Structural Basis for EarP-Mediated Arginine Glycosylation of Translation Elongation Factor EF-P

ABSTRACT Glycosylation is a universal strategy to posttranslationally modify proteins. The recently discovered arginine rhamnosylation activates the polyproline-specific bacterial translation elongation factor EF-P. EF-P is rhamnosylated on arginine 32 by the glycosyltransferase EarP. However, the enzymatic mechanism remains elusive. In the present study, we solved the crystal structure of EarP from Pseudomonas putida. The enzyme is composed of two opposing domains with Rossmann folds, thus constituting a B pattern-type glycosyltransferase (GT-B). While dTDP-β-l-rhamnose is located within a highly conserved pocket of the C-domain, EarP recognizes the KOW-like N-domain of EF-P. Based on our data, we propose a structural model for arginine glycosylation by EarP. As EarP is essential for pathogenicity in P. aeruginosa, our study provides the basis for targeted inhibitor design.

T ranslation elongation is a nonuniform process and directly depends on the amino acids (aa) to be incorporated into the growing polypeptide chain (1). Due to its chemical and physical properties, proline delays the peptidyl transfer reaction (2), and ribosomes can even stall upon translation of distinct diprolyl-containing sequence motifs ( Fig. 1) (3,4). Such ribosome stalling is alleviated by the eukaryotic and archaeal elongation factor 5A (e/aEF-5A) (5-7) and its prokaryotic orthologue the bacterial translation elongation factor P (EF-P) (8)(9)(10)(11)(12)(13)(14). The L-shaped EF-P is composed of three ␤-barrel domains and structurally resembles tRNA in both size and shape (15). EF-P binds to the polyproline-stalled ribosomes between the binding sites of peptidyl-tRNA (P-site) and the exiting tRNA (E-site) (16) and stimulates peptide bond formation by stabilization of the CCA end of the P-site prolyl-tRNA ( Fig. 1) (17,18). A conserved positively charged residue-located at the tip of the EF-P KOW-like N-domain-is essential for function (11,17). However, for full EF-P activity, this residue is posttranslationally elongated (19). Certain bacteria-including Escherichia coli and Salmonella enterica-ß-lysinylate a conserved lysine, K34 EF-P , by EpmA. This EF-P-specific ligase cross-relaxation effects and to decrease the signal line width. Nonetheless, using transverse relaxation-optimized spectroscopy (TROSY)-based experiments, we were able to assign 62% of the EarP Ppu backbone.
The two domains are interconnected by a bipartite helix (␣6, ␣7) comprising aa 156 to 176. This linker region together with an unstructured segment that positions ␣15 in the vicinity of the N terminus defines the floor of the cleft that separates the domains ( Fig. 2A and see Fig. 4).
Analysis of the TDP-␤-L-rhamnose binding site in the EarP C-domain. In Leloirtype GT-B glycosyltransferases, the nucleotide-sugar binding site is canonically located in the protein C-domain (40). Similarly, TDP-Rha in the EarP Ppu crystal structure is located in a binding pocket that is composed of residues located in the C-domain (Fig. 3A). F191 EarP , F252 EarP , and F258 EarP side chains form an aromatic cage that stacks against the base of the nucleotide moiety of TDP-Rha. The sugar ring of the nucleotide is then specifically recognized by a hydrogen bond between the hydroxyl group on C3= of the sugar and the side chain of Q255 EarP . The diphosphate is recognized by hydrogen bonds formed with the side chain guanidine of R271 EarP , the Y193 EarP side chain hydroxyl, and backbone amides of E273 EarP and D274 EarP . The binding pocket is closed by the bulky side chain of Y193 EarP , which may sterically ensure proper positioning of the rhamnose sugar (Fig. 3A). The rhamnose sugar itself does not seem to make any contact with the protein and is solvent exposed. We further confirmed this by saturation transfer difference (STD) NMR experiments (41), where we did not observe any difference signal from the rhamnose moiety but did observe one from the TDP moiety of TDP-Rha (Fig. S5A).
In parallel, small-angle X-ray scattering (SAXS) of free EarP Ppu and EarP Ppu bound to TDP-Rha has been performed (Fig. S3D). The overall shape of the molecule could be validated to be the same in solution. Protein backbone conformational changes upon TDP-Rha binding are confirmed by chemical shift perturbations (see Fig. 7B); however, SAXS indicates that there are no large (Ͼ10-Å) conformational changes or movements of the two Rossmann fold domains with respect to each other upon binding of TDP-Rha, as the scattering density does not change from that in the free state. To show that TDP-Rha is bound to EarP under SAXS experimental conditions, STD NMR experiments were performed. They confirm again that TDP-Rha binding occurs with the ligand at a 7-fold excess compared to the amount of protein (Fig. S5B).
In order to identify conserved amino acids, we used Clustal Omega (42) and generated a multiple-sequence alignment (Fig. 4A). We found 49 residues with a sequence conservation of Ն95%. Mapping of these residues onto the crystal structure revealed an accumulation at or near the interdomain cleft (Fig. 4B), including the binding pocket for the nucleotide sugar donor substrate (Fig. 3A), which is highly supportive of the correctness of the solved structure.
To substantiate our structural findings with biochemical data, we prepared EarP Ppu constructs with single-amino-acid substitutions of the individual residues forming the binding pocket and tested the activities of the EarP Ppu variants both in vivo and in vitro (Fig. 5). This included F191 EarP , F252 EarP , and F258 EarP , which form the aromatic pocket, as well as Y193 EarP , Q255 EarP , R271 EarP , and D274 EarP , which are involved in hydrogen bond networking (Fig. 5B).
Previously, we could show that the heterologous expression of efp and earP from Shewanella oneidensis in E. coli can fully complement a lack of EF-P (17) with respect to the activation of the lysine-dependent acid stress response by the transcriptional activator CadC (11). Similarly, coproduction of wild-type EF-P Ppu and wild-type EarP Ppu (WT EarP ) can restore ␤-galactosidase activity in an E. coli P cadBA ::lacZ Δefp strain ( Fig. 5A and S1B). From the nine tested EarP Ppu substitution variants, we measured reduced ␤-galactosidase activities for the variants F191A EarP , Y193A EarP , R271A EarP , S275A EarP , and Y291A EarP . The variants R271A EarP and Y291A EarP failed to induce ␤-galactosidase expression at all ( Fig. 5B and S1B).
In parallel, the enzymatic activity of EarP Ppu was investigated in vitro by employing an anti-Arg Rha antibody. The antibody was raised against a chemically synthesized glycopeptide antigen (SGR Rha NAAIVK) and specifically detects arginine rhamnosylation Ribbon representation of the nucleotide-sugar binding pocket with stick representation of bound TDP-Rha (blue sticks) as well as the three invariant residues D13, D17, and E273 (green sticks), which are presumably involved in catalysis. (Right) Surface representation of the nucleotide-sugar binding pocket with stick representation of bound TDP-Rha (blue). Surfaces of D13, D17, and E273 are in green. Ribbons are color coded according to their degree of conservation, as follows: yellow, 100%; black, Ն95%; dark grey, Ն90%; light grey, Ն50%; and white, Ͻ50% identical residues in all analyzed EarP orthologues. The electron density for TDP-Rha bound to EarP is shown in Fig. S3C. All illustrations were generated with UCSF Chimera (82).
(see Materials and Methods) (Fig. S1A). This in turn enabled the quantification of rhamnosylation rates of EF-P Ppu by Western blot analysis ( Fig. 5C and D). In a first step, the K m and k cat of WT EarP were determined to be 53 M and 35 min Ϫ1 , respectively ( Fig. 5B, C, and D).
We wondered whether this K m makes sense physiologically and therefore analyzed the cellular TDP-Rha levels in P. putida, P. aeruginosa, and E. coli, which were 3.5 mM, 2.0 mM, and 4.0 mM, respectively (see Materials and Methods and Fig. S6). In good accordance with our measurements, the physiological TDP-Rha concentration in Lactococcus lactis was previously determined to be as high as 1 mM (43). Thus, within a bacterial cell, the donor substrate reaches saturating concentrations, according to the WT EarP K m measurements.
Next, the K m and k cat of EarP Ppu substitution variants were determined and compared to those of the wild-type protein. Strikingly, all earP mutations affected enzymatic activity ( Fig. 5B and S2B). Depending on the substituted residue, the K m increased up to 60-fold for the F252A EarP variant (K m ϭ 3.4 mM). Conversely, the k cat decreased up to 3,500 times when we measured the kinetics of the F191A EarP and Y193A EarP variants.
To exclude the possibility that decreased enzyme activity was due to fold disruption, selected EarP Ppu variants (F191A EarP , Y193A EarP , F252A EarP , R271A EarP , D274A EarP , and Y291A EarP ) were analyzed by NMR 1 H-15 N heteronuclear single quantum coherence (HSQC) experiments (Fig. S7). All tested substitution variants showed no structural alterations from the wild-type protein, except for the D274A EarP variant. The structural instability of this EarP variant might be a result of disrupting a salt bridge that is formed between the side chains of D274 EarP in the protein C-domain and an equally conserved arginine at position 23 (R23 EarP ) in the protein N-domain (Fig. 4). This salt bridge might be of importance in clamping both EarP domains together, and a lack of it might therefore destabilize the protein. Indeed, further purification of the D274A EarP variant by size exclusion chromatography (SEC) revealed an elution pattern with three distinct EarP peaks, indicating a certain degree of protein aggregation. However, the lowest molecular peak in the D274A EarP SEC profile is congruent with the one that we found when subjecting WT EarP to SEC. Accordingly, K m (TDP-Rha) and k cat values were determined from this protein fraction to be 206 M and 0.74 min Ϫ1 , respectively (Fig. 5B).
In parallel, a bacterial two-hybrid analysis (44) was set up to investigate interactions between EF-P Ppu and WT EarP as well as the above-mentioned nine substitution variants At low pH and with the concomitant presence of lysine, the transcriptional activator CadC activates the promoter of its two downstream genes (P cadBA ) and with this induces expression of lacZ in an E. coli MG1655 P cadBA ::lacZ strain (11). Proper translation of CadC is dependent on the presence of EF-P and its corresponding modification system, and thus ␤-galactosidase activity can be taken as a direct readout for EF-P and EarP functionality. (B) Degree of conservation (identity/similarity) in percent, in vivo activities, and kinetic parameters of tested single-amino-acid exchange variants of EarP Ppu . In vivo EarP Ppu activities were determined by measuring the ␤-galactosidase activities of an E. coli MG1655 P cadBA ::lacZ Δefp strain heterologously expressing efp Ppu together with wild-type or mutant earP Ppu genes from o/n cultures in LB (pH 5.8). Background corrected mean values from three independent measurements are shown. Standard deviations were determined from three independent experiments to be Յ10%; the K m and k cat of wild-type EarP Ppu (WT EarP ) and variants with single-amino-acid substitutions are given in micromolar concentrations and per minute, respectively. Standard errors were determined by SigmaPlot to be Ͻ20%. (C, top) 2,2,2-Trichlorethanol (TCE) protein stain (75) of a representative SDS gel used for determination of kinetic parameters. Fixed amounts of EF-P Ppu (2.5 M) and WT EarP (0.1 M) were incubated with various concentrations of TDP-Rha for 20 s and subjected to SDS-PAGE. (Bottom) Detection of rhamnosylated EF-P Ppu . EF-P Ppu was visualized after Western blotting using 0.25 g/ml anti-Arg Rha . (D) TDP-Rha saturation curve of WT EarP . Band intensities from panel C were quantified using ImageJ (76). Reaction rates were calculated as means of four independent measurements. Standard deviations are shown as error bars for each concentration. (Fig. 5B). Therefore, fusions were generated with two complementary fragments, T25 and T18, encoding segments of the catalytic domain of the Bordetella pertussis adenylate cyclase CyaA. If EF-P Ppu and WT EarP do interact, then CyaA is reconstituted, which in turn allows induction of the lac promoter and results in lacZ expression. Accordingly, ␤-galactosidase activity is a measure of the interaction strength. When coproducing EF-P Ppu with WT EarP , we determined ca. 250 MU, whereas combinations with solely T25 and T18 resulted in Ͻ60 MU, thus defining the threshold of the assay (Fig. S1C). Except for the R271A EarP and Y291A EarP proteins, all other variants were below this threshold, indicating that alterations in the donor binding site might also affect acceptor binding (Fig. S1C).
The KOW-like EF-P N-domain is sufficient for EarP-mediated rhamnosylation. To test which part of EF-P is involved in the interaction with EarP, NMR chemical shift perturbation experiments were performed by comparing 1 H-15 N HSQC results between unbound EF-P Ppu and EarP Ppu -bound EF-P Ppu (Fig. 6A). Triple-resonance experiments of EF-P Ppu enabled backbone assignment, with a sequence coverage of 97%. Missing assignments are for residues S123, R133, N140, V164, D175, and G185. The assignment also enabled secondary-structure determination from secondary chemical shifts and confirmed the validity of the EF-P model for P. putida, based on the crystal structure of P. aeruginosa EF-P ( Fig. S3E) (45). The titration experiment showed clear chemical shift perturbations in the N-terminal acceptor domain of EF-P Ppu ( Fig. 6B and C). However, R32 EF-P and residues surrounding the rhamnosylation site (e.g., S30 EF-P , G31 EF-P , R32 EF-P , To analyze the interaction, CSPs were calculated as described in Materials and Methods and plotted against residue numbers. Color coding is indicated in the upper right corner. Full lines indicate median CSPs, dashed lines indicate median CSPs plus standard deviations, and residues with CSPs higher than the median plus standard deviation are shown in brighter shades of the colors. The N-terminal loop containing rhamnosylation target R32 EF-P is indicated. (C) CSPs of unmodified EF-P Ppu titrated by EarP Ppu plotted on the model of EF-P from P. aeruginosa (45) (PDB accession number 3OYY) using a white-to-orange gradient, where white represents the weakest CSP and orange depicts the strongest CSP. The position of R32 EF-P is indicated. (D) Rhamnosylation experiments using full-length EF-P Ppu and C-terminally truncated variants (EF-P Ppu with aa 1 to 128, EF-P Ppu with aa 1 to 65). EF-P was detected using 0.2 g/ml anti-EF-P. Rhamnosylation of purified protein was detected using 0.25 g/ml anti-Arg Rha . The domain structure of the respective protein variants is indicated as in panel B. N33 EF-P ) are severely line broadened beyond detection. Therefore, chemical shift perturbation values cannot be determined for these and vicinal residues. This line broadening is an indication that they are bound by EarP Ppu and thus have rotational correlation times expected for a complex of that size. Several residues located in the S1-like OB-domain are also slightly affected. However, this is not necessarily due to direct contacts with EarP Ppu but might also be propagating effects. Therefore, we also investigated in vitro rhamnosylation of truncated EF-P Ppu variants comprising either amino acids 1 to 128 or amino acids 1 to 65 (Fig. 6D). Both truncations were readily rhamnosylated by EarP Ppu , further corroborating that EF-P contact sites are predominantly located in the KOW-like N-domain.
In addition, we compared NMR interactions between EarP Ppu and unmodified EF-P Ppu or rhamnosylated EF-P Ppu . This experiment clearly showed that chemical shift perturbations for unmodified EF-P are stronger than for rhamnosylated EF-P (Fig. 6B). Thus, EarP releases EF-P after rhamnosylation due to decreased affinity, while unmodified EF-P binds with higher affinity to enable efficient posttranslational modification.
Mutational analysis of the three invariant EarP residues D13, D17, and E273. We and others previously showed that EarP inverts the anomeric configuration on the sugar moiety from TDP-␤-L-rhamnose to ␣-rhamnosyl arginine (26,27). Reportedly, inverting glycosyltransferases employ a direct-displacement S N 2-like reaction (46). The molecular basis for inverted N-linked glycosylation was elucidated for the oligosaccharyl transferase PglB (47). Here the catalytic site features three acidic side chains (29). As with PglB, three negatively charged residues-aspartates D13 EarP and D17 EarP and glutamate E273 EarP -were identified as potential candidates to catalyze the glycosylation reaction (Fig. 3B). All three residues are invariant in all EarP orthologues ( Fig. 4A; Data Set S3). Moreover, the D13 EarP and D17 EarP variants as well as the E273 EarP variant are in the vicinity of the rhamnose moiety and might therefore be proximal to the putative active center and R32 of EF-P (Fig. 3B). The distances of these three residues to rhamnose atoms range from 2.5 to 4.5 Å (the carboxyl group of D13 is the closest, with a distance of 2.5 Å to the methyl group of the rhamnose, followed by the side chains of D17 and E273, with distances of 3.9 and 4.5 Å to the hydroxyl group of C4 and C2, respectively). Consequently, we constructed the corresponding alanine substitution variants D13A EarP , D17A EarP , and E273A EarP and investigated their enzymatic activities in vitro. In line with the idea that these residues might be involved in catalysis, EF-P rhamnosylation could not be detected even after 8 h of incubation, and accordingly these EarP variants are inactive (Fig. 7A).
To exclude misfolding being causative for the nonfunctional EarP Ppu protein variants, 15 N HSQCs were measured for D13A EarP , D17A EarP , and E273A EarP . The spectra show no structural alterations from WT EarP (Fig. 7B, C, and D and see Fig. S7). Additionally, the variants D13A EarP and D17A EarP were titrated with TDP-Rha being indistinguishable from WT EarP perturbations. Interestingly, although D13 EarP and D17 EarP resonances could not be assigned, other residues in close proximity (G16 EarP and G19 EarP ) exhibited strong perturbations not only in WT EarP but also in the D13A EarP and D17A EarP variants upon TDP-Rha binding, despite not forming direct ligand contacts (Fig. 7E). Similarly, we could measure TDP-Rha binding for E273A/D/N EarP variants using STD NMR (Fig. S5C). This confirms that these mutations do not affect donor substrate binding.
To investigate interactions between EF-P Ppu and the D13A EarP , D17A EarP , and E273A EarP variants, we again performed a bacterial two-hybrid analysis and were able to show that all substitution variants are capable of acceptor binding, demonstrated by a blue colony on X-Gal (5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside)-containing LB plates (Fig. 7F, S1C).
To further corroborate our findings on the in vitro-inactive D13A EarP , D17A EarP , and E273A EarP variants, they were subjected to an in vivo experiment in which we investigated their ability to activate EF-P Ppu (Fig. 5A). Additional substitutions-D13N/E EarP , D17N/E EarP , and E273Q/D EarP -were also included in the study. Expectedly, coproduction of the D13A EarP , D17A EarP , and E273A EarP variants with EF-P Ppu phenocopies Δefp with respect to P cadBA activation and in vivo rhamnosylation ( Fig. 7G; Fig. S1B). Similar results were obtained with the D17N/E EarP and E273Q EarP variants, whereas the D13E EarP and E273D EarP variants were drastically impaired in function, although they retained some residual activity. Their impairment is indicated by a certain degree of P cadBA activation as well as a band in the in vivo rhamnosylation blot ( Fig. 7G; Fig. S1B). In contrast, a variant with a change of D13 to asparagine was indistinguishable from WT EarP , implying an importance of the chain length over charge.
Our thorough analysis of these EarP variants suggests that they are promising candidates to be involved in catalysis.

DISCUSSION
Activation of the proline-specific translation elongation factors EF-P and IF-5A is usually achieved by posttranslational elongation of the -amino group of a conserved lysine (20-23, 48, 49). The resultant noncanonical amino acids-␤-lysinyl-hydroxylysine, hypusine, and 5-amino-pentanolyl-lysine-appear to be chemically and structurally analogous. We recently showed that in a subset of bacteria, a so-far-unappreciated form of posttranslational modification plays an important role in the activation of EF-P. Here, instead of lysine, the guanidine group of a conserved arginine is modified with a rhamnose moiety by a glycosyltransferase termed EarP (17). This type of modification not only contrasts with the other known EF-P/IF-5A activation strategies but is also one of only two reported cases of enzyme-mediated arginine glycosylation. In canonical N-linked glycosylation, the sugar is attached to the amide nitrogen of an asparagine in an N-X-S/T consensus sequence (X is any amino acid except for a proline) (46,50). In contrast, the effector glycosyltransferase NleB of enteropathogenic E. coli N-acetylglucosaminylates (GlcNAc) specifically the arginines at positions 117 and 235 in the death domain-containing proteins FADD and TRADD, respectively (31,51). This in turn antagonizes the apoptosis of infected cells, thereby blocking a major antimicrobial host response. Notably, EarP shows neither sequential nor structural homologies to the GT-A-type glycosyltransferase NleB, and thus the arginine glycosylation of death domains and EF-P are examples of convergent evolution. Instead EarP seems to be structurally related to MurG. Moreover, and despite the lack of a significant overall sequence similarity, certain residues important for function remain the same. According to these facts, one might speculate that EarP is not simply analogous to MurG but a distinct homologue. Note that MurG is essential for cell wall biosynthesis in both Gram-negative and Gram-positive bacteria, and due to its degree of conservation, it is most likely more ancient then EarP. Although there is no real evidence for this, one might hypothesize about the possibility of a duplication of MurG in a betaproteobacterial progenitor, which is the presumed origin of EarP (17). Subsequently, the sequences of both proteins more and more diverged in consequence of distinct donor and acceptor substrates. This assumption is at least also in line with the theory that NleB (GT-A type) and EarP (GT-B type) are phylogenetically nonrelated enzymes. Accordingly, one can also assume that the molecular mechanisms of the glycosyl transfer reactions in both arginine glycosyltransferases differ. In 2016, Wong Fok Lung and coworkers mutated nleB and identified certain residues in NleB either interfering with FADD binding or preventing GlcNAcylation (52). They confirmed the importance of two invariant aspartate residues, D221 and D223, from among the nonfunctional NleB protein variants (31). A catalytic Asp-X-Asp motif is featured by various GT-A glycosyltransferases. Here, the two negatively charged aspartate side chains coordinate a divalent cation that facilitates departure of the nucleoside phosphate. Negatively charged amino acids also play important catalytic roles in inverting GT-B glycosyltransferases (46). In the case of the metal-independent fucosyltransferase FucT (53), for example, the side chain carboxyl groups of D13 and E95 may work as base catalysts (46). Also, the activation of the acceptor amide nitrogen by the lipid donor utilizing bacterial oligosaccharyltransferase PglB depends on the two negatively charged amino acids D56 and E319. These residues abolish the conjugation of the nitrogen electrons and allow the positioning of a free electron pair for the nucleophilic attack onto the anomeric center of the donor substrate (29,47). Analogously, the invariant negatively charged residues D13 EarP , D17 EarP , and E273 EarP in the EarP glycosyltransferase family might play a role in activating the R32 guanidino group of EF-P. Especially D17 EarP and E273 EarP -both in close proximity to each other-may form a catalytic dyad (Fig. 3B).
While activation of the acceptor substrate might be driven by the essential amino acids D13 EarP , D17 EarP , and E273 EarP , the nucleotide sugar donor TDP-Rha is bound in a highly conserved cavity of the protein C domain. A cocrystal structure of the putative structural EarP analogue MurG Eco with its cognate substrate reveals that aromatic amino acid side chains play important roles in UDP binding (PDB accession number 1NLM) (54). Similar interactions were reported for the protein O-fucosyltransferase POFUT1 (PDB accession number 3ZY6), where F357 is involved in -stacking with the respective nucleobase (55). Stacking interactions also play a role in EarP, in which the aromatic side chains of F252 EarP and F258 EarP bind the thymine and ribose moiety of TDP-Rha, respectively. In contrast, contacts with the ribose or the phosphate moieties frequently occur via interactions with side chain amines, hydroxyl groups, and backbone amides (37,54,55). Accordingly, this is also the case for EarP.
In GT-B glycosyltransferases, positively charged amino acids are often involved in facilitating leaving group departure. This is achieved by neutralization of evolving negative charges on the phosphate moiety during the glycosyl transfer reaction, as described, e.g., for R261 of MurG Eco (PDB accession number 1F0K) (37). Notably, earP Ppu encodes an invariant R271 EarP in the equivalent position and a substitution to alanine (R271A EarP ) strongly impairs protein function, all of which suggests that they have similar roles in product stabilization.
In GT-B glycosyltransferases, the two Rossmann folds can generally be divided into one donor and one acceptor substrate binding domain (40). As with other glycosyltransferases, the nucleotide sugar is bound by the protein C-domain of EarP. Accordingly, it is worth assuming important binding sites for EF-P in the protein N-domain. Conversely, EF-P presumably contacts EarP by amino acids that are in close proximity to the glycosylation site R32 EF-P . In agreement with this hypothesis, the EF-P ␤-lysine ligase EpmA, for example, recognizes EF-P via identity elements in a region located around the E. coli EF-P modification site K34 (21,22,56). Along the same line, the deoxyhypusine synthase (DHS) can efficiently modify a human eIF-5A fragment comprising only the first 90 amino acids of the protein (57). Similarly, we could show that the KOW-like N-terminal domain of EF-P (Fig. 6B) is sufficient to be glycosylated by EarP (Fig. 6D), being congruent with the NMR titrations of EF-P with EarP ( Fig. 6A to C). Upon titration with EarP, the chemical shift perturbations observed were (with a few exceptions) restricted to the first 65 residues.
Taking all of this together, we propose a three-step model for the rhamnosylation of EF-P by its cognate modifier EarP. In the ground state, both the nucleotide sugar binding site in the C-domain and the putative acceptor binding site in the N-domain are unoccupied.
In the donor-bound state, TDP-Rha is coordinated within a highly conserved cavity in the protein C-domain, including an aromatic pocket that surrounds the thymine ring (Fig. 3). Previous studies showed that binding of the donor substrate induces structural alterations in both the N and C-domains of glycosyltransferases (40,58,59). In MurG, these rearrangements include rotation of F244, which stacks over the nucleobase to cap the donor binding pocket (37). Notably, in the crystal structure of EarP, a phenylalanine, F252, is in the equivalent position, indicating that this capping interaction is conserved (Fig. 3A) (54).
In the catalytic state, the R32 guanidino group of EF-P might be activated by a mechanism analogous to the one that was reported for the oligosaccharyltransferase PglB (47). Hence, in the EF-P rhamnosylation reaction, R271 EarP might stabilize the nucleotide product, thereby facilitating leaving group departure. Upon successful inverting glycosyl transfer from TDP-Rha to R32 EF-P , presumably by a single S N 2 displacement reaction, the products are released from the active site of EarP, in turn reverting to the unbound ground state.
We point out that there is most likely no strict sequence of binding events, as NMR measurements demonstrate that EarP can interact with either substrate independently.
Altogether, our structural and biochemical investigation of EarP provides first insights into arginine glycosylation and improves our general understanding of N-linked glycosyl transfer reactions. Additionally, our research might open up new avenues for the development of antimicrobial drugs in order to fight, e.g., P. aeruginosa infections.

MATERIALS AND METHODS
Bacterial strains and growth conditions. Strains and plasmids used in this study are listed in Data Set S1 in the supplemental material. P. putida and E. coli were routinely grown in lysogeny broth (LB) (60,61) according to the Miller modification (62) at 30°C (for P. putida) and 37°C (for E. coli), unless indicated otherwise. When required, media were solidified by using 1.5% (wt/vol) agar. If necessary, media were supplemented with 50 g/ml chloramphenicol, 100 g/ml kanamycin sulfate, and/or 100 g/ml ampicillin sodium salt. For promoter induction from P BAD -containing plasmids (63), L-arabinose was added to a final concentration of 0.2% (wt/vol) in liquid medium. For promoter induction from plasmids comprising the lac operator sequences, isopropyl ␤-D-1-thiogalactopyranoside (IPTG) (Sigma-Aldrich) was added to a final concentration of 1 mM.
Molecular biology methods. Enzymes and kits were used according to the manufacturers' directions. Genomic DNA was obtained according to the protocol of Pospiech and Neumann (64), and plasmid DNA was isolated using a Hi Yield plasmid minikit (Süd-Laborbedarf GmbH). DNA fragments were purified from agarose gels by employing a Hi Yield PCR cleanup and gel extraction kit (Süd-Laborbedarf). Restriction endonucleases were purchased from New England Biolabs (NEB). Sequence amplifications by PCR were performed utilizing the Q5 high-fidelity DNA polymerase (NEB) or the OneTaq DNA polymerase (NEB). Mutations were introduced into the earP gene by overlap extension PCR (65,66). Oligonucleotides used in this study are listed in Data Set S1. All constructs were analyzed by Sanger sequencing (LMU Sequencing Service). Standard methods were performed according to the instructions of Sambrook and Russel (67).
␤-Galactosidase activity assay. Cells expressing lacZ under the control of the cadBA promoter were grown in buffered LB (pH 5.8) overnight (o/n) and harvested by centrifugation. ␤-Galactosidase activities were determined as described in reference 68 in biological triplicates and are given in Miller units (MU) (69). The significance of the results was determined by applying a two-sided Student t test and stating a result as significantly different if P was Ͻ0.05.
For quantification of interaction strength, which corresponds to the ␤-galactosidase activity, cells were inoculated in 1.5 ml LB medium containing 0.5 mM IPTG as well as 50 g/ml kanamycin sulfate and 100 g/ml ampicillin sodium salt. After incubation in 2-ml reaction tubes under microaerobic conditions at 30°C for 42 h, cells were harvested and ␤-galactosidase activities were determined as described above.
Protein purification. C-terminally His 6 -tagged EarP Ppu variants (pBAD33-earP Ppu ) were overproduced in E. coli LMG194 by addition of 0.2% arabinose to exponentially growing cells and subsequent cultivation at 18°C o/n. N-terminally His 6 -tagged EarP (pACYC-DUET-earP Ppu ) and His 6 -SUMO-tagged EF-P Ppu (pET-SUMO-efp Ppu ) were overproduced in E. coli BL21(DE3) by addition of 1 mM IPTG to exponentially growing cells. Subsequently, cells were incubated at 18°C overnight. Rhamnosylated EF-P Ppu (EF-P Rha ) was produced by cooverproduction with His 6 -tagged EarP Ppu . Cells were lysed by sonication, and His 6 -tagged proteins were purified using Ni-nitrilotriacetic acid (Ni-NTA; Qiagen) according to the manufacturer's instructions. The His 6 -SUMO tag was removed by incubation with 1 /mg His 6 -Ulp1 (71) overnight. Subsequently, tag-free EF-P Ppu was collected from the flowthrough after metal chelate affinity chromatography. For biochemical analyses, cells were cultivated in LB. For use in NMR spectroscopy, cells were grown in M9 minimal medium (62). If necessary, 15 N-labeled nitrogen ( 15 NH 4 Cl) and 13 C-labeled glucose were used. For NMR backbone assignment of EarP Ppu , additionally 99.8%-pure heavy water D 2 O (Sigma-Aldrich) was used instead of H 2 O in growth medium to allow partial deuteration of the protein in order to reduce cross-relaxation effects and increase the signal-to-noise ratio. Size exclusion chromatography of EarP Ppu and the D274A EarP variant was performed in 100 mM NaP i (pH 7.6) 50 mM NaCl using a Superdex 200 Increase 10/300-Gl column with a flow rate of 0.3 ml/min on an Äkta purifier (GE Healthcare). Four milligrams of protein was loaded in a volume of 0.5 ml (8 mg/ml). Eluting protein was detected at 280 nm. Fractions of 0.5 ml were collected.
For the production of selenomethylated EarP Ppu , E. coli BL21(DE3) cells expressing N-terminally His 6 -tagged EarP Ppu were cultivated in 1 liter M9 minimal medium at 37°C to an optical density at 600 EarP Structure and Biochemistry nm (OD 600 ) of 0.6. One hundred micrograms of threonine, 100 g lysine, and 50 g isoleucine were added to feedback inhibit methionine biosynthesis (72). Additionally, 50 g L-(ϩ)-selenomethionine was added 15 min prior to induction. Protein production was induced by addition of 1 mM IPTG, and cells were incubated at 18°C overnight. Protein concentrations were determined as described by Bradford (73). For biochemical analyses, EarP Ppu and EF-P Ppu were dialyzed against 100 mM NaP i , pH 7.6, 5 mM dithiothreitol (DTT), whereas a buffer composed of 100 mM NaP i , pH 7.6, 50 mM NaCl, and 5 mM DTT was used when the proteins were subjected to NMR analysis. Synthesis of a single rhamnosyl-arginine containing glycopeptide. Moisture-and air-sensitive reactions were conducted in flame-dried glassware under an argon atmosphere. Commercially available reagents and solvents were used without further purification. CH 2 Cl 2 was distilled from calcium hydride, and tetrahydrofuran (THF) was distilled from sodium benzophenone immediately prior to use. Dimethylformamide (DMF) was stored under argon in a flask containing 4 Å molecular sieves. Reactions were monitored by thin layer chromatography (TLC) with precoated Silica Gel 60 F 254 aluminum plates (Merck, Darmstadt, Germany) using UV light and methoxyphenol reagent (100 ml 0.2% ethanolic methoxyphenol solution and 100 ml 2 M ethanolic sulfuric acid) as the visualizing agent. Flash chromatography was performed using silica gel (35 to 70 m) from Acros Organics. Peptide purification by reverse-phase high-performance liquid chromatography (RP-HPLC) was performed on a JASCO purification system with a UV-visible-light detector (model UV-2075Plus) using a Phenomenex Aeris Peptide 5-m XB-C 18 column (250 by 21.2 mm). Analytical RP-HPLC was measured on a JASCO system with a Phenomenex Aeris Peptide 5-m XB-C 18 column (250 by 4.6 mm). In all cases, mixtures of water (eluent A) and acetonitrile (eluent B) were used as eluents; if required, 0.1% formic acid (FA) or 0.1% trifluoroacetic acid (TFA) was added. High-resolution electrospray ionization (HR-ESI) mass spectra were recorded on a Thermo Finnegan LTQ FT mass spectrometer or on a Bruker maxis apparatus equipped with a Waters ACQUITY ultrahigh-performance liquid chromatograph (UPLC) using a Kinetex C 18 column (2.6 m, 100 Å) at 40°C (Fig. 8).
Glycopeptide SGR Rha NAAIVK was synthesized using a Liberty Blue automated microwave peptide synthesizer, followed by on-resin glycosylation and deprotection (Fig. 8). For construction of peptide 1, 0.1 mmol of preloaded H-Lys(Boc)-2-chlorotrityl resin (loading concentration, 0.78 mmol/g) was applied. Cleavage of the Fmoc-protecting group was achieved with 20% piperidine in DMF (75°C, 35 W, 3 min). Fmoc-protected amino acids (5 eq) were activated for peptide coupling using 5 eq of ethyl (hydroxyimino)cyanoacetate (Oxyma Pure), 0.5 eq of N,N-diisopropylethylamine (DIPEA), and 5 eq of N,N=diisopropylcarbodiimide. All coupling reactions were conducted at 75°C and 28 W for 5 min. Removal of the allyloxycarbonyl-protecting group and subsequent coupling of the sugar moiety, as well as deprotection of the acetyl groups, were performed according to established procedures (26). Final deprotection gave the desired glycopeptide, SGR Rha NAAIVK, yielding 39% after HPLC purification. The amino acid sequence of the glycopeptide corresponds to the primary structure of the S. oneidensis acceptor loop, which is highly similar to the consensus sequence of EarP-arginine-type EF-Ps (17).
Antibody generation. Polyclonal antibodies were raised commercially by Eurogentec according to the Rabbit Speedy 28-day (AS superantigen) program. The mono-rhamnosyl-arginine-containing peptide was coupled to bovine serum albumin (BSA) according to an internal protocol (AS-PECO 05). Antibodies capable of binding to rhamnosyl-arginine were purified from rabbit sera by affinity chromatography (AS-PURIϪMED) against the glycopeptide SGR Rha NAAIVK. To test the specificity of the purified polyclonal antibodies toward EF-P Rha , 1.5 g of unmodified and 0.5 g of modified EF-P were transferred to a nitrocellulose membrane by Western blotting. While polyclonal antibodies that were raised against EF-P from S. oneidensis detect both unmodified and modified EF-P Ppu , anti-Arg Rha specifically detects the modified protein variant (Fig. S1A).
SDS-PAGE and Western blotting. Electrophoretic separation of proteins was carried out using SDS-PAGE as described by Lämmli (74). Separated proteins were visualized in gel using 0.5% (vol/vol) 2-2-2-trichloroethanol (75) and transferred onto a nitrocellulose membrane by vertical Western blotting. Antigens were detected using 0.1 g/ml anti-His 6 tag (Abcam, Inc.), 0.2 g/ml anti-EF-P, or 0.25 g/ml of anti-Arg Rha . Primary antibodies (rabbit) were targeted by 0.2 g/ml alkaline phosphatase-conjugated anti-rabbit IgG (H&L) (goat) antibody (Rockland). Target proteins were visualized by addition of substrate Determination of kinetic parameters. Kinetic parameters were determined by varying TDP-Rha concentrations while keeping concentrations of EarP Ppu (0.1 M) and unmodified EF-P Ppu (2.5 M) constant. A mixture of EarP Ppu and unmodified EF-P Ppu was equilibrated to 30°C in 100 mM NaP i (pH 7.6). The reaction was started by the addition of TDP-Rha and was stopped after 20 s of incubation at 30°C by the addition of 1 vol of 2ϫ Lämmli buffer (74) and incubation at 95°C for 5 min. Samples were subjected to SDS-PAGE, and rhamnosylated EF-P Ppu was detected as described above. Band intensities were quantified using ImageJ (76). Product formation (in nanomoles per milligram) was calculated relative to fully (in vivo) rhamnosylated EF-P Ppu . K m and k cat values were determined by fitting reaction rates (in nanomoles per milligram per second) to the Michaelis-Menten equation using SigmaPlot. Time course experiments conducted at a TDP-Rha concentration of 500 M show that the rhamnosylation reaction is not saturated after 20 s of incubation (Fig. S2A).
Fold recognition. Fold recognition models were generated using the online user interface of Phyre 2 (33,77), SWISS-MODEL (78)(79)(80)(81), and the I-TASSER server (34)(35)(36) as instructed on the websites. Model structures were selected from the array of results according to best confidence, Q mean, and z scores, respectively. All images of tertiary protein structures in this work were generated using the UCSF Chimera package developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (82). Protein structures were obtained as .pdb files from http://www.rcsb.org (83) or the respective modeling platforms mentioned above.
Determination of intracellular TDP-Rha concentrations. Cells were grown in 1 liter LB to an OD 600 of 0.5 (5 ϫ 10 8 cells/ml), harvested by centrifugation, and resuspended in 25 ml 100 mM NaP i (pH 7.6) (2 ϫ 10 10 cells/ml). After disruption of cells with a Constant Systems Ltd. continuous-flow cabinet at 1.35 kb, cell debris were removed by centrifugation and lysates were sterilized by filtration (Steriflip). A mixture of EarP Ppu (0.1 M) and unmodified EF-P Ppu (2.5 M) was equilibrated to 30°C in 10 l 100 mM NaP i (pH 7.6). The reaction was started by addition of 10 l lysate from~2 ϫ 10 7 or~2 ϫ 10 8 cells and stopped after 20 s of incubation at 30°C by addition of 1 vol 2ϫ Lämmli buffer (74) and incubation at 95°C for 5 min. In parallel, a TDP-Rha calibration series was generated by addition of TDP-Rha at final concentrations ranging from 5 M to 160 M, including the linear range of the rhamnosylation reaction rate (Fig. 5D). Samples were subjected to SDS-PAGE, and rhamnosylated EF-P Ppu was detected as described above. Band intensities were quantified using ImageJ (76). TDP-Rha concentrations in samples containing lysate were calculated by dividing the respective relative band intensities by the slope of the corresponding calibration curve (5 M to 80 M TDP-Rha). Intracellular TDP-Rha concentrations were calculated from the amount of substance (in moles) per cell, with an assumption of equal distribution of TDP-Rha across all cells as well as an average cell volume of 3.9 m 3 for E. coli (84) and 2.1 m 3 for P. putida and P. aeruginosa (85).
NMR spectroscopy and backbone assignment of EF-P and EarP. All NMR experiments were performed at 298 K on Bruker Avance III spectrometers with a magnetic field strength corresponding to a proton Larmor frequency of 600 MHz (equipped with a Bruker TXI cryogenic probe head), 700 MHz (equipped with a Bruker room temperature probe head), or 800 MHz (equipped with a Bruker TXI cryogenic probe head). All data sets were processed using NMRPipe (91).
Before NMR measurements of 15 N-and 13 C-labeled EF-P (700 M) in 100 mM NaP i , 50 mM NaCl, and 5 mM DTT (pH 7.6), 0.02% NaN 3 was added to the sample. Sequential resonance assignment was obtained from two-dimensional (2D) 1H-15N HSQC and three-dimensional (3D) HNCA, CBCACONH, and HNCACB backbone experiments, using a constant time during 13 C evolution (86). The assignment process was assisted by CARA (http://cara.nmr.ch) and CcpNmr Analysis (63), and 98% of the backbone resonances could be assigned. Missing assignments for residues other than prolines are S123, R133, N140, V164, D175, and G185. Secondary chemical shift analysis was performed based on the difference between measured 13 C ␣ and 13 C ␤ chemical shifts and random coil chemical shifts of the same nuclei to assign a secondary structure to the EF-P sequence (Fig. S3E) and confirm the validity of the model shown in Fig. 6 (87, 88). EarP Structure and Biochemistry ® Due to the size of EarP (43 kDa), backbone resonance assignment was possible only for 2 H-, 15 N-, and 13 C-labeled samples to reduce the number of protons and thus cross-relaxation effects, which also enables efficient acquisition of backbone assignment experiments in TROSY mode (89). TROSY-HNCA, -HNCACB, and -CBCACONH experiments (90), processed by NMRPipe (91) and analyzed using CARA (http://cara.nmr.ch), enabled backbone resonance assignment of 62% of all assignable residues (excluding prolines).
The NMR titrations were always performed by adding an unlabeled interaction partner to the 15 N-labeled protein sample and monitoring the progress of the titration by recording 1 H-15 N HSQC. First, 15 N-labeled 150 M unmodified EF-P was titrated with unlabeled EarP to a 1:2 EF-P/EarP molar ratio with intermediate steps at 1:0, 1:0.5, 1:1, and 1:1.5 EF-P/EarP molar ratios. 15 N-labeled 41 M rhamnosylated EF-P was titrated with unlabeled EarP to a 1:2 EF-P/EarP molar ratio without any intermediate steps. 15  STD NMR experiments were performed with 10 M WT EarP or mutants and either 70 M (1:7 ratio of protein to ligand to mimic SAXS conditions) or 1 mM TDP-Rha in 100 mM potassium phosphate buffer, pH 7.5, 150 mM NaCl, 1 mM DTT, and 10% D 2 O. The experiments were performed on a Bruker Avance III 700-MHz spectrometer equipped with a triple resonance (TXI) room temperature probe head at 277 K. Protein was saturated with 49-ms Gaussian pulses at the resonance frequency of methyl resonances at 0.592 ppm. The experimental results were collected after a total saturation time of 20 s, with 1,596 scans performed for the WT EarP sample with a 100-fold excess of ligand, and after a total saturation time of 5 s, with 4,096 scans performed for the WT EarP sample with a 7-fold excess of ligand. For EarP mutants, the experimental results were collected after a total saturation time of 4 s and with 128 scans.
Small-angle X-ray scattering. Thirty microliters of EarP, EarP plus TDP-rhamnose, and buffer (with and without TDP-rhamnose) were measured at 20°C at BioSAXS beamline BM29 at the European Synchrotron Radiation Facility using a 2D Pilatus detector. For each measurement, 10 frames with a 1-s exposure time per frame were recorded for each EarP and buffer sample, using an X-ray wavelength () of 0.9919 Å. Measurements were performed in flow mode, where samples are pushed through a capillary at a constant flow rate to minimize radiation damage. The protein concentrations measured were 1.0, 2.0, 4.0, and 8.0 mg/ml. TDP-Rha was used in a 7:1 excess (ligand to protein). The buffer measurements were subtracted from each protein sample, and the low Q range of 1.0 mg/ml was merged with the high Q range of the 8.0-mg/ml sample, using PRIMUS (92). The merging was done due to the rising scattering density at low Q ranges for the more highly concentrated samples, indicative of aggregation. CRYSOL (93) was used to fit the back-calculated scattering densities from the crystal structure to the experimental data.
X-ray crystallography. For crystallization, N-terminally His 6 -tagged EarP Ppu expressed as a selenomethionine derivative was used. The protein was dialyzed to 50 mM Tris, 100 mM NaCl, 1 mM DTT, pH 7.6, and concentrated to 183 M. TDP-Rha was added to a final concentration of 10 mM. The crystallization condition was 0.2 M ammonium acetate, 0.1 M bis-Tris (pH 6.0), and 27% (wt/vol) polyethylene glycol 3350. A full data set was collected at the ID29 beamline, ESRF, Grenoble, France, at a wavelength of 0.97 Å (the absorption peak for selenium) and with a 15.05% beam transmission with a 0.15°oscillation range, 0.037-s exposure time, and 2,400 frames. The space group was determined to be I4. The data set was phased using single-wavelength anomalous diffraction (SAD) by the Crank2 (94) automatic pipeline in CCP4 (95), using Afro provided by N. S. Pannu (unpublished) for substructure factor amplitude (FA) estimation, Crunch2 (96) for substructure detection, and Solomon (97) for density modification. The anomalous signal extended to a 3.4-Å resolution (in a data set with a 3-Å resolution). We could successfully find 3 Se-Met signals with an occupancy of 1 located in the C-terminal domain and 2 Se-Met signals with an occupancy of~0.5 located in the N-terminal domain. The initial structure was built in Phenix Autobuild (98), completed with several rounds of manual model building in Coot (99), and used as the model for molecular replacement (MR) of a native data set extending to 2.3 Å. Despite our rigorous efforts in manual model building, which included extreme density modification, use of homology models to model the N-terminal domain, Rosetta modeling, and refinement strategies with different refinement software (Phenix [98], refmac [100], and CNS [101,102] [and CNS-DEN-assisted refinement]), the structure displays an R-free of 35% at 2.3 Å, with large parts of the electron density in the N-domain not interpretable. No crystallographic pathology (twinning, anisotropy) could be identified in any of the multiple data sets that we obtained, and trying to interpret crystallographic symmetry as noncrystallographic symmetry by deliberately choosing space groups with lower symmetry (C2, P1) did not improve the density. This indicates intrinsic crystal disorder caused by the N-terminal domain adopting several conformations in different unit cells.
Accession number(s). Atomic coordinates and structure factors for the reported crystal structures have been deposited with the Protein Data Bank under accession number 5NV8.