Structural basis for arginine glycosylation of host substrates by bacterial effector proteins

The bacterial effector proteins SseK and NleB glycosylate host proteins on arginine residues, leading to reduced NF-κB-dependent responses to infection. Salmonella SseK1 and SseK2 are E. coli NleB1 orthologs that behave as NleB1-like GTs, although they differ in protein substrate specificity. Here we report that these enzymes are retaining glycosyltransferases composed of a helix-loop-helix (HLH) domain, a lid domain, and a catalytic domain. A conserved HEN motif (His-Glu-Asn) in the active site is important for enzyme catalysis and bacterial virulence. We observe differences between SseK1 and SseK2 in interactions with substrates and identify substrate residues that are critical for enzyme recognition. Long Molecular Dynamics simulations suggest that the HLH domain determines substrate specificity and the lid-domain regulates the opening of the active site. Overall, our data suggest a front-face SNi mechanism, explain differences in activities among these effectors, and have implications for future drug development against enteric pathogens.

8. page 6, line 130 and 136: Instead, please state the form in which the crystal structure was obtained for each protein. For example: NleB2 was solved in its unliganded form. SseK1 was solved in complex with UDP. SseK2 was solved in its unliganded form and in complex with UDP and UDP-GlcNAc. 9. page 6, line 132: 'the N-terminus of SseK1 and SseK2 were truncated, respectively.' Please explain the reason of that.
10. page 7, line 143: 'because their datasets were obtained at a higher resolution." And the authors have three different snapshots of the active site.
11. page 7, line 145: 'root-mean-square deviation value of the structure is ~2.0 Å based on DALI pairwise comparison'. Please include the r.ms.d. values, Z-score as well as number of compared residues.
13. page 7, line 156: 'However, in the complex structure with the UDP or UDP-GlcNAc, the substrate leads to an unambiguous electron density map for the C-terminal lid domain,'. Please specify the protein complexes.
14. page 8, line 186: "In the SseK3 structure, SseK2-like π-π stacking interaction is conserved and Phe190SseK3 and Trp52SseK3 (corresponding to Phe203SseK2 and Trp65SseK2) participate in an interaction with the uridine group in the same orientation." A good opportunity to introduce a Figure 3 panel showing the arrangement of these aromatic residues into the SsekK3 active site.
15. page 9, line 214: What is the experimental evidence that support the substrate binding/product release order for SseK1, SseK2, SseK3 or NleB2? I agree with the authors that in most of the cases, GTs follow an order mechanism with the donor substrate binding first. However, in the absence of experimental enzymatic/kinetic data, the authors should be cautious with the interpretation of the structural data. A statement should be added accordingly.
16. page 9, line 229: 'Based on these data, we suggest that the conformation of the C-terminal domain is highly similar between SseK2 and SseK3.' Please remove this sentence.
17. page 10, line 245: 'STD NMR experiments of SseK1 and SseK2 in the presence of the peptides FADD110-118,  showed that all three peptides bound to both enzymes in the presence and absence of Mn2+ and UDP'. Please rephrase the sentence.
18. page 10, line 254: 'notion that that a' by 'notion that a' 19. page 13, line 323: I would suggest the authors to introduce here a section focused into the 'Catalytic mechanism of SseK1, Sske2 and NleB2'. In that context, please move/introduce the entire 'Anomeric configuration of glycosylated peptides' section here, and renumber Figure 1 accordingly. After the discussing the catalytic mechanism, the authors could introduce the section 'Catalytic importance of HEN motif'.  -Space group -Please follow international table of Crysrtallography, -Cell dimensions -Please check the use of points, -I/sigmaI: The high resolution shells in NleB2 and SseK1-UDP might be overestimated. I would suggest to process the data to high resolution and run few more cycles of refinement, -Please add Wilson B-factor, Refienement: -In Resolution, please add the range for the high resolution shell, -In number of reflections, please state if the number of reflections used in refinement and number of reflections for the high resolution shell, -In Rwork/Rfree please add % unit and the corresponding values for hte high resolution shell, -Please add Number of reflections used for Rfree, Ramachandran favoured (%), Ramachandran outliers (%) and Clashscore.
Reviewer #2 (Remarks to the Author): The authors have done a great job attempting to refine the aMD by inclusion of GaMD functions. The work is well executed and manuscript is well structured and written. The findings of this report would definitely assist in the understanding of the arginine glycosylation protein structures and function. But there are minor issues with the paper in the present form. #1. The abbreviation of "Gaussian Accelerated Molecular Dynamics" is "GaMD". #2. In figure 4a, the authors compared the open and close forms of SeeK. How about the delta PMF between the open and close forms of SeeK ? The authors can use the GaMD results to calculate the delta PMF. #3. In the method section, the authors should show the detail setting and information of GaMD.
The manuscript describes structural and functional studies of an important class of bacterial glycosyltransferases that are used by these bacteria to specifically suppress host defense. This topic is of interest to a broad readership. The crystallographic studies are sound and reveal many details of donor-substrate enzyme interactions. Complementary biophysical methods, mainly NMR, have been used to gain more insight into the function of these interesting enzymes. The two most important conclusions of the studies are 1/ the retaining character of the enzymes can be experimentally proven, and 2/ the data support a front-face SNi type mechanism. MD simulations are finally used to present a comprehensive functional model.
The data presented are new and deserve publication. I also think that Nature Communications would be the right journal to present this study. However, there are some shortcomings and questions, which need to be addressed prior to publication as this is detailed in the following.

General comments:
I suggest that the authors phrase their statements on the proposed SNi mechanism more carefully, as no direct experimental evidence has been presented.
In general, ITC experiments provide enthalpic and entropic signatures of binding reactions. Unfortunately, this analysis has not been done although the raw data seem to be available. Some of the ITC thermograms in Supplementary Figure 1 look certainly like they are suitable for analysis of binding enthalpies and entropies. This full analysis should be done where possible. On the other hand, some of the thermograms, e.g. the ones reflecting binding of UDP-Glc and UPD-Gal seem so suffer from artifacts. What is the reason for the discontinuities and could these be corrected? Experimental details for the ITC experiments are missing and should be included. Data analysis with Origin also provides Chi2 values for the dissociation constants, which should also be included. It would be interesting to compare the thermodynamic signatures of binding of activated sugars to these glycosyltransferases with existing literature data. If for some reason a full data analysis of the ITC data is impossible the authors should explain this.
Given that the importance of the experimental determination of the alpha-configuration of the D-GlcNAc-Arg linkage is so much stressed in the paper, why wasn't the complete set of chemical shifts of the glycopeptide obtained enzymatically reported? From the reading the paper I would assume this data set exists. Why not report it?
On p. 11, first para, it is mentioned that STD NMR experiments indicate a "significant conformational rearrangement of the peptide ligand". I do not agree with this conclusion. A change in binding epitopes may have many causes. Only one of them would be a conformational change.
On p. 11, second para, it is stated that "Hence, it is clear that differences in glycosylation of full length ...are not due to differences in binding modes ...". I simply cannot follow this argument, as glycosylated peptides or proteins have not been studied here.
On p. 15, Proposed mechanism: It is not quite correct that there is no experimental evidence for double displacement SN2. For human blood group B galactosyltransferase, a covalent intermediate has been described for the E303M mutant of this enzyme. There is a relatively recent review on this general topic by Ardevol et al. (2016). Maybe this is a good reference to get the non-expert reader involved.

Questions regarding Online Methods:
p. 20, NMR spectroscopy: Which reference has been used for chemical shifts? For measurements in H2O: how much D2O has been added for the lock?
p.20, line 533: The strength of the B1 field and the flip angle used for the Gaussian pulse in the cascade should be given.
p.20, line 534: "spoil sequence" probably means spinlock filter. How long was the spinlock field?
p. 21, line 539: I find it a little surprising that at 2mM MnCl2 good NMR spectra have been obtained. Is the concentration cited really OK? Compare the settings for the STD NMR experiments.
p. 21, line 538-547: What was the digital resolution of the 2D spectra used to determine the one-bond C-H coupling of C1 of GlcNAc?
Minor points: p. 6, line 122: It should read "UDP-GlcNAc/MnCl2" p. 28, line 705: Here and throughout the text "a-GlcNAc" should be substituted with "a-D-GlcNAc" Fig. 2: Replace "bounded" with "bound" or "bound to"; applies to Fig. 3, too. p.6., line 130: "substrate" should be specified. It is the "donor substrate". p. 8, line 182 to 185: The Kd for binding of UDP-GlcNAc to SseK2 wt vs. the F203A mutant was not reduced but increased. Sentence should be rephrased, e.g. "While the Kd measured for binding of UDP-GlcNAc to SseK1 F187A was almost the same ..., the Kd for binding of UDP-GlcNAc to ..." p. 8, line 192: What is a "weaker Kd"?. Please correct the subscript "s". p. 10, line 235: Subsitute "bounded" with "bound". p. 10, line 247: I could not find STD NMR spectra in the absence of Mn2+ and UDP. Either these spectra are included or it should be said "data not shown". In this paper, Park et al reported the type III secretion system effector proteins of NleB family are retaining glycosyltransferases following a SNi mechanism. Recently, the structure of SseK3 was determined, revealing a GT-A fold. However, the specific enzyme mechanism and the identification of the catalytic base remain unclear. There are also discrepancies regarding whether these enzymes are retaining or inverting GTs because this has not been experimentally probed. In addition, details regarding substrate specificity based on structural evidence are also limited due to the lack of ternary complexes. Park et al want to address these remaining major questions of this field. However, there are several flaws need to be properly addressed. Validation the linkage form of the glycoside on the substrate protein is the key step to analyzing the enzymatic mechanism. The authors did not get the real structure of enzyme-sugar ligand-substrate ternary complex. And the substrate should be the full death domain proteins but not the peptides, because there is no evidence to show the peptides can be used as substrates and be hundred percent modified as the death domains (such as TRADD DD, FADD DD). So the methodology and the data cannot fully support the conclusion of the enzymatic mechanism of this newly reported post translational modification (GlcNAcylation on Arginine). I would like to recommend the editors of Nature communication considering a resubmission when the authors can provide crystal or NMR structures of the NleB/SseK-UDP-GlcNAc-DDs ternary complex. Other points: Q1: Line116. Based on protein sequence similarity, NleB belongs to the GT-8 family of enzymes， but one cannot predict the glycosyltransfer mechanism base on the protein sequence similarity since they have different modification residue, which means different niche around the modification site.
Q2: Line120. Utilize other method to verify the peptide modification site and percentage. Q3: Line129. Show data of the protein quality. Do the mutations or truncation have defects on the enzyme activity of NleB family proteins? Q4: If the authors use SseK1 to test the biochemistry activity and use SseK2 to do structural analyses, how to explain why sseK1 expressed in the cytoplasm whereas the SseK2 and SseK3 localized on Golgi apparatus? Q5: Line175. Since the pi-pi stacking mode is not necessary for other GTs and slightly different in NleB/SseK123, is it required for the biochemistry and functional activities of NleB family? which pistacking is more or equally crucial ? Q6: what is the corellation between the flexilibity of lid domain and the enzyme activity? Q7: what about binding affinity between peptides and the AxA mutant or other mutants? More experimental data are required to verify the molecular basis for peptide substrate recognition which the authors proposed. Q8: SseK2 has the relatively weakest enzyme activity in all the reported family members, why the authors choose SseK2 to do the molecular docking? Can it represent the physilogical situation? Only the open lid form of SseK2 can be used in docking, which means ternary complex and binary complex might have a lot of differences, especially in the enzyme-substrate binding aspect. If a short peptide cannot be docked in, we can speculate a full length death domain protein may induce much more conformational changes after ternary complex formation. So the docking model is not enough to clarify the enzymatic mechanism and enzyme-substrate coordination mechanism. The crystal structure of the ternary complexes are required. Q9: Since WR motif is important for the binding of SseKs and peptides form death domains or GAPDH, loss of function data (in binding and modification) were largely lacking. Q10: in Fig.6g, The E253A and HEN mutants showed different moleculat weight pattern on the SDS PAGE, whereas in other gel (such as in Fig.6def) they did not show as this pattern, the authors should clarify it. The data quality of the immunoblotting in Fig6 should be improved.
Q14. page 8, line 186: "In the SseK3 structure, SseK2-like π-π stacking interaction is conserved and Phe190SseK3 and Trp52SseK3 (corresponding to Phe203SseK2 and Trp65SseK2) participate in an interaction with the uridine group in the same orientation." A good opportunity to introduce a Figure 3 panel showing the arrangement of these aromatic residues into the SseK3 active site.

Response: We changed Figure 3 as suggested.
We added a panel of uridine binding mode of SseK3 in figure 3a and please check the upper picture. The red rectangle is the part we added. In addition, we described the related content in manuscript line 195-198 (Change-marked ver: line 220-221).
Q15. page 9, line 214: What is the experimental evidence that support the substrate binding/product release order for SseK1, SseK2, SseK3 or NleB2? I agree with the authors that in most of the cases, GTs follow an order mechanism with the donor substrate binding first. However, in the absence of experimental enzymatic/kinetic data, the authors should be cautious with the interpretation of the structural data. A statement should be added accordingly.
Response: We agree with the reviewer that we do not have any experimental evidence on whether these enzymes follow an ordered mechanism. However, we do not mention in the text anything related to an ordered mechanism. We only state that in the presence of the sugar nucleotide, we see large conformational changes of the C-terminal lid domain. This also occurs in similar virulence factors such as the Legionella pneumophila glucosyltransferase and the toxin A and B (PMIDs: 20030628 and 17901056), which share this C-terminal lid.
Q16. page 9, line 229: 'Based on these data, we suggest that the conformation of the C-terminal domain is highly similar between SseK2 and SseK3.' Please remove this sentence.
Response: We deleted that sentence.  Tables. 3-5). For each peptide, four different enzyme systems were prepared: apo-SseK1, apo-SseK2, holo-SseK1, and holo-SseK2, where apo and holo stand for the enzyme without and with Mn2+ and UDP, respectively. We observed that all three peptides bound to both SseK1 and SseK2, irrespectively of the forms used in the experiments (Supplementary Figs. 3, 4). These data imply that binding of the short peptide ligands occurs independently of enzymatic activity and can also take place in the absence of the sugar nucleotide". The text is more elaborated than before and it is clear that the peptides can bind independently of the presence of the nucleotide and cofactor. As you can see we have elaborated more what was stated before. We think that the sentence is more understandable now and can be unambiguously interpreted.
Q18. page 10, line 254: 'notion that that a' by 'notion that a' Response: We corrected this sentence. (page 11, line 275/ Change-marked ver: line 302) Q19. page 13, line 323: I would suggest the authors to introduce here a section focused into the 'Catalytic mechanism of SseK1, SseK2 and NleB2'. In that context, please move/introduce the entire 'Anomeric configuration of glycosylated peptides' section here, and renumber Figure 1 accordingly. After the discussing the catalytic mechanism, the authors could introduce the section 'Catalytic importance of HEN motif'.
Response: We thank the reviewer for this suggestion. However, we believe that the current order of our manuscript is easier to read and we prefer to discuss the HEN motif before the mechanism because we can then explain why we think an S N i mechanism is more plausible than an S N 2 mechanism. The results for HEN motif mutants rule out completely the possibility of the S N 2 mechanism and support the S N i mechanism, together with the molecular dynamics simulations.
The manuscript describes structural and functional studies of an important class of bacterial glycosyltransferases that are used by these bacteria to specifically suppress host defense. This topic is of interest to a broad readership. The crystallographic studies are sound and reveal many details of donorsubstrate enzyme interactions. Complementary biophysical methods, mainly NMR, have been used to gain more insight into the function of these interesting enzymes. The two most important conclusions of the studies are 1/ the retaining character of the enzymes can be experimentally proven, and 2/ the data support a frontface SNi type mechanism. MD simulations are finally used to present a comprehensive functional model. The data presented are new and deserve publication. I also think that Nature Communications would be the right journal to present this study. However, there are some shortcomings and questions, which need to be addressed prior to publication as this is detailed in the following.
General comments: I suggest that the authors phrase their statements on the proposed SNi mechanism more carefully, as no direct experimental evidence has been presented.

Response:
We have revised this in the updated version of the manuscript. Our data conflicts with an S N 2 mechanism because there are no residues that could act as nucleophiles in the vicinity of the GlcNAc moiety. The neighboring residues forming the HEN motif are not essential for catalysis, implying that these enzymes do not follow an S N 2 mechanism. In addition, we also discuss that a similar mechanism to the S N i was proposed earlier that also involves a front face mechanism. However, this mechanism has never been probed experimentally. Thus, we believe that these enzymes might follow the most accepted and typical S N i mechanism for the retaining glycosyltransferases. In any case, we have toned down our claims regarding the proposed S N i mechanism.
We think it is clear that we have toned down our claims with respect to the mechanism because we now suggest that they might follow the S N i mechanism. We think that this is now well explained in the revised version of our manuscript (especially line 441-470/ Change-marked ver: line 468-478).
In general, ITC experiments provide enthalpic and entropic signatures of binding reactions. Unfortunately, this analysis has not been done although the raw data seem to be available. Some of the ITC thermograms in Supplementary Figure 1 look certainly like they are suitable for analysis of binding enthalpies and entropies. This full analysis should be done where possible. On the other hand, some of the thermograms, e.g. the ones reflecting binding of UDP-Glc and UPD-Gal seem so suffer from artifacts. What is the reason for the discontinuities and could these be corrected? Experimental details for the ITC experiments are missing and should be included. Data analysis with Origin also provides Chi2 values for the dissociation constants, which should also be included. It would be interesting to compare the thermodynamic signatures of binding of activated sugars to these glycosyltransferases with existing literature data. If for some reason a full data analysis of the ITC data is impossible the authors should explain this. Figure 1' as suggested. We purified the enzyme again and performed ITC assays with UDP-Glc and UDP-Gal. These data support our earlier analysis and lack artifacts.

Response: We thank the reviewer for this suggestion. We added extra data that include the stoichiometry, enthalpy, entropy, Gibbs free energy, and binding affinity in 'Supplementary
Given that the importance of the experimental determination of the alpha-configuration of the D-GlcNAc-Arg linkage is so much stressed in the paper, why wasn't the complete set of chemical shifts of the glycopeptide obtained enzymatically reported? From the reading the paper I would assume this data set exists. Why not report it?
Response: Chemical shift assignments were made for the 9-mer peptides (FADD 110-118 , TRADD 229-237 , and GAPDH 195-203 ) used for STD NMR. We have now updated the manuscript to include these assignments in the supplementary material (Supplementary Tables 3-5). However, assignment of the glycopeptide was not possible since a very low amount of the glycosylated 16-mer of GAPDH was obtained from the enzymatic reaction for the NMR study, as the reaction did not proceed quantitatively, which meant that the final sample consisted of a complex mixture which additionally contained the non-glycosylated peptide, a large excess of free UDP-GlcNAc, and the hydrolysis products UDP and alpha-and beta-GlcNAc (as shown in Fig.1

-left in the main text), making extremely difficult to carry out a full chemical shift assignment of the glycopeptide. Instead, we were able to confirm the linkage between GlcNAc and arginine in the resulting glycopeptide through a 2D 1 H, 1 H-TOCSY experiment which showed a correlation between the GlcNAc anomeric proton and a η-proton of an arginine. Given that the glycosylation has been confirmed through other biophysical techniques within this paper (i.e.
kinetics experiments with GAPDH wt and mutants), we believe that this is sufficient evidence to show that the measured coupling indeed belongs to the GlcNAc that is covalently linked in alpha configuration to an arginine residue of the peptide ligand. We have now added to figure 1 an expansion of the TOCSY experiment showing the above mentioned correlation between the anomeric proton and the arginine proton.
Q1.On p. 11, first para, it is mentioned that STD NMR experiments indicate a "significant conformational rearrangement of the peptide ligand". I do not agree with this conclusion. A change in binding epitopes may have many causes. Only one of them would be a conformational change.
Response: We agree with the reviewer that, in general, a change in epitope is not sufficient to claim a conformational change, since the differences may be due to a different binding mode. However, in this case, we believe that a change in conformation can be claimed, since the WR-motif is preserved as the predominant contact, ruling out the possibility of a significant change in binding mode. This is further evidenced by the fact that there is no substantial change in absolute STD intensities.
Q2.On p. 11, second para, it is stated that "Hence, it is clear that differences in glycosylation of full length ...are not due to differences in binding modes ...". I simply cannot follow this argument, as glycosylated peptides or proteins have not been studied here.
Response: Based on the similarities of how the peptides interacted with the enzymes from STD experiments and the differences in substrate glycosylation, we suggest that the differences in activity for the SseK1 and SseK2 for the protein substrates are due to contacts with the substrates far away from the active binding site. Although we do not have any direct experimental evidence for this, our molecular dynamics simulations for SseK2 in complex with FADD death domain suggest that FADD residues interact with regions of the SseK2 far distant from the active site. In any case, we have toned down our argument in the revised version of our manuscript.
We now suggest that our claims are more likely to happen because as we explain we only have indirect evidences of it. These are our own STD-NMR experiments and molecular dynamics simulations. As well as with the S N i mechanism, we have toned down our claims because we now suggest that interactions far away from the active site between SseK1/2 and FADD/TRADD/GAPDH are likely to happen explaining the differences among these enzymes regarding the glycosylation of protein substrates.
Q3.On p. 15, Proposed mechanism: It is not quite correct that there is no experimental evidence for double displacement SN2. For human blood group B galactosyltransferase, a covalent intermediate has been described for the E303M mutant of this enzyme. There is a relatively recent review on this general topic by Ardevol et al. (2016). Maybe this is a good reference to get the non-expert reader involved.
Response: The reviewer is partially right. That study claimed they demonstrated a S N 2 mechanism for the human blood group B galactosyltransferase. However, more robust experimental work has clearly demonstrated that the S N 2 mechanism was not right. In fact, the results with E303M appear to be an artifact because the later study (see PMID:28960760 or DOI:10.1002/anie.201707922) clearly demonstrates that this enzyme follows the typical S N i mechanism. In fact, they suggest that most retaining glycosyltransferases must follow the S N i mechanism. The authors claim that E303 must play an important role in the stabilization of the transition state and product release.
Questions regarding Online Methods: Q4.p. 20, NMR spectroscopy: Which reference has been used for chemical shifts? For measurements in H2O: how much D2O has been added for the lock?

Response: the residual HDO signal was used as a reference. For the measurements made in H 2 O, the ratio was 90% H 2 O/10% D 2 O. The methods section has been modified accordingly (page 21-22, lines 576-582/ Change-marked ver: line 614-627)
Q5.p.20, line 533: The strength of the B1 field and the flip angle used for the Gaussian pulse in the cascade should be given. Q6.p.20, line 534: "spoil sequence" probably means spinlock filter. How long was the spinlock field?

Response
Response: The spoil sequence described here is a sequence of two trim pulses of 2.5 and 5 ms followed by a 40% z-gradient applied for 3 ms at the beginning of the experiment to destroy any residual x,y-magnetisation from the previous scan. A spinlock filter was also used to filter out protein signals and was applied for 40 ms. The methods section has been modified accordingly (lines 585-588/ Change-marked ver: line 614-627).
Q7.p. 21, line 539: I find it a little surprising that at 2mM MnCl2 good NMR spectra have been obtained. Is the concentration cited really OK? Compare the settings for the STD NMR experiments.
Response: We agree with the referee that this issue was a bit confusing in the way it was written in the methods section. This concentration of MnCl 2 is referring to the experimental conditions for the preparation of the glycosylated peptide, not for the NMR experiments. The glycopeptide was purified previous to the NMR study. Nevertheless, some remaining amount of Mn 2+ was evident as broadening of the NMR signals, but not to such an extent as to preclude observation of cross peaks in the 2D NMR spectra. We have modified the starting sentence at line 590 (Change-marked ver: line 630) to "Samples for peptide glycosylation were prepared…", as well as the caption of Figure 1.
Q8.p. 21, line 538-547: What was the digital resolution of the 2D spectra used to determine the one-bond C-H coupling of C1 o GlcNAc?
Response: The original experiment at 800 MHz had a resolution in the direct dimension of 20 Hz, which, although still qualitatively valid to distinguish the coupling, was not of enough resolution as to quantitatively report the coupling. We have now repeated the experiment at 500 MHz with a resolution of 1.6 Hz and measured the coupling to be 169 Hz. The manuscript has been updated to include the spectrum acquired at 500 MHz (Figure 1, and lines 595-

599/ Change-marked ver: line 633-638)
Minor points: In this paper, Park et al reported the type III secretion system effector proteins of NleB family are retaining glycosyltransferases following a SNi mechanism. Recently, the structure of SseK3 was determined, revealing a GT-A fold. However, the specific enzyme mechanism and the identification of the catalytic base remain unclear. There are also discrepancies regarding whether these enzymes are retaining or inverting GTs because this has not been experimentally probed. In addition, details regarding substrate specificity based on structural evidence are also limited due to the lack of ternary complexes. Park et al want to address these remaining major questions of this field. However, there are several flaws need to be properly addressed. Validation the linkage form of the glycoside on the substrate protein is the key step to analyzing the enzymatic mechanism. The authors did not get the real structure of enzyme-sugar ligand-substrate ternary complex. And the substrate should be the full death domain proteins but not the peptides, because there is no evidence to show the peptides can be used as substrates and be hundred percent modified as the death domains (such as TRADD DD, FADD DD). So the methodology and the data cannot fully support the conclusion of the enzymatic mechanism of this newly reported post translational modification (GlcNAcylation on Arginine). I would like to recommend the editors of Nature communication considering a resubmission when the authors can provide crystal or NMR structures of the NleB/SseK-UDP-GlcNAc-DDs ternary complex.
Other points: Q1: Line116. Based on protein sequence similarity, NleB belongs to the GT-8 family of enzymes, but one cannot predict the glycosyltransfer mechanism base on the protein sequence similarity since they have different modification residue, which means different niche around the modification site.

Response: We have deleted this statement.
Q2: Line120. Utilize other method to verify the peptide modification site and percentage.
Response: In addition to the determination of the configuration of the anomeric carbon of the transferred GlcNAc, by undecoupled HSQC experiments, we have now been able to confirm that the linkage takes place between GlcNAc and arginine in the resulting glycopeptide through a 2D 1 H, 1 H-TOCSY experiment which showed a correlation between the GlcNAc anomeric proton and a η-proton of an arginine. An expansion of this spectrum is now included in Figure 1.
Q3: Line129. Show data of the protein quality. Do the mutations or truncation have defects on the enzyme activity of NleB famly proteins? Response: As we described in manuscript, we truncated the flexible N-termini and mutated a Cys residue to make protein crystals in obtain the 3D structure. However, for other In-vivo and In-vitro assays, we have used the wild-type form of the proteins. In addition, the mutated Cys residue is located on the opposite side from the active site, implying that this Cys is a scaffold residue and is not involved in catalysis.
Q4: If the authors use SseK1 to test the biochemistry activity and use SseK2 to do structural analyses, how to explain why sseK1 expressed in the cytoplasm whereas the SseK2 and SseK3 localized on Golgi apparatus? Response: We did not make such a statement nor did we assay for the subcellular localization of these enzymes. Differences in subcellular localization reported by others may be due to protein-protein interactions other than those that are the subject of investigation here.
Q5: Line175. Since the pi-pi stacking mode is not necessary for other GTs and slightly different in NleB/SseK123, is it required for the biochemistry and functional activities of NleB family? which pi-stacking is more or equally crucial ? Response: We thank the reviewer for this suggestion. We have added ITC data and NF-kB assay result in Supplementary Figure 1 and have updated this in the revised manuscript. Based on our data, the pipi stacking mode is important for binding to UDP-GlcNAc and enzyme activity (Supplementary Figure  1b, c and d). In addition, combined with the ITC assay and the NF-kB activation assay, Trp65 SseK2 and Trp51 SseK1 play a more crucial role than Phe203 SseK2 and Trp331 SseK1 to UDP-GlcNAc binding and enzyme activity. . Regarding other residues, we also demonstrate by kinetic experiments the importance of the HEN motif in binding and catalysis towards the substrates. Finally, our STD-NMR experiments and the importance of the WR motif in the peptides are supported by molecular dynamics simulations. To fully understand the molecular basis of the protein substrate recognition, we would need the crystal structure of a ternary complex and this is something that we have not been able to obtain, despite 18 months of effort (please see below).
Q8: SseK2 has the relatively weakest enzyme activity in all the reported family members, why the authors choose SseK2 to do the molecular docking? Can it represent the physilogical situation? Only the open lid form of SseK2 can be used in docking, which means ternary complex and binary complex might have a lot of differences, especially in the enzyme-substrate binding aspect. If a short peptide cannot be docked in, we can speculate a full length death domain protein may induce much more conformational changes after ternary complex formation. So the docking model is not enough to clarify the enzymatic mechanism and enzymesubstrate coordination mechanism. The crystal structure of the ternary complexes are required.
Response: SseK1 preferentially glycosylates GAPDH and TRADD while SseK2 glycosylates FADD. The targeted region of GAPDH is a random coil, so it is difficult to model. Conversely, the targeted regions of TRADD and FADD are helical, making modeling feasible. This is why we chose SseK2 for molecular docking studies. FADD was chosen because a high-resolution crystal structure exists. Regarding the likelihood of conformational changes upon formation of the ternary complex, our 500 ns GaMD simulation of the SseK2/ FADD (full protein) complex, showed theoretically that at that timescale there are no significant conformational rearrangements.
We agree with the reviewer that the crystal structure of the ternary complexes would clarify the issue about conformational changes upon complex formation. However, we have been attempting for nearly 18 months to obtain this complex and have exhausted all known combinations of protein truncations, point mutants, and buffer conditions. We have attempted 1) co-crystallisation experiments of SseK1/SseK2 with UDP-MnCl2 and different peptides, 2) soaking experiments of peptides on crystals of SseK1-UDP-Mn +2 or SeeK2-UDP-Mn +2 , 3) cocrystallization experiments of SseK1/SseK2 with UDP and the FADD death domain, 4) two fusion proteins containing SseK1/2 coupled to the death domain of FADD, and 5) several SseK1/2 fusion proteins coupled to the peptide substrates.
Finally, we have attempted similar experiments for other GT-A fold glycosyltransferases. In particular, we worked with the Legionella pneumophila glucosyltransferase (binary complexes were published in PMID: 20030628) and the toxin A and B glycosyltansferases, and in both cases we were never able to obtain ternary complexes.
Q9: Since WR motif is important for the binding of SseKs and peptides form death domains or GAPDH, loss of function data (in binding and modification) were largely lacking.
Response: We thank the reviewer for this question. To reveal the importance of the GAPDH WR motif, we did additional experiment for measuring the enzyme kinetics by using GAPDH peptides including the wild type and WR mutant peptides. As a result, the single mutation form of each WR motif decreases the catalytic efficacy (40% and 47% to Trp and Arg, respectively) as compared to the wild type and the double mutant catalytic efficacy synergistically decreased (17%) (Supplementary Figure  16b). This data supports the STD-NMR experiment and combined with the kinetics assay and STD-NMR results, it can be concluded that the WR motif is important for the interaction with SseK. Additionally, we dissected the contribution of the WR-motif to binding affinity, by running STD NMR competition experiments between WT and single/double mutants of the TRADD peptide interacting with SseK1. The relevance of the WR-motif was demonstrated as any modification on the WR-motif impacted negatively the affinity of the peptide for the enzyme (new text included; lines 287-310 (Change-marked ver: line 313-336); Supplementary Figures 19 and 20) Q10: in Fig.6g, The E253A and HEN mutants showed different molecular weight pattern on the SDS PAGE, whereas in other gel (such as in Fig.6def) they did not show as this pattern, the authors should clarify it. The data quality of the immunoblotting in Fig6 should be improved.
Response: While not the subject of this work, we and others (e.g. Esposito, JBC, 2018), have observed that some, but not all of the NleB/SseK orthologs are self-glycosylated. We believe that selfglycosylation may contribute to the differential migration of some proteins in SDS-PAGE. We have carefully checked our immunoblotting data for rigor and reproducibility and believe that the images provided are of high quality and support the overall conclusions reached in the manuscript. o In FADD peptide Val116 was wrongly labelled: it has been corrected to Ala116 o In TRADD peptide Ala233 was wrongly labelled: it has been corrected to Val233 o The STD NMR experiments for TRADD have been repeated with the correct sequence, ending in Leucine (KWRKVGRSL). Previously, the sequence ending with Isoleucine (KWRKVGRSI) was wrong.