Ribosomal Oxygenases are Structurally Conserved from Prokaryotes to Humans

Abstract

2-Oxoglutarate (2OG)-dependent oxygenases play important roles in the regulation of gene expression via demethylation of N-methylated chromatin components^1,2, hydroxylation of transcription factors³, and of splicing factor proteins⁴. Recently, 2OG-oxygenases that catalyze hydroxylation of tRNA^5-7 and ribosomal proteins⁸, have been shown to play roles in translation relating to cellular growth, T_H17-cell differentiation and translational accuracy^9-12. The finding that the ribosomal oxygenases (ROX) occur in organisms ranging from prokaryotes to humans⁸ raises questions as to their structural and evolutionary relationships. In Escherichia coli, ycfD catalyzes arginine-hydroxylation in the ribosomal protein L16; in humans, Mina53 (MYC-induced nuclear antigen) and NO66 (Nucleolar protein 66) catalyze histidine-hydroxylation in ribosomal proteins rpL27a and rpL8, respectively. The functional assignments of the ROX open therapeutic possibilities via either ROX inhibition or targeting of differentially modified ribosomes. Despite differences in residue- and protein-selectivities of prokaryotic and eukaryotic ROX, crystal structures of ycfD and ycfD_RM from E. coli and Rhodothermus marinus with those of human Mina53 and NO66 (hROX) reveal highly conserved folds and novel dimerization modes defining a new structural subfamily of 2OG-oxygenases. ROX structures in complex with/without their substrates, support their functional assignments as hydroxylases, but not demethylases and reveal how the subfamily has evolved to catalyze the hydroxylation of different residue sidechains of ribosomal proteins. Comparison of ROX crystal structures with those of other JmjC-hydroxylases including the hypoxia-inducible factor asparaginyl-hydroxylase (FIH) and histone N^ε-methyl lysine demethylases (KDMs) identifies branchpoints in 2OG-oxygenase evolution and distinguishes between JmjC-hydroxylases and -demethylases catalyzing modifications of translational and transcriptional machinery. The structures reveal that new protein hydroxylation activities can evolve by changing the coordination position from which the iron-bound substrate oxidizing species reacts. This coordination flexibility has likely contributed to the evolution of the wide range of reactions catalyzed by iron-oxygenases.

To investigate the structural basis of catalytic differences within the ROX subfamily of JmjC-enzymes and their relationship with JmjC-KDMs, we conducted structural analyses on both prokaryotic (initially ycfD from E. coli and subsequently from the thermophile R. marinus) and human ROX (Mina53_26-465 and NO66_183-641). We used the R. marinus ycfD_RM to obtain an ycfD-substrate structure. All 4 ROX display striking similarities in their folds: the JmjC domain is followed by helical dimerization and C-terminal ‘winged-helix’ (WH)-domains¹³ (Fig. 1b). The ROX JmjC domains consist of 11-12 β-strands, 8 of which (I-VIII) form a double-stranded-β-helix (DSBH), which is stereotypical of 2OG-oxygenases (Fig. 1c and Extended Data Fig. 1)^14,15.

Figure 1. The overall folds of the ribosomal oxygenases.

a, Reactions catalyzed by ROX and related oxygenases. CAD: C-terminal transactivation domain of HIF-α; ARD: Ankyrin repeat domain. b, Ribbons representations of ycfD, ycfD_RM, Mina53 and NO66 homodimers. The monomers contain a JmjC domain with the double-stranded-β-helix (DSBH) core present in all 2OG-oxygenases (blue) followed by dimerization (yellow) and C-terminal ‘winged-helix’ domains (red). Domain architecture and a schematic representation of the DSBH core β-strands (βI-VIII) that form major (grey, βI, VIII, III and VI) and minor sheets (blue, βII, VII, IV and V) is shown boxed. The insert between βIV and βV (purple) is involved in substrate binding. The 3 Fe-coordinating residues are on the βII and βVII strands (black circles). 2OG is in green sticks; the 2OG C5-carboxylate binding residue, Arg (ycfDs) or Lys (hROX) from βIV is a black circle.

The dimerization domains have a 2-fold symmetry and comprise a bundle of 3 α-helices (Extended Data Fig. 2); dimers are stabilized by electrostatic/hydrogen-bonding and hydrophobic interactions. Consistent with a catalytic role for this domain, dimerization blocking substitutions, I211R_ycfD and R313E_Mina53 decrease activity. Hydrogen-bonding/electrostatic interactions are substantially more important in ycfD_RM dimerization than for the other ROX, consistent with the increased occurrence of electrostatic interactions in thermophiles¹⁶. The ROX C-terminal domains, which are required for activity (Extended Data Fig. 3), are reminiscent of WH-domains involved in protein-protein and protein-nucleic-acid interactions¹³; however their overall negative charge suggests they do not directly bind nucleic acids. In contrast to ROX, other JmjC-hydroxylases (FIH¹⁷, tRNA wybutosine-synthesizing enzyme 5/TYW5⁶, JmjC-domain containing protein 4/JMJD4⁹, JMJD5¹⁸, JMJD6⁴) and KDMs do not contain a WH domain (Fig. 2). The combined structures lead to the proposal that ROX fold evolved into those of JmjC-hydroxylases and KDMs partly via loss of the WH-domain which enabled the C-terminal helical bundle to take on other roles as in KDMs or the dimerization mode as observed in FIH¹⁷.

Figure 2. Comparison of the substrate structures for ROX/JmjC enzymes.

Ribbons representations from ROX and related 2OG-oxygenase-substrate complexes: a, Mina53·Mn·2OG·rpL27a_(32-50) (P2₁2₁2₁, 2.05 Å), b, NO66·Mn·NOG·rpL8_(205-224) (C2, 2.35 Å), c, ycfD_RM·Mn·NOG·L16_(72-91) (P2₁2₁2₁, 3.0 Å), d, FIH·Fe·NOG·HIF-1α_(786-826) (PDB: 1H2K), e, PHF8·Fe·NOG·histone H3K4me3K9me2_(2-25) (PDB: 3KV4), f, KDM4A·Ni·NOG·histone H3K9me2_(7-14) (PDB: 2OX0). For comparison, the DSBH core of each structure is in a similar orientation. Note the directionality of substrate binding in the JmjC domains. The active site metals (Fe/ surrogate) are color-coded spheres. Analyses of the structures reveal that the ROX overall folds (a-c), oligomerization states and active site architectures are evolutionarily conserved.

ROX structures were determined in complex with Mn(II) and 2OG/N-oxalylglycine (NOG), replacing Fe(II) and 2OG. As for most 2OG-oxygenases, the metal is octahedrally coordinated by a 2-His-1-carboxylate triad from DSBH-βII and - βVII^14,15 (Fig. 3); 2 coordination sites are occupied by the 2OG/NOG oxalyl group leaving one for H₂O/O₂ binding (Fig. 4 and Extended Data Fig. 4). With the ycfDs the NOG C5-carboxylate is positioned to salt bridge with Arg140_ycfD/Arg148_ycfDRM on DSBH-βIV (Extended Data Fig. 4). This arrangement is notable because with other 2OG-oxygenases where the 2OG C5-carboxylate interacts with an Arg-residue, it is located on βVIII^14,15. In hROX, the 2OG C5-carboxylate interacting residue is a lysine (Lys194_Mina53/Lys355_NO66) from βIV, as in most JmjC-hydroxylases and KDMs. These observations lead to the proposal that the eukaryotic JmjC-hydroxylases/KDMs evolved from prokaryotic ycfDs/ROX.

Figure 3. Features of ROX-substrate binding.

Ribbons representations of Mina53 (a), NO66 (b) and ycfD_RM (c) monomers showing difference electron density (Fo-Fc OMIT) for substrates contoured to 3σ (right panels). Left panels show active site surface representations, showing key hydrogen-bonds/polar interactions (dotted lines) with substrates. a, With Mina53, the His39_rpL27a imidazole nitrogens form hydrogen-bonds with Tyr167/Ser257 (Nδ_His39-OH_Tyr167 2.9 Å; Nε_His39-Oγ_Ser257 3.1 Å). b, In NO66, His216_rpL8 is similarly bound in a deep pocket; the His216_rpL8 imidazole nitrogens form hydrogen-bonds with Tyr328/Ser421 (Nδ_His216-OH_Tyr328 3.2 Å; Nε_His216-Oγ_Ser421 2.7 Å) and hydrophobic interactions with Ile244, that project its pro-S hydrogen toward the metal (metal-β-CH₂, 4.4 Å). While Mina53 (a) uses 4 primary amides, Asn101, Gln136, Gln139 and Asn165 to interact with rpL27a backbone amides, NO66 (b) uses 2 arginines (272, 297) to hydrogen-bond with the Asn215_rpL8 sidechain and His216_rpL8 backbone. In the ycfD_RM·L16 complex (c), Arg82_L16 binds in a pocket defined by the Tyr137/Met120 sidechains, which form π-cation and hydrophobic interactions with Arg82_L16 sidechain. The Arg82 guanidino group makes electrostatic interactions with the Asp118_ycfDRM carboxylate (O-NH, 2.8-3.1 Å) and hydrogen-bonds to Ser208_ycfDRM (Nε_Arg82-OH_Ser208 3.5 Å; Nη_Arg82-CO_Ser208 3.2 Å). Although Tyr167_Mina53/Tyr328_NO66 are not positionally related to Tyr137_ycfDRM, the role of the serine (Ser257_Mina53, Ser421_NO66, Ser208_ycfDRM, β-VIII) in binding the hydroxylated His/Arg is conserved in ROX. Substitutions of these residues cause significant loss of activity (see Extended Data Fig. 6).

Figure 4. Proposed sequence of evolution of active metal chemistry of ROX and related JmjC 2OG-oxygenases.

The figure compares views from active sites of representative JmjC-enzymes and suggests how the ROX fold evolved into JmjC-hydroxylases and -KDMs. Structurally informed cross-genomic bioinformatic analyses imply that the ROX are the earliest identified JmjC 2OG-oxygenases³⁰; ycfD and NO66 both exist in prokaryotes but only NO66 is identified in eukaryotes. Coupled to the analyses of the active sites, these analyses imply NO66/close-relatives are the precursors of Mina53 and other JmjC-hydroxyalses and KDMs. a, Upper panel: structure based alignment of ROX, FIH, PHF8 and KDM4A with DSBH core labeled βI-VIII, iron-coordinating and the 2OG C5-carboxylate binding residues in red and green. Lower panels: analyses of active sites suggest conservation of metal-/2OG-binding in ROX, FIH and KDMs, note the 2OG C5-carboxylate binding residue (usually from βIV in JmjC-enzymes), changes from an Arg (in ycfDs) to a Lys (in hROX, JmjC-hydroxyalses/KDMs) (Extended Data Fig. 4). b, Overlays of the NO66/ycfD_RM, NO66/Mina53, NO66/FIH, NO66/KDM4A active site views. The hydroxylated β-methylenes nearly superimpose in ROX, such that the oxidized C-H bonds (red arrows, 3-pro-R in Arg82_L16 and 3-pro-S in His39_rpL27a and His216_rpL8) project toward the metal. The spatial relationship of the hydroxylated C3/Nε-methyl carbon with respect to the metal (and associated reactive oxidizing species) is conserved in ROX and the demethylases, e.g. KDM4A, but not in FIH. Note the different hydroxylation positions, but the similar orientation of Asn803_CAD/FIH (hydroxylated) and Asn215_rpL8/NO66 (not hydroxylated).

Initial attempts to obtain substrate complexes by co-crystallization/soaking crystals were unsuccessful. We therefore pursued alternatives, one involving using a thermostable ycfD homolog, which we considered may have a relatively low substrate K_d, enabling complex crystallization. The R. marinus ycfD_RM (31% identity with ycfD) catalyzes L16 fragment (20-mer, aa Lys72-Glu91) Arg82-hydroxylation with ~7 fold lower K_m relative to ycfD (268 μM and 1.9 mM, respectively). An ycfD_RM·L16_72-91 structure, obtained by co-crystallization was solved by molecular replacement using the apo-ycfD structure (PDB: 4CCL). The overall ycfD/ycfD_RM structures are similar (Cα rmsd: 1.58 Å); L16 residues Lys77-Lys85 are visible in the electron density map (Fig. 3c).

For the hROX, we employed electrospray-ionization mass spectrometry guided disulfide cross-linking^19,20 to obtain substrate complexes (Extended Data Fig. 5). Structures were obtained for wild-type (wt) NO66·rpL8-G220C (complex 1), L299C/C300S-NO66·rpL8-G220C (complex 2), and S373C-NO66·rpL8-G214C (complex 3) pairs. Electron density corresponding to rpL8 residues 215-223 (complex 1), 213-223 (complex 2) and 212-223 (complex 3) was observed at the active site (Fig. 3b and Extended Data Fig. 5). The rpL8 residues (215-219) including that hydroxylated, i.e. His216, adopt near identical conformations (Cα rmsd, 0.29-0.36 Å) implying all 3 structures represent catalytically functional complexes (Extended Data Fig. 5). In the light of NO66·rpL8 structures, we identified a Mina53 residue (Tyr209) suitable for cross-linking: Y209C-Mina53 crystallized in complex with rpL27a-G37C with electron density observed for rpL27a residues 36-44 (Fig. 3a). Further validation of the functional relevance of the cross-linked structures comes from comparisons with the wt-ycfD_RM·L16 structure and kinetic studies demonstrating activities with most variants (Extended Data Fig. 6).

Mina53/NO66 bind their rpL27a/rpL8 substrates in a conserved manner (Cα rmsd, rpL27a_38-43, rpL8_215-220, 0.8 Å). Comparison of hROX and ycfD_RM complexes reveals similarities in substrate binding particularly for the hydroxylated-residue and that to its N-terminus. In all ROX complexes, substrates bind with the same N-/C-directionality, as for FIH¹⁷ and for one KDM, plant homeo domain finger 8/PHF8²¹ (and likely other KDM2/7 subfamily members) but differing from that for most KDMs (JMJD2A/KDM4A²², JMJD3/KDM6B²³, KDM6A²⁴) (Fig. 2). The substrates bind in shallow channels on the ROX surfaces and form multiple interactions/hydrogen-bonds with residues from DSBH-βI, -βII, -βVIII, and the extended βIV-βV loop. While the N-terminal regions of rpL27a (aa 36-39), rpL8 (213-216) and L16 (78-81) bind similarly, the C-terminal regions of rpL27a (40-44) and rpL8 (217-223) form more extensive interactions with hROX than does L16 (83-85) with ycfD_RM (Fig. 3). Notably, both rpL27a and rpL8 substrates make hydrophobic contacts with the WH-domains in Mina53 and NO66 (Extended Data Fig. 3). In addition, Mina53 forms a catalytically important salt-bridge interaction between Arg42_rpL27a and Asp333_Mina53 located on the α-helix connecting the dimerization and WH-domains (Extended Data Figs 6 and 7).

The general binding mode of the hydroxylated residues is conserved between prokaryotic and hROX, i.e. they bind in deep pockets and the positions of the hydroxylated β-methylenes nearly superimpose (Fig. 4). There are, however, clear differences in the way hROX and ycfD_RM bind their target residue sidechains (Fig. 3). With hROX, the binding of His39_rpL27a/His216_rpL8 involve a series of hydrogen-bonds to backbone amides/sidechains of hROX residues: Gln136_Mina53/Arg297_NO66, Asn165_Mina53/Asn326_NO66, Tyr167_Mina53/Tyr328_NO66 and Ser257_Mina53/Ser421_NO66 (Fig. 3a, b). With ycfD_RM, the Arg82 ‘slots’ into a hydrophobic cleft defined by Tyr137_ycfDRM and Met120_ycfDRM sidechains and hydrogen-bonds to Asp118_ycfDRM and Ser208_ycfDRM (Fig. 3c). Mutagenesis studies on ROX support the observed binding modes of the substrate residues (Extended Data Figs 6 and 8).

There are conflicting reports as to the catalytic activities of some JmjC-hydroxylases, including NO66 which has been assigned as both a hydroxylase⁸ and a KDM²⁵. Comparison of the ROX and KDM/FIH (Figs 2 and 4a) identifies distinctive structural features characteristic of JmjC-hydroxylases and -KDMs, in addition to the roles of the WH-domains (see below). This is important because it supports the assignment of hydroxylase (but not demethylase) activities for ROX and other human JmjC-hydroxylases e.g. FIH¹⁷, JMJD6⁴. In our assays with isolated Mina53 and NO66 we have consistently not observed enzyme-catalyzed demethylation under conditions where JmjC-KDMs are active⁸. While we cannot rule out the possibility that some of the JmjC-hydroxylases may have KDM activities under different conditions or in cells, the multiple structures reported here suggest that for this to occur, substantial active-site rearrangements would be required on substrate binding. In ROX and KDM4A·H3K9me₂ (PDB: 2OX0)²² complex structures the different substrates bind with ‘opposite’ N- to C-directionalities with respect to the catalytic machinery. The histone K9me_n side chain is positioned similarly to the ROX-hydroxylated residue sidechains; however, because KDM-catalyzed hydroxylations occur at N^ε-methyl lysine-residue termini, their target residues do not penetrate as far into the enzyme active site (Fig. 4a). The ROX also lack 2 flexible loops linking α4-βI (aa 164-175) and α9-α10 (aa 302-317) in KDM4A, which are conserved in KDMs^21-24 and which form important interactions with the Kme_n sidechain, illustrating how the ‘core’ ROX-fold has been modified by evolution to accommodate the Kme_n sidechain.

Like ROX, FIH catalyzes β-hydroxylation of an Asn-residue in its HIF-α transcription factor substrate¹⁷ and of other residues including histidines in ankyrins²⁶. Superimposition of hROX/FIH-substrate structures is interesting from catalytic and evolutionary perspectives. Although both FIH/hROX catalyze histidine 3S-hydroxylation, the positions of their substrate imidazoles is strikingly different (Fig. 4 and Extended Data Fig. 9). The positioning of hydroxylated methylenes relative to the metal differs substantially: in the overlaid structures, the angle between the metal and the Cβ atoms of the His39_rpL27a/His216_rpL8 (hROX substrate) and Asn803_HIF1α (FIH substrate) is ~50° (Fig. 4 and Extended Data Fig. 9), demonstrating that the reactive oxidizing intermediates (Fe(IV)=O)^15,27 react from different coordination positions in different oxygenases. Studies with 2OG-dependent halogenases have led to the proposal that iron-bound reactive intermediates abstract a hydrogen from substrate and deliver a halogen/hydroxyl from different coordination positions to form products²⁸. In contrast, our work implies flexibility in the coordination positions with respect to the hydrogen abstracted in different JmjC-hydroxylases from which the ferryl-oxo reacts. Together with other structural considerations, this observation has consequences for the evolution of the JmjC-enzymes.

The rpL8 (NO66 substrate) has an Asn at the −1 position relative to the hydroxylated His216 (ycfD/Mina53 substrates have hydrophobic residues at the analogous positions). The Asn215_rpL8 methylene is only slightly (0.5 Å) further from the metal than that of His216_rpL8, revealing exquisite sensitivity of oxygenase catalysis to geometric positioning. There is a striking correlation in the binding of Asn215_rpL8/Asn803_HIF1α to NO66/FIH, even though one residue is hydroxylated and one not; the primary amides of both Asn215_rpL8/Asn803_HIF-1α hydrogen-bond with primary amides, i.e. Asn376_NO66 and Gln239_FIH. Collectively these observations reveal 2OG-oxygenases can evolve new activities not only by ‘directly’ altering the nature of enzyme-substrate interactions (including by altering the directionality of substrate binding), but also by changing the coordination position from which the ferryl-intermediate reacts.

The combined structures reveal that the observed mode of ROX hydroxylations have likely evolved into the FIH/related JmjC Lys-hydroxylation and KDM type hydroxylation/demethylation modes both by altering the coordination position from which the ferryl-oxo reacts and by engineering the depth of substrate penetration. Structurally informed phylogenetic analyses (Extended Data Fig. 10), coupled to the observation that NO66 is more widely distributed than FIH/Mina53, reveal that the prokaryotic ycfDs evolved into NO66, which is a branchpoint leading to the eukaryotic JmjC-hydroxylases/demethylases. 2OG-oxygenases are amongst the most catalytically flexible of all enzyme-families. Recent work reveals that FIH manifests remarkable catalytic promiscuity, including the ability to oxidize Asn- and His-residues²⁹. Our structural studies reveal that ROX react with substrates via a different but evolutionarily related binding mode to FIH. The catalytic capabilities of 2OG-oxygenases for protein oxidations thus likely extend beyond those presently identified.

METHODS

Recombinant protein production and enzyme assays

cDNA sequences encoding N-terminally truncated Mina53 (aa 26-465) and NO66 (aa 183-641) were PCR amplified from Mammalian Gene Collection (MGC) (accession no.: BC014928 and BC011350, respectively) and cloned into pNIC28-Bsa4 vector. Full length ycfD was cloned into pET-28a(+) vector (Novagen, Madison, WI, U.S.A.) as described⁸. ycfD_RM gene (NCBI GENE ID: 8566662) was amplified by PCR from genomic DNA of R. marinus, and was cloned into pGEM®-T Easy Vector and then into pET-28a(+). Stratagene’s QuickChange site-directed mutagenesis kit was used to make all ROX mutations using the above constructs as templates.

Wild-type ROX enzymes/variants were produced as native His₆-tagged proteins in Escherichia coli BL21(DE3) as described⁸. For crystallization experiments, selenomethionine (SeMet) derivatized enzymes, SeMet-Mina53 and SeMet-ycfD were produced in E. coli BL21(DE3)-R3-pRARE2 and BL21(DE3) strains, respectively. In general, cells were grown in Le Master media³¹ (alternatively in SelenoMethionine Medium Base plus Nutrient Mix) supplemented with selenomethionine (40-50 mg·mL⁻¹) and kanamycin (30 μg·mL⁻¹) at 37 °C (while shaking at 200 rpm) until an OD₆₀₀ (optical density at 600 nm) of 1.2 (SeMet-Mina53) or 0.6 (SeMet-ycfD) was reached. Protein expression was then induced with 0.2 mM (SeMet-Mina53) or 1.0 mM (SeMet-ycfD) isopropyl β-D-1-thiogalactopyranoside (IPTG) and allowed to continue for 18 h at 18 °C. All native/SeMet-derivatized proteins were purified from cell lysates using immobilized Ni²⁺ affinity chromatography with gradient elution using imidazole and/or ion-exchange chromatography. For ycfDs, imidazole was removed by buffer exchange to 50 mM Hepes-Na pH 7.5 using a PD10 desalting column followed by a further purification using Q-Sepharose HP (ycfD) or SourceQ 16 (ycfD_RM) anion exchange chromatography. For Mina53 and NO66, the His₆-tag was removed by incubation with TEV protease followed by a final-step purification using size-exclusion chromatography in 50 mM Hepes-Na pH 7.5, 500 mM NaCl, 5% (v/v) glycerol, 0.5 mM TCEP (tris(2-carboxyethyl)phosphine). Proteins were concentrated to 10-30 mg·ml⁻¹ and were of >95 % purity, as determined by SDS-PAGE. All columns were supplied by GE Healthcare. Assays were performed as described⁸.

Crystallization, data collection and processing

Crystals of Mina53, NO66, ycfD and ycfD_RM complexes were grown as described in Supplementary Table 1. In general, crystals were cryoprotected by transferring to a solution of mother liquor supplemented with 20% (v/v) ethylene glycol (Mina53/NO66) or 25% (v/v) glycerol (ycfDs) before being cyro-cooled in liquid N₂.

As described in Supplementary Tables 2-4, data on native and SeMet-derivatized crystals were collected at 100K using synchrotron radiation at the Swiss Light Source (SLS) beamline X10SA, European Synchrotron Radiation Facility (ESRF) beamline BM16 and Diamond Light Source (DLS) beamlines. The data were processed as outlined in Supplementary Tables 2-4.

Structure solution and refinement

Mina53 structures

SHAKE-AND-BAKE³² was used to identify 5 Se-positions in the SeMet-Mina53 dataset (P4₃32 space group); refinement of heavy atom parameters and phasing was carried out with SHARP³³ using the single ismorphous replacement with anomalous scattering (SIRAS) method with Mina53.NOG (native) as the native and SeMet-Mina53 as the derivative dataset (Supplementary Table 2). The electron density map after density modification with SOLOMON³⁴ was of good quality; automated model building with ARP/wARP resulted in a >80% complete model with one Mina53 molecule per asymmetric unit, which corresponds to an unusually high solvent content of ~75%. Refinement was carried out with BUSTER³⁵ and after several cycles of manual rebuilding with COOT³⁶, the model converged to 19.7% R_cryst and 22.9% R_free. Atomic coordinates and structure factors for this structure are deposited in the PDB database with the accession code 2XDV.

SeMet-MINA53–2OG structure was solved by using phases from a highly redundant single-wavelength anomalous dispersion (SAD) data set collected around the Se absorption edge. Using Patterson seeding and dual-space direct methods, SHELXD (SHELXCDE pipeline³⁷/CCP4 suite³⁸) located 6 out of 8 possible Se-sites using a SeMet-Mina53 SAD data set. Refinement of substructure solution followed by density modification with SHELXE³⁷ resulted in good-quality initial phases to 2.8 Å resolution. Automated model building with BUCCANEER³⁹ resulted in a model where core regions including the JmjC and dimerization domains were built. Iterative refinement using CNS 1.3⁴⁰ and model building using COOT³⁶ continued until R_free was around 30%. Final rounds of manual fitting using COOT³⁶ and refinement using a combination of CNS 1.3⁴⁰ and PHENIX⁴¹ continued until R_cryst/R_free no longer improved (Supplementary Table 2). This structure (deposited with PDB ID: 4BU2) was then used as a search model to solve the structure of Y209C Mina53 in complex with rpL27aG37C by molecular replacement with PHASER⁴² (P2₁2₁2₁ space group, resolution, 2.05 Å). The quality of all Mina53 structures was validated using MOLPROBITY⁴³ with >95% of the residues in the favored region of the Ramachandran plot.

NO66 structures

An N-terminally truncated form of Mina53 (aa 30-260), comprising the JmjC domain, was used as a search model for MR using PHASER⁴². The two molecules in the asymmetric unit of NO66 were readily located, but the electron density away from the JmjC core of NO66 was ambiguous. Density modification with RESOLVE⁴⁴ as implemented in PHENIX⁴¹, which took advantage of the 2-fold non-crystallographic symmetry (in a P2₁2₁2 space group), led to a significant map improvement and allowed automated model building with BUCCANEER³⁹. Refinement was carried out with REFMAC⁴⁵; after several cycles of manual rebuilding with COOT³⁶, the model converged to 18.5% R_cryst and 23.1% R_free. Atomic coordinates and structure factors for this structure are deposited in the PDB (accession code 4DIQ). The remaining NO66 structures including those in complex with substrate rpL8, were solved in P2₁ or C2 space groups (resolution, 2.15-2.50 Å) with 2-4 molecules per asymmetric unit (Supplementary Table 3) using the NO66/P2₁2₁2 structure (PDB ID: 4DIQ) as a search model. Iterative rounds of model building using COOT³⁶ and refinement using PHENIX⁴¹ and/or CNS 1.3⁴⁰ were performed until the decreasing R_cryst and R_free no longer converged (Supplementary Table 3). All residues were in acceptable regions of Ramachandran plots as calculated by MOLPROBITY⁴³.

YcfD structures

SOLVE was used to locate 17 out of 22 possible Se-sites using the SeMet-ycfD dataset. 8 pairs of sites were related by non-crystallographic symmetry. The initial electron density map after solvent flattening density modification with RESOLVE⁴⁴ was of good quality and automated model building resulted in a model where core regions (60% of residues in the crystallized protein’s sequence) of both molecules in the asymmetric unit were built. Refinement and fitting cycles were performed using PHENIX⁴¹ and COOT³⁶ that converged to a final 19.5% R_cryst and R_free 25.0%. Phasing and refinement statistics are summarized in Supplementary Table 4. Structures of ycfD_RM in complex with IOX3⁴⁶ or substrate L16 were solved by MR using the ycfD structure as the search model. The structural refinement was carried out with PHENIX with iterative rebuilding of the models using COOT until R_cryst/R_free converged to final values (Supplementary Table 4).

Supplementary Material

ED Fig1

Extended Data Figure 1 | Schematic protein topologies of ROX and related 2OG oxygenases. Protein topologies of (a) Mina53·Mn·2OG·rpL27a_(32-50), (b) NO66·Mn·NOG·rpL8_(205-224), (c) ycfD_RM·Mn·NOG·L16_(72-91), (d) FIH·Fe·NOG·HIF-1α_(786-826) (PDB: 1H2K), (e) PHF8·Fe·NOG·histone H3K4me3K9me2_(2-25) (PDB: 3KV4), and (f) KDM4A·Ni·NOG·histone H3K9me2_(7-14) (PDB: 2OX0) (substrates are not shown). DSBH core elements, labeled βI-βVIII, are in smudged green, helices in cyan, additional β-strands in red, random coils in black and the insert between the fourth and fifth β-strands in blue. Note that not all the DSBH oxygenases maintain antiparallel hydrogen-bond pairing between βII and βVII even though the ϕ/ψ angles (βII) are within the β-region of the Ramachandran plot. Figures were generated using TopDraw⁴⁷.

ED Fig2

Extended Data Figure 2 | ROX dimerization domains. a, Comparison of the dimerization domains in ROX and FIH. b, Intermolecular interactions observed at dimerization interfaces (monomer A, grey; monomer B, yellow). Validation of the functional relevance of the ROX dimers comes from biochemical/kinetic studies demonstrating loss of activities with most variants. The dimer interfaces in the ROX are related to that of FIH; we propose the FIH dimerization fold evolved from that of the ROX ^17,48. The large buried surface area (>3000 Ų) within all ROX dimerization domains is sufficient for dimerization in solution, as reported for NO66⁴⁹. The interactions observed in dimerization include both hydrogen bonds/electrostatic interactions and hydrophobic interactions.

In the ycfD/ycfD_RM dimerization domains, residues involved in hydrophobic interactions are mainly from α2 and are well conserved (ycfD_RM residues in parentheses): Phe214 (Met223), Val242 (Ile250), Met247 (Leu255), Leu250 (Ile258), Met253 (Leu261), Met254 (Leu262), Leu257 (Leu265), Ile258 (Ile257). Hydrogen bonding/electrostatic interactions are more important in ycfD_RM dimerization than in ycfD/hROX. The network of hydrogen bonds between the two ycfD_RM monomers A and B includes Asp256_A-Arg269_B-Gln259_A-Asp267_B-Arg263_A which, due to two-fold symmetry, creates a total of 8 hydrogen bonds. In ycfD, Leu255 (Arg263 in ycfD_RM) is positioned at the center of the equivalent network. Further, in ycfD_RM Gln216 is positioned hydrogen bond with the backbone amide N of Arg234 and carbonyl O of Leu261. Hydrogen bonding in ycfD dimerization is less extensive, with only Asn226 amide-N positioned to form a hydrogen bond to the hydroxyl group O of Thr207 and Arg208 hydrogen bonding with carbonyl O of Gly224. However, hydrophobic/aromatic clusters are involved in ycfD dimerization, including by the sidechains of Leu210_A, Leu223_A, Tyr217_A (α1), Phe264_A, Trp267_A, Phe268_A and Phe271_A (α3) from monomer A and Val242_B, Met247_B, Leu250_B (α2) from monomer B.

As in the ycfDs, in NO66 there is only one apparent salt-bridge interaction at the dimer interface, i.e. between Arg474 and Asp495 (Arg474_A NH1-Asp495_B Oδ1, 2.9Å; Arg474_A NH2-Asp495_B Oδ2, 2.7Å), which links the α2 and α3 helices of opposite monomers. Similarly a ‘complex salt-bridge’ is observed in Mina53 between Arg313 and Glu320/Asp317 (Arg313_B NH1-Glu320_A Oε2, 3.2Å; Arg313_B NH2-Glu320 Oε1, 2.7Å; Arg313_A NH2-Asp317 Oδ1, 2.9Å) that connects α2 helices of different monomers. Backbone amide hydrogen bonding additionally occurs between NO66 residues, Asn426 and Leu454, Arg452 and Trp428, Phe450 and Gly429. Mina53 also has backbone-to-sidechain interactions between residues from flexible loops connecting α1-α2 and α2-α3 helices (Gln297_B O-Lys331_A Nζ, 3.7 Å; Ser300_B Oγ-Glu324_A O, 3.1 Å). The role of hydrophobic/aromatic clusters in dimerization is apparent in NO66 where the α2 helices from different monomers are further apart when compared with those of ycfDs and Mina53 and hence have less buried surface area. However, in NO66, an apparent hydrophobic cluster forms between the N-terminal part of α1 and the C-terminal part of α2. Trp428_NO66 (Trp264 in Mina53) is positioned at the start of the α1 helix of monomer A and forms the center of a hydrophobic cluster, interacting with residues Phe431_A, Ile435_A and Leu432_A on monomer A, and Val481_B, Leu484_B, Met462_B, Phe477_B and Pro455_B on monomer B. NO66 Trp428 also forms an apparent cation-π interaction with residue Lys480. A similarly positioned Trp264 in Mina53 maintains hydrophobic contacts with Phe267 and Leu268 of the same monomer and with Ile290, Pro291 and Leu294 of the other in addition to a cation-π interaction with Arg307. Other hydrophobic contacts observed in Mina53 dimerization, involving the α1 and α2 helices of different monomers include between the sidechains of residues Leu308/α2 (interacting with Leu319/α2 and Phe267, Leu268, Thr271 of α1), Leu312/α2 (interacting with Ile272/α1 and Leu315/α2) and Phe277/α1 (interacting with Val276, Leu269 and Ile272 of α1).

Disruption of ROX dimerization leads to loss of activity as observed for Mina53 R313E and ycfD I211R variants as well as for truncated Mina53 (1-265, 1-299) without dimerization and the C-terminal domains. Non-denaturing gel electrophoresis was used to investigate ROX oligomerization states in solution, which demonstrates disruption of dimerization in I211R ycfD and Mina53 R313E. The loss of activity via destabilizing ROX dimerization is reminiscent of similar roles of FIH dimerization in catalysis (An FIH L340R variant that was predominantly monomeric is inactive)⁵⁰. Data are mean and s.e.m. (n=3).

ED Fig3

Extended Data Figure 3 | Interaction of the ROX C-terminal ‘winged helix’ (WH) domains with their respective ribosomal protein substrates. The figure shows how the ROX C-terminal domains interact with their substrates. A DALI search⁵¹ indicates that a close structural homolog of the ROX C-terminal domain is the ‘peptide clamp’ (‘winged helix’ WH) domain of MccB, an enzyme involved in the biosynthesis of the microcin C7 antibiotic⁵². WH domains, a subtype of the helix-turn-helix (HTH) family, are nucleic acid/protein interacting domains and occur in different cellular pathways from transcriptional regulation to RNA processing¹³. Although the overall negative charge of ROX WH domains suggests that they may not directly interact with nucleic acids, it is notable that the prokaryotic ribosomal proteins L6, which is located proximate to L16 in intact ribosomes⁵³, and the transcriptional regulator PhoP contain WH folds⁵⁴; the latter is interesting because in the E. coli K12 genome the ycfD gene is located adjacent to those for the PhoP/PhoQ two component signaling system, which is involved in stress responses⁵⁵. a, General topology of the C-terminal WH domain showing two distinct binding sites for L16 (yellow) and rpL27a (magenta)/rpL8 (orange) involving residues either from an N-terminal loop connecting the WH and dimerization domains (as in ycfD_RM) or from an extended loop between WH β3-β4 (as in hROX). Figures b-e compare the WH domains in MccB (b), Mina53 (c), NO66 (d) and ycfD_RM/ycfD (e) showing the interactions observed between this domain and the substrate(s). Note that whilst both the rpL27a and rpL8 substrates make hydrophobic contacts with the WH domains in Mina53 (Met405 and Met406) (c) and NO66 (Val576 and Tyr577) (d), ycfD_RM employs Arg285 to form hydrogen bond with the Met83_L16 (Arg285_ycfDRM NH2-Met83_L16 O, 2.5 Å) (e). Right panels show the partial loss of activity with mutations of Mina53 (M405A), NO66 (Y577A) and ycfD (H277C) residues from WH domains. Data are mean and s.e.m. (n=3).

ED Fig4

Extended Data Figure 4 | Comparison of 2OG/co-substrate binding in ROX and representative 2OG oxygenases. The identity of the basic residue (Arg or Lys) that binds the 2OG C5-carboyxlate via electrostatic interactions is indicated along with which of the 8 DSBH (I to VIII) strands it is located on. The occurrence and positioning of the basic Arg/Lys is subfamily characteristic^14,15. 2OG binding also involves other polar residues including alcohols, i.e. a Ser (βVIII, part of ‘RXS’ motif as present in e.g. DAOCS, ANS, FTO, algal P4H) or Thr (βII, e.g. as in some KDMs: JMJD3, JMJD6, PHF8, UTX) or Tyr (non-DSBH β-strand, e.g. as in FIH, KDM4A, ABH2, PHD2) and sometimes, water molecule(s) (reviewed in ^15,56,57). In an analogous position to the serine of ‘RXS’ motif (βVIII), the hROX have histidine-residues, His253_Mina53/His417_NO66 (βVIII), that form part of a hydrogen-bond network involving Thr255_Mina53/Thr419_NO66 (βVIII), a water molecule, and the 2OG carboxylates. Although ycfD/ycfD_RM has Asn197/Thr206 at this position (βVIII), it is the conserved serine from βI (122/ycfD_RM and 114/ycfD) that is positioned to hydrogen bond with the 2OG C5-carboxylate. Abbreviations: DAOCS, deacetoxycephalosporin C synthase; ANS, anthocyanidin synthase; FTO, fat mass and obesity associated protein; algal P4H, prolyl-4-hydroxylase from Chlamydomonas reinhardtii; FIH, asparaginyl hydroxylase factor inhibiting HIF (hypoxia-inducible factor); JMJD3 and 6: JmjC-domain containing protein 3 and 6; PHF8: PHD (plant homeo domain) finger protein 8; UTX: ubiquitously-transcribed X chromosome tetratricopeptide repeat protein; ABH2, alkylated DNA repair protein (AlkB) homolog 2; PHD2, prolyl hydroxylase domain 2.

ED Fig5

Extended Data Figure 5 | hROX-substrate complexes showing disulfide cross-linking sites (red arrows) and difference electron density (F_o-F_c OMIT) for the substrate residues contoured to 3σ. (a) Strategy adopted to obtain the cross-linked structures (the same strategy can be used for other protein hydroxylases/KDMs). (b-d) Different disulfide cross-linking sites (red arrows) that form NO66·rpL8 cysteine-disulfide pairs under equilibrating conditions. Analyses of the 2OG-oxygenase-substrate complexes reveal that substrate residues at ±2 positions relative to the hydroxylated residues make interactions with enzyme residues within an ~12 Å radius of the metal. To obtain stable NO66-rpL8 complexes, we engineered NO66 variants substituting Cys-residues within ~12 Å radius of the metal at positions considered likely involved in substrate binding based on the analyses of other 2OG-oxygenase-substrate structures^21,22,26 and the evolutionary/phylogenetic analyses of NO66/NO66-like proteins in eukaryotes. We also substituted Cys residues at ±2 positions on the peptide substrate sequence, relative to the hydroxylated residue. ESI-MS assays were used to identify best cross-linking yields for the NO66-rpL8 pairs under equilibrating conditions. The following cross-linked pairs were used for crystallization: wild-type (wt) NO66 with rpL8G220C, a double NO66 variant L299C/C300S with rpL8G220C, and a single NO66 variant S373C with rpL8G214C. Structures were obtained for wt NO66·rpL8G220C (complex 1, b), L299C/C300S_NO66·rpL8G220C (complex 2, c), and S373C_NO66·rpL8G214C (complex 3, d) in combination with NOG/Mn(II) in C2 space group, 2.25-2.50 Å resolution with 2 molecules/asymmetric unit; rpL8 residues 215-223 (complex 1), 213-223 (complex 2) and 212-223 (complex 3) were observed bound to the NO66 active site. Figure e shows superimposition of the three complex structures; note that the key rpL8 residues (215-219) including the hydroxylated His216 are observed in near identical conformations (rmsd, 0.29-0.36 Å for Cα atoms); the similarity of the substrate positions in all the three structures suggests they all likely represent functional complexes. Based on the NO66·rpL8 structures, we identified a Mina53 residue, Y209C suitable for cross-linking which we crystallized in complex with rpL27aG37C (g).

To test whether the wt/mutant enzymes and altered substrates still function catalytically we carried out endpoint and time-course assays using variable enzyme/substrate ratios. The biochemical data show that for both wt NO66 and Mina53 (wt and Y209C), all the Cys-substituted peptides function as substrates (f and h, respectively). In the case of Mina53, the Y209C variant with which we obtained the Mina53·rpL27a complex structure is ~4-fold more active than wt Mina53. Data are mean and s.e.m. (n=3). We also tested wt NO66 for reaction between enzyme cysteines and the cysteines of modified substrate peptides by ESI-MS. Despite testing multiple combinations, we only observed disulfide formation in cases where we were also able to obtain crystal structures for substrate complexes. All possible combinations of wt hROX/variants and peptides containing Cys at variable positions were used for the cross-reactivity tests; NO66: wt, R297C, L299C/C300S, S373C, S421C; rpL8: wt, G214C, H218C and G220C; Mina53: wt and Y209C; rpL27a: wt and G37C. The combined activity and MS analyses suggest that in order to form stable/crystallizable cross-linked complexes, the substrates need to be recognized by the enzyme active sites in a catalytically relevant manner (a).

ED Fig6

Extended Data Figure 6 | Mutagenesis analyses of the substrate binding residues located on the JmjC catalytic domains of Mina53 (a), NO66 (b) and ycfD_RM (c). Left panels show views from the active sites of ROX-substrate complexes and the right panels show the effects of mutations on ROX catalysis. Data are mean and s.e.m. (n=3).

Analyses of ROX substrate complexes reveal important interactions between ROX and their ribosomal protein substrates. With hROX, the binding of ribosomal His39_rpL27a (light blue)/His216_rpL8 (orange) involve a series of hydrogen-bonds to backbone amides and the sidechains of Mina53/NO66 residues: Gln136_Mina53/Arg297_NO66, Asn165_Mina53/Asn326_NO66, Tyr167_Mina53/Tyr328_NO66 and Ser257_Mina53/Ser421_NO66. In addition in the Mina53·rpL27a complex, Leu38 and Arg42 of rpL27a make hydrophobic contacts with Leu176_Mina53 and a salt-bridge interaction with Asp333_Mina53 (respectively). We produced variants of all these residues to investigate their roles on substrate binding. The results of the endpoint assays as well as kinetic studies on the variants (right panels) show that substitution of these residues causes substantial losses of activity. c, In the case of ycfD_RM, the hydroxylated residue Arg82_L16 binds in a hydrophobic cleft lined by Tyr137_ycfDRM and Met120_ycfDRM sidechains and hydrogen-bonds to Asp118_ycfDRM and Ser208_ycfDRM. To test the crystallographically observed binding mode, variants of ycfD_RM residues (Asp118, Met120, Tyr137 and Ser208, highlighted) were prepared in ycfD (corresponding to Asp110, Met112, Tyr129 and Ser199, respectively). Mutagenesis studies on all ROX support the crystallographically observed binding modes of the substrate residues. The combined biochemical and structural data also provide insights into substrate selectivity of ROX over other oxygenases.

ED Fig7

Extended Data Figure 7 | Conformational changes on substrate binding in ROX. The figure shows conformational changes at the domain and residue levels in (a) Mina53 (dark salmon and red with/without rpL27a, light blue), (b) NO66 (slate and cyan with/without rpL8, orange) and (c) ycfD_RM (grey and split pea with/without L16, yellow). Although the overall movement observed for the C-terminal WH domain on substrate binding is more significant in Mina53 as compared to other ROX, the ycfD_RM structures with and without substrate display significant local changes in the sidechains of substrate binding residues (see below). a, The inset highlights local changes to the active site region in Mina53 in the presence (green sticks)/absence (yellow sticks) of substrate; Mina53 uses an acidic residue, Asp333 located on an α-helix connecting the dimerization and WH domains, to form a catalytically important salt-bridge interaction with Arg42_rpL27a. Support for this statement comes from activity analyses on variants of both rpL27a and Mina53. We have previously reported that a mutation of Arg42_rpL27a to Ala results in <5% hydroxylation⁸. The variant of D333A Mina53 ablates hydroxylation (almost completely) of native rpL27a in all tested substrate-enzyme ratios (Extended Data Fig. 6). In the substrate-unbound form, Asp333_Mina53 has two alternate conformations indicating flexibility. The NO66 substrate, rpL8 has an Ile219 at the analogous position to Arg42_rpL27a that makes hydrophobic contacts with the Tyr577 sidechains from WH domain of NO66 (b). In the case of ycfD_RM, the substrate interacting residues located on the βII-βIII loop (Tyr137), the βIV-βV insert (Arg169), the dimerization domain (Arg212 and Glu218) and on the loop connecting the dimerization and WH domains (Arg284) are observed in different conformations in the structures with and without substrate, likely reflecting induced fit on substrate binding (c). Substitutions of these residues have variable effects on ROX catalyses (Extended Data Fig. 6).

ED Fig8

Extended Data Figure 8 | Comparison of ycfDs from E. coli and R. marinus. The figure shows differences between ycfDs from E. coli (green) and R. marinus (grey). (a) Superimposition of ycfD and ycfD_RM·L16 complex structures showing crystallographically observed differences particularly in the dimerization and βIV-βV loop regions. The βIV-βV insert is highlighted in crimson red and pink in ycfD and ycfD_RM, respectively. (b) Residue numbering is according to ycfD_RM with the ycfD numbering shown in brackets. Note that all of the directly identified substrate-binding residues are strictly conserved between ycfD and ycfD_RM. However, residues particularly those located on the βIV-βV insert including Asp118, Tyr137 and Arg212 in ycfD_RM (Asp110, Tyr129 and Arg203 in ycfD) are observed in different conformations suggesting potential roles for these residues in catalysis. (c-d) Predicted binding mode of L16 (yellow) to ycfD from E. coli (green). A model complex of ycfD with Mn(II), NOG and L16 (residues Pro77-Lys84) was generated using ycfD-SeMet as the template and by comparison with ycfD_RM·L16 and Mina53·rpL27a_(32-50) structures. (d) Surface representations of the ycfD·Mn·NOG·L16_(77-84) complex, predicting key hydrogen-bonds/polar interactions (dotted lines) with L16. The hydroxylated Arg81_L16 is predicted to bind in a pocket defined by the Tyr129 and Met112 sidechains, which likely form π-cation and hydrophobic interactions with Arg81_L16 sidechain as observed in the ycfD_RM·L16 crystal structure. The Arg81 guanidino group is predicted to make electrostatic interactions with the Asp110_ycfD carboxylate and hydrogen-bonds to Ser199_ycfD. ycfD residues Asp110, Met112, Tyr129 and Ser199 were substituted to test the predicted mode of binding; the assay results are given in the Extended Data Fig. 6c.

ED Fig9

Extended Data Figure 9 | Comparison of active site chemistry of ROX and related enzymes. The figure compares active site chemistry in representative 2OG oxygenases and directionality of the peptide substrate binding through the active site. Red/blue arrows indicate hydroxylation/demethylation sites. The active site metals (Fe/Fe-surrogates, Mn or Ni) are in color-coded spheres.

ED Fig10

Extended Data Figure 10 | Phylogenetic relationships of human JmjC 2OG-oxygenases. The figure shows a parsimony tree constructed using Archaeopteryx v0.9812⁵⁸ from ClustalW⁵⁹ aligned protein sequences of human JmjC 2OG-oxygenases showing that distinct branches of JmjC-oxygenases exist for hydroxylases (red), demethylases/hydroxylases (light green) and demethylases (blue). Abbreviations used, JHDM1A, 1B, 1D, 1E and 3A: JmjC domain-containing histone demethylation protein 1A, 1B, 1D, 1E and 3A; JHD2C: JmjC domain-containing histone demethylation protein 2C; KDM2A-2B, 3A-3B, 4A-4E, 5A-5D, 6A-6B and 7 Lysine-specific demethylase 2A-2B, 3A-3B, 4A-4E, 5A-5D, 6A-6B and 7; RBBP2: Retinoblastoma binding protein 2; PTDSR: phosphatidylserine receptor; HIF: Hypoxia-inducible factor; HSPB1: 27 kDa heat shock protein; PASS1: Protein associated with small stress protein 1.