Substrate recognition by the Pseudomonas aeruginosa EF-Tu–modifying methyltransferase EftM

The opportunistic bacterial pathogen Pseudomonas aeruginosa is a leading cause of serious infections in individuals with cystic fibrosis, compromised immune systems, or severe burns. P. aeruginosa adhesion to host epithelial cells is enhanced by surface-exposed translation elongation factor EF-Tu carrying a Lys-5 trimethylation, incorporated by the methyltransferase EftM. Thus, the EF-Tu modification by EftM may represent a target to prevent P. aeruginosa infections in vulnerable individuals. Here, we extend our understanding of EftM activity by defining the molecular mechanism by which it recognizes EF-Tu. Acting on the observation that EftM can bind to EF-Tu lacking its N-terminal peptide (encompassing the Lys-5 target site), we generated an EftM homology model and used it in protein/protein docking studies to predict EftM/EF-Tu interactions. Using site-directed mutagenesis of residues in both proteins, coupled with binding and methyltransferase activity assays, we experimentally validated the predicted protein/protein interface. We also show that EftM cannot methylate the isolated N-terminal EF-Tu peptide and that binding-induced conformational changes in EftM are likely needed to enable placement of the first 5–6 amino acids of EF-Tu into a conserved peptide-binding channel in EftM. In this channel, a group of residues that are highly conserved in EftM proteins position the N-terminal sequence to facilitate Lys-5 modification. Our findings reveal that EftM employs molecular strategies for substrate recognition common among both class I (Rossmann fold) and class II (SET domain) methyltransferases and pave the way for studies seeking a deeper understanding of EftM's mechanism of action on EF-Tu.

Protein post-translational modifications (PTMs) 3 add an additional level of complexity to the proteome and are critical to the biological functions of proteins in all domains of life. In bacteria, PTMs such as methylation are used to respond to environmental cues and may be critical for adaptation to or evasion of host immune systems during infection (1)(2)(3).
The bacterial methyltransferase CheR serves as the paradigm for environmental adaptation behavior arising from protein methylation (4). In many enteric bacteria, response to chemical stimuli occurs via O-methylation of glutamic acid residues on chemotaxis receptors by CheR. Bacterial methyltransferases can also directly influence host gene expression and host/ pathogen interaction. For example, Mycobacterium tuberculosis expresses a secreted 5-methylcytosine-specific DNA methyltransferase (Rv2966c) that directly modifies the host epigenetic machinery to alter transcription (5). Establishment of infection can also be both negatively and positively regulated by methylation of bacterial surface proteins by host-or pathogen-derived methyltransferases. The human protein SUV39H1, for example, methylates the crucial surface-exposed mycobacterial protein HupB, which reduces bacterial adhesion and survival inside the host cell (6). In Rickettsia, methylation of outer membrane protein B (OmpB) directly controls virulence (7): virulent species display lysine trimethylation of OmpB, which mediates host adhesion, attachment, and invasion, whereas avirulent strains possess monomethylated OmpB. Finally, our previous work has revealed a role in the establishment of Pseudomonas aeruginosa infection for the trimethylation of translation elongation factor thermo-unstable (EF-Tu) on its Lys-5 residue by the EF-Tu methyltransferase, EftM (8 -10).
Whereas the roles for protein methylation in bacterial virulence and host/pathogen interactions are emerging, there is generally less known about the molecular mechanisms that control substrate recognition and catalysis by the bacterial methyltransferase enzymes that incorporate these PTMs. The Gram-negative environmental bacterium P. aeruginosa is an opportunistic pathogen that causes serious respiratory, eye, and skin infections in vulnerable individuals, such as those with cystic fibrosis, compromised immune systems, or severe burns. P. aeruginosa is also intrinsically resistant to many antibiotics making infections difficult to treat once established. However, precisely how P. aeruginosa establishes infection in a host and adapts to its new environment is still not fully understood. A more complete understanding of this process could potentially identify novel strategies to restrict the establishment of infection.
Toward this goal, we previously described trimethylation of EF-Tu on Lys-5 as a PTM that enhances P. aeruginosa host-cell adherence and infection (8,9). In P. aeruginosa isolates from acute infections, as well as the laboratory strain PAO1, this EF-Tu modification is present in bacteria cultured at 25°C but absent at 37°C (8,9,11). This thermoregulation of PTM incorporation was found to occur at the level of transcription initiation as well as the direct thermolability of the modifying enzyme, EftM (8,10). In contrast, some chronic isolates, such as PAHM4, retain their ability to methylate EF-Tu at 37°C (11). However, whether this arises from increased thermal stability of EftM has not been tested. Together, these observations suggest that the major role of the Lys-5 PTM may be to modulate a "moonlighting" function of EF-Tu in host-cell adhesion by P. aeruginosa.
Here, we define the mechanism of EF-Tu substrate recognition by EftM. Evolutionary sequence analyses and computational modeling of EftM and the EftM/EF-Tu complex were used to direct mutagenesis and functional studies to identify the protein/protein binding interface, revealing that EftM requires more than just the N-terminal sequence of EF-Tu to bind and methylate its substrate. We also identify conserved residues in a substrate-binding channel that position the N-terminal sequence of EF-Tu for trimethylation on Lys-5. Collectively, our findings lead to a new mechanistic model for EftM/EF-Tu recognition reliant on both protein/protein docking and protein conformational changes to accommodate the EF-Tu N-terminal sequence, which together direct specific substrate recognition and modification. These functional studies represent an important step toward a better understanding of the molecular recognition mechanism of EftM and bacterial lysine methyltransferases in general.

EftM HM4 is functionally equivalent to EftM PAO1 but with increased thermostability
We previously reported that the temperature sensitivity of EF-Tu methylation by EftM in P. aeruginosa strain PAO1 (EftM PAO1 ) is due, at least in part, to the direct unfolding of the methyltransferase upon transition from environment-like (25°C) to host-like (37°C) temperatures (8). This direct thermolability also likely contributed to challenges we encountered in expression, purification, and handling of EftM PAO1 for biochemical and biophysical studies. Therefore, we decided to examine the folding and functional properties of the same enzyme (EftM HM4 ) from the chronic infection strain of P. aeruginosa PAHM4 in which EF-Tu methylation was previously observed at 37°C (9).
Using differential-scanning fluorimetry (DSF) we found that the T m of EftM HM4 is increased by ϳ5°, to 37°C (Fig. 1A), compared with the 31.5°C that we determined previously for EftM PAO1 (8). Additionally, as observed for EftM PAO1 , S-adenosyl-L-methionine (SAM) stabilizes EftM HM4 by ϳ5°C resulting in an unfolding T m of 41.5°C in the presence of the cosubstrate. Next, we asked whether this additional thermal stability of EftM HM4 directly correlates with its activity at an elevated temperature. As expected, EftM HM4 retained enzymatic activity after incubation at either 25 or 37°C, as observed previously, but its activity was ablated after incubation at 42°C for 5 min or more (Fig. 1B). Collectively, these data show the retained methylation of EF-Tu in PAHM4 at 37°C is due to an increase in the thermostability of EftM methyltransferase by ϳ5°, which allows the enzyme to remain active at the elevated temperature. Critically, these findings also show that EftM PAO1 and EftM HM4 are functionally equivalent enzymes and that the latter still exhibits thermoregulation but shifted to higher temperature.
As there is no known structure of any EftM family member, we searched for structures of the closest paralogs and identified the N,N-dimethyltransferase DesVI from Streptomyces venezuelae (PDB code 3BXO), which methylates a sugar intermediate in macrolide antibiotic biosynthesis, dTDP-3-amino-3,6-dideoxy-␣-D-glucopyranose (12). DesVI is 25.3% identical and 36.7% similar to EftM HM4 as calculated in EMBL EMBOSS, with both values ϳ1% higher compared with EftM PAO1 . A combined approach of homology modeling and threading was used, followed by energy minimization, to generate structural models of EftM PAO1 and EftM HM4 . This combination of methods, which is applicable to modeling in instances of relatively mod-Figure 1. EftM from P. aeruginosa clinical isolate strain PAHM4 is more thermodynamically stable compared with EftM from PAO1 strain. A, differential scanning fluorimetry of EftM HM4 (red circle) and EftM PAO1 (black circle) alone or in the presence of 1 mM SAM (EftM HM4 , tan triangle and EftM PAO1 blue triangle). The data for EftM PAO1 were previously reported (8) and are shown here for comparison with EftM HM4 . B, in vitro methylation activity of EftM HM4 after incubation at a designated temperature.

EF-Tu recognition by EftM
est sequence identity, together with the high-structural conservation of the Class I methyltransferase fold allowed generation of models of good quality as the basis for experimental analyses. The essentially identical modeled EftM PAO1 and EftM HM4 structures each consist of two domains: a Class I (Rossmann fold) methyltransferase domain containing a central sevenstranded ␤-sheet and an auxiliary domain comprising the N-terminal ␣-helix and a three-stranded antiparallel ␤-sheet formed by an extended sequence that links the fifth and sixth strands (␤5/6 linker) of the core fold (Fig. 2). The EftM models displayed the expected similarities with the structure of the DesVI modeling template but also revealed some important potential differences. In particular, the substrate-binding channel of EftM is substantially enlarged, consistent with its binding of an extended substrate (i.e. the N-terminal sequence of EF-Tu). As the structure of DesVI was solved with the methyltransferase cosubstrate SAM, the model also revealed the likely location of cosubstrate and the surrounding catalytic center of EftM (Fig. 2, B and C).
The amino acid sequences of EftM HM4 and EftM PAO1 are 96.4% similar and 88.7% identical, with 28 amino acids that vary between the two enzymes (Fig. 3). Almost all of these 28 amino acid differences occur on the periphery of the enzyme suggesting that a network of interactions may be responsible for the increased stability of the EftM HM4 , rather than one or a small number of individual amino acid changes. However, it is noteworthy that the changes involving the most physicochemically dissimilar amino acids are clustered in a region that connects the core methyltransferase fold and the auxiliary domain in the EftM model (Fig. 3, red spheres). This suggests a possible difference in regulation of protein flexibility between EftM HM4 and EftM PAO1 , which might be responsible, at least in part, for the thermostabilization of EftM HM4 .
In summary, EftM HM4 exhibits conserved activity, retained thermoregulation but with increased thermostability, and is predicted to have identical structure compared with the previously characterized thermolabile methyltransferase EftM PAO1 . Coupled with its greater ease of expression, purification, and handling, these factors lead us to use EftM HM4 as the optimal context to complete the detailed study described herein of EF-Tu recognition by EftM.

EftM requires more than just the EF-Tu N-terminal peptide for binding
To begin defining EF-Tu recognition by EftM, we first asked whether important interactions are made with the surface of the globular protein fold of EF-Tu or whether the isolated N-terminal peptide ( 1 MAKEKFERNKP 11 ; trimethylated Lys-5 is in bold font and underlined) alone is a suitable substrate for modification. First, a variant of P. aeruginosa EF-Tu lacking the first 11 amino acids of the N terminus (⌬11-EF-Tu) was generated to assess contributions of the EF-Tu surface to binding by EftM. To qualitatively assess binding, we compared the ability of EftM to form a stable complex with EF-Tu or ⌬11-EF-Tu using gel-filtration chromatography (Fig. 4, A and B). Complexes of both EftM/EF-Tu and EftM/⌬11-EF-Tu were observed to elute from the column before either free component due to their larger size, suggesting that EftM is able to bind the truncated EF-Tu protein.
We next quantified the binding affinity of EftM to WT P. aeruginosa EF-Tu and the ⌬11-EF-Tu truncated construct using isothermal titration calorimetry (ITC). The binding affinity (K D ) measured for EftM and WT EF-Tu was relatively weak (ϳ27 M; Table 1) as would be anticipated given the very high levels of the translation factor in bacterial cells. Remarkably, a similar affinity was also obtained with the truncated EF-Tu protein (Table 1), consistent with retained interaction even in the absence of the EF-Tu N-terminal sequence. Interestingly, however, although the weak binding precludes accurate determination of binding enthalpy (⌬H), binding of EftM to ⌬11-EF-Tu appears to have a significantly lower ⌬H (Ͼ10-fold lower) than for WT EF-Tu (Fig. 4C). This difference suggests binding-induced changes in the EF-Tu N-terminal peptide or EftM structure occur upon protein/protein interaction even though the EF-Tu N-terminal peptide does not appear to directly contribute to the binding affinity of the complex.

EF-Tu recognition by EftM
In a complementary approach, we used mass spectroscopy (MS) to discern whether peptides corresponding to the isolated N-terminal peptide sequence of P. aeruginosa EF-Tu could be methylated by EftM. Samples containing 10-mer peptide ( 1 MAKEKFERNK 10 ) alone or after incubation with EftM/SAM were subject to MALDI-TOF MS analysis. In both spectra, peaks corresponding to the unmethylated peptide (m/z 1280) were identified, but no methylated peptides were detected (pre-dicted m/z 1294, 1308, and 1322 for mono-, di-, and trimethylated peptide, respectively), even in the presence of enzyme and SAM (Fig. 4D). We additionally tested a shorter peptide ( 1 MAKEKF 6 ) that corresponds more precisely to the sequence we predict is accommodated within the EftM peptide-binding channel (see below). Although synthetic trimethylated peptide ( 1 MAKEK Me3 F 6 ) could be detected, demonstrating our ability to measure Lys-5-modified peptides by MS, only unmethylated peptide was detected after treatment with EftM and SAM (Fig.  S1), as for the longer sequence.
In summary, gel-filtration and ITC data show that EftM can bind EF-Tu lacking its N-terminal sequence, whereas MS analysis also reveals that EftM cannot methylate an isolated EF-Tu N-terminal peptide sequence. Together, these observations point to a critical role for an extended surface of the globular body of EF-Tu in substrate recognition by EftM.  . EftM recognizes and binds to more than just the EF-Tu N terminus. A, gel-filtration chromatography of EftM/EF-Tu complexes: EftM/EF-Tu (black), EF-Tu alone (gray-shaded), and EftM/⌬11-EF-Tu (green). B, Coomassie-stained SDS-polyacrylamide gel of gel-filtration fractions confirming elution of the EftM/EF-Tu in the first peak in the chromatogram (fractions at 13 and 14 ml). C, ITC analysis of binding of EftM and EF-Tu/⌬11-EF-Tu. Although weak-binding affinity precludes accurate determination of ⌬H, i.e. the intercept of the fit curve on the y axis in the lower plot for each titration, note the difference in the scale of this plot for each complex, which suggests that enthalpy of binding is significantly reduced in the absence of the EF-Tu N-terminal peptide in ⌬11-EF-Tu. D, MS analysis of a 10-mer peptide corresponding to the N-terminal peptide without (black) or with (orange) treatment with EftM and SAM in a methylation reaction.

EF-Tu recognition by EftM EftM interacts with an extensive surface of EF-Tu
We next generated a computational model of the EftM/ EF-Tu complex using the protein/protein docking software HEX 6.0 (13). To generate this model, a structure corresponding to ⌬11-EF-Tu was first created by deletion of N-terminal residues in the structure of P. aeruginosa EF-Tu (PDB code 4ZV4). Potential models of ⌬11-EF-Tu bound to EftM HM4 generated by HEX 6.0 were ranked according to calculated energies for shape and electrostatic complementarity. The top-scoring prediction was manually adjusted, e.g. by selection of optimal side-chain rotamers, to remove obvious clashes before energy minimization to generate the final model of the EftM/⌬11-EF-Tu complex (Fig. 5, A and B).
The final model was assessed based on amino acid evolutionary conservation within the EftM family, revealing a moderate-ly-conserved surface of EftM HM4 that is predicted to interact with EF-Tu (Fig. 5, B and C). In contrast, the predicted solventexposed surface of EftM exhibited lower overall conservation, whereas the predicted openings to the EF-Tu N-terminal peptide-binding channel have much higher amino acid conservation as would be expected for this functionally critical region of the enzyme (Fig. S2). The putative EftM/⌬11-EF-Tu binding interface is stabilized by multiple electrostatic interactions made by EftM HM4 residues Asp-76, Arg-81, Asp-91, and Asp-98. To experimentally test the role of these residues in EF-Tu binding, each was altered with a charge-reversal substitution (i.e. Arg to Glu or Asp to Lys). Additionally, His-86 within the same surface was substituted with Ala. This group of residues is distributed across most of the predicted interaction interface (Fig. 5B). Variant EftM HM4 proteins were expressed and puri- of EftM (yellow) and EF-Tu (light purple), which were mutated to test the predicted protein/protein interface. C, top, cartoon of EftM HM4 highlighting key structural features, colored as in Fig. 2. Bottom, surface representation with amino acid conservation of residues on the predicted EF-Tu-interacting surface of EftM shown with a gradient of green-white coloring (most conserved to most variable). D, representative reaction time course of EF-Tu methylation assay immunoblot (top) and Coomassie staining of methylation reactions (center). Bottom, quantification of immunoblot data from two replicate experiments. The reaction approaches completion at ϳ30 min (black arrowhead) for EftM HM4 , EF-Tu, and SAM concentrations of 0.25 M, 10 M, and 1 mM, respectively. E, representative trimethyllysine immunoblot (top) and quantification of replicates (bottom) for EF-Tu methylation with EftM HM4 amino acid substitutions on the interacting surface: D91K, D98K, and H86A. F, as in E but for EftM HM4 variants D76K, R81E, and D76K/R81E, and EF-Tu variants E69K, R208E, and E69K/R208E.

EF-Tu recognition by EftM
fied as for the WT protein, and their folding was assessed using thermal denaturation monitored by intrinsic fluorescence. Each protein had a similar unfolding profile and inflection point melting temperature (T i ; within Ϯ5°C) compared with WT EftM HM4 consistent with their correct folding. Additionally, because each substituted residue is located on the EftM HM4 protein surface and distant from both the SAM-binding pocket and EF-Tu N-terminal peptide-binding channel, we therefore ascribe any observed decrease in trimethylation activity to weakening of the binding interaction between EftM HM4 and its EF-Tu substrate at this predicted surface.
To measure trimethylation of EF-Tu by the variant EftM HM4 proteins, we used the established immunoblot assay with an anti-trimethyllysine antibody. First, to determine the optimal reaction conditions for comparison with WT EftM HM4 , a timecourse activity assay was performed using WT EftM HM4 (0.25 M) in the presence of excess EF-Tu substrate (10 M) and SAM cosubstrate (1 mM). This experiment revealed that under the conditions used, EF-Tu trimethylation neared completion at ϳ30 min (Fig. 5D), and this single time point was selected for use in all subsequent methylation assays. D91K, D98K, and H86A substitutions in EftM HM4 resulted in a severely reduced ability to trimethylate EF-Tu (Fig. 5E), with activity decreased to ϳ16, 13, and 17%, respectively, compared with WT EftM HM4 (Fig. 5E). D76K and R81E substitutions also caused a reduced ability to methylate EF-Tu, to ϳ40, and 15%, respectively, although double substitution of these residues in EftM HM4 -D76K/R81E further lowered activity to ϳ7%.
We next made individual substitutions of residues Glu-69 and Arg-208 in EF-Tu, which are predicted by the model of the complex to be near EftM HM4 residues Asp-76 and Arg-81, respectively. Consistent with the effects of the corresponding changes in EftM HM4 , charge-reversal substitutions E69K and R208E in EF-Tu also resulted in reduced methylation activity by WT EftM, dropping to ϳ55 and 20%, respectively (Fig. 5E). The double mutant, EF-Tu-E69K/R208E, again showed a more pronounced decrease in methylation activity (to ϳ11% of WT), mirroring the effects of the corresponding substitutions in EftM. Collectively, the activities of the EftM HM4 variants with targeted substitutions, and of WT EftM on variant EF-Tu proteins with corresponding changes, experimentally validate the predicted protein/protein interaction interface in our computational model of the EftM/⌬11/EF-Tu complex.

EF-Tu N terminus is accommodated in a conserved substratebinding channel within EftM
Using our model of the EftM/EF-Tu complex, we next asked how the EF-Tu N-terminal peptide might be positioned in EftM's substrate-binding channel. The EftM HM4 homology model has its substrate-binding channel in a similar location to that observed in the DesVI structure but with an overall larger volume and distinct surface shape and charge distribution. The residues lining the substrate-binding channel, Tyr-7, Asp-15, Met-74, Trp-170, and Trp-196 ( Fig. 6A and Fig. S2B), are highly conserved (Ͼ90%), whereas Asp-20 near the opening of the channel is also conserved (82.4%) in EftM family proteins. Additionally, although these residues are conserved in the EftM family, they are less highly conserved in DesVI and other paralogs, highlighting their potential role in interactions with the EF-Tu N-terminal peptide. Near the center of the channel, EftM also accommodates the SAM cosubstrate in an orientation consistent with other Rossmann fold methyltransferases, in which the methyl group is oriented toward the opening of the substrate channel adjacent to the modeled EF-Tu N terminus (Fig. 6B).
The cavity within EftM that forms the EF-Tu N-terminal peptide-binding channel appears large enough to accommodate the first six residues of EF-Tu ( 1 MAKEKF 6 ) with some distinctive surface pockets that can fit specific EF-Tu residues. We therefore next attempted to model this N-terminal sequence backbone path through the channel, positioning specific residues where possible (Fig. 6B). Two EF-Tu residues, Lys-5 and Phe-6, could be readily positioned with minimal clashes; in other cases, although the backbone path is clear, expansion of the channel would be necessary to accommodate the residue side chains. As discussed further below, we propose that interaction of EftM and EF-Tu may induce such widening of the peptide-binding channel.

EF-Tu recognition by EftM
To test the predicted roles of EftM residues within the substrate-binding channel in recognizing and stabilizing the EF-Tu N-terminal peptide, further EftM variants were generated. Tyr-7 and Met-74 are located next to the predicted path of the first three residues ( 1 MAK 3 ) of EF-Tu and may orient and stabilize the N-terminal sequence. Consistent with such a role, individual alanine substitution of each residue reduced EftM activity to 6.5 and 3.5%, respectively (Fig. 6, C and D). EF-Tu Lys-5 is placed in the model near the EftM catalytic center, as required for its modification, and proximal to Asp-15. Substitution of this EftM residue with alanine (EftM-D15A) also dramatically reduced EF-Tu methylation activity to ϳ5% of the WT enzyme (Fig. 6, C and D). The modeling also places EftM Trp-170 adjacent to EF-Tu Phe-6 and Trp-196 at the opposite end of the peptide-binding channel adjacent to the EF-Tu N terminus; as described in the following section, these residues also appear to play particularly critical roles in EftM activity. Finally, Asp-20 is located just outside the peptide-binding channel and could potentially interact with EF-Tu residues following Phe-6. Substitution of Asp-20 with alanine led to a more modest decrease in activity compared (ϳ40% of WT EftM activity; Fig. 6, C and D). These results suggest that interactions made to the first six N-terminal residues of EF-Tu are the most critical for positioning Lys-5 for methylation.

EftM Trp-170 and Trp-196 flank the EF-Tu N-terminal peptidebinding channel and are critical for EftM activity
The highly-conserved EftM residues Trp-170 and Trp-196 are positioned at each end of the EF-Tu N-terminal peptidebinding channel suggesting they may perform "gatekeeping" functions at the channel openings. We therefore investigated their potential roles in cosubstrate binding and EF-Tu recognition and modification. EftM HM4 Trp-170 and Trp-196 were individually substituted to phenylalanine or alanine, and the activity of each variant was assessed as before. First, in the immunoblot methylation assay, both Trp-170 substitutions, W170A and W170F, completely ablated EftM HM4 activity, whereas the equivalent changes at Trp-196 (W196A and W196F) resulted in a very low level of trimethylation activity, ϳ4 and 10%, respectively, compared with WT EftM HM4 (Fig. 6,  C and D).
We next used ITC to establish whether substitution of either residue disrupts SAM binding. The binding affinities (K D ) of SAM and S-adenosylhomocysteine (SAH) ( Table 1) for WT EftM HM4 were first determined and found to be essentially identical to those previously measured for EftM PAO1 (8). Similarly, no major difference was observed in SAM-binding affinity to either tryptophan to phenylalanine variant (Table 1), indicating that neither Trp-170 nor Trp-196 contribute meaningfully to cosubstrate binding.
To address whether either Trp-170 or Trp-196 contributes to EF-Tu binding affinity, the interaction of the translation factor with each phenylalanine substitution variant of EftM HM4 was also assessed by ITC. As discussed above, WT EftM HM4 binds EF-Tu with a K D of ϳ28 M, and the binding affinities of both W170F and W196F are essentially identical (Table 1). However, as observed for EftM HM4 interaction with the ⌬11-EF-Tu variant, a large decrease (Ͼ10-fold) in ⌬H was again apparent for EftM-W170F and EftM-W196F interaction with WT EF-Tu (Fig. S3). These findings are also consistent with our earlier interpretation that a large enthalpic component of binding arises from binding-induced re-organization of the EF-Tu N terminus into the substrate channel and, furthermore, that these two EftM Trp residues play an important role in this process.
Together, these data reveal that Trp-170 and Trp-196 play functionally critical roles in EftM activity, with alteration of the former residue, in particular, fully ablating EF-Tu methylation. These residues do not, however, contribute directly to the binding affinity for either substrate or cosubstrate. The proximity of Trp-196 to a conserved opening on the EftM surface (Fig. S2B ) suggests this residue could be required for interactions or conformational changes necessary for exchange of SAM/SAH during catalytic turnover or in stabilizing the most N-terminal residues of EF-Tu, together with Tyr-7 and Met-74. The absolute requirement for Trp-170 in EftM activity, together with our modeling of residues in the peptide-binding channel, suggests that Trp-170 interacts with the highly-conserved EF-Tu residue Phe-6 to precisely orient Lys-5 for modification. Consistent with this idea, alteration of EF-Tu with an F6A substitution also results in a dramatic loss of methylation by EftM (Fig. 6, C and D).

Binding of the EF-Tu N-terminal peptide for Lys-5 methylation
We next addressed the question of how the N terminus of EF-Tu accesses the EftM substrate-binding channel. At least two distinct potential mechanisms can be envisaged. First, following EftM/EF-Tu interaction, the EF-Tu N-terminal sequence could enter via an adjacent opening on the EftM surface (Fig. S2B) and thread into the substrate channel until fully accommodated (a "threading" model). Alternatively, binding to EF-Tu could induce conformational changes in EftM that open the peptide-binding channel allowing "placement" of the 1 MAKEKF 6 sequence directly into its final position (Fig. S4A). Given that EftM can methylate an N-terminally His 6 -tagged EF-Tu construct with similar efficiency to EF-Tu with an authentic N terminus (Fig. S5), the latter scenario appears most plausible, and we sought further evidence to more fully support this placement model.
To gain initial insight into potential points of flexibility in EftM that could be influenced by interaction with EF-Tu, we used the El Nemo server (14) to perform a normal mode analysis (NMA) on the EftM HM4 homology model structure. NMA provides a coarse model of large-scale protein movements and can identify potential hinge points and the directionality of domain movements. The NMA suggests that EftM HM4 can undergo a large-scale conformational change between the methyltransferase core fold and the ␤5/6 linker of the auxiliary domain ( Fig. 7A and Fig. S4B). This motion could potentially generate a more open conformation as well as a more closed conformation of the peptide-binding channel in EftM (Fig. 7A), compared with the starting homology model structure. We speculate that this movement toward the "open" conformation could be induced by EF-Tu binding and thus expand the substrate-binding channel to allow direct accommodation of the 1 MAKEKF 6 peptide sequence.

EF-Tu recognition by EftM
To further experimentally test the placement model, we used a construct of EF-Tu fused at its N terminus to the SUMO domain (Fig. 7B, SUMO-L0-EF-Tu). Our expectation was that if the channel is rigid and does not open upon EF-Tu binding then the relatively large, folded SUMO domain attached to the N terminus of EF-Tu would preclude entry through the opening on the EftM surface (Fig. S4C). Although methylation was reduced, EftM HM4 is still able to trimethylate Lys-5 in SUMO-L0-EF-Tu with reasonable efficiency (ϳ30% of WT EftM activity; Fig. 7, C and D). We next increased the linker length between SUMO and EF-Tu domains, to adjust the distance between the SUMO domain and the EftM/EF-Tu complex. Two constructs with either a three-residue (SUMO-L3-EF-Tu) or a six-residue (SUMO-L6-EF-Tu) glycine/serine-rich linker both showed a further modest recovery of activity compared with SUMO-L0-EF-Tu (ϳ50% compared with WT EF-Tu; Fig.  7, C and D). Collectively, these results support the model of placement of the EF-Tu N terminus into a widened peptidebinding channel in EftM upon its binding to EF-Tu.

Discussion
Some bacteria that infect the host respiratory tract present the phosphorylcholine (or, choline phosphate; ChoP) modification on their cell surface to enhance airway epithelial cell adhesion and thus promote the establishment of infection (11, 15,16). P. aeruginosa, in contrast, does not present ChoP but instead uses the surface-exposed protein synthesis factor EF-Tu with trimethylated Lys-5 as a molecular mimic of this PTM that can similarly bind to the epithelial cell platelet-activating factor receptor as a key adhesin in infection (9, 11). How EF-Tu is exported is unclear as it lacks a canonical secretion signal, suggesting an alternative mechanism of selective transport to the cell surface. In this regard, EF-Tu's association with the P. aeruginosa type VI secretion system (17,18) is intriguing and warrants further investigation. Whether trimethylation of Lys-5 is required for transport to the cell surface is also not definitively known. However, EF-Tu is cell-surface-associated in numerous microbes (19), including many that lack EftM. Furthermore, studies of intact P. aeruginosa PAO1 have revealed that EF-Tu is cell-surface-associated in both trimethylated and unmethylated forms as it is found in this moonlighting role regardless of growth temperature and thus Lys-5 methylation status (11,20). Collectively, these results suggest that, unlike the established role of the Lys-5 PTM in host cell adherence once on the cell surface, transport of EF-Tu to that extracellular location is likely not dependent on trimethylation.
Prior to this work, the mechanism of EF-Tu substrate recognition by EftM was unclear. More broadly, the mechanisms of action of bacterial lysine methyltransferases are currently not well-defined in sharp contrast to their eukaryotic counterparts, which play a prominent role in epigenetics via histone protein modification (21). Here, we showed that EftM from the chronic infection strain PAHM4 possesses conserved activity and retained thermoregulation, albeit with modestly (ϳ5°C) increased thermostability, compared with the previously studied EftM PAO1 enzyme, making it suitable for detailed structural and mechanistic studies of this enzyme family. Our studies revealed that the isolated EF-Tu N-terminal peptide sequence is not a suitable substrate for EftM, but instead, specific sub-

EF-Tu recognition by EftM
strate recognition and modification are reliant on interaction via an extended protein/protein binding interface. Overall, our results evoke a model in which protein/protein docking via an extended surface is required for interaction of enzyme and substrate, and subsequent modification of EF-Tu Lys-5 (Fig. 8). Furthermore, this interaction may induce conformational changes that open EftM peptide-binding channel to allow access to the EF-Tu N-terminal peptide sequence. Within this channel, several highly-conserved residues, including two functionally critical Trp residues, appear to act in concert to organize the 1 MAKEKF 6 sequence and thus precisely orient the Lys-5 target residue for modification (Fig. 8).
The mechanism of EF-Tu recognition and Lys-5 methylation by EftM revealed by our work may also have broader implications for our understanding of other lysine methyltransferases belonging to both the Class I Rossmann fold-and Class II SET domain-containing methyltransferases. Specifically, our findings suggest that EftM may exploit features of molecular mechanisms common to both the Rossmann fold methyltransferases, in its requirement for an extended protein surface for substrate recognition, and the SET domain methyltransferases, with its substrate-binding channel that orients the modified residue within in a flexible loop or disordered protein region (22).
Despite recent discoveries of extensive methylation in bacterial proteins, including lysine methylations, and their roles in bacterial physiology and infection (4,7,23,24), there still exists a significant gap in our mechanistic understanding of bacterial lysine methyltransferases. The majority of Class I Rossmann fold protein methyltransferases recognize the globular body of their protein substrate and possess a shallow and relatively surface-exposed catalytic center. For example, the ribosomal protein L11 methyltransferase, PrmA, modifies three target sites in its substrate, including two lysine residues (Lys-3 and Lys-39), exploiting recognition of the globular L11 protein fold and accommodating its structurally distinct target sites in a shallow catalytic center (24). This enzyme-substrate recognition mechanism thus allows the necessary flexibility in target selection for both Lys-3, which resides in an unstructured N-terminal region, and Lys-39, which is located within an ␣-helix. The OmpB methyltransferase in Rickettsia also recognizes the globular body of its substrate to allow catalysis of multiple lysine methylations within its shallow catalytic site (25). A similar rec-ognition mechanism is also observed in eukaryotic Class I Rossmann fold methyltransferases such as DOT1L, which methylates histone H3 Lys-79 within the nucleosome using its N-terminal domain to bind the substrate (26). The recognition mechanism and subsequent catalysis of these methyltransferases follows a "catch and catalyze" strategy, which is mostly dependent on globular body recognition, associated conformational changes, and orienting substrate in the shallow catalytic site for methylation (26). Although our studies have shown that EftM also requires an extensive protein/protein interface to bind EF-Tu, its catalytic center is located within a deep substrate-binding channel. This may reflect the fact that, unlike the bacterial lysine methyltransferases noted above, EftM is very specific with a single trimethylation target site at Lys-5. As such, in this latter aspect of substrate recognition, EftM functions in a manner more similar to the SET domain methyltransferases.
Many SET domain methyltransferases can bind and catalyze modification of short, isolated peptide sequences containing their target site residue. The histone-lysine N-methyltransferase SETD2 specifically trimethylates Lys-36 of histone H3 (H3-K36 me3 ), and the enzyme is active on a 14-residue dimethylated (H3-K36 me2 ) peptide as substrate (27). Upon binding to the H3 peptide, the catalytic tunnel is proposed to undergo stepwise conformational changes consistent with a placement model, accommodating and orienting the peptide (27). Similarly, EHMT1, which specifically mono-and dimethylates Lys-9 of histone H3 (H3K9 me1/me2 ), can recognize and methylate an 11-residue peptide from its target as the substrate, again with a mechanism consistent with the placement model (28). Other SET domain methyltransferases have additional domains fused to the catalytic SET domain or act as a part of multiprotein complexes, in which both the recognition of globular domain of protein(s), induced conformational rearrangements, and placement of substrate motif into deep channels are observed (29). Our findings reveal that EftM has a deep peptidebinding channel like the SET domain methyltransferases, and the EF-Tu N-terminal sequence is accommodated directly, likely with the aid of conformational changes induced by protein/protein interaction (Fig. 8).
Our current studies have revealed important details of substrate recognition by EftM and suggest that its mechanism of substrate recognition and modification are more complex than other known bacterial lysine methyltransferases. Specifically, Figure 8. Model of EF-Tu recognition by EftM. EftM binds EF-Tu via an extended protein/protein interface, promoting conformational changes in EftM that open the peptide-binding channel of the free form to a more open state that allows direct placement of the EF-Tu N-terminal peptide. Within the substratebinding channel, the N-terminal peptide of EF-Tu is oriented by a group of highly-conserved residues that position Lys-5 for trimethylation. The dotted arrows denote the subsequent steps needed for additional methylation of Lys-5 to generate dimethyllysine and then the final trimethyl modification. SAM/SAH must be recycled for each methylation reaction but whether EF-Tu remains fully bound (in a processive mechanism) or is partially or fully released (in a distributive mechanism) remains to be clearly defined.

EF-Tu recognition by EftM
EftM appears to more closely resemble the SET domain methyltransferases of eukaryotic origin in key aspects such as its EF-Tu N-terminal peptide-binding channel. We also note that the likely catalytic center of EftM is composed of multiple tyrosine and histidine residues, which are also found in the SET domain methyltransferases. Further investigation using highresolution structural and careful biochemical analyses will be required to fully define the molecular basis of specific substrate recognition as well as other aspects of EftM's mechanism of catalysis ( Fig. 8).

Computational modeling of EftM structure and the EftM/EF-Tu complex
A hybrid molecular modeling strategy was used to generate homology models of the EftM PAO1 and EftM HM4 protein structures. First, homology modeling was performed using SWISS-MODEL (30) to generate complete structures for each protein.
Inspection of these structures revealed likely incorrect modeling of the ␤5/6 linker region (e.g. multiple aromatic residues exposed to solvent), and this region was thus remodeled using protein threading with ITASSER (31). The model for each EftM protein was further examined in Coot (32) to confirm favorable side-chain conformations and minimize clashes before a round of energy minimization in UCSF Chimera (33).
A structure corresponding to ⌬11-EF-Tu from P. aeruginosa PAO1 was generated by deletion of the residues 8 -11 from PDB code 4ZV4 (note, the first seven residues are disordered and were already absent in the deposited structure). This ⌬11-EF-Tu structure and the EftM HM4 homology model were used as target and ligand, respectively, in the protein/protein docking software HEX 6.0 (13) to identify an unbiased docking interface. The models were ranked based on their total energy score, which combines calculated energies derived from both from shape complementarity and electrostatics. The top-scoring model of the complex (E ϭ 212.6 kcal/mol) was energy-minimized in the UCSF Chimera (33) and then examined in Coot (32) to confirm favorable side-chain conformations. This model exhibits high electrostatic and surface shape complementarity between EftM HM4 and ⌬11-EF-Tu, and the close placement of the EF-Tu N-terminal sequence to EftM gives further confidence in its likely accuracy.
The CAVER 3.0.1 plugin (34) in PyMOL with shell depth and a radius of 4 Å each, clustering threshold of 3.5 Å, and the channel origin set at the SAM binding pocket, was used to visualize substrate-binding channel in EftM.
NMA, a fast and simple method to calculate vibrational modes and protein flexibility, was performed on the EftM HM4 homology model using the El Nemo server (14). NMA is an implementation of an elastic network model in which the protein residues are modeled as point masses or pseudo-atoms (C␣ only) connected by springs, which represent the interatomic force fields. The springs connecting each node to all other neighboring nodes are of equal strength, and only the atom pairs within a cutoff distance (within 8 Å to each other, its interaction sphere) are considered connected. The normal modes of this "ball and spring" system are calculated using a Hessian matrix. Although this treatment is inherently simple and does not capture detailed protein dynamics, comparisons of low-frequency normal modes and the directions of largeamplitude fluctuations in molecular dynamics simulations have revealed qualitatively similar results (35). The first six NMA modes are rotational and translational, whereas modes 7-12 reflect overall long-range motions within the protein. For EftM HM4 , the resulting modes 7-12 were analyzed to identify potential regions of flexibility and possible conformational changes in the protein structure.

EftM evolutionary sequence analysis
EftM sequences were retrieved by BLAST search using EftM PAO1 as the query in UniProt. A set of 177 unique amino acid sequences was then selected by applying a 90% sequence identity cutoff in CD-HIT (36). Calculations of site-specific conservation were made using Geneious Prime (37) for all EftM sequences and used to calculate the position-specific Shannon entropy of the EftM protein family. The normalized Shannon entropy is given by Equation 1, where p i (l) represents the frequency of i class of residues at position l in the multiple sequence alignment, and n represents the number of amino acid groups depending the classification criteria. For our calculation, n ϭ 20 was chosen (considering each amino acid unique). The higher entropy implies higher variability in the given position in multiple sequence alignment and vice versa. The calculated amino acid conservations were mapped onto the surface of the EftM HM4 homology model using the PDB B-factor column with a white to dark green gradient representing increasing evolutionary conservation.

EftM and EF-Tu cloning and site-directed mutagenesis
Plasmids encoding EftM PAO1 /EftM HM4 (pColdII vector) and EF-Tu (pJP04, pDEST14-N-His 6 tufB) for recombinant protein expression in Escherichia coli or P. aeruginosa were previously reported (9). Site-directed mutagenesis of EftM HM4 and EF-Tu was performed using a mega-primer whole-plasmid PCR approach (38,39), and the variant proteins were expressed from the same parent vectors as the WT proteins. A construct for expression of tag-free EF-Tu was generated by PCR amplification and subcloning the P. aeruginosa PAO1 EF-Tu (tufB) coding sequence from plasmid pJP04 (9) into the pE-SUMO vector (LifeSensors). This construct produces His 6 -SUMO-EF-Tu protein ("SUMO-L0-EF-Tu") from which the authentic EF-Tu N terminus is generated by cleavage of the His 6 -SUMO tag by the ubiquitin-like protease (Ulp). Equivalent constructs with linker sequences extended by three or six additional amino acids between the SUMO domain and EF-Tu ("SUMO-L3-EF-Tu" and "SUMO-L6-EF-Tu", respectively) were generated using the mega-primer whole-plasmid PCR mutagenesis approach.
For expression of His 6 -EF-Tu, P. aeruginosa PAO1⌬eftM cells were freshly streaked on an LB-agar plate. A single colony was used to inoculate LB (50 ml), and the culture was incubated at 37°C for ϳ8 h (to A 600 ϳ0.8). Cells were then pelleted in microcentrifuge tubes and washed three times with 300 mM sucrose (1 ml) at room temperature. Sucrose-washed cells (50 l) were transformed with pJPO04 plasmid (1.5 l) by electroporation and allowed to recover in SOC medium (1 ml) at 37°C for 1 h before plating on LB-agar plates containing 300 g/ml carbenicillin. A single colony was then grown in LB supplemented with 0.1% arabinose and 300 g/ml carbenicillin for 18 h at 25°C before harvesting. Expression of His 6 -⌬11-EF-Tu was accomplished in E. coli BL21-AI cells transformed with pDEST14-⌬-11tufB N-His 6 fusion in ZYM-5052 medium supplemented with 0.2% arabinose (w/v) and carbenicillin (100 g/ml) and incubated with shaking at 25°C for 14 h. The His 6 -EF-Tu and His 6 -⌬11EF-Tu proteins were extracted and purified using the same chromatographic procedures as described above for the EftM proteins.
SUMO-fusion EF-Tu proteins were expressed in E. coli BL21 (DE3) transformed with the pE-SUMO-EF-Tu vector (L0, L3, or L6) and grown at 37°C in LB supplemented with ampicillin (100 g/ml) to mid-log phase (A 600 ϳ0.4 -0.6). Protein expression was induced by addition of IPTG (1 mM) and cultures were grown for 4 -5 h at 37°C. SUMO-fusion EF-Tu proteins were purified using the same chromatographic procedures as described above for EftM. Where needed, the His 6 -SUMO tag was removed from partially purified His 6 -SUMO-L0-EF-Tu following the Ni 2ϩ -affinity chromatography step by overnight digestion with Ulp (0.1 g of protease per 1.25 mg of fusion protein) at 37°C. The free His 6 -SUMO tag was removed from EF-Tu by passing the protease cleavage reaction back over a HisTrap-FF column before further purification of EF-Tu by gelfiltration chromatography.
Methylation assays for testing the substrate recognition model were performed at 25°C and contained EF-Tu (10 M), EftM (0.25 M), and 1 mM SAM in the same buffer as above. For the methylation reaction time course with WT EftM HM4 (Fig.  5D), samples were removed at each time point (2,4,8,15,30, and 60 min) for analysis. All subsequent EF-Tu methylation assays with WT and variant EftM HM4 and EF-Tu proteins were performed for 30 min. Reactions were quenched in 2ϫ SDS-PAGE buffer and resolved on two 10% SDS-polyacrylamide gels, one for immunoblotting and a second for staining with Coomassie. After transfer onto a polyvinylidene difluoride membrane, blots were blocked using 5% nonfat milk in PBS with 2% Tween 20 at room temperature and then probed overnight at 4°C with a rabbit polyclonal anti-trimethyllysine antibody (Immunchem, ICP0601) diluted 1:10,000 in the same buffer. Blots were probed for 1 h at room temperature with a horseradish peroxidase-conjugated goat anti-rabbit secondary antibody (Sigma; A0545) diluted 1:10,000 in the same buffer. The blots were treated with enhanced chemiluminescence reagent (Thermo Fisher Scientific), and the intensity of bands was analyzed on a Bio-Rad ChemiDoc TM Imager and quantified using ImageQuant software (GE Healthcare). Two replicates using independent preparations of protein for each methylation reaction were performed and quantified, and figures show one representative blot for each protein variant.

Protein stability measurement
DSF was used to compare the thermal stability of EftM PAO1 and EftM HM4 . Proteins samples (24 M) were subject to a linear temperature gradient (0.5°C/min from 25 to 75°C), and fluorescence of the hydrophobic residue-binding dye SYPRO Orange (5000-fold dilution) measured on a Step One Plus Real-Time PCR instrument (Applied Biosystems). DSF measurements were performed for each protein in the absence and presence of SAM (150 M) in three replicates, together with corresponding control experiments without protein under each condition. The first derivative of the melting curve from each replicate was calculated using GraphPad Prism software to determine the mid-point melting temperature (T m ). The data for EftM PAO1 were reported previously (8) and are shown here for comparison with EftM HM4 .
For protein quality control and analysis of variant protein folding compared with the corresponding WT protein, intrinsic fluorescence (at 350 and 330 nm) was monitored during unfolding over a linear temperature gradient (35-95°C) over 3 min on a Tycho NT.6 instrument (NanoTemper). The temper-

EF-Tu recognition by EftM
ature at which a transition occurs in a plot of the first derivative of the fluorescence ratio (350/330 nm), termed inflection temperature (T i ), was determined by the instrument software and compared for each protein.

Analysis of protein binding by gel-filtration chromatography
Gel-filtration chromatography was performed to qualitatively assess the binding of EF-Tu or ⌬11-EF-Tu with EftM HM4 . EftM and EF-Tu were mixed in a 500-l reaction and resolved on a Superdex 200 column (GE Healthcare) using 50 mM Tris, pH 7.5, buffer containing 150 mM sodium chloride, 5 mM ␤-mercaptoethanol, and 20% glycerol. One-ml fractions were collected, and a sample of each was run on an SDS-polyacrylamide gel and stained with Coomassie to confirm coelution of EftM and EF-Tu in the first peak.

Isothermal titration calorimetry
SAM and SAH were each dissolved in gel-filtration buffer to 1 mM final concentration for EftM-cosubstrate binding experiments. EF-Tu and ⌬11-EF-Tu were concentrated to ϳ500 M in the same buffer for protein/protein binding experiments. These ligands (SAM, SAH, or EF-Tu proteins) were titrated into EftM (30 -50 M) in 16 ϫ 2.4-l injections using an Auto-iTC 200 microcalorimeter (Malvern/MicroCal) at 25°C. After accounting for the heat of dilution by subtraction of the residual heat measured at the end of the titration, the data were fit using a model for one set of sites to determine the binding affinity (K d ).

Mass spectrometry analysis of EF-Tu peptide modification
Reactions containing 10 M EftM, 10 M peptide, and 1 mM SAM were incubated at 25°C for 30 min. Peptide or methylation reaction was spotted onto the source target with 10 mg/ml ␣-cyano-4-hydroxycinnamic acid matrix in a 50% acetonitrile, 0.1% TFA solution. MS data were collected in positive ion mode on a Bruker UltraFlex II MALDI instrument.