Development of a selective inhibitor of Protein Arginine Deiminase 2

Abstract

Protein arginine deiminase 2 (PAD2) plays a key role in the onset and progression of multiple sclerosis, rheumatoid arthritis and breast cancer. To date, no PAD2-selective inhibitor has been developed. Such a compound will be critical for elucidating the biological roles of this isozyme and may ultimately be useful for treating specific diseases in which PAD2 activity is dysregulated. To achieve this goal, we synthesized a series of benzimidazole-based derivatives of Cl-amidine, hypothesizing that this scaffold would allow access to a series of PAD2-selective inhibitors with enhanced cellular efficacy. Herein, we demonstrate that substitutions at both the N-terminus and C-terminus of Cl-amidine result in >100-fold increases in PAD2 potency and selectivity (30a, 41a, and 49a) as well as cellular efficacy 30a. Notably, these compounds use the far less reactive fluoroacetamidine warhead. In total, we predict that 30a will be a critical tool for understanding cellular PAD2 function and sets the stage for treating diseases in which PAD2 activity is dysregulated.

Graphical abstract

Introduction

The protein arginine deiminase (PAD) family of enzymes, which catalyze protein citrullination (Figure 1), have garnered significant recent interest because aberrantly upregulated protein citrullination is associated with a variety of autoimmune diseases (e.g., rheumatoid arthritis (RA), multiple sclerosis, lupus, and ulcerative colitis) as well as certain cancers.^{1, 2} In addition to playing an important role in human pathology, the deiminase activity of the PADs regulates a diverse array of cellular processes including NET formation, the epigenetic control of gene transcription, differentiation, and the maintenance of pluripotency.^{1, 3–7} There are five PAD isozymes (PADs 1–4 and 6) that are uniquely distributed in various tissues.⁸ PADs 1–4 all possess deiminase activity, whereas PAD6 has a number of mutations that render it inactive.^{1, 9} The active PAD enzymes (i.e., PADs 1–4) are all regulated by calcium, wherein calcium binding triggers a conformational change that moves a nucleophilic cysteine residue into the active site, resulting in a >10,000-fold increase in PAD activity.^10–12

Figure 1.

Peptidyl-arginine (1) to peptidyl-citrulline (2) hydrolysis reaction catalyzed by the PADs.

PAD2, which is expressed in most tissue and cell types,⁸ contributes to the development of certain inflammatory diseases and cancers.^{1, 2} Specifically, PAD2 is upregulated in multiple sclerosis (MS) where it citrullinates myelin basic protein (MBP) leading to demyelination.^{13, 14} PAD2 is also released into the synovial fluid of arthritic joints where it citrullinates a number of extracellular targets, thereby promoting the development of anti-citrullinated antibodies (ACPA), which is a characteristic feature of RA.^{8, 15–18} Additionally, PAD2 is highly expressed in luminal breast cancers and its expression correlates with the levels of the HER2 protooncogene.^19–21 When PAD2 activity is inhibited in MCF10-DCIS cells, a breast cancer cell line, proliferation is decreased with a corresponding increase in apoptosis.¹⁹ This appears to be an epigenetic phenomenon as PAD2 is recruited to ERα promoters where it citrullinates histone H3 at R26 to trigger the localized decondensation of chromatin to facilitate ERα binding to its promoters to drive gene transcription.^{19, 20}

Given the strong evidence linking dysregulated PAD2 activity to MS, RA and breast cancer, we and others have focused on developing inhibitors targeting this enzyme, as well as PAD4, whose activity is also upregulated in a variety of inflammatory diseases including RA^{2, 22–24} and lupus.^25–27 Notably, our first-generation pan-PAD inhibitor Cl-amidine (3b, Figure 2) demonstrates efficacy in animal models of RA,²⁴ lupus,²⁷ ulcerative colitis,²⁸ spinal cord injury,²⁹ breast cancer,¹⁹ and atherosclerosis.³⁰ Despite these successes, many peptide-based inhibitors similar to Cl-amidine (3b) have displayed less than optimal potency in cellulo, which has been attributed to their poor metabolic stability and cell membrane permeability. In an effort to address these issues, the C-terminal carboxamide was replaced with a benzimidazole moiety to yield BB-Cl-amidine (4b, Figure 2).²⁷ This second-generation PAD inhibitor is at least 10-fold more potent than Cl-amidine (3b) in cells and animal models of lupus and RA, which represents a significant advance over Cl-amidine (3b).^{22, 23, 27}

Figure 2.

Structures and k_inact/K_I values for the 1^st-Generation PAD inhibitors F-amidine (3a) and Cl-amidine (3b) and the 2^nd-Generation PAD inhibitors BB-F-amidine (4a) and BB-Cl-amidine (4b).

Interestingly, the fluoroacetamidine counterparts of Cl-amidine (3a, F-amidine) and BB-Cl-amidine (4a, BB-F-amidine) have yet to find much utility in animal models as they display a marked decrease in PAD inhibition as compared to their chloroacetamidine analogs. This has been attributed to the less electrophilic nature of the fluoroacetamidine warhead, which reacts with an active site thiol conserved in all PAD isozymes.^{31, 32} It is also important to note that while BB-Cl-amidine (4b) brought a significant advance to PAD inhibitor development, as a pan PAD inhibitor,²⁷ it does not provide utility in investigating the roles of specific isozymes.

Prior to the development of BB-Cl-amidine (4b) advances had been made in improving isozyme-specific inhibition. Specifically, TDFA provided PAD4-selective inhibition³³, D-Cl-amidine provided PAD1-selective inhibition³⁴, while Cl4-amidine^{10, 31} and a series of hydantoin-based inhibitors^{35, 36} provided PAD3-selective inhibition. While these inhibitors have brought about significant advances in PAD inhibitor development, two main issues remain: 1) most of these inhibitors exhibit poor efficacy in cell-based assays (potentially due to metabolic instability and poor membrane permeability) and 2) a PAD2-selective inhibitor has yet to be developed. The deficiencies of current PAD inhibitors underscore a pressing need for PAD2-selective inhibitors with enhanced cellular efficacy. Herein, we report the development of benzimidazole-based PAD2-selective inhibitors that inhibit cellular PAD2 activity.

Results and Discussion

Inhibitor Design

Given the favorable cellular efficacy observed with BB-Cl-amidine (4b), we hypothesized that further elaboration of this scaffold would identify compounds with enhanced potency, selectivity and cellular efficacy. To aid these efforts we first generated the fluoro-containing analog, BB-F-amidine (Figure 2, 4a). We then determined the crystal structure of BB-F-amidine (4a) bound to PAD4 (Figure 3A). Similar to crystal structures of F-amidine,³¹ TDFA³³, and o-F-amidine³⁷ bound to PAD4, the amidine forms an intricate hydrogen bonding network with D350, H471, and D473. Interestingly, while R374 forms a hydrogen bond with the C-terminal carboxamide of the 1^st-Generation PAD inhibitors, it instead forms a cation-π interaction with the benzimidazole. Overlays of the PAD2 structure¹² onto the BB-F-amidine·PAD4 co-crystal structure (Figure 3B) shows a similar network of favorable interactions with D473, D351, and H471. These interactions properly orient the warhead for covalent modification by C647. By contrast to the PAD4 structure, R373, which corresponds to R374 in PAD4, orients away from the benzimidazole moiety such that it does not form the favorable cation-π interaction observed with PAD4.

Figure 3.

(A) Co-crystal structure of BB-F-amidine (4a) bound to PAD4 (PDB ID: 5N0M). (B) PAD2 (PDB ID: 4N2C)¹² was superimposed on the PAD4·BB-F-amidine (4a) co-crystal structure and then PAD4 was removed leaving BB-F-amidine. (C) Evolution of 2^nd-Generation PAD inhibitors BB-F-amidine (4a) and BB-Cl-amidine (4b) to 3^rd-Generation benzimidazole-based PAD inhibitors.

Ortho-carboxylate Series

Using these structures as a guide, we hypothesized that further elaboration of the benzimidazole scaffold could lead to compounds with enhanced potency. Given our previous demonstration that the installation of an ortho-carboxylate (o-F-amidine and o-Cl-amidine)³⁷ enhanced potency by ≥10-fold, due to the formation of a key hydrogen bond with W348 in PAD2 or W347 in PAD4,³⁷ we sought to combine the benzimidazole scaffold with the ortho-carboxylate functionality, expecting that the ortho-carboxylate would enhance in vitro potency and the benzimidazole would promote cellular uptake. This approach had previously been highly successful in developing highly potent tetrazole-based inhibitors.³⁸ Combining these two features (compounds 9a and 9b, Figure 4) increased PAD1 (18- and 1.8-fold, respectively) and PAD4 (6.5- and 2.3-fold, respectively) inhibition as compared to BB-F-amidine (4a) and BB-Cl-amidine (4b), however, PAD2 potency was relatively unaffected.

Figure 4.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 1 compounds 9–12. (B) Summary of isozyme selectivity for series 1 compounds 9-12 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

N-alkyl Series

We next explored other aspects of the benzimidazole scaffold to improve PAD2 potency and selectivity. Focusing on the nitrogen atoms of the benzimidazole scaffold, we noted that when BB-F-amidine (4a) is bound to PAD4, each nitrogen atom is involved in a water-mediated hydrogen bonding network; similar interactions are also observed with F-amidine (3a). While no crystal structure of BB-F-amidine (4a) bound to PAD1 exists, these water molecules are conserved in its structure. However, this hydrogen bonding network is absent in PAD2 and is replaced with a unique hydrophobic binding region created by V469, L467, L642, and F573 (Figure S1). This insight was intriguing and we hypothesized that N-alkylation of the benzimidazole could target this binding region for potent and selective PAD2 inhibition. N-alkylation would also increase the overall hydrophobicity of the inhibitor, which we expected to enhance cell permeability.

As discussed above, the ortho-carboxylic acid containing series (compounds 9a and 9b) improved PAD1 and PAD4 inhibition as compared to BB-Cl-amidine (4b) and BB-F-amidine (4a), but surprisingly did not improve PAD2 inhibition. By contrast, N-alkylation of the benzimidazole exhibited a marked increase in PAD2 inhibition and selectivity (Figure 4). Specifically, the N-ethyl derivative (11a) showed a 26-fold increase in PAD2 inhibition as compared to BB-F-amidine (4a) while also exhibiting 31-fold selectivity for PAD2 over PAD4 and 18-fold selectivity for PAD2 over PAD3. Also noteworthy is the fact that the fluoroacetamidine warhead (9a–12a) demonstrated similar or better PAD2 inhibition than its chloroacetamidine counterpart (9b–12b).

While this improvement in PAD2 potency and selectivity is not completely understood, the co-crystal structure of BB-F-amidine (4a) with PAD4 and the subsequent PAD2 overlay do provide some insight (Figure 3). These structures suggest that there is a key water-mediated hydrogen bond that exists for PAD4 and not PAD2 with the nitrogen atom of the benzimidazole moiety. Introduction of an N-alkyl group may disrupt this hydrogen bond causing a decrease in PAD4 inhibition. The PAD2 overlay also shows a hydrophobic binding region unique to PAD2 which the N-alkyl group could be oriented toward. This gained hydrophobic interaction could explain the improved PAD2 inhibition of compounds 10–12 with the N-ethyl substitution (11a) proving to be optimal for this binding region. The fluoroacetamidine warhead demonstrates similar (and in some instances improved) PAD2 inhibition as compared to the chloroacetamidine warhead potentially due to a smaller binding region near C647, as fluorine is smaller than chlorine. The differences in PAD2 selectivity, may also be attributed to the less electrophilic nature of the fluoroacetamidine warhead as compared to the chloroacetamidine warhead leading to a higher degree of specificity for the intended C647 target.

Lactam Series

Hypothesizing that the hydrophilic nature of the ortho-carboxylate might limit cellular uptake, we next replaced this moiety with a lactam, yielding series 2 compounds (14–17). We further hypothesized that the enhanced conformational constraint of this amide would further improve PAD2 inhibition. Rewardingly, introduction of the lactam moiety resulted in a ~7-fold increase in PAD2 inhibition for 14a as compared to both 9a and ~9-fold compared to BB-F-amidine (4a). By contrast, there was only a minor increase in PAD4 inhibition observed for 14a as compared to 9a (~1.4-fold). Similar to the series 1 compounds, N-alkylation of the benzimidazole caused a marked increase in PAD2 inhibition and selectivity. Specifically, the N-ethyl benzimidazole (16a) displayed the most potent PAD2 inhibition (k_inact/K_I = 61,600 M⁻¹ min⁻¹) utilizing the fluoroacetamidine warhead, representing a ~41-fold increase in potency compared to 9a and ~51-fold increase compared to BB-F-amidine (4a). Furthermore, 16a has the greatest degree of PAD2 selectivity (56-fold over PAD3 and ~33-fold over PAD4). The fluoroacetamidine warhead again proved superior with an optimal combination of PAD2 potency and selectivity.

A co-crystal structure of 14a bound to PAD4 was then determined (Figure 6A); we used PAD4 because efforts to co-crystallize PAD2 with 14a as well as other inhibitors proved unsuccessful. The PAD4•14a complex is highly similar to the PAD4•o-F-amidine structure (Figure 6B). Notably, these structures show that the carbonyl of either the ortho-carboxylic acid or the lactam form favorable hydrogen bonds with the indole nitrogen of W347. This interaction is likely a key determinant of the enhanced potency observed with this series. A potential hydrogen bond (either direct or water mediated) between R639 and the ortho-carboxylate or the nitrogen of the lactam may also contribute to the observed potency. Overlays of these PAD4 structures with PAD1³⁹ (Figures 6C–D) suggest similar hydrogen bonding networks with W349 and R348 also led to the marked increase in PAD1 inhibition observed for these compounds. However, modeling studies with PAD2¹² suggest that the increased potency for PAD2 is due solely to a hydrogen bond with W348 as an arginine residue corresponding to R639 in PAD4 is far removed from this binding region. The improved PAD2 inhibition for series 2 compounds (14–17) as compared to series 1 compounds (9–12) may be attributed to the replacement of the carboxylic acid with an amide functionality, the introduction of conformational constraint, or a synergistic effect of the two changes. Efforts to fully understand this are underway through the pursuit of a PAD2 co-crystal structure.

Figure 6.

(A) Co-crystal structure of AFM-14a bound to PAD4. (B) Co-crystal structure of o-F-amidine bound to PAD4³⁷ (PDB: 3B1T). (C) PAD1³⁹ (PDB: 5HP5) overlay with PAD4•AFM-14a co-crystal structure in PyMol. (D) PAD1³⁹ (PDB: 5HP5) overlay with PAD4•o-F-amidine co-crystal structure in PyMol. (E) PAD2¹² (PDB: 4N2C) overlay with PAD4•AFM-14a co-crystal structure in PyMol. (F) PAD2¹² (PDB: 4N2C) overlay with PAD4•o-F-amidine co-crystal structure in PyMol.

N-alkyl Lactams

To further optimize this scaffold, we next focused on N-alkylation of the lactam moiety (Figure 7) because the PAD4•14a co-crystal, as well as our PAD1 modeling studies, suggested that modification of this position would disrupt the hydrogen bonding network with R639 and with R348 in PAD1,³⁹ neither of which appear to aid in PAD2¹² binding. Furthermore, this substitution would further increase the overall hydrophobicity (19a ClogP = 1.61 and 14a ClogP = 1.00), which we expected to further improve the cell permeability.

Figure 7.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 3 compounds 19–22. (B) Summary of isozyme selectivity for series 3 compounds 19–22 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

Interestingly, this series of compounds (19–22) demonstrated similar PAD2 potencies and PAD2-selectivity as compared to series 2 compounds (14–17). Specifically, compounds 21a and 16a exhibited identical PAD2 vs PAD4 selectivity (33-fold), PAD2 vs PAD1 selectivity (~2-fold) and similar PAD2 vs PAD3 selectivity (>33-fold). While the N-ethyl lactam does not influence PAD2 potency or selectivity, it does improve the overall hydrophobicity of the compounds.

Alkoxy-substituted Benzimidazoles

Having demonstrated that the lactam enhanced potency and the N-ethyl benzimidazole modification enhanced PAD2 selectivity, we next explored substitutions of the 6-membered ring of the benzimidazole. Specifically, alkoxy substitutions were introduced to probe potential interactions with D351, R373 and H471 in PAD2, all of which surround the 6-membered ring of the benzimidazole.

This investigation began with a methoxy substitution at the 4-position giving series 4 compounds 29a and 29b which both showed improvements in PAD2 inhibition and selectivity over PAD4 as compared to their parent compounds 14a and 14b. This improvement was then coupled with N-alkylation of the benzimidazole to give series 4 compounds 30–31. Rewardingly, compound 30a (N-methyl benzimidazole) showed a dramatic increase in PAD2 inhibition (k_inact/K_I = 210,300 M⁻¹ min⁻¹) while also displaying very good selectivity for PAD2 (1.6-fold over PAD1, 47-fold over PAD3 and ~15-fold over PAD4). Interestingly, the N-ethyl benzimidazole (compound 31a) displayed a 4-fold decrease in PAD2 potency (compared to 30a) while improving PAD2 selectivity over PAD4 to 32-fold. This result suggests that introduction of the 4-methoxy group shifts the position of the benzimidazole so that the smaller N-methyl group would better access the unique hydrophobic binding region of PAD2 than the larger N-ethyl group. This is in contrast to series 1, 2, and 3 compounds which showed that the N-ethyl group was optimal at this position. Utilization of a 4-ethoxy group (32a and 32b) also showed excellent PAD2 inhibition, albeit ~2-fold less than 30a. However, 32a did demonstrate superior selectivity for PAD2 over PAD4 (95-fold) and PAD2 over PAD3 (79-fold).

Further SAR of the benzimidazole moiety then explored a 5-methoxy substitution. As seen in Figure 9, compounds 36a and 36b showed an increase in PAD2 inhibition compared to series 2 compounds 14a and 14b. Similar to series 4 compounds, N-methyl alkylation of the benzimidazole resulted in a significant increase in PAD2 inhibition (albeit ~2-fold lower than 30a) as well as excellent PAD2 selectivity over PAD4 (95-fold) and PAD3 (114-fold). The N-methyl substitution again displayed a greater degree of PAD2 inhibition and PAD2 selectivity than the N-ethyl substitution (38a and 38b).

Figure 9.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 5 compounds 36–38. (B) Summary of isozyme selectivity for series 5 compounds 36–38 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

In order to improve the hydrophobic nature of these compounds, we then reintroduced the N-ethyl lactam moiety in conjunction with the 4-alkoxy substitutions on the benzimidazole (30a ClogP=1.26 and 41a=1.62) to give series 6 compounds. Similar to 29a, compound 40a showed enhanced PAD2 potency and selectivity as compared to 14a (Figure 10). Series 6 compounds also demonstrated that the N-methyl benzimidazole substitution (41a) was optimal for PAD2 inhibition resulting in the most potent PAD2 inhibitor to date. Notably, the k_inact/K_I for this compound, which is 365,400 M⁻¹ min⁻¹, represents a >300-fold improvement over F-amidine (3a) and BB-F-amidine (4a). Compound 41a also demonstrated excellent selectivity for PAD2 over PAD4 (85-fold) and PAD3 (85-fold), while improving selectivity for PAD2 over PAD1 (~5-fold). The 4-methoxy benzimidazole substitution also proved superior over the 4-ethoxy and 5-methoxy substitutions utilized by series 6 compounds 43a and 43b (Figure 10) and series 7 compounds 45–47.

Figure 10.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 6 compounds 40–43. (B) Summary of isozyme selectivity for series 6 compounds 40–43 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

Series 6 compound 41a demonstrated that a combination of a 4-methoxy N-methyl benzimidazole with an N-ethyl lactam presented an ideal combination for potent and selective PAD2-inhibition. However, further investigation of lactam N-alkylation was needed to fully understand this phenomenon as well as finding a good balance between optimal potency and hydrophobicity. Therefore, series 8 compounds 49–51 were synthesized to explore N-alkylation of the lactam moiety. These data show that an N-methyl substitution of 49a was similar in PAD2 potency to the unsubstituted lactam (30a) while also exhibiting similar PAD2-selectivity to the N-ethyl lactam, 41a. N-isopropyl and N-cyclopropyl lactams (50a and 51a, respectively) also proved to be highly potent and selective PAD2 inhibitors, although neither were as potent and selective as 30a, 41a or 49a. Despite this, 50a and 51a are more hydrophobic than other lactam N-alkyl substitutions.

PAD1, PAD2 and PAD4 binding of 30a and 41a

Co-crystal structures of 30a and 41a bound to PAD4 were determined to shed light on how these compounds bind PAD4 (Figure 13). Overlays of these structures with those determined for PADs 1 and 2 further highlight common and unique interactions. For example, these compounds interact with PAD1 (D474, H472, W349, D352 and C645), PAD2 (D473, H471, D351, W348 and C647) and PAD4 (D473, H471, W347, D350 and C645) in a similar fashion to 14a. Close inspection of these structures reveal three main reasons for the PAD2-selectivity versus both PAD1 and PAD4. First, the N-methyl benzimidazole appears to orient into a unique hydrophobic binding region of PAD2, while simultaneously eliminating a water mediated hydrogen bonding network that is important for PAD4 binding. Second, the 4-methoxy group of the benzimidazole appears to be ideal for this binding region in PAD2 and tolerated for PAD1. However, the PAD4 cation-π with R374 is no longer optimal as the methoxy group forces the benzimidazole moiety to move and it is not able to properly interact with R374. Third, the N-ethyl lactam appears to cause R639 in PAD4 to orient away from the lactam functionality. Since the corresponding residue in PAD2 is not an arginine, this likely explains the higher selectivity for PAD2. Similarly, the enhanced PAD2-selectivity over PAD1 is likely explained by the fact that PAD1 does place an arginine residue close to this position. Despite these findings, selective inhibition of PAD2 over PAD1 proved difficult and will be a focus for future inhibitor development.

Figure 13.

(A) Co-crystal structure of AFM-30a bound to PAD4 (PDB: 5N0Y). (B) Co-crystal structure of AFM-41a bound to PAD4 (PDB: 5N0Z). (C) PAD1³⁹ (PDB: 5HP5) overlay with PAD4•AFM-30a co-crystal structure in PyMol. (D) PAD1³⁹ (PDB: 5HP5) overlay with PAD4•AFM-41a co-crystal structure in PyMol. (E) PAD2¹² (PDB: 4N2C) overlay with PAD4•AFM-30a co-crystal structure in PyMol. (F) PAD2¹² (PDB: 4N2C) overlay with PAD4•AFM-41a co-crystal structure in PyMol.

Target Engagement Assays

To measure target engagement, we used a stably transfected HEK293T cell line that overexpresses PAD2 (HEK293T/PAD2). Briefly, cells were incubated with an inhibitor of interest for 15 min, ionomycin (a calcium ionophore) for an additional 15 min, and calcium chloride was then added and the cells incubated for 1 h. Cells were then harvested, lysed, and cell lysates incubated with RIBFA, 55, a fluorescently tagged PAD probe (see Supporting Information) and calcium chloride for 1 h. Proteins were then resolved by SDS-PAGE and fluorescently tagged proteins imaged. For proof of concept, BB-Cl-amidine (4b) was first evaluated in this assay due to its high cell permeability and efficacy in other cell-based assays. As seen in Figure 14, BB-Cl-amidine (4b) was effective at entering the cell and covalently modifying PAD2, giving an EC₅₀ of 7.5 μM in this target engagement assay. Figure 14 also shows that BB-F-amidine (4a) was ~3-fold less potent in the target engagement assay than BB-Cl-amidine (4b), which was expected as it was less potent in vitro as well as less hydrophobic.

Figure 14.

(A) Target engagement assay in HEK293/PAD2 cells for BB-Cl-amidine (4b). (B) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by BB-Cl-amidine (4b). (C) Target engagement assay in HEK293/PAD2 cells for BB-F-amidine (4a). (D) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by BB-F-amidine (4a). Note: Panel C, BB-Cl-amidine (4b) was evaluated at 25 μM.

This target engagement assay was then used a primary screen to evaluate the cellular efficacy of compounds that showed a k_inact/K_I for PAD2 of ≥50,000 M⁻¹ min⁻¹. Specifically, the most potent PAD2 inhibitors (16a, 30a–32a, 37a, 41a, and 49a–51a) were first tested at 25 μM for their ability to enter HEK293T/PAD2 cells and covalent modify PAD2 (Figure S2). The unsubstituted lactams 30a, 31a, 32a, and 37a exhibited good potency in this assay with >60% occupancy at 25 μM, while 16a was roughly half as potent. Interestingly, N-alkyl lactams 41a, 49a, 50a, and 51a were all ~2-fold less potent than their unsubstituted counterparts (<37% occupancy). This was unexpected as the N-alkyl lactams displayed a similar or greater ability to inhibit PAD2 in vitro while also possessing a higher ClogP, or higher degree of hydrophobicity. These attributes were expected to enable better cell penetration and inactivation of PAD2. Mass spectrometry analysis of compound entering the cell showed that ~6-fold less of 41a was found internalized by the cell than unsubstituted lactams 30a, 31a, 32a, and 37a (see Supporting Information, Table S1). The exact reason for this difference is not completely understood but is an active area of investigation.

For the best compounds (30a, 31a, 32a, and 37a) we next obtained a 5-point dose response using our target engagement assay (Figure 15). Excitingly, these compounds all proved to be more potent than BB-F-amidine (4a) while also exhibiting similar potencies to BB-Cl-amidine (4b); 37a proved to be the most potent. Notably, however, BB-Cl-amidine (4b) was 25-35-fold better at gaining access to the cell than these four unsubstituted lactams (Supporting Information, Table S1). This apparent discrepancy is readily explained by the fact that 30a, 31a, 32a, and 37a are ~12–50-fold more potent PAD2 inhibitors than BB-Cl-amidine (4b) in vitro. Thus, the enhanced potency of these compounds overcomes their relatively poor ability to enter cells.

Figure 15.

(A) Target engagement assay in HEK293/PAD2 cells for 30a. (B) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by 30a. (C) Target engagement assay in HEK293/PAD2 cells for 31a. (D) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by 31a. (E) Target engagement assay in HEK293/PAD2 cells for 32a. (F) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by 32a. (G) Target engagement assay in HEK293/PAD2 cells for 37a. (H) Quantified results of target engagement assay plotted as percent occupancy of PAD2 by 37a. Note: BB-Cl-amidine (4b) was evaluated at 25 μM.

Histone H3 Citrullination in HEK293T/PAD2 Cells

Based on the promising results of our target engagement assay, BB-Cl-amidine (4b, Figure S3), 30a, 31a, 32a, and 37a were then evaluated for their ability to inhibit histone H3 citrullination in HEK293T/PAD2 cells (Figure 16). For these studies, we used a stably transfected HEK293T cell line that overexpresses PAD2 (HEK293T/PAD2). Cells were incubated with the inhibitor of interest, ionomycin and calcium chloride for 3 h. Cells were then lysed and proteins were resolved by SDS-PAGE and transferred to PVDF membrane. The membranes were then incubated with antibodies for histone H3 or citrullinated histone H3 (Cit 2,8,17), followed by incubation with the appropriate secondary antibodies and imaged by Licor analysis. Similar to the target engagement assay, BB-Cl-amidine (4b) strongly inhibited histone H3 citrullination with an EC₅₀ of 1.2 μM. Rewardingly, compounds 30a, 31a, 32a, and 37a showed similar efficacy to BB-Cl-amidine (4b) with 30a demonstrating the greatest ability to inhibit H3 citrullination (Figure 16A, EC₅₀=0.4 μM). Notably, while BB-Cl-amidine (4b) demonstrated similar efficacies to 30a, 31a, 32a, and 37a in both the target engagement and histone citrullination assays, these compounds were >30-fold less cytotoxic than BB-Cl-amidine (Table S1).

Figure 16.

(A) Histone H3 citrullination (CitH3) assay in HEK293/PAD2 cells for 30a. (B) Quantified results of target engagement assay plotted as percent relative intensity of CitH3 for 30a. (C) Histone H3 citrullination (CitH3) assay in HEK293/PAD2 cells for 31a. (D) Quantified results of target engagement assay plotted as percent relative intensity of CitH3 for 31a. (E) Histone H3 citrullination (CitH3) assay in HEK293/PAD2 cells for 32a. (F) Quantified results of target engagement assay plotted as percent relative intensity of CitH3 for 32a. (G) Histone H3 citrullination (CitH3) assay in HEK293/PAD2 cells for 37a. (H) Quantified results of target engagement assay plotted as percent relative intensity of CitH3 for 37a. Note: Lane 1 is HEK293T/PAD2 cells treated with DMSO only, lane 2 is HEK293T/PAD2 cells treated with DMSO and 10 μM ionomycin, and lanes 3–6 show HEK293T/PAD2 cells treated with increasing amounts of compound along with 10 μM ionomycin.

Conclusions

Motivated by the pathological roles of PAD2 in numerous inflammatory diseases and cancer, coupled with the dearth of highly potent and selective PAD2 inhibitors, we generated a detailed SAR of a benzimidazole-based scaffold to explore several key features for PAD2 inhibition, including: replacement of the N-terminal ortho-carboxylate with a lactam, N-alkylation of the lactam, N-alkylation of the benzimidazole, 4- and 5-alkoxy substitution of the benzimdazole, and the electrophilic warhead. While 30a, 41a and 49a were the three most potent PAD2 inhibitors that also exhibit excellent PAD2-selectivity (up to 106-fold), only 30a proved to be potent in multiple cell-based assays. This was surprising since 41a and 49a are more hydrophobic than 30a, which we expected to enhance cellular uptake. This discrepancy is partially explained by uptake studies showing that 6-fold more 30a was able to enter HEK293T cells than 41a.

These compounds represent a significant advance over BB-Cl-amidine (4b) due to their enhanced selectivity for a single PAD isozyme and their use of the less reactive fluoroacetamidine warhead, which significantly limits the number of off targets.³² Although PAD inhibitors have been developed to inhibit specific isozymes, most of these demonstrate poor efficacy in cell-based assays and none have shown specific and potent inhibition of PAD2. By contrast, 30a is both highly potent and selective for PAD2 while also demonstrating excellent efficacy in multiple PAD2 cell-based assays. Furthermore, 30a possesses a large therapeutic window as it is equipotent to BB-Cl-amidine (4b) in multiple cell-based assays while exhibiting >30-fold less non-specific toxicity. Ultimately, we expect that 30a will prove to be a powerful tool for validating the therapeutic relevance of inhibiting PAD2 in multiple disease states, i.e., breast cancer, multiple sclerosis and rheumatoid arthritis. Furthermore, this compound will enable studies to probe the role of this enzyme in numerous cellular pathways where PAD2 is presumed to play a role, including the epigenetic control of gene transcription.

EXPERIMENTAL SECTION

Chemistry

¹H NMR were recorded at either 400 MHz using a Bruker DRX-400 with H/C/P/F QNP gradient probe spectrometer or 500 MHz using a Bruker BioSpin 500MHz Avance AV-III Digital NMR Spectrometer and ¹³C NMR spectra were recorded at either 100 MHz or 125 MHz: Chemical shifts are reported in δ (ppm) relative to the internal chloroform-d (CDCl₃, 7.26 ppm) or methanol-d (CD₃OD, 3.31 ppm). ESI-HRMS signals were recorded with a Micromass Q-TQF I at the Mass Spectrometry Center at the University of South Carolina (Columbia, SC). The purity of all compounds was determined to be >95% by ¹H NMR and ¹³C NMR spectra. TLC was performed on glass backed silica get plates (Uniplate) with spots visualized by UV light. All solvents were reagent grade and, when necessary, were purified and dried by standard methods. Concentrations of solutions after reactions and extractions involved the use of a rotary evaporator operating at reduced pressure. Synthetic procedures and spectral characterizations of intermediates and final compounds are provided in the supporting information.

Proteins and Cell Culture

PADs 1, 2, 3 and 4 were purified as previously reported.^{10, 11, 40} HEK293T and HEK293T cells stably expressing human PAD2 (HEK293T/PAD2) were cultured as previously described.^{41, 42} For protein crystallography, recombinant GST-PAD4⁴³ was expressed in LEMO21 (DE3) cells overnight at 16 °C and cells were lysed in buffer (20mM Tris pH 8.1, 400 mM NaCl, 2 mM DTT, 5 mM EDTA, 20% glycerol, protease inhibitor, 0.5 mg/mL lysozyme, 5 U/mL benzonase) by sonication. Protein was incubated with glutathione sepharose resin and eluted with 10 mM glutathione. The GST tag was removed by cleavage with HRV 3C protease (PreScission Protease) and ion exchange with HiTrapQ FF column (protein eluted on gradient of 0.1–2 M NaCl). Pooled fractions were applied to a Superdex S200 26/60 column, equilibrated with buffer (20 mM Tris pH 8.1, 400 mM NaCl, 0.5 mM TCEP) before concentration to 4 mg/mL.⁴³

Crystallography

Native PAD4 crystals were grown by vapour diffusion at 10 °C by mixing equal volume of protein with precipitant (0.1 M imidazole pH 7.5, 0.2 M Li₂SO₄, 8–11% PEG3350). For soaking, crystals were transferred to soaking solution (0.1 M Imidazole pH 7.5, 0.2 M Li₂SO₄, 10% PEG3350, 1 mM compound (prepared at 100 mM stock in DMSO), 5 mM CaCl₂, 0.5 mM TCEP) and incubated overnight. Prior to freezing in liquid nitrogen, crystals were cryoprotected by brief transfer to a solution of mother liquor supplemented with 25% glycerol.

Data Collection

Data were collected on beamlines at the Diamond Light Source (Oxford, UK) and ESRF (Grenoble, France). Automated data processing was performed using Xia2^44–47 or GrenADES.^{46, 48, 49} Molecular replacement was performed using DIMPLE⁵⁰ (using PDB ID: 3B1U as a reference model) and refinement was performed with REFMAC5⁵¹ and coot.⁵² Compound dictionaries were generated using AFITT.⁵³ Refinement statistics are provided in Table S2.

Inactivation Kinetics

Inactivation kinetic parameters were obtained via established methods.^{31, 54} Briefly, values were determined by incubating PAD1, PAD2, PAD4 (2.0 μM) or PAD3 (5.0 μM) in a pre-warmed (10 min; 37 °C) inactivation mixture (50 mM HEPES, 10 mM CaCl₂, and 2 mM DTT, pH 7.6, with a final volume of 60 μL) containing various concentrations of inhibitor. Aliquots were removed at various time points and added to a pre-warmed (10 min, 37 °C) reaction mixture (50 mM HEPES, 10 mM CaCl₂, and 2 mM DTT, and 10 mM BAEE or 10 mM BAA in the case of PAD3; pH 7.6). After 15 min, reactions were quenched in liquid nitrogen and citrulline production was quantified using the COLDER assay.^{11, 55} Data were plotted as a function of time and fit to eq 1,

$$v=v_o{e}^{-kt}$$ eq 1

using Grafit version 5.0.11, where v is velocity, v_o is initial velocity, k (or k_obs) is the pseudo-first order rate constant of inactivation, and t is time. When saturation was reached upon plotting k_obs versus inactivator concentration, the data were fit to eq 2,

$$k_{\mathit{obs}}=k_{\text{inact}}\left[\mathrm{I}\right]/\left(K_{\mathrm{I}}+\left[\mathrm{I}\right]\right)$$ eq 2

using GraFit version 5.0.11, where k_inact corresponds to that maximal rate of inactivation and K_I is the concentration of inhibitor that gives half-maximal inactivation. If the plot of k_obs versus [I] was linear and did not saturate, then the value for k_inact/K_I equaled the slope of the line. All experiments were performed in at least duplicate. k_inact/K_I values are depicted graphically in the main text and summarized in Tables S3–S10 in the Supporting Information.

Target engagement assay with HEK293T/PAD2 cells

HEK293T cells stably expressing human PAD2 (HEK293T/PAD2) were cultured as previously described.^{41, 42} Cells were grown to ~80% confluence in a Nunc Cell Culture Treated EasYFlask™ T175 flask, trypsinized, and trypsin activity quenched with complete media. The cells were harvested by centrifugation at 1000 × g for 3.5 min, and resuspended in complete media. Cells were then added to two 6-well plates (2 mL/well) and incubated overnight. After overnight incubation, cells reached 80–90% confluence and were ready for the target engagement assay.

The complete media was removed from each well of the 6-well plate and carefully replaced with 1x HBS (2 mL/well) containing 2 mM CaCl₂ and incubated at 37 °C with 5% CO₂ for 15 min. Cells were then treated with either DMSO (1 μL) or varying concentrations of inhibitor (1 μL in DMSO) and incubated at 37 °C with 5% CO₂ for another 15 min. Cells were then treated with 1 μL of ionomycin (0.5 μM final concentration) and incubated at 37 °C with 5% CO₂ for 1 h.

A cell scraper was then used to harvest the cells, followed by centrifugation at 1000 × g for 3.5 min. The cell pellets were then washed 4x with 1x HBS, transferred to 1.6 mL tubes, and the HBS was removed. The cell pellets were then resuspended in 1x HBS with 1% Triton X-100 and incubated for 1 h at 0 °C. Lysates were cleared by centrifugation at 21,000 × g for 15 min. The soluble protein fraction was isolated and quantified by the DC assay (Bio-Rad). Lysates (2 μg/μL, 50 μL) were then treated with CaCl₂ (2 mM final concentration) and RIBFA (2 μM final concentration). The tubes were then incubated at 37 °C for 1 h followed by quenching with 6x SDS loading buffer and separated by SDS-PAGE (4–15% gradient gel). The bands were visualized by scanning the gel in a typhoon scanner (excitation/emission maxima ~546/579, respectively).

Histone H3 Citrullination in HEK293T/PAD2 cells

HEK293T cells stably expressing human PAD2 (HEK293T/PAD2) were cultured as previously described.^{41, 42} Cells were grown to ~80% confluence (8 × 10⁶ cells), trypsinized, and quenched with complete media. The cells were harvested by centrifugation at 1000 × g for 3.5 min and washed 4× with 1× HBS. Cells were resuspended in 1x HBS at 8 × 10⁶ cells/mL, and 4 × 10⁵ cells were added to 0.65 mL tubes for the subsequent assay. Ionomycin (10 μM), CaCl₂ (1 mM) and either DMSO or varying concentrations of inhibitor were added to the cells and incubated for 3 h. The final concentration of DMSO was 1% in each sample.

Triton X-100 was then added to the cell suspensions (1% final concentration) and sonicated at 0 °C for 1 h. Lysates were cleared by centrifugation at 21,000 × g for 15 min, and soluble proteins were removed and quantified by DC assay (Bio-Rad). Lysates were normalized and 1 μg of protein was separated (4–15% gradient gel) and transferred to PVDF membranes (Bio-Rad) at 80 V for 60 min. Membranes were then blocked with TBST and BSA (5%) for 1 h at 23 °C. Blocked membranes were incubated with antibodies for histone H3 (1:2000) or histone H3 Cit 2,8,17 (1:1000) in TBST with 5% BSA for 12 h at 4 °C. Membranes were then washed with TBST (9x) and incubated with anti-Rabbit IgG Licor conjugate (1:5000) for 1 h at 23 °C. Membranes were washed with TBST (9x) and imaged by Licor analysis.

Cellular Uptake Assays

HEK293T/PAD2 cells stably expressing PAD2 were cultured and treated with various inhibitors similarly to the target engagement assay described above. After the cells were treated with CaCl₂ (2 mM), ionomycin (0.5 μM) and various inhibitors (25 μM), a cell scraper was then used to harvest the cells, followed by centrifugation at 1000 × g for 3.5 min. The cell pellets were then washed 4x with 1× HBS, transferred to 1.6 mL tubes, and the HBS was removed. The cell pellets were then resuspended in acetonitrile and sonicated for 1 h at 0 °C. Lysates were cleared by centrifugation at 21,000 × g for 15 min. The soluble protein fraction was isolated, filtered, and injected onto the LCMS. The area under the curve was then taken to be the ‘area of analyte’ as described in Table S1. The retention time and ‘area of standard’ were both determined by injecting 25 μM of inhibitor directly onto the LCMS and integrating the area under the curve.

Cytotoxicity Studies

Human HEK293T cells were plated (2.5 × 10⁶ cells/well) in a 96-well plate. The next day the cells were treated with DMSO (1 μL), Triton X-100 (1 μL), or various concentrations of BB-F-amidine (4a), BB-Cl-amidine (4b), 30a, 31a, 32a, 37a or 41a and incubated for 24 h. Cell viability was measured using the XTT reagent kit (ATCC) by reading the absorption at 475 nm using a Spectramax plate reader. EC₅₀ values for cell growth inhibition were determined by fitting an eight-point dose-response curve to eq 3,

$$\mathrm{Y}=\text{Range}/{\left(1+\left(\left[\mathrm{I}\right]/{\mathrm{EC}}_{50}\right)\right)}^{\mathrm{s}}+\text{background}$$ eq 3

using GraFit 5.0.11, where range is the uninhibited value minus the background and s is the slope factor.

Figure 5.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 2 compounds 14–17. (B) Summary of isozyme selectivity for series 2 compounds 14–17 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

Figure 8.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 4 compounds 29–32. (B) Summary of isozyme selectivity for series 4 compounds 29–32 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

Figure 11.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 7 compounds 45–47. (B) Summary of isozyme selectivity for series 1 compounds 45–47 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

Figure 12.

(A) k_inact/K_I values for BB-F-amidine (4a), BB-Cl-amidine (4b), and series 8 compounds 45–47. (B) Summary of isozyme selectivity for series 1 compounds 45–47 based on their respective k_inact/K_I ratios setting the enzyme with the lowest value as 1.

ABBREVIATIONS USED

RA

Rheumatoid arthritis

PAD

Protein arginine deiminase

ACPA

Anti-citrullinated protein antibodies

NET

Neutrophil extracellular trap

RFA

Rhodamine-conjugated F-amidine

TAMRA-N₃

Tetramethylrhodamine azide

Fmoc

Fluorenylmethyloxycarbonyl

Orn

Ornithine

Boc

tert-butyloxycarbonyl

HBTU

(2-(1H-benzotriazol-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate

HOBt

Hydroxybenzotriazole

AcOH

Acetic acid

DIPEA

N,N-Diisopropylethylamine

DMF

Dimethylformamide

HCl

Hydrochloric acid

Et₂O

Diethyl ether

MeOH

Methanol

TEA

Triethylamine

SDS-PAGE

Sodium dodecyl sulfate polyacrylamide gel electrophoresis

kDa

Kilodalton

MBP

myelin basic protein

MS

multiple sclerosis

TDFA

threonine-aspartic acid-F-amidine

HEK293T

human embryonic kidney 293 cells

PBS

phosphate buffered saline

HBS

HEPES buffered saline

PVDF

polyvinylidene fluorine

DMSO

dimethyl sulfoxide

TBTA

Tris[(1-benzyl-1H-1,2,3-triazol-4-yl)methyl]amine