Comparative Metabolomics and Transcriptomics Reveal Multiple Pathways Associated with Polymyxin Killing in Pseudomonas aeruginosa

Pseudomonas aeruginosa has been highlighted by the recent WHO Global Priority Pathogen List due to multidrug resistance. Without new antibiotics, polymyxins remain a last-line therapeutic option for this difficult-to-treat pathogen. The emergence of polymyxin resistance highlights the growing threat to our already very limited antibiotic armamentarium and the urgency to understand the exact mechanisms of polymyxin activity and resistance. Integration of the correlative metabolomics and transcriptomics results in the present study discovered that polymyxin treatment caused significant perturbations in the biosynthesis of lipids, lipopolysaccharide, and peptidoglycan, central carbon metabolism, and oxidative stress. Importantly, lipid A modifications were surprisingly rapid in response to polymyxin treatment at clinically relevant concentrations. This is the first study to reveal the dynamics of polymyxin-induced cellular responses at the systems level, which highlights that combination therapy should be considered to minimize resistance to the last-line polymyxins. The results also provide much-needed mechanistic information which potentially benefits the discovery of new-generation polymyxins.

treatment. In the present study, correlative metabolomics and transcriptomics were conducted to investigate cellular metabolic perturbations and differentially expressed genes in paired polymyxin-susceptible and -resistant P. aeruginosa strains in response to polymyxin treatment. This integrated omics approach provides detailed mechanistic insights into the mechanisms of antibacterial killing and resistance to polymyxins, as well as potential intracellular targets to tackle resistance to this last-line class of antibiotics.

RESULTS
Metabolic and lipidomic perturbations in response to polymyxin B. According to the Clinical and Laboratory Standards Institute (CLSI) and European Committee on Antimicrobial Susceptibility Testing (EUCAST) guidelines (2017), the clinical resistance breakpoint of polymyxin B against P. aeruginosa is Ն4 mg/liter (25). Therefore, 4 mg/ liter polymyxin B was used to examine the metabolic responses in the paired polymyxin-susceptible P. aeruginosa PAK (polymyxin B MIC, 1 mg/liter) and polymyxinresistant PAKpmrB6 (polymyxin B MIC, 16 mg/liter) strains at 1, 4, and 24 h (see Fig. S1 in the supplemental material). To ensure that both the hydrophilic and lipophilic metabolites were detected, hydrophilic interaction liquid chromatography (HILIC) and reversed-phase liquid chromatography (RPLC) were employed, yielding 871 (see Data Set S1 in the supplemental material) and 427 (Data Set S2) putatively identified metabolites, respectively. These metabolites were involved in a wide range of pathways, including amino acids, carbohydrates, lipids, nucleotides, and secondary metabolites. The technical performance was monitored based on periodic analysis of pooled biological quality control (PBQC) samples with the median relative standard deviation (RSD) values of 10.5% (HILIC) and 12.9% (RPLC), which were well within the acceptable limits for metabolomics (Table S1) (26). The reproducibility of metabolomics data are potentially affected by analytical techniques, sampling procedures, and natural biological variability; the median RSD values for all sample groups were between 25 and 40% (Table S1), which were generally within acceptable limits (26). Moreover, the principalcomponent analysis (PCA) score plots showed that the PBQC samples were clustered tightly by both methods (Fig. S2), indicating minimal analytical variation and consistency with the RSD values.
From PCA score plots, the most significant overall metabolic change induced by polymyxin B (4 mg/liter) treatment was observed at 1 h, whereas little difference was observed at 24 h between the polymyxin-susceptible and -resistant P. aeruginosa strains. The perturbations of metabolites in the wild-type PAK strain with polymyxin B treatment at 4 h were still distinguishable, but not significant in strain PAKpmrB6 (Fig. 1A). With regards to the number of significantly changed metabolites (fold change [FC] Ͼ 2; P Ͻ 0.05) at 1 and 4 h, polymyxin B treatment resulted in 12.9% and 9.0% metabolic changes in the polymyxin-susceptible wild-type PAK strain, respectively, but only 4.9% and 3.7% metabolic changes in the polymyxin-resistant PAKpmrB6 strain, respectively ( Fig. 1B and 2; Data Set S3). Intriguingly, polymyxin B treatment (4 mg/liter) dramatically perturbed a wide range of key pathways, including lipids, carbohydrates, nucleotides, and amino acids, particularly in the wild-type PAK strain. Moreover, the volcano plots revealed that lipids and the associated metabolites were the most significantly perturbed features with 4 mg/liter polymyxin B at 1 and 4 h in both strains (Fig. 2). It is notable that the intracellular levels of two intermediates responsible for the synthesis of UDP-4-amino-4-deoxy-L-arabinose (UDP-L-Ara4N) were significantly elevated by polymyxin B in strain PAK, but not in strain PAKpmrB6 ( Fig. 2 and Table 1). Moreover, our metabolomics results also showed that 4 mg/liter polymyxin B induced dramatic depletions of nucleotides (27.8%; 10 out of 36 nucleotides) in PAK only at 1 h, but no significant perturbations at 4 h. However, nucleotides were not significantly changed in PAKpmrB6 due to polymyxin B treatment at 4 mg/liter over 24 h (Fig. 1B and  2 and Table 1). Polymyxin B induced differential transcriptomic changes between polymyxinsusceptible and -resistant P. aeruginosa strains. Polymyxin B at 4 mg/liter also induced markedly different transcriptomic responses between strains PAK and PAKpmrB6 (Fig. 3). The PCA score plots revealed that 4 mg/liter polymyxin B successfully distinguished the treated samples from those of untreated PAK at 1 h; however, no significant difference was detected at 4 h (Fig. 3A). Notably, polymyxin B treatment led to 558/226 (up/down) differentially expressed genes (DEGs) in strain PAK at 1 h, which sharply decreased to 94/7 at 4 h ( Fig. 3B) (FC Ͼ 2; false-discovery rate [FDR] Ͻ 0.05). Our results demonstrated that bacterial responses to polymyxin B were rapid in polymyxinsusceptible PAK. In the polymyxin-resistant PAKpmrB6 strain, a minimal transcriptomic response was induced by 4 mg/liter polymyxin B, with only 8/1 (up/down) and 11/1 (up/down) DEGs observed at 1 and 4 h, respectively (Fig. 3). Even in the absence of polymyxin B treatment, a number of different DEGs (40/4) were observed between the wild-type PAK and the pmrB mutant PAKpmrB6 (Table S2). Notably, transcriptional regulators PA4581-4585, PmrAB (PA4776-4777), and the regulated arnBCADTEF-pmrE operon (PA3552-3558), spermidine biosynthesis (PA4773-4774), ferrous transfer (PA4357-4359), ABC transporter (PA3396-3399), as well as heme biosynthesis (PA0510-0515) were all significantly upregulated (FC Ͼ 2) in polymyxin-resistant PAK-pmrB6 compared to polymyxin-susceptible PAK. Conversely, in PAKpmrB6, the arginine biosynthesis gene cluster (PA5171-5173) was downregulated (FC Ͻ Ϫ2) compared to the wild-type strain. Our transcriptomic data also showed significantly increased expression of crpA in PAKpmrB6 than in PAK, which is known to play a role in high-level polymyxin resistance in P. aeruginosa (27).
Polymyxin B induced alterations in lipid profiles and the related metabolites. Polymyxin B treatment at 4 mg/liter significantly perturbed the relative abundance of lipids and the related metabolites in both PAK and PAKpmrB6 strains mainly at 1 and 4 h (Data Set S3A). Specifically, a number of fatty acids containing 8 to 18 carbons were significantly depleted by polymyxin B treatment in both strains at 1 and 4 h ( Table 1), whereas the dramatic depletion of lysophospholipids (phospholipids that have lost one fatty acyl chain from di-acyl phospholipids), including lysophosphatidylethanolamine (LPE), lysophosphatidylcholine (LPC), and lysophophatidylglycerol (LPG), was mainly  observed in strain PAK after polymyxin B treatment (4 mg/liter) at both 1 and 4 h ( Table 1). Intriguingly, further analysis through RPLC revealed that in PAK after 4 mg/ liter polymyxin B treatment, the levels of di-acyl phospholipids, in particular PC, PE, PG. and phosphatidylserines (PS) were dramatically increased at 1 h but decreased at 4 h (Fig. 4A). However, 4 mg/liter polymyxin B did not induce significant changes in the levels of phospholipids and lyso-phospholipids in PAKpmrB6 (Fig. 4A). Moreover, it is notable that polymyxin B at 4 mg/liter significantly altered the levels of specific metabolites associated with phospholipid metabolism in both strains. Specifically, sn-glycerol-3-phosphate (an important precursor in phospholipid synthesis) was depleted (FC ϭ Ϫ3.1 to Ϫ4.8) in both PAK and PAKpmrB6 strains with polymyxin B treatment (4 mg/liter) at 1 and 4 h ( Table 1 and Fig. 4B) (28). Consistently, its upstream substrates, sn-glycerol-3-phosphocholine (FC ϭ 3.7 to 8.8) and sn-glycerol-3phosphoethanolamine (FC ϭ 1.9 to 3.9), accumulated in both strains in response to 4 mg/liter polymyxin B (Fig. 4B). Moreover, the decreased levels of choline (FC ϭ Ϫ2.1 to Ϫ2.8) which is a by-product of sn-glycerol-3-phosphocholine degradation were observed in both strains after 4 mg/liter polymyxin B treatment for 1 and 4 h (Table 1) (28,29).  their levels were dramatically decreased by polymyxin B at 8ϫ MIC in both strains at the same time point (Fig. 5).
Polymyxin B caused metabolic and transcriptomic changes in central carbon metabolism and stress response pathways. Polymyxin B (4 mg/liter) differentially altered the levels of metabolites related to central carbon metabolism in both PAK and PAKpmrB6 strains. Specifically, coenzyme A (CoA), which is associated with the tricarboxylic acid (TCA) cycle and plays an important role in the synthesis of fatty acids (30), was significantly decreased in its relative abundance in strain PAK (FC ϭ Ϫ13.5 and Ϫ4.9, respectively) but increased in strain PAKpmrB6 (FC ϭ 6.7 and 2.1, respectively) under polymyxin B treatment (4 mg/liter) at 1 and 4 h (Table 1). Notably, a critical metabolite in both pentose phosphate and glycolysis pathways, D-glucose-6phosphate, was depleted by polymyxin B (4 mg/liter) in PAK at 1 h (FC ϭ Ϫ2.7), but it had no dramatic change in PAKpmrB6. It is also known that D-glucose-6-phosphate functions as a substrate in the synthesis of trehalose-6-posphate as well as the formation of UDP-glucose (31,32). In our study, the levels of trehalose-6-phosphate were significantly increased in response to 4 mg/liter polymyxin B in both PAK and PAKpmrB6 at 1 h (FC ϭ 4.0 and 2.3, respectively) ( Fig. 2 and Table 1).
Metabolic and transcriptomic perturbations in lipid A remodelling. The most common mechanism of polymyxin resistance in P. aeruginosa is the modification of lipid A phosphate groups with positively charged L-Ara4N (11). From our untargeted metabolomics study, two key intermediates related to L-Ara4N biosynthesis, UDPglucuronate (FC ϭ 7.8) and UDP-4-deoxy-4-formamido-L-arabinose (UDP-L-Ara4FN) (FC ϭ 41.6) were significantly enriched in strain PAK after 4 mg/liter polymyxin B treatment even at 1 h (Table 1). Consistently, extensive modifications of lipid A phosphates with an L-Ara4N group (peaks at m/z 1497, 1577, and 1747) rapidly occurred even within 1 h in response to polymyxin B (4 mg/liter) (Fig. 8)  for pmrB) (Fig. 6). This upregulation activated the L-Ara4N biosynthesis pathway which was responsible for the enriched abundance of UDP-glucuronate (FC ϭ 7.8) and UDP-L-Ara4FN (FC ϭ 41.6) ( Table 1) as well as the L-Ara4N modifications of lipid A (Fig. 8) (38). In addition, in our recent study, we reported an upregulation of pagL (PA4661) (FC Ͼ 3) in both PAK and PAKpmrB6 strains in response to 4 mg/liter polymyxin B over 24 h, which resulted in the deacylation of lipid A in both strains (39). The relationships of lipid A deacylation with polymyxin resistance have been confirmed using neutron reflectometry (39). Moreover, the effect of polymyxin B at a higher concentration (i.e., 8ϫ MIC) on the lipid A modification pathway in both PAK (8 mg/ liter) and PAKpmrB6 (128 mg/liter) was also investigated. Remarkably, in L-Ara4N synthesis, UDP-glucuronate and UDP-L-Ara4FN were not significantly altered by polymyxin B at 8ϫ MIC in either strain PAK or PAKpmrB6 (Data Set S3B). Accordingly, unlike 4 mg/liter polymyxin B treatment, 8ϫ MIC polymyxin B did not cause dramatic lipid A modifications (Fig. 8).

DISCUSSION
For the first time, the present study elucidated different metabolic and transcriptomic changes between paired polymyxin-susceptible and -resistant P. aeruginosa strains associated with both polymyxin killing and the development of resistance (Fig. 9). The current literature supports the idea that polymyxins affect the OM physical integrity and lead to phospholipid exchange (4). As predicted, polymyxin B treatment (4 mg/liter) led to profound perturbations of several key lipid metabolites mainly in the wild-type PAK strain (Fig. 4A), possibly due to the membrane-targeted killing by polymyxins (5). Our metabolomics data indicated that, after treatment with a clinically relevant concentration of polymyxin B, the decreased level of lyso-phospholipids in strain PAK was probably due to the suppression of phospholipid degradation, and therefore resulted in an accumulation of phospholipids at 1 h. As the decreased levels of fatty acids and sn-glycerol-3-phosphate might contribute to the inhibition of phospholipid synthesis (40), a decreased level of phospholipids observed in PAK at 4 h supports this hypothesis. Our results are consistent with the metabolomic findings for In strain PAK, polymyxin B (4 mg/liter) significantly upregulated the PmrAB regulatory system and related genes, resulting in the increased L-Ara4N synthesis. Meanwhile, polymyxin B induced severe oxidative stress and osmotic imbalance and significantly decreased metabolite levels related to LPS and peptidoglycan synthesis in the wild-type PAK strain. In contrast, in strain PAKpmrB6, 4 mg/liter polymyxin B did not activate the PmrAB system and related genes; therefore, the L-Ara4N synthesis pathway was not affected. Moreover, polymyxin B significantly elevated the metabolites associated with LPS and peptidoglycan synthesis pathways in the polymyxin-resistant PAKpmrB6 strain. In addition, polymyxin B induced osmotic stress in both PAK and PAKpmrB6 strains. It is also notable that in both strains, a high concentration of polymyxin B (8ϫ MIC) significantly decreased fatty acid and phospholipid levels but increased lipoamino acid levels. The red and blue arrows indicate that the levels of metabolites were significantly increased and decreased, respectively (fold change Ͼ 2, P Ͻ 0.05, FDR Ͻ 0.05, Student's t test).
Acinetobacter baumannii that colistin treatment significantly decreased the OM lipid levels and disrupted membrane asymmetry (18,20,41). Importantly, this finding supports the notion proposed in our recent study that decreased phospholipid levels observed in the pmrB mutants PAKpmrB6 and PAKpmrB12 possibly play important roles in polymyxin resistance in P. aeruginosa (42). On the other hand, a high concentration of polymyxin B (8ϫ MIC) substantially disrupted the OM phospholipids and their synthesis in both P. aeruginosa PAK and PAKpmrB6 (Fig. 5). Therefore, we speculate that the dramatically elevated lipoamino acid levels may contribute to the stabilization of the OM by counteracting the negative charge of LPS (43), thereby diminishing the interaction with positively charged polymyxin molecules. However, the precise biological functions of lipoamino acids in Gram-negative bacteria and the relationship with polymyxin resistance are not clear and are under investigation in our laboratory.
In addition to disorganizing the OM, our results indicated that polymyxin B also interfered with the biosynthesis of both LPS and cell wall in the wild-type PAK strain (Fig. 7); this finding is consistent with our previous study in A. baumannii (18). However, the different metabolic changes in strain PAKpmrB6 suggested that the polymyxinresistant strain responded to polymyxin treatment by promoting the biosynthesis of LPS and peptidoglycan, possibly to cope with the cell envelope damages by polymyxins. The central carbon metabolism plays an essential role in generating metabolic precursors in bacteria (30). Our metabolomics data showed different perturbations in the levels of CoA and D-glucose-6-phosphate between the polymyxin-susceptible PAK strain and the polymyxin-resistant PAKpmrB6 strain (Table 1). Notably, elevated trehalose-6-phosphate levels were evident after 4 mg/liter polymyxin B treatment at 1 h. As the dephosphorylated form trehalose is a well-known osmoprotectant when bacterial cells are under osmotic stress (44), the elevated trehalose-6-phosphate levels suggest that 4 mg/liter polymyxin B induced osmotic imbalance in both polymyxinsusceptible and -resistant strains at 1 h.
Our transcriptomics results indicated that the dramatically increased expression of PmrAB-regulated speDE, feoABC, and ssuAC operons as well as several ABC transporters in strain PAK was related to oxidative stress due to polymyxin B treatment (Fig. 6). Spermidine has been reported to stabilize and protect the bacterial OM against antibiotic and oxidative damage (33). The upregulation of the speDE operon in strain PAKpmrB6 was possibly related to oxidative stress caused by polymyxin B and contributed to polymyxin resistance (33,45). Consistently, our recent study revealed perturbed spermidine levels and the methionine salvage cycle in the pmrB mutants compared to the wild-type PAK strain even without polymyxin treatment (42). The FeoABC transporter is a well-conserved system in the transport of ferrous iron in bacteria (46). It is known that feoABC plays an important role in promoting bacterial growth in response to low iron and Mg 2ϩ concentrations (35). Therefore, it was very likely that the expression of the feoABC operon played a key role in the bacterial survival from polymyxin B treatment. Our results also suggested that polymyxin B treatment potentially led to sulfate starvation, while the upregulation of the ssu operon helped to counter the oxidative stress (36). In addition, the significant upregulation of several genes regulated by ParRS (e.g., pagL, PA1797, and mexXY) due to polymyxin B treatment in both the wild-type and pmrB mutant strains is indicative of the interactions between the two TCRs PmrAB and ParRS (13,14,27).
Our correlated metabolomic and transcriptomic data demonstrated, for the first time, that bacterial cells rapidly responded to polymyxins by lipid A modifications, thereby minimizing the interaction and subsequent cellular damage and developing resistance (Fig. 8). This finding is also consistent with the minimal metabolic and transcriptomic perturbations observed in strain PAKpmrB6 following polymyxin B treatment (4 mg/liter), which was very likely due to the dramatically diminished interaction between polymyxin molecules and L-Ara4N modified lipid A (39). Considering the pharmacokinetics/pharmacodynamics of polymyxins, the 4-mg/liter polymyxin B concentration employed in the present study is higher than the unbound average steadystate concentration (fC ss,avg ) of 1.17 mg/liter (i.e., C ss,avg 2.79 mg/liter ϫ unbound fraction 0.42) in patients with the currently recommended dosage regimens (47); combination therapy should be strongly recommended to minimize any potential emergence of resistance to this important last-line class of antibiotics. Polymyxin B at a super MIC (e.g., 8ϫ MIC) severely damaged bacterial cells and halted the remodelling of their LPS (e.g., adding L-Ara4N to lipid A) (Fig. 8), indicating that both strains were not able to generate polymyxin resistance at super MICs. Unfortunately, for P. aeruginosa isolates with MICs Ն 0.5 mg/liter, an fC ss,avg of 8ϫ MIC is difficult to achieve in patients with the current dosage regimens (47)(48)(49).
Collectively, our findings on the complex and dynamic interactions of multiple cellular pathways provide key mechanistic insights into understanding the mechanisms of activity and resistance to polymyxins. This study highlights the urgency of developing rational combination therapy to reduce polymyxin resistance due to rapid lipid A modifications. Our results may also benefit the discovery of much-needed newgeneration polymyxins targeting polymyxin-resistant Gram-negative 'superbugs'.

MATERIALS AND METHODS
Chemical and reagents. Polymyxin B was purchased from Sigma-Aldrich (Sydney, New South Wales, Australia). The stock solution of polymyxin B (1 mg/ml) was prepared using Milli-Q water (Millipore Australia, North Ryde, NSW, Australia) and filtered through 0.22-m syringe filters (Sartorius, Melbourne, VIC, Australia).
Bacterial strains and culture. P. aeruginosa PAK and PAKpmrB6 strains were obtained from the Moskowitz laboratory (Massachusetts General Hospital, MA, USA) (50). The polymyxin B MICs of P. aeruginosa PAK and PAKpmrB6 strains were determined by broth microdilution (25). Prior to experiments, strain PAK was subcultured on Mueller-Hinton agar plates, while the mutant strain PAKpmrB6 was subcultured on Mueller-Hinton agar plates containing 4 mg/liter polymyxin B and incubated for 16 to 18 h at 37°C. A single colony was then inoculated into 10 ml of cation-adjusted Mueller-Hinton broth (CaMHB) (Oxoid) and incubated overnight at 37°C with shaking at 150 rpm. The overnight culture was diluted 1:100 into three different reservoirs with 100 ml fresh CaMHB medium and then grown to an optical density at 600 nm (OD 600 ) of 0.50 Ϯ 0.02 (ϳ10 8 CFU/ml). The bacterial cultures of strains PAK and PAKpmrB6 were then treated with polymyxin B at 4 mg/liter, 8 mg/liter (PAK), and 128 mg/liter (PAKpmrB6) for 1, 4, and 24 h; the untreated bacterial culture served as a control sample.
Preparation of metabolomics samples. Cellular metabolites of PAK and PAKpmrB6 strains were extracted by a previously optimized method with slight modifications (18). Briefly, both treated and untreated samples were collected at 0, 1, 4, and 24 h for metabolite extraction. Bacterial cultures (20 ml) were collected and immediately transferred into 50-ml ice-cold Falcon tubes. Samples were then rapidly quenched in a dry ice/ethanol bath for ϳ30 s to stop the metabolic processes and normalized according to OD 600 at 0.50 Ϯ 0.02 to ensure that the bacterial cell counts were at ϳ10 8 CFU/ml. Cell pellets were then collected from 10 ml normalized culture after centrifugation at 3,220 ϫ g at 4°C for 10 min. After the cell pellets were washed twice with 2 ml ice-cold 0.9% NaCl, they were resuspended in 500 l chloroform/methanol/water (CMW) (1:3:1 [vol/vol]) containing 1 M generic internal standards (CHAPS, CAPS, PIPES, and TRIS). A freeze-thaw process was performed three times to lyse the cells and release cellular metabolites. The extracted samples were centrifuged at 3,220 ϫ g at 4°C for 10 min, and a 300-l supernatant was collected, which was followed by a further centrifugation at 14,000 ϫ g for 10 min at 4°C to obtain particle-free supernatants (200 l) for LC-MS analysis.
LC-MS analysis of metabolites. Both hydrophilic interaction liquid chromatography (HILIC) and reversed-phase liquid chromatography (RPLC) coupled to high-resolution mass spectrometry (HRMS) were employed to ensure the detection of both hydrophilic and hydrophobic metabolites. Samples were analyzed on a Dionex U3000 high-performance liquid chromatography system (HPLC) in tandem with a Q-Exactive Orbitrap mass spectrometer (Thermo Fisher) in both positive and negative ion modes with a resolution at 35,000. The HILIC method was described previously in detail (18). Briefly, samples maintained at 4°C were eluted through a ZIC-pHILIC column (5 m, polymeric, 150 by 4.6 mm; SeQuant, Merck) by mobile phase A (20 mM ammonium carbonate) and mobile phase B (acetonitrile). The gradient started with 80% mobile phase B at a flow rate of 0.3 ml/min and was followed by a linear gradient to 50% mobile phase B over 15 min. The Ascentis Express C8 column (5 cm by 2.1 mm, 2.7 m) (catalog no. 53831-U; Sigma-Aldrich) was applied in the RPLC method. The samples were controlled at 4°C and eluted by mobile phase A (40% of isopropanol and 60% of Milli-Q water with 8 mM ammonium formate and 2 mM formic acid) and mobile phase B (98% of isopropanol and 2% of Milli-Q water with 8 mM ammonium formate and 2 mM formic acid). The linear gradient started from 100% mobile phase A to a final composition of 35% mobile phase A and 65% mobile phase B over 24 min at 0.2 ml/min. All samples were analyzed within a single LC-MS batch to avoid variations. The pooled quality control samples (QC), internal standards, and total ion chromatograms were assessed to evaluate the chromatographic peaks, signal reproducibility, and stability of analytes. To assist the identification of metabolites, a mixture of ϳ600 metabolite standards was analyzed within the same batch.
Data processing and statistical analyses. Metabolomics data analyses were performed using mzMatch and IDEOM (http://mzmatch.sourceforge.net/ideom.php) (51). The quantification of each metabolite was based on the chromatogram raw peak height (relative intensity). Univariate and multivariate statistical analyses were conducted using MetaboAnalyst 3.0 (52). Prior to analysis, the data set of relative peak intensity was normalized by the median, log transformed, and auto-scaled. Unsupervised principalcomponent analysis (PCA) was applied for the analysis of global metabolic profiles, while Student's t test (P Ͻ 0.05; false-discovery rate [FDR] Ͻ 0.05) was used to identify significantly changed metabolites in polymyxin B-treated samples relative to untreated control samples at each time point. Metabolites that showed a fold change of Ͼ2 were further analyzed and subjected to metabolic pathway analysis using the KEGG pathway (53), BioCyc (54), and Visualisation and Analysis of Networks containing Experimental Data (VANTED) software (55).
RNA extraction and analysis of RNA sequencing data. Polymyxin B-treated and untreated samples (1.5 ml) were collected at 0, 1, 4, and 24 h for RNA extraction. RNAprotect (Qiagen) was used for the sample collection in order to preserve gene expression profiles. RNA was isolated using an RNeasy minikit (Qiagen) in accordance with the manufacturer's instructions. RNA-Seq was undertaken using Illumina HiSeq (56). RNA-seq data were analyzed according to the methods described previously (39). Briefly, the transcriptome was assembled based on the RNA-Seq data using Trinity RNA-Seq software, and the RNA-Seq reads were aligned according to the genome sequences of P. aeruginosa PAK and PAKpmrB6 strains using Subread (57). The RNA-Seq data were analyzed using voom and limma linear models through Degust interactive Web-based RNA-Seq visualization software (http://degust.erc.monash.edu) (58).
Isolation and structural characterization of lipid A. Lipid A was isolated by mild acid hydrolysis as previously described (59,60). Briefly, P. aeruginosa PAK and PAKpmrB6 were treated by polymyxin B at 4 mg/liter and 8ϫ MIC for 1 h. The bacterial cell pellets were harvested from 100 ml of normalized culture (OD 600 ϭ 0.50 Ϯ 0.02) via centrifugation at 3,220 ϫ g for 20 min and washed twice with 5 ml PBS. Cell pellets were then resuspended in 4 ml PBS, which was followed by resuspension in 5 ml chloroform and 10 ml methanol to make a single-phase Bligh-Dyer (chloroform/methanol/water, 1:2:0.8 [vol/vol]). After 15-min centrifugation at 3,220 ϫ g, the supernatant was removed, leaving LPS in the pellets. After washing once with 5 ml single-phase Bligh-Dyer solvent, the LPS pellets were resuspended in 10.8 ml of hydrolysis buffer (50 mM sodium acetate [pH 4.5] with 1% sodium dodecyl sulfate [SDS]) and homogenized via sonication with a probe tip sonicator (Misonix, USA) at a constant duty cycle (20 s at 50% output). The samples were then incubated in a boiling water bath for 45 min and allowed to cool to room temperature. To extract lipids after hydrolysis, 12 ml of chloroform and 12 ml of methanol were added to the 10.8-ml hydrolysis solution to make a double-phase Bligh-Dyer (chloroform/methanol/water, 1:1:0.9 [vol/vol]). The lower phase containing lipid A was finally collected and dried under nitrogen gas stream. Structural analysis of lipid A was performed using mass spectrometry in negative ion mode on a Q-Exactive Hybrid Quadrupole-Orbitrap mass spectrometer.
Data availability. All transcriptomic raw data were deposited in GenBank under accession numbers SRX4714399 to SRX4714440. The metabolomic data set is publicly available at MetaboLights under the study identifier MTBLS751.