The Atomic Resolution Structure of Human AlkB Homolog 7 (ALKBH7), a Key Protein for Programmed Necrosis and Fat Metabolism*

Background: Human ALKBH7 is a mitochondrial protein required for alkylation- and oxidation-induced programmed necrosis and short chain fatty acid metabolism. Results: Atomic resolution structures of ALKBH7 in complex with two ligands were determined. Conclusion: Structural analyses and observation of self-hydroxylation demonstrate that ALKBH7 may function as a protein hydroxylase. Significance: ALKBH7 structures provide a foundation for drug design in cell death and metabolic diseases. ALKBH7 is the mitochondrial AlkB family member that is required for alkylation- and oxidation-induced programmed necrosis. In contrast to the protective role of other AlkB family members after suffering alkylation-induced DNA damage, ALKBH7 triggers the collapse of mitochondrial membrane potential and promotes cell death. Moreover, genetic ablation of mouse Alkbh7 dramatically increases body weight and fat mass. Here, we present crystal structures of human ALKBH7 in complex with Mn(II) and α-ketoglutarate at 1.35 Å or N-oxalylglycine at 2.0 Å resolution. ALKBH7 possesses the conserved double-stranded β-helix fold that coordinates a catalytically active iron by a conserved HX(D/E) … Xn … H motif. Self-hydroxylation of Leu-110 was observed, indicating that ALKBH7 has the potential to catalyze hydroxylation of its substrate. Unlike other AlkB family members whose substrates are DNA or RNA, ALKBH7 is devoid of the “nucleotide recognition lid” which is essential for binding nucleobases, and thus exhibits a solvent-exposed active site; two loops between β-strands β6 and β7 and between β9 and β10 create a special outer wall of the minor β-sheet of the double-stranded β-helix and form a negatively charged groove. These distinct features suggest that ALKBH7 may act on protein substrate rather than nucleic acids. Taken together, our findings provide a structural basis for understanding the distinct function of ALKBH7 in the AlkB family and offer a foundation for drug design in treating cell death-related diseases and metabolic diseases.

Cell death, one of the most vital cellular responses of multicellular organisms, can be divided into apoptosis, autophagy, and necrosis based on different morphological appearance (1). Although apoptosis is well studied and regarded as programmed cell death, necrosis has been considered as an uncontrolled process that occurs as a result of infection or injury for a long time. However, the accumulating evidence reveals that necrosis is also well regulated as apoptosis in many cases. For example, the tumor necrosis factor (TNF)-induced programmed necrosis has unique signaling mechanisms mediated by receptor-interacting protein kinase 1 (RIP1) and RIP3. Programmed necrosis plays an important role in various pathological processes, including ischemic brain injury, neurodegenerative diseases, and viral infections (2). Necrosis-mediated via the poly(ADP-ribose) polymerase pathway is the other extensively studied model of programmed necrosis. In response to DNA damage caused by DNA-alkylating agents such as 1-methyl-3-nitro-1-nitrosoguanidine, the DNA repair enzyme poly(ADP-ribose) polymerase 1 is hyperactivated, resulting in massive depletion of nicotinamide adenine dinucleotide (NAD ϩ ), dysfunction of mitochondria, abolishment of ATP production, and execution of programmed necrosis (3)(4)(5). Because many kinds of cancer cells are resistant to apoptosis, programmed necrosis emerges as an attractive target to activate the death of these cancer cells (6).
Recently, the mitochondrial protein human AlkB homolog 7 (ALKBH7) was found to be essential for alkylation-and oxidation-induced programmed necrosis, but it had no effect on apoptosis (7). ALKBH7-depleted cells were much more likely to survive than wild-type cells when suffering DNA damage caused by oxidizing or alkylating agents. Both types of cells underwent poly(ADP-ribose) polymerase 1 hyperactivation, NAD ϩ , and ATP depletion after severe DNA damage, only the ALKBH7-depleted cells exhibited rapid recovery of intracellular NAD ϩ and ATP levels. In addition, ALKBH7-depleted cells maintained their mitochondrial membrane potential and homeostasis, indicating a vital role of ALKBH7 in triggering the collapse of mitochondrial membrane potential and dysfunction of mitochondria in alkylation-and oxidation-induced programmed necrosis. Besides, another study showed that the knock-out Alkbh7 in mice led to increased body weight and body fat. Alkbh7 was found to facilitate the utilization of shortchain fatty acids (8). All of this research demonstrated the key roles of ALKBH7 in mitochondrial function.
ALKBH7 does not exhibit any demethylase activity for both m 1 A and m 3 C lesions in single-stranded DNA/dsDNA (7,24). Its substrate is still unknown after 10 years of assiduous research. In contrast to ALKBH7, inactivation of other AlkB family proteins leads to sensitivity rather than resistance to DNA damage (25)(26)(27)(28). Furthermore, deletion of the fto gene in mice protects from obesity, whereas the Alkbh7 knock-out mouse develops obesity (8,29). Taken together, ALKBH7 may play a unique role distinct from the known ALKBH functions such as DNA repair or RNA modification. However, the structure and activity of ALKBH7 remain unclear.
Here, we present the atomic resolution structure of ALKBH7 in complex with Mn(II) and ␣-KG or N-oxalylglycine (NOG). Structure analyses reveal that ALKBH7 possesses the conserved active site motifs and residues of Fe(II)/␣-KG-dependent dioxygenase. We demonstrate that ALKBH7 can self-hydroxylate Leu-110. Furthermore, our data reveal the differences between substrate-binding sites of ALKBH7 and other AlkB family members. Our studies not only provide an explanation why ALKBH7 has no repair activity for alkylation-damaged oligonucleotides but also indicate a potential protein hydroxylation activity of ALKBH7.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-DNA fragments encoding various fragments of wild-type ALKBH7 and their mutants were amplified by PCR and ligated into pET-28a (Novagen) with a tobacco etch virus protease cleavage site. The final clones were verified by DNA sequencing. All proteins were expressed in E. coli BL21 (DE3) cells (Novagen). Cells were grown in LB at 37°C until A 600 reached 0.6 -0.8 and then were induced for 12 h at 18°C with 0.4 mM isopropyl ␤-D-1-thiogalactopyranoside. For the selenomethionine (SeMet)-labeling protein, a single colony grown at 37°C in LB was diluted 1:1000 into M9 minimal medium containing 0.2% (w/v) glucose, 50 mg/liter of each amino acid other than methionine. Cells were grown at 37°C until the A 600 reached 0.3, and then 50 mg/liter SeMet (Sigma) was added. After 0.5 h of additional growth at 37°C, cells were induced with 0.4 mM isopropyl ␤-D-1-thiogalactopyranoside for 16 h at 18°C. The cultures were harvested by centrifuging at 4000 ϫ g for 10 min at 4°C and then resuspended in lysis buffer (20 mM Tris, pH 8.0, 1 M NaCl, 1 mM ␤-mercaptoethanol) supplemented with Triton X-100 and PMSF (Invitrogen). After sonication and centrifugation at 20,000 ϫ g for 30 min at 4°C, the supernatant was applied to His affinity beads (Bio-Rad). The column was washed with lysis buffer containing 10 mM imidazole, and the protein was eluted with lysis buffer containing 200 mM imidazole. Tobacco etch virus protease was added to the protein-containing eluent to remove the His tag at a 1:10 weight ratio overnight. The digestion production was reloaded onto His affinity beads to remove the His tag and Histagged tobacco etch virus protease. Further purification was performed with size exclusion chromatography (Superdex TM -200, GE Healthcare) using gel filtration buffer (20 mM Tris, pH 8.0, 150 mM NaCl, 2 mM dithiothreitol for native protein or 5 mM tris(2-carboxyethyl)phosphine for SeMet-labeled protein). Fractions were analyzed by SDS-PAGE, and the target protein was pooled and concentrated to 15 mg/ml for crystallization.
Crystallization and Data Collection-Mn(II) was used to replace Fe(II). To obtain ALKBH7 complex crystals, 1 mM protein was mixed with 10 mM MnCl 2 , 10 mM ␣-KG, or ␣-KG analogs such as NOG or 2,4-pyridine dicarboxylate and then incubated on ice overnight. Initial crystallization trials were performed at 4 and 16°C using the sitting drop vapor diffusion method. A series of truncations was tested for crystallization. The apo-ALKBH7 protein failed to crystallize, and the ALKBH7(17-215)-protein mixed with MnCl 2 and ␣-KG/NOG formed rod-shaped crystals at 4°C in the reservoir solution of 10% PEG3K, 200 mM MgCl 2 , 100 mM sodium cacodylate, pH 6.2. Although diffracted to 2.5 Å, these rod-shaped crystals contained a kind of crystal disorder called lattice-translocation defect (30) as revealed by the sharp and diffuse diffraction pattern (Fig. 1A). The smeary diffraction spots led to problems in indexing and integrating reflection intensities. The diffraction qualities of these crystals were not improved after extensive optimization of crystallization conditions or various postcrystallization treatments such as annealing and dehydration. Mutants of the ALKBH7 protein according to many strategies were applied. Finally, the Q90R mutant engineered by ALKBH7 sequence alignment between human and different species remarkably improved the diffraction quality ( Figs. 1B and 2). The high quality crystals were obtained by hanging drop vapor diffusion against 150 mM MnCl 2 , 50 mM MgCl 2 , 100 mM sodium cacodylate, pH 6.5, 10% glycerol, and 8% PEG3350 for 7 days. SeMet-labeled crystals were obtained under the same condition. For data collection, crystals were gradually transferred to a cryo-buffer (reservoir buffer supplemented with 20% ethylene glycol) and flash-frozen in liquid N 2 .
The anomalous data of the selenium atom were collected on beamline BL-5A at the Photon Factory (High Energy Accelerator Research Organization). The other data were collected on beamline BL17U1 at the Shanghai Synchrotron Radiation Facility and beamline 3W1A at the Beijing Synchrotron Radiation Facility. All the data were integrated and scaled with the HKL2000 suite of programs (31). Data collection and processing statistics are shown in Table 1.
Structure Determination and Refinement-Native crystals of ALKBH7(17-215)-Q90R diffracted to 1.35 Å resolution ( Table  1). Because of the low sequence homology with other structuresolved proteins (the highest sequence identity is 16%), molecular replacement was unsuccessful getting the proper phase of ALKBH7. Multiwavelength anomalous diffraction data collected from SeMet-labeled crystals were used to solve the structure.
The diffraction data of SeMet crystals were collected at the selenium peak and edge (0.97929 and 0.97943 Å) with an oscillation of 0.5°per image. However, pseudo-merohedral twinning was detected, which increased the difficulty in locating the selenium sites and structural determination. Because of low anomalous diffraction signal and twinning, no clear selenium site pattern was found using SOLVE (32). Because of only one methionine in the predicted secondary structure of this fragment, two extra methionine sites added by the R86M/L89M double mutation were found to facilitate the phasing. Initial phases were calculated by SOLVE, and phases were subsequently improved using RESOLVE (32). The figure of merit was 0.358, and the Z score was 17.1 for the solution of SOLVE. A few ␤-strand bundles could be automatically modeled into the electron density map by RESOLVE. Then a polypeptide model was built using COOT (33) and refined using REFAMC5 (34) in iterative cycles. Finally, the best solution was refined to an R work /R free values of 0.350/0.428, and the figure of merit was 0.455. Using this model, the high resolution structure was determined and built using COOT (33). After numerous rounds of model building and refinement, the structure was refined to an R work of 14.7% and an R free of 16.3%.
The complex structure of ALKBH7-Mn(II)-NOG was determined by molecular replacement using the ALKBH7-Mn(II)-␣-KG structure as a model. The final structure was refined to an R work of 15.9% and an R free of 16.5%. All figures in this article displaying molecular structures were made using PyMOL (35).
LC-MS Analysis-Purified proteins (100 M in 20 mM Tris, pH 8.0, 150 mM NaCl), 500 M (NH 4 ) 2 Fe(SO 4 ) 2 , and 500 M ␣-KG were incubated aerobically for 24 h and loaded onto SDS-PAGE. Samples were excised from Coomassie Blue-stained gels and ground to small pieces. The gel pieces were soaked in 200 l of MilliQ-H 2 O for 30 min at 37°C and extensively washed with 100 l of 100 mM NH 4 HCO 3 three times. Then gel pieces were shrunk in 100 l of 100 mM NH 4 HCO 3 for 15 min. After removal of residual solvent and drying, the pieces were rehydrated in 10 l of trypsin (10 ng/l) (sequence grade, Roche Applied Science) on ice for 45 min, and then 10 l of 25 mM NH 4 HCO 3 , 5% CH 3 CN was added. Digestion was carried out at 37°C overnight. The digested peptide was separated by a nano-ACQUITY UPLC BEH C 18 1.7 m column (Waters), using a CH 3 CN/H 2 O (0.1% formic acid) gradient. MS utilized a Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Scientific) that was set to detect positive ions over the 300 -1800 m/z range.
Microscale Thermophoresis-The binding affinity of ␣-KG toward ALKBH7 was measured using microscale thermophoresis with Monolith NT.115 (NanoTemper Technologies GmbH, München, Germany). ALKBH7 was labeled with a fluorescent dye NT-647 (cysteine-reactive) according to the manufacturer's manual. For the assay, 10 l of 10 M labeled protein in a buffer containing 20 mM Tris, pH 8.0, 150 mM NaCl, 0.005% Tween 20 was mixed with 10 l of ␣-KG at various concentrations. The final concentrations were from 0.75 to 6250 M for ␣-KG. After incubation for 30 min at room temperature, 10 l of the mixture were loaded into standard treated capillaries. Thermophoresis was measured at 25°C for 30 s with 20% LED power and 40% infrared laser power. Data of three independent measurements were combined and analyzed using the NTAnalysis software (NanoTemper Technologies GmbH).

RESULTS
Structure Determination of ALKBH7-Predicted as a mitochondrial targeting sequence (7), the N-terminal 17 residues of ALKBH7 were deleted to promote prokaryotic expression. After extensive screening, truncation of the C-terminal 6 residues was found to facilitate protein crystallization. The apoprotein failed to crystallize, whereas the ALKBH7⅐Mn(II)⅐␣-KG/ NOG complex did, indicating a more ordered conformation after binding ligands. However, the wild-type protein crystal produced a poor diffraction pattern (Fig. 1A). After further screening, a Q90R mutant selected according to the protein sequence of Bos taurus ( Fig. 2) was found to greatly improve the diffraction quality (Fig. 1B). However, the structure was not solved because no similar structures were available, and the anomalous diffraction signal of SeMet-labeled protein crystals was low. The R86M/L89M double mutation introducing two additional methionine residues was found to facilitate structural determination. The structure of ALKBH7 in complex with Mn(II) and ␣-KG was solved using the multiwavelength anomalous diffraction signal of SeMet-labeled protein crystals and was refined to 1.35 Å resolution for native protein crystals. There are three protein molecules in the asymmetric unit. We rebuilt Mn(II), ␣-KG, and 189 residues for each ALKBH7 molecule, although no electron density was observed for residues 100 and 207-215. The structure of ALKBH7 in complex with Mn(II) and NOG was solved using molecular replacement and refined to 2.0 Å resolution. Interestingly, only two NOG molecules were found in the trimeric structure. An overlay of ALKBH7-Mn(II)-␣-KG and ALKBH7-Mn(II)-NOG structures revealed little difference in the overall conformation with a root mean square deviation (r.m.s.d.) of 0.4 Å. In addition, the R86M/L89M double mutation (Table 1) also did not alter the overall structure with an r.m.s.d. of 0.3 Å.
Although wild type and Q90R mutant were monomeric in solution, they crystallized as a homotrimer in the space group P2 1 (Fig. 4A). The Q90R mutation introduces a stabilization network between Arg-90 of one ALKBH7 molecule and Glu-44 and Arg-91 of the adjacent ALKBH7 molecule (Fig. 4B), which might facilitate the crystal packing. This mutation locates on the helix ␣3 and is far away from the active site. Except in the active center, Mn(II) ions are also found at the molecule-molecule interface. Interaction between residues His-65 of one ALKBH7 molecule and Glu-62 and Glu-75 of the neighboring ALKBH7 molecule via a Mn(II) ion promotes the formation of the crystallographic trimer (Fig. 4C). Another significant lattice contact involves a Mn(II) ion surrounded by three Glu-189 residues from three ALKBH7 molecules (Fig. 4D).
ALKBH7 Has a Unique Active Center-The HX(D/E) . . . X n . . . H facial triad motif of ALKBH7 coordinates the metal ion at the active site, which is conserved in the Fe(II)/␣-KG-dependent dioxygenase superfamily (36). The differential electron density map clearly shows that the Mn(II) is coordinated in an octahedral manner by two oxygen atoms of ␣-KG, one water molecule and the side chains of His-121, Asp-123, and His-177 (Fig. 5A). The observed distances for residues coordinating with the metal ion correspond to normal values as other ALKBH family members. The cofactor ␣-KG chelates Mn(II) in a bidentate manner via its c-1 carboxylate trans to His-177 and c-2 keto group trans to Asp-123. ␣-KG is further stabilized by electrostatic interaction with Arg-197 and Arg-203 at the both ends of ␤10 (␤VIII), which is conserved in AlkB family (Figs. 2 and 5A).
In most of the AlkB family proteins, the ␣-KG c-5 carboxylate hydrogen bonds to the side chain of tyrosine from DSBH ␤I (Fig. 5, D-H). However, the ␣-KG-binding tyrosine comes from DSBH ␤VI in ALKBH7 and FTO (Fig. 5, A and I). Moreover, this ␣-KG c-5 carboxylate-binding tyrosine (Tyr-165) is highly conserved among ALKBH7 from different species (Fig. 2). The ␣-KG c-5 carboxylate also forms another hydrogen bond with the side chain of asparagine from ␤VIII in AlkB, ALKBH2, or ALKBH3 (Fig. 5, D-F) and with that of Ser-199 from ␤VIII (␤10) in ALKBH7 (Fig. 5A). Furthermore, the ␣-KG c-1 carboxylate oxygen usually forms a hydrogen bond with asparagine

Structure and Function of ALKBH7
OCTOBER 3, 2014 • VOLUME 289 • NUMBER 40 from ␤I in AlkB family. But no such hydrogen bond was observed in ALKBH7, a unique feature of it (Fig. 5, A and D-I).
A highly conserved Leu-110 among ALKBH7 from different species lies in the equivalent position of the above asparagine (Fig. 2).
The absence of a hydrogen bond between ␣-KG c-1 carboxylate and ALKBH7 would lead to weaker binding affinity than those previously reported ␣-KG-dependent oxygenases. Consistently, the dissociation constant (K d ) was determined to be 80.9 Ϯ 8.12 M using microscale thermophoresis (Fig. 5C). The weak binding of cofactor is further proven by the rotation of the c-1 carboxylate of NOG in the crystal structure (Fig. 5B).
The distance between c-1 carboxylate and metal ion increases from 2.0 to 2.8 Å in the ALKBH7⅐Mn(II)⅐NOG complex structure compared with that in the ALKBH7⅐Mn(II)⅐␣-KG complex structure.
Besides the ␣-KG binding residues, other residues in the ␣-KG binding cavity of ALKBH7 are mainly hydrophobic. These residues are derived from ␤-strands ␤4 Another interesting discovery is the post-translational modification of Leu-110 (Fig. 5A). A similar but not the same post-

Structure and Function of ALKBH7
OCTOBER 3, 2014 • VOLUME 289 • NUMBER 40 translational modification was reported in ALKBH3 (19). Differences will be discussed in detail below.

Comparison with Other AlkB Family Members Reveals Several Unique Features of ALKBH7-Despite the vital role in alkylation-
and oxidation-induced programmed necrosis and metabolism of short-chain fatty acids, the main substrate of ALKBH7 remains unknown. Comparison of ALKBH7 and other AlkB family members may shed light on the open question. Although ALKBH1 was reported as a histone H2A demethylase (40) and ALKBH4 was shown to mediate demethylation of a monomethylated site (K84me1) in actin in vivo (41), all structuresolved AlkB family members are nucleic acid oxygenases whose substrates are either DNA or RNA. Here, structural comparison of ALKBH7 and other AlkB homologs demonstrates several unique features that make ALKBH7 significantly different from the nucleic acid oxygenases.
Superimposed to the solved AlkB family structures, the most significant structural difference of ALKBH7 rests on the motif referring to as the "nucleotide recognition lid (NRL)." All structurally characterized nucleic acid oxygenases possess a conserved NRL that constitutes the nucleobase-binding cavity (Fig.  6A). NRL is further divided into two parts named "Flip1" and "Flip2." Although Flip1 determines double-strand versus single-strand substrate recognition of ALKBH2 and ALKBH3 (18), Flip2 provides residues interacting with the modified nucleobase. In AlkB, ALKBH2, FTO, or ALKBH5, Trp-69, Phe-124, Tyr-108, or Tyr-141 (Fig. 5, D, E, I, and G) in Flip2 binds the modified nucleobase through aromatic stacking interaction (20,38,(42)(43)(44). Y108A mutant in FTO and Y141A mutant in ALKBH5 abolished their activity (43,44). Furthermore, Tyr-76 of AlkB and Tyr-122 of ALKBH2 in Flip2 interact with the N6 atom of the m 1 A substrate (Fig. 5, D and E) (20). Tyr-143 of ALKBH3, Tyr-139 of ALKBH5, and Tyr-106 of FTO are in the equivalent position (Fig. 5, F, G, and I). However, structural comparison and sequence alignment of AlkB family members clearly reveal the absence of Flip2 in ALKBH7 (Figs. 2 and 6A). The lack of Flip2 in ALKBH7 leads to a solvent-exposed active site and a loss of nucleobase binding ability.
Moreover, the conformation of ␣2-␤3 loop in ALKBH7 is different from that of Flip1 in other AlkB family members and the residues in Flip1, which are important in binding nucleobase or phosphate backbone, are also absent in the ␣2-␤3 loop (Fig. 6B). Arg-110 was found to interact with the substrate phosphate backbone in ALKBH2⅐dsDNA complex (20). Arg-131 of ALKBH3 in the equivalent site is positioned to make contact with the N3 atom of m 1 A, and R131A mutant was shown to be inactive (19). However, Gly-72 is located in the equivalent position of ALKBH7 (Figs. 2 and 6B). The absence of a nucleic acid-binding residue may greatly affect the binding to nucleic acid. Furthermore, structural superimposition of ALKBH7 and the AlkB⅐dsDNA or ALKBH2⅐dsDNA complex demonstrates that the ␣2-␤3 loop of ALKBH7 would crash with the modified DNA strand (Fig. 6C). The ␣2-␤3 loop does not seem to have enough flexibility to undergo conformational change due to the rather low temperature factor in the solved structures.
Previous studies have demonstrated that a positively charged groove and the ␤IV-␤V loop are needed to bind nucleic acids (19,20,38,42,44,45). The minor ␤-sheet plus ␤IV-␤V loop of DSBH create the nucleic acid binding groove in nucleic acid oxygenases (Fig. 6, D and E). Arg-161 located at the apex of the ␤IV-␤V loop of AlkB is important in the recognition of the damaged nucleobases. ALKBH2 has a long ␤IV-␤V loop carrying nucleic acid-binding residues Arg-198 and Lys-205 and a short ␤VII-␤VIII loop with a positively charged RKK sequence (Arg-241-Lys-243). The ␤IV-␤V loop is not visible in the crystal structure of ALKBH3, but three consecutive prolines in this loop form a proline-rich pinch that has been shown to play a role in DNA damage scanning. In ALKBH5, the ␤IV-␤V loop includes basic residues (Lys-231, Lys-235, and Arg-238), which form a positively charged groove. However, the counterpart (␤6-␤7 loop) of ALKBH7 only contains 6 residues (Figs. 2 and 3), and this short loop is negatively charged with two acidic residues (Glu-150 and Glu-153). In addition to this, the electrostatic potential map of ALKBH7 demonstrates a continuous negatively charged surface compared with the positively charged groove of nucleic acid oxygenases (Fig. 6E).
Structural comparison of the DSBH domain of AlkB family members reveals an additional long loop between ␤VII and ␤VIII (␤9 and ␤10) in ALKBH7, which forms the outer wall of the minor ␤-sheet (Fig. 6D). Several conserved acidic residues (Asp-182, Glu-183, Glu-184, and Glu-189) in the ␤VII-␤VIII (␤9-␤10) loop lead to the negatively charged surface of ALKBH7 (Figs. 2 and 6E). It is worth mentioning that ALKBH8 has a long ␤VII-␤VIII loop as well (46). Because the binding of tRNA is through its RNA recognition motif domain, this loop of ALKBH8 is proposed to mediate regulatory interactions remotely from the active site. There are also several acidic residues (Glu-306, Asp-316, and Asp-319) in the loop of ALKBH8. The similarity of the ␤VII-␤VIII loop between ALKBH7 and ALKBH8 indicates this loop may also serve as a binding site for substrates or partner proteins in ALKBH7.
Similar Conformation of ALKBH7 and Fe(II)/␣-KG-dependent Dioxygenase PHD2-Significant differences between ALKBH7 and other AlkB family members in both structure and assay suggest a distinct function of ALKBH7 from other known ALKBH members. Therefore, a structural similarity search with the DALI server (47) was carried out to investigate the potential function of ALKBH7 by structural comparison.
Besides AlkB family members, the best match is the human PHD2, the HIF prolyl 4-hydroxylase (PDB code 3HQR; r.m.s.d. 2.7 Å for 155 C␣ atoms, Z-score ϭ 14.3 and sequence identity 10%) (48). PHD2 is a cellular oxygen sensor which catalyzes the post-translational formation of 4-hydroxyproline in HIF-1␣ proteins (49,50). Structural alignment of ALKBH7 and PHD2 exhibits a similar conformation except for the longer substrate binding lid of PHD2 in the equivalent position of ␣2-␤3 loop of ALKBH7 and an extra ␣-helix in the C terminus of PHD2 (Fig.   7A). Interestingly, no major steric hindrance is found between the FIH1-␣ CODD peptide and ALKBH7 (Fig. 7B). Although residues forming the ␣-KG binding pocket are not strictly conserved between PHD2 and ALKBH7, the active site cavity of PHD2 is also predominantly lined by hydrophobic residues as ALKBH7.
Binding of HIF-1␣ COOD peptide to PHD2 is mainly through hydrophobic interactions (48). In ALKBH7, three acidic residues (Glu-62, Asp-64, and Asp-67) in the ␣2-␤3 loop along with two acidic residues (Glu-75 and Glu-77) in strand ␤3 create a negatively charged cleft opening into the active site (Figs. 2 and 6E). Furthermore, there are two acidic residues (Glu-207 and Glu-213) in the disordered C terminus, which may be involved in binding substrates. These features indicate that ALKBH7 could interact with a substrate containing positive charges, which is different from the FIH1-␣ CODD peptide. Combined with the structural similarity between ALKBH7 and PHD2, proline and lysine/arginine-containing peptides, FIGURE 5. ALKBH7 has an unusual active center. A, view of the ALKBH7 active site residues, ␣-KG and Mn(II). F o Ϫ F c (3.0, magenta mesh) is calculated after omitting ␣-KG and Mn(II). The ␣-KG and its interacting residues are shown as green sticks. The Mn(II) ion is shown as a red sphere and water is shown as a blue sphere. B, conformation comparison of ␣-KG and NOG in the two structures. ALKBH7-NOG complex is colored cyan, and ALKBH7-␣-KG complex is colored green. F o Ϫ F c electron density map (blue mesh) of NOG is contoured at 2.0. C, equilibrium binding analysis of ALKBH7 with ␣-KG in the presence of the Mn(II) ion using microscale thermophoresis (MST). D-I, active center of other AlkB family members. AlkB (lime green, PDB code 3BIE), ALKBH2 (blue, PDB code 3BUC), ALKBH3 (pink, PDB code 2IUW), ALKBH5 (magenta, PDB code 4NRO), ALKBH8 (orange, PDB code 3THT), FTO (yellow, PDB code 3LFM), respectively. Residues of the nucleotide recognition lid (NRL) Flip2, which are important in binding nucleobase (white sticks), are colored cyan. Note that Flip2 of ALKBH8 is completely disordered so the residues are not shown. The asparagine that forms a hydrogen bond with ␣-KG c-1 carboxylate oxygen is highlighted in red. The Mn(II) ion is shown as a red sphere and Fe(II) ion is shown in brown sphere.

Structure and Function of ALKBH7
OCTOBER 3, 2014 • VOLUME 289 • NUMBER 40 which could be the substrates of ALKBH7, need to be further identified.
Leu-110 Is Hydroxylated in the ALKBH7⅐␣-KG Complex-An extra electron density of Leu-110 was found in the ALKBH7⅐␣-KG complex (Fig. 8), indicating a post-translational modification for Leu-110. LC-MS was further used to identify this modification. The tryptic digest of ALKBH7 was separated by nanoACQUITY UPLC and detected via MS analysis. A mass shift of ϩ16 Da was found in MS analyses of the co-eluted modified and unmodified peptides (Fig. 9A). This result demonstrated Leu-110 was partially hydroxylated. MS analysis of the L110A mutant only showed the unmodified peptide, further confirming that Leu-110 is the hydroxylated residue (Fig. 9B).
Self-hydroxylation of Leu-110 was furthermore supported by loss of self-hydroxylation in the H121A/D123A and R197A/ R203A mutants (Fig. 9, C and D). These two mutants disturb the metal ion or ␣-KG binding and thus lack catalytic activity. Observation of self-hydroxylation suggests that ALKBH7 could catalyze post-translational modification of its substrate. Moreover, the Q90R mutation did not alter the self-hydroxylation.
The modification of ALKBH7 Leu-110 may have resulted from an attack by a hydroxyl radical formed by the uncoupled reaction. It is commonly thought that this uncoupled reaction could arise from binding inhibitors or poor substrates and could also occur in the absence of a primary substrate. There is evidence that autocatalyzed oxidative modifications happen to FIGURE 6. ALKBH7 lacks the nucleotide recognition lid, which is essential for binding nucleobase and exhibits a solvent-exposed active site. A, structural superimposition of ALKBH7 and other AlkB family members. The conserved nucleotide recognition lids of AlkB (green), ALKBH2 (blue), ALKBH3 (pink), and ALKBH5 (magenta) are highlighted. The Flip1 and Flip2 of the nucleotide recognition lid are labeled, and the corresponding motifs of ALKBH7 are colored red. Structure comparison clearly reveals the absence of Flip2 in ALKBH7. B, comparison of Flip1 of the nucleic acid oxygenases and the ␣2-␤3 loop of ALKBH7. The conformation of ␣2-␤3 loop is different from that of Flip1 and the ␣2-␤3 loop lacks arginine residues that are important in binding nucleobase and phosphate backbone. C, superimposition of ALKBH7, AlkB-dsDNA (green), and ALKBH2-dsDNA (blue) complexes. The ␣2-␤3 loop of ALKBH7 (red) overlaps with the repaired DNA strands. Note: AlkB and ALKBH2 are not shown. D, comparison of the DSBH fold of AlkB family members. The ␤IV-␤V loop and ␤VII-␤VIII loop are highlighted. E, electrostatic surface representation (basic in blue; acidic in red) of ALKBH7 and ALKBH2. The surface of ALKBH7 in equivalent position with the nucleic acid binding groove of other AlkB family members is negatively charged. The active site of ALKBH7 is solvent-exposed, and the cleft opening into the active site is negatively charged.
Fe(II)/␣-KG-dependent dioxygenases (51). Hydroxylation of Leu-177 was also found in ALKBH3 (19). Leu-177 locates in the central position of the active center of ALKBH3 serving as a "buffer stop" to prevent penetration of alkylated purines too deep in the catalytic pocket. The leucine is conserved in AlkB, ALKBH2, and FTO. It is proposed that formation of the hydrogen bond between the oxidized Leu-177 and Arg-275 negatively affects the coordination of ␣-KG and inactivates ALKBH3 to avoid generation of reactive oxygen species. L177E and L177Q mutants mimicking the oxidized leucine abolished the catalytic activity of ALKBH3. Fu et al. (7) generated the L110Q mutant of ALKBH7 to inactivate the protein according to the L177Q mutant of ALKBH3. However, the mutant could promote necrosis as well as the wild type. Structural superimposition of ALKBH3 and ALKBH7 exhibits a different location of ALKBH3 Leu-177 and ALKBH7 Leu-110 (Fig. 8). Actually, the corresponding residue of ALKBH3 Leu-177 is His-108 in ALKBH7 and not Leu-110. The hydroxylated Leu-110 is 6.0 Å away from the ␣-KG-binding Arg-203 of ALKBH7, so there is no interaction between them. Moreover, the hydroxyl group of the hydroxylated Leu-110 is 6.7 Å away from the catalytic iron center. Because the hydroxylated Leu-110 cannot negatively affect ␣-KG and the catalytic iron ion in ALKBH7, it is not surprising   that L110Q mutant of ALKBH7 could function as the wild-type protein.

DISCUSSION
The nonheme Fe(II)-and ␣-KG-dependent dioxygenases catalyze a wide range of oxidative reactions. Among them, AlkB family members, which mediate the oxidative DNA/RNA repair, are extensively studied both functionally and structurally. The function and substrate selection mechanisms of AlkB family members as nucleic acid oxygenase are clear because of their solved structures. They use Fe(II) to activate the dioxygen molecule for oxidation of the aberrant alkyl groups. A conserved nucleotide recognition lid covering the DSBH creates a cavity for modified nucleobase binding. This nucleotide recognition lid is also involved in the interaction with the phosphate backbone of nucleic acids. The minor ␤-sheet of DSBH plus several loops form the positively charged DNA/RNA binding groove. Although AlkB family members are mainly known as nucleic acid oxygenase in the past decade, several recent studies extend their function. ALKBH1 was reported to demethylate histone H2A (40), and ALKBH4 was shown to mediate demethylation of a monomethylated site (K84me1) in actin (41).
ALKBH7 is required for alkylation-and oxidation-induced programmed necrosis (7). It is unique because the ALKBH7depleted cells are more resistant to DNA damage caused by alkylating agents than normal cells. ALKBH7 plays a role totally different from the protective role of AlkB, ALKBH2, ALKBH3, and ALKBH8 upon exposure to DNA-damaging agents. To understand the unusual function mechanism of ALKBH7, we solved the structure of ALKBH7 in complex with Mn(II), ␣-KG, or NOG. The crystal structure of ALKBH7 exhibits several unique features compared with other AlkB family members.
First and foremost, the conserved nucleotide recognition lid of AlkB family members is absent in ALKBH7 (Fig. 6A). Without the essential nucleotide recognition lid, ALKBH7 displays a solvent-exposed active site incapable of positioning the modified nucleobase, which basically rules out the possibility that ALKBH7 functions as a nucleic acid oxygenase. The solventexposed active site may not only abolish the binding of a modified nucleobase but may also potentially promote promiscuous catalysis of ␣-KG oxidation in the absence of main substrates. Release of reactive oxygen species due to the uncoupled reaction is harmful in cells, and Leu-110 is hydroxylated because of this. Loss of the conserved ␣-KG c-1 carboxylate-binding asparagine in ALKBH7 (Figs. 2 and 5A) results in weaker binding affinity of ␣-KG, which may lower the uncoupled turnover of ␣-KG in adaptation to the solvent-exposed active site. Moreover, the hydrophobic nature of residues at the active site may reflect the tolerance of ALKBH7 to potential oxidative damage because they are less susceptible to oxidation. Second, ALKBH7 has a negatively charged surface in equivalent position with the nucleic acid binding groove of other AlkB family members (Fig.  6E). An additional loop between strands ␤9 (␤VII) and ␤10 (␤VIII) of ALKBH7 creates the outer wall of the minor ␤-sheet of DSBH (Fig. 6D and supplemental movie S1). Four acidic residues (Asp-182, Glu-183, Glu-184, and Glu-189) in this loop combining with Glu-150 and Glu-153 from the ␤6-␤7 loop, Glu-156 from strand ␤7, and Glu-178 from strand ␤9 are responsible for the negatively charged surface of ALKBH7. Binding of nucleic acid would be impossible because of the charge repulsion between the phosphate backbone and ALKBH7. In addition, the cleft opening into the active site is also negatively charged because of Glu-62, Asp-64, and Asp-67 in ␣2-␤3 loop and Glu-75 and Glu-77 in strand ␤3. The negatively charged surface of ALKBH7 indicates that it may interact with a positively charged molecule, such as the lysine/argininerich protein. Further studies are in urgently needed to uncover the substrates of ALKBH7 in programmed necrosis and fat metabolism.
Structural comparison of ALKBH7 and other Fe(II)-and ␣-KG-dependent dioxygenases reveals the potential role of ALKBH7 as a protein hydroxylase. The conformations of ALKBH7 and PHD2 are similar, and they all prefer hydrophobic residues in the active center. Hydroxylation of either Pro-402 or Pro-564 in human HIF-1␣ by PHD2 enables human HIF-1␣ binding to the von Hippel-Lindau protein (49,50). Hydroxylation of the target protein by ALKBH7 may enable its binding to the downstream elements and trigger the collapse of mitochondrial membrane potential. As a trigger of cell death, the interaction between ALKBH7 and its target protein and the hydroxylation reaction must be strictly controlled. No high confidence interaction partners of ALKBH7 was found by yeast two-hybrid screens (52), and the absence of stimulus might be the reason. CypD is a mitochondrial member of the cyclophilin family of peptidylprolyl cis-trans isomerases and has a crucial role in the mitochondrial permeability transition. CypD-deficient cells died normally in response to various apoptotic stimuli but showed resistance to necrotic cell death (53), and deletion of CypD in mice caused an abnormal accumulation of white adipose tissue resulting in obesity (54). These similar phenotypes of CypD and ALKBH7 indicate a potential relationship between them.
In summary, we solved the unique atomic resolution structures of ALKBH7 with a solvent-exposed active site and negatively charged surface. These features exclude the possibility that ALKBH7 functions as a nucleic acid oxygenase and indicate that ALKBH7 may interact with a positively charged molecule. The substrate of ALKBH7 may be peptides or proteins containing proline and lysine/arginine.
Our results point the way for future studies to address the signaling mechanism of the alkylation-and oxidation-induced programmed necrosis. The structure of ALKBH7 can also be exploited for the development of selective inhibitors to avoid unwarranted cell death in wounded or sick patients. Moreover, it also offers a foundation for drug design in treating metabolic diseases in humans and increasing the body weight of animals in livestock.