Laser desorption rapid evaporative ionization mass spectrometry (LD-REIMS) demonstrates a direct impact of hypochlorous acid stress on PQS-mediated quorum sensing in Pseudomonas aeruginosa

ABSTRACT To establish infections in human hosts, Pseudomonas aeruginosa must overcome innate immune-generated oxidative stress, such as the hypochlorous acid (HOCl) produced by neutrophils. We set out to find specific biomarkers of oxidative stress through the development of a protocol for the metabolic profiling of P. aeruginosa cultures grown in the presence of different oxidants using a novel ionization technique for mass spectrometry, laser desorption rapid evaporative ionization mass spectrometry (LD-REIMS). We demonstrated the ability of LD-REIMS to classify samples as untreated or treated with a specific oxidant with 100% accuracy and identified a panel of 54 metabolites with significantly altered concentrations after exposure to one or more of the oxidants. Key metabolic changes were conserved in P. aeruginosa clinical strains isolated from patients with cystic fibrosis lung infections. These data demonstrated that HOCl stress impacted the Pseudomonas quinolone signal (PQS) quorum sensing system. Ten 2-alkyl-4-quinolones (AHQs) associated with the PQS system were significantly lower in concentration in HOCl-stressed P. aeruginosa cultures, including 2-heptyl-3-hydroxy-4(1H)-quinolone (PQS), the most active signal molecule of the PQS system. The PQS system regulates the production of virulence factors, including pyocyanin and elastase, and their levels were markedly affected by HOCl stress. No pyocyanin was detectable and elastase concentrations were reduced by more than 75% in cultures grown with sub-lethal concentrations of HOCl, suggesting that this neutrophil-derived oxidant may disrupt the ability of P. aeruginosa to establish infections through interference with production of PQS-associated virulence factors. IMPORTANCE This work demonstrates that a high-throughput ambient ionization mass spectrometry method can be used successfully to study a bacterial stress response. Its application to the opportunistic pathogen Pseudomonas aeruginosa led to the identification of specific oxidative stress biomarkers, and demonstrated that hypochlorous acid, an oxidant specifically produced by human neutrophils during infection, affects quorum sensing and reduces production of the virulence factors pyocyanin and elastase. No pyocyanin was detectable and elastase levels were reduced by more than 75% in bacteria grown in the presence of hypochlorous acid. This approach has the potential to be widely applicable to the characterization of the stress responses of bacteria.

B acteria encounter oxidative stress in aerobic environments.The innate immune systems of both animals and plants use reactive oxygen species to attack and to attempt to kill invading microbes.Upon phagocytosis of bacteria, the phagocyte or NADPH oxidase of human neutrophils is induced, assembled, and catalyzes the formation of the superoxide radical (O 2  .-).This is known as the respiratory burst and is associated with a 10-to 20-fold increase in O 2 consumption (1).The importance of superoxide generated by NADPH oxidase is demonstrated through the susceptibility of people with defective NADPH oxidase to infection by a broad range of microbes, as seen in conditions such as chronic granulomatous disease (CDG) (2).However, superoxide has low reactivity and is membrane impermeable, and so the antimicrobial action of the respiratory burst is thought to be mediated through the conversion of superoxide to stronger oxidants, including H 2 O 2 , hypochlorous acid (HOCl) and hypothiocyanous acid (HOSCN) (3).O 2 .
-dismutates either spontaneously or catalyzed by superoxide dismutase to form H 2 O 2 .HOCl and HOSCN can then be formed by the reaction of H 2 O 2 with chloride or thiocyanate ions respectively, catalyzed by myeloperoxidase (MPO).While there exists a large body of evidence to indicate that H 2 O 2 is important for the killing of pathogens by neutrophils (4) (5) the antimicrobial actions of HOCl and HOSCN are relatively poorly studied, despite MPO comprising around 5% of neutrophil proteins (1) and that neutrophil-mediated killing of bacteria has been shown to be MPO-dependent (6).
To further understand their bactericidal action and the corresponding bacterial protective response, we previously performed a genetic and transcriptomic analysis of the response of Pseudomonas aeruginosa to HOCl and HOSCN (7).We identified a diverse range of genes required for protection against HOCl and HOSCN, including the transcriptional regulator RclR, a homolog of the Escherichia coli HOCl-specific sensor RclR.We demonstrated that RclR regulates the expression of rclX, a putative peroxiredoxin, in P. aeruginosa and that both RclR and RclX are required specifically for protection against HOCl and HOSCN, with RclX shown to be a HOCl reductase (8).Another key study showed that hypohalous acids, such as HOCl, caused protein aggregation and that P. aeruginosa responded by increasing polyphosphate levels, which improved survival by protecting against protein aggregation (9).
We wanted to complement these studies by investigating the metabolic changes that occur following oxidative stress to provide additional mechanistic insights and to identify biomarkers of the stress to provide a way of examining bacterial stress exposure in vivo during infection.In parallel we were interested in establishing whether laser desorption rapid evaporative ionization mass spectrometry (LD-REIMS) can be used to study a bacterial stress response.
REIMS is a novel ionization method for mass spectrometry, originally developed as an intra-operative, in situ method for tumour margin detection, in which rapid heating of a sample generates an aerosol containing gas phase ions of metabolites and biologi cal molecules that can be analyzed to provide real time, semi-quantitative metabolic profiling (10).Rapid heating can be achieved through the application of electrical diathermy, for example in the form of a heated electrosurgical knife, or by use of a laser, as in LD-REIMS.REIMS is an ambient ionization mass spectrometry (AIMS) technique, meaning it allows for mass spectrometry analysis of biological samples in their natural state (11).Established AIMS techniques such as desorption electrospray ionization (DESI) and direct analysis in real time (DART) have previously been successfully applied to microbial metabolomics, allowing for imaging of microbial interactions (12)(13)(14) and high throughput identification of bacteria (15)(16)(17)(18)(19). Based on these promising applications of AIMS techniques, REIMS has been expanded from a surgical tool to applications in microbiology.The Takats group used electrical diathermy REIMS to characterize the metabolomes of P. aeruginosa clinical isolates by analyzing colonies grown on agar plates, successfully classifying isolates to a strain level with 81% accuracy and identifying a number of P. aeruginosa metabolites (20), while LD-REIMS has been used successfully for metabolic fingerprinting of biofluids and for screening of libraries for synthetic biology applications, able to detect a panel of diverse biological molecules in a yeast model, demonstrating its potential for broad applicability to metabolite detection (21)(22)(23)(24)(25).
The aim of this study was to determine whether LD-REIMS can be used as a high-throughput approach to study bacterial stress responses.We wanted first to test whether LD-REIMS can distinguish untreated from oxidatively stressed P. aeruginosa, and subsequently whether it can be used to reliably identify changes in the oxidant-specific biomarkers in this bacterium, as well as to provide mechanistic insights into the response to oxidative stress.

Bacterial strains and growth conditions
The P. aeruginosa lab strain PA14 was used for all experiments unless otherwise stated.The transposon mutant pqsE came from a P. aeruginosa transposon mutant library (26).The clinical isolates C1, C2, C3, C9, C13, C14, C25, C28, and C50 were supplied by the Royal Brompton Hospital (Table S1).All liquid cultures were grown in LB medium (5 g/L NaCl, 5 g/L yeast extract, and 10 g/L tryptone) at 37°C, shaking at 700 rpm.

Sodium hypochlorite
HOCl solutions were prepared by diluting NaOCl (Sigma) directly into the media.The molar concentration of OCl − was determined using iodometric titration.Potassium iodide (KI) and hydrochloric acid (HCl) were added to NaOCl, and the resulting solution titrated to a colorless end-point using sodium thiosulphate (Na 2 S 2 O 3 ) and a starch solution indicator.

Hypothiocyanous acid
HOSCN solutions were generated as described in reference (27) with modifications.Briefly, a solution containing 6.5 mM potassium thiocyanate (KSCN, Fluka) and 6 U/mL lactoperoxidase (LPO, Sigma) was prepared in LB.Aliquots of 1 mM H 2 O 2 were added over 5 min, followed by the addition of 100 U/mL catalase (Sigma).To remove proteins, the solution was centrifuged at 4,000 rpm for 10 min using a 10-kDa filter (Ambion).The concentration of HOSCN was determined by monitoring the loss of signal at 412 nm (ε412 nm = 14,150 M −1 cm −1 ) that occurs when HOSCN reacts with 5-thio-2-nitrobenzoic acid (TNB) (28).

Hydrogen peroxide (H 2 O 2 ) solutions
Solutions were prepared by diluting 30% wt/wt in H 2 O (9.79 M, Sigma) directly into the growth medium.

Methylglyoxal solutions
MGO solutions were prepared by diluting MGO (Sigma, 40% in H 2 O) directly into the growth medium.

Methyl viologen solutions
MV solutions were prepared by diluting MV (Sigma, 10 mM) directly into the growth medium.

Preparation of liquid-culture grown bacterial samples for LD-REIMS analysis
LB medium was inoculated with P. aeruginosa in a 96-deep-well plate and cultured overnight at 37°C, shaking at 700 rpm on a THERMOstar microplate incubator and shaker.
Each replicate was an independent culture.The overnight cultures were subcultured 1:50 into 190 µL of fresh media spiked with the oxidant of interest in a 96-well plate and incubated at 37°C, shaking at 700 rpm.Once balanced growth was achieved in the mid-exponential phase, the cultures were transferred to a V-shaped-bottom 96-well plate (Sigma) and centrifuged at 3,270 × g for 20 min at 4°C.The supernatant was removed, the pellets resuspended in 100 µL of sterile ammonium acetate solution (150 mM) and again centrifuged at 3720 × g for 20 min, at 4°C.The supernatant was removed, and the plate containing the bacterial cell pellet was introduced into the LD-REIMS plate reader for analysis.

LD-REIMS analysis
LD-REIMS analysis was performed by adapting the methodology described by reference (29).The pellets were subject to laser ablation using an Opolette 2940 laser system (Opotek).One laser burn was performed per pellet, with a fluence of 5 J/cm 2 and a laser firing duration of 3 s.Samples were analyzed at 3 s intervals.The aerosol produced was transported by vacuum through polytetrafluorethylene tubing (3.2 mm O.D., 1.6 mm I.D.) connected to the REIMS interface through a T-shaped piece, also permitting inflow of isopropanol (IPA) at a flow rate of 0.1 mL/min, as described by reference (30).All mass analysis was carried out in negative ionization mode using a Xevo G2-S QToF instrument with the cone voltage of 40 V, heater bias voltage of 80 V, and m/z scan range from 50 to 1,200 Da.The REIMS instrument was calibrated in sensitivity mode according to the manufacturer's standard instructions (Waters Corporation), using 0.1 mM sodium formate solution in isopropanol and water (90/10, vol/vol) at a flow rate of 0.1 mL/min.

Data processing and statistical analysis
Peak picking was performed using an in-house workflow based in R and Python and developed by members of the Takats group.The script is available on request from the authors.In summary, first, the raw data were background subtracted and lock mass corrected using one of three known P. aeruginosa metabolites as standards depending on the mass range of focus (m/z = 199.1698/529.3382/716.5231).Fast Fourier Transform filtering was applied to remove high frequency noise in the Fourier domain and the data peak-picked and mass recalibrated using the R package MALDIquant (31).The resulting data were saved in .csvformat, ready for post-processing.Multivariate statistical analysis was performed on the data using the MetaboAnalyst 5.0 platform (32).Sum normalization, logarithmic transformation, and Pareto scaling were applied.Principal component analysis (PCA) and orthogonal projections to latent structures discriminant analysis (OPLS-DA) were used for data visualization and support vector machines (SVMs) for classification and feature selection.SVM models were validated using F1 to measure model accuracy, calculated from the model precision and recall.

LC-MS/MS analysis
P. aeruginosa samples were prepared for LC-MS analysis by subculturing overnight cultures 1:50 into fresh LB medium, untreated or spiked with HOCl (3.2 mM).Once an OD 600 of 1.0 was achieved, 100 µL of each culture was transferred to a microcentrifuge.Hank's Balanced Salt Solution (100 µL, Gibco) was added and the mixture incubated at room temperature for 30 min.The mixture was then centrifuged at 3,270 × g for 5 min at 4°C, the supernatant removed, the remaining pellet washed using 1 mL ice-cold ammonium acetate solution (150 mM) in LC-MS grade H 2 O, centrifuged at 3,270 × g for 5 min at 4°C.The wash was repeated four more times.After the final supernatant removal, the pellet was stored at −80°C.LC-MS/MS analysis was performed by the National Phenome Centre (NPC) service using the lipidomics protocol in negative ion mode (33).Prior to analysis, the pellets were thawed and resuspended in ammonium acetate solution (150 mM).The final ratio of sample to IPA prior to injection was 1:4, as described in the protocol for human serum samples.

Virulence factor assays
P. aeruginosa strains were grown overnight in 5 mL of LB broth in a 50-mL centrifuge tube (Falcon) and incubated overnight at 37°C at 180 rpm, and pyocyanin and elastase measured in the P. aeruginosa culture supernatants as previously described (34,35).

Development of an LD-REIMS protocol to study the response of P. aeruginosa to oxidative stress
As REIMS has previously only been used to analyze bacterial colonies grown on agar, our first aim was to develop a protocol to analyze stressed, exponentially growing liquid cultures, with minimal disruption of cellular physiology between sampling and analysis, to investigate the metabolic changes occurring during the adaptive response of P. aeruginosa to oxidative stress.We grew P. aeruginosa cultures in a 96-well microtiter plate format with or without oxidative stress to exponential growth phase (OD 600 ~0.8).The cells were then pelleted and washed with 150 mM ammonium acetate before transfer to the high-throughput LD-REIMS autosampler (Fig. 1).
A typical LD-REIMS spectrum of P. aeruginosa samples prepared using this protocol (Fig. 2) showed a similar range of features to those described previously for analysis of P. aeruginosa colonies (20), with predominantly 2-alkyl-4-quinolones (AHQs) associated with the Pseudomonas quinolone signal (PQS) quorum sensing (QS) system detected in the range from m/z 240-320, rhamnolipids from m/z 330-710, and phospholipids from m/z 600 to 900.

LD-REIMS analysis can distinguish between untreated and oxidatively stressed P. aeruginosa cultures
We then tested our LD-REIMS protocol for the detection of metabolic differences between untreated P. aeruginosa cultures and cultures stressed by growth in the presence of the oxidants HOCl, HOSCN, and H 2 O 2 , the superoxide generator MV and the reactive electrophilic compound MGO that has been shown to promote oxidative stress (36).One non-lethal concentration was chosen for each oxidant which led to a 1-h growth lag of the treated compared to the untreated cultures.This criterion was chosen to provide evidence that the bacteria were responding to the stress, adapting, and then resuming growth in its presence (Fig. S1).The selected concentrations were 3.2 mM HOCl, 0.35 mM HOSCN, 60 mM H 2 O 2 , 1.8 mM MGO, and 0.6 mM MV.Ten biological replicates were analyzed for each of the six sample classes (untreated, HOCl, HOSCN, H 2 O 2 , MGO, and MV).
Given the complexity of the LD-REIMS spectra across the different conditions, we first used PCA, a dimensionality reduction technique, to condense the data, allowing us to visualize high-level variance patterns in score plots (Fig. 3).Clear clustering of the untreated and treated classes can be observed for each oxidant, indicating that growth in their presence causes substantial metabolic divergence in the bacteria for the biomolecules detected by LD-REIMS.
We then employed orthogonal projection to latent structures discriminant analysis (OPLS-DA) to confirm the discriminative power of LD-REIMS seen with PCA.OPLS-DA is a supervised modelling method, meaning the class labels are provided and the model identifies the variations in the data set that are most able to predict the given classes.While this approach provides greater power for class differentiation than PCA, it can be prone to overfitting, meaning the model fits the training data too well and cannot be generalized to new data sets.As one of the aims of this study was to identify biomark ers of oxidative stress that are universal to P. aeruginosa samples, it was particularly important to avoid overfitting.Therefore, we cross-validated the OPLS-DA models using the Q 2 coefficient, a statistic that describes the predictive quality of the model, providing a measure of its reliability.Our analysis generated a Q 2 value greater than 0.5 for each oxidant (Fig. 4), a commonly employed threshold for validation that suggests the model is less likely to be overfitted (37).
To provide further confirmation that the model was not overfitted, we tested the reproducibility of the data by performing an identical experiment on new cultures with the same level of biological replication (10 independent cultures) and then compar ing the two independent data sets.OPLS-DA generated Q 2 coefficients similar to the single-day analysis, indicating the discriminative power of the analysis was retained over different days, despite a reduction in separation in PCA scores plots for HOSCN, H 2 O 2 , and MGO that can be attributed to a batch effect (Figs S2 and S3).Overall, these analyses show that LD-REIMS can discriminate clearly and consistently between untreated and oxidatively stressed bacteria for a range of oxidants.

Can LD-REIMS reliably classify P. aeruginosa based on stress exposure?
We next wanted to determine whether LD-REIMS can reliably classify samples based on the specific stress they have been exposed to.To achieve this, we first split the merged data set at random into a training set and a test set, in an 80:20 ratio.We then used the training set to build a support vector machine (SVM) for classification and validated this model using k-fold cross-validation (k = 10).SVMs are commonly employed for large metabolomics data sets due to their resistance to outliers and ability to handle noisy data (38).Using the model, the samples were classified depending on oxidant exposure with 100% accuracy (Fig. 5).The model was then tested using the test data set and the pellets washed with ammonium acetate buffer and pelleted again.The time between starting to harvest the cultures and LD-REIMS analysis of the pellets was standardized to 60 ± 10 min.(2) Rapid heating and evaporation of the surface of each pellet was performed by moving the 96-well plate under a laser using a 2D moving platform.The resulting aerosol was transported to the LD-REIMS interface and analyzed using a QTOF mass analyzer.(3) Once the analysis was complete, the data were first processed using a peak-picking script.(4) Statistical analysis was performed on the processed data to determine which spectral features were significantly different between treated and untreated samples.Support vector machine (SVM) classification models were used to classify samples depending on treatment.(5) Metabolite identification was performed using a previous P. aeruginosa metabolite database generated in our lab and online databases.A selection of the most significantly impacted metabolites had their identity confirmed using LCMS/MS.samples were again classified with 100% accuracy, demonstrating that LD-REIMS analysis provides sufficient discriminatory metabolic information to classify samples as treated with a specific oxidant with very high confidence.

REIMS analysis can identify biomarkers of general oxidative stress and biomarkers specific to individual oxidants
We next investigated whether the metabolic changes detected in P. aeruginosa were specific to one or common to two or more oxidative stresses.Features were selected based on the relative importance of their linear regression coefficients obtained during SVM model training.In total, 333 spectral features were highlighted as significantly impacted by exposure to at least one oxidant.Of these, 54 (16.2%) were identified as unique metabolites (Table S2).Thirty-four of these were previously identified as P. aeruginosa metabolites using the same mass spectrometer (20).Twenty-four had their identity confirmed using LC-MS/MS (Table S3).
The changes in metabolite concentration observed in oxidatively stressed samples and the impact of each oxidant on the three classes of biomolecule are shown in Fig. 6 (see Table S4 for a list of the fold-changes in intensity and associated P values for each metabolite).These data allow us to draw conclusions about the specific metabolic impact of each oxidant.Exposure to HOCl significantly decreased the concentration of 10 AHQs (Fig. 6 and 7) to an extent not seen with any other oxidant, while also causing a decrease in the concentration of three rhamnolipids (Fig. 6 and 8).In contrast, 14 of the 21 phospholipids impacted by HOCl stress showed an increase in concentration, with the most notable exception being the phosphatidylglycerols PG(35:0), PG(35:1), and PG(35:2) which were significantly lower in concentration for all oxidants beside MV.No AHQ or rhamnolipid was impacted by exposure to HOSCN.However, 19 phospholipids were impacted by HOSCN stress.The phospholipids PE(30:0), PE(32:1), and PG(32:0) had higher concentrations in the HOSCN-treated samples than any other oxidant.Exposure to H 2 O 2 had the strongest impact on rhamnolipid concentrations of all the oxidants (Fig. 8), while the impact on phospholipids was similar to the other treatments besides MV.The metabolite profiles of MGO-treated samples were broadly similar to those of H 2 O 2 -treated samples, having an almost identical phospholipid profile and having the second strongest impact on the detected rhamnolipids after H 2 O 2 .MV had a unique profile compared to the other treatments in the phospholipid region, causing an increase in concentration of several phospholipids that decreased in concentration for all other oxidants, and vice versa.The phospholipids PE(33:2), PE(33:1), and PE(30:0) increased in concentration only upon exposure to MV. MV also had a distinct impact on the growth of P. aeruginosa cultures, causing a reduction in the final OD 600 reached in the stationary phase compared to untreated samples (Fig. S1).Each oxidant except HOSCN elicited at least one oxidant-specific metabolic change.

Biomarkers of HOCl and HOSCN stress are conserved in P. aeruginosa clinical isolates
Our data have established that LD-REIMS can discriminate between oxidatively stressed and untreated and identified biomarkers specific to exposure to different oxidants in PA14 samples, a commonly used lab strain of P. aeruginosa.For these findings to have potential use in the investigation of P. aeruginosa infections in vivo, for example, to establish whether bacteria are exposed to these stresses during infection, we needed to demonstrate that these specific metabolomic responses are conserved in clinical strains.Therefore, we carried out LD-REIMS analysis on 10 P. aeruginosa strains isolated from cystic fibrosis (CF) lung infections and compared their untreated and HOCl-treated metabolomes using LD-REIMS.
Separation in PCA scores plots between treated and untreated samples was reduced compared to the lab strain analysis (Fig. S4 and S5).However, this included the PA14 control, suggesting the reduced discriminative power was due to the lower number of replicates (n = 6 compared to n = 20).
We then tracked in the clinical isolates the biomarkers of HOCl and HOSCN established using PA14.The decrease in concentration of AHQs was conserved in the HOCl-treated clinical isolates (Table S5).C9:db-NHQ, C11:db-UHQ, and UHQ were conserved in all or all but one isolate, suggesting any impact of HOCl on the PQS system is not limited to the laboratory strain.The rhamnolipids and phospholipid biomarkers were poorly conserved in both the clinical isolates and the PA14 control.The HOSCN biomarkers were very well conserved (Table S6).All biomarkers besides PA(34:1) and PS(38:0) were conserved in the majority of clinical isolates.Only C13 and C25 presented with fewer than 50% of the biomarkers.

HOCl-stress reduces quorum sensing-mediated virulence factor levels
The decrease in concentration of 10 AHQs associated with the PQS quorum sensing system on exposure to HOCl was one of the strongest patterns observed in our data.While it is not possible to say whether the decrease in concentration is due to a reduction in their production or to direct oxidation of the metabolites, the outcome is likely to be a decrease in the efficacy of the molecules to modulate the PQS-mediated quorum sensing response of P. aeruginosa.To investigate this, we measured the concentration of pyocyanin and elastase, virulence factors known to be regulated by the PQS system, in wild-type cultures grown in the presence of HOCl.We compared the concentrations to those in pqsE mutants, which are mutated in genes required for PQS synthesis and are expected to produce little or no pyocyanin and elastase (39).Pyocyanin and elastase were clearly made by the wild-type strain but when grown in the presence of HOCl the levels were significantly affected (Fig. 9), with pyocyanin not detectable at concentrations above those in the pqsE mutant and elastase present at less than 25% of the concentration seen in the untreated bacteria.To test whether the decrease in pyocyanin was due to a decrease in production or direct oxidation, we treated stationary phase cultures in which pyocyanin was already present with HOCl and measured the concentration over time.We did not see a difference in the concentration of pyocyanin in treated cultures (Fig. S8), suggesting that HOCl does not cause direct oxidation of pyocyanin.

LD-REIMS can be used to confidently distinguish between oxidatively stressed P. aeruginosa cultures
In this study, we tested LD-REIMS as a novel technique for the metabolic profiling of P. aeruginosa, with the aim to identify biomarkers of oxidative stress and to provide insight into the oxidative stress protection systems employed by the bacteria.REIMS has been employed for a range of microbiological studies, most frequently for bacterial identification and classification (22,(40)(41)(42)(43), as well as metabolic screening of yeast colonies (21) and imaging (44).A study in the Takats' lab successfully classified P. aeruginosa isolates to a strain level with 81% accuracy and identified a number of P. aeruginosa metabolites (20).
While REIMS has been established for the analysis of bacterial colonies, the LD-REIMS protocol reported here allows for the analysis of pelleted samples, meaning bacterial samples grown in liquid, including high salt media that REIMS would be intolerant toward, can be analyzed in a high-throughput manner.We analyzed and compared the metabolomes of P. aeruginosa grown in the absence and presence of oxidative stresses, with the only sample preparation requirement being the centrifugation and washing of the samples just prior to analysis.Using our approach, 96 samples can be analyzed in 20 min, and samples are analyzed within 60 min of the start of sample preparation.Additionally, this was the first reported application of a 3 µm laser for high-throughput REIMS, after ongoing research found that a 3 µm, shorter pulse width laser led to increased ion yields compared to the original CO 2 setup (30).The limitation of REIMS is that it has a restricted metabolic coverage, largely detecting hydrophobic molecules in the surface layers of bacteria, but molecules that will be first to come into contact with an externally applied stress.
We used an SVM model to classify P. aeruginosa PA14 samples based on the specific oxidative stress they were grown in with 100% accuracy.This result was reproducible, with classification accuracy consistent for samples prepared and analyzed on different days.
These experiments were performed on the lab strain PA14, a burn wound isolate that represents the most common clonal group worldwide and is more virulent than other commonly used strains, PAO1 and PAK (45) (45).It is now recognized that it is important not to rely on analysis of a limited number of laboratory reference strains but to extend to natural isolates, including clinical strains (46).For this reason, we extended our LD-REIMS study to clinical isolates from CF patients.The biomarkers of HOCl and HOSCN stress were conserved in the clinical isolates.While the small number of clinical strains used in this study do not provide sufficient statistical power to look for conserved lifestyle specific biomarkers, the speed and simplicity of the LD-REIMS approach make this a feasible future research objective.

A unique metabolic profile was recorded for each oxidant
The observed classification accuracy of 100% indicates that growth in the presence of different oxidants impacts the P. aeruginosa metabolome in a manner unique to each oxidant, at least for the metabolites detectable using LD-REIMS.We were able to use a combination of univariate and multivariate analyses to find the metabolites with significantly different intensities between sample classes.These included representatives from all three of the main classes of P. aeruginosa metabolites detected using LD-REIMS: AHQs associated with the PQS quorum sensing system, rhamnolipids, and phospholipids (20).
Of the oxidants, HOCl and MV (superoxide) had the greatest impact on the P. aeruginosa metabolome, with both seen to cluster away from the untreated samples using PLS-DA.HOSCN, H 2 O 2 , and MGO did not separate from untreated when all oxidants were analyzed together; however, clear separation from untreated samples was observed when analyzed independently using two-group PCA and OPLS-DA, and all samples were classified with 100% accuracy.This suggests HOCl and MV have a broader impact on the metabolome, disrupting the concentrations of a greater range of molecules, while the other three impact fewer metabolites but still leave a consistent and unique mark on the metabolome.The greater impact of HOCl matches with the reaction profiles of the oxidants.HOCl is a stronger oxidant than H 2 O 2 or HOSCN, and able to react with a wide range of bacterial substrates including thiols and amines (47).HOSCN reacts almost solely with thiol groups, while H 2 O 2 only reacts directly with thiols and transition metal centres, due to the high activation energy of its reactions (48).Reaction profiles are not sufficient to explain the greater impact of MV, as superoxide has a low reactivity in aqueous solutions (1).However, the use of methyl viologen as an intermediary for superoxide generation may explain the impact on the metabolome.MV (also known as paraquat) is non-polar and can therefore pass through bacterial membranes, where superoxide is generated within the cell.The intracellular production of MV would likely have a significantly different impact on the metabolome compared to the other oxidants, which may cause more damage at the cell membrane.This is particularly relevant for this study, as the detected molecules are membrane-associated or extracellular.

HOCl has a specific impact on AHQs associated with the PQS quorum sensing system
Our most striking finding was the decrease in concentration of 10 AHQs associated with the PQS quorum sensing system after exposure to HOCl.Among these are 2-heptyl-3hydroxy-4(1H)-quinolone, also known as the PQS, the most active signal molecule of the PQS system that regulates the production of virulence factors including elastase, rhamnolipids, pyocyanin, and siderophores, such as pyoverdine, alongside inducing its own production (49).The decrease in concentration of AHQs was also seen in the clinical isolates (Table S6), clearly showing that the impact of HOCl on the PQS system is replicated in clinical strains.
PQS and other AHQs are released extracellularly by P. aeruginosa where their concentration acts as a proxy of the local bacterial population and subsequently dictates the expression of PQS-dependent genes.Due to their alkyl chain moiety, AHQs are hydrophobic and rely on membrane vesicles (MVs) for transport to and excretion from the cell.Their association with MVs and the cell envelope explains their prevalence in LD-REIMS P. aeruginosa mass spectra alongside phospholipids and rhamnolipids.The PQS system is associated with the oxidative stress response of P. aeruginosa, having been shown to act as both an antioxidant and pro-oxidant in different contexts (49)(50)(51).Its antioxidant effect is attributed to its nature as a strong electron donor due to its bisoxygenated aromatic structure, able to reduce intracellular reactive oxygen species (ROS) levels through direct reaction with oxidants.However, it has also been shown to sensitize P. aeruginosa to killing by H 2 O 2 and oxidative stress-mediated via ultraviolet radiation (50,52).This pro-oxidant effect is thought to be due to the ability of PQS to chelate Fe 3+ ions, resulting in the formation of free radicals.
It is unclear whether the decrease in concentration of the AHQs is due to HOCl inhibiting their production or to their direct oxidation.However, irrespective of the mechanism, the phenotypic impact is clear as the PQS-regulated virulence factor pyocyanin was undetectable and elastase concentration was reduced by >75% in HOCl stressed cultures.In addition, we showed that this was not due to oxidation of pyocyanin by HOCl, supporting a model whereby HOCl prevents the production of AHQs.However, we demonstrated previously that the transcriptional regulator RclR, which is required for protection of P. aeruginosa against HOCl stress, positively regulates the expression of the phz genes that encode the pyocyanin biosynthesis pathway in the presence of HOCl (7).While the pyocyanin phenotype is clear, these data indicate complexity in the effect of HOCl on pyocyanin formation and the associated regulatory circuits that need further investigation.An additional consideration is that the decrease in concentration of the AHQs in the presence of HOCl would additionally remove the potential antioxidant properties of the PQS system.Multiple lines of evidence support the importance of PQS-mediated quorum sensing and its upregulation of virulence factor production in the pathogenicity of P. aeruginosa (39,53,544,41,54).Therefore, our findings lead us to hypothesize that, during infection, neutrophil-generated HOCl formation results in the inhibition of virulence factor formation by P. aeruginosa due to its effect on PQS-medi ated quorum sensing, and that this augments the impact of direct neutrophil-killing of bacteria by reducing their pathogenicity.

FIG 1
FIG1 Protocol for high-throughput LD-REIMS analysis of P. aeruginosa exposed to oxidative stresses.(1) P. aeruginosa samples were grown in a 96-well plate, untreated or treated following balanced growth in exponential phase.This was achieved by inoculating 10 µL of P. aeruginosa stationary phase cultures into 200 µL of LB to give an initial OD 600 of approximately 0.05, followed by growth for 4 h to an OD 600 of approximately 0.8.The cultures were then pelleted, the (Continued on next page)

FIG 3
FIG 3 Statistical analysis of mass spectral data obtained by LD-REIMS by a two-group PCA of untreated and treated P. aeruginosa samples.(A) 3.2 mM HOCl, (B) 0.35 mM HOSCN, (C) 60 mM H 2 O 2 , (D) 1.8 mM MGO, and (E) 0.6 mM MV. Statistical analyses on samples (n = 10) were completed on the 50 to 1200 m/z range, after background subtraction and mass drift correction, using the MetaboAnalyst 5.0 platform.The plots all represent principal components 1 and 2. Shaded areas show 95% confidence intervals of the sample groups.Separation between untreated and treatment samples is clear for all oxidants.

FIG 4
FIG 4 Statistical analysis of P. aeruginosa mass spectral data obtained by LD-REIMS by OPLS-DA analysis.Score plots generated based on OPLS-DA of untreated and treated P. aeruginosa samples (n = 10), (A) 3.2 mM HOCl, (B) 0.35 mM HOSCN, (C) 60 mM H 2 O 2 , (D) 1.8 mM MGO, and (E) 0.6 mM MV. Statistical analyses were completed on the 50 to 1200 m/z range, after background subtraction and mass drift correction, using the MetaboAnalyst 5.0 platform.(F) The model can separate the untreated and treated samples for each oxidant, cross-validated using the Q 2 coefficient.

FIG 5
FIG 5 Confusion matrix for the SVM classification of P. aeruginosa samples exposed to oxidative stress conditions.The whole m/z ranges of the mass spectra acquired using LD-REIMS were employed to classify each sample into one of the six classes.The model was validated using k-fold cross-validation (k = 10).The samples were classified with 100% accuracy.

FIG 6
FIG 6 Heatmap visualisation of metabolite intensity changes in P. aeruginosa following treatment with oxidants.Color represents the variation in the normalized intensity of metabolite peaks in mass spectra, relative to the mean, where dark red means a higher intensity and blue a lower intensity.Metabolites are arranged into the three detected classes: PQS-associated AHQs, rhamnolipids, and phospholipids.

FIG 7
FIG 7 Changes in the intensities of PQS-associated quorum sensing molecules in P. aeruginosa exposed to oxidants.The box and whisker plots demonstrate the changes in AHQ intensities between different oxidative stresses.Exposure to HOCl caused a significant decrease in concentration of all 10 PQS-associated quorum sensing molecules identified using LD-REIMS, including the Pseudomonas quinolone signal (PQS) and its immediate precursor HHQ, the two most active signaling molecules for the PQS system.MV and H 2 O 2 caused a decrease in the concentration of four AHQs and one AHQ, respectively.The threshold criterion for significance was a P value <0.05, determined using ANOVA and Fisher's LSD post hoc analysis.Classes significantly different from untreated are highlighted with an asterisk.The box and whisker plots summarize the normalized intensity values (mean ± SD) for the metabolite.

FIG 8
FIG 8 Changes in the intensities of rhamnolipids in P. aeruginosa exposed to oxidants.Exposure to H 2 O 2 caused a decrease in concentration of all nine detected rhamnolipids in P. aeruginosa, while MGO, HOCl, and MV caused a decrease in concentration of five, three, and one rhamnolipid, respectively.The threshold criterion for significance was a P value <0.05, determined using ANOVA and Fisher's LSD post hoc analysis.Classes significantly different from untreated are highlighted with an asterisk.The box and whisker plots summarize the normalized intensity values (mean ± SD) for the metabolite.

FIG 9
FIG 9HOCl stress reduces the levels of the virulence factors pyocyanin and elastase in P. aeruginosa.P. aeruginosa wild-type cultures and a pqsE mutant, that is defective in a gene essential for PQS synthesis, were grown in the presence and absence of 3.2 mM HOCl, and their pyocyanin (A) and elastase (B) levels were analyzed (n = 3).Statistical analysis was performed using one-way analysis of variance with Tukey's multiple-comparison post hoc test.*, P < 0.05; **, P < 0.01; ****, P < 0.0001.