Alterations of Metabolic and Lipid Profiles in Polymyxin-Resistant Pseudomonas aeruginosa

ABSTRACT

Multidrug-resistant Pseudomonas aeruginosa presents a global medical challenge, and polymyxins are a key last-resort therapeutic option. Unfortunately, polymyxin resistance in P. aeruginosa has been increasingly reported. The present study was designed to define metabolic differences between paired polymyxin-susceptible and -resistant P. aeruginosa strains using untargeted metabolomics and lipidomics analyses. The metabolomes of wild-type P. aeruginosa strain K ([PAK] polymyxin B MIC, 1 mg/liter) and its paired pmrB mutant strains, PAKpmrB6 and PAKpmrB12 (polymyxin B MICs of 16 mg/liter and 64 mg/liter, respectively) were characterized using liquid chromatography-mass spectrometry, and metabolic differences were identified through multivariate and univariate statistics. PAKpmrB6 and PAKpmrB12, which displayed lipid A modifications with 4-amino-4-deoxy-l-arabinose, showed significant perturbations in amino acid and carbohydrate metabolism, particularly the intermediate metabolites from 4-amino-4-deoxy-l-arabinose synthesis and the methionine salvage cycle pathways. The genomics result showed a premature termination (Y275stop) in speE (encoding spermidine synthase) in PAKpmrB6, and metabolomics data revealed a decreased intracellular level of spermidine in PAKpmrB6 compared to that in PAKpmrB12. Our results indicate that spermidine may play an important role in high-level polymyxin resistance in P. aeruginosa. Interestingly, both pmrB mutants had decreased levels of phospholipids, fatty acids, and acyl-coenzyme A compared to those in the wild-type PAK. Moreover, the more resistant PAKpmrB12 mutant exhibited much lower levels of phospholipids than the PAKpmrB6 mutant, suggesting that the decreased phospholipid level was associated with polymyxin resistance. In summary, this study provides novel mechanistic information on polymyxin resistance in P. aeruginosa and highlights its impacts on bacterial metabolism.

INTRODUCTION

Pseudomonas aeruginosa is an opportunistic Gram-negative pathogen that can cause both acute and chronic infections, particularly in people with cystic fibrosis (1, 2). Moreover, it is notorious for developing high-level resistance to the majority of antibiotics, often leaving clinicians with few therapeutic options (2–5). Polymyxins (i.e., polymyxin B and colistin), a family of cyclic lipopeptides, are among the very few antibiotics that retain activity against multidrug-resistant P. aeruginosa (6, 7). The putative mode of action of polymyxins involves an initial electrostatic interaction between positively charged l-α,γ-diaminobutyric acid (Dab) residues of polymyxins and negatively charged phosphates within the lipid A and inner core moieties of lipopolysaccharide (LPS) in the Gram-negative outer membrane (OM) (7, 8). This is followed by hydrophobic interactions between the fatty acyl chains and positions 6 and 7 of polymyxins and the bacterial OM, which are also essential for the bacterial killing (9).

Although the current incidence of polymyxin resistance in P. aeruginosa is relatively low, suboptimal use and poor antibiotic stewardship have led to a notable increase in polymyxin-resistant clinical isolates (10–13). In particular, the recent emergence of the mcr-1 plasmid indicates that polymyxin resistance can readily spread (14). At a genetic level, polymyxin resistance in P. aeruginosa often results from mutations in the two-component regulatory systems (TCRs) PhoPQ, PmrAB, and/or ParRS (15–18). Specific mutations in these TCRs can trigger constitutive expression of the arnBCADTEF-pmrE operon, which encodes enzymes responsible for covalent attachment of 4-amino-4-deoxy-l-arabinose (l-Ara4N) to lipid A (15–17, 19, 20). ColRS and CprRS TCRs also interact with PhoPQ, presumably regulating additional genes that contribute to polymyxin resistance in P. aeruginosa (21, 22). Moreover, disruptions in galU, lptC, wapR, or ssg genes responsible for LPS biosynthesis in P. aeruginosa have been reported to affect the permeability of the OM and/or to inhibit LPS modifications, causing increased polymyxin susceptibility (19). Collectively, the current literature on polymyxin resistance is largely focused on the studies of genetic alterations and indicates elaborate mechanisms of polymyxin resistance in P. aeruginosa; unfortunately, metabolic perturbations associated with polymyxin resistance are poorly understood. The present study employed untargeted metabolomics and lipidomics to investigate the altered metabolism in P. aeruginosa due to polymyxin resistance.

RESULTS

Genomics analysis and lipid A structural characterization.

Genomes of the wild-type P. aeruginosa strain K ([PAK] polymyxin B MIC, 1 mg/liter) and its paired mutants PAKpmrB6 (polymyxin B MIC, 16 mg/liter) and PAKpmrB12 (polymyxin B MIC, 64 mg/liter) (23) were sequenced and assembled. Compared to that in PAK, pmrB mutations L243Q (locus PRK_00328) and A248V (locus PRK_00328) were observed in PAKpmrB6 and PAKpmrB12, respectively (23). Moreover, a missense mutation (P415L) in locus PRK_00106, encoding transcriptional regulator RtcR, and a premature termination (Y275stop) in speE (locus PRK_00325 in the pmrAB operon, encoding spermidine synthase) were identified in PAKpmrB6, whereas no additional mutations were observed in PAKpmrB12 compared to the wild-type PAK.

Mutations in pmrB can cause polymyxin resistance via the 4-amino-4-deoxy-l-arabinose (l-Ara4N) modification of lipid A through the activation of arn operon (17, 24). We therefore characterized lipid A profiles of each strain using electrospray ionization-mass spectrometry (ESI-MS). Penta-acylated (peaks at m/z 1445.85 and 1461.85) and hexa-acylated (peaks at m/z 1599.99, 1615.99, and 1631.98) forms of lipid A were predominant in the wild-type PAK, with a Δm/z of 170 attributable to the enzymatic removal of R-3-hydroxydecanoate at the 3 position of lipid A (Fig. 1A; see also Fig. S1 in the supplemental material). In comparison, the polymyxin-resistant PAKpmrB6 strain predominantly produced lipid A species modified with one (m/z 1576.92, 1748.05, and 1764.05) or two (m/z 1708.98, 1863.11, 1879.11, and 1895.10) l-Ara4N moieties (Δm/z = 131.05 or 262.10, respectively) (Fig. 1B; Fig. S1). The PAKpmrB12 strain displayed a very similar lipid A profile to that of PAKpmrB6 (Fig. 1C). Dephosphorylated lipid A species were detected in PAK (peaks at m/z 1365.89, 1381.89, 1520.02, 1536.02, and 1552.01) (Fig. 1A) and both mutants (peaks at m/z 1480.96, 1496.95, 1512.94, and 1667.08) (Δm/z = −79.97) (Fig. 1B and C).

FIG 1.

Mass characterization of lipid A profiles in polymyxin-susceptible and -resistant P. aeruginosa strains. Wild-type PAK (A), PAKpmrB6 (B), and PAKpmrB12 (C) mutants and the predominant structures of lipid A from the wild-type PAK and/or both pmrB mutants (D). The lipid A samples were analyzed by LC-MS/MS in negative ion mode, which obtained most lipid A peaks with the loss of one hydrogen ([M-H]⁻). R-3-hydroxydecanoate is shown in blue and l-Ara4N is shown in red.

Evaluation of the effects of extraction solvents on the recovery of bacterial lipidomes.

Both Bligh-Dyer (25) and methyl-tert-butyl ether (MTBE)/methanol/water (MMW) (26) extraction methods were evaluated on the basis of the recovery of lipidomes from the wild-type PAK. To avoid the loss of certain hydrophilic lipids (e.g., phospholipids), the single-phase Bligh-Dyer method was also performed in tandem. The performance of extraction solvents was assessed on the basis of the number and intensity of the peaks detected by reversed-phase liquid chromatography (RPLC) (27). Principal-component analysis (PCA) score plots were constructed to depict the impact of the solvent system used for extraction on the resultant lipidome, showing general differences among the three solvent systems (Fig. 2A) (28). As shown in the heatmap, the double-phase Bligh-Dyer method (chloroform [CHCl₃]/methanol [MeOH]/water [H₂O] [CMW], 1:1:0.9 [vol/vol]) yielded a greater abundance of fatty acids but without the detection of a number of lipophilic glycerophospholipids (GPLs) compared to those with single-phase CMW (CHCl₃/MeOH/H₂O, 1:2:0.8 [vol/vol]) and MMW (MTBE/MeOH/H₂O, 10:3:2.5 [vol/vol]) methods (Fig. 2B). A minimal overall difference was apparent between the single-phase CMW and MMW methods in general; however, several specific phospholipids, putatively identified as phosphatidylserines [PS(26:0) and PS(31:0)] and phosphatidylinositols [PI(29:0) and PI(31:1)], were extracted only by the MMW method (Fig. 2B). However, the MMW extraction method displayed higher data variations in the relative abundances (log scale) from each sample than the single-phase CMW method (Fig. 2C). Therefore, the single-phase CMW method was employed in the lipidomics experiment in the present study.

FIG 2.

Method comparison in the extraction of lipidomes from P. aeruginosa strain K (PAK). (A) PCA score plot shows that metabolites are grouped due to different extraction methods: red, CHCl₃/MeOH/H₂O ([CMW] 1:1:0.9 [vol/vol], double-phase Bligh-Dyer); green, CMW (1:2:0.8 [vol/vol], single-phase Bligh-Dyer); and blue, MTBE/MeOH/H₂O ([MMW] 10:3:2.5 [vol/vol], MTBE extraction). The colored dots on the PCA score plot represent five biological replicates. (B) The heat map illustrates the relative peak intensities of lipids with three major lipid classes extracted by three different solvents, CMW (left, 1:2:0.8 [vol/vol]), CMW (middle, 1:1:0.9 [vol/vol]), and MMW (right, 10:3:2.5 [vol/vol]). Five independent biological replicates of each extraction method are shown. Colors indicate relative abundance of lipids in each sample based on the relative peak intensity for each lipid: red, high; yellow, mean; blue, low or undetectable. (C) The bar charts show the relative abundance (log scale) of each sample between different extraction methods as well as the quality control (QC) samples. Box plots show the upper and lower quartiles (tops and bottoms of boxes), medians (lines within the boxes), and the spread of data that are not outliers (whiskers).

Perturbations of lipids and the related metabolites in the pmrB mutants.

A lipidomics analysis of the wild-type PAK and the polymyxin-resistant PAKpmrB6 and PAKpmrB12 strains yielded 221 putatively identified lipids with major classes in fatty acids and glycerophospholipids (see Data Set S1). Our results revealed profound alterations (fold change of >2, P < 0.05, false discovery rate [FDR] < 0.05, one-way analysis of variance [ANOVA]) of these lipids, especially GPLs (see Data Set S2A). A high-level identification of GPLs based on molecular formula revealed significant variations in the major species, phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylglycerol (PG), and phosphatidylserines (PS). In general, the relative abundances of double-chain GPLs were decreased in both pmrB mutants compared to that in the wild-type PAK (Fig. 3A). The comparison between the two mutant strains indicated a further depletion of GPLs in PAKpmrB12 relative to PAKpmrB6. In addition, the levels of fatty acids and acyl-coenzyme A (acyl-CoA) were significantly decreased in both pmrB mutants compared to that in the wild-type PAK (Fig. 3B).

FIG 3.

Lipidomic perturbations between pmrB mutants and the wild-type PAK. (A) Fold changes of GPLs in PAKpmrB6 (blue) and PAKpmrB12 (red) compared to PAK collected from the reversed-phase liquid chromatography (RPLC) method (P < 0.05, FDR < 0.05, one-way ANOVA). (B) Fold changes indicate the decreased levels of fatty acids and acyl-CoA in PAKpmrB6 (blue) and PAKpmrB12 (red) compared to PAK analyzed by the hydrophilic interaction chromatography (HILIC) method (P < 0.05, FDR < 0.05, one-way ANOVA).

Metabolomics analysis of polymyxin-susceptible and -resistant P. aeruginosa strains.

The metabolomics analysis performed for this study putatively identified 578 metabolites (see Data Set S3). The pooled quality control (QC) samples that indicate analytical reproducibility clustered tightly in the PCA score plots, reflecting minimal analytical variations (see Fig. S2A). Furthermore, the median relative standard deviation (RSD) values of the metabolites in the QC and bacterial samples were 15.7% and 25.8 to 32.3%, respectively (Fig. S2B), within the acceptable limits for metabolomics analysis (29).

Global metabolic differences between the pmrB mutants and the wild-type PAK were analyzed by multivariate and univariate statistics (Fig. 4). The PCA score plots demonstrated that the metabolomes of the wild-type PAK and its pmrB mutants differed significantly in the first principal component, which accounts for 19.6% of the variance in the data set; however, the distance between PAKpmrB6 and PAKpmrB12 is much less dramatic (Fig. 4A). The volcano plots revealed that, compared to the wild-type PAK, 55 and 71 metabolites were dramatically perturbed in PAKpmrB6 and PAKpmrB12, respectively (>1 log₂-fold, P < 0.05, FDR < 0.05, Student's t test), and more than two-thirds of the metabolites decreased significantly. It is notable that most of the significantly perturbed metabolites detected in PAKpmrB6 were also observed in PAKpmrB12. Interestingly, the results highlighted that metabolites related to amino sugars as well as methionine metabolism were significantly enriched in both PAKpmrB6 and PAKpmrB12 (Fig. 4B and Table 1). In addition, significantly depleted nucleotide levels were observed in both mutants compared to that in the wild-type strain, and this depletion was greater in PAKpmrB12 than in PAKpmrB6 (Table 1).

FIG 4.

Multivariate and univariate statistical analyses of metabolic perturbations between wild-type PAK and the pmrB mutants, PAKpmrB6 and PAKpmrB12. (A) PCA score plots of the two principle components for metabolite levels from the three strains: red, PAK; blue, PAKpmrB6; green, PAKpmrB12. (B) Volcano plots show the fold change and significance of metabolites in both pmrB mutants compared to the wild-type strain. Red plots represent metabolites having a fold change of >2 and a P value of <0.05 (Student's t test), while gray plots represent metabolites that are not significantly changed. Fold changes relative to the untreated control are based upon mean values from five biological replicates in all three strains.

TABLE 1.

Fold changes (relative intensities) in the abundance of amino acid and nucleotide metabolites in the polymyxin-resistant PAKpmrB6 and PAKpmrB12 relative to the wild-type PAK

Pathway	Metabolitea	Formula	Mass (Da)	RTb (min)	Confidencec	Log2 fold changed
						PAKpmrB6	PAKpmrB12
Amino acid
Amino sugar metabolism	N-Acetyl-d-glucosamine 6-phosphate	C8H16NO9P	301.056	14.26	6	−1.06	−1.77
Histidine metabolism	Hercynine	C9H15N3O2	197.116	11.98	6	1.19	0.79
Lysine biosynthesis	l-Lysine	C6H14N2O2	146.105	18.73	8	0.53	1.54
	l-Carnitine	C7H15NO3	161.105	11.73	8	0.58	1.43
	l-Pipecolate	C6H11NO2	129.079	18.96	8	0.59	1.30
Methionine metabolism	S-Methyl-5-thio-d-ribose	C6H12O4S	180.046	7.41	6	10.92	10.71
	S-Methyl-5-thio-d-ribose 1-phosphate	C6H13O7PS	260.012	11.57	8	6.54	6.11
	S-Adenosyl-l-methionine	C15H22N6O5S	398.138	15.84	8	−6.63	−5.81
	S-Methyl-5′-thioadenosine	C11H15N5O3S	297.090	7.42	6	−2.56	0.05
	1,2-Dihydroxy-3-keto-5-methyl-thiopentene	C6H10O3S	162.035	7.68	6	9.28	7.09
	l-Methionine	C5H11NO2S	149.051	11.17	8	−0.56	−1.16
	S-Adenosyl-l-homocysteine	C14H20N6O5S	384.122	13.13	6	−0.93	−1.34
Carbohydrate
Glycolysis	2-(α-Hydroxyethyl)thiamine diphosphate	C14H22N4O8P2S	468.064	13.58	6	−0.53	−1.21
l-Ara4N biosynthesis	UDP-l-Ara4N	C14H23N3O15P2	535.060	16.15	8	7.18	6.24
	UDP-l-Ara4FN	C15H23N3O16P2	563.055	15.32	8	13.78	13.41
	UDP-glucuronate	C15H22N2O18P2	580.035	18.41	8	5.56	5.91
Pyruvate metabolism	S-Lactoylglutathione	C13H21N3O8S	379.105	15.63	8	−0.57	3.71
Nucleotide
Purine metabolism	GDP	C10H15N5O11P2	443.025	17.30	8	−0.88	−1.17
	Adenosine	C10H13N5O4	267.097	8.87	8	−0.51	−1.21
	5-Phosphoribosylamine	C5H12NO7P	229.035	14.21	8	−2.52	0.19
	Adenine	C5H5N5	135.054	9.28	8	−1.05	−1.06
	AMP	C10H14N5O7P	347.063	13.19	8	−0.46	−1.12
Pyrimidine metabolism	CDP	C9H15N3O11P2	403.019	16.46	8	−0.99	−1.03
	CMP	C9H14N3O8P	323.052	15.09	8	−0.67	−1.34
	dTMP	C10H15N2O8P	322.057	12.44	8	−0.37	−1.61
	UDP	C9H14N2O12P2	404.002	15.29	6	4.56	4.35

^aPutative metabolite, identified by the exact mass and retention time, with at least 2-fold differences at an FDR of <0.05 and a P value of <0.01 between pmrB mutants and the wild-type PAK. The metabolic data were collected based on five biological replicates.

^bRT, retention time.

^cConfidence, confidence level that gives an indication of the confidence of identification. Confidence level at 8 indicates that the metabolites were putatively identified in the preferred database and related peaks with the variation of standard retention time of <5%, while metabolites with confidence level at 6 were putatively identified based on preferred database and related peaks with calculated retention time within 50%.

^dFold changes in bold indicate those metabolites with significant change (fold change ≥ 1 log₂, P < 0.05) compared to the wild-type PAK.

Perturbations of carbohydrate metabolism associated with l-Ara4N modification of lipid A.

A pathway analysis of the metabolomics data revealed that the relative abundances of the intermediates in the biosynthesis of l-Ara4N were significantly increased in both pmrB mutants compared to that in the wild-type PAK (Fig. 5). In particular, the intracellular levels of the key l-Ara4N precursor UDP-glucuronate were markedly increased (>5 log₂-fold) in two pmrB mutants compared to that in PAK. Notably, three key intermediates, UDP-l-Ara4N, UDP-4-deoxy-4-formamido-l-arabinose (UDP-l-Ara4FN), and the substrate for ArnT, undecaprenyl phosphate-l-Ara4N, were detected only in the two polymyxin-resistant strains. In addition, as a by-product in the generation of undecaprenyl phosphate-l-Ara4N, UDP was observed to be significantly enriched (>4 log₂-fold) in both PAKpmrB6 and PAKpmrB12 compared to that in the wild-type strain (Table 1).

FIG 5.

Perturbations of the intermediates in l-Ara4N synthesis pathway. UDP-glucuronate, UDP-l-Ara4N, and UDP-l-Ara4FN were detected using the HILIC method, whereas undecaprenyl phosphate-l-Ara4N was detected using the RPLC method. The green box indicates the metabolite that was significantly increased by >2-fold, while the red boxes show metabolites that were exclusively detected in the pmrB mutants compared to that in the wild-type PAK. Metabolites in gray boxes were not detected through either the HILIC or RPLC method. **, P < 0.01; ***, P < 0.001 (Student's t test).

Metabolic perturbations of methionine salvage cycle and spermidine synthesis metabolism.

The intracellular levels of metabolites involved in amino sugar, histidine, lysine, methionine, and tryptophan metabolic pathways were significantly perturbed (>1 log₂-fold) in the pmrB mutants compared to that in the wild-type PAK (Table 1). Remarkably, significant changes were observed in the methionine salvage cycles of both mutant strains. Specifically, the relative abundances of three metabolites, S-methyl-5-thio-d-ribose, S-methyl-5-thio-d-ribose-1-phosphate, and 1,2-dihydroxy-3-keto-5-methyl-thiopentene, were significantly higher (>7 log₂-fold), while two metabolites, the universal methyl group donor S-adenosyl-l-methionine and its methyltransferase reaction product S-adenosyl-l-homocysteine, were in lower abundance (<−1 log₂-fold) compared to that in the wild-type PAK (Fig. 6 and Table 1). Intriguingly, two key metabolites in the synthesis of spermidine, S-adenosylmethioninamine and putrescine, were significantly increased (4.9 log₂-fold and 2.7 log₂-fold, respectively). Concomitantly, the intracellular levels of spermidine and its by-product S-methyl-5-thioadenosine were dramatically decreased (2.6 log₂-fold and 3.5 log₂-fold, respectively) in PAKpmrB6 compared to those in both PAK and PAKpmrB12 (Fig. 6 and Table 1).

FIG 6.

Metabolic perturbations in methionine salvage cycle and spermidine synthesis pathway. S-Adenosylmethioninamine, putrescine, and spermidine were analyzed using the RPLC method, while the other metabolites were analyzed by the HILIC method. Metabolites in red boxes indicate increased abundance in PAKpmrB6 and/or PAKpmrB12, while blue boxes indicate decreased abundance compared to that in the wild-type PAK. Metabolites in gray boxes were not detected through either the HILIC or RPLC method. *, P < 0.05; **, P < 0.01; ***, P < 0.001 (Student's t test). Dashed arrow shows the conversion of putrescine to spermidine catalyzed by SpeE in the presence of the cofactor S-adenosylmethioninamine.

DISCUSSION

In view of the challenges associated with the ever-worsening spread of resistance and the development of antibiotics with novel modes of action, an understanding of the mechanisms of resistance to last-resort antibiotics such as polymyxins is paramount (30). Global metabolomics and lipidomics are of considerable utility for elucidating the mechanism of antimicrobial killing and resistance. To date, metabolomics has identified key metabolic differences between antibiotic-susceptible and -resistant strains of Acinetobacter baumannii, Mycobacterium tuberculosis, and Edwardsiella tarda (31–33). In the present study, we successfully employed metabolomics and lipidomics to elucidate key metabolic perturbations between paired polymyxin-susceptible and -resistant P. aeruginosa strains.

Mutations in the pmrB locus of P. aeruginosa result in constitutive phosphorylation of PmrA, which in turn activates the expression of the arnBCADTEF-pmrE operon and eventuates in the covalent modification of the 1 and 4′ phosphates of lipid A with l-Ara4N (20, 23). Concordantly, the lipid A profiles from both pmrB mutants PAKpmrB6 and PAKpmrB12 indicated the addition of one or two l-Ara4N moieties to the predominant hexa- and penta-acylated lipid A species (Fig. 1), which is consistent with the reported l-Ara4N-modified hexa-acylated lipid A in the pmrB mutants (23). Our metabolomics study provides an in-depth understanding of polymyxin resistance by systematically elucidating the changes in key intermediates in the l-Ara4N biosynthetic pathway (Fig. 5). Striking increases were observed in the levels of UDP-glucuronate, UDP-l-Ara4N, UDP-l-Ara4FN, and undecaprenyl phosphate-l-Ara4N in the synthesis of l-Ara4N in both pmrB mutants, which are attributed to the activated arn operon due to the mutations L243Q and A248V in pmrB. In addition, the conversion of UDP-l-Ara4FN to undecaprenyl phosphate-l-Ara4N involves the generation of UDP, which showed dramatically enriched abundances in both mutants compared to that in the wild-type strain.

The asymmetric Gram-negative OM bilayer consists of an outer leaflet predominantly composed of LPS and an inner leaflet predominantly comprising GPLs (34). The OM acts as an efficient permeability barrier against a number of antimicrobial agents. Apart from the aforementioned lipid A modifications that are associated with polymyxin resistance, we hypothesized that changes in GPLs may affect the interaction between polymyxins and the Gram-negative OM and therefore play important roles in polymyxin resistance. To investigate such perturbations, it is important to have an effective extraction procedure that reproducibly yields a broad range of total cellular lipids. Folch and Bligh-Dyer methods utilizing single- and double-phase chloroform/methanol/water (CMW) solvents are most commonly applied in lipid extraction (25, 35). The methyl-tert-butyl ether (MTBE)/methanol/water (MMW) extraction method enables a cleaner recovery of lipids (26). Our results showed that a large number of fatty acids were extracted using the double-phase CMW method, but several hydrophilic GPLs were not detected (Fig. 2B). The MMW method yielded more GPLs but was associated with higher statistical variations in the extraction than in both CMW methods, possibly due to the predrying and reconstitution steps. Overall, the single-phase CMW (1:2:0.8 [vol/vol]) method provided a high recovery and reproducibility for the extraction of total intracellular lipids and is ideal as a one-step method for untargeted lipidomics in P. aeruginosa.

Compared to that of the polymyxin-susceptible PAK, the polymyxin-resistant PAKpmrB6 and PAKpmrB12 strains produced lower levels of total cellular GPL species, including PC, PE, PG, and PS (Fig. 4A). A similar phenomenon was reported in Brucella melitensis, in which polymyxin resistance is associated with decreased PE content in the cell envelope (36). In contrast, LPS-deficient polymyxin-resistant isolate A. baumannii 19606R produces higher levels of GPLs than its paired polymyxin-susceptible wild-type strain, A. baumannii ATCC 19606 (33). Although the precise mechanism of polymyxin resistance related to the changes in GPL levels remains unclear, these observations indicate potential species-specific OM remodelling events in response to polymyxin treatment. For example, P. aeruginosa and several other Gram-negative pathogens (e.g., Salmonella enterica serovar Typhimurium) employ l-Ara4N modification of lipid A to develop resistance to polymyxins (20, 37), while Brucella is capable of either reducing divalent cations or lowering the acyl chain fluidity in its LPS to increase polymyxin resistance (38–40). In A. baumannii, a complete loss of LPS in polymyxin-resistant isolates may in turn account for the increased production of GPLs as a compensatory mechanism to fortify the OM (33, 41). Our results also showed decreased levels of fatty acids and acyl-CoA in both polymyxin-resistant mutants (Fig. 4B). It is known that fatty acids and acyl-CoA are crucial in the synthesis of phosphatidic acid (PA), which serves as a key intermediate in the synthesis of all membrane phospholipids (42). Notably, the depleted abundance of total intracellular lipids in both pmrB mutants suggests a key role in polymyxin resistance and warrants further investigation. Interestingly, PAKpmrB12 is highly resistant to polymyxins (polymyxin B MIC, 64 mg/liter) and expressed lower levels of GPLs than PAKpmrB6 (polymyxin B MIC, 16 mg/liter). Our findings support the notion that the decreased GPL level also plays a role in l-Ara4N modification-mediated polymyxin resistance.

Apart from dramatic lipidomic changes, our metabolomics results showed remarkable perturbations in the methionine salvage cycle, also called the S-methyl-5′-thioadenosine (MTA) cycle (Fig. 6) (43). This pathway is regulated in response to sulfur availability and also takes part in the regulation of polyamine synthesis (43–45). MTA and spermidine are synthesized from S-adenosylmethionine (SAM) and regulated by speE (43, 46). Our genome sequencing data revealed a nonsense mutation in speE (locus PRK_00325) in PAKpmrB6; this mutation very likely caused the dramatically depleted MTA level and enriched putrescine level and consequently led to a decreased intracellular spermidine concentration in this strain compared to those in both PAK and PAKpmrB12 (Fig. 6). Notably, the upregulation of speE homolog PA4774 mediated by PmrAB in P. aeruginosa PAO1 was suggested to contribute to reduced outer membrane permeability and increased antibacterial peptide resistance (46, 47). Therefore, the truncation of speE in PAKpmrB6 but not in PAKpmrB12 explains the difference in the levels of polymyxin resistance between the two polymyxin-resistant strains. In addition, our results showed significantly increased levels of S-adenosylmethioninamine, S-methyl-5-thio-d-ribose, S-methyl-5-thio-d-ribose-1-phosphate, and 1,2-dihydroxy-3-keto-5-methyl-thiopentene in both pmrB mutants, indicating the activation of the methionine salvage cycle. Altogether, the dramatic perturbation in the methionine salvage cycle is possibly associated with polymyxin resistance by altering the level of spermidine and therefore protecting bacteria from polymyxin killing.

In conclusion, our study is the first to use comparative metabolomics and lipidomics to elucidate significant global metabolic alterations in P. aeruginosa associated with polymyxin resistance. In addition to lipid A modifications with l-Ara4N, the pmrB mutants displayed significant metabolic perturbations in the levels of total intracellular lipids, especially GPLs, the methionine salvage cycle, and spermidine synthesis. This study provides novel insights into the multifaceted mechanisms of polymyxin resistance which may benefit the development of new-generation polymyxins.

MATERIALS AND METHODS

Chemicals and reagents.

Polymyxin B (sulfate), lipid A (L5399-5 mg), and super-DHB (50862-1g) were purchased from Sigma-Aldrich (Sydney, NSW, Australia). All other reagents and solvents were of the highest purity commercially available. The stock solution of polymyxin B was prepared in Milli-Q water (Millipore, North Ryde, NSW, Australia) and filtered through 0.22-μm syringe filters (Sartorius, Melbourne, VIC, Australia).

Bacterial strains and culture.

P. aeruginosa strain K (PAK), PAKpmrB6, and PAKpmrB12 were obtained from the Moskowitz laboratory (Massachusetts General Hospital, MA, USA) and were generated spontaneously from a late-exponential-phase culture containing polymyxin B at 20 to 50 mg/liter (23). All bacterial strains were stored at −80°C in tryptone soya broth. Prior to experiments, PAK was subcultured on Mueller-Hinton agar plates, and the mutants were subcultured on plates containing 4 mg/liter polymyxin B (Medium Preparation Unit, University of Melbourne, VIC, Australia). Overnight cultures were subsequently grown in 5 ml of cation-adjusted Mueller-Hinton broth ([CaMHB] Oxoid), from which a 1 in 100 dilution was performed in fresh medium to prepare 100 ml of mid-log-phase cultures according to an optical density at 600 nm (OD₆₀₀) of 0.5. All liquid cultures were incubated at 37°C in a shaking incubator (180 rpm).

Determination of MICs.

Polymyxin B MICs of all strains were determined with the broth microdilution method (36). Briefly, experiments were performed in 96-well polypropylene microtiter plates (42) with concentrations of polymyxin B (0 to 128 mg/liter) in CaMHB. MICs were defined as the lowest concentration that inhibited visible bacterial growth after a 20-h incubation at 37°C.

DNA sequencing and genomic analysis.

A DNeasy blood and tissue kit (Qiagen) was employed to extract genomic DNA from 2 ml log-phase bacterial culture. Electrophoresis and a NanoDrop 1000 (Thermo Fisher Scientific) were used to assess the quality and quantity of genomic DNA samples. Paired-end 250-bp DNA sequencing of PAKpmrB6 and PAKpmrB12 were performed on an Illumina HiSeq 2500 platform (MicrobesNG, Birmingham, UK) with 30× coverage; paired-end 150-bp DNA sequencing of PAK was performed on a HiSeq X platform (Novogene, Hong Kong) with 35× coverage. Quality-trimmed reads of PAK were then used for de novo genome assembly with SPAdes (48), and Prokka was employed for genome annotation. The reads of PAKpmrB6 and PAKpmrB12 were aligned to the assembled PAK genome (GenBank assembly accession GCA_000408865.1) using SubRead (49). The genetic variations were determined using nesoni and annotated using snpEff (50, 51).

Isolation and structural characterization of lipid A.

Lipid A was isolated by mild acid hydrolysis as previously described (52). Briefly, cells were harvested from 200 ml of culture (OD₆₀₀ of 0.5) via centrifugation at 3,220 × g for 20 min and washed twice with 5 ml phosphate-buffered saline (PBS). The cells were resuspended in 4 ml PBS, and then 5 ml chloroform and 10 ml methanol were added to the suspension, making a single-phase Bligh-Dyer mixture (chloroform/methanol/water, 1:2:0.8 [vol/vol]). The mixture was centrifuged at 3,220 × g for 15 min to remove the supernatant. 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 incubated in a boiling water bath for 45 min. To extract the lipids after hydrolysis, the SDS solution was converted into a two-phase Bligh-Dyer mixture by adding 12 ml chloroform and 12 ml methanol (25). The lower phase containing lipid A was then collected, and the samples were dried and stored at −20°C. A structural analysis of lipid A was performed using mass spectrometry in the negative mode on a Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer and a Shimadzu MALDI-7090 matrix-assisted laser desorption ionization–tandem time of flight (MALDI-TOF/TOF) mass spectrometer. The matrix used for lipid A analysis was super-DHB (25 mg/ml in CHCl₃/MeOH, 1:1 [vol/vol]).

Sample preparation for lipidomics and metabolomics experiments.

Intracellular metabolites were extracted as described previously (53). The general procedure for sample preparation was as follows. Ten milliliters of mid-log-phase culture (OD₆₀₀ of 0.5) was centrifuged at 3,220 × g for 10 min at 4°C, and the supernatant was kept for footprint analysis. After washing twice with 2 ml of 0.9% NaCl, the cell pellets were resuspended in 0.5 ml of extraction solvents. Liquid nitrogen was used to permeabilize the cells and release the intracellular metabolites. The mixtures were centrifuged at 3,220 × g for 10 min at 4°C, and 300 μl of the supernatant containing the extracted metabolites was collected in 1.5-ml centrifuge tubes and stored at −80°C. The samples were further centrifuged at 14,000 × g for 10 min at 4°C, and 200 μl of particle-free supernatant was transferred into an injection vial for liquid chromatography-mass spectrometry (LC-MS) analysis.

To evaluate the effect of different extraction solvents on the recovery of lipids from whole bacterial cells, three classic extraction methods were examined. (i) For the single-phase Bligh-Dyer method (CHCl₃/MeOH/H₂O, 1:2:0.8 [vol/vol]), the cell pellets were resuspended in 0.5 ml solvent and 300 μl of supernatant was collected using the aforementioned method. (ii) For the double-phase Bligh-Dyer method (CHCl₃/MeOH/H₂O, 1:1:0.9 [vol/vol]) (25), the cell pellets were resuspended in 380 μl of the solvent from method “i” and 100 μl CHCl₃ and 100 μl H₂O were added after vortexing and freeze-thawing to form a double-phase Bligh-dyer mixture. The lower phase (100 μl) was finally collected after centrifugation. For LC-MS analysis, 200 μl MeOH was added to the CHCl₃ extracts, which were centrifuged at 14,000 × g for 10 min at 4°C, and 200 μl of the particle-free supernatant was transferred into an LC-MS glass vial. (iii) For the method using methyl-tert-butyl ether (MTBE)/MeOH/H₂O (10:3:2.5 [vol/vol]) (26), the cell pellets were resuspended in 400 μl MTBE and 120 μl MeOH, followed by vortexing and freeze-thawing to release metabolites from the bacterial cells. Milli-Q water (100 μl) was added to the suspension to separate the phases; 300 μl of the upper phase was collected after centrifugation and dried under a nitrogen stream. For further analysis, the samples were reconstituted in 300 μl CHCl₃/MeOH (1:2 [vol/vol]) and centrifuged at 14,000 × g for 10 min at 4°C to obtain the particle-free supernatants.

The procedures for metabolic sample preparation were similar to those for lipidomic samples extracted by the single-phase Bligh-Dyer method. The only difference was the three-solvent system, CHCl₃/MeOH/H₂O (1:3:1 [vol/vol]), for extracting intracellular metabolites. For the preparation of footprint samples, an aliquot of approximately 1.5 ml of supernatant from the culture after centrifugation was rapidly filtered through a 0.22-μm filter, and the extraction solvent (250 μl) was then added. The supernatant (200 μl) was collected for LC-MS analysis after centrifugation (14,000 × g for 10 min at 4°C).

LC-MS analysis for lipids and polyamine analyses.

The lipidomics analysis was conducted on a Dionex U3000 high-performance liquid chromatography (HPLC) system in tandem with a Q Exactive Orbitrap mass spectrometer (Thermo Fisher) in both positive and negative modes with a resolution at 35,000. The mass scanning range was from 167 to 2,000 m/z. The electrospray voltage was set at 3.50 kV, and nitrogen was used as collision gas. The Ascentis Express C₈ column (5 cm by 2.1 mm, 2.7 μm, 53831-U; Sigma-Aldrich) was maintained at 40°C, and the samples were controlled at 4°C. The flow rate was 0.2 ml/min for the first 24 min and then increased to 0.5 ml/min from 25 to 30 min. For lipidomics studies, the linear gradient started from 100% mobile phase A (40% isopropanol and 60% Milli-Q water with 8 mM ammonium formate and 2 mM formic acid) to a final composition of 35% mobile phase A and 65% mobile phase B (98% isopropanol and 2% Milli-Q water with 8 mM ammonium formate and 2 mM formic acid). For polyamine (mass scanning range 67 to 900 m/z) analysis, 100% mobile phase A (0.1% formic acid in Milli-Q water) was performed in the first 3 min followed by a linear gradient to 90% mobile phase B (0.1% formic acid in acetonitrile) within 5 min at a flow rate of 0.2 ml/min.

LC-MS analysis for metabolomics.

The hydrophilic interaction liquid chromatography (HILIC)-high resolution mass spectrometry (HRMS) employed in this experiment was described previously (33, 53). Samples maintained at 4°C were eluted by mobile phase A (20 mM ammonium carbonate) and B (acetonitrile) through a ZIC-pHILIC column (5 μm, polymeric, 150 mm by 4.6 mm; SeQuant, Merck). The gradient started with 80% mobile phase B at a flow rate of 0.3 ml/min, followed by a linear gradient to 50% mobile phase B over 15 min. All samples (five biological replicates each) were analyzed in a single LC-MS batch. The assessment of pooled quality control (QC) samples, internal standards, and total ion chromatograms was used for monitoring the chromatographic peaks, signal reproducibility, and the stability of analytes. Mixtures of pure metabolic standards containing over 250 metabolites were analyzed within the batch for assisting the identification of metabolites.

Data processing.

Untargeted lipidomic and metabolic analyses were performed through mzMatch (54) and IDEOM (55) (http://mzmatch.sourceforge.net/ideom.php). Raw LC-MS data files were converted to the mzXML format through a proteowizard tool, Msconvert. Automated chromatography peaks were picked by XCMS (56) and then converted to peakML files, which were combined and filtered by mzMatch on the basis of the intensity (1,000), reproducibility (RSD for all replicates, <0.8), and peak shape (codadw > 0.8). The mzMatch program was used for retrieving intensities for missing peaks and the annotation of related peaks. Unmatched peaks and noises were rejected through IDEOM. Putative metabolites were identified on the basis of the neutral exact mass within 3 ppm and confirmed by the retention time of authentic standards (<5%) and the calculated retention time (<50%). The databases used in IDEOM included KEGG, MetaCyc, and Lipidmaps, with preference for bacterial metabolites noted on PseudoCyc (57). Each metabolite was quantified by the intensity of its peak and compared between the wild-type and mutant strains. Univariate statistical analysis was performed using both one-way analysis of variance (ANOVA) and Student's t test (P < 0.05), while multivariate analysis was conducted using the metabolomics R package and MetaboAnalyst 4.0 (http://www.metaboanalyst.ca/faces/ModuleView.xhtml). Metabolic pathways were analyzed based upon the KEGG pathway (58), BioCyc (59), and Visualization and Analysis of Networks containing Experimental Data (VANTED) software (60).