Transition state mimics are valuable mechanistic probes for structural studies with the arginine methyltransferase CARM1

Significance

The posttranslational methylation of arginine is a widespread epigenetic modification catalyzed by the family of protein arginine methyltransferases (PRMTs). Dysregulation of PRMT expression is implicated in the pathogenesis of many diseases including human cancers. An atomic-scale understanding of the PRMT catalytic mechanism is crucial for both fundamental biological and pharmacological applications. Despite intense efforts, crystal structures of PRMT complexes with long peptides and full-length substrates have not been solved because of their inherent instability. To address this issue, we describe peptide-based transition state mimics that form stable complexes with the PRMT enzyme coactivator associated arginine methyltransferase 1 resulting in high-resolution cocrystal structures. Our findings provide an exciting approach to understanding PRMT substrate recognition and the regulation of arginine methylation.

Abstract

Coactivator associated arginine methyltransferase 1 (CARM1) is a member of the protein arginine methyltransferase (PRMT) family and methylates a range of proteins in eukaryotic cells. Overexpression of CARM1 is implicated in a number of cancers, and it is therefore seen as a potential therapeutic target. Peptide sequences derived from the well-defined CARM1 substrate poly(A)-binding protein 1 (PABP1) were covalently linked to an adenosine moiety as in the AdoMet cofactor to generate transition state mimics. These constructs were found to be potent CARM1 inhibitors and also formed stable complexes with the enzyme. High-resolution crystal structures of CARM1 in complex with these compounds confirm a mode of binding that is indeed reflective of the transition state at the CARM1 active site. Given the transient nature of PRMT–substrate complexes, such transition state mimics represent valuable chemical tools for structural studies aimed at deciphering the regulation of arginine methylation mediated by the family of arginine methyltransferases.

The methylation of arginine residues by protein arginine N-methyltransferases (PRMTs) plays a vital role in a variety of key cellular functions including gene regulation, signal transduction, RNA processing, and DNA repair (1, 2). Conversely, dysregulated PRMT activity is known to be linked to a variety of cancers and other diseases (3, 4). Coactivator associated arginine methyltransferase 1 (CARM1, also known as PRMT4) is directly involved in numerous cellular processes via methylation of the histone H3 tail peptide, splicing factors, RNA binding proteins, and coactivation of nuclear receptors (1, 5). Growing evidence indicates associations with the up-regulation of CARM1 in breast (6, 7), colon (8, 9), prostate (9, 10), and liver (11) cancers, making it an appealing potential therapeutic target.

Like all lysine and arginine methyltransferases, CARM1 uses the cofactor S-adenosine-l-methionine (AdoMet) as the source of the methyl group that is transferred to the guanidine group of the arginine side chain in a target protein or peptide with concomitant generation of S-adenosine-l-homocysteine (AdoHcy) (1). As for all type 1 PRMTs, the first methylation step catalyzed by CARM1 gives monomethyl-arginine (MMA), followed by a second methylation to form asymmetric dimethyl-arginine (aDMA). Although the different PRMTs vary significantly in overall size and sequence, they share a common active site architecture defined by specific amino acids (belonging to the so-called motifs I to IV) known to be key for catalysis (Fig. 1A) (1–4, 12). Most notable in this regard are residues involved in hydrogen bonding to the adenosine moiety of AdoMet and two conserved glutamate residues that form the so-called “double-E loop” critical for chelating and orienting the guanidine group of the target arginine residue. In doing so, the target guanidine moiety is precisely presented to the electrophilic methylsulfonium group of AdoMet, facilitating an “S_N2-like” substitution reaction.

Fig. 1.

The CARM1 active site and design of transition state mimics. (A) CARM1 employs an “S_N2-like” bisubstrate mechanism. Key active site residues are indicated including the conserved Glu₂₆₇ and Glu₂₅₈ residues comprising the guanidine-binding “double-E loop” (Mus musculus CARM1 numbering scheme). The AdoMet cofactor is highlighted in red and the target arginine-containing peptide in blue. (B) Structures of PABP1-derived constructs prepared in this study as CARM1 transition state mimics.

Recently, we reported a series of small molecule PRMT inhibitors wherein a guanidine moiety was covalently linked to an adenosine group meant to mimic that of AdoMet (13). Interestingly, some of these inhibitors displayed potent and specific inhibition of CARM1 with IC₅₀ values in the nanomolar range (13). As a means of enhancing CARM1 selectivity, we report here an approach wherein known CARM1 substrate peptides are covalently tethered to an adenosine unit to create transition state mimics of the CARM1 methylation reaction (Fig. 1B). Such transition state mimics were also envisioned to serve as mechanistic probes and chemical tools for structural studies. Previous reports have highlighted the challenges associated with cocrystallizing PRMTs with their various peptide substrates (14–18). Because of these challenges, a complete structural understanding of how PRMTs achieve specificity in arginine methylation at different target sites remains elusive. Furthermore, although structures of isolated PRMTs have been published, few studies exist that provide an atomic-level characterization of the functional macromolecular complexes formed between PRMTs and their targets. Importantly, the transition-state mimicking compounds we here describe were found to cocrystallize with CARM1, forming stable complexes suitable for structural studies.

Results

Design of Transition State Mimics and Inhibition Studies.

In selecting the CARM1-specific peptide substrates to be used in designing the transition state mimics, we focused on the nonhistonic poly(A)-binding protein 1 (PABP1), one of the most highly methylated CARM1 targets. Previous studies using peptide fragments derived from PABP1 identified the strongest CARM1 methylation sites as R455 and R460 (19). We therefore selected two PABP1 fragments, comprised of residues 447–459 and 456–466 (each containing a single arginine) to serve as the peptide segment of the transition state mimics (Fig. 1B). Building on methodology recently developed in our group (20–22), we here report a convenient on-resin approach for preparing peptides in which an adenosine group can be covalently linked to the guanidine moiety of a specific arginine residue (compounds 1–6; Fig. 1B, see SI Appendix, Materials and Methods for complete synthetic details). Spacer lengths of two and three carbon atoms were explored because previous investigations indicated these closest mimic the PRMT transition state (13). In addition, unsaturated analogs of the three-carbon spacer constructs were also prepared to examine the effect of reduced linker flexibility. Compounds 1–6 were specifically designed so as to mimic the transition state of the first methylation reaction catalyzed by CARM1. Given that the first catalytic step is common to all PRMTs, this strategy represents a general approach that should be applicable for studying substrate binding by any PRMT enzyme (type I, II, or III) and its corresponding substrate(s). Also prepared were compounds 7 and 8, in which the target arginine of the PABP1 fragments was incorporated as the asymmetrically dimethylated species. Compounds 7 and 8 thus represent product inhibitors of CARM1 and were included as a means of evaluating the relative inhibitory potency of transition state mimics 1–6.

The inhibitory activity of compounds 1–8 toward CARM1 was investigated and compared with the ability to inhibit PRMT1, the most abundant arginine methyltransferase. An initial inhibition screen performed at a fixed inhibitor concentration of 50 µM revealed that compounds 1–6 were all effective CARM1 inhibitors, whereas the dimethylated PABP1-derived peptides 7 and 8 showed no effect under these conditions. Based on these preliminary findings, complete IC₅₀ curves were generated for compounds 1–6 against both CARM1 and PRMT1 (Table 1). Analogs containing three-atom saturated linkers (compounds 3 and 4) and unsaturated linkers (compounds 5 and 6) display the most potent inhibition of CARM1 with IC₅₀ values <100 nM. Shortening the spacer from three to two carbon atoms results in a significant decrease in activity, suggesting that a three-atom linker between the guanidine moiety and the adenosine group is optimal for mimicking the transition state geometry. When tested against PRMT1, compounds 1–6 displayed significantly reduced inhibitory activity. In light of these findings, we next examined the application of the more potent CARM1 inhibiting compounds 3–6 as transition state mimics for use in cocrystallization studies.

Table 1.

IC₅₀ values measured for compounds 1–8 against CARM1 and PRMT1

	IC50 values*
Compound	CARM1	PRMT1
AdoHcy	0.276 ± 0.052	7.65 ± 2.98
1 (Peptide 1, n = 1)	1.93 ± 0.55	16.28 ± 4.49
2 (Peptide 2, n = 1)	5.90 ± 0.87	16.29 ± 5.68
3 (Peptide 1, n = 2)	0.0920 ± 0.0132	12.34 ± 3.65
4 (Peptide 2, n = 2)	0.0901 ± 0.0107	25.54 ± 7.40
5 (Peptide 1, n = 2, unsat.)	0.0817 ± 0.0116	4.29 ± 1.04
6 (Peptide 2, n = 2, unsat.)	0.0876 ± 0.0092	16.62 ± 5.96
7 (Peptide 1, aDMA)†	>50	>50
8 (Peptide 2, aDMA)†	>50	>50

^*IC₅₀ values reported in micromolars. IC₅₀ values from triplicate data obtained at a range of 7–10 concentrations ± SD (see SI Appendix, IC₅₀ curves).

^†In compounds 7 and 8, the central arginine residue is present in asymmetrically dimethylated form.

Crystallization and Structure Determination of CARM1 Complexes with Transition State Mimicking Compounds 3–6.

Cocrystallization studies were performed with the PABP1-derived transition state mimics and an isolated catalytic domain of mmCARM1 (Mus musculus CARM1, residues 130–487) (15, 23). All structures were solved and refined to 2-Å resolution in the space group P2₁2₁2 as seen for mmCARM1 catalytic domain (15) and contain one copy of the CARM1 tetramer in the asymmetric unit (SI Appendix, Table S1). In all cases, each CARM1 monomer binds one molecule of the transition state mimic investigated (Fig. 2A). As indicated in Fig. 2B, the structures solved include near-complete assignment of all residues in the peptide fragments of compounds 3–6. Compounds 3 and 5 are comprised of a 13-aa sequence from PABP1 (residues 447–459) centered around R455, flanked by eight residues on the N-terminal side and four residues on the C-terminal side. The electron density maps obtained in the cocrystallization studies with both 3 and 5 reveal the conformation of 12 of 13 residues in two monomers and for the full-length peptide (13 residues) in the two other CARM1 monomers (Fig. 3A). Compounds 4 and 6 are comprised of an 11-aa sequence from PABP1 (residues 456–466) centered around R460 flanked by four residues on the N-terminal side and six residues on the C-terminal side. The electron density maps obtained in the cocrystallization studies with both 4 and 6 reveal the conformation of 8 of 11 residues in three of the monomers and 9 of 11 residues in the remaining CARM1 monomer (Fig. 3B). The last two residues on the C-terminal side of 4 and 6 are not seen in the electron density map. As expected for the structures obtained with compounds 3–6, the adenosine moiety occupies the AdoMet binding pocket. In the other CARM1 structures reported to date, this same pocket is occupied by either AdoHcy or sinefungin (15–18).

Fig. 2.

Crystallization and structure determination of CARM1-transition state mimic complexes. (A) Dimer of mmCARM1 with compound 3 bound. CARM1 monomers within the dimer (green, blue) are represented as a cartoon. Compound 3 is represented as sticks. N-terminal helices of each monomer (not shown in other figures) are here highlighted in light blue. (B) Peptides sequences used and results from crystallization studies with compounds 3–6: Residues visible in all four monomers of the asymmetric unit indicated in green and those seen in at least one monomer are shown in red. Compounds 3 and 4 contain a saturated three-atom spacer, whereas compounds 5 and 6 contain an unsaturated three-atom spacer.

Fig. 3.

Electron density (2F_obs−F_calc) weighted maps. Compound 5 (A) and compound 6 (B) bound to subunit B of mmCARM1. CARM1 is represented as cartoon, and peptide-based transition state mimics 5 and 6 are represented as sticks. Maps are represented as a mesh, contouring level set to 1σ.

The three carbon-atom linker appears to provide the ideal spacer to position and orient both the adenosine group and the modified arginine residue present in the peptide-based transition state mimics (Fig. 4). Specifically, the four PRMT signature motifs surround the binding pocket occupied by the covalently linked adenosine and arginine components of compounds 3–6. The catalytic site is locked on one side by the PRMT motif I (Y₁₅₀F₁₅₁xxY₁₅₄ mmCARM1 numbering) and the guanidine moiety of the arginine side chain is hydrogen-bonded to E258 and E267 belonging to the double-E loop and H415 of the THW loop. Of particular interest is the comparison of the binding modes for the transition state mimics that contain a fully saturated spacer (3 and 4) with those containing an unsaturated spacer (5 and 6), revealing both to have similar conformations (Fig. 4 and SI Appendix, Figs. S3 and S4). These findings indicate that the unsaturated spacers in 5 and 6 lead to a preformed conformation suitable for binding in the CARM1 active site. The additional hydrogen atoms present in the saturated linkers of compounds 3 and 4 result in a modification of the conformation of M269 and a slightly different network of water molecules (Fig. 4 and SI Appendix, Fig. S3).

Fig. 4.

Cocrystal structures of CARM1 with transition state mimics 5 and 6 revealing interacting active site residues and water molecules. Cartoon and stick representation of the transition state mimics bound to mmCARM1. Compound 5 is derived from PABP1_447–459 (A) and compound 6 is derived from PABP1_456–466 (B). Both compounds contain an unsaturated linker between the guanidine moiety and the adenosine unit. For clarity, N-terminal helices of CARM1 are not shown.

Another striking result is the finding that the conformations observed for the peptide components of the CARM1-bound transition state mimics are identical to those observed with unmodified PABP1 fragments bound to CARM1 in a recently reported cocrystallization study (18). Superimposition of the structures of the binary complexes formed between mmCARM1 and transition state mimics 3 or 5 with the structures recently reported for the ternary complexes of hsCARM1, sinefungin, and a PABP1-derived peptide containing R455 (18) reveal an identical conformation for the visible PAPB1 substrate peptide in both complexes (Fig. 5A and SI Appendix, Figs. S5 and S6). A similar result is obtained when superimposing the structures of binary complexes of mmCARM1 and transition state mimics 4 and 6 with the ternary complex of hsCARM1, sinefungin, and another PABP1-derived peptide containing R460 (Fig. 5B and SI Appendix, Fig. S6). Moreover, the electron density maps of the complexes of mmCARM1 with 3 and 5 reveal the conformation of four more residues at the N-terminal side of the peptide compared with the ternary complexes of hsCARM1, sinefungin, and the PABP1-derived peptide containing R455 (Fig. 5A and SI Appendix, Fig. S5). In the case of the complexes formed between mmCARM1 and transition state mimics 4 and 6, two additional N-terminal residues are also visible compared with the ternary complex of hsCARM1, sinefungin, and a PABP1-derived peptide containing R460 (Fig. 5B and SI Appendix, Fig. S6). Furthermore, it is interesting to note that the recently reported hsCARM1-PABP1 peptide complexes could only be solved in the presence of the AdoMet-competitive methyltransferase inhibitor sinefungin (18). The inclusion of sinefungin in the AdoMet binding site has an impact on the conformation of R169 that normally interacts with the methionine moiety of AdoMet. This distortion in turn leads to a small alteration in the conformation of E258, resulting in a change in the arrangement of bound water molecules. By comparison, transition state mimics 3–6 do not introduce such disturbances in the active site. The three carbon-atom linker used in these conjugates serves to properly orient and position the substrate peptide while simultaneously docking the adenosine moiety in the AdoMet binding site. In doing so, a “frozen” image of the transition state of the methylation reaction is effectively achieved.

Fig. 5.

Structures of transition state mimics overlaid with isolated peptides in presence of SFG. Superimposition of transition state analogs 5 and 6 and free PABP1 peptides on CARM1. (A) Superimposition of compound 5 (blue sticks) bound to mmCARM1 on SFG-PABP1(R455) (orange sticks) bound to hsCARM1 (PDB ID code 5DX1). (B) Superimposition of compound 6 (blue sticks) bound to mmCARM1 on SFG-PABP1(R460-MMA) (orange sticks) bound to hsCARM1 (PDB ID code 5DXA). SFG molecule is not shown for the hsCARM1 structures.

Discussion

Given the challenges associated with obtaining stable complexes of PRMTs and their corresponding protein substrates, the majority of PRMT crystal structures reported to date are of free enzyme, often bound to an analog of the AdoMet cofactor. These structures reveal a strong conservation of catalytic domain architecture and specifically the AdoMet binding pocket. In contrast, no structures have been solved for PRMTs in complex with full-length target proteins, and only a few structures have been reported for PRMTs in complex with small, well-resolved, peptide substrate fragments (18, 24, 25). The covalent tethering of substrates to enzymes has proven to be an effective strategy for capturing difficult to characterize enzyme–substrate complexes, leading to their structural characterization (26, 27). In a related approach, we describe here the application of transition state mimics for enabling structural studies with the PRMTs. Specifically, we designed transition state mimics 1–6 based on CARM1 peptide substrates wherein the arginine that is normally methylated is instead covalently linked to an adenosine moiety mimicking that present in the AdoMet cofactor. Compounds 3–6 were found to be particularly potent inhibitors of CARM1, and the subsequently solved cocrystal structures reveal that they are effective transition-state mimics capable of revealing biologically relevant peptide conformations. This assertion is supported by the near-identical conformations seen when superimposing the binary complexes described here with the recently reported ternary structures of hsCARM1, sinefungin, and a PABP1-derived peptide (18). Furthermore, it is interesting to note that ternary hsCARM1–PABP1 peptide complexes could only be solved in the presence of the AdoMet-competitive methyltransferase inhibitor sinefungin. The inclusion of sinefungin in the AdoMet binding site has an impact on the conformation of several residues. By comparison, the transition state mimics here described do not introduce any unnatural perturbations of the AdoMet binding site.

It has been shown that long distance interactions are crucial for substrate recognition and binding by PRMTs (15–18, 28, 29). However, despite extensive efforts, crystal structures of PRMT complexes with longer peptides or full-length substrates have not yet been solved. The transition state mimic approach here described presents a strategy toward stabilizing PRMT–substrate complexes for use in structural studies aimed at deciphering substrate recognition and binding by PRMTs at the atomic level. The specific peptide constructs used exhibit potent CARM1 inhibition and effectively bind to the enzyme despite having limited secondary structure relative to what would be expected for full-length substrates. In this regard, the transition state mimic approach is also expected to be applicable in preparing much larger constructs (i.e., via protein total synthesis using native chemical ligation strategies) as a means of stabilizing PRMT–substrate complexes in their biologically relevant conformations. The aim of the present study was to generate transition state mimics of the first methylation reaction catalyzed by CARM1 and represents an approach that should be generally applicable to all PRMTs. It is conceivable that similar transition state mimics, bearing an additional methyl group on the guanidine unit of the modified arginine, might also be able to provide subtype-specific structural insights into PRMT catalysis. Specifically, incorporation of a methyl group at the nitrogen atom connected to the adenosine moiety could provide a transition state mimic for characterizing the active site features leading to asymmetric arginine dimethylation as for the type I PRMTs. Conversely, a mimic bearing a methyl group at the other terminal guanidine nitrogen, not connected to the adenosine group, would provide a tool with which to study symmetric dimethylation as catalyzed by the type II PRMTs.

Given the transient nature of complexes formed between PRMTs and their peptide/protein substrates, tool compounds like those presented here are expected to be of great value in achieving a comprehensive structural understanding of how PRMT methylation is regulated, an important and outstanding challenge in the field of epigenetics. In addition, insights gleaned from the approaches described here may also provide inspiration for the design of smaller, peptidomimetic compounds that retain the potent and selective bisubstrate inhibitory properties of the transition state mimics.

Materials and Methods

Synthetic Procedures.

Compounds 1–8 were synthesized by using a newly developed methodology allowing for the on-resin preparation of peptides containing N^G-substituted arginine residues. Specifically, PABP1-derived peptides were synthesized by using standard solid-phase techniques after which the adenosine group was introduced. Detailed synthetic protocols are provided in the accompanying SI Appendix, Materials and Methods, including experimental procedures and analytical data for all new compounds.

Enzyme Assays.

Methyltransferase inhibition assays were performed as described (30) by using commercially available chemiluminescent assay kits for PRMT1 and PRMT4/CARM1 (purchased from BPS Bioscience). The enzymatic reactions were conducted in triplicate at room temperature for 1 h in substrate-coated well plates at a final reaction volume of 50 µL containing the manufacturer’s proprietary assay buffer, AdoMet (at a concentration of 5 times the respective K_m value for each enzyme), the methyltransferase enzyme: PRMT1 (80 ng per reaction, final enzyme concentration 38 nM) and CARM1 (200 ng per reaction, final enzyme concentration 63 nM), inhibitors 1–6, and AdoHcy (in the range of concentrations: 0.001–250 µM in water). Before addition of the AdoMet, the enzyme was first incubated with the inhibitor for 15 min at 37 °C. Positive controls were performed in the absence of inhibitors by using water to keep the final volume consistent. Blanks and substrate controls were performed in the absence of the enzyme and AdoMet, respectively. Following the enzymatic reactions, 100 μL of primary antibody (recognizing the respective immobilized asymmetrically dimethylated arginine product) was added to each well, and the plate was incubated at room temperature for an additional 1 h. Then, 100 μL of secondary horseradish peroxidase (HRP)-conjugated antibody was added to each well, and the plate was incubated at room temperature for additional 30 min. Finally, 100 μL of an HRP substrate mixture was added to the wells, and the luminescence was measured directly by using a standard microplate reader. In all cases, enzyme activity measurements were performed in triplicate at each of the inhibitor concentrations evaluated. The luminescence data were analyzed by using GraphPad Prism (version 6.02).

The luminescence data were normalized with the highest value obtained in a concentration range defined as 100% activity and the lowest value defined as 0%. Blank and positive controls were performed to analyze the validity of the assay method. Blank values obtained were 0.4–2.2% for PRMT1 and 4.3–29.3% for CARM1. Because the blank values were generally higher than the lowest value obtained in a concentration range, the lowest value in each series was set at 0%. Positive controls were generally within ±15% of the highest values obtained in a concentration range, but occasionally the highest values were up to 33% higher than the positive control. The percent activity values were plotted as a function of inhibitor concentrations and fitted by using nonlinear regression analysis of the Sigmoidal dose–response curve generated by using normalized data and a variable slope following Eq. 1,

$$\mathit{Y}=\frac{100}{\left(1+{10}^{\left(\left(LogIC50-X\right)*HillSlope\right))}\right)},$$ [1]

where Y = percent activity, X = the logarithmic concentration of the compound, and Hill Slope = slope factor or Hill coefficient. The IC₅₀ value was determined by the concentration resulting in a half-maximal percent activity. The IC₅₀ values measured for AdoHcy, which served as a reference compound, are similar to those reported (13). Full IC₅₀ curves are presented in SI Appendix, IC₅₀ curves.

CARM1 Cloning, Expression, and Purification.

The Mus musculus CARM1 gene sequence corresponding to the PRMT core (residues 130–487, mmCARM1_130–487) was amplified by PCR from the original GST-CARM1 construct (15, 31) (see SI Appendix, Materials and Methods for full experimental details).

X-Ray Data Collection and Structure Determination.

Crystallization.

Transition state mimics 3–6 were solubilized in water before addition to the protein solution (2 mg⋅mL⁻¹) at the final concentration of 2 mM. The protein–peptide solution was incubated 20 min at room temperature before use. Vapor diffusion method using hanging drop trays with a 0.5-mL reservoir was used for crystallization. Typically, 2 μL of protein-ligand solution were added to 1 μL of well solution consisting of 14–16% (v/v) PEG 3350, 100 mM Tris⋅HCl pH 8.5, and 200 mM (NH₄)₂SO₄. Crystals grew in a few days at 293 K.

Data collection and structure solution.

Crystals were flash-frozen in liquid nitrogen after a brief transfer to 5-µL reservoir solution containing 15% (vol/vol) PEG 400 as a cryoprotectant and were stored in liquid nitrogen. The diffraction datasets were collected on the SOLEIL PROXIMA1 and ESRF ID-29 beamlines, using a PILATUS 6M (Dectris) or a Quantum 315 (ADSC) detector and processed with XDS (32) and HKL-2000 (33). The crystals belonged to the P2₁2₁2 space group with four monomers of CARM1 in the asymmetric unit. The structures were solved by molecular replacement using CARM1 structure as a probe (15). Model building and refinement were carried out by using Coot (34) and PHENIX (35). TLS refinement with three groups per polypeptide chain was used. All other crystallographic calculations were carried out with the CCP4 package (36). Structure figures were generated with PyMol.

Data availability.

All structures and structure factors have been deposited in the Protein Data Bank (CARM1-Compound 3 = 5LGP; CARM1-Compound 4 = 5LGQ; CARM1-Compound 5 = 5LGR; CARM1-Compound 6 = 5LGS).

Additional information.

Supplementary information, chemical compound information, and source data are available in SI Appendix, Materials and Methods.