An integrated genomic regulatory network of virulence-related transcriptional factors in Pseudomonas aeruginosa

The virulence of Pseudomonas aeruginosa, a Gram-negative opportunistic pathogen, is regulated by many transcriptional factors (TFs) that control the expression of quorum sensing and protein secretion systems. Here, we report a genome-wide, network-based approach to dissect the crosstalk between 20 key virulence-related TFs. Using chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-seq), as well as RNA-seq, we identify 1200 TF-bound genes and 4775 differentially expressed genes. We experimentally validate 347 of these genes as functional target genes, and describe the regulatory relationships of the 20 TFs with their targets in a network that we call ‘Pseudomonas aeruginosa genomic regulatory network’ (PAGnet). Analysis of the network led to the identification of novel functions for two TFs (ExsA and GacA) in quorum sensing and nitrogen metabolism. Furthermore, we present an online platform and R package based on PAGnet to facilitate updating and user-customised analyses.

1-Not all known virulence is captured: I realize that these authors, like all of us, are limited by time and/or resources, and cannot do everything. However, it was strange that regulators known to dramatically alter virulence were not included… I'm specifically thinking about PhoB, GbdR, AlgR, FecI, PchR, SphR, FleQ, among many others. While the authors grabbed most of the major regulators, they missed too many to justify calling this a real integrated network of Pa virulence. 2-RpoN is not a transcription factor, it is a sigma factor. Since a sigma factor is a required component of RNA polymerase, it is not surprising that this particular sigma factor regulates the largest number of genes of all of the "TFs". This is the only sigma factor on the authors virulence list. They either need to remove it from the analysis and focus just on strict TFs, or, if the authors are going to include sigma factors into their analysis, they need to add FoxR, AlgU, FliA, and RpoH minimally, since all are important for virulence. This should be an easy fix, since the data used in this current manuscript to generate the network have already been generated by Schulz et al. 2015 (PLoS Pathogens), including ChIP-Seq, RNA-Seq, and binding site analysis for most of the non-essential sigma factors in Pa. 3-The authors provide no statistical analysis of their interaction networks. Since these regulators are known to have different scales of binding sites in the genome and there are limited upstream regions on which to bind, by simple network statistics, we'd predict overlapping binding at some frequency of promoters. The co-bound and co-regulated extinction curves (captured readily in the data from Supp Tables 2-4) seem to match a standard Gaussian extinction curve. Please provide statistical support that these overlaps are not simply by chance. 4-Transcriptomes vs Reality -One of the issues with the transcriptomes as analyzed in this manuscript is that most of the transcriptomes are derived from different experimental conditions, timepoints, and even strains of Pa. This really makes the analysis of co-regulated genes meaningless, since they might link more to shared responses of the mutants to experimental variables than direct regulation by those regulators. 5-Oddly, the authors have listed LipA, and LipC, known Type II secreted lipases, in their T6SS list… what is the rationale for this inclusion? 6-The methods section is not complete. The authors state they used anti-vsvG to do their ChIP, but do not describe the generation of the vsvG-tagged strains (or it is not detailed appropriately), and do not demonstrate that the vsvG-tagged versions complement the deletion strains. Without genetic complementation by the vsvG-tagged TFs, the ChIP-Seq is meaningless. 7-The grammar and syntax are nearly unreadable, mostly for the Introduction. Things get a little better in the Results, but this manuscript needs very heavy editing by a scientifically literate editor with English fluency.
Minor issues: -The tap(sheet) for Supplemental Table 2 contains the right data but is labeled within the sheet as Table S7 -Many of the EMSA figs, while convincing, are of poor resolution and should be improved.
-RpoN association with promoters is known to be altered when it is in the complex as a portion of RNA polymerase, this potentially invalidates the EMSAs conducted with purified RpoN. -The PaVIRnet site does not appear to be much more useful than the supplemental figures and tables are, at least for the data presented. The website is capable of receiving user input data but I have two problems -since this is not sourced on an academic server, is there a source for sustained support for this site? Second, what assurances do users have that any network data they upload will not be used by the authors or others? - Fig 1a should have a log2 or log10 scaled y-axis to make distinction amongst the TFs apparent.
Reviewer #2 (Remarks to the Author): In this manuscript, the authors present the results of a very comprehensive analysis of Pseudomonas aeruginosa regulatory genes, including 19 regulators analyzed by ChIP-seq and transcriptomics methods. The ChIP-data are corroborated by EMSA analysis of selected targets showing the the results are largely valid. Alltogether, the data represent a strong contribution for the attempts to elucidate the highly complex regulatory networks of P. aeruginosa, a model organism that harbors around 500 regulatory genes (which is ~ 10% of the genomic contenct), which should also be of interest to a broader community. However, despite the high value of this work, many key aspects of the presentation should be reconsidered and I will concentrate on these in this review. 1) I find both the term "PAVIRnet" and the way it is introduced highly misleading. Firstly, it is not clear, why the network should be considered as a network regulating virulence. It is never clearly stated, why the authors selected the 19 regulators that were included here and the selection seems rather arbitrary. There is nothing wrong with that but this large number of regulators covers a major part of the genome which of course regulates virulence but also all kinds of other pathways. RpoN for example is mainly described as a regulator for nitrogen metabolism, which has many connections to virulence genes but not exclusively. I'd argue that it is even hard to find any regulatory gene affecting multiple operons that has not any connection to virulence at all. Therefore, I think that the term "virulence" is close to meaningless in the title and might as well be replaced by "genomic". Secondly, speaking of "the PAVIRnet" implies that it represents a functional unit (which the authors claim to have "constructed"), while in fact it is a rather arbitrary subset of the complete genomic regulatory network of P. aeruginosa. All this claims much more than the data can actually delivered and it should be tuned down accordingly. 2) Complex regulatory networks of P. aeruginosa have been published before, covering different aspects of the story and using different types of analyses (e.g., Schulz et al, 2015Schulz et al, , doi: 10.1371Balasubramanian et al, 2013Balasubramanian et al, , doi: 10.1093. What are the new aspects covered by this work that have been missing before? It is not clearly stated and the discussion should better put this work in the context of previous knowledge. 3) The writing appears somewhat sloppy in parts of the manuscript. The first sentences of the introduction are difficult to read due to mixed up wording or grammar. The nomenclature in very inconsistent, e.g., the use of the term "transcription factor" (TF). Not all of the mentioned regulators are TFs, some regulate on post-transcriptional levels as the authors state themselves (e.g. line 104) and the list of regulators itself seems to vary: line 95 mentions WspR, VqsR and BfmR that are not mentioned at all in the next paragraph. Instead, in line 109 ExsA appears, which was not mentioned before. The text should be carefully proofread and checked for such inconsistencies and also to remove the many smaller typos and language errors. 4) There are also inconsistencies between the descriptions of results. i) The network was built on 19 "TFs" (line 177) by assigning function targets based on two criteria, one of which was direct binding of the TF to the upstream region. This is in contradiction to the statement that at least three of the regulators act post-transcriptionally. Or another example ii) Table S4 and the text (line 185) claim that rhlR is co-regulated by multiple genes, however the ChIP-results only show binding of the rhlR promoter by RhlR itself. It appears as if binding was not really used as a hard criterion but the text does not reflect this. 5) There is no definition of "promoter region" or "upstream of genes" and "downstream of genes", which all is absolutely necessary to understand, how the ChIP-results were associated with gene loci.
6) The section "Experimental verification of PAVIRnet" contains misleading statements. Firstly, it is not an experimental verification of the network as such but of a subset of the results from ChIPseq and transcriptome analysis. Secondly, no "predicted regulatory relationships" are presented (line 230). The network is deduced from the data and does not predict any additional relationships. 7) The discussion is mainly repeating many results instead of putting the data into context. 8) The supplementary tables are not sufficiently explained, which limits their use

RNA-seq, 5 ChIP-seq, biochemical verifications and statistical analyses) and
carefully made substantial revisions based on these comments. We updated all figures and tables. We have also significantly revised the main text per the reviewers' suggestions, which are highlighted in red in the revised manuscript. The

following are our point-by-point responses to the reviewers' comments.
Reviewer #1 (Remarks to the Author): In this manuscript, Huang et al. present new ChIP-Seq and RNA-Seq results that they combine with previous data from their group and others in the field to generate what they term is the virulence network of P. aeruginosa (Pa). I think that this data will be useful to the field and this manuscript brings together and extends much useful data on virulence in Pa. There are a number of caveats, however, that diminish my enthusiasm for the manuscript:

Response: Thank you for your positive and critical comments. We have extensively revised the following part based on your constructive comments.
1-Not all known virulence is captured: I realize that these authors, like all of us, are limited by time and/or resources, and cannot do everything. However, it was strange that regulators known to dramatically alter virulence were not included… I'm specifically thinking about PhoB, GbdR, FecI, PchR, SphR, FleQ, among many others.
While the authors grabbed most of the major regulators, they missed too many to justify calling this a real integrated network of Pa virulence.

Based on the empirical background distribution, we calculated the number of co-occurrence (or co-binding) events by chance (termed N) on the promoter of interest.
4. P-values were derived from N/10 5 .

The raw p-values were finally adjusted for multiple hypothesis testing by the
Benjamini-Hochberg procedure. 2014), we did not show its RNA-seq data in LB, OD 600 = 0.6. Figure 3, Figure S3 and

Response: Response: We apologise for the errors. We have removed them in the revised manuscript.
6-The methods section is not complete. The authors state they used anti-vsvG to do their ChIP, but do not describe the generation of the vsvG-tagged strains (or it is not detailed appropriately), and do not demonstrate that the vsvG-tagged versions complement the deletion strains. Without genetic complementation by the vsvGtagged TFs, the ChIP-Seq is meaningless. Figure S1. Please see Lines 131-134.

Nucleic Acids Res. 2017; Zaborina et al. PLoS Pathog. 2007). A Congo red assay also showed that FLAG-tagged FleQ indeed complemented the colony morphology of fleQ strain (Hickman et al. Mol Microbiol. 2008). The results are displayed in
7-The grammar and syntax are nearly unreadable, mostly for the Introduction. Things get a little better in the Results, but this manuscript needs very heavy editing by a scientifically literate editor with English fluency.

Introduction section has been largely re-written. Please see Lines 44-112.
Minor issues: -The tap(sheet) for Supplemental Table S2 contains the right data but is labeled within the sheet as Table S7. -RpoN association with promoters is known to be altered when it is in the complex as a portion of RNA polymerase, this potentially invalidates the EMSAs conducted with purified RpoN.

Response: We agree with your comment. In the revised manuscript, we have removed all results on RpoN in accordance with the reviewer #1's major comment
#2.
-The PaVIRnet site does not appear to be much more useful than the supplemental figures and tables are, at least for the data presented. The website is capable of receiving user input data but I have two problems -since this is not sourced on an academic server, is there a source for sustained support for this site? Second, what assurances do users have that any network data they upload will not be used by the authors or others?

To provide more flexibility, we also made PAGnet available as an R package 'PAGnet' in Github that can be freely downloaded, locally installed and used by the user on his/her personal computer. Thus, the user can choose to use the R package locally should they have further concerns about the confidentiality of their data. All functions available on our website are also provided in the 'PAGnet' R package by running a local shiny GUI within R. In addition, the master regulator analysis can be performed using the function 'pagnet.mra' in the R console without running the shiny GUI. Detailed instructions about installation and step-by-step use of the package are provided in the vignette of the 'PAGnet' package. Please see Lines 347-
355.
- Fig 1a should have a log2 or log10 scaled y-axis to make distinction amongst the TFs apparent.

Response: We agree with the comment. Accordingly, we have revised the y-axis
showing log2 in Figure 1a. Please see the Figure 1a and Line 636.

Reviewer #2 (Remarks to the Author):
In this manuscript, the authors present the results of a very comprehensive analysis of Pseudomonas aeruginosa regulatory genes, including 19 regulators analyzed by ChIPseq and transcriptomics methods. The ChIP-data are corroborated by EMSA analysis of selected targets showing the results are largely valid. Altogether, the data represent a strong contribution for the attempts to elucidate the highly complex regulatory networks of P. aeruginosa, a model organism that harbors around 500 regulatory genes (which is ~ 10% of the genomic content), which should also be of interest to a broader community.

Response: Thank you for your positive comments!
However, despite the high value of this work, many key aspects of the presentation should be reconsidered and I will concentrate on these in this review.
1) I find both the term "PAVIRnet" and the way it is introduced highly misleading.
Firstly, it is not clear, why the network should be considered as a network regulating virulence. It is never clearly stated, why the authors selected the 19 regulators that were included here and the selection seems rather arbitrary. There is nothing wrong with that but this large number of regulators covers a major part of the genome which of course regulates virulence but also all kinds of other pathways. RpoN for example is mainly described as a regulator for nitrogen metabolism, which has many connections to virulence genes but not exclusively. I'd argue that it is even hard to find any regulatory gene affecting multiple operons that has not any connection to virulence at all. Therefore, I think that the term "virulence" is close to meaningless in the title and might as well be replaced by "genomic".
Secondly, speaking of "the PAVIRnet" implies that it represents a functional unit (which the authors claim to have "constructed"), while in fact it is a rather arbitrary subset of the complete genomic regulatory network of P. aeruginosa. All this claims much more than the data can actually delivered and it should be tuned down accordingly. 3) The writing appears somewhat sloppy in parts of the manuscript. The first sentences of the introduction are difficult to read due to mixed up wording or grammar. The nomenclature in very inconsistent, e.g., the use of the term "transcription factor" (TF). Not all of the mentioned regulators are TFs, some regulate on post-transcriptional levels as the authors state themselves (e.g. line 104) and the list of regulators itself seems to vary: line 95 mentions WspR, VqsR and BfmR that

Response: We appreciate the critical and constructive comments and agree
are not mentioned at all in the next paragraph. Instead, in line 109 ExsA appears, which was not mentioned before. The text should be carefully proofread and checked for such inconsistencies and also to remove the many smaller typos and language errors. 5) There is no definition of "promoter region" or "upstream of genes" and "downstream of genes", which all is absolutely necessary to understand, how the ChIP-results were associated with gene loci.

Response: Thank you for the suggestion to clarify the definitions of genomic loci.
We have provided clarifications in the revised manuscript. To illustrate these definitions, please see the following schematic diagram and Lines 140-144.

2014; Kong et al. Nucleic Acids Res. 2015; Zhao et al. PLoS Biol. 2016).
6) The section "Experimental verification of PAVIRnet" contains misleading statements. Firstly, it is not an experimental verification of the network as such but of a subset of the results from ChIP-seq and transcriptome analysis. Secondly, no "predicted regulatory relationships" are presented (line 230). The network is deduced from the data and does not predict any additional relationships.

Response: We agree with your comment and apologise for the misleading statements. In the revised manuscript, we have revised the expression 'Experimental verification of PAGnet' to 'Experimental verification of the functional targets of TFs'. Please see Line 264. We have also revised the expression 'predicted regulatory relationships' to 'validation' and have removed the word 'predicted' throughout the manuscript. Please see Line 277.
7) The discussion is mainly repeating many results instead of putting the data into context.