Chemical Biology of Protein Citrullination by the Protein Arginine Deiminases (PADs)

Abstract

Citrullination is a posttranslational modification (PTM) that converts peptidyl-arginine into peptidyl-citrulline; citrullination is catalyzed by the Protein Arginine Deiminases (PADs). This PTM is associated with several physiological processes, including the epigenetic regulation of gene expression, neutrophil extracellular trap formation, and DNA-damage induced apoptosis. Notably, aberrant protein citrullination is relevant to several autoimmune and neurodegenerative diseases and certain forms of cancer. As such, the PADs are promising therapeutic targets. In this review, we discuss recent advances in the development of PAD inhibitors and activity-based probes, the development and use of citrulline-specific probes in chemoproteomic applications, and methods to site-specifically incorporate citrulline into proteins.

Introduction:

Posttranslational modifications (PTMs) of histone proteins are at the forefront of the epigenetic regulation of gene expression [1,2]. One such PTM is citrullination, which involves the hydrolysis of the positively-charged guanidium side chain of an arginine residue to form the neutral urea present in citrulline (Figure 1A) [3,4]. Histone citrullination leads to both activation and repression of gene transcription. For example, the citrullination of H3R26 at estrogen receptor α (ERα) target genes enhances the expression of >200 genes, whereas citrullination of H3R17 at the ERα-regulated pS2 promoter leads to transcriptional repression [5,6]. Histone citrullination also plays a pivotal role in neutrophil extracellular trap (NET) formation or NETosis [7,8]. This PTM is catalyzed by a group of cysteine hydrolases called Protein Arginine Deiminases (PADs) [9,10]. Despite evidence for decitrullination in MCF-7 cells [5], no decitrullinase has been discovered to date and, therefore, citrullination is considered an irreversible PTM.

Figure 1.

(A) Conversion of peptidyl-arginine to peptidyl-citrulline by PADs. (B) Crystal structure of PAD2, indicating the head-to-tail homodimer and different domains in each protomer (PDB: 4N20) [20]. (C) Overlay of Ca²⁺-binding sites (Ca1-Ca6) in PAD1, PAD2 and PAD4 (cartoon colors, pale blue: PAD1, pale green: PAD2, pale yellow: PAD4) (PDB IDs: 5HP5 (PAD1), 4N2C (PAD2), 1WDA (PAD4)) [20–22]. (D) Overlay of apo- and holo-PAD2 active sites, indicating the Ca²⁺-induced conformational changes (PDB: 4N20 (apo) and 4N2C (holo)) [20]. (E) Crystal structure of PAD4 C645A mutant active site in complex with the small molecule substrate BAA (PDB: 1WDA) [21].

Aberrant protein citrullination is a hallmark of multiple autoimmune diseases, including rheumatoid arthritis (RA), multiple sclerosis (MS), ulcerative colitis (UC), lupus, neurodegenerative diseases, sepsis and certain cancers [3,4,10]. These disease links establish the PADs as promising therapeutic targets. We developed several pan-PAD inhibitors, including Cl-amidine and BB-Cl-amidine that exhibit excellent efficacy in multiple animal models of these diseases [11,12]. We also developed several activity-based probes that enabled fluorescence labeling of the PADs and the evaluation of the off-target toxicity of PAD inhibitors [13–15]. Furthermore, using phenyl glyoxal-based probes and chemoproteomics, we identified several non-histone citrullinated proteins, including RNA-binding motif protein X chromosome (RBMX), U2 auxiliary factor 2 (U2AF2), serine/arginine-rich splicing factor 7 (SRSF7), RNA-binding protein 39 (RBM39), heterogeneous nuclear ribonucleoprotein A/B (HNRNPAB) and polypyrimidine tract binding protein 1 (PTBP1) from a PAD2-overexpressing HEK293T cell line [16]. While most of these proteins are involved in RNA-binding and mRNA splicing, the impact of this PTM on their function remains elusive. Interestingly, Lee et al. also identified several citrullinated RNA-binding proteins that are present various human tissues, indicating a potential role for citrullination in modulating RNA biology [17]. Additionally, using our citrulline-specific probes and proteomics studies, we identified various classes of citrullinated non-histone proteins (described later) from the synovial fluid and synovial tissue of RA patients [18]. More recently, we developed a novel technology to site-specifically incorporate citrulline into proteins in mammalian cells [19]. This technology enables studies to understand the downstream implications of citrullination on the structure and activity of a given protein. In this review, we discuss recent advances in developing PAD inhibitors, activity-based probes, citrulline-specific probes, and the site-specific incorporation of citrulline into proteins.

Protein Arginine Deiminases (PADs):

Protein citrullination was first detected in trichohyalin, an insoluble, cross-linked protein abundant in the medulla and inner root sheath of hair follicles [23]. Subsequent studies demonstrated that citrulline is generated from arginine by an enzyme present in the hair follicles [24]. This enzyme, later denoted as a protein arginine deiminase (PAD), was found to be distinct from prokaryotic arginine deiminases and eukaryotic nitric oxide synthase (NOS) as it cannot convert free L-arginine to L-citrulline and requires calcium for activity. The PAD family consists of five highly related isozymes (PAD1, 2, 3, 4 and 6). However, only four (PADs1–4) are catalytically active [3,4,10]. PAD structures are comprised two N-terminal immunoglobulin-like domains, IgG1 and IgG2, that are connected in series to a C-terminal catalytic domain (Figure 1B). Furthermore, PAD2 and PAD4 independently crystallize as head-to-tail homodimers with both active sites on the same face of the dimer [20,21]. Notably, PADs contain a nucleophilic cysteine (C645 in PAD1, 4; C647 in PAD2; C646 in PAD3), a histidine (H471), and two aspartates (D350 and D473) in the active site. While the cysteine and histidine play direct roles in catalysis, the aspartates stabilize the positively-charged substrate arginine between the active site cysteine and histidine. As mentioned earlier, calcium increases PAD activity by >10,000 fold and PADs contain 4 (PAD1), 5 (PAD3, 4) and 6 (PAD2) Ca²⁺-binding sites (Figure 1C). Ca²⁺-binding at these sites leads to a series of conformational changes that afford a catalytically competent active site. For example, in apo-PAD2, the active site is shielded by a gatekeeper arginine, R347, and C647 remains away from the catalytic center (Figure 1D) [20]. Ca²⁺-binding at Ca3, Ca4 and Ca5 sites generates the Ca2 site, and Ca²⁺-binding at the latter site brings C647 into the catalytic center and swings R347 out of the active site, allowing the entry of a substrate arginine. W348 also moves into a suitable position to stabilize the side chain of a substrate arginine via hydrophobic interactions (Figure 1D) [20]. Although PADs exhibit a broad substrate scope, histones are a major PAD substrate and PADs 1, 2 and 4 all enter the nucleus and catalyze histone citrullination [3,4,10]. Notably, PAD4 is the only isozyme with a canonical nuclear localization sequence, indicating that other mechanisms promote the nuclear localization of PADs 1 and 2. Historically, PAD activity has been studied with either histone-derived peptides or arginine analogs [21,25–27]. The crystal structure of one such PAD substrate, N^α-benzoyl arginine amide (BAA), in complex with the PAD4 C645A mutant indicates that R374, D350 and D473 play important roles in substrate recognition (Figure 1E) [21]. Structures of PAD4 bound to histone-derived peptides indicate that larger substrates form a β-turn conformation upon interacting with the enzyme. Less is known about how the other PADs bind their substrates [28].

PAD Inhibitors:

The role of PADs in various disease pathologies has attracted significant interest for developing PAD inhibitors both from academia and industry. We previously developed the pan-PAD inhibitors, F-amidine and Cl-amidine, which covalently modify the catalytic cysteine by reaction with their fluoro- and chloro-acetamidine warheads, respectively, (Figures 2A–C) [12,29]. Structure-activity relationship (SAR) studies on these inhibitors afforded o-F-amidine and o-Cl-amidine with significantly improved potency [30]. A cocrystal structure of the PAD4-o-F-amidine complex revealed that the o-carboxylate hydrogen bonds (H-bond) with W347 and a water that is also H-bonded to Q346, accounting for the tighter binding than parent scaffold (Figure 2D). Further optimizations afforded the more hydrophobic pan-PAD inhibitors, BB-F-amidine, BB-Cl-amidine, compound 1 and compound 2 (Figure 2A) with improved cellular efficacy and metabolic stability [11,31]. For example, BB-Cl-amidine and 2 exhibit 20- and 16-fold, respectively, higher cytotoxicity in U2OS cells than Cl-amidine. Also, BB-Cl-amidine has a superior in vivo half-life than Cl-amidine (1.75 h versus ~15 min). Moreover, BB-Cl-amidine at 10 mg/kg is more efficacious than Cl-amidine in the murine collagen-induced arthritis (CIA) model, where only a 50% reduction in disease severity was observed after 35 days treatment with 50 mg/kg Cl-amidine [32,33].

Figure 2.

(A) Chemical Structures of pan-PAD inhibitors and PAD3-selective inhibitors, F4-amidine and Cl4-amidine. (B) Potencies (k_inact/K_I) of pan-PAD and isozyme-selective PAD inhibitors for the inhibition of PADs1–4. Crystal structures of PAD4-F-amidine (C, PDB: 2DW5) [29], PAD4-o-F-amidine (D, PDB: 3B1U) [30], PAD4-AFM30a (F, PDB: 5N0Y) [39] and PAD4C645A-GSK199 (G, PDB: 4X8G) [40] complexes, indicating the noncovalent interactions responsible for inhibitor binding. (E) Chemical Structures of isozyme-selective PAD inhibitors. (H) Chemical Structures of photoswitchable inhibitors and their potencies for the inhibition of PAD2.

The differential roles of the PADs in various disease pathologies next led us to develop isozyme-selective PAD inhibitors. Attempts to increase metabolic stability afforded D-Cl-amidine as a PAD1-selective inhibitor (Figure 2E, A) [34]. The fact that PAD1 binds to D-Cl-amidine suggested that it possesses a larger active site that might accommodate bulky substituents. To that end, we developed SM26 and SM91 which possess large iodo substitutions that selectively form a halogen bond (X-Bond) with PAD1 [35–38] (Figure 2E, B) [14]. Interestingly, removal of the iodines or replacement of them with other hydrophobic alkyl groups led to a dramatic decrease in PAD1-potency and isozyme-selectivity. Detailed SAR established that the iodine at 5-position forms an X-Bond with PAD1. Notably, these compounds efficiently inhibit histone H3 citrullination in PAD1-overexpressing HEK293T cells. They also dramatically arrest early embryo development of mouse zygotes at the 4–8 cell stage.

Detailed SAR studies on BB-Cl-amidine also afforded AFM30a and AFM41a as PAD2-selective inhibitors (Figures 2E, B) [39]. The fluoroacetamidine warhead, and the 4-methoxy and 1-methyl substitutions on the benzimidazole ring contribute significantly to PAD2-selectivity. A cocrystal structure of AFM30a bound to PAD4 revealed that the lactam ring forms a H-bond with W347 (Figure 2F) and superposition of this structure on PAD2 indicates that the benzimidazole group is positioned in a unique hydrophobic pocket on PAD2, accounting for PAD2-selectivity. AFM30a exhibits excellent efficacy in animal models of TLR-7–dependent lupus and endotoxic shock [41,42]. Furthermore, inhibition with AFM30a indicates a novel role for PAD2 in NLRP3 inflammasome assembly and pro-inflammatory IL-1β release in pyroptotic macrophages [43].

Only a few PAD3-selective inhibitors have been developed. F4-amidine and Cl4-amidine are two of them (Figure 2A) [27]. Using a substrate-based fragment method and SAR studies, Jamali et al. developed 3 and 4 with >10-fold PAD3-selectivity (Figure 2E) [44,45]. Notably, 4 rescued thapsigargin-induced cell death in PAD3-overexpressing HEK293T cells.

The first PAD4-selective inhibitor, TDFA was identified from a 264-membered peptide library (Figures 2E, B) [46]. The selectivity for PAD4 is due in part to an H-bond between the aspartate in TDFA and R639, which is unique to PAD4. TDFA is also significantly more potent than Cl-amidine for the inhibition of histone H3 citrullination in HL-60 cells [46]. Screening of a DNA-encoded small molecule library ultimately led to the identification of GSK-199 as the first PAD4-selective reversible inhibitor [40]. GSK199 binds to the calcium-free form of PAD4 (apo-PAD4) and forms H-bonds with D473 and H471. The PAD4-selectivity originates from the close packing of the benzimidazole moiety against F634, which is unique to PAD4 (Figure 2G). Notably, GSK199 inhibits NETosis in mouse and human neutrophils, and reduces disease severity in the CIA-model of RA [40,47].

Recently, we developed SM7T as photo-activatable PAD2 inhibitor (Figure 2H) [48]. This molecule exploits the trans-cis photoisomerism of azobenzenes and conversion of SM7T to SM7C with 350 nm light led to a 10-fold increase in potency. Surprisingly, the fluoroacetamidine-containing analogue, SM9T led to a 45-fold drop in activity upon irradiation. In agreement with the in vitro potencies, SM7C dose-dependently inhibits histone H3 citrullination in HEK293TPAD2 cells, while SM7T is completely inactive, indicating that this compound is suitable for developing photopharmaceuticals for PAD-related diseases.

Chemical Probes:

The covalent nature of our probes led us to develop fluorophore-tagged versions that can be used as activity-based probes. The first probe, rhodamine-conjugated F-amidine (RFA) can be used to monitor inhibitor potency by in-gel fluorescence and fluorescence polarization (Figure 3A, B) [13]. Using these platforms, we identified ruthenium red, minocycline, chlorotetracycline, NSC95397 and streptonigrin as PAD inhibitors from high-throughput screens [49–51]. Recently, we developed alkyne-tagged versions of BB-Cl-amidine, BB-Cl-Yne and BB-F-Yne, that can be conjugated with fluorescent or biotin reporter tags using click chemistry (Figure 3A, C) [15]. These probes and a chemoproteomic platform enabled us to identify off-targets for BB-Cl-Yne. These include highly abundant proteins heterogeneous nuclear ribonucleoprotein U, clathrin heavy chain 1, bifunctional glutamate/proline-tRNA ligase, tubulin- and the actin β-chain in PAD2-expressing HEK293T cells. By contrast, BBF-Yne was highly selective for PAD2, indicating the potential of using the fluoroacetamidine warhead to develop isozyme- and proteome-selective PAD inhibitors. Similar selectivity was observed for the SM26-derived probe, SM119, which failed to identify off-targets in HEK293TPAD1 cells. This data supports the remarkable proteome-wide selectivity of the fluoroacetamidine warhead and this inhibitor scaffold (Figure 3A) [14].

Figure 3.

(A) Chemical structures of PAD-targeted activity-based probes. (B) High-throughput screening platforms using RFA. (C) Labelling of cellular PADs and proteomic identification of cellular off-targets by the clickable probes, BB-Cl-Yne, BB-F-Yne and SM119. (D) Condensation of phenyl glyoxal-based citrulline-specific probes, Rhodamine-PG and Biotin-PG, with citrulline at acidic pH.

To additionally monitor PAD activity, we developed a set of probes (Rh-PG and Biotin-PG) that can chemically tag citrullinated proteins. The chemical logic behind these compounds relates to the ability of phenyl glyoxal (PG) to condense with citrulline at acidic pH (Figure 3D) [16,52]. Using Rh-PG, we visualized extensive protein citrullination in the serum of mice with colitis, and in the synovial fluid and synovial tissue of RA patients. Biotin-PG in combination with chemoproteomics permitted the identification of citrullinated proteins from HEK293TPAD2 cells. This platform also allowed us to identify various classes of novel citrullinated proteins, including serine protease inhibitors (SERPINs), serine proteases, transport proteins and complement system components along with known citrullinated proteins (e.g., vimentin, enolase, keratin and fibrin) from RA synovial fluid and synovial tissue [18]. In this study, we further demonstrated that citrullination of SERPINs and nicotinamide N-methyl transferase (NNMT) dramatically abolishes their activity.

Incorporation of Citrulline into Proteins:

Despite our success in understanding the biochemistry of citrullination and its contribution to various disease pathologies, the effect of citrullination on the structure and activity of a given protein remain elusive mainly due to the lack of a suitable method to incorporate citrulline into proteins. Although a citrullinated protein of interest can theoretically be generated by its treatment with a PAD, this treatment most often leads to citrullination at all the available sites with a range of stoichiometries, which does not necessarily recapitulate the scenario in vivo. Other techniques, including in vitro translation, in vivo nonsense suppression using chemically-acylated tRNAs, and post-translational mutagenesis are limited in many aspects [53–55]. For example, post-translational mutagenesis involves the treatment of dehydroalanine-containing proteins with radical generators under denaturing conditions and affords a mixture of D/L-citrulline (Figure 4A). Additionally, this method cannot be performed in cells.

Figure 4.

(A) Post-translational mutagenesis to incorporate citrulline into proteins. (B) Chemical structure of SM60. (C) Site-specific incorporation of citrulline into proteins in mammalian cells using genetic code expansion technology. Posttranslational conversion of SM60 to citrulline needs irradiation with 365 nm UV.

To overcome these challenges, we developed a novel genetic code expansion approach to incorporate citrulline. This approach exploits an E. coli-derived leucyl tRNA synthetase (LeuRS)/tRNA^Leu pair that can site-specifically incorporate a photocaged-citrulline, SM60, into proteins in response to a nonsense codon[19]. SM60 can subsequently be converted to citrulline by 365 nm UV (Figures 4B, C). Notably, SM60 and its photodecaged products are completely nontoxic at the concentration used for nonsense suppression. Using this technique, we incorporated citrulline in PAD4 at two known autocitrulllination sites, R372 and R374. Using benzoylarginine ethyl ester (BAEE, a small molecule substrate of PADs) and a traditional COLDER assay [56] to detect citrulline formation, we found that the R372Cit and R374Cit mutants are 181- and 9-fold less active than WT PAD4, indicating that citrullination can have profound effects on PAD4 activity. We also assessed the cellular activity of WT PAD4 and the R374Cit mutant and showed that citrullination of R374 leads to 6-fold less citrullinated histone H3, consistent with the in vitro data. Despite the lower activity of citrulline-containing mutants, autocitrullination does not impact the activity of PAD4 as we found that only a minor fraction of PAD4 undergoes autocitrullination, and amongst the 19 sites of autocitrullination, R372 and R374 are only minor sites [19].

Conclusions:

In this review, we highlighted recent advances in the biochemistry and chemical biology of protein citrullination. Although the field has matured and our new technology to site-specifically incorporate citrulline into proteins in mammalian cells will further reveal the biology of citrullination at the molecular level, many questions remain unanswered. These include but are not limited to, how PAD activity is activated in cells in the presence of submicromolar concentrations of calcium, the full repertoire of proteins that are PAD substrates, and the full impact of citrullination on protein function. Knowledge of these areas will result in further advances in the development of novel drugs that block diseases associated with aberrant citrullination.