Interaction of JMJD6 with single-stranded RNA

Abstract

JMJD6 is a Jumonji C domain-containing hydroxylase. JMJD6 binds α-ketoglutarate and iron and has been characterized as either a histone arginine demethylase or U2AF65 lysyl hydroxylase. Here, we describe the structures of JMJD6 with and without α-ketoglutarate, which revealed a novel substrate binding groove and two positively charged surfaces. The structures also contain a stack of aromatic residues located near the active center. The side chain of one residue within this stack assumed different conformations in the two structures. Interestingly, JMJD6 bound efficiently to single-stranded RNA, but not to single-stranded DNA, double-stranded RNA, or double-stranded DNA. These structural features and truncation analysis of JMJD6 suggest that JMJD6 may bind and modify single-stand RNA rather than the previously reported peptide substrates.

JMJD6 was first characterized as a receptor for phosphatidylserine (PSR), which facilitates the phagocytosis of dead and dying cells by macrophages and fibroblasts (1). Targeted deletion of gene encoding PSR in mice and morpholino knock-downs of PSR in zebrafish resulted in embryonic lethality, with severe defects in hematopoiesis and aberrant development of eye, brain, and heart (2–5). In contrast, knock-down of PSR expression in Caenorhabditis elegans produced only a mild phenotype (5). Somewhat surprisingly, sequence analysis suggested that JMJD6 contains a Jumonji C (JMJC) domain, which places it within a highly conserved, cupin fold-containing enzyme family (6–8). Further analysis demonstrated that the protein is localized specifically in the nucleus (7–9). Despite the significant effects of JMJD6 deficiency, knockout mice engulfed apoptotic cells normally (9). Based on these studies and additional sequence analysis, the protein was recategorized as an α-ketoglutarate- and Fe²⁺-dependent hydroxylase and was named JMJD6 (10).

Recent studies demonstrated that most JMJC domain-containing proteins function as histone demethylases by specifically acting on lysine residues in histone tails (11–14). For example, the specific interactions between enzymes from the JMJD2 subfamily and methylated peptides have been structurally characterized (15–18). Interestingly, JMJD6 was reported to demethylate arginine residues in histone tails (10). Several laboratories including ours, however, have been unable to reproduce these results. In other studies, JMJD6 was identified as a lysine hydroxylase that specifically recognizes the protein tail of U2AF65, a mediator of RNA splicing (19).

To resolve the disparate results and further elucidate the structure and functions of JMJD6, we determined X-ray crystallographic structures of the protein with and without α-ketoglutarate. To obtain these structures, JMJD6 was cocrystallized with a Fab fragment derived from a JMJD6-specific hamster monoclonal antibody. Intriguingly, the structure of JMJD6 is dramatically different from known structures of other JMJC domain superfamily proteins including FIH (20, 21), JMJD2A (16), and AlkB (22). Our structural and biochemical analyses suggest that JMJD6 may recognize substrates including nucleic acids in addition to the known peptide tails.

Results

Overall Structure.

As described in Methods, full-length human JMJD6 was crystallized in the presence of Fab fragments obtained from a JMJD6-specific monoclonal antibody. Due to the flexibility of the C terminus of JMJD6, the Fab fragments are essential to obtain crystals of the entire JMJD6 protein. Briefly, the initial phases and structure were determined using the single wavelength anomalous dispersion (SAD) method and a mercury derivative. For refinement, data from multiple additional crystals with or without α-KG were used to obtain structures both at 2.7-Å resolution. In the final models, residues 1 to 334 of JMJD6 are well defined; however, the C-terminal, serine-rich region (residues 335 to 403) is completely disordered (Fig. 1 and Fig. S1 A and B). The structure contains a total of 15 α-helices, with α2, α3, α5, α6, α10, and α12 containing only one-turn helix. These one-turn helices distribute all over the surface of the molecule and are connected by a variety of coil regions, a unique feature for JMJD6 with unknown function (Fig. 1). With the exceptions of β3 and β4, 11 of the 13 β-strands in JMJD6 contribute to the cupin fold, a hallmark of this enzyme family (Fig. 1) (6). The structure can be divided into an N-terminal domain and C-terminal domain, which associate via β13 and α9 of the N-terminal domain and α13 of the C-terminal domain. Several hydrophobic residues are involved in these interactions, including Leu160, Phe161, and Tyr163 of the N-terminal domain and residues Trp298, Phe294, Leu308, Trp312, Leu316, and Leu323 from the C-terminal domain (Fig. S1C). Two consecutive proline residues between α9 and β6 and the hydrophobic core assembled between α9 and the C-terminal domain suggest a relatively rigid association between the N-terminal and C-terminal domains. An Fe²⁺ ion is chelated by three residues that are highly conserved among cupin domains: His187, Asp189, and His 273. The cofactor α-ketoglutarate is bound to the Fe²⁺ ion and the side chains of residues Lysine 204 and Asn277, as well as the main chain amide of Ser184. The Fab fragment was found binding at the back tail of the C-terminal domain (Fig. S1A); conformational changes caused by Fab fragments could be limited due to the rigid connection between the N-terminal domain and C-terminal domain.

Fig. 1.

The overall structure of JMJD6, a new structure of JMJC family. (A) The stereo view of the structure of JMJD6 at present of alpha-KG and Fe²⁺. The secondary structures are colored as follows: alpha-Helix, red; beta-sheet, yellow; coil, green. (B) The electrostatic surface potentials of JMJD6, positive charged surface is colored red, negative charged surface is colored blue, and neutral surface is colored gray.

Structural Comparisons Among JMJD6, FIH, JMJD2, and AlkB.

JMJD6 has been characterized as an arginine demethylase and U2AF65 tail lysine hydroxylase (10, 19). Therefore, we compared the structure of JMJD6 with representative structures from these two protein families. Overlapping features were observed between the catalytic cores of the lysine demethylase JMJD2A (16) and FIH (20, 21), a well-characterized asparaginyl hydroxylase (Fig. 2 and Fig. S2). Nevertheless, with the exception of the cupin fold, we did not detect much similarity among the three proteins. This suggests that each protein may represent a distinct subfamily. Comparisons with the structure of AlkB—a known DNA/RNA demethylase (22–24)—also revealed fairly limited structural homology (Fig. 2 and Fig. S3). However, if only the cupin folds are involved, high similarity can be found. The rmsds between the cupin fold of JMJD6 and those of JMJD2A, FIH, as well as AlkB, are 1.14 Å, 1.11 Å, and 1.20 Å, respectively (Fig. 2A).

Fig. 2.

The similarity and difference of JMJD6 among the cupin fold family members. (A) The cupin fold of JMJD6 is compared to those of Jmjd2a, FIH, and AlkB. JMJD6 is colored magenta, Jmjd2a is colored blue, FIH is colored cyan, and AlkB is colored green. (B) The comparison of JMJD6 on the catalytic core of Jmjd2a. JMJD6 is colored magenta, and Jmjd2a is colored blue.

Unique Structural Features of JMJD6.

Our analysis revealed a number of interesting structural features in JMJD6. As described above, JMJD6 contains an N-terminal cupin fold and a C-terminal helix-turn-helix-like motif (Fig. 1). The two domains are connected by an inflexible region containing two proline residues and the hydrophobic core, which likely limit the relative movement of the two domains (Fig. 3). These features suggest that substrate binding may not cause significant conformational changes in JMJD6. Furthermore, the two domains of JMJD6 create a large groove, characterized by a diameter of ∼20 Å at its narrowest and ∼30 Å at its widest (Fig. 1). Of note, the helix-turn-helix-like motif within the C-terminal domain and the β-hairpin comprising β3 and β4 from the N-terminal domain are exposed along the open groove (Fig. 1).

Fig. 3.

Unique structural features around the active center of JMJD6. (A) The active center without binding of alpha-KG. Three aromatic residues Tyr131, Phe133, and Trp174 are aligned but not stacking on each other. (B) The overlay of the active center of JMJD6 with and without alpha-KG. In the presence of alpha-KG, the side chain of PHe133 stacks with those of Tyr131 and Trp174. This may represent a mode of substrate recognition.

JMJD6 contains three aromatic residues close to the catalytic center, which are not found in other family members. Interestingly, the three side chains from Tyr131, Phe133, and Tyr174 are stacked against each other in the α-ketoglutarate- and iron (II)-bound complex (Fig. 3). The side chain of Phe assumes two conformations: the stacked arrangement and a conformation that is rotated by 90° (Fig. 3).

RNA and DNA-Binding Properties of JMJD6.

A surface potential map was built based on the distribution of charged residues (Fig. 1B). The map showed positively charged areas on the flat side of the molecule and within the groove containing the helix-turn-helix-like motif (Fig. 1B). These positively charged structural features suggest JMJD6 could bind RNA or DNA.

The full-length JMJD6 protein was used in DNA and RNA EMSAs as describe in Methods. JMJD6 strongly bound a 27-nt single-stranded RNA (ssRNA) probe with an approximate affinity of ∼40 nM (Fig. 4A), but failed to bind to the equivalent ssDNA (Fig. S4A), double-stranded RNA (dsRNA) (Fig. S4B), or dsDNA (Fig. S4C) probe. To identify requirements for the binding of JMJD6 to ssRNA, we compared its binding to progressively shorter probes (Fig. 4B). JMJD6 bound similarly to ssRNA probes of 27, 24, and 21 nt. Binding decreased dramatically to an 18-nt probe and was not detected using probes of 15 or 13 nt. To identify regions of JMJD6 that are necessary for ssRNA binding, truncated proteins were generated. Only full-length JMJD6 including the unstructured serine-rich C terminal domain (residues 334–403) bound the 27-nt ssRNA probe with high affinity (Fig. 4C and Fig. S1B). The structured fragment containing only residues 1–337 exhibited weak binding to ssRNA, whereas fragments containing residues 290–403 or 334–403 did not bind ssRNA detectably.

Fig. 4.

JMJD6 binds ssRNA. (A) Binding of full-length JMJD6 to ssRNA. Increasing concentrations of JMJD6 were incubated with a constant amount of 27-nt ssRNA probe prior to fractionation in an EMSA. (B) Progressively shorter ssRNA probes were incubated with (+) or without (−) full-length JMJD6 prior to fractionation in an EMSA. (C) Binding of truncated JMJD6 proteins to the 27-nt ssRNA probe. 1. JMJD6 (1–403), 2. JMJD6 (1–337), 3. JMJD6 (290–403) and 4. JMJD6 (334–403). All protein concentrations are about 200 nM. (D) The potential model of JMJD6 binding to ssRNA.

Discussion

Members of the α-ketoglutarate-dependent hydroxylase protein family participate in different reactions, including antibiotic biosynthesis, detection of hypoxia, and metabolite processing (25). Moreover, these enzymes include DNA, RNA, and histone tail demethylase functions, which contribute to nucleobase, nucleoside, nucleotide, and chromatin metabolism (25). Each of these subfamilies is characterized by unique structural features that allow the enzymes to recognize and process their cognate substrates.

Sequence similarities among the members of the JMJC domain-containing protein family and the overall structure of JMJD6 were used to identify the catalytic core domain of the JMJD2 histone demethylase subfamily (15, 26); the structure of JMJD6 was determined before these activities of JMJD2 were identified in the laboratory. Although proteins from this family are thought to mostly function as histone demethylases, so far we have been unable to demethylate various histone tails using JMJD6 in vitro. Also, we have not detected DNA demethylase or deaminase activities. It has been reported that JMJD6 hydroxylates lysine residues specifically in the tail of U2AF65, suggesting a potential role of JMJD6 in the regulation of mRNA splicing (19). Hydroxylase activities were also identified during the characterization of JMJD6 as an arginine demethylase (10), suggesting that JMJD6 hydroxylates protein tails nonspecifically. In this study, we have shown that JMJD6 contains a large groove around the catalytic center, which should be accessible to elongated peptides. This may facilitate the access of lysine side chains of any flexible protein tails to the catalytic center.

A question remains as to whether JMJD6 acts on other substrates in addition to proteins, or what is the real cognate substrate? Several lines of evidence led us to hypothesize that ssRNA may serve as substrates. The groove around the JMJD6 catalytic center would accommodate RNA or DNA. The C-terminal groove region includes a helix-turn-helix-like motif, which is typical for DNA or RNA binding proteins (Fig. 3). The antiparallel β3- and β4-sheet, which forms a β-hairpin, is also located close to the groove (Fig. 1). β-hairpins, including acidic residues at the apex, often make contacts with nucleotide bases. Meanwhile, the positively charged surface within the groove is consistent with the property required for a RNA- or DNA-binding domain. Moreover, a striking feature is the stack of aromatic side chains (Fig. 3), a feature similar to that found in several cap-binding proteins such as the -cap dimethyltransferase TGS1, the eukaryotic initiation factor 4E, and the cap-binding complex, etc. (27). The flexibility of the side chain of the middle Phe within the stack may allow for interactions with bases, which may be important for binding to nucleic acid substrates during the enzymatic reaction. Furthermore, direct experimental data demonstrated the binding of JMJD6 to ssRNA. Finally, further domain characterization demonstrated that the N-terminal domain, C-terminal domain, and unstructured serine-rich region in the C-terminal of the protein were required for this binding (Fig. 4C). None of these regions alone produced detectable binding to ssRNA; however, the structured core containing residues from 1 to 337 (including the N-terminal domain and C-terminal domain) binds to ssRNA weakly (Fig. 4C). This result suggests that the groove region within the structured core is essential for the ssRNA binding; the unstructured C-terminal serine-rich motif is also required to enhance the interaction. Here, we hypothesize that ssRNAs are cognate substrates of JMJD6. As a potential mechanism of action, bases within the ssRNA substrates could insert into the stack for enzymatic modifications similar to the interation of mRNA with cap-binding proteins, although the enzymatic consequences are likely to be different. Based on our data, a model of JMJD6 binding to ssRNA can be built (Fig. 4D). It should be noted that the ssRNA sequence used in this study was selected randomly, suggesting a lack of base sequence specificity in binding to JMJD6. However, how promiscuous this specificity is will require additional studies.

A key issue that still remains to be resolved is the relationship between the nucleic acid-binding and catalytic activities of JMJD6. All these features, the severe phenotype of JMJD6 knockout, the special structural features, and the ssRNA binding activity, differentiate JMJD6 from other members of the JMJC protein family, which work mainly as histone demethylases and protein hydroxylases. The nature of the catalytic activity, however, is still unclear. Based on the structural information, biochemistry data generated so far, and the enzymatic activity nature for JMJC proteins, methyl groups on nucleic acids are likely candidates for the enzymatic target moiety. More specifically, JMJD6 may function as a component of the splicesome (including U2AF65) at the branch site to affect the alternative splicing of pre-mRNA. Future studies will address the identities of these chemical targets and their regulatory roles in cells.

Methods

Protein Expression, Purification, and Crystallization.

The cDNA clone encoding wild-type human JMJD6 was described previously (1). DNA fragments encoding wild-type human JMJD6 (amino acids 1–403) or the mutant variants were amplified in PCRs and ligated into pET23b (Novagen) with a C-terminal 6X His tag or without a tag. The final clones were verified by restriction enzyme digestion and DNA sequencing. Rosetta (DE3) cells (Novagen) were transformed with the recombinant plasmid containing the JMJD6 gene. JMJD6 expression was induced by adding IPTG to an 8-L growing culture (37 °C) at an OD₆₀₀ of 0.8. After 4 h of additional growth, cells were harvested and resuspended in buffer A [20 mM Tris-HCl (pH 8.0), 500 mM NaCl, and 5 mM imidazole] supplemented with protease inhibitors (Invitrogen). After cell lysis using a continuous-flow French press and a low-speed spin, the soluble fraction was loaded onto His affinity beads (Novagen). After intensive washing with buffer A, JMJD6 was eluted from the beads with buffer A containing 1 M imidazole. The JMJD6 sample was loaded onto a MonoS column (Pharmacia). After elution with a NaCl gradient, the homogeneity of the protein sample was evaluated using Coomassie blue-stained SDS-polyacrylamide gels. JMJD6 was further purified using a Superdex 200 column (Amersham). All other truncated versions of JMJD6 were produced similarly as the entire version of JMJD6. Purified anti-JMJD6 hamster monoclonal antibody (http://www.scbt.com/datasheet-32740-psr-apsr-14-4-antibody.html) was digested into Fab fragments using papain (p4406, Sigma) at 37 °C for 4 h in buffer B [20 mM phosphate buffer (pH 7.0), 10 mM EDTA, and 25 mM cysteine]. Fab fragments were loaded onto a MonoS column. Purified JMJD6 and the Fab fragments were mixed at 4 °C for 2 h and loaded on a Superdex 200 column. Purified JMJD6-Fab complexes (6 mg/mL) were crystallized at 4 °C using vapor diffusion against 2.1 M ( and 100 mM Bis-Tris (pH 5.6). For data collection, crystals were gradually transferred into cryobuffer (reservoir buffer supplemented with 20% glycerol) and flash cooled in liquid N₂.

Structure Determination and Refinement.

Native crystals of the JMJD6–Fab complexes diffracted poorly (∼3.2 Å) when examined in a synchrotron. The crystal diffraction quality dramatically improved to 2.6 Å after the crystals were incubated for one week in crystallization solution saturated with methylmercury chloride, although the crystals remained sensitive to the X-ray beam. Because the crystal was highly symmetric, a complete dataset could be collected from a single crystal. Data were processed using HKL2000 (28). The initial phases were derived from a native dataset and anomalous derivative data using SOLVES (29) (Table S1). The phases were extended from 4.5 Å to 3.2 Å using SOLOMON (30). From the initial calculated map, β-strands and α-helixes were identified. The Fab fragment was quickly built into the model. The map of JMJD6 improved after adding the Fab fragments via the Sigma program (30). The final model, which contained residues 1 to 334 of JMJD6, was further refined using higher resolution data combined from data of two crystals by the Crystallography and NMR System program (CNS) (31) (Table S1). Cocrystals of α-ketoglutarate and JMJD6 were grown at a similar condition as the native form. The complex structure of α-ketoglutarate and JMJD6 was determined using a Fourier transformation and refined using a dataset from three crystals by CNS (31) (Table S1). All structural models were built and adjusted in the O program (32). All structural figures were made using the PyMOL program (http://pymol.sourceforge.net). Crystallographic data are briefly summarized in Table S1.

EMSA for ssRNA and Others.

All RNA probes were purchased from Integrated DNA Technologies. Single-stranded RNA (5′-rAUACGAUGCUUUACGGUGCUAUUUUGU-3′; 27 nt) was 5′ end labeled using T4 polynucleotide kinase (New England Biolabs) and γ³²P-ATP (6,000 Ci/mmol; PerkinElmer Life and Analytical Sciences) according to the manufacturers’ instructions. Following extraction with phenol:chloroform, unincorporated nucleotides were removed using illustra™ MicroSpin™ G-25 columns. Progressively shorter RNA probes included 5′-AUACGAUGCUUUACGGUGCUAUUU-3′ (24 nt), 5′-AUACGAUGCUUUACGGUGCUA-3′ (21 nt), 5′-AUACGAUGCUUUACGGUG-3′ (18 nt), 5′-AUACGAUGCUUUACG-3′ (15 nt), and 5′-GCUUUACGUGCU-3′ (13 nt). For EMSA, 15,000 cpm of each probe were incubated with increasing amounts of recombinant JMJD6 protein in 20 μL of 10 mM Hepes pH7.6, 100 mM NaCl, 1 mM MgCl₂, 1 mM dithiothreitol, 5% glycerol, 100 μg/mL BSA, and 0.5 μL RNasin® (Promega Corporation). Binding reactions were incubated at 4 °C for 30 min; 8 μL of each reaction were fractioned by electrophoresis on a 5% polyacrylamide/1X Tris/glycine/EDTA gel (4 °C, 240 V for 45 min) and autoradiographed as described (33). Data were quantitated using a Typhoon 9200 PhosphorImager system (Molecular Dynamics/Amersham/GE Healthcare). DsRNA and dsDNA binding assays follow the above procedure.