Strain collection for metabolomic analyses

To identify molecular mechanisms of metabolic specialization which modify the interactions between the host and the pathogen occurring during the infection of pwCF, we selected from our longitudinal collection of 474 P. aeruginosa clinical strains [39] pairs of early (*_E; first P. aeruginosa isolated from the person) and late (*_L) clinical isolates, longitudinally isolated from eight different pwCF (Fig 1A). Previous studies have shown that slow growth is an important phenotype associated with long term adaptive evolution to the CF airways, appearing within the first 2 to 3 years of the infection [9]. Therefore, we used growth rate as a proxy for within-host evolution. This allowed us to increase the odds of selecting strains presenting unexplored mechanisms of metabolic specialization. Accordingly, we investigated metabolic adaptation in a diverse collection of 16 clinical strains from 8 distinct pwCF (Fig 1A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Collection of P. aeruginosa clinical strains.

(A) Growth rate (hour^-1) of early (circles), late (triangles), and PAO1 (square) strains colored by their clone type. The fold change of growth rate from early to late is shown as bar chart on the right y axis and as number value at the base of each bar. (B) The maximal optical density (maxOD) of all isolates, grouped by early vs. late comparison. Statistical significance was calculated by unpaired Welch t test and indicated as *** (p < 0.001). (C) Growth rate (hour^-1) of each strain in M9 minimal media containing 20 mM of a single carbon source glucose (Glu), lactate (Lac), or succinate (Scc). The data underlying this figure can be found in S4 Data. maxOD, maximal optical density.

https://doi.org/10.1371/journal.pbio.3002781.g001

As by selection criteria, when grown in synthetic cystic fibrosis medium 2 (SCFM2) (which has been formulated to resemble the CF mucus composition [40]), early isolates showed a high growth rate, comparable with the reference strain, PAO1. Conversely, late isolates showed a 2.2- to 5.1-fold reduction in growth rate, in addition to a lower maximal optical density (paired t test, p = 0.002) (Fig 1A and 1B). Moreover, some late isolates were unable to grow in minimal medium in presence of single carbon sources such as glucose, lactate, or succinate, indicative of a metabolic constraint such as auxotrophy [41] (Fig 1C).

The genome of each isolate had been previously sequenced, and their genotype and phylogenetic relationship determined [39]. Specifically, each pair of early and late clinical isolates are phylogenetically related, share a common ancestor, and belong to the same “clone type” (defined as genomes sharing more than 10,000 SNPs) (Fig 1A). Each pwCF was colonized by a distinct clone type with evolutionary histories spanning between 1.4 and 7 years (S1 Table). On average, the difference in mutations between early and late strains was 46 ± 5 with the exception of strains of the DK36 clone type which show an increased number of mutations (402 mutation differences) due to hypermutation (S1 Table) [39]. As previously reported [42], each single isolate represents the most abundant representative of the population from a sputum sample. These selection criteria, therefore, allowed us to analyze isolates from different individuals with distinct clinical and evolutionary histories, rather than focusing on the heterogeneous population present in pwCF.

Within-host evolution selects for specialized metabotypes

To evaluate the degree of metabolic specialization in the clinical isolates, dynamic exo-metabolomics was performed on cells growing in SCFM2 (S1 Data). This approach allowed for analysis of the assimilation and secretion patterns of specific metabolites during the growth of each isolate, providing a dynamic profile of their metabolic activity. Principal component analysis (PCA), k-means clustering, and hierarchical cluster analysis (HCA) of the extracellular metabolomes revealed the presence of 3 distinct specialized metabolic configurations in late strains, hereafter defined as “adapted metabotypes” (DK15 and DK55 metabotype 1; DK12 metabotype 2; DK03, DK13, and DK36 metabotype 3). These metabotypes separate the late isolates from each respective early isolate and from those strains not exhibiting metabolic changes, hereafter referred to as “naïve metabotypes” (Fig 2A). Note that whereas PCA emphasizes the largest differences in the metabolomic profiles, the HCA clearly separates the early strains from the late strains, indicating a certain degree of metabolic specialization in all late isolates (Fig 2B). Indeed, both the hierarchy of assimilation (metabolite half-lives, OD₅₀) which represents the order of assimilation of the available carbon sources, and the secretion of metabolites differ between early and late isolates (Fig 2C and 2D). Of note, late strains of DK12, DK36, and DK55 secreted high amounts of pyruvate which is a key metabolite connecting glucose metabolism to the TCA cycle. Detailed information on the specific metabolic preferences of each metabotype is presented in S1 Text. Furthermore, the net balance between assimilated and secreted metabolites (total mM) varies between early and late strains (Fig 2E), which positively correlates with the lower biomass of the late isolates (Pearson’s r = 0.9; p = 0.0023) (Figs 1B and 2E). In other words, a high percentage of assimilated carbon sources are catabolized in the late isolates and secreted back into the culture medium, thus limiting their availability for biomass production. This suggests either a specific metabolic configuration for the late isolates—the objective of which is not biomass accumulation—or an apparent inefficient metabolism.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Exo-metabolomics analysis reveals adapted metabotypes.

(A) PCA showing separation of strains based on their time-resolved exo-metabolomes. Strains are shown as circles (early), triangles (late), and square (PAO1), with dashed arrows indicating the most notable trajectories from early to late, colored by clone type. Metabotypes were designated based on iterative k-means clustering analysis and HCA. See Material and methods for details. (B) HCA of the exo-metabolomics data, showing a general separation between early and late strains. Branches for early isolates are indicated in blue, late isolates in red and PAO1 in black. Accuracy of the HCA was tested by bootstrapping where gray values within the branches represent the % of bootstrap values for 10,000 replicates. (C) Assimilation hierarchies of the analyzed metabolites. Each symbol represents the half-life (OD₅₀) of a specific metabolite while each line connects the same metabolite for each pair of early and late isolates. Missing connecting curves indicates that either the strain did not assimilate the metabolite, or that the OD₅₀ was outside the analyzed assimilation window. (D) Secretion plots showing variations in the concentration of acetate, pyruvate, and formate (mM) relative to normalized growth (OD₆₀₀). Shaded areas indicate the 95% confidence intervals. (E) Table showing the total amount (mM) of carbon sources assimilated for each strain. Secreted metabolites were subtracted from the total to account for their excretion in the medium. Clone types are separated by gray boxes with “E” indicating early and “L” indicating late isolate. The data underlying this figure can be found in S1 and S4 Data. HCA, hierarchical cluster analysis; PCA, principal component analysis.

https://doi.org/10.1371/journal.pbio.3002781.g002

Changes in proteome allocation supports the metabolic specialization of clinical isolates

To test the hypothesis that the observed metabolic specialization is rooted in changes in expression of proteins involved in cellular metabolism, we analyzed the proteome of each clinical strain. To this end, we performed whole cell proteomics to compare protein expression profiles between PAO1, early, and late strains (comparisons being early versus PAO1, late versus PAO1, and late versus early). This allowed us to evaluate proteome changes in the clinical isolates relative to a laboratory reference strain (early versus PAO1, late versus PAO1) and to identify in the clinical isolates possible molecular mechanism(s) underpinning the metabolic specialization we observed (late versus early). Of the 2,061 proteins identified, 740 were differentially expressed in at least 1 comparison (early versus PAO1 or late versus early) (S2 Data). Similar to the metabolomic analysis, when comparing the normalized expression profile of all identified proteins, PCA readily separated the proteome of the late isolates from that of the early isolates, suggesting that changes in metabolite profiles might be rooted in the proteome (Fig 3A and 3B). Most of the early and some of the late proteomes co-localized with the reference strain PAO1, indicating little or no changes in proteome allocation (Fig 3A and 3B). However, the proteome of the late isolates of lineages DK12, DK17, DK36, and DK55 separated from the respective early strains to form a cluster of adapted proteomes (Fig 3A and 3B). Notably, the DK03 strains formed an independent cluster, indicative of a lineage-specific proteomic signature (Figs 1A, 3A, and 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Proteome composition changes support metabolism specialization.

(A) PCA showing separation of strains based on their mid-exponential proteomes. Strains are shown as circles (early), triangles (late), and square (PAO1), with dashed arrows indicating most notable trajectories from early to late, colored by clone type. (B) HCA of the proteomic data containing 2,061 proteins identified, showing a general separation between early and late strains. Branches for early isolates are indicated in blue, late isolates in red, and PAO1 in black. Accuracy of the HCA was tested by bootstrapping where gray values within the branches represent the % of values for 10,000 replicates. (C) HCA of the 235 differentially expressed proteins when comparing late vs. early strains within the COG categories related to metabolism. Proteomes are represented by dashed lines for “cluster A” and continuous lines for “cluster B.” Accuracy of the HCA was tested by bootstrapping where gray values within the branches represent the % of bootstrap values for 10,000 replicates. (D) Parallel plots of the number of DE proteins within the COG categories related to metabolism in cluster A (blue, shaded) and B (red, transparent). For each category, the mean of each cluster is indicated. The difference in the total number of DE proteins between clusters when matching by category was computed by two-tailed paired t test where ** indicates p = 0.0047. The data underlying this figure can be found in S2 Data. COG, Clusters of Orthologous Groups; DE, differentially expressed; HCA, hierarchical cluster analysis; PCA, principal component analysis.

https://doi.org/10.1371/journal.pbio.3002781.g003

A similar relationship between the proteomes is obtained through HCA, specifically when considering the 235 differentially expressed proteins (late versus early comparison) belonging to the Clusters of Orthologous Groups (COG) categories involved in metabolism (Fig 3C). When comparing differentially expressed proteins between late versus early strains, 2 clusters of proteomes are identified. Cluster A represents proteomic changes in clone types without any apparent metabolic specialization. In contrast, cluster B represents proteomic changes in strains which underwent metabolic specialization in the adaptive evolution process from early to late (DK12, DK36, and DK55). This result supports our hypothesis that metabolic specialization is likely rooted in changes in the expression of proteins involved in cellular metabolism (Fig 3C). Notably, cluster B is characterized by a significantly higher number of differentially expressed proteins in the COG categories involved in metabolism relative to cluster A (two-tailed paired t test p = 0.0047) (Figs 3D and S1A). Similarly, several categories of proteins involved in the metabolism of amino acids, fatty acids, and sugars are statistically enriched in early and late strains (S1B Fig) providing a molecular explanation for the altered hierarchy of assimilation of the carbon sources and the reduced growth rate of the adapted metabotypes (Fig 2C and 2D). Detailed information on the convergent expression at the pathway level of metabolic proteins is presented in S1 Text (S2 Fig).

Although late strains of DK12, DK36, and DK55 belong to different metabotypes, their proteomes move in the same direction (Fig 3A–3C). This indicates that largely similar proteomes can sustain distinct metabolic configurations which ultimately depend on the metabolic fluxes thorough specific pathways and on their regulation. We note that, in the case of the DK17 strains comparison (cluster B) (Fig 3A–3C), although our metabolomic analysis did not detect any major metabolic rewiring (Fig 2A and 2B), proteome reorganization could be related to other metabolic processes which were not analyzed in our study.

Changes in virulence traits during within patient evolution

To evaluate the relationship between metabolic specialization and virulence, we analyzed which differentially expressed proteins were enriched when comparing late versus early isolates. Interestingly, most of the changes (based on their KEGG and GO categories) were related to adaptation to the infection environment, redox balance, and virulence (Figs 4A and S1B). For example, most of the lineages showed expression changes in phenazine biosynthesis (KEGG) and secondary metabolite(s) biosynthetic process (GO), both of which are deeply involved in redox-balance, cell homeostasis, metabolism, and virulence [43,44].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Expression of virulence factors differs in clinical strains vs. PAO1.

(A) Enrichment analysis for KEGG and GO terms by the specific comparisons (Early vs. PAO1; Late vs. PAO1; Late vs. Early). Icons represent individual lineages and are colored by fold-enrichment. (B) Differential expression of virulence factors in early clinical strains vs. PAO1. Only differentially expressed proteins are represented and are colored by clone type. (C) HCA of differentially expressed proteins of virulence factors in late vs. early clinical strains. The data underlying this figure can be found in S2 Data. HCA, hierarchical cluster analysis.

https://doi.org/10.1371/journal.pbio.3002781.g004

Several changes in protein expression are already apparent in early strains relative to PAO1 indicating a different response of clinical strains to the airway-like conditions accompanying growth in SCFM2 (Fig 4A). Specifically, early strains show increased expression of proteins involved in alginate production, hydrogen cyanide (HCN), phenazine biosynthesis, and PQS (the Pseudomonas quinolone signal), and decreased expression of proteins involved in flagella biogenesis, lipopolysaccharide (LPS) O-antigen metabolism, and pyoverdine production, which are all hallmarks of an acute infection phenotype (Fig 4B). Early stage colonization and acute infection are thought to require expression of several virulence factors promoting host tissue injury and immune response impairment [45]. By contrast, when comparing late versus early strains, the adapted metabotypes—and specifically strains belonging to cluster B (DK12, DK17, DK36, and DK55)—show convergent up-regulation of proteins related to alginate, phenazine, PQS, and the type VI secretion system (T6SS), suggesting that metabolism plays an active role in regulating virulence and the chronic infection phenotype (Figs 4C and S4). Reduced virulence and cytotoxicity are advantageous for long-term infection and the establishment of a persistent infection [46]. Interestingly, the DK12 late strain also shows lower expression (-3-fold) of the T3SS toxin, ExoT, which is known to play a role in preventing phagocytosis, in the induction of cytoskeletal reorganization, and in host cell apoptosis [16]. Overall, the identified pattern of proteomic changes suggests that early isolates show reduced expression of virulence factors relative to PAO1, with an even greater reduction being associated with the late isolates. This is consistent with their persistence during CF airway infections [46].

aceE and aceF mutations leads to metabolic specialization and impaired virulence

From the previously published genome sequences [39], we searched for mutations in the genome of each clinical isolate which might explain the secretion of pyruvate and the metabolic specialization of late isolates. In particular, we focused on those genes encoding the phosphoenolpyruvate-pyruvate-oxaloacetate node (S2 Fig), since such mutations might be expected to lead to elevated pyruvate secretion. This led to the identification of mutations in genes encoding the PDHc, aceE and aceF, in late strains of DK12 and DK36, respectively. The PDHc catalyzes the conversion of pyruvate to acetyl-CoA, connecting sugar metabolism with the TCA cycle (S2 Fig). Interestingly, the aceE and aceF genes are candidate pathoadaptive genes, suggesting positive selection for such mutations in pwCF [39]. In our strain collection of 474 isolates from 34 pwCF [39], we identified the presence of 18 different and independent aceE/aceF mutations (4 indels and 14 SNPs, of which 3 were synonymous) in 18 separate lineages. These lineages were present in more than half of the patients, supporting the hypothesis that modulation of the PDHc might be selected for during airway infections (S5 Fig). Moreover, in situ expression of the aceE and aceF genes is reduced in sputum samples collected from chronically infected pwCF [47]. This suggests an undescribed role of the PDHc activity which connects central carbon metabolism and pathogenicity during infection.

The DK12_L and DK36_L strains contained a +TCCC duplication at position 813 in aceF and a T→C transition at position 551 in aceE, respectively (Fig 5A). To evaluate the contribution of these mutations to the bacterial phenotype, independent of the underlying historical contingency of the clinical strains, we generated recombinant PAO1 derivative strains containing the same mutations. The aceF mutation leads to a frameshift starting from Lys 273, whereas the aceE mutation leads to a Phe→Ser amino acid change at position 184. The recombinant aceE and aceF mutant strains show reduced growth rate (Fig 5B) and increased secretion of pyruvate (maximal pyruvate accumulation by the aceE mutant was 0.5 mM, whereas for aceF, the value was 13 mM) (Fig 5C). By comparison, maximal pyruvate accumulation by the DK12_L and DK36_L late strains was 8.5 mM and 4.6 mM, respectively (Fig 2D). The PAO1-derived aceE mutant strain clearly has a milder phenotype (a smaller reduction in growth rate and lower pyruvate secretion) compared with the aceF mutant strain, suggesting partial functionality of the PDHc. As previously reported, in P. aeruginosa, acetate is catabolized into acetyl-CoA, and can, therefore, metabolically complement the growth defects of PDH mutants [31,48,49]. Indeed, we were able to fully restore the growth phenotype of the aceE strain by supplementing the bacterial culture with acetate to replenish the pool of acetyl-CoA, whereas this only partially restored growth in the aceF mutant strain (Fig 5B). A similar result was obtained by expressing a wild-type copy of the aceE gene under its native promoter which complements the growth defect of the strain (S6A Fig). These data confirm that the frameshift in aceF has a much greater impact on PDHc activity than the SNP in aceE. Surprisingly, in laboratory conditions, phenotypes such as biofilm formation, motility, redox susceptibility, pyoverdine production, and antibiotic susceptibility showed no statistical difference between the PAO1 wild-type strain and the aceE and aceF mutant strains, except for a slight decrease in twitching motility and mildly increased tobramycin susceptibility (S6 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. aceE and aceF mutations cause metabolic specialization.

(A) Schematic of the aceE and aceF genes including location and type of mutation found in the DK36 (aceE) and DK12 (aceF) late strains. (B) Growth rate (hour^-1) in SCFM2 (blue) and SCFM2 supplemented with 5 mM acetate (red) for PAO1, aceE and aceF mutant strains, and DK12 and DK36 late clinical isolates. Bars indicate mean ± SEM, with icons representing biological replicates. Statistical significance is assessed by two-tailed unpaired parametric Welch t test and indicated as * (p < 0.05), ** (p < 0.01), or **** (p < 0.0001). (C) Pyruvate secretion (mM) for PAO1 wt (yellow), aceE (cyan), and aceF (magenta) mutant strains over 24 h. Icons represent biological replicates. (D) PCA and HCA of whole-cell proteomics for PAO1 wt and aceE and aceF mutant strains. Filled icons indicate without (−) and unfilled icons indicate with (+) 5 mM acetate. E Number of differentially expressed proteins for mutant strain vs. PAO1 wt comparisons: in orange in absence of acetate (+/− mutation) and in green in presence of acetate (+/− mutation (ace)). In purple, the number of differentially expressed proteins for either the PAO1 wt or the aceE or aceF mutant strain in presence vs. absence of acetate (+/− acetate). (F) Metabolic map of enzymes related to pyruvate and acetyl-CoA metabolism for mutant strains vs. PAO1 in absence (left) or presence (right) of acetate. Reactions are colored by their pathway if the underlining enzyme is differentially expressed (dashed if down-regulated or continuous if up-regulated). Enzymes responsible for each reaction are indicated by numbered circles. For details on individual enzymes, see S7 Fig. (G) Differentially expressed proteins involved in virulence in aceE and aceF mutant strains vs. PAO1 in absence of acetate. Jittered icons indicate specific proteins that are up-regulated (red plus) or down-regulated (blue circle). The data underlying this figure can be found in S3 and S4 Data. HCA, hierarchical cluster analysis; PCA, principal component analysis; SCFM2, synthetic cystic fibrosis medium 2.

https://doi.org/10.1371/journal.pbio.3002781.g005

To characterize the effect of the aceE and aceF mutations more broadly, we performed whole cell proteomics on the aceE and aceF recombinant strains (and on the PAO1 progenitor) in presence and absence of acetate. In total, we quantified 3,246 proteins and identified 449 as differentially expressed in at least 1 comparison of either the mutant versus wild-type or of the mutant in presence versus absence of acetate (S3 Data). In the absence of acetate, the proteome of the aceE and aceF mutants was clearly different compared with that of the wild-type progenitor, PAO1 (Fig 5D). PC1, which encompasses >50% of the variance in the data set, separated the aceF mutant proteome from that of the wild type, whereas the proteome of the aceE mutant was more similar to that of the wild type, separating along PC2 (accounting for just 20% of the total variance in the data set, Fig 5D). As previously noted, supplementation of the growth medium with acetate altered both the aceE and aceF mutant proteomes, with both moving closer to that of the wild type (Fig 5D). This effect was greater in the aceF mutant, where growth in acetate decreased the number of up-regulated proteins (cf. PAO1) by more than half (from 301 to 126) (Fig 5E). Unsurprisingly, the expression of several enzymes involved in pyruvate and acetyl-CoA metabolism (e.g., lactate dehydrogenase (41), acetyl-coenzyme A synthetase (6), citrate synthase (28)) return to wild-type levels following the addition of acetate (Figs 5F and S7A). Similarly, for the clinical isolates DK12_L and DK36_L, the categories of proteins involved in amino acid/lipid metabolism and energy conversion presented the largest number of differentially expressed proteins (S7B Fig). Moreover, proteins involved in terpene, propionate, isoprenoid and branched-chain amino acid metabolism (all of which are directly connected to pyruvate and acetyl-CoA metabolism) were statistically enriched in the PAO1-derived aceF mutant, indicating a reorganization of both central and peripherical pathways to cope with reduced synthesis of acetyl-CoA (S7C Fig). The PAO1-derived aceE mutant also shows an extreme down-regulation of the oxygen-sensing transcriptional regulator Dnr, which is known to be required for denitrification but also regulates acetate metabolism and the T6SS [50–52]. This mirrors the down-regulation seen specifically for cluster B of the clinical isolates (Fig 3C), which may suggest a role for Dnr down-regulation in metabolic specialization (S7D Fig).

Importantly, the aceE and aceF mutants showed increased expression of proteins involved in alginate production and T6SS, as well as decreased expression of proteins involved in the T3SS, including the secreted factor ExoT (in strains aceF and DK12_L) (Fig 5G and S3 Data). This indicates that the aceE and aceF mutations contribute to the expression profile of virulence determinants shown by the late clinical isolates of DK12 and DK36 (Figs 4 and 5G and S3 Data).

Pyruvate dehydrogenase mutations modulate pathogenicity in ALI culture infections

To test whether the secretion of pyruvate and the reduced expression of virulence determinants in the PAO1-derived aceF and aceE mutants lead to reduced infectivity, we performed host-bacteria infections using an ALI infection model system. This model system is composed of mucociliated differentiated airway epithelial cells which represent the airways and provides insights into the host response including epithelial damage and recruitment of the immune system. Overall, the aceF mutant displayed a broad suppression of virulence, including reduced epithelium damage and innate immune recognition during the infection (Fig 6). The transepithelial electrical resistance (TEER) which quantifies the integrity and permeability of the epithelial layer, LDH release which quantifies the epithelium cellular damage, and the bacterial count which quantifies the growth and penetration of the bacteria through the epithelium to the basolateral side of the ALI transwells, all showed reduced values after 14 h of infection by the aceF mutant, compared with the PAO1 progenitor (Fig 6A). This is in line with the behavior of a ΔpscC mutant defective in T3SS, which also shows severely reduced virulence (Fig 6A). By contrast, the aceE mutant elicited epithelial damage similar to PAO1, indicating that the mutation does not influence bacterial penetration (Fig 6A). This is consistent with the proteomic analysis showing no significant differential expression in this strain of proteins involved in the T3SS, which is the leading cause of epithelial damage (Fig 5G). However, both the aceE and aceF mutants, together with the ΔpscC mutant, elicit lower interleukin 8 (IL-8) release, a cytokine that is secreted by epithelial cells and is necessary for the recruitment of the immune system at the site of infection (Fig 6B). These results are corroborated by confocal microscopy data, which show a similar colonization profile on the airway epithelium of the wild-type and aceE mutant, and of the aceF and ΔpscC mutant, respectively. It is worth noting that nuclei shedding, seen as nuclei (in blue) being pushed to the apical side of the epithelium (in red), was observed only for PAO1 and not for the aceE mutant (Fig 6C), which may suggest a slight reduction in the aggressiveness of the infection. This mechanism of reduced cellular damage and reduced inflammation seems not to depend on the growth defect associated with the aceF mutant, since the ΔpscC mutant shows comparable aggressiveness to the aceF mutant but a wild-type–like growth rate.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Aggressiveness of aceE and aceF mutants in ALI airway infections.

(A) Mean ± SEM of TEER (Ω∙cm²), LDH release, and CFUs in the apical and basolateral ALI compartments after 14 h of infection in fully differentiated BCi-NS1.1 cells. (B) IL-8 release into the basolateral media indicating inflammation caused by the invading bacteria. Icons represent each biological replicates. Mock represents un-infected control cells. Statistical significance was determined by two-way ANOVA for TEER and CFU measurements and one-way ANOVA for LDH and IL-8 measurements and indicated as * (p < 0.05), ** (p < 0.01), *** (p < 0.001), and **** (p < 0.0001). (C) Confocal images of ALI transwells following infection with P. aeruginosa in green (GFP), epithelium in red (Phalloidin), and nuclei in blue (To-pro). Scale bar = 40 μm. The data underlying this figure can be found in S4 Data. ALI, air–liquid interface; CFU, colony-forming unit; TEER, transepithelial electrical resistance.

https://doi.org/10.1371/journal.pbio.3002781.g006

Importantly, in our longitudinal collection of clinical P. aeruginosa strains, 18 different lineages (specific clone types obtained from different pwCF; 89 isolates) harbored mutations in the aceE and aceF genes and/or carried additional mutations in T3SS genes [39]. In 9 lineages, PDHc mutations are present alone or anticipate additional T3SS mutations. In 7 lineages, T3SS mutations precede PDHc mutation or both systems are mutated at the same time (4 lineages) already in the first isolate of the lineage (S8 Fig).

Altogether, these results suggest a role for the aceE and aceF mutations in persistent infections and a link between pyruvate metabolism and virulence.

Strains and media

The collection of clinical isolates used in this study is a subset of the collection published in Marvig and colleagues and Bartell and colleagues [9,39]. The local ethics committee at the Capital Region of Denmark (Region Hovedstaden) approved the use of the stored P. aeruginosa isolates (registration number H-21078844). For information about specific clinical isolates, see S1 Table.

For construction of aceE, aceF, and ΔpscC mutant strains, derivatives of pACRISPR were constructed with the Uracil Specific Excision Reagent (USER) cloning [72]. The plasmids and primers used are listed in S1 and S2 Tables, respectively. Target DNA fragments were amplified from gDNA of clinical strains DK12 Late and DK36 Late using the Phusion U polymerase kit (Thermo Fisher Scientific, United States of America). For the ΔpscC mutant strain primers were designed to create matching short overlaps between the DNA fragments that should be stitched together in the final plasmid. The resulting products were treated with FastDigest DpnI enzyme (Thermo Fisher Scientific, USA) and ligated with the USER Enzyme (New England Biolabs, USA). PCR was performed on random E. coli DH5alpha colonies using OneTaq 2X Master Mix (New England Biolabs, USA) to confirm the correct insertions and further sequenced with Sanger method (Eurofins Scientific, Luxembourg). aceE, aceF, and ΔpscC mutations in the genome of P. aeruginosa PAO1 were introduced using the previously developed CRISPR-Cas9 system and following the indicated protocol [73] with the addition of 0.2% arabinose in the growth medium during transformation. To confirm the presence of desired mutations, PCR fragment of the genome around the mutations was amplified and sequenced with Sanger sequencing (Eurofins Scientific, Luxembourg). To perform confocal microscopy analyses, strains were tagged with GFP using 4 parental mating using a mini-Tn7 delivery method [74] but for strains aceE and aceF with the additional supplementation of 5 mM acetate to ensure growth of PDHc dysregulated target strains. Transformants were identified by green fluorescence and validated by comparing growth rate to the untagged target strain.

To complement the mutation in the PDH (aceE gene), we constructed a complementation strain aceE(rev) containing a wild-type copy of the aceE gene under the control of its native promoter and integrated it into the genome of the aceE mutant using the mini-Tn7 delivery method [74] as above. The plasmid pIP281 carrying the aceE gene and the Tn7 system was assembled with USER cloning [72]. The plasmids and primers used are listed in S1 and S2 Tables, respectively.

Bacteria were grown in Synthetic Cystic Fibrosis Media 2 (SCFM2) [75]. To reduce viscosity and allow for HPLC analysis, DNA and mucins were excluded [76]. Cultures were grown at 37°C and 250 rpm.

Dynamic exo-metabolomics analyses

Sampling of supernatants was performed in 96-well deep well plates (Cat. No. 0030502302; Eppendorf, Hamburg, Germany) with an air:liquid ratio of 1:1 using a high-throughput dilution-based growth method as previously described [77]. Immediately before sampling, OD₆₀₀ of the cultures were measured in a Synergy MX microtiter plate reader (BioTek Instruments, Winooski, Vermont, USA). Supernatants were stored at −80°C until HPLC-analysis. For organic acids and sugars (glucose, lactate, formate, acetate, and pyruvate), a Dionex Ultimate 3000 system (Thermo Scientific, Waltham, USA) with a HPx87H ion exclusion column (125–0140, Aminex, Dublin, Ireland), equipped with a guard column (125–0129, BioRad, Hercules, California, USA) and guard column holder (125–0131, BioRad, Hercules, California, USA) was used. Samples were injected with an injection volume of 20 μl and eluted, using a 5 mM H₂SO₄ mobile phase, at an isocratic flow of 0.6 ml min⁻¹ at 45°C for 30 min. Pyruvate was analyzed by UV detection at a wavelength of 210 nm, using a System Gold 166 UV-detector (Beckman Coulter, Brea, USA), while the rest of the metabolites were analyzed by RI detection, using a Smartline RI detector 2300 (KNAUER Wissenschaftliche Geräte, Berlin, Germany). For amino acids (aspartic acid, glutamic acid, serine, histidine, glycine, threonine, arginine, alanine, tyrosine, valine, phenylalanine, isoleucine, leucine, lysine, and proline), 20 μl of the thawed sample was diluted 1:10 by mixing with 80 μl Ultrapure MilliQ water and 100 μl of internal standard (20 μg/ml 2-aminobutyric acid and sarcosine) before injection of 56.5 μl into the instrument. Prior to injection into the column, derivatization of amino acids was performed in the HPLC-instrument by automatic mixing with the following eluents: (i) 0.5% (v/v) 3-mercaptopropionic acid in borate buffer 0.4 M at pH 10.2; (ii) 120 mM iodoacetic acid in 140 mM NaOH; (iii) OPA reagent (10 mg/ml o-phtalaldehyde and 3-mercaptopropionic acid in 0.4 M borate buffer); (iv) FMOC reagent (2.5 mg/ml 9-fluorenylmethyl chloroformate in acetonitrile); and (v) buffer A (40 mM Na₂HPO₄, 0.02% (w/v) NaN₃ at pH = 7). Following derivatization, the samples were separated isocratically on a Dionex Ultimater 3000 HPLC with fluorescence detector (Thermo Scientific, Waltham, USA) through a Gemini C18 column (00F-4439-E0, Phenomenex, Værløse, Denmark) equipped with a SecurityGuard Gemini C18 guard column (AJ0-7597, Phenomenex, Værløse, Denmark) with 5 mM H₂SO₄ at a flowrate of 1 ml min⁻¹ at 37°C for 31 min. Amino acids were detected using an UltiMate 3000 Fluorescence variable wavelength UV detector (FLD-3400RS, Waltham, Massachusetts, USA).

Proteomic analyses

To maintain a 1:1 air:liquid ratio, as in the metabolomic analysis, 25 ml SCFM2 cultures were inoculated from ON cultures in 50 ml Falcon tubes at OD₆₀₀ of 0.05 and then incubated in an orbital shaker at 37°C and 250 rpm until sampling by centrifugation at mid-exponential phase. Five biological replicates were analyzed for PAO1 and the clinical isolates while 3 biological replicates for the PAO1 derivative strains aceE and aceF both in presence and absence of 5 mM acetate. Pellets were washed twice with PBS and stored at −80°C until protein extraction. No specific preparation or enrichments were performed to identify secreted proteins. For experiment with clinical isolates, protein extraction was done in Lysis buffer A (100 mM Tris, 50 mM NaCl, 1 mM tris(2-carboxyethyl)phosphine (TCEP), 10% glycerol, pH = 7.5) with cOmplete Mini protease inhibitor (Roche) by sonication at amplitude 10 for 3 × 10 s cycles with 20 s cooling between, followed by acetone precipitation and resuspension in Lysis buffer B (6 M Guanidinium hydrochloride, 5 mM TCEP, 10 mM chloroacetamide, 100 mM Tris-HCl, pH = 8.5). For experiment with PAO1 derivative mutant strains, protein extraction was done in Lysis buffer B by bead beating with 3-mm zirconium oxide beads at 99°C for 5 min in a Tissuelyzer (Retsch, MM 400) at 25 Hz, then boiled, still at 99°C, in heat block (Eppendorf Thermomixer C) for 10 min while shaking/mixing at 2,000 rpm. In both cases, protein concentrations were determined by micro BCA Protein Assay Kit (Thermo Scientific, prod #23235) and 100 μg protein was used for trypsin digest in Trypsin/Lys-C Buffer (0.1 μg/μl trypsin, 50 mM Ammonium Bicarbonate). The reaction was stopped by addition of 10 μl 10% TFA and samples were desalted by stagetipping with C18 filter plugs (Solaμ HRP 2 mg/1 ml 96-well plate, Thermo Scientific). Peptide samples were stored in 40 μl of 0.1% formic acid at 4°C until LC-MS analysis.

For the experiment with clinical isolates, LC-MS/MS was carried out using a CapLC system (Thermo Fisher Scientific, Waltham, USA) coupled to an Orbitrap Q-exactive HF-X mass spectrometer (Thermo Fisher Scientific, Waltham, USA). The first samples were captured at a flow of 10 μl/min on a precolumn (μ-pre-column C18 PepMap 100, 5 μm, 100 Å) and then at a flow of 1.2 μl/min. Peptides were separated on a 15 cm C18 easy spray column (PepMap RSLC C18 2 μm, 100Å, 150 μm × 15 cm). The applied gradient went from 4% acetonitrile in water to 76% over a total of 60 min. MS-level scans were performed with Orbitrap resolution set to 60,000; AGC Target 3.0e6; maximum injection time 50 ms; intensity threshold 5.0e3; dynamic exclusion 25 s. Data-dependent MS2 selection was performed in Top 20 Speed mode with HCD collision energy set to 28% (AGC target 1.0e4, maximum injection time 22 ms, Isolation window 1.2 m/z). For the experiment with the aceE and aceF mutants, peptides were loaded onto a 2 cm C18 trap column (Thermo Fisher 164946), connected in-line to a 15 cm C18 reverse-phase analytical column (Thermo EasySpray ES904) using 100% Buffer A (0.1% formic acid in water) at 750 bar, using the Thermo EasyLC 1200 HPLC system, and the column oven operating at 30°C. Peptides were eluted over a 70 min gradient ranging from 10% to 60% of 80% acetonitrile, 0.1% formic acid at 250 nL/min, and the Orbitrap Exploris instrument (Thermo Fisher Scientific) was run in DIA mode with FAIMS Pro Interface (Thermo Fisher Scientific) with CV of −45 V. Full MS spectra were collected at a resolution of 120,000, with an AGC target of 300% or maximum injection time set to “auto” and a scan range of 400 to 1,000 m/z. The MS2 spectra were obtained in DIA mode in the orbitrap operating at a resolution of 60.000, with an AGC target 1,000% or maximum injection time set to “auto,” and a normalized HCD collision energy of 32. The isolation window was set to 6 m/z with a 1 m/z overlap and window placement on. Each DIA experiment covered a range of 200 m/z resulting in 3 DIA experiments (400 to 600 m/z, 600 to 800 m/z, and 800 to 1,000 m/z). Between the DIA experiments a full MS scan is performed. MS performance was verified for consistency by running complex cell lysate quality control standards, and chromatography was monitored to check for reproducibility.

Phenotypic characterizations

Growth curves were performed by inoculating 1 μl of overnight culture in 149 μl of media, using 96-well microtiter plates (Cat. No. 650001; Greiner Bio-One, Kremsmünster, Austria), covered with plate seals (Ref. 4306311, Thermo Fisher Scientific, United Kingdom) and incubated at 37°C and 250 rpm in a BioTek ELx808 Absorbance Microtiter Reader (BioTek Instruments, Winooski, Vermont, USA) for 24 to 48 h. Antibiotic MICs were determined by microdilution. ON cultures were standardized to OD₆₀₀ = 0.5 and diluted 1:2,500 to reach 5 × 10⁵ CFU/ml in fresh SCFM2 media (LB for Azithromycin). Growth assays were performed at increasing antibiotic concentrations and MIC determined based on final OD. Resistance to oxidative stress was measured as the diameter of clearance zones around diffusion disks saturated with 5 μl of fresh 30% H₂O₂ after 24 h incubation at 37°C on LB agar plates cast with 3 ml overlay agar containing 100 μl of LB ON culture standardized to OD₆₀₀ = 1. Pyoverdine production was measured as the fluorescence at 400/460 nm excitation/emission on a Synergy H1 Hybrid Multi-Mode Reader (BioTek Instruments, Winooski, Vermont, USA) of supernatants normalized against OD₆₀₀ of ON cultures in King’s B medium. Biofilm formation assay was done as previously described for NUNC peg lids [9] (NUNC cat no. 445497). Motility was measured as the diameter of the motility zone around single colonies deposited in middle layer of 0.3% (swimming), surface layer of 0.6% (swarming), or bottom layer of 1.5% (twitching) LB agar motility plates, after incubation at 37°C for 24 to 48 h. For PAO1 wild-type and aceEF mutant strains, pyruvate secretion was determined from supernatants of samples taken after 0, 4, 8, and 24 h of growth in SCFM2. Supernatants were stored at −80°C and analyzed by the HPLC method also used for exo-metabolomics.

Infection of ALI cultures

For the ALI infections, the BCi-NS1.1 cells were used [78]. Cells were cultured in Pneumacult-Ex Plus medium (STEMCELL Technologies, 05040) supplemented with Pneumacult-Ex 50x supplement (STEMCELL Technologies, 05008), 96 ng/ml hydrocortisone (STEMCELL Technologies, 07925), and 10 μm Y-27632 ROCK inhibitor (Bio-Techne #1254/10) in a 37°C, 5% CO₂ humidified incubator. Following expansion, 1.5 × 10⁵ cells were seeded onto 6.5-diameter-size transwells with 0.4 μm pore polyester membrane inserts (Corning Incorporated, 3470) previously coated with human type I collagen (Gibco, A1048301). ALI was established once cells reached full confluency by removing media from the apical chamber and replacing media in the basolateral chamber with Pneumacult-ALI maintenance medium (STEMCELL Technologies, 05001). Pneumacult-ALI maintenance medium was supplemented with Pneumacult-ALI 10× supplement (STEMCELL Technologies, 05003), Pneumacult-ALI maintenance supplement (STEMCELL Technologies, 05006), 480 ng/ml hydrocortisone, and 4 μg/ml heparin (STEMCELL Technologies, 07980). ALI cultures were grown in a 37°C, 5% CO₂ humidified incubator for 30 days, with media replacement every 2 to 3 days. Epithelial polarization was monitored by measurements of the TEER using a chopstick electrode (STX2; World Precision Instruments). Following 15 days under ALI conditions, the apical surface was washed with PBS every media change to remove accumulated mucus. Biological replicates of bacterial strains were obtained from single colonies on LB agar plates grown ON as precultures in LB and then diluted to an OD₆₀₀ of 0.05 before sampling at mid-exponential phase by centrifugation, washing with PBS and resuspending in PBS at a density of 10⁵ CFU/ml. Fully differentiated BCi-NS1.1 cells were inoculated with 10³ CFU from the apical side, diluted in 10 μl PBS. Control wells were incubated with bacteria-free PBS. Cells were incubated for 14 h at 37°C, followed by addition of 200 μl PBS to the apical side and measurement of the TEER. CFUs were determined by platting 10 μl of 6-fold serial dilutions on LB-agar plates in technical triplicates both for the initial inoculum, as well as for the apical and basolateral solutions following TEER measurements. The basolateral media was also used for measurements of LDH and IL-8 release in technical triplicates, using the Invitrogen CyQUANT LDH Cytotoxicity Assay Kit (Invitrogen, C20301) and Human IL-8/CXCL8 DuoSet ELISA Kit (R&D Systems, DY208) according to the manufacturer’s instructions. BCi-NS1.1 cells on transwell inserts were rinsed once with PBS and fixed by adding 4% (wt/vol) paraformaldehyde (PFA) to both apical and basolateral chambers for 20 min at 4°C. After washing, cells were permeabilized and blocked for 1 h with a buffer containing 3% BSA, 1% Saponin, and 1% Triton X-100 in PBS. Cells were stained on the apical side with Phalloidin-AF488 (Invitrogen, 65-0863-14) and TO-PRO3 (Biolegend, 304008) diluted in a staining buffer (3% BSA and 1% Saponin in PBS) at a 1:500 dilution for 2 h at room temperature. Transwells were removed from their supports with a scalpel and mounted on glass slides with VECTASHIELD Antifade Mounting Medium (VWR, VECTH-1000). Images were acquired with a Carl Zeiss LSM 510 Confocal Laser Scanning Microscope (40× magnification, 1.3 oil) and analyzed using the ImageJ software.

Data analysis

For exometabolomics, all chromatograms were analyzed and used to construct standard curves constructed for absolute quantification of concentrations for all 20 metabolites, using software Chromeleon v7.2.9. All other analysis of exo-metabolomic data was done in JMP Pro 15.0. To compare between different strains and media batch-effects, OD₆₀₀ was normalized against final OD₆₀₀ of that strain in the experiment and concentrations were normalized against specific concentrations in SCFM2 controls for each batch. Missing values were replaced with 20% of lowest value detected for any metabolite in any sample. PCA and HCA were done on C_metabolite of all quantified metabolites for 16 samples (8 time points from biological duplicates) of each strain. PCA and iterative k-means clustering was done on covariance using JMP Pro 15.0 software. HCA was performed using the “average” clustering method and “correlation” for distance using the R package pvclust 2.2. Reliability of the clusters were analyzed by 10,000 bootstraps. Metabolite half-life (OD₅₀), defined as the normalized OD₆₀₀-value, where 50% of the starting concentration of a metabolite is present, was calculated from the sigmoidal mechanistic growth model (Equation: a(1 –bExp(–cx)) where a = asymptote, b = scale, and c = rate) [79]. Naïve and adapted metabotypes were designated based on PCA, iterative k-means clustering (range 3 to 10) followed by curation based on HCA. Specifically, iterative k-means clustering readily identified adapted metabotypes 2 and 3 based on 3 clusters and the highest Cubic Cluster Criterion statistical analysis. Cluster 1 is determined based on 5 k-means and agrees with the grouping generated by HCA followed by bootstrapping. Differences of metabolic preferences between metabotypes were analyzed by comparing the assimilation and secretion profiles between strains using OD₅₀ values. The net balance of carbon sources assimilated during the growth was calculated by summing the amount (mM concentration) of assimilated metabolites and subtracting the secreted ones. Metabolomics data are enclosed in S1 Data.

For the experiment with clinical isolates, proteomic raw files were analyzed using Proteome discoverer 2.4 software with the following settings: Fixed modifications: Carbamidomethyl (C) and Variable modifications: oxidation of methionine residues. First search mass tolerance 20 ppm and a MS/MS tolerance of 20 ppm. Trypsin as enzyme and allowing 1 missed cleavage. FDR was set at 0.1%. The Match between runs window was set to 0.7 min. Quantification was only based on unique peptides and normalization between samples was based on total peptide amount. For the experiment with mutant strains, raw files were analyzed using Spectronaut (version 16.2). Dynamic modifications were set as Oxidation (M) and Acetyl on protein N-termini. Cysteine carbamidomethyl was set as a static modification. All results were filtered to a 1% FDR, and protein quantitation done on the MS1 level. The data was normalized by RT dependent local regression model [80] and protein groups were inferred by IDPicker. In both cases, spectra were matched against the P. aeruginosa PAO1 reference proteome (UniProt Proteome ID UP000002438). In both experiments, any protein that was not quantified in at least 3 of 5 or 2 of 3 biological replicates for all strains were excluded from the analysis. Using JMP Pro 15.0, abundances were normalized by Log₂-transformation and biological replicates used for missing value imputation. PCA Wide was done on correlations using JMP Pro 15.0 and HCA was performed on mean Log₂(abundance) and Log₂(Fold change) for the relevant comparisons. HCA was performed using the “average” clustering method and “correlation” for distance using the R package pvclust 2.2. Reliability of the clusters were analyzed by 10,000 bootstraps. Differential expression was determined by two-way ANOVA with Tukey’s multiple comparisons test, using GraphPad Prism 9.3.1, and defined as those protein-comparisons where adjusted P value ≤0.05 and Log₂(Fold Change) ≥ |0.6|. Enrichment analysis was done using the DAVID Functional Annotation Bioinformatics Microarray Analysis tool from lists of Locus Tags of proteins that were differentially expressed, separated into lists of up-regulated and down-regulated proteins, respectively, for each of the relevant strain comparisons. The reference genome was set as Pseudomonas aeruginosa. Proteomics data are enclosed in S2 and S3 Data.

Growth rates and maxOD were calculated in JMP Pro 15.0. Blanks were first subtracted from OD₆₀₀-measurements and values converted to cm^-1 (using Greiner dimensions for pathlength). The stationary phase was excluded, and growth rates were calculated by fitting the Exponential 3P model to the exponential phase (r² > 0.99). Mean ± SD was calculated from biological replicates. GraphPad Prism 9.3.1 was used for statistical analysis of biofilm formation, motility, redox sensitivity, pyoverdine production, and MICs (see S6 Fig for details), as well as for ALI infection experiments. ALI data are represented as mean ± SEM. Replicates represent independent experiments performed with cells from different passages. Statistical comparisons were calculated using two-way ANOVA for TEER and CFU measurements and one-way ANOVA for LDH and IL-8 measurements. Statistical significance was considered for p value < 0.05. All figures were finalized in Adobe Illustrator Artwork 27.0.