Metabolic specialization drives reduced pathogenicity in Pseudomonas aeruginosa isolates from cystic fibrosis patients

Metabolism provides the foundation for all cellular functions. During persistent infections, in adapted pathogenic bacteria metabolism functions radically differently compared with more naïve strains. Whether this is simply a necessary accommodation to the persistence phenotype or if metabolism plays a direct role in achieving persistence in the host is still unclear. Here, we characterize a convergent shift in metabolic function(s) linked with the persistence phenotype during Pseudomonas aeruginosa colonization in the airways of people with cystic fibrosis. We show that clinically relevant mutations in the key metabolic enzyme, pyruvate dehydrogenase, lead to a host-specialized metabolism together with a lower virulence and immune response recruitment. These changes in infection phenotype are mediated by impaired type III secretion system activity and by secretion of the antioxidant metabolite, pyruvate, respectively. Our results show how metabolic adaptations directly impinge on persistence and pathogenicity in this organism.


Comments:
1. Overall, it is unclear the sta9s9cal methods used, if any, for a number of the observa9ons.For example, for the metabotypes, which sta9s9cal tests allowed the authors to define these?For figure 3d, e, these are great visualiza9ons of complex data.Were sta9s9cs performed on the highlighted phenotypes with differently regulated metabolites?If so, which are significant?Showing paNerns in pathways is interes9ng, but sta9s9cs are needed to validate observa9ons.This goes for most figures, please describe the sta9s9cal tests used for each figure, as they are likely quite different, given the many approaches used in the study.Without this informa9on, it is very hard to rigorously review the study.
We acknowledge the comment from the reviewer, and we recognize that the sta9s9cal methods in our manuscript are not always easy to find and read.In material and methods under "Data analysis" (see line 670), we present the sta9s9cal methods altogether.To avoid confusion and make the specific analyses easier to find, we have included specific informa9on on the sta9s9cal methods in each result sec9on and figure legend.
Naïve and adapted metabotypes were designated by using PCA and itera9ve K-mean clustering first, followed by cura9on according to HCA.We have expanded the Data analysis sec9on in the material and methods to include specific informa9on on the metabotypes designa9on (see lines 686-690).
For figure 3d, to confirm that the difference in the number of differen9ally expressed proteins between clusters is significant, we performed two-sided paired t-test which showed a P value of 0.0047.We have included this informa9on in the results, figure and figure legend (see lines [212][213][214].
As also suggested by reviewer 3, we moved figure 3e and its corresponding text to supplementary text 1 and supplementary figure S2.Moreover, we have specifically indicated that convergent proteomic changes at the pathway level require further analyses and confirma9ons.
We hope this clarifies the confusion and allows for easier recogni9on of the sta9s9cal method used.
2. Some terminology is confusing.The authors use the term "naïve" metabotypes to describe cluster A, yet this cluster has a combina9on of early and late isolates.The term "adapted" also is used for other metabotypes.These terms when applied without sta9s9cal tests shape the conclusions and need to be rigorous tested to be used.Especially as naïve vs adapted clusters are driving later experiments and greatly impac9ng the conclusions drawn from the study.
We agree with the reviewer that the use of the terms 'naïve' and 'adapted' both for describing metabotypes and proteomic changes creates confusion.We have, therefore, rephrased the results (see lines 203-210) and verified the used of this terminology in the en9re manuscript.Moreover, for the HCA presented in figure 3 b and c, and all the other HCA in our manuscript, we used the sta9s9cal test of bootstrapping (10,000 replicates) with % of bootstrap values presented on the branches.Separa9on between cluster A and B is consistently shown in 99% of the replicates making us confident of the conclusions.

3.
In figure 2b, what is driving the two clusters of late isolates?Can you deduce this from the variables driving PC2 in fig 2A ?The sta9s9cal analysis performed on the hierarchical clustering (bootstrap analysis for 10,000 replicates) indicates that this specific clustering reappears in only 18% of the replicates.This separa9on is therefore highly uncertain, meaning those strains could also be clustered together with the rest of the early strains as is the case in the PCA in Fig. 2a.For this reason, we are not sure of the specific drivers of the separa9on of DK17_L and DK41_L.4. In figure 3, are there any pa9ent characteris9cs unique to cluster B? An9bio9c regimen differences?Mucoidy of the isolates?Lung func9on?Unfortunately, we do not have data on treatment or lung func9on.Therefore, we are unable to correlate pa9ent data with specific clusters.None of the isolates are mucoid.4b, what does the fold change look like for the early vs late isolates?This comparison seems very interes9ng to the study and would be nice to see, in addi9on to the PAO1 comparison shown.

In figure
This comparison is included in fig 4c for a set of virulence categories where sta9s9cally significant differences between early vs late and between clone types are present.We have specified this comparison in the results (see lines 244-248).Moreover, in fig 4a we show enrichment analysis for the comparison Late vs Early for the KEGG and GO terms.Finally, in Supplementary Data 2 we present all the fold changes for the Late vs Early comparison.
6. Lines 257-258, this statement seems like a testable hypothesis.This comment should be either be tested experimentally or moved to the discussion, if there is no data.
We agree with the reviewer that we do not have experimental data on the strains' immunogenicity.Therefore, we modified the sentence to underline the changes in virulence factors expression profiles rather than specula9ng on the strains' immunogenicity.
7. Also lines 356-358, this commentary is more appropriate for the discussion than in the results, as it doesn't set up data to be shown.
We agree with the reviewer's comment and therefore we have moved the commentary to the discussion.See lines 438-441.
8. In general, the authors are using the term immunogenicity, when they are just measuring IL-8 secre9on.Inflamma9on is more descrip9ve of the results presented, since the authors are just focusing on IL-8.Changed to inflamma9on.9.The final paragraph in the results is interes9ng, but needs more experimental data to support the conclusions the authors are making.Does an aceE/F dele9on mutant have altered T3SS ac9vity?Changes in exotoxin gene expression?Muta9ons in both the PDHc pathway and T3SS in clinical isolates can be true and not related.Experimenta9on is needed to link these two in the strains discussed, if the authors want to present the conclusions in lines 373-376.Or move this commentary to the discussion sec9on.
We agree with the reviewer's comment and therefore we have rephrased the conclusions of the paragraph and moved the commentary to the discussion.See lines 366-373 and 408-413. 10.What is the pyruvate concentra9on in CF sputum?Addi9on of this informa9on would support the authors results for a role for pyruvate in establishing persistent Pa infec9ons in CF.
Pyruvate has been iden9fied in sputum samples during exacerba9on events at an average concentra9on of 0.3 mM.We have included this informa9on in the text.See lines 425-426. 11.For the comment in lines 450-452, were an9bio9c resistance genes dysregulated in the aceE or aceF proteomics studies?If the metabolic phenotype precedes the development of an9microbial resistance, you might expect an9bio9c resistance gene expression to be altered in the aceE/F mutant backgrounds.
We did not find resistance genes with dysregulated expression.Moreover, the aceE and aceF strains do not show altered MICs rela9ve to PAO1 wild type.We only found a slight increase in Tobramycin MIC which most likely depends on the reduced growth rate of the strains (McKenney et al., 1997(McKenney et al., doi: 10.1093/jac/40.3.415;/jac/40.3.415;La Rosa et al., 2021doi:10.1038/s41467-021-23451-y). Since in our collec9on of clinical strains (Bartell et al., 2019 doi:10.1038/s41467-019-08504-7)an9bio9c resistance occurs only arer years of treatment, we argue that adapta9on to the host environment and specifically metabolic specializa9on might be more important at the beginning of the infec9on to counterbalance the host and its immune system.

Reviewer #2:
The manuscript by Pedersen et al tackles the role that metabolic adapta9on plays in the cys9c fibrosis context.The manuscript first uses a metabolomics approach to assess differences between early and late isolates from 8 pa9ents.The metabolomic data suggests that late isolates oren have similar metabolic profiles compared to their early counterparts.Proteomic data also supports similarity in late strains compared to early.Both metabolomic and proteomic analyses point to the secre9on of pyruvate as a key factor associated with late strains.Analysis of the proteomic data suggests that late strains have predicted changes in virulence protein expression that agree with previous work.They also find muta9ons in genes that code for the pyruvate dehydrogenase complex (aceE and aceF).They then generate mutants in the PA01 background that harbor these muta9ons, which they then use for whole cell proteomics.The proteomic data suggest far reaching consequences for gene expression in these mutants, and they were defec9ve for virulence as measured in an ALI model.This is an interes9ng study that tries to 9e together metabolic adapta9on with other known adapta9ons, and is tantalizing in sugges9ng that metabolic changes likely precede other adapta9ons.The finding need to be supported with further clarifica9ons, a clear exposi9on of limita9ons, and gene9c complementa9on of the mutants.Overall the findings of this study, especially the metabolomics and proteomics could be presented much more clearly, and these sec9ons could be shortened.
We thank the reviewer for the comment.We have addressed his/her concerns and clarified where requested (see below).We have expanded the discussion sec9on to address further limita9ons of our study, and shortened the metabolomics and proteomics sec9ons to introduce a supplementary text with a more extended and detailed presenta9on of the specific differences between the strains.Moreover, as requested, we have performed complementa9on assays of one of the mutant strains.
Major concerns: 1.It is clear that heterogeneous popula9ons of PA can coexist for long periods of 9me in pwCF.Therefore, the selec9on of single isolates from early and late is a limita9on unless the inves9gators can say defini9vely that the late strain was not present at the early 9mepoint.Likewise, the assump9on that early isolates are not adapted to the CF airway should also be treated as a limita9on.
We appreciate the reviewer's comment and sugges9on on the heterogeneous popula9ons present in pwCF.We previously compared the single isolates collec9on from certain pa9ents to their corresponding metagenomic profile (Sommer et al., 2016(Sommer et al., doi:10.1186(Sommer et al., /s12864-016-2873-1)-1).Our data showed that single isolates collected from sputum samples are usually the most abundant representa9ve of the popula9on.This confirms that single isolate collec9ons can be used to characterize the adap9ve and evolu9onary process of P. aeruginosa in pwCF.Indeed, while we do agree that pwCF are colonized by a heterogeneous popula9on, we believe that single isolates wellrepresent the most abundant sub-popula9on of the person.Moreover, the goal of our work is not to describe the metabolic changes occurring in the en9re popula9on, but rather to iden9fy molecular mechanisms of metabolic adapta9on which change the host-pathogen interac9ons.For this reason, we priori9zed to include different pa9ents and different clone types instead of several isolates from the same pa9ent.We have clarified this point in the result sec9on (lines 150-154).
2. The selec9on of early and late strains is a bit unclear.In the results, it seems as if the late strains were specifically chosen based on decreased growth rates.This might bias the data and confuse "late" for "growth retarda9on".
As indicated in the first paragraph of the result sec9on, growth rate reduc9on is a key phenotype associated with adap9ve evolu9on in pwCF.Therefore, to iden9fy new mechanisms of metabolic specializa9on occurring during within host evolu9on and to increase the likelihood of selec9ng interes9ng isolates, we chose late isolates which also had a reduced growth rate.This allowed us to put together a collec9on of early and late adapted strains where to analyse metabolic specializa9on.S9ll independently of the specific selected strains, we iden9fied mechanisms of metabolic specializa9on occurring in late slow growing isolates which are used in different pa9ents and infec9on scenarios to modulate the host response.We have rephrased most of the first result paragraph to clarify the strain selec9on approach (lines 126-154).
3. The authors do not address the phylogene9c relatedness of the strain and only refer to previous work in ref 36.However, it would be helpful to present that data here in order to understand how closely each of the strains is related to each other to understand whether the observed changes in metabolism are generalizable across PA.It would also help with interpreta9on of HCA and the clustering of metabolomic and proteomic profiles.
We acknowledge the request from the reviewer on the phylogene9c rela9onship between the strains.However, since the genome sequences of the clinical isolates have already been published and their rela9onship defined elsewhere (Marvig et al., 2014 doi:10.1038/ng.3148),we have not included this informa9on here.Moreover, since each pair of early and late strains belongs to the same clone types and we examined pairs from several different pwCF, when performing phylogene9c analysis on the strain's muta9ons, they will always cluster based on their muta9onal signature which by defini9on of clone types is more than 10,000 (Marvig et al., 2014(Marvig et al., doi:10.1038/ng.3148)/ng.3148).This can be evidenced by a recent publica9on from our group where we performed phylogene9c analysis on isolates from different pwCF and different clone types (Espaillat et al., 2024 doi:10.1093/molbev/msae022,Fig. 1B).
4. The aceF and aceE mutants should be complemented especially since many of the tests are on phenotypes not directly related to the ac9on of these two enzymes.
We acknowledge the comment from the reviewer, and we have performed complementa9on assays for the aceE mutant strain.Moreover, previous studies in P. aeruginosa have indeed demonstrated that complementa9on of muta9ons in the pyruvate dehydrogenase pathway can be achieved by supplemen9ng acetate in the culture medium (Jeyaseelan and Guest, 1980doi: 10.1099/00221287-120-2-385, Rae et al., 1997doi: 10.1128/jb.179.11.3561-3571.1997and Decheux et al., 2002doi: 10.1128/IAI.70.7.3973-3977.2002).For this reason, we have metabolically complemented the mutant strains and performed growth analysis and proteomics in presence and absence of acetate (lines 287-293).Lines 134-135 It would also be important to know how many SNP differences were found between each pair of genomes.
We have included this informa9on in Table S1.Line 272: The muta9ons are listed but not the haplotypes.Did each muta9on occur in it's own allele or were some found together.Also, define "silent" We did not find haplotypes for the aceE and aceF genes.Each muta9on occurred independently in each clone type and none of them were found together.We have changed 'silent' to 'synonymous' to avoid confusion.
Lines 409-411.The conten9on that aceE and aceF mutants occur prior to T3SS muta9ons is not well supported by the data.This could be addressed using a phylogene9c approach and other exis9ng whole genome sequences.
We agree with the reviewer that our data are not conclusive.Therefore, as also suggested by reviewer 1, we have modified this statement and indicated the need for further inves9ga9on.See lines 407-411.Data are jiNered to avoid overlap between symbols and therefore exclusively for visualiza9on purposes.

Reviewer #3:
The evolu9on of an9bio9c resistance dominates the research landscape involving microbial pathogens, but the mechanisms of microbial persistence are similarly important.In this study, the authors examine mechanisms of persistence, with a focus on metabolic rewiring/specialisa9on.They demonstrate that metabolic rewiring widely occurs during adapta9on to the host, and showcase an example that re-wiring metabolism related to pyruvate can enhance the persister phenotype in P. aeruginosa.To achieve this, the authors use a myriad of advanced techniques including metabolomics, proteomics, and confocal microscopy.
Due to the small sample size used in the study (n = 8 for most groupings e.g.'early vs late' and 'early/late vs PA01') many of the metabolomics and proteomics highlight correla9ons and trends, but are restricted to broad associa9ons.However the results rela9ng to Figs 5 and 6 -where the authors compare genomic sequence data for 34 pa9ents, make targeted genomic altera9ons, and assess phenotypes in an environmental context that emulates the human lung -are focussed and highly interes9ng.In these results the authors show how muta9ons affec9ng pyruvate metabolism can directly change phenotypes associated with persistence, showcasing that metabolic rewiring muta9ons may not only occur to alleviate trade-offs from other muta9ons.
Overall I feel that the manuscript is 9mely and presents some exci9ng results.I have only a few comments and ques9ons regarding the manuscript.I do not believe any of these comments are essen9al to complete prior to publica9on, but addressing these may help improve the clarity and readability: We thank the reviewer for his posi9ve and construc9ve comments on our manuscript.
1. Lines 90-93: "previous metabolic characterisa9ons of clinical strains were carried out on only a limited number of isolates and/or on bacterial cultures on one specific growth phase".This statement requires cita9ons.

Added.
2. Lines 153-155: "These metabotypes separate the late isolate from each respec9ve early isolate and from the rest of the non-adapted metabotypes".I think using the word 'non-adapted' here is misleading.We do not know if they have not adapted, only that they haven't adapted with regards to the screened phenotypes.Please provide a statement clarifying this next to the 'non-adapted' descrip9on.
We agree with the reviewer, so we have modified the sentence to specifically point out that 'nonadapted' refers to metabolic adapta9on.See lines 164-166.
3. Lines 178-179: "To test hypothesis that the observed metabolic specialisa9on is par9ally rooted in changes in expression of proteins involved in cellular metabolism".Why would the hypothesis be that metabolism changes are only par9ally rooted in changes to protein expression?This statement seems to be wriNen so that the hypothesis directly matches the results observed, rather than an organic hypothesis made at the beginning of the experiment.I would suggest either removing the word 'par9ally' or adding an explana9on as to why we would expect proteins rela9ng to metabolism to only partly govern metabolic changes.
6. Lines 295-299: The authors describe here the lack of associa9on between pathogenic phenotypes and the ace muta9ons, but they later describe how ace mutants exhibit different persister phenotypes.I was expec9ng an explana9on for these discrepancies to feature more prominently in the discussion.
The lack of associa9on between pathogenic phenotypes and aceE and aceF muta9ons refers to the laboratory condi9ons where most of the phenotypes are usually tested.We have included an explana9on of such lack of associa9on in the results (line 295) and discussion (lines 396-398 and lines 470-472).We agree with the reviewer that fig 1b contextualises 1a.However, we want to first focus on the general characteris9cs of the strain collec9on used in our study and then underline the specific difference between the growth rate of early vs late strains.

Fig 2C:
Why are there missing datapoints between some early and late condi9ons?Some data points and connec9ng curves are missing because, either the specific metabolites were not assimilated, or the OD50 was not possible to calculate since the assimila9on window was longer than the 9me we tested.We have included an explana9on in the figure legend.See lines 780-782.We realized that the confidence intervals for acetate got missing in the figure.We have now included them.
10. Fig 3E: I would advocate moving this panel to the supplementary informa9on, as I did not find this associa9on par9cularly convincing.740/2061 proteins exhibited differen9al expression (line 185), of which 235 belonged to those related to metabolism (line 196).When so many proteins exhibit differen9al expression, we should not be surprised that there are a handful of differences affec9ng this pathway between clinical strains and a laboratory strain.Similarly, we should also not be surprised to see differences between these strains and those that have spent several more years adap9ng in a host.How do we know that these changes are all adap9ve, could some not be compensatory muta9ons or products of gene9c drir?It may be that the pathways have been altered to "op9mise" metabolism in the lung, but as currently presented I think the authors draw too many conclusions from this panel.
As suggested by the reviewer, we have moved figure 3e to supplementary figure S2.We have shortened the corresponding text and introduced a detailed presenta9on of the proteomic changes at the pathway level in supplementary text 1.Furthermore, we have indicated that further analyses are required to confirm whether changes at pathway level leading to metabolic specialisa9on are convergent.

Fig
Fig 5g is unclear.Why are the + and o at slightly different heights?

7.
Fig 1: I would suggest swapping the order of Figs 1A and 1B.1B contextualises 1A by showing the nega9ve correla9on between colonisa9on 9me in the lung and growth rate, which nicely outlines why we see a difference between the early and late growth rates shown in 1A.I don't see what it adds to the results by going arer 1A, however.
9. Fig 2D: Why are confidence intervals only shown for pyruvate late and formate late?If the confidence intervals fall within the thick lines in the other condi9ons -or if the other lines have no replicates -please describe this in the figure legend arer confidence intervals are discussed.