthologs, but also in all other AlkB family members (Figs. 1 and   6). Additionally, we were able to crystallize ALKBH5 in complex with a modest inhibitor citrate. Surprisingly, the citrate binds to ALKBH5 in a different manner from that observed in FTO. In the FTO-citrate complex, the citrate replaces 2OG, FIGURE 7. Modeled structure of ALKBH5 complexed with an m 6 A-containing substrate compared with analogous complexes for other AlkB family members. Substrates and the residues involved in substrate binding are shown in stick mode. Hydrogen bonds are shown as black dashes. Bonds with the metal ion are shown as red dashes. A, structure of E. coli AlkB (cyan) and m 1 A containing ssDNA (yellow) derived from PDB entry 2FD8. B, structure of ALKBH2 (green) and m 1 A containing dsDNA (yellow) derived from PDB entry 3BUC. C, structure of FTO (red) and m 3 T (yellow) derived from PDB entry 3FLM. D, electrostatic surface of ALKBH5 binding to m 6 A (yellow). E, modeled structure of ALKBH5 (gray) and m 6 A (yellow). Residues involved in recognizing m 6 A and cofactor/cosubstrate are shown as blue and gray sticks, respectively. and the metal ion is still intact. In the ALKBH5-citrate complex, the citrate excludes both the metal ion and the 2OG cosubstrate (Fig. 4, A and B). The different modes of inhibition caused by a TCA cycle intermediate, such as citrate, may be pathophysiologically relevant because inhibition of some 2OG oxygenases as caused by up-regulation of TCA cycle is proposed to be prooncogenic (54,55).
Although citrate is only a modest inhibitor of the ALKBH family dioxygenases, the different binding modes of citrate to FTO and ALKBH5 may be taken advantage of in the design of new types of target-specific inhibitor or chemical probes for the Fe(II)/2OG oxygenases. Although inhibition via competition with Fe(II) can be achieved of a noncatalytically active metal, e.g. inhibition of the hypoxic inducible factor 2OG-dependent prolyl hydroxylases can be achieved in isolated protein form and in vivo by Co(II) ions (including to stimulate erythropoiesis in animals) (56), such metal competition is not specific. Thus, organic compounds that compete with Fe(II) binding at 2OG oxygenase active sites could constitute a valuable new class of 2OG oxygenase inhibitors.
Although all identified AlkB family members have similar folds, they have different substrates and functions. E. coli AlkB, ALKBH1, and ALKBH3 prefer single-stranded nucleic acids, and ALKBH2 acts on double-stranded DNAs (21)(22)(23)(24)(25). ALKBH5 exhibits a strict preference for m 6 A ssRNA (8,30), whereas FTO can demethylate m 6 A ssRNA, as well as m 3 T ssDNA (9,29). To understand why ALKBH5 acts on singlestranded, but not double-stranded nucleic acids, we compared the structure of ALKBH5 with that of ALKBH2-dsDNA and found that a rigid loop in ALKBH5 will cause a steric clash with dsDNA. FTO also prefers m 3 T or m 6 A in single-stranded nucleic acids and also utilizes a loop to discriminate against binding of dsDNA.
Although ALKBH5 and FTO both contain specific loops that apparently discriminate single-stranded from double-stranded nucleic acids, the underlying mechanisms differ. FTO contains a longer loop between its ␤5 and ␤6 strands, which corresponds to the loop between ␤4 and ␤5 of ALKBH5. The longer loop of FTO could cause a steric clash with the complementary strand of the nucleic acids ( Fig. 1A) (34). ALKBH5 likely utilizes a different structural feature to exclude the binding of dsDNA. The loop between ␤7 and ␤8 of ALKBH5 adopts a different conformation from that observed in ALKBH2, because it is linked to the ␤10 strand via a disulfide bond formed between Cys-230 and Cys-267 (Fig. 5D). Hence the ␤7-␤8 linking loop is rigidified and is proposed not to tolerate the conformational changes required to accommodate dsDNA. Furthermore, two regions of ALKBH2 that contact the backbone of the complementary strand of the dsDNA and a base pair of the dsDNA, respectively, are missing in ALKBH5 (Fig. 5, A and B). Collectively, these structural differences rationalize the preference of ALKBH5 for single-stranded over double-stranded nucleic acids.
To identify key residues of ALKBH5 involved in m 6 A recognition (30), we modeled the m 6 A into the ALKBH5-2OG-Mn(II) complex structure based on a few ALKBH-substrate complex structures. Analogous to the structures of E. coli AlkB, ALKBH2, and FTO complexes, m 6 A is also predicted to occupy a pocket close to 2OG and Mn(II) in the modeled structure. In ALKBH5, the m 6 A base is predicted to pack against His-204 and is in a pocket composed of Arg-130 and Tyr-139. The combined substitution and enzymatic assay results support the essential roles of Arg-130 and Tyr-139 in substrate recognition. Although the two variants can still bind to 2OG and meta, neither of them binds to the m 6 A RNA substrate, nor do they display activity toward the m 6 A RNA substrate. Although further investigations on the structure of ALKBH5 complexes are required, our biochemical and structural data not only provide insights into understanding the molecular mechanism of ssRNA demethylation by ALKBH5 but also identify citrate as a modest inhibitor for ALKBH5.