Structures of prokaryotic ubiquitin-like protein Pup in complex with depupylase Dop reveal the mechanism of catalytic phosphate formation

Pupylation is the post-translational modification of lysine side chains with prokaryotic ubiquitin-like protein (Pup) that targets proteins for proteasomal degradation in mycobacteria and other members of Actinobacteria. Pup ligase PafA and depupylase Dop are the two enzymes acting in this pathway. Although they share close structural and sequence homology indicative of a common evolutionary origin, they catalyze opposing reactions. Here, we report a series of high-resolution crystal structures of Dop in different functional states along the reaction pathway, including Pup-bound states in distinct conformations. In combination with biochemical analysis, the structures explain the role of the C-terminal residue of Pup in ATP hydrolysis, the process that generates the catalytic phosphate in the active site, and suggest a role for the Dop-loop as an allosteric sensor for Pup-binding and ATP cleavage.

M ycobacteria and other actinobacteria evolved a posttranslational protein modification pathway termed pupylation that is functionally related to eukaryotic ubiquitination and can target proteins to a bacterial proteasome complex for degradation [1][2][3] . Mycobacterium tuberculosis (Mtb), one of the deadliest human pathogens to date, benefits from this pathway for its survival inside the host [4][5][6] . Protein quality control mechanisms and protein degradation pathways have recently garnered high interest as targets for the development of new drugs against increasingly drug-resistant Mtb strains that cause untreatable tuberculosis infections 7,8 . An understanding of the regulation of pupylation on the molecular level is therefore not only of fundamental scientific interest but also of potential medical relevance for drug development to interfere with this bacterium's survival in the host.
During the process of pupylation, prokaryotic ubiquitin-like protein (Pup) is covalently attached via the γ-carboxylate of its C-terminal glutamate to a lysine side chain in the target protein by the formation of an isopeptide bond 9,10 . While the bacterial proteasome is homologous to the eukaryotic proteasome and was likely obtained by horizontal gene transfer 11,12 , the modification enzymes that catalyze pupylation and depupylation evolved from an ancient bacterial glutamine synthetase-like enzyme 13 . A single Pup ligase enzyme, PafA (proteasome accessory factor A), is responsible for modifying all pupylation substrate proteins 3,9 and likewise a single depupylase enzyme, Dop (deamidase of Pup), removes Pup from Pup-protein adducts 10,14,15 . In some actinobacterial species including Mtb, where Pup is encoded with a C-terminal glutamine (PupQ), Dop also acts as deamidase to generate ligation-competent Pup with a C-terminal glutamate (PupE) 9 . The ligase and depupylase enzymes are close sequence and structural homologs, featuring a large N-terminal domain that contains the active site and is homologous to glutamine synthetase (GS) and other γ-glutamyl-amine ligases 16 . A small C-terminal domain unique to the pupylation enzymes lies adjacent to the active site and closes it off on one side 16 . The active sites of Dop and PafA feature a curved anti-parallel β-sheet cradle with nucleotide bound at one end of the cradle and the binding site for the C-terminal glutamate residue of Pup located at the other end 16,17 . The phosphate chain of the nucleotide runs along the β-strands of the cradle towards the glutamate 17 .
In PafA, ATP is required for activation of the side-chain carboxylate of Pup's C-terminal glutamate by transferring the γ-phosphate to Pup, forming a mixed glutamyl-phosphate anhydride and turning over ATP to ADP in the process 18 . The carboxyl-carbon of the phospho-Pup intermediate is then attacked by the amino group of a lysine side-chain in the target protein to form the isopeptide bond 18 . ATP turnover in PafA is therefore part of the catalytic cycle and is stoichiometric with substrate turnover. Dop on the other hand does not turn over ATP stoichiometrically with substrate 9,19 . Rather, before the catalytic cycle of Dop can begin, Dop-mediated cleavage of ATP in the active site must take place to generate the Pi species important for catalysis 19 . The turnover of ATP is therefore stoichiometric with Dop active sites rather than substrate turnover, and ATP hydrolysis is not part of the depupylation/deamidation reaction cycle but serves to generate the active site Pi species. Orthophosphate remains in the active site for multiple rounds of catalysis until it eventually dissociates and ATP must rebind 19 . We furthermore demonstrated that ATP hydrolysis in the active site of Dop is dependent on the C-terminal residue of Pup, be it glutamate or glutamine 19 . We previously solved a crystal structure of Dop from Acidothermus cellulolyticus ( Acel Dop) with ADP and Pi bound in the active site 19 (PDB 5LRT), but a structure of Dop in complex with Pup has not been determined so far, and the mechanistic role played by the C-terminal residue of Pup in ATP hydrolysis is not known.
The unique features of the two homologous enzymes, PafA and Dop, that allow them to catalyze opposing reactions in the pupylation-depupylation cycle are still not fully understood. A recent study demonstrated that multiple mutational paths likely led to the emergence of the depupylase from the ligase activity, and that this path cannot easily be reenacted in vitro by reciprocal mutagenesis of uniquely conserved residues 20 .
One characteristic feature of the depupylase Dop that is absent in the homologous PafA ligase is a highly conserved sequence stretch of 40-65 residues preceding the β2 strand of the β-sheet cradle, which has been termed the Dop-loop 16 . Although it was shown that deletion of this region does not abolish Dop activity 16 , the high degree of sequence conservation in the first half of the loop would suggest that it plays a functional role. In the Dop structures solved so far, the Dop-loop could not be resolved, indicating that it was disordered 16,19 . However, it is possible that the Dop-loop undergoes a disorder-to-order transition during the catalytic cycle or under specific conditions. For example, the intrinsically disordered Pup undergoes such a disorder-to-order transition, forming two orthogonal helices in its last 27 residues upon binding to a long groove on PafA 17 . A homologous groove is also present in Dop. Differences in the shape of the groove and residues lining it as well as the fact that Pup binds to Dop an order of magnitude more tightly than to PafA 16 indicate that Pup and Dop must exhibit some distinct interaction features.
In this study, we present a series of crystal structures of the depupylase Dop in complex with Pup in different intermediate states along the catalytic cycle. In combination with biochemical experiments, our analysis reveals the mechanism of ATP hydrolysis that is required to generate the catalytic phosphate in the active site, and provides insights into the allosteric regulatory role of the Dop-loop on Pup-binding and ATP hydrolysis.

Results and discussion
The Pup C-terminal residue forms extensive contacts in the Dop active site. In order to deduce the catalytic mechanism of the depupylase enzyme, a structure of Dop in complex with Pup is crucial. This is particularly important, since the binding of Pup to Dop stimulates the generation of the inorganic phosphate that must be present in the active site for catalysis 19 . As previous experiments have shown that the N-terminal portion of Pup does not participate in binding to Dop 16 , an N-terminally truncated Pup-fragment, Acel PupQ ΔN43 , was used for co-crystallization experiments with Acel Dop. Furthermore, in order to prevent the turnover of PupQ to PupE, the non-hydrolyzable ATP analog AMP-PCP was used, since ATPγS and AMP-PNP can still be hydrolyzed by Dop, albeit much more slowly. We obtained crystals of the enzyme in complex with the PupQ fragment and AMP-PCP and determined the structure at 1.65 Å resolution by molecular replacement (Fig. 1a; Table 1). The final atomic model contained all residues of PupQ ΔN43 bound to the enzyme (Supplementary Fig. 1).
Upon binding to Dop, Pup undergoes a disorder-to-order transition adopting two well-resolved helices (helix 1: D44-V53; helix 2: N57-S64) (Fig. 1a, b). Comparison with the Pup-ligase complex ( Cglu PafA D64N -Cglu PupE ΔN37 -ATP, PDB 4BJR) 17 shows that the overall conformation of Pup in the enzyme binding groove is conserved (Fig. 1b). However, Pup in complex with Dop exhibits a significantly more extensive network of interactions than Pup bound to the ligase PafA, with differences especially pronounced at the C-terminal residue of Pup (Fig. 1c, d). The C-terminal glutamine residue of Pup in the Dop-Pup complex is well defined in the electron density map (Supplementary Fig. 1) and forms extensive interactions with Dop ( Fig. 1d, upper panel), whereas in the PafA-Pup complex, the C-terminus of Pup is poorly ordered, as indicated by the weak electron density 17 , and only weakly forms a salt bridge with the side chain of R201 (Fig. 1d, lower panel).
Depupylase Dop is structurally homologous to Pup ligase PafA, although the two evolutionarily related enzymes catalyze opposing reactions. It is possible that the tighter binding of the C-terminal Pup residue to the Dop active site compared to PafA contributes to the opposite directionality of the reaction catalyzed by Dop. A similar phenomenon was observed for the regulation of phosphoribosyl-linked (PR) serine ubiquitination by SidE (ubiquitinase) and DupA/B (deubiquitinase), which have catalytic domains that are highly homologous to each other yet catalyze chemically opposite reactions 21 . The deubiquitinase activity of DupA/B is favored by the high affinity of the PR-ubiquitinated substrate, whereas the ubiquitination activity of SidE is conferred by low-affinity interactions with ubiquitin. DupA can be converted into a SidE-type ubiquitin ligase by weakening the binding affinity of the ubiquitinated peptide. The Dop-Pup complex structure solved here shows that the loop preceding strand β7, referred to in the literature as α-loop (F208-K220) 20 , is involved in the binding of the C-terminal residue of Pup (Q71) through a strong hydrogen bond formed with T217 (Fig. 1d, upper panel). In comparison, the equivalent region (H188-S200) in the PafA-Pup complex structure (PDB 4BJR) forms an α-helix which does not interact with Pup 17 . Substitution of the α-loop with the equivalent region from PafA was reported to contribute to the conversion of the depupylase into a ligase 20 , thus supporting our hypothesis that the observed network of interactions constraining the C-terminal residue of Pup in the active site contributes importantly to the directionality of the catalyzed reaction.
Allosteric interplay between Pup-binding and the Dop-loop. The sequence stretch preceding strand β2 and referred to as the Dop-loop is a distinguishing feature between the depupylase and the Pup ligase, since it is present in all Dop orthologs but absent in all PafA homologs 16 . In previously published Dop structures, the Dop loop was not resolved, however, given its high degree of conservation ( Supplementary Fig. 2a Supplementary Fig. 2b), including residues P40-R65 and L73-L80 and thereby all of the highly conserved residues in the loop ( Supplementary Fig. 2a). The Dop-loop features two very short helical stretches (DLη1 and DLα1) flanking a long zigzagging loop (Fig. 2a). Helix DLη1 (η stands for 3 10 helix) consists of residues 48-52 (DFEEE) and is located nearby the nucleotide-binding pocket (NBP) of Dop, while DLα1 is formed by residues 75-78 (DEDL) and is positioned distal to the active site. The Dop-loop forms extensive interactions with residues surrounding the active site ( Fig. 2b, c, upper panel). In particular, insertion of the highly conserved Dop-loop-residue W47 into the NBP leads to the formation of a triple π-stack, where W47 interacts with H231 that in turn stacks with W453 of the C-terminal domain (Fig. 2b). Another π-stacking interaction is formed by Dop-loop-residue F49 and H155. Furthermore, a network of salt bridges is formed between acidic residues in the Dop-loop and several arginine side chains surrounding the active site.

S L L D E I D G L L E N N A E E F V R S Y V Q K G G E D A I L D E I D D V L E E N A E E F V R S Y I Q K G G
In contrast, in the Pup-bound structure described in the previous section and in the Pup-bound structures described in later sections of this article, which we obtained in crystallization   efforts with metal fluorides, the Dop-loop is mostly disordered except for the highly conserved residue W47 and six residues flanking it ( Supplementary Fig. 3b). Interestingly, in the Pupbound structures, W47 stacks with the side chain of R90 and forms a hydrophobic interaction with the conserved W453 residue of the C-terminal domain (Fig. 2c, lower panel, and 2d; Supplementary Fig. 2a), indicating that Pup-binding significantly reorders the Dop-loop. Superimposition of the Dop-loop-inserted structure with any of the Pup-bound structures reveals dramatic conformational changes of the Dop-loop and several active site residues. Although the conformation of the visible stretch of the Dop-loop is congruent in all the Pup-bound structures we solved, the Pup-bound complex with ADP-MgF 3 (H 2 O) − in the active site, which is discussed more extensively in a later paragraph of this article, is used here for superimposition due to the fact that it exhibits the highest quality map for the partially resolved Doploop ( Fig. 2e; Supplementary Fig. 3b). The comparison shows that, upon binding of Pup, H95 swings upward by about 90°out of the way of Pup's C-terminal residue and displaces the Doploop, including residue W47 that moves around 11 Å away from the NBP. To complement our crystallographic study with biochemical analysis, we generated a Dop-loop variant of Dop from Corynebacterium glutamicum ( Cglu Dop), referred to as Cglu DopGS, where the Dop-loop (H43-I77) was replaced with a short linker sequence (GS) 4 . Cglu Dop was chosen for the in vitro biochemical and mutational analysis due to its higher solubility compared to the mycobacterial orthologs and because C. glutamicum, in contrast to A. cellulolyticus, is not a thermophile. Using isothermal titration calorimetry (ITC) binding experiments, we show that PupE binds to the DopGS variant with an affinity~2.5 times tighter than to the wild type (WT) enzyme ( Supplementary Fig. 3c and 3d), supporting the notion that Pup binding is coupled to the displacement of the Dop-loop.

e P A Q R Q A R W D F E E E N P L R D A R G F E V A R E A A D P S Q L T D E D L G L
The Dop-loop affects ATP hydrolysis allosterically. When Pup is bound to the active site, the highly conserved Dop-loop residue W47 stacks with the side chain of R90 via a cation-π interaction ( Supplementary Fig. 4a). R90 in turn stacks with the side chain of Y92, stabilizing the binding of nucleotide in the active site of Dop through a salt bridge with the α-phosphate group. In this conformation W47 influences the positioning of nucleotide in the active site and could thereby affect the ATP hydrolysis that generates the catalytic phosphate and must take place before the catalytic cycle of Dop can begin.
Using the fluorescent model substrate Pup-(5-FAM-Lys) (Pup-Fl) it is possible to follow the isopeptidase activity of Dop by measuring the resulting decrease in fluorescence anisotropy 22 . In the presence of ATP but not ADP/Pi, Cglu DopWT exhibits a lag phase that was previously shown to correlate with the production of ADP and Pi in the active site 19 (Fig. 3a). The DopGS variant exhibits a prolonged lag phase, suggesting an allosteric influence of the Dop-loop on ATP hydrolysis. Pup-Fl turnover catalyzed by the DopGS variant is slower than the turnover catalyzed by WT Dop, as is evident from the steeper slope in the anisotropy time course of the WT enzyme. In order to determine the effect of a missing Dop-loop on the steady-state parameters of isopeptide bond cleavage by Dop, we carried out turnover measurements as a function of Pup-Fl concentration and analyzed them according to Michaelis-Menten kinetics (Fig. 3b). While the K m remains unchanged, the k cat of depupylation in the presence of ATP exhibited by the DopGS variant is 2.3-fold lower than that of the WT enzyme. We also compared WT Dop and the DopGS variant by following depupylation of pupylated ketopantoate hydroxymethyltransferase (PanB-Pup) using a gel-based assay (Fig. 3c, d). Although this assay does not offer the same resolution as the anisotropy assay, our data show that the DopGS variant is lagging behind the WT enzyme in the PanB-Pup depupylation time course. Taken together, our results indicate that the Dop-loop aids formation of the active site Pi species and also enhances the overall turnover of pupylated substrate.
A recent study suggested that the Dop-loop serves as a regulatory element that inhibits the depupylation activity 23 , however, such an inhibitory role of the Dop-loop on depupylation was not observed in our experiments (Fig. 3). Our results are more in line with the observation that the removal of the Doploop in addition to the substitution of the α-loop from PafA can convert the depupylase Dop into a ligase 20 , as our results indicate that the Dop-loop stimulates the production of the active site phosphate, an important step for enabling C-N bond cleavage.
ATP γ-phosphate and the Pup C-terminal residue are poised for nucleophilic attack. Although we previously demonstrated that ATP hydrolysis in the active site of Dop is dependent on Pup binding 19 , the mechanistic role played by Pup in ATP hydrolysis remained unknown. Structural analysis of Pup-bound Dop complexes should provide the molecular basis for understanding the role of Pup in the production of the catalytic phosphate.
In the active site of the Dop-PupQ complex ( Supplementary  Fig. 4a), the non-hydrolyzable ATP analog AMP-PCP displays strong and unambiguous density with an occupancy of about 72 % (Supplementary Fig. 4b). Correspondingly, two of the three canonical Mg 2+ -binding sites (n1-n3) defined for the carboxylate-amine/ammonia ligase superfamily also show only partial occupancies (n2 and n3).
The γ-phosphate of AMP-PCP is kinked slightly out of the active site and away from the side chain of the C-terminal Pup residue ( Supplementary Fig. 4a), likely due to the inability of the AMP-PCP-analogue-specific carbon atom between the βand γ-phosphates to fill the empty coordination position of Mg 2+ at the n3 site. In the structure of GS solved in the presence of AMP-PNP (PDB 2D3B) 24 , the equivalent γ-phosphate is oriented towards the side chain of the bound inhibitor methionine sulfoximine (MSO) and the bridging nitrogen atom coordinates the manganese (Mn 2+ ) ion at the n3 site ( Supplementary Fig. 4c). In the presence of ATP, the NH group of MSO makes a nucleophilic attack on the γ-phosphorus of ATP producing phosphorylated MSO 24 . As Dop features a highly similar active site configuration as GS (compare Supplementary Fig. 4a and 4c), it is expected that AMP-PNP binds to Dop in the same configuration as to GS. In order to determine experimentally if the γ-phosphate group adopts a similar conformation in Dop, we tested if Dop is also able to react with a sulfoximine inhibitor. We used a Pup derivative ( Acel PupBSO ΔN43 ) whose C-terminal residue was replaced with buthionine sulfoximine (BSO), a potent inhibitor of glutamate cysteine ligase (GCL) 25,26 . Using intact mass spectrometry, we show that Dop indeed forms a phosphorylated PupBSO species in the presence of ATP (Fig. 4), demonstrating that the γ-phosphate is oriented towards the C-terminal residue of Pup and stabilized by coordination with Mg 2+ at the canonical n1-n3 sites.
Production of Pi is preceded by phosphoryl transfer to C-terminal Pup residue. The mechanism of enzyme-catalyzed ATP hydrolysis has been widely studied by using metal fluorides to substitute the phosphoryl group in order to stabilize distinct catalytic intermediates 27 . While the tetrahedral beryllium trifluoride (BeF 3 − ) mimics the ground state, the octahedral aluminum tetrafluoride (AlF 4 − ) and trigonal bipyramidal magnesium trifluoride (MgF 3 − ) present transition state analogs 28,29 . In order to gain structural insights into ATP hydrolysis generating the active site inorganic phosphate species 19 , we carried out extensive screens to co-crystallize the Dop-Pup complex with different metal fluorides. Magnesium fluoride (MgF x ) proved to be best suited, yielding soluble complexes and producing well-diffracting crystals. We solved the structure of a Dop-Pup complex with a square planar MgF x species bound between ADP and Pup (Fig. 5a, b; Table 1). The same unexpected square planar configuration was recently observed in the structure of Zika virus NS3 helicase crystallized in the presence of MgF x 30 . It was identified as containing three fluorines and one water molecule, MgF 3 (H 2 O) − , in the equatorial plane. In Dop, detailed examination of the unbiased omit map of the compound revealed weaker electron density at the site closest to D94, thereby identifying it as oxygen (Fig. 5a, b).
The MgF 3 (H 2 O)‾ species mimicking the dissociating γ-phosphate in the Dop active site is sandwiched at equal distance of around 2.4 Å from the β-phosphate of ADP and from the carboxylate side chain of the C-terminal Pup residue, forming an octahedral coordination geometry (Fig. 5a). Two of the fluorine atoms are coordinated to Mg 2+ ions in the active site, one to the n2 site Mg 2+ and the other held between the n1 and n2 site Mg 2+ ions, which helps to neutralize the charges developing on the γ-phosphate of ATP during hydrolysis. The configuration of MgF 3 (H 2 O) − is suggestive of a transition state for phosphoryl transfer to the C-terminal glutamate residue of Pup 30 (Fig. 5a, c). Indeed, our structure agrees with a transition state model proposed previously for GS-catalyzed glutamine synthesis 24 . The PupE fragment used here for cocrystallization with Dop represents the product generated after deamidation or depupylation. We previously showed that both PupQ and PupE can stimulate the formation of the active site inorganic phosphate species from ATP 19 . It is therefore expected that the analogous transition state complex is formed with PupQ and magnesium fluoride. Taken together, our structure suggests that the C-terminal amide or carboxylate side chain of Pup is directly involved in ATP cleavage and presents the acceptor group in the transition state for phosphoryl transfer.   The phosphorylated Pup intermediate resulting from phosphoryl transfer to the amide side-chain of the C-terminal residue should only exist transiently in solution and be readily hydrolyzed after its formation 19 . However, when crystals of Dop-PupE complex were soaked with ATP for 19 h, we serendipitously produced Dop with a bound phosphorylated Pup fragment and ADP in the active site (Fig. 5d, e; Table 1). Presumably due to the high concentration of KH 2 PO 4 in the crystallization solution, a potassium (K + ) ion instead of a Mg 2+ ion was identified at the n2 site, as indicated by the strong anomalous signal peak from data collected at around 2.0 Å wavelength (Fig. 5e)  by a Mg 2+ ion (Fig. 5f). However, the binding of K + at the n2 site does not affect the binding of ADP and the C-terminal Pup residue, as both bind in a similar conformation as observed in the transition state complex (Supplementary Fig. 5). Furthermore, the active site with the phosphorylated Pup intermediate displays a similar configuration as the active site of GS complexed with ADP, Mn 2+ and phosphorylated MSO (PDB 2D3A) 24 . This further supports a mechanism involving the nucleophilic attack of the side chain carboxylate oxygen of the C-terminal Pup residue on the γ-phosphate of ATP proceeding through a transition state of phosphoryl transfer 24 as observed in the Dop-Pup structure in complex with ADP and octahedral trifluoromagnesate (Fig. 5a).
Active site Pi is held in place for multiple rounds of catalysis by two low-barrier hydrogen bonds. In order to observe the Dop active site after generation of the catalytic phosphate from the phospho-Pup intermediate, we attempted to obtain crystals of Dop-Pup-ADP in complex with sulfate or tetrafluoromagnesate, both of which could theoretically serve as Pi mimics 31,32 . Dop-Pup-ADP crystals were soaked with buffers containing high concentrations of MgSO 4 and NaF, allowing us to determine a structure of the product state with a clear tetrahedral electron density observed for the Pi analog (Fig. 5g, h; Table 1). Structure refinement with SO 4 2− or MgF 4 2− placed in the tetrahedral electron density (Supplementary Fig. 6a and 6b) showed that although both of the ligands were refined to partial occupancies (65 % vs 74 %, respectively), MgF 4 2− fits the observed density better than SO 4 2− and was hence modeled in the final structure of the product state (Fig. 5g, h). The binding of the tetrahedral MgF 4 2− in the active site of Dop is mediated via an intricate polar interaction network. The contacts include electrostatic interactions formed with the side chain of the highly conserved R227 and the Mg 2+ ion bound at the site where the square planar magnesium fluoride is located in the transition state complex (referred to here as n7 site) ( Supplementary Fig. 6c), as well as hydrogen bonds established with the side chain of H241, the C-terminal residue of Pup and likely the β-phosphoryl oxygen of ADP (Fig. 5g). Notably, the distance between MgF 4 2− and the β-phosphate of ADP is only 2.4 Å. Although we cannot confirm the existence of a hydrogen atom in a density map at 1.75 Å resolution, it stands to reason that in the absence of a hydrogen bond, such a short distance would lead to strong electrostatic repulsion. We, therefore, conclude that MgF 4 2− interacts with the β-phosphate of ADP through a hydrogen bond. Hydrogen bonds with a length below 2.6 Å are particularly strong, and this type of hydrogen bond, referred to as low-barrier hydrogen bond (LBHB), is frequently found in the active sites of enzymes [33][34][35] . Another LBHB identified in this structure is located between MgF 4 2− , the Pi mimic, and R227, separated by a distance of around 2.3 Å (Fig. 5g). Therefore, it appears that Pi in the active site of the Dop-Pup complex is held by two strong LBHBs (Fig. 5g, i), likely preventing fast dissociation of Pi, which explains the observation that the generated ADP/Pi in the active site can support multiple rounds of depupylation/deamidation 19 .
Our earlier hypothesis for the catalytic role of Pi during depupylation was that Pi acts as a nucleophile attacking the isopeptide bond of a pupylated substrate to form a phosphorylated Pup intermediate 19 . Based on the structure presented here of the Dop-PupE complex with the Pi mimic in the active site, Pi could alternatively serve to make the carbonyl carbon of the amide a better electrophile by stabilizing the amide resonance form with the carbon-nitrogen double bond. This could allow the direct nucleophilic attack of a water molecule to cleave the C-N bond. Supporting this alternative model is the location of MgF 4 2− , where an ideal Bürgi-Dunitz angle of attack (α BD ) 36 towards the side chain carbonyl group of the C-terminal Pup residue is not achieved, so that Pi might not be able to initiate a nucleophilic attack on Pup. However, we cannot rule out that Pi attacks the isopeptide bond, as we cannot be sure of the exact positioning of Pi in the Dop-PupQ complex (as opposed to the Dop-PupE complex we have solved here). Both scenarios are possible and in line with the mechanistic features of the γ-carboxylate/amine ligase superfamily, and in both cases, catalysis depends on the presence of the active site phosphate species.
Mechanism of Pup-stimulated ATP cleavage to generate the catalytic Pi species. Before the catalytic cycle of Dop can begin, Dop-mediated cleavage of ATP in the active site has to take place to generate the Pi species important for catalysis 19 Fig. 4a and 4b), the transition state of phosphoryl transfer (Dop-PupE-ADP-MgF 3 (H 2 O) − ) (Fig. 5a, c), the Pup-phosphorylated state (Dop-Pup P -ADP) (Fig. 5d, f) and the product state (Dop-PupE-ADP-MgF 4 2− ) (Fig. 5g, i), we can develop a full mechanistic model for the generation of the catalytic Pi species (Fig. 6). We previously showed that the stimulation of Dop-mediated ATP hydrolysis by the C-terminal residue of Pup can be supported by both glutamate (E) and glutamine (Q) as C-terminal residue 19 . We, therefore, propose that both PupQ and PupE employ their C-terminal side-chain carbonyl oxygen to serve as the nucleophile attacking the γ-phosphorus of ATP (Fig. 6a). In case of PupE, this is the analogous reaction mechanism as occurs in GS during glutamine synthesis 24 . With the deamidation substrate PupQ (or the substrate-linked Pup) the C-terminal side chain amide would carry out this reaction in a similar fashion as is observed for argininosuccinate synthesis, where the side chain carbamide oxygen of citrulline nucleophilically attacks the αphosphorus of ATP to form the activated acyl-AMP intermediate 37 . The C-terminal side chain amide of PupQ in the active site of Dop might exist as a significant resonance form of the carbon-nitrogen double bond, which is stabilized via coordination with the Mg 2+ bound at the n1 site, thus making the amide carbonyl oxygen a good nucleophile for the γ-phosphorus of ATP. The carbamidyl phosphate moiety produced in the active site of Dop through a transition state of phosphoryl transfer is stabilized through coordination with three Mg 2+ ions bound at the n1, n2 and n3 sites (Figs. 5c, f and 6). In the subsequent step, a nearby water molecule, most likely activated by the side chain of D94, is responsible for hydrolysis of the glutaminyl-phosphate intermediate 19 . The resulting Pi product is stabilized in the active site through strong LBHBs and through coordination with a Mg 2+ bound at the n7 site (Figs. 5g, i and 6). In order to test our hypothesis, we mutated the C-terminal residue of Pup to methionine (PupM), which is of similar length as glutamine, but does not possess the amide functional group. ATP hydrolysis was measured in the absence of Pup, in the presence of PupE, and in the presence of PupM ( Supplementary Fig. 7). PupM is unable to stimulate ATP hydrolysis in the Dop active site, demonstrating that the carboxylate/amide side chain of the C-terminal Pup residue is directly involved in ATP hydrolysis.
Taken together, our high-resolution X-ray crystallographic data supported by biochemical analysis provide comprehensive insights into the catalytic mechanism of Dop-mediated ATP hydrolysis, which proceeds via a transition state of phosphoryl transfer from the γ-phosphate of ATP to the C-terminal side chain carbonyl oxygen of Pup. Furthermore, we demonstrate that a characteristic loop in Dop (the Dop-loop) plays a key role in the process by serving as an allosteric sensor to detect Pup binding and subsequently promote ATP hydrolysis and catalysis.

Methods
Protein expression and purification. Acel Dop was expressed from isopropyl-β-D-1-thiogalactopyranoside (IPTG)-inducible vector pET21 in Escherichia coli Rosetta (DE3) cells (Invitrogen) as a C-terminal tobacco etch virus (TEV) protease cleavage site-His 6 fusion for 16 h at 20°C. For purification, cleared lysate was loaded on a 5 mL Hi-Trap immobilized metal affinity chromatography (IMAC) HP column (GE Healthcare Life Sciences) charged with Ni 2+ . The column was washed with 50 mL of buffer W (50 mM Tris-HCl pH 8.0 at 23°C, 300 mM NaCl, 10 % (v/v) glycerol and 40 mM imidazole) and the protein was eluted with buffer W containing 300 mM imidazole. Dop-containing fractions were pooled and dialyzed overnight at 4°C against buffer D (50 mM Tris-HCl pH 8.0 at 4°C, 150 mM NaCl, 10 % (v/v) glycerol and 1 mM EDTA). The C-terminal His 6 -tag was cleaved at the TEV protease cleavage site by the addition of His-tagged TEV protease. TEV protease was subsequently removed via Ni 2+ -affinity chromatography. Acel Dop was further purified by size exclusion chromatography using a Superdex 75 gel filtration column in 20 mM HEPES-NaOH pH 8.0 at 4°C and 50 mM NaCl. Cglu Dop and its variants with an N-terminal His 6 -TEV protease cleavage site fusion were expressed from pET24 and purified similarly to Acel Dop except that the buffer used for size exclusion chromatography was changed to buffer D containing 300 mM NaCl, 10 % (v/v) glycerol, and 1 mM DTT. For biochemical assays, Cglu Dop and its variants were further purified by anion exchange chromatography on a Mono Q 5/50 GL column (Cytiva) to reduce the amount of co-purified E. coli adenylate kinase (Adk). Cglu Dop was eluted with a linear gradient from 0 to 1 M NaCl in 20 mM Tris-HCl pH 7.0 at 4°C and exchanged into buffer D by dialysis.
The Pup fragments ( Acel PupE/Q ΔN43 ) used for crystallization were synthesized by GenScript, while Cglu PupE or Cglu PupM was expressed and purified as described previously 9 with the addition of an anion exchange step to remove co-purified E. coli Adk. Briefly, Pup was expressed as a His 6 -Thioredoxin-TEV-Pup fusion protein and purified by a 5 ml Hi-Trap IMAC HP column (GE Healthcare Life Sciences) charged with Ni 2+ . After cleavage of the fusion protein with His-tagged TEV protease, His 6 -Thioredoxin and TEV protease were removed by Ni 2+ -affinity chromatography. Pup was further purified by size-exclusion chromatography on a Superdex 200 column (GE Healthcare) and an anion-exchange column (Mono Q 5/50 GL column, Cytiva), and stored in 50 mM HEPES-NaOH pH 8.0 at 23°C, 150 mM NaCl, 1 mM EDTA and 0.5 mM TCEP. Acel PupBSO ΔN43 peptide was synthesized by PEPSCAN.
Gel-based depupylation assays. 3 µM Mtb PanB-Cglu Pup was incubated with 0.5 µM Cglu Dop or Cglu DopGS variant at 30°C in buffer R (50 mM HEPES-NaOH pH 8.0 at 23°C, 150 mM KCl, 10 % (v/v) glycerol, 1 mM DTT and 20 mM MgCl 2 ) supplemented with 0.5 mM nucleotide and additional 10 mM Na/K phosphate (pH 8.0 at 23°C) in the case of ADP. The formation of PanB was monitored by SDS-PAGE followed by Coomassie staining and was analyzed densitometrically using the GelAnalyzer software (version 2010a). The fraction of PanB is expressed with respect to the total PanB (sum of PanB and PanB-Pup). The contribution from Pup is negligible due to its small size and poor staining.   Fig. 6 Proposed mechanism for the generation of the catalytic Pi in the active site of Dop. Active site residues are numbered according to Acel Dop and shown in simplified representation omitting magnesium coordinating residues (Glu8, Glu10, Tyr92, Glu99, His155, and His241) and waters. Shared protons are bonded as hashed lines. R is equal to H or substrate.
spectrometer. The Sampling Cone energy was set at 40 V, and the capillary voltage was set at 3.0 kV. The neutral-mass spectra were inferred from the mass over charge measurement deconvolution using the MaxEnt3 Software.