Structural insights into protein arginine symmetric dimethylation by PRMT5

Abstract

Symmetric and asymmetric dimethylation of arginine are isomeric protein posttranslational modifications with distinct biological effects, evidenced by the methylation of arginine 3 of histone H4 (H4R3): symmetric dimethylation of H4R3 leads to repression of gene expression, while asymmetric dimethylation of H4R3 is associated with gene activation. The enzymes catalyzing these modifications share identifiable sequence similarities, but the relationship between their catalytic mechanisms is unknown. Here we analyzed the structure of a prototypic symmetric arginine dimethylase, PRMT5, and discovered that a conserved phenylalanine in the active site is critical for specifying symmetric addition of methyl groups. Changing it to a methionine significantly elevates the overall methylase activity, but also converts PRMT5 to an enzyme that catalyzes both symmetric and asymmetric dimethylation of arginine. Our results demonstrate a common catalytic mechanism intrinsic to both symmetric and asymmetric arginine dimethylases, and show that steric constrains in the active sites play an essential role in determining the product specificity of arginine methylases. This discovery also implies a potentially regulatable outcome of arginine dimethylation that may provide versatile control of eukaryotic gene expression.

Protein arginine methyltransferase 5 (PRMT5) catalyzes the evenly addition of two methyl groups to the two ω-guanidino nitrogen atoms of arginine, resulting in ω-N^G, N^′G symmetric dimethylation of arginine (sDMA) of the target protein (1–5). PRMT5 functions in the nucleus as well as in the cytoplasm, and its substrates include histones, spliceosomal proteins, transcription factors, and proteins involved in piRNA biogenesis (6). Symmetric dimethylation of these proteins profoundly impact many biological processes; e.g., epigenetic control of gene expression (7), splicing regulation (2, 3, 8, 9), circadian rhythms (9, 10), DNA damage response (11, 12), and germ cell development and pluripotency (13–16). Interestingly, both PRMT5 and a group of asymmetric (type-I) arginine dimethylases, which add two methyl groups to the same ω-guanidino nitrogen atom (aDMA), share common recognition sequences, and the target arginine can often be symmetrically or asymmetrically dimethylated. Yet, these isomeric modifications have distinct biological effects. One such example occurs at arginine-3 of histone H4 (H4R3). Symmetric dimethylation of H4R3 has been linked to repression of gene expression (17–19), while asymmetric dimethylation of H4R3 is associated with gene activation (20, 21). The startling difference in biological effects of sDMA and aDMA modifications necessitates the understanding of the enzymatic mechanisms differentiating the two chemically isomeric but functionally antagonistic posttranslational modifications.

Results

Overall Structure.

We have determined the crystal structures of full-length PRMT5 from Caenorhabditis elegans, alone and in complex with S-Adenosyl-L-homocysteine (SAH). The nematode enzyme shares high sequence homology with its human counterpart, and the recombinant protein displays robust and specific symmetric arginine dimethylase activity in vitro (Fig. 1). The structure shows the PRMT5 is composed of four clearly defined domains, a previously unsuspected TIM-barrel at the N-terminal end, a middle Rossmann-fold domain, a C-terminal β-barrel domain, and a ∼60 residue dimerization domain inserted between β1 and β2 of the β-barrel domain (Fig. 2A). The first three domains are packed in a triangular manner, with direct contacts between sequential domains, and the oligomerization domain bridges the TIM-barrel and β-barrel domains. The structure of the SAH-bound PRMT5 differs from that of free protein in that a N-terminal loop (L0) and helix (αA) are ordered in the SAH-bound structure. We will use the SAH-bound structure for analysis unless explicitly noted.

Fig. 1.

Structural and functional conservation of PRMT5. (A) A schematic representation of domain structures of C. elegans and human PRMT5, and a representative type-I arginine methylase, PRMT1 of rat. The lengths of the boxes are approximately drawn in scale with the protein lengths, and the residue numbers at domain boundaries are labeled. Areas filled in tan, cyan, green, and yellow represent TIM-barrel, Rossmann-fold, β-barrel, and oligomerization domains, respectively. Levels of amino acid identity and similarity of the individual domains between C. elegans and human PRMT5s, and that between human PRMT5 and rat PRMT1 are shown. (B) Sequence alignment. The full-length sequences of C. elegans and human PRMT5s, and the regions of the solved structures of rat PRMT1 and mouse CARM1 are aligned. Residues conserved in all four proteins are shown in white letters over purple background, and similar residues are indicated with red letters. Residues conserved in PRMT5 proteins and type-I arginine methylases are highlighted tan and yellow, respectively. Blue stars mark the CePRMT5 residues subjected to mutagenesis. At the top of the sequences, a schematic representation of the secondary structure elements of CePRMT5 is shown. Every ten residues are indicated with a “·” sign. (C) Enzymatic activity assay. Top box, coomassie-stained gel of enzymes and substrate (histone H4) used. GST-tagged rat PRMT1 and poly(His)-tagged C. elegans PRMT5 were expressed in E. coli, and flag-tagged human PRMT5 was purified from HEK293 cells. Approximately 5 μg of enzymes and histone H4 each were used in the assay. Top 2nd box, autoradiograph generated with the use of 0.25 mCi of SAM with tritiated methyl group. Top 3rd box, Western blot detection of asymmetrically dimethylated histone H4R3. Bottom box, Western blot detection of symmetrically dimethylated histone H4R3.

Fig. 2.

Overall structure of PRMT5. (A) A ribbon representation of PRMT5 monomer. Domains are colored as in Fig. 1A. Helices and strands are labeled TA to TH and T1 to T8, respectively, for the TIM-barrel domain; αA to αF and β1 to β5, respectively, for the Rossmann-fold domain; and OA-OB and O1-O2, respectively, for the oligomerization domain. Strands in the β-barrel domain are labeled b1 to b10. L0 indicates the N-terminal loop of the Rossmann-fold domain, and the SAH molecule is shown in a stick model. (B) A PRMT5 dimer is shown as a ribbon model superimposed onto a surface representation. The surface for one monomer is colored light green and the other in light blue. The red line approximately traces the dimeric interface.

PRMT5 exists as a homodimer, shown both in the crystal structure and in solution, as determined by analytic ultracentrifugation (Fig. 2B, Fig. S1). The dimeric interface buries a total pair wise surface area of 2,305 Å², and the intermolecular interactions occur between the dimerization domains and that between the TIM-barrel and β-barrel domains. Human PRMT5 has a shorter oligomerization domain, but extensive conservation of amino acids involved in intermolecular interactions implies that it also forms a dimer, consistent with the report of dimeric and higher oligomeric forms of human PRMT5 (5). In fact, all arginine methylases with known structures dimerize via a homologous dimerization domain, also known as the dimerization “arm” (Fig. 3A), (22–26). Thus, protein dimerization appears to be an evolutionarily conserved property of arginine methylases, although the functional significance remains poorly understood.

Fig. 3.

Structural features of the PRMT5 active site. (A) Structural comparison with type-I arginine methylases PRMT1 (Pdb id: 1OR8; magenta) and CARM1 (Pdb id: 2V74; light blue). For visual clarity, only three regions of major differences, enclosed in red circles and labeled I, II and III, are superimposed onto the structure of PRMT5. (B) An up-close view of the active site. Key residues of PRMT5 (carbon: yellow; oxygen: red; nitrogen: blue) and the SAH molecule (carbon: orange; sulfur: gold) are shown in a stick model superimposed with a ribbon representation of PRMT5. The double-E loop of PRMT1 (while ribbon), Glu153 on it, and the bound substrate arginine (stick model; carbon: white) are also shown. (C) Methylase activities of Phe379 mutants. Top box: coomassie-stained gel of enzymes and substrate used. Bottom box: autoradiography detection. (D) Circular dichroism spectra of the wild-type and Phe379 mutants of PRMT5.

Comparison with Arginine Asymmetric Dimethylases.

The overall fold and spatial positioning of the Rossmann-fold and β-barrel domains are similar to that of type-I enzymes, represented by PRMT1 and CARM1 (Fig. 3A) (24–26). The root-mean-squared deviation of Cα positions between PRMT5 and PRMT1 is approximately 2.1 Å when both the Rossmann-fold and β-barrel domains are compared, whereas it is 1.4 Å for the Rossmann-fold domain alone. In particular, a segment including a N-terminal loop (L0) and a following helix (αA) (a.a. 359–380) became ordered upon SAH binding. Helix αA is positioned similar to that found in type-I enzymes, sheltering SAH from exposing to the solvent and creating a secluded catalytic active site. However, key residues responsible for the disordered-to-ordered conformational transition upon SAH/SAM binding are separately conserved among PRMT5 family members (Fig. 1B). Among which, Tyr376 and Phe379 on αA interact with the ribose and homocysteine moieties via hydrogen bonds and van der Waals contacts, respectively. Loop L0 contains several amino acids uniquely conserved in PRMT5s across species. The corresponding region in PRMT1 is disordered, and that of CARM1 adopts a helical conformation. This PRMT5 loop has two apparent functions, (i) contact the SAH/SAM molecule via residues conserved in PRMT5 proteins (Pro366, Leu367, and Leu371) and forms a solvent inaccessible area for catalysis; (ii) the N-terminal end of L0 makes a U-turn and contacts the dimerization domain, which stabilizes the loop in a conformation endowed with the ability to influence substrate binding (Figs. 3 A and B). Hence, the highly conserved PRMT5 loop is important for SAH/SAM binding, as well as in a position to regulate substrate binding.

The Active Site.

The active site of PRMT5 is identified by the location of the sulfur atom of SAH and a pair of invariant glutamate residues, Glu499 and Glu508, found in all protein arginine methylases (Fig. 3B). These two residues are located on a hairpin loop connecting β4 and αF, also known as the “double-E” loop (24), and are absolutely required for enzymatic activities (Fig. S2). An examination of the active site reveals that there are four residues separately conserved among PRMT5 proteins. These four residues are Phe379, Lys385, Ser503, and Ser669 (Fig. 1B). The corresponding residues in type-I arginine methylases are Met, Arg, Tyr, and His, respectively, and they are also conserved among type-I enzymes. We reasoned that some of these differences might be important for PRMT5’s catalytic activity, and carried out analyses of these residues by mutagenesis. Interestingly, changing Phe379 to a methionine (F379M) resulted in a more active enzyme, while an F279Y mutant is inactive, and mutations to an alanine or a glycine pronouncedly reduced the enzymatic activity (Fig. 3C). Circular dichroism spectra show that the observed differences in enzymatic activities are not due to gross structural alterations caused by mutations (Fig. 3D).

Mutation of other conserved residues near the active site, such as changing Ser503 to a tyrosine (S503Y) or a double substitution of Val668 and Ser669 to a threonine and a histidine (V668T/S669H), greatly diminished the enzymatic activity (Fig. S2). These residues are unlikely to be directly involved in catalysis, as suggested by their distances from the SAH molecule. Val668 and Ser669 are situated on a β-barrel domain loop between strands b5 and b6 that interacts with critical elements of the Rossmann-fold domain: the αA-αB junction and the double-E loop. The conformation of this loop is considerably different from those in type-I arginine methylases (Fig. 3A, region II). Ser503 is located on the tip of the double-E loop and makes hydrogen bonds with the carbonyl and amide groups of Phe671 of this β-barrel domain loop. Thus, the mutagenesis data of Ser503 and Val88/Ser669 indicate that interdomain contacts mediated by the b5–b6 loop is critical for precise positioning of key residues for catalysis.

Determinants for Arginine Symmetric Dimethylation.

An in-depth evaluation of the enzymatic property of the F379M mutant shows that it is more active than the wild-type enzyme over a broad range of enzyme concentrations (Fig. 4A). Enzyme kinetic measurements indicate that a drop of the Km value is largely responsible for the elevated enzymatic activity of the mutant enzyme (Fig. 4C). Furthermore, we probed whether the F379M mutant exclusively carries out symmetric dimethylation of arginine. Surprisingly, both symmetric and asymmetric dimethylation of H4R3 were detected (Fig. 4B). This phenomenon appears to be specific to the change to a methionine, as no such activity was detected for the F379A and F379G mutants (Fig. S3). These observation imply that: (i) symmetric and asymmetric dimethylation of arginine shares a common catalytic mechanism, as the same active site is involved; (ii) Phe379 occupies a key position for PRMT5’s sDMA product specificity. The above observation should hold for all type-II PRMTs, as the phenylalanine is absolutely conserved among them. The corresponding mutation (F327M) in human PRMT5 also resulted in the gaining of asymmetric arginine dimethylase activity (Fig. 4D).

Fig. 4.

Enzymatic properties of the F379M mutant. (A) Top box: coomassie-stained gel showing varying amounts of the wide-type and the F379M mutant of PRMT5, and a constant amount of histone H4 (5.0 μg) used for enzymatic assay. “C” indicates no enzyme added. Bottom box: autoradiography detection with 0.25 mCi of [³H]-SAM used in each reaction. (B) Western blot detection of asymmetric (middle box) and symmetric (bottom box) dimethylation of histone H4R3. Top box is a coomassie-blue stained gel showing proteins used in the activity assays. Please note, for a comparable level of Werstern blot signal, the amount of the F379M protein is adjusted to ∼1/10 of the wild-type protein, and the amount of PRMT1 is even smaller. (C) Double reciprocal plot analysis of the wild-type and F379 mutant of PRMT5. Derived kinetic parameters are tabulated. (D) Conserved property of the Phe-to-Met mutants of human and nematode PRMT5s. Top two boxes: coomassie staining of the wild-type and mutant enzymes and the substrate used. The right pointing arrow indicates the position of human PRMT5 proteins; an asterisk indicates the position of rat PRMT1; and the left pointing arrow marks the position of C. elegans PRMT5. Third box: Western blot detection of the flag-tagged wild-type and mutant human PRMT5. Fourth box: Western blot detection of asymmetrically dimethylated H4R3. Bottom box: Western blot detection of symmetrically dimethylated H4R3.

Discussion

A surprising finding of this study is the role of Phe379 in specifying sDMA specificity of PRMT5. Phe379’s conformation is tightly fixed in the structure, and its closest distance to the sulfur atom of SAH is ∼4 Å. The structure suggests that Phe379 will be juxtaposed with the methyl group of SAM and the substrate arginine, thus, serving as a steric factor restricting the reaction product to sDMA. A methionine in this position has more conformational flexibility, as seen from the structures of PRMT1 and CARM1, which may allow the production of both sDMA and aDMA. We also mutated the corresponding methionine in PRMT1 to a phenylalanine to see if the reverse is true, but the result is inconclusive due to a much lowered level of the mutants’s overall enzymatic activity. It should be noted that changing Phe379 of PRMT5 to a methionine appears to relax PRMT5's sDMA constraint rather than a complete switch to aDMA. It is conceivable that other factors are needed in conjunction with Phe379 to achieve a complete switch. For example, Lys385 and Tyr386 are in the vicinity of Phe379 and SAH, and they are exclusively conserved in PRMT5 proteins (Fig, 3B). The two amino acids make hydrogen bonds with the carboxylate group of SAH and the catalytic Glu499. In type-I enzymes, a conserved arginine serves the functions of the pair of residues in PRMT5. Further structure and function studies are needed to fully delineate requirements for symmetric vs. asymmetric arginine dimethylation. Another interesting point is that Phe379 is located on αA, which undergoes large conformational changes upon SAH binding. It is conceivable that the conformation of αA may be regulated, such as by protein-protein interactions, and the conformational dynamics of αA is likely to impact the enzyme activity and product specificity of the arginine dimethylases.

This first structure of a symmetric arginine dimethylase also reveals that the N-terminal domain of PRMT5 has a TIM-barrel fold, and this domain is important for homodimerization. As noted previously, human PRMT5 lacking the N-terminal region encompassing the TIM-barrel domain is still capable of forming a dimer at high protein concentrations, but the catalytic activity is severely compromised (27). This observation suggests that the TIM-barrel domain has other essential functions in addition to being important for PRMT5 dimerization. Human PRMT5 was purified as a large protein complex, known as the methylosome, that contains pICLn and MEP50 (2, 3, 28). pICLn interacts with the PRMT5 region now known to have a TIM-barrel structure, and it stimulates PRTM5’s activity towards Sm proteins (27). Another protein, RioK1 competes with pICLn for binding to PRMT5 and directs PRMT5’s activity towards the RNA-binding protein nucleolin (29). Thus, the TIM-barrel domain of PRMT5 also serves as a scaffold for the binding of adaptor proteins, such as pICLn and RioK1. Hence, this domain is important for the assembly of PRMT5 complexes and their substrate selectivities.

In summary, the structural and biochemical properties of the PRMT5 uncovered here will serve as a guide for understanding the biochemical mechanisms of all type-II arginine methylases, and advance the understanding of symmetric arginine dimethylation in a wide-spectrum of biological processes.

Materials and Methods

Expression, Purification, and Crystallization of PRMT5.

A bacterial expression plasmid of C. elegans PRMT5 was constructed by cloning the full-length cDNA into a pET21a vector (Novagen). The poly(His)-tagged recombinant protein was produced in the BL21(DE3) strain of Escherichia coli. Protein expression was induced with 0.4 mM IPTG when the cell density reached OD₆₀₀ = 0.8, after which point the temperature was shifted to 16 °C for 30 h. The poly(His)-tagged PRMT5 was first purified using Ni-IDA resins, followed by anion exchange and gel-filtration column chromatography. High purity fractions were pooled and concentrated to ∼8–10 mg/mL for crystallization. PRMT5 crystals were grown by vapor diffusion in hanging drops at 16 °C in a condition with 100 mM Tris (pH 7.0), 9% PEG-5000MME and 5% Tacsimate.

Selenyl-methionine (SeMet) substituted PRMT5 was expressed using a defined medium supplemented with 25 mg/L of SeMet. SeMet PRMT5 was purified and crystallized in a manner identical to the wild-type protein.

Single or multiple point mutations of PRMT5 were generated by PCR and verified by DNA sequencing. Mutant proteins were expressed and purified following the same protocol for the wild-type protein.

Data Collection and Structure Solution.

X-ray diffraction data were collected at liquid nitrogen cryogenic temperature using a cryoprotectant with the well solution supplemented with 30% glycerol. A 3.0 Å single wave length anomalous dispersion (SAD) dataset was collected at a wave length of 0.9792 Å at beamline BL17U of Shanghai Synchrotron Radiation Facility (SSRF) using a Mar CCD-225 detector (Mar Research). Diffraction data were processed with the HKL2000 software (30). The crystal belongs to the P2₁2₁2₁ space group, and there are two PRMT5 molecules per asymmetric unit. Se sites were found by SHELXD (31) and PHENIX (32) was used for phasing and generation of an initial model. Iterative cycles of model building and refinement were carried out using COOT (33) and PHENIX. The data used for refinement were subjected to anisotropic scaling performed using the UCLA Anisotropy Server. Noncrystallography symmetry restrains, secondary structure restrains, and TLS refinement (four TLS groups: A46-330, A359-734, B46-330, B359-734) were applied to improve the electron density map. The coordinates for the refined models have been deposited in PDB under the accession codes 3UA3 and 3UA4. Detailed statistics of data collection and refinement are shown in Table S1.

Analytical Ultracentrifugation.

Protein sample prepared after gel-filtration was diluted to OD₂₈₀ = 0.8 in a buffer of 20 mM Tris-HCl (pH 7.4) and 150 mM NaCl. Sedimentation velocity experiment was carried out with the ProteomeLab XL-I (Beckman Coulter). Experiment was performed at 4,800 rpm at 20 °C for 5 h. Velocity data were collected in a continuous scan mode at 280 nm, and sedimentation coefficients were calculated with the program Sedfit (34).

Circular Dichroism (CD) Measurements.

CD spectra were recorded using an Applied Photophysics Pi-Star 180 spectropolarimeter at 20 °C, with the protein samples at 0.1 mg/mL in PBS buffer. Measurements were made under nitrogen with a 0.1 cm path length. Data were collected at 1 nm wave length interval from 200 to 260 nm. Mean residue molar ellipticity was calculated using the formula [θ] = θ × 10³/(c × l × N), where c is the mean protein concentration in millimolar, θ is the ellipticity in millidegrees (mdeg), l is the path length in millimeters, and N is the number of amino acids.

In Vitro Methylation Assay.

Wild-type and various mutants of His-tagged C. elegans PRMT5, and GST-tagged rat PRMT1 were expressed in E. coli and purified using a standard protocol. Flag-tagged human PRMT5 (hPRMT5) or its F327M mutant was prepared from HEK293 cells transfected with a pCMV2b vector carrying the wild-type or mutant cDNA, and the recombinant enzyme was purified by resins conjugated with anti-FLAG antibody. FLAG-pull down of hPRMT5 samples from cells grown to ∼70% confluency in a 60 mm-diemeter dish and transfected with 2 μg of plasmid DNA were divided into two equal aliquots for enzymatic reactions. Approximately 5 μg of E. coli expressed proteins, or an aliquot of hPRMT5 was individually incubated with 5 μg of histone H4, and 0.25 mCi of adenosyl-L-[methyl-³H] methionine (Amersham Biosciences, Inc.) at 30 °C for 2 h in a final volume of 20 μL. After the reaction, the mixtures were boiled in the SDS sample buffer and separated by SDS-PAGE. The gels were stained by Coomassie Blue and destained, then treated with Amplify (Amersham Biosciences, Inc.), dried, and exposed to a film.

Enzyme Kinetics Measurements.

Enzyme kinetics parameters were determined using the Methyltransferase Colorimetric Assay kit (Cayman, 700140). His-tagged PRMT5 and PRMT5-F379M proteins (0.5 μM), in the presence of saturating SAM (≥100 μM), and varying concentrations of Histone H4 were incubated for 45 min at 37 °C in a 200 μL reaction mixture. During incubation, color formation was assessed at 515 nm with Thermo Scientific Varioskan Flash (Thermo). Initial velocity data, measured as a function of substrate concentration, were analyzed using the Michaelis-Menten equation, V = V_max[S]/([S] + K_m); and K_cat = V_max/[E], where [E] is the total enzyme concentration. All measurements were done in triplicate.

Western Blot Analysis.

Western blot analysis was performed with antibodies against histone H4 with symmetric dimethyl Arg 3 (ab5823/97454, Abcam), H4 asymmetric dimethyl Arg 3 (39705, Active Motif), and an anti-FLAG antibody (F1804, Sigma).