Protein citrullination: inhibition, identification and insertion

Abstract

Protein citrullination is a post-translational modification (PTM) that is catalysed by the protein arginine deiminase (PAD) family of enzymes. This PTM involves the transformation of an arginine residue into citrulline. Protein citrullination is associated with several physiological processes, including the epigenetic regulation of gene expression, neutrophil extracellular trap formation and DNA damage-induced apoptosis. Aberrant protein citrullination is relevant to several autoimmune and neurodegenerative diseases and certain forms of cancer. PAD inhibitors have shown remarkable efficacy in a range of diseases including rheumatoid arthritis (RA), lupus, atherosclerosis and ulcerative colitis. In RA, anti-citrullinated protein antibodies can be detected prior to disease onset and are thus a valuable diagnostic tool for RA. Notably, citrullinated proteins may serve more generally as biomarkers of specific disease states; however, the identification of citrullinated protein residues remains challenging owing to the small 1 Da mass change that occurs upon citrullination. Herein, we highlight the progress made so far in the development of pan-PAD and isozyme selective inhibitors as well as the identification of citrullinated proteins and the site-specific incorporation of citrulline into proteins.

This article is part of the Theo Murphy meeting issue ‘The virtues and vices of protein citrullination’.

1. Introduction

The protein arginine deiminases (PADs) are a family of five enzymes that convert arginine residues in proteins into citrulline [1–3] (figure 1a). Protein citrullination came into prominence in the late 1990s when Schellekens [4,5] and colleagues first reported their findings that rheumatoid arthritis (RA) patients generated antibodies to citrullinated proteins such as enolase, vimentin [6] and filaggrin [7]. Subsequent work by Schellekens and others demonstrated that these anti-citrullinated protein antibodies (ACPA) are the most specific diagnostic for RA (approx. 90% specificity) and that ACPA can be detected prior to disease onset [8–10]. This finding suggested that ACPA are pathogenic and contribute to disease onset and progression. Much work has been subsequently devoted to studying immunity to citrullinated antigens including their use as neoantigens in anti-tumour vaccines. This work is reviewed elsewhere [11–14].

Figure 1.

Protein citrullination and catalytic mechanism. (a) PAD catalysed hydrolysis of peptidyl-arginine to generate peptidyl-citrulline. (b) PAD4 uses a reverse protonation mechanism to convert arginine into citrulline. (Online version in colour.)

Two other findings further drove interest in protein citrullination and the enzymes that catalyse this reaction, i.e. the PADs. First, Coonrod and Kouzarides showed that protein citrullination could antagonize the methylation of arginine residues in histones and contribute to gene regulation by modulating chromatin structure [15–17]. This finding indicated that the PADs were histone modifying enzymes that could contribute to the so-called histone code hypothesis of epigenetic gene regulation. The second major driver was a genome-wide association study studying identifying single nucleotide polymorphisms in PAD4 that increased the risk of developing RA [18]. This finding was particularly important because it provided a strong link between dysregulated PAD4 activity and the generation of ACPA and provided a strong impetus for developing PAD4 selective inhibitors as potential therapies for RA.

It was around this time that the Thompson laboratory began their work to characterize the activity of the five PAD isozymes (PADs 1–4 and PAD6) and develop inhibitors for these isozymes [17,19,20]. Our initial efforts focused on characterizing the substrate specificity and catalytic mechanism of PAD4 [17,21]. These efforts showed that PAD4 did not accept a methylated arginine as a substrate in contrast to a previous report [15]. Our mechanistic studies were particularly aided by the disclosure of the structure of PAD4 by the Sato group in 2004 [22]. This structure identified key catalytic residues including Asp 350, His471, Asp 473 and Cys645. We subsequently showed that PAD4 employs a reverse protonation mechanism to convert arginine into citrulline [21] (figure 1b). In this mechanism, the cysteine exists in the thiolate form and the histidine is protonated. The substrate binds and the cysteine thiolate attacks the guanidinium carbon. The resulting tetrahedral intermediate is probably stabilized by the donation of a proton from His471 [23]. Collapse of the tetrahedral intermediate leads to the loss of ammonia and the formation of an electrophilic thiouronium intermediate that is subject to nucleophilic attack by an active site water that is deprotonated by H471. The resulting tetrahedral intermediate then collapses to generate citrulline and restore the enzyme to the catalytically competent state.

2. Amidine-based protein arginine deiminase inhibitors

Given that the catalytic mechanism employed by the PADs resembles that employed by other cysteine hydrolases, including numerous cysteine proteases, we focused our initial efforts on developing irreversible inhibitors for the PADs because there is a long history of developing irreversible inhibitors for cysteine hydrolases [24]. Our substrate specificity studies on PAD4 showed that positive charge was essential for catalysis via interactions with Asp350 and Asp473 and prompted us to think about inhibitors that could mimic the positive charge of the guanidinium but contain an electrophilic group that could react with the cysteine. After initial failures, we hypothesized that a haloacetamidine warhead would not only provide favourable interactions with Asp350 and Asp473 but also possess an electrophilic moiety that would react with Cys645. As proof of concept, we synthesized F-amidine [20] (figure 2a) and showed that it is a highly potent inhibitor of PAD4. Next, we developed a structure activity relationship and generated a variety of analogues [19]. These efforts led to the development of chloro (Cl)-amidine [19,25] (figure 2a) which showed significantly higher cellular potency [19]. Fast and colleagues [26] also reported that 2-chloroacetamidine, the warhead alone, could also inactivate PAD4.

Figure 2.

Structures of PAD inhibitors and mechanism of inactivation by haloacetamidine containing inhibitors. (a) Structures of F-amidine, chloro (Cl)-amidine, tetrazole Cl-amidine and biphenyl-benzimidazole (BB)-Cl-amidine, respectively. Structures of PADs specific inhibitors including threonine aspartate F-amidine (TDFA) for PAD4, SM26 for PAD1 and AFM30a for PAD2. (b) Mechanisms by which haloacetamidine inhibitors irreversibly inhibit the PADs. (c) Structures of the PAD4 selective inhibitors GSK484, BMS-P5 and JBI-589. (Online version in colour.)

We also obtained the first structure of PAD4 bound to an inhibitor [19] (i.e. F-amidine) and developed the first activity-based probe targeting PAD4 [20]. F-amidine is classified as a mechanism-based inhibitor because nucleophilic attack on the amidine carbon generates an intermediate that facilitates an intramolecular halide displacement reaction [23] (figure 2b). The reaction with Cl-amidine also appears to proceed via this route but direct displacement of the chloride is also possible (figure 2b) because of the higher leaving group potential of chloride [23]. These disclosures further catalysed our efforts to develop more potent analogues including biphenyl-benzimidazole (BB)-Cl-amidine [27], which along with Cl-amidine have been the most widely used PAD inhibitors in both cellular and in vivo studies (table 1).

Table 1.

Select in vivo and cellular studies of PAD inhibitors developed by the Thompson laboratory.

PAD inhibitor	cellular and in vivo studies	reference
BB-Cl-amidine	peptidylarginine deiminase inhibition disrupts neutrophil extracellular trap (NET) formation and protects against kidney, skin and vascular disease in lupus prone MRL/lpr mice	[27]
BB-Cl-amidine	peptidylarginine deiminase inhibition prevents diabetes development in non-obese diabetic mice	[28]
BB-Cl-amidine	BB-Cl-amidine as a novel therapeutic for canine and feline mammary cancer via activation of the endoplasmic reticulum stress pathway	[29]
AFM-32a	inhibition of PAD2 improves survival in a mouse model of lethal lipopolysaccharide-induced endotoxic shock	[30]
TDFA	PAD 4 selective inhibitor TDFA protects lipopolysaccharide-induced acute lung injury by modulating nuclear p65 localization in epithelial cells	[31]
Cl-amidine	inhibiting PAD2 enhances the anti-tumour effect of docetaxel in tamoxifen-resistant breast cancer cells	[32]
Cl-amidine	peptidylarginine deiminase inhibition is immunomodulatory and vasculoprotective in murine lupus	[33]
Cl-amidine	suppression of colitis in mice by Cl-amidine: a novel peptidylarginine deiminase inhibitor	[34]
Cl-amidine	molecular targeting of protein arginine deiminases to suppress colitis and prevent colon cancer	[35]
Cl-amidine	peptidylarginine deiminase inhibitor Cl-amidine attenuates cornification and interferes with the regulation of autophagy in reconstructed human epidermis	[36]
Cl-amidine	N-α-benzoyl-N5-(2-chloro-1-iminoethyl)-L-ornithine amide, a protein arginine deiminase inhibitor, reduces the severity of murine collagen-induced arthritis	[37]
Cl-amidine	peptidylarginine deiminase inhibition reduces vascular damage and modulates innate immune responses inmurine models of atherosclerosis	[38]
Cl-amidine	a novel role for peptidylarginine deiminases in microvesicle release reveals therapeutic potential of PAD inhibition in sensitizing prostrate cancer cells to chemotherapy	[39]

3. Efficacy in animal models

Cl-amidine was the first PAD inhibitor to be tested in an animal model of RA [37]. The collagen-induced model of arthritis was used for these studies and the data showed that PAD inhibition could prevent disease onset and development when given at 10 or 50 mg kg⁻¹ d⁻¹ over the duration of the disease course. Subsequently, Cl-amidine was tested in a variety of other models including the dextran sodium sulfate model of ulcerative colitis where it dose-dependently reduced disease severity [34]. Notably, better efficacy was observed when Cl-amidine was provided in the drinking water than when given intra peritoneally [35]. The enhanced efficacy probably relates to the relatively short in vivo half-life of Cl-amidine when given intra peritoneally. Cl-amidine also demonstrated efficacy in models of lupus [33], sepsis [40] and atherosclerosis [38]. Subsequent work to develop inhibitors with enhanced cellular potency and in vivo stability led to the development of a series of tetrazole [41] and benzimidazole containing compounds (figure 2a) that include BB-Cl-amidine [38]. Notably, BB-Cl-amidine shows enhanced efficacy in models of lupus, RA and type I diabetes [28,38,42,43].

4. Mechanistic studies on protein arginine deiminases 1, 2 and 3

Given that Cl-amidine and BB-Cl-amidine inhibit the four active PAD isozymes with similar efficacies (PAD6 is not active) [2], we were interested in developing selective inhibitors of the PAD isozymes with the thought being that selective PAD inhibitors would show fewer potential side effects in vivo. Towards that end, we began efforts to study the catalytic mechanisms and substrate specificities of PAD1, 2, and 3 and compare them to PAD4 [21,44,45]. From these efforts, we showed that the PADs possess overlapping substrate specificities and that like PAD4, PADs 1 and 3 employ a reverse protonation mechanism to facilitate hydrolysis of the guanidinium [45]. By contrast, PAD2 employs a substrate-assisted mechanism [44] wherein the positive charge on the substrate guanidinium depresses the pK_a of the active site thiolate, Cys647 in PAD2, to facilitate deprotonation of the thiol. In addition to these mechanistic efforts, we also evaluated the calcium dependence of the PADs [17,21,44–46]. Notably, all four active PAD isozymes require high micromolar concentrations of calcium (0.1–1 mM) for full activity. Based on structural data, it is now known that PAD1 binds four calcium ions, PADs 3 and 4 bind five and PAD2 binds to six calcium ions [22,46–48].

To further understand how calcium binds and activates enzymatic activity, we determined high-resolution crystal structures of PAD2 in the absence and presence of increasing concentrations of calcium [46]. In the absence of calcium, PAD2 binds tightly to two calcium ions (Ca1 and Ca6). The Ca1 site is directly adjacent to the active site and helps position key catalytic residues including D351, H471 and D473. The Ca6 site is present far from the active site in immunoglobulin-like domain 2 (figure 3). When the concentration of calcium soaked into the crystals was increased, we were able to detect corresponding increases in the electron density at the sites that bind Ca3, Ca4 and Ca5. Plots of electron density versus calcium concentration were virtually identical indicating that calcium binding at these sites occurs cooperatively with a dissociation constant of approximately 200 µM. Amazingly, this value is nearly identical to that obtained for when activity is measured as a function of calcium concentration [46]. Further studies showed that calcium binding at the Ca3, Ca4 and Ca5 sites triggers a conformational change that generates the Ca2 site. Calcium binding to this site then triggers the movement of the catalytic cysteine into a position that is competent for catalysis [46].

Figure 3.

PAD2 undergoes a series of calcium dependent conformational changes. Wild-type PAD2 structures soaked with 0 mM (apoenzyme (a), PDB: 4N20) and 10 mM CaCl₂ (b), PDB: 4N2B) and the PAD2 F221/222A mutant soaked in 10 mM CaCl₂ (holoenzyme (c), PDB: 4N2C). (Online version in colour.)

5. Selective protein arginine deiminase inhibitors

The first PAD-selective inhibitor was developed for PAD4. This compound, threonine aspartate F-amidine (TDFA; figure 2a), was identified from a tripeptide-based library that incorporated a fluoroacetamidine warhead [49]. Despite being a tripeptide, TDFA shows cellular efficacy and more notably protects against injury in the lipopolysaccharide-induced model of acute lung injury [31]. Subsequently, GlaxoSmithKline reported the discovery of a benzimidazole-based series of compounds that selectively inhibit PAD4 [50]. Notably, the initial lead was identified from a DNA-encoded library and further elaborated to generate GSK484 (figure 2c) which inhibits PAD4 with an IC₅₀ of 50 nM [50]. We showed that inhibition is competitive with respect to calcium and that GSK484 is significantly less potent when the concentration of calcium is increased [50]. Structures of GSK484 bound to PAD4 confirmed our findings and showed that this compound preferentially binds the calcium free form of the enzyme at a site that is adjacent to the active site to allosterically inhibit PAD4 activity. GSK484 and GSK199, a less potent derivative, could inhibit neutrophil extracellular trap formation in human and mouse neutrophils [50,51]. Additionally, GSK199 showed efficacy in the collagen-induced model of arthritis [52]. These studies were highly significant because they showed for the first time that a PAD4 selective inhibitor could ameliorate disease severity. Notably, however, the decrease in disease severity was less than that observed for BB-Cl-amidine suggesting the possibility that additional isozymes played a role in disease. Several more advanced PAD4 selective inhibitors based on the GSK484 scaffold have been developed including BMS-P5 [53] and JBI-589 (figure 2c) [54]. Notably, BMS-P5 efficiently blocks NETosis and improves survival in a murine model of multiple myeloma [53]. JBI-589 has also shown efficacy in tumour metastasis models wherein this molecule reduces the growth of primary tumours and reduces lung metastases, in part by inhibiting NETosis and neutrophil chemotaxis [54].

We also reported selective inhibitors for PADs 1 and 2. These compounds include SM26 (figure 2a) which selectively inhibits PAD1 [55]. Notably, the potency and selectivity of this compound is owing in part to the formation of halogen bonds between the enzyme and PAD1 [55]. SM26 is highly potent in cells and inhibits histone H3 citrullination in HEK293T cells engineered to overexpress PAD1 as well as in mouse embryos where PAD1 inhibition blocks embryonic development at the morula stage. PAD1 knockout embryos are also arrested at this stage. We also reported several PAD2 selective inhibitors [56] including AFM30a (figure 2a) which Cayman Chemicals markets as CAY10723. This compound is highly selective for PAD2 over PAD4 (47-fold) and derives its potency via the formation of a hydrogen bond between the indole nitrogen in W347 and the lactam moiety. Selectivity is derived from the O-methyl group as well as methylation of the benzimidazole. AFM30a also inhibits PAD2 catalysed histone H3 citrullination (EC₅₀ = 400 nM). We also developed AFM32a, which is 95-fold selective for PAD2 over PAD4 [56]. This compound differs from AFM30a by the replacement of the O-methyl with an O-ethyl. Although AFM32a is approximately sixfold less potent than AFM30a in its ability to inhibit histone citrullination, AFM32a increases survival in multiple models of sepsis including the cecal ligation puncture model of septic shock and the lipopolysaccharide-induced model of endotoxic shock [30,57]. Notably, these effects mimic the effects observed in PAD2 knockout mice which show enhanced survival in these models. Moreover, it is important to recognize that PAD2 knockout attenuates vascular permeability and acute lung injury in the cecal ligation puncture model as well as inhibiting non-canonical pyroptosis [58]. These data suggest the therapeutic potential of targeting PAD2 in sepsis.

6. Chemoproteomic approaches

A key challenge in the field of protein citrullination has been the ability to detect this post-translational modification. Ground-breaking work led to the development of the COLDER assay which detects citrulline via a reaction between the urea and diacetylmonoxime in the presence of thiosemicarbazide and ammonium iron (II) sulfate under highly acidic conditions [59]. This assay is suitable for many inhibition experiments and to evaluate bulk effects on citrullination but offers little information on specific proteins that are citrullinated. In 1992, Senshu described the first antibody-based method to detect citrullinated proteins [60]. This antibody was raised by injecting rabbits with citrullinated histones that had been chemically derivatized with diacetyl monoxime and antipyrine under strongly acidic conditions. This anti-modified citrulline antibody is classically used by separating proteins on an SDS-PAGE, transferring the proteins to polyvinylidene difluoride membranes, and then modifying the proteins in situ with diacetyl monoxime and antipyrine. While this strategy works well, there have been issues with lot-to-lot variability in the performance of the antibody as well as challenges with uniform labelling of the proteins on the blot. In 2011, Moelants et al. generated an antibody to a 2,3-butanedione-modified citrulline which was used in a sandwich ELISA format to detect as low as 1 ng of citrullinated cytokines, with high specificity [61]. To generate additional reagents that could detect citrullinated proteins, Nicholas developed the F95 antibody [62]. This monoclonal antibody was generated by injecting mice with a citrullinated peptide consisting of 10 citrulline residues and a carboxyl Gly-Gly-Cys tripeptide. This antibody has been widely used and is now marketed by Millipore. Several protein-specific antibodies (e.g. anti-citrullinated histone H3) have also been generated and are widely used by the community.

In an effort to develop a chemical tool that could detect citrullinated proteins in an unbiased manner, we developed rhodamine-phenylglyoxal [63] (Rh-PG; figure 4a). The glyoxal group on this compound selectively reacts with the urea group on citrulline under acidic conditions (i.e. 20% trichloroacetic acid) to form an imidazalone ring structure (figure 4b) that irreversibly modifies a citrullinated protein. Note that a key technological challenge was the need to use reduced temperatures (i.e. 37°C) to facilitate protein modification, as higher temperatures lead to irreversible aggregation of the protein such that proteins could not be separated by SDS-PAGE. Rh-PG has been widely used to detect changes in protein citrullination and even monitor the kinetics of this process [63–66]. Moreover, Rh-PG was used to show that protein citrullination undergoes dynamic changes in a mouse model of colitis [63].

Figure 4.

Development of phenylglyoxal-based probes to visualize protein citrullination. (a) Chemical structures of rhodamine-phenylglyoxal (Rh-PG) and biotin-phenylglyoxal (biotin-PG). (b) Phenylglyoxal fragment preferentially reacts with citrulline (over arginine) at acidic pH to form an imidazalone ring structure. (Online version in colour.)

To further develop this technology and ease the identification of citrullinated proteins, we subsequently developed a biotinylated version of this compound, i.e. biotin-PG (figure 4a), which facilitates the enrichment of citrullinated proteins on streptavidin agarose [67]. Proteins can subsequently be identified by either Western blotting with an appropriate antibody or by tandem mass spectrometry. Tutteren et al. developed a similar compound that was used to enrich for citrullinated peptides [68,69]. A key advantage of our approach is that it readily enables protein identifications because multiple peptides are detected for any individual protein rather than just relying on a single citrullinated peptide. A key disadvantage of this approach, however, is that we do not get information on which sites are citrullinated. The lack of site-specific information primarily relates to the fact that in tandem mass spectrometry biotin fragmentation dominates over fragmentation of the peptide backbone making it challenging to identify the specific sites of citrullination [69]. We have recently overcome this issue by replacing the biotin with desthiobiotin.

Traditional proteomic approaches to identify citrullinated proteins have also been challenging in part because the mass change (i.e. 0.984 Da) is so small and identical to the mass change observed for a deamidated glutamine or asparagine. Consequently, incorrect assignments of citrullinated peptides can occur when a deamidated glutamine or asparagine is present in the peptide sequence. A tell tale sign of an incorrectly annotated citrulline is its presence at the C-terminus of the peptide. This is the case because the rate at which trypsin cleaves after a citrulline is markedly lower than the rate at which the protease cleaves after arginine. To aid the proteomic identification of citrullinated proteins, we and others have recommended that sites of citrullination be validated by detecting ions associated with the neutral loss of isocyanic acid which results in the generation of ions in the MS2 spectra that are 43.0058 Da smaller than the corresponding b or y ions [70–73]. We also recommend that researchers confirm that the monoisotopic species displays the correct + 0.984 mass shift [70,71]. To automate this process, we recently reported two open-source programs, i.e. ionFinder and envoMatch, that rapidly identify diagnostic neutral loss ions in Cit-containing peptides and match the isotopic envelopes of the monoisotopic species [70].

7. Biological effects of citrullination

In addition to its effects on gene transcription and skin biology, recent proteomic efforts have revealed several new roles for protein citrullination. For example, biotin-PG was used to identify the citrullinated proteins present in RA sera, synovial fluid and synovial tissue [74]. Notably, many of the citrullinated proteins are extracellular proteins, consistent with the fact that PAD activity is routinely detected in sera and synovial fluid of patients with RA. Among these proteins were several serine protease inhibitors (SERPINs) including antithrombin, antiplasmin and tissue plasminogen activator inhibitor [65,74]. SERPINs control several protease cascades involved in blood clotting and complement deposition by acting as pseudo substrates for their cognate proteases [75]. For example, thrombin will bind antithrombin wherein the active site serine attacks the peptide bond C-terminal to the ‘P1' position in the so-called reactive centre loop. The resulting tetrahedral intermediate collapses to form an acylenzyme intermediate with the concomitant loss of the C-terminus of the SERPIN. This reaction triggers the insertion of the reactive centre loop into the body of the SERPIN. This complex is quite stable and results in the irreversible inhibition of the protease [75]. Notably, in many SERPINs, the P1 residue is an arginine and we hypothesized that citrullination of this residue would block the ability of a SERPIN to inhibit its cognate protease. We subsequently confirmed this hypothesis by showing the citrullinated forms of antithrombin, antiplasmin and tissue plasminogen activator inhibitor could not inhibit thrombin, plasmin and tissue plasminogen activator, respectively [74]. We further demonstrated that citrullination of antithrombin and tissue plasminogen activator inhibitor decreased their binding to their cognate proteases [65]. By contrast, citrullination of antiplasmin converted it from an inhibitor to a substrate. To demonstrate the physiological relevance of this phenomenon, we evaluated antiplasmin and antithrombin citrullination in a mouse model of deep vein thrombosis and showed the citrullination of antithrombin is increased in the thrombus consistent with the increase in thrombin activity that is associated with the formation of a blood clot [65]. Wagner and colleagues also demonstrated a role for citrullination in blood clotting by showing that PAD4 could citrullinate and thereby inactivate ADAMTS13 [76]. ADAMTS13 is a metalloprotease that degrades von Willebrand factor (VWF) multimers. VWF binds to platelets and increases clot formation. Thus, inactivation of ADAMTS13 by PAD4 leads to an accumulation of VWF and an increase in clot formation. These studies were validated by directly injecting PAD4 into mice which leads to an increase in the half-life of VWF multimers and increased thrombosis [77]. Recent proteomic studies in HL60 cells stimulated with a calcium ionophore led to the identification of more than 14 000 citrullination sites [78]. Many of the citrullinated proteins are involved in RNA processing and splicing consistent with earlier findings by Lewallen et al. [67] suggesting a role for citrullination in RNA biology.

8. Site-specific incorporation of citrulline into proteins

Although tremendous progress has been made to characterize protein citrullination in vitro and in vivo, determining the specific effects of citrullination on the activity of an individual protein has been challenging. Traditional approaches have included enzymatic citrullination of specific proteins and evaluating the effects in vitro [64,65,74,79,80]. This approach has several limitations because the PADs lack a well-defined consensus sequence and can modify most arginine residues in unstructured regions. Consequently, multiple arginine residues are then modified in any one protein and there is no guarantee that stoichiometric citrullination will be achieved. Nevertheless, we and others have used this approach to generally gauge the effects of citrullination. For example, we used this approach to demonstrate that citrullination of nicotinamide N-methyltransferase promotes its unfolding leading to a loss in enzymatic activity [64]. Additionally, and as noted above, we used this approach to show that citrullination inactivates the protease inhibitory activity of the SERPINs [65,74].

Other routine approaches to study protein citrullination include the use of site-directed mutagenesis to study the effects of citrullinating an individual residue. In this case, we and others have incorporated glutamine in place of the citrullinated arginine residue [64,79]. The thought being that the neutral amide is a reasonable mimic of the urea group on citrulline. We used this approach to evaluate the effect of autocitrullination on PAD4 activity. Alternatively, one can generate a non-citrullinatable Arg to Lys mutant to evaluate the effect of blocking citrullination of a specific residue while maintaining positive charge. These approaches have several drawbacks, however, as the amide is not a perfect mimic of the urea and the side chain of glutamine is one methylene unit shorter. With respect to lysine, this residue is one methylene unit larger and lacks the ability to form bidentate delocalized electrostatic interactions with either an aspartate or a glutamate.

To overcome these issues, we developed a methodology [81] to site specifically insert citrulline into any position in a protein using a variant of the unnatural amino acid mutagenesis technology pioneered by Schultz and colleagues [82]. In our approach, we use a leucyl transfer RNA (tRNA) synthetase that recognizes a photocaged citrulline to charge a tRNA that recognizes the UAG stop codon in RNA. In this manner, the photocaged citrulline is incorporated whenever the ribosome encounters a UAG stop codon. Using this approach, we demonstrated that we can incorporate the photocaged citrulline into multiple proteins in mammalian cells and that the photocage is efficiently removed with 350 nm light. To demonstrate this methodology further, we evaluated the effect of citrullination on the catalytic activity of PAD4. We were motivated to study this problem because others had reported that autocitrullination of PAD4 leads to its inactivation [83]. By contrast, we had shown that there was no effect on the activity of PAD4 [79]. Although PAD4 autocitrullinates at multiple sites with most being surface expose residues, two residues that are often found to be citrullinated in proteomic studies, i.e. R372 and R374, are active site residues. Notably, R374 hydrogen bonds with the substrate backbone and R372 forms a stabilizing salt bridge with D345. Given the location of these resides, we expected that their citrullination would adversely impact PAD4 activity. Indeed, Gln and Lys mutants both perturb PAD4 activity [79]. However, as mentioned above, these residues are imperfect mimics of citrulline. Therefore, we incorporated citrulline in place of R372 and R374 and showed that citrullination of these positions led to a 180- and 9-fold reduction in catalytic activity, consistent with autocitrullination potentially inactivating the enzyme [81]. Given that these data are inconsistent with our own data showing that citrullination does not inactivate PAD4 activity, we sought to reconcile these findings by evaluating the stoichiometry of citrullination. For these studies, we used a quantitative proteomic approach to measure the abundance of citrullinated peptides as well as the loss of the corresponding arginine containing peptide [81]. Notably, we detected 18 arginine residues that were converted to citrulline. However, we did not observe complete conversion of any particular arginine residue and in most cases, we only saw a small loss in the arginine containing peptide, indicating that citrullination of any particular residues is far from stoichiometric. These data suggest that most autocitrullination events do not lead to enzyme inactivation consistent with our bulk autocitrullination experiments.

9. Conclusion and future perspectives

The field of protein citrullination has made considerable advances in the past 20 years and it is now clear that multiple PAD isozymes represent bona fide therapeutic targets for autoimmunity and cancer. For example, pan-PAD inhibitors have demonstrated remarkable efficacy in a range of animal models of inflammatory diseases including RA, lupus, multiple sclerosis, ulcerative colitis and type I diabetes. Moreover, inhibitors that selectively target an individual PAD isozyme have efficacy in a range of diseases including PAD4 selective inhibitors (e.g. GSK199) which shows efficacy in a murine model of RA and JBI-589 which reduces tumour growth and lung metastases in models of metastatic lung cancer. Additionally, PAD2 selective inhibitors improve survival in several models of sepsis, a disease for which no viable therapy exists.

A key question that is routinely asked is whether a PAD4 selective inhibitor is sufficient to treat RA in humans. Without performing specific clinical studies, it is difficult to definitively answer that question. Nevertheless, it is noteworthy that elevated levels of both PAD2 and PAD4 can be detected in the RA synovium suggesting that both enzymes contribute to disease pathology [84]. Moreover, our pan-PAD inhibitor BB-Cl-amidine displays greater efficacy than GSK199 in the collagen-induced model of RA suggesting that the inhibition of multiple isozymes (most likely PAD2 and PAD4) is required for optimal efficacy. PAD2/PAD4 double knockout mice have recently been described [85] and these mice will undoubtedly be useful for ascertaining the relative contributions of these enzymes to disease. It is also worth recalling that the relative contribution of each isozyme to a specific disease may differ. This is the case because PAD2 knockout mice or treatment with a PAD2-selective inhibitor promotes survival in sepsis models whereas PAD4 inhibitors or genetic knockout is not protective [57].

Another key question that remains outstanding is the relative merits of inhibiting intracellular versus extracellular PAD activity. The fact that the PADs are externalized during NETosis (and probably other forms of cell death) and remain active, strongly suggests that an extracellular inhibitor of PAD activity would be efficacious. Notably, AFM32a, which shows efficacy in sepsis models, showed relatively poorer cellular potency than AFM30a but was more potent in vivo, consistent with the preferential inhibition of extracellular PAD activity [56,57]. If extracellular PAD activity is the primary culprit, it should be readily possible to generate antibodies or small molecules that primarily act in serum. Finally, with the advent of advanced proteomic tools and the ability to site specifically insert citrulline into any mammalian protein, the future of the field is bright and much remains to be learn about how this post-translation modification contributes to normal human physiology and disease, especially with recent proteomic studies identifying 14 000 sites of citrullination in HL60 cells alone [78].

Data accessibility

This article has no additional data.

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

L.B.: writing—original draft, writing—review and editing; P.R.T.: supervision, writing—original draft, writing—review and editing.

Both authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

P.R.T. is a scientific founder of Danger Bio. P.R.T. co-founded Padlock Therapeutics which was acquired by Bristol-Myers Squib.

Funding

This work was supported in part by National Institutes of Health (grant no. R35 GM118112) to P.R.T.