2.1. Protein expression and purification

The synthetic gene of the recombinant human PAD6 (V2-P694 S10E S446E, Fig. S1 of the supporting information) with a TEV protease-cleavable 6×His-GST tag fused to its N-terminus was codon-optimized for mammalian expression and cloned into the pcDNA3.4 vector using TOPO cloning strategy (GenScript). The resulting construct was used for transient transfection in HEK cells using PEI-MAX (Sigma–Aldrich) and FreeStyle media (Gibco). To purify PAD6, the cell pellet was resuspended in lysis buffer containing 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM TCEP, 2 mM MgCl₂, 10 U ml⁻¹ benzonase and complete protease inhibitor tablets (Roche Applied Science). Cells were lysed by sonication, 5 cycles of 20 s on at 40% amplitude and 20 s off on ice, and the clear lysate was obtained by centrifugation. The His-GST tagged protein was bound to gluta­thione Sepharose 4FF resin (Cytiva) for 2 h at 4°C with gentle rocking rotation. PAD6 was washed with 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM TCEP and eluted with 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM TCEP, 10 mM reduced gluta­thione. The tagged protein was treated with TEV protease (with a ratio of 1 mg TEV per 20 mg PAD6) overnight at 4°C, and the untagged protein was purified with HisTrap FF (Cytiva) reverse nickel-affinity chromatography followed by size-exclusion chromatography on Superdex 200 with 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 1 mM TCEP before concentrating to 3.2 mg ml⁻¹.

The recombinant human PAD4 construct (Fig. S1), used for DSF and aSEC, was adapted from Muth et al. (2017). Briefly, the GST-PAD4 was expressed by autoinduction in ZYM-5052 media (Teknova) in BL21(DE3) pLysS over 18 h at 18°C. Cells were lysed in 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM TCEP, protease inhibitor, 10 U ml⁻¹ benzonase, 2 mM MgCl₂. The GST-tagged protein was bound to a GSTrap 4B column (Cytiva) and eluted with 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM TCEP, 10 mM reduced gluta­thione. The protein was further purified using a Resource Q column (Cytiva) before tag removal using HRV 3C protease (PreScission) incubation overnight at 4°C (with a ratio of 1 mg HRV 3C protease per 40 mg PAD4). The untagged protein was further purified by reverse GSTrap affinity chromatography followed by size-exclusion chromatography using a Superdex 200 with 20 mM Tris–HCl pH 7.5, 400 mM NaCl, 0.5 mM TCEP before concentrating to 3.1 mg ml⁻¹.

2.2. Analytical size-exclusion chromatography

The oligomerization of PAD6 in solution was examined by analytical size-exclusion chromatography on a Superdex 200 increase 5/150GL (Cytiva). The column was equilibrated with 20 mM Tris–HCl pH 7.5, 200 mM NaCl, 1 mM TCEP and calibrated using a set of molecular weight protein standards (Bio-Rad) composed of bovine thyroglobulin (670 kDa), bovine gamma globulin (158 kDa), chicken ovalbumin (44 kDa) and horse myoglobulin (17 kDa). The PAD6 sample (1 mg ml⁻¹) and the standards were all run at 0.3 ml min⁻¹.

2.3. Differential scanning fluorimetry

Differential scanning fluorimetry (DSF) was performed using SYPRO Orange (Sigma) as the shift reporter dye. Briefly, 4 µM of protein was incubated in buffer (10 mM Tris–HCl pH 7.5, 200 mM NaCl with and without 10 mM of CaCl₂) for 30 min on ice. SYPRO Orange dye (Sigma) was diluted to 2× final concentration from 5000× stock. The reactions were monitored in real time (Stratagene MX3005P; excitation, 490 nm; emission, 575 nm) from 25 to 95°C with a rate of change of 0.5°C min⁻¹.

2.4. Determination of PAD6 crystal structure

2.5. Detection of citrullinated proteins by 4-azido­phenyl glyoxal cyclo­addition on membranes (on-blot assay)

100 µg ml⁻¹ histone H3 (Abcam) or 20 µg ml⁻¹ cytokeratin 5 (Abcam) was incubated with 200 nM PAD6 or 100 nM PAD4 (Cayman) ± 50 µM GSK484 (Sigma) in citrullination buffer (50 mM Tris–HCl pH 7.6, 1 mM CaCl₂, 200 mM NaCl₂, 2 mM DTT) in a 384-well Nunc MaxiSorp plate (Thermo Scientific) for 3 h at 37°C and 300 r.p.m. shaking. The reaction was stopped by the addition of EGTA, pH 8.0, to a final concentration of 50 mM. The citrullination was detected using the alkyne-biotin based method as described by Hensen et al. (2015). The blots were scanned using the LiCor Odyssey CLx and quantification of protein bands was performed with the integrated Image Studio software.

2.6. Detection of citrullinated histone H3 by enzyme-linked immunosorbent assay

Enzyme-linked immunosorbent assay (ELISA) assay described by Verheul et al. (2018) was adapted as follows: 1 µg ml⁻¹ histone H3 (Abcam) was incubated with 200 nM PAD6 or 310 pM PAD4 (Cayman) ± 50 µM GSK484 (Sigma) in citrullination buffer (50 mM Tris–HCl pH 7.6, 1 mM CaCl₂, 200 mM NaCl and 2 mM DTT) in a 384-well Nunc MaxiSorp plate (Thermo Scientific) for 4 h at 37°C and 300 r.p.m. shaking. After overnight storage at 4°C, plates were blocked with 1% BSA in PBS for 2 h at 37°C and 300 r.p.m. shaking, followed by incubation with the primary antibody (anti-citrullinated H3: Cayman, No. 17939) diluted 1:500 in 1% BSA and 0.05% Tween 20 in PBS, and secondary antibody (HRP-conjugated anti-mouse: Sigma, No. A9044) diluted 1:10 000 in 1% BSA and 0.05% Tween 20 in PBS, for 1 h at 37°C and 300 r.p.m. shaking. After every incubation step the plate was washed 3× with 0.05% Tween 20 in PBS. Finally, the secondary antibody was detected through oxidation of 3,3′,5,5′-tetra­methyl­benzidine (Sigma), incubating the reaction for 20 min at 37°C and 300 r.p.m. shaking. The reaction was stopped by adding an equal volume of 0.5 M sulfuric acid. The absorbance at 450 nm was measured immediately after (Perkin Elmer Envision 2104).

3.1. Construct design of human PAD6

From the five human PAD homologs, four (PAD1 to PAD4) have been previously purified and structurally characterized by X-ray crystallography (Saijo et al., 2016; Slade et al., 2015; Rechiche et al., 2021; Funabashi et al., 2021; Arita et al., 2004). To better understand the relationship between the function and structure of human PAD6, we designed a full-length construct V2-P694 with a TEV-cleavable N-terminal 6×His-GST tag. Two phospho­rylation sites on Ser10 and Ser446 have been identified in human PAD6 (Rose et al., 2012). In order to understand the impact of these post-translational modifications on its activity, we designed two phospho­mimetic mutations, S10E and S446E, and included them in our final construct. The construct (V2-P694 S10E S446E) was purified to homogeneity, crystallized and characterized biochemically.

3.2. PAD6 shows no citrullination activity on proteins in vitro

Previous work suggests that PAD6 can citrullinate either histones or cytokeratins in oocytes, as indicated by assays using citrulline-specific antibodies and immunocytochemistry (Esposito et al., 2007). However, other studies using in vitro citrullination assays failed to demonstrate PAD6 activity (Raijmakers et al., 2007; Taki et al., 2011). To test the hypothesis that PAD6 is responsible for the citrullination of histone H3 or cytokeratin 5, both potential targets for PAD6 (Schwarz et al., 1995; Nachat et al., 2005; Esposito et al., 2007), we first analyzed the citrullination of histone H3 by an ELISA assay using a specific anti-citrullinated histone H3 antibody. PAD4, which was used as a positive control, clearly increased the citrullination signal for histone H3 [Fig. 1(a)]. The activity of PAD4 was blocked by the addition of the PAD4-specific inhibitor GSK484 (Lewis et al., 2015). In the presence of PAD6 carrying phospho­mimetic mutations S10E and S446E, the citrullination signal was similar to that of PAD4:GSK484, showing that no detectable citrullination activity could be observed with PAD6 for histone H3. Given the challenge to identify specific antibodies directed against additional citrullinated proteins of interest, we adopted an antibody-independent assay based on the on-blot technology (Hensen et al., 2015) which allowed us to test additional substrates, such as cytokeratin 5. Data obtained with this orthogonal method clearly confirmed the data obtained with the ELISA method showing the inability of phospho­mimetic PAD6 to citrullinate histone H3. Similar enzymatic citrullination inactivity was detected towards cytokeratin 5 as the substrate for PAD6 [Fig. 1(b) and Fig. S2]. Based on these data, we conclude that, unlike PAD4, PAD6 does not exhibit citrullination activity on the substrates tested in vitro.

Figure 1 In vitro citrullination assays of different substrates. (a) ELISA assay for measuring citrullination activity of PAD4 and PAD6 on the histone H3 substrate. PAD6 (blue) and PAD4 (grey) were tested for histone H3 citrullination activity in the presence and absence of 1 µg ml−1 histone H3 (two independent experiments with n = 3). PAD4 citrullination activity was inhibited by 50 µM GSK484 inhibitor (green). All data were normalized to the citrullination signal of the substrate in the absence of PAD. Error bars indicate the standard deviation. (b) On-blot assay for measuring citrullination activity of PAD6 and PAD4 on different substrates. Citrullination of two substrates (histone H3 and cytokeratin 5) was evaluated in presence of PAD6 (blue, n = 7 for histone H3, n = 3 for cytokeratin 5) or PAD4 (grey, n = 5 for histone H3, n = 2 for cytokeratin 5) and normalized to the citrullination signal of the substrate in absence of PAD. PAD4 activity was inhibited by 50 µM GSK484 inhibitor (green, n = 4 for histone H3, n = 1 for cytokeratin 5).

3.3. Tertiary and quaternary structure of PAD6

PAD6 crystallized in the C2 space group, and the structure was determined to a resolution of 1.7 Å (Table 1). A single chain is present in the asymmetric unit. The tertiary structure of PAD6 is typical of the PAD family and consists of two consecutive immunoglobulin-like domains (IgG1 and IgG2) followed by a α/β propeller C-terminal domain (Saijo et al., 2016; Slade et al., 2015; Funabashi et al., 2021; Arita et al., 2004). This C-terminal domain contains the catalytic citrullination site in other PADs (Mondal & Thompson, 2019) [Fig. 2(a)].

Figure 2 Overall structure of PAD6. (a) Domain organization of PAD6. The N-terminal IgG1, N-terminal IgG2 and the C-terminal catalytic domains are coloured in cyan, green and blue, respectively. (b) Model of the PAD6 dimer constructed with the symmetry mate of the crystal lattice. Overlay and comparison with the PAD4 dimer and r.m.s.d. calculated from PDB entry 2dew. (c) Analytical size-exclusion chromatography of PAD6 using a Superdex 200 5/150GL column equilibrated with 20 mM Tris pH 7.5, 200 mM NaCl, 1 mM TCEP. The elution volumes for PAD6 and bovine gamma globulin standard (158 kDa) were 1.7 ml for both samples. Inlet: standards in white circles and PAD6 in the black triangle. The molecular weight for PAD6 is estimated at ∼182 kDa. (d) Comparison between interacting residues involving the I-loops of PAD4 (orange) or that of PAD6 (blue). The I-loop is coloured red in PAD4 and purple in PAD6. In PAD6, the I-loop contains a disordered fragment (Pro445–Gly449) represented as a dashed line. The equivalent residues Tyr435 (PAD4) and Tyr444 (PAD6) are indicated with a star. Left panel: an overlay between PAD4 and PAD6 focused on the I-loop, showing different interface interactions involving Tyr435 (PAD4) and Tyr444 (PAD6). Right panels: detailed descriptions of the interactions involving Tyr435 in PAD4 (top) and those involving Tyr444 in PAD6 (bottom).

PAD4 was found to form a head-to-tail homodimer, with one monomer being related to another PAD molecule by a crystallographic twofold axis (Arita et al., 2004), and it has been shown that this dimerization is important for optimal activity (Liu et al., 2011; Lee et al., 2017). The crystal structures of PAD2 (Slade et al., 2015) and PAD3 (Funabashi et al., 2021; Rechiche et al., 2021) exhibit the same head-to-tail assembly. Examination of the PAD6 crystal packing showed that PAD6 adopts a nearly identical dimeric arrangement across the crystal packing, with a buried surface of 1986.5 Ų at the interface [Fig. S3(a)]. This value suggests that the two molecules are involved in a physiological dimerization rather than a crystal lattice (Janin & Chothia, 1990). Interestingly, the PAD6 assembly closely resembles the PAD4 dimer, since the superposition gives an r.m.s.d. of 2.0 Å with 982 aligned Cα [Fig. 2(b), Fig. S3(b)]. This small r.m.s.d. value is in the range of those obtained after pairwise superimpositions between the different dimeric assemblies of PAD isoenzymes (PAD2, PAD3, PAD4 and PAD6), indicating that this quaternary structure organization is well conserved in all these members of the PAD family. An analytical size-exclusion chromatography assay indicates that PAD6 elutes as a dimer in solution, with a molecular weight calculated at ∼182 kDa [Fig. 2(c)]. In addition, a chemical cross-linking experiment showed a unique band observed at ∼150 kDa, consistent with inter-molecular interactions among PAD6 molecules to form dimers [Fig. S3(c)]. Altogether, these data suggest that PAD6 most likely dimerizes in the same way as PAD4 and other dimeric PAD isoenzymes.

The dimeric interface is formed by residues from the three domains [Fig. S3(d)]. It has been shown in PAD4 that the hydro­phobic nature of several interfacial residues is important for the dimeric stability (Liu et al., 2011; Lee et al., 2017). Notably, hydro­phobic residues are also found at equivalent positions in PAD6. Among these, in PAD4, Tyr435 belongs to the `interface-loop' (I-loop) which critically influences both the dimeric stability and the catalytic activity (Lee et al., 2017). The Tyr435 residue is conserved in PAD6 (Tyr444). The structure reveals a different conformation and interfacial interactions to those in PAD4. This Tyr435 directly interacts with several residues from the facing monomers as it establishes hydrogen bonds with Glu281 and Tyr237 side chains and is involved in a network of hydro­phobic contacts involving Val200 and Leu272. However, in PAD6, the I-loop is partly disordered and Tyr444 interacts with Tyr561 of the same chain, which in turns establishes hydro­phobic contacts with Ile288–Pro289 from the facing monomer [Fig. 2(d)]. This shows that the dimer interface, while being overall well conserved, reveals subtle differences regarding the I-loop.

3.4. PAD6 is calcium-free

Previous studies on PAD4 revealed the capacity of the enzyme to bind five Ca²⁺ ions cooperatively to transition to the active conformation (Arita et al., 2004; Liu et al., 2011). The Ca1 and Ca2 are located in the C-terminal catalytic domain and Ca²⁺ binding at these sites is crucial to shape the substrate-binding site and assist the catalysis (Arita et al., 2004). Ca3–Ca5 are located further from the active site in the IgG2 domain. Although not essential for activity, Ca²⁺ binding at these sites enhances PAD4 catalytic efficiency (Liu et al., 2013). PAD1–3 bind calcium at equivalent sites to PAD4 except PAD1, which lacks the binding site for Ca5 (Saijo et al., 2016; Slade et al., 2015; Rechiche et al., 2021). The sequence examination shows that many of the acidic (seven Asp or Glu) and polar (one Asn) residues involved in the Ca²⁺ coordination in PAD4 are not conserved in PAD6 [Fig. 3(a)].

Figure 3 Comparing calcium binding of PAD6 and PAD4. (a) Sequence alignment of human PAD6 and other PADs focused on regions containing PAD4 residues involved in Ca1, Ca2, Ca3–5. The arrows point to residues involved in Ca2+ coordination in PAD4. The red arrows, with associated letters, highlight residues not conserved and unable to coordinate Ca2+ in PAD6. (b) Close-up views of PAD6 regions equivalent to Ca2+-binding sites in PAD4. Below, PAD4 Ca2+-binding sites are shown for reference (PDB entry 2dew). The yellow spheres represent the Ca2+ ions in PAD4. Residues that differ between the two proteins for which the side chain cannot coordinate Ca2+ in PAD6 are highlighted in red (equivalent positions between PAD6 and PAD4 are indicated with letters in brackets). (c) Thermal stability profiles of PAD4 and PAD6 in the presence of 0 or 10 mM CaCl2. The table provides the average protein melting temperatures (determined as the inflection point of the thermal transition) and the standard deviation from triplicate measurements.

Fig. 3(b) maps the corresponding residues in the PAD6 crystal structure. The regions equivalent to the five Ca²⁺-binding sites in PAD4 are more exposed to solvent than in Ca²⁺-bound PAD4, and they are occupied by water molecules. Within the region equivalent to PAD4 Ca3–Ca5, the dis­ordered segment 170–176 also illustrates its flexibility.

To support the sequence and structural analyses, we performed thermal shift assay to examine the effect of calcium to PAD4 and PAD6. We observed that the addition of 10 mM CaCl₂ significantly increases the melting temperature of PAD4 [Fig. 3(c)], from T_m = 46.8°C ± 0.7 to T_m = 69.6°C ± 0.2, indicating that the conformational stability of PAD4 is dependent on Ca²⁺ ions. Conversely, PAD6 was not thermostabilized in the presence of 10 mM CaCl₂. Additionally, isothermal titration calorimetry (ITC) experiments provide further evidence that PAD6 is unlikely to bind Ca²⁺ in contrast to PAD4 (Fig. S4). Altogether, the present analysis suggests that PAD6 is unlikely to coordinate calcium ions.

3.5. Inactive form of PAD6

In the current apo PAD6 structure, the loops surrounding the putative active site could be entirely traced from the electron density [Fig. S5(a)]. In the PAD6 sequence, Ala676 is found at a position equivalent to the active cysteine Cys645 in PAD4, which is conserved in all other PAD isoenzymes [Fig. 4(a)]. Thus, this position cannot serve as a potential catalysis of citrullination in PAD6. However, Ala676 is flanked by two cysteines (Cys675 and Cys677) that could potentially act as active residues. In the structure, neither Cys675 nor Cys677 are favourably positioned and oriented for effective reactivity with the substrate in PAD6, and the I661–A678 loop occludes access to either of these cysteines [Figs. 4(b) and 4(c)]. Fig. 4(c) illustrates the differences between PAD6 and the holo PAD4 structure in complex with a substrate, the histone H3 N-terminal tail in which Arg8 is the target for citrullination (PDB entry 2dew; Arita et al., 2006). In the current PAD6 structure, loop I661–A678 would sterically hinder substrate binding as seen in 2dew [Fig. 4(c), top right]. Additionally, while in the holo PAD4 structure the active cysteine (which is intentionally mutated to alanine in 2dew) is part of a small α-helix, Ala676 and Cys677 in PAD6 are part of a β-strand [Fig. 4(c), bottom right]. A B-factor analysis indicates that loop I661-A678 has low thermal motion compared with other regions of the structure, suggesting that its conformation is relatively stable [Fig. S5(b)]. In the apo PAD4 structure, (PDB entry 1wd8; Arita et al., 2004), the equivalent loop (I630-G646) is dis­ordered, as well as the surrounding loops. In the apo PAD2 structure (PDB entry4n20; Slade et al., 2015), the corresponding loop I635–G648 was modelled and exhibited a conformation characteristic of an inactive state, akin to what is observed in PAD6. Notably, in this loop the PAD2 active cysteine (Cys646) is not in a position suitable for catalytic citrullination [Fig. 4(d)]. Additionally, in this structure, Arg347 shields access to the catalytic centre formed by Asp351, His471 and Asp473 by occupying the substrate-binding cleft (Slade et al., 2015). In PAD6, the residue corresponding to Arg347 of PAD2, Arg355, adopts a similar position in the vicinity of Asp353, His480 and Asp482 [Fig. 4(e)]. The role of Arg355 as a pseudo-substrate is further supported by a direct interaction with Asp359. Interestingly, a non-productive form of Ca²⁺-bound PAD3 has been crystallized (PDB entry 7d8n; Funabashi et al., 2021), showing a similar configuration of the active site where the equivalent Arg346 shields access to the catalytic centre [Fig. 4(e)].

Figure 4 PAD6 structural analysis of the C-terminal domain. (a) Sequence alignment between five human PADs, focusing on the PAD6 loop I661–A678. Within this loop, the active or putative active cysteines are underlined and labelled: blue for PAD6, red for the other PADs. (b) Superimposition between structures of PAD6 and PAD4 in complex with substrate (PDB entry 2dew). The substrate is in yellow stick representation. (c) Same superimposition as in (b), showing a close-up view of the segment containing the I661–A678 loop and the β-strands connected by this loop, with the equivalent segment in PAD4 and the substrate. This view shows that the conformation of the PAD6 I661–A678 loop (blue) is different from the equivalent loop in PAD4 (I630–T647, red), and occludes the active site, so that the substrate would clash if positioned as in PAD4 (right panels). (d) PAD6 I661–A678 loop in blue superimposed with the equivalent segment of the apo PAD2 structure in green (PDB entry 4n20) and the inactive Ca2+-PAD3 in cyan (PDB entry 7d8n). Flexible loop I631–T648 (PAD3) represented by a dashed line. (e) Similarities in the active sites of the PAD6 structures and the apo PAD2 and non-productive form of Ca2+-bound PAD3, showing that equivalent Arg residues (Arg355 in PAD6, Arg347 in PAD2 and Arg346 in PAD3) occupy the active site, in lieu of the substrate. Ca2+ binding triggers the displacement of Arg347 in holo PAD2 (PDB entry 4n2c) together with other residues to shape a functional active site (right panel).

Altogether, our structure analysis shows that the PAD6 structure corresponds to an inactive form of PAD that resembles the inactive apo PAD2 structure. If PAD6 is to be activated to catalyze citrullination of substrates in the physiological environment, it may be achieved through a specific mechanism, independent to Ca²⁺. The activation of PAD6 would involve the displacement of the loop I661–A678 so that the substrate could access the active centre. In our structure, a PEG molecule from the crystallization solution was found in the vicinity of Cys675 and Cys677 [Fig. S5(c)], suggesting that this conformation is not rigid, and the putative active site can be accessed by a potential substrate.