Metabolomic profiling of Burkholderia cenocepacia in synthetic cystic fibrosis sputum medium reveals nutrient environment-specific production of virulence factors

Infections by Burkholderia cenocepacia lead to life-threatening disease in immunocompromised individuals, including those living with cystic fibrosis (CF). While genetic variation in various B. cenocepacia strains has been reported, it remains unclear how the chemical environment of CF lung influences the production of small molecule virulence factors by these strains. Here we compare metabolomes of three clinical B. cenocepacia strains in synthetic CF sputum medium (SCFM2) and in a routine laboratory medium (LB), in the presence and absence of the antibiotic trimethoprim. Using a mass spectrometry-based untargeted metabolomics approach, we identify several compound classes which are differentially produced in SCFM2 compared to LB media, including siderophores, antimicrobials, quorum sensing signals, and various lipids. Furthermore, we describe that specific metabolites are induced in the presence of the antibiotic trimethoprim only in SCFM2 when compared to LB. Herein, C13-acyl-homoserine lactone, a quorum sensing signal previously not known to be produced by B. cenocepacia as well as pyochelin-type siderophores were exclusively detected during growth in SCFM2 in the presence of trimethoprim. The comparative metabolomics approach described in this study provides insight into environment-dependent production of secondary metabolites by B. cenocepacia strains and suggests future work which could identify personalized strain-specific regulatory mechanisms involved in production of secondary metabolites. Investigations into whether antibiotics with different mechanisms of action induce similar metabolic alterations will inform development of combination treatments aimed at effective clearance of Burkholderia spp. pathogens.

growth media such as LB is inadequate in modeling bacterial physiology during infection, as nutrient availability can vary markedly between infection sites and standard growth media. To overcome this challenge, the previously developed SCFM2 was used to model the physical and chemical environment of human sputum from patients with cystic fibrosis. In this study, we employed an untargeted metabolomic approach to compare the metabolomes of three different strains of B. cenocepacia namely, C5424, K56-2, and J2315 when cultured in SCFM2 and LB in triplicate (Fig. 1a). These strains are clonal and are associated with the highly transmissible, epidemic ET12 lineage of B. cenocepacia 19 . In addition, comparative metabolomic analysis was carried out in the presence and absence of trimethoprim, which is an antibiotic used clinically for treatment of Burkholderia infections 40 . This antibiotic is known to upregulate biosynthetic pathways involved in production of secondary metabolites in B. thailandensis and in the selected strains of B. cenocepacia in LB, but its effect has not been investigated in SCFM2 [41][42][43] . In order to capture a diverse range of compound classes, extractions were performed on bacterial cultures using both liquid-liquid (with ethyl acetate, EtOAc) and solid-phase extraction (SPE) methods. Extracts were then analyzed using high resolution tandem mass spectrometry coupled with ultra-highperformance liquid chromatography (UHPLC-HRMS/MS). Metabolite features representing analytes detected at a unique m/z and retention time were extracted, aligned, and quantified using the open-source MZmine2 software and feature-based molecular networking was performed (Fig. 1b) 44,45 .
We first compared the metabolome of each strain irrespective of media type, extraction method, and exposure to antibiotic (Fig. 2a). The largest number of unique metabolite features (20.5%) was detected exclusively in the extracts of the strain K56-2, whereas 8.5% unique features were detected in the extracts of J2315 strain and 1.1% in the extracts of the strain C5424. While 30.2% of features were shared by all three strains, the largest number of features were shared between the strains J2315 and C54524 (31.7%). Among other factors, the similar metabotype of these two strains may be reflective of their ability to produce a brown pigment known as pyomelanin, which is not produced by the strain K56-2 46,47 . Next, we compared the metabolomes acquired using the SPE and EtOAc extraction methods separately at 48 h post-inoculum. This analysis revealed that more metabolite features were exclusively detected in extracts generated using SPE as compared to extracts generated with EtOAc (Fig. 2b). A total of 32.5% of the metabolite features were detected only with the SPE method, while 9.5% of the total features were unique to EtOAc extraction method (Fig. 2b). Thus, the extraction method employed results in biased metabolomics comparisons when only one type of extraction strategy is employed. Lastly, we generated an UpSet plot to visualize the number of features unique to each strain under different growth conditions as well as exposure to sub-lethal dose of antibiotic trimethoprim after subtraction of media background (Fig. 2c) 48 . This analysis revealed that media specific differences were the largest driver of metabolomic diversity within this study, with 878 features shared between all LB samples and 563 features shared between all SCFM2 samples . In comparison, 513 features were detected in all samples. The UpSet plot also demonstrates that exposure to trimethoprim results in a unique metabolomic response by the K56-2 strain, with 362 metabolomic features exclusively detected in SCFM2 samples , 59 exclusive to LB samples , and 44 detected in both media types . Interestingly, 456 metabolomic features were uniquely detected in all LB samples except for K56-2 with trimethoprim and 37 in all SCFM2 samples except for K56-2 with trimethoprim . While our prior studies have shown that K56-2 exhibits a unique metabolomic response to trimethoprim, our analysis demonstrates that this response is even more apparent in an environment representative of CF sputum 43 . Another pattern observed using UpSet plot highlighted that many features are uniquely detected as shared between the pigmented J2315 and C5424 strains, with 203 features uniquely detected in LB , 90 in SCFM2 , and 93 across both media types . An additional 59 features were uniquely detected in C5424 and J2315 strains cultured in LB and SCFM2 in the presence of trimethoprim , hinting at a distinct response to trimethoprim that is associated with pyomelanin production. Such responses are hallmark of the strain-specific phenotypes of Burkholderia observed while growing under conditions encountered during infection [49][50][51][52] . Future investigations into phenotypic differences as well as differences in gene expression via transcriptomics will provide insights into the biochemical underpinnings driving these observations. Specific metabolites underlying these personalized chemotypes are discussed below.
Differences in siderophore production between LB and SCFM2 cultures. Siderophores are compounds secreted by bacteria to acquire iron from the surrounding environment, and were among the metabolites which were differentially produced by B. cenocepacia strains in LB and SCFM2 in this study. In healthy mammalian hosts, the pool of free iron is limited due to poor solubility of iron in its ferric state under physiological conditions, and because the majority of iron is either located in intracellular compartments or bound by host proteins such as hemoglobin, transferrin, lactoferrin, and ferritin 53,54 . In contrast, levels of both free and ferritinbound iron are higher in CF sputum compared to sputum from healthy hosts 55,56 . Since this cofactor is essential for many important biological processes, iron-acquisition is required for survival in the host environment and can influence microbe-microbe interactions 54,57 . Siderophores which are known to be produced by Burkholderia include pyochelin, ornibactins, salicylic acid, and cepabactin 43,54,58 . Among these, B. cenocepacia have been shown to produce ornibactin and pyochelin 54 . In our study, various structural analogs of pyochelin were detected exclusively in SCFM2 in the presence of trimethoprim (Fig. 3). Pyochelin itself was not detected in this www.nature.com/scientificreports/ study, despite prior reports of low-level production by B. cenocepacia. These prior reports did not use mass spectrometry or NMR to report pyochelin production, but relied on fluorescence-based thin layer chromatography 59 .
Using MS n analysis, we report production of a methylated derivative of pyochelin rather than pyochelin by B. cenocepacia strains used in this study as described below. The feature with m/z 339.083 had a database annotation as a pyochelin methyl ester in GNPS ( Supplementary  Fig. S1a) 60 . However, pyochelin methyl ester has only been reported as a synthetic product generated to facilitate NMR characterization, and there is no available biosynthetic evidence that would support methylation on the carboxylic acid to produce an ester 61 . This feature was detected exclusively in cultures grown in SCFM2 supplemented with trimethoprim in both the K56-2 and C5424 strains, albeit with lower levels observed in C5424 (Fig. 3a). Comparisons with the GNPS library MS 2 spectra of pyochelin revealed that the fragment peaks with m/z 120.045, 180.048, and 190.032 are shared between the two molecules while others were shifted by 14.015 Da (-CH 2 ), supporting the annotation as a methylated analog of pyochelin ( Supplementary Fig. S1b). A structure search of pyochelin in SciFinder revealed thiazostatin as a potential candidate. Thiazostatin A/B (1) are previously reported stereoisomeric natural products that are related to pyochelin by an additional C4″-methylation of the thiazolidine ring ( Fig. 3a) 62,63 . This thiazolidine C4″-methylation has also been observed in structurally homologous metabolites including isopyochelin, watasemycin, and yersiniabactin [64][65][66] . By MS 2 analysis alone, the location of the methylation cannot be unambiguously determined, although the mass shifts of the fragment ion with m/z 146.027 in the MS 2 spectrum of pyochelin to m/z 160.043 in our unknown metabolite's MS 2 spectrum suggests it is found on the terminal thiazolidine ring. To confirm the position of methyl group, we conducted b) a) c) 187   www.nature.com/scientificreports/ MS 3 analysis on the fragment ion with m/z 186.058. This analysis revealed that fragmentation of 186.058 yields a MS 3 ion with m/z 158.063, which indicates methylation on the thiazolidine ring rather than at the carboxylic acid ( Supplementary Fig. S2). As this detected metabolite is likely produced by same biosynthetic gene cluster as pyochelin, it is likely that the stereochemistry is also the same. Thus, this feature is putatively annotated as enantiothiazostatin, labeled in Fig. 3 as methylated. Thiazostatin A/B have been found to display antioxidant activity, and further screening is necessary to determine if the detected compound possesses additional bioactivities 62 . Using MASST through the GNPS platform, we searched the MS 2 spectrum of this compound against all public spectral datasets ( Supplementary Fig. S3) 67 . This MASST search found dataset matches in extracts of Pseudomonas spp. grown in vitro, as well as in datasets analyzing the metabolomes of humans with various inflammatory diseases (including CF, diabetes, irritable bowel syndrome, rheumatoid arthritis, and HIV). Detection of this molecule in humans with inflammatory disease raises the possibility that this molecule may be important in influencing microbiome structure in the host, although further studies will be needed to explore whether this observation is truly associated with any biological significance. Another metabolite with m/z 307.021 was detected exclusively in SCFM2 medium extracts in both the B. cenocepacia strains K56-2 and J2315. In our molecular network, this feature matched with 2′-(2-hydroxyphenyl)-4′-thiazolyl-2,4-thiazolinyl-4-carboxylic acid (HPTzTn-COOH, 2) in the GNPS spectral library, which was further verified by manually comparing experimental MS 2 spectra with a previously published MS 2 spectra (Fig. 3a, Supplementary Fig. S4a) 68 . Like pyochelin, HPTzTn-COOH is a siderophore that is dependent on salicylic acid and cysteine as biosynthetic precursors and is capable of chelating Fe 3+ in addition to other metal ions including Al 3+ , Ni 2+ , and Ca 2+68,69 . HPTzTn-COOH was detected exclusively in SCFM2 medium, and primarily in the K56-2 strain, although also at low levels in J2315 (Fig 3a). This metabolite exhibited increased production in the presence of trimethoprim for the K56-2 strain and was not detected in the absence of trimethoprim for the J2315 strain.
Next, a feature with m/z 222.022 was detected exclusively in K56-2 cultures grown in SCFM2 medium in the presence of trimethoprim (Fig. 3b, Supplementary Fig. S4b). This feature was annotated as aeruginoic acid (3), which is a shunt product in the pyochelin biosynthesis pathway observed in Pseudomonas and Burkholderia spp. 70,71 . Aeruginoic acid is the oxidized form of aeruginaldehyde (also known as the integrated quorum sensing signal, or IQS), which has been proposed as a "fourth QS molecule" in P. aeruginosa, although this claim has been disputed 72,73 . Due to the presence of the reactive aldehyde moiety in aeruginaldehyde, it reacts with complex natural products such as malleonitrone or mindapyrrole B 74,75 . Interestingly, pyochelin has been shown to spontaneously undergo cleavage and subsequent transformation into aeruginaldehyde when incubated in buffer solution at 30 °C 74   same conditions as aeruginoic acid, and annotated as aerugine (4) (Fig. 3b, Supplementary Fig. S4c). Aerugine has been previously isolated from Pseudomonas fluorescens and exhibits selective antifungal activity 76 . Recent work by Kaplan et al. has indicated that aeruginaldehyde, aerugine, and aeruginoic acid have iron-binding activity with aeruginoic acid binding to Fe 3+ with a 1:1 ratio, aeruginaldehyde binding with a 2:1 ratio, and aerugine binding with a 3:1 ratio 71 . Unlike pyochelin, aeruginoic acid has a specific affinity for iron compared to other biologically relevant metals 71 . The largest abundance of methylated pyochelin is observed in the strain K56-2, which is likely why these intermediates are also detected in the strain K56-2.
In contrast to metabolites from the pyochelin pathway described above, the ornibactin class of siderophores was detected in the culture extracts of LB medium and not detected in culture extracts of SCFM2 ( Supplementary  Fig. S5). The SCFM2 medium is prepared by adding 3.60 µM iron in the form of Fe 3 SO 4 , which was observed to be the average iron concentration present in expectorated sputum collected from CF patients 23 . Recently, it has been reported that commercial sources of mucin can be contaminating sources of iron that lead to altered siderophore production in P. aeruginosa cultured in SCFM2 22 . To determine whether the concentration of iron might account for the differential production of siderophores observed between SCFM2 and LB media, inductively coupled plasma mass spectrometry (ICP-MS) was performed on media aliquots. ICP-MS analysis revealed that SCFM2 contained a mean iron concentration of 5.25 µM (standard deviation of 0.52 µM) while LB contained a similar mean concentration of 4.73 µM (standard deviation of 0.62 µM) ( Supplementary Fig. S6). Therefore, factors other than iron availability likely play a role in the expression of the pyochelin and ornibactin biosynthetic gene clusters 11,25 . QS systems have been previously implicated in regulating production of both pyochelin and ornibactin, with the CepR transcriptional regulator repressing production of ornibactin and the CepR2 regulator activating production of pyochelin in B. cenocepacia H111 77 . In our experiment, production of siderophores was observed to be dependent on strain and nutritional environment. The production of metabolites from the pyochelin pathway was further induced by the antibiotic trimethoprim, while ornibactins were not. Thus, a specific chemical cue in SCFM2 in presence of trimethoprim might play a role in induction of pyochelin production under these conditions and warrants detailed investigation in the future with knockout strains that lack QS circuitry. This observation is noteworthy as pyochelin production by B. cepacia was previously suggested to be correlated with morbidity and mortality in patients with CF 78 . Mechanistic investigations into selective induction of siderophore biosynthesis pathways are critical to understanding their relevance to infections, and as such require further inquiry.

N-acyl-homoserine lactones production in LB compared to SCFM2. QS mediates bacte-
rial response to changing environmental conditions through cell-density dependent global changes in gene expression 79,80 . Burkholderia can utilize QS to coordinate various metabolic processes and modulate cellular phenotypes such as swarming, aggregation, spatial structuring, and biofilm formation 81,82 . QS also plays an important role in infection by regulating genes involved in virulence, and so establishing how the external environment influences production of QS signals is important to understanding their role in pathogenesis [83][84][85] . The CepIR and CciIR QS systems have been described in B. cenocepacia strains which primarily produce and sense N-octanoyl-homoserine lactone (C8-AHL, 5) and N-hexanoyl-homoserine lactone (C6-AHL) respectively. In addition, an orphan LuxR-type regulator called CepR2, which is antagonized by C8-AHL, is also present in these strains 77,[86][87][88] . The gene for the orphan CepR2 regulator lacks an adjacent gene required for synthesis for a cognate N-acyl-homoserine lactone. In a previous untargeted metabolomic experiment, we observed production of a wide diversity of AHLs by Burkholderia spp. grown in LB medium, as well as their corresponding acylhomoserine products formed by hydrolysis of the lactone ring (referred to hereafter as "hydrolyzed AHLs") 43 . In this study, we queried whether these signals are differentially detected in SCFM2 compared to LB media. The C8-AHL (m/z 228.160, 5) was detected in both LB and SCFM2 media, and its hydrolyzed form (6) was detected in LB medium alone (Fig. 4). Additionally, we detected hydrolyzed C13-AHL (7), hydrolyzed C13-AHL:1db (8), the sodium adduct of hydrolyzed C13-AHL (9), and the sodium adduct of hydrolyzed 3-OH C13-AHL (10) exclusively in SCFM2 when trimethoprim was present (Fig. 4). Thus, the production of these C13-AHLs were induced by trimethoprim only in SCFM2. Detection of C8-AHL, hydrolyzed C8-AHL and hydrolyzed C13-AHL was confirmed using commercial AHL standards. These AHL standards were treated with sodium hydroxide to promote hydrolysis of the lactone ring to verify the detection of hydrolyzed AHLs (Supplementary Fig. S7). Naturally produced AHLs with an odd number of carbons in the acyl sidechain are relatively rare, and their functions are not well-characterized 89 . In a previous untargeted metabolomics study of 10 different Burkholderia strains grown in LB medium, we detected hydrolyzed 3-oxo-C13-AHL:1db exclusively in extracts of B. thailandensis E264 43 . To our knowledge, the present study represents the first description of C13 AHLs being produced by B. cenocepacia. Further studies will be needed to explore the function and biochemical basis for production of C13-AHLs by B. cenocepacia in SCFM2 medium, and identify the mechanism by which trimethoprim upregulates the production of this AHL.
Fragin and pyrazine secondary metabolites. Fragin (11) is a diazeniumdiolate metallophore with antifungal activity that is produced by Burkholderia and Pseudomonas spp. 90 . In a previous untargeted metabolomics study, we reported significantly increased production of fragin and its structural analogs in B. cenocepacia K56-2 cultures grown in LB medium when supplemented with the antibiotic trimethoprim 43 . Fragin production was only observed in the K56-2 strain, despite the fact that the ham gene cluster responsible for fragin biosynthesis is identical in the closely related J2315 and C5424 strains. This result highlighted that even genetically similar Burkholderia strains can exhibit markedly different responses to external stimuli, such as trimethoprim exposure. In the current study, fragin was detected in both media conditions, but the majority of other nodes in the cluster (11/17) were detected exclusively in either LB or SCFM2 media (Fig. 5a). These analogs differ in the www.nature.com/scientificreports/ length of the acyl group added by the HamF enzyme, likely a result of differential availability of compounds containing variable acyl chain lengths which act as substrates for HamF. The MS 2 spectra of these analogs showed a characteristic fragment corresponding to the loss of NO group (29.998 Da), that was also observed in the MS 1 spectrum as an in-source fragment 43 . Fragin analogs with differential production in the two media types include nodes with m/z 302.243 (and the corresponding in-source fragment with m/z 272.246), 316.223 (in-source fragment with m/z 286.225), and 318.239 (in-source fragment with m/z 288.240) (Fig. 5a). We discovered another cluster in our molecular network containing several molecules which, like fragin, are exclusively detected in K56-2 samples and show increased production in the presence of trimethoprim (Fig. 5b). These metabolites were reported by our group in a previous study as unknown metabolites 43 . Here we employed the recently developed CANOPUS tool to classify these unknown compounds into ClassyFire chemical classes, leading to their annotation as pyrazines 91,92 . This information led us to conduct a literature search of pyrazines produced by Burkholderia, enabling annotation of these metabolites as pyrazine N-oxides (PNOs), which was verified using standards provided by Li and colleagues (Supplementary Figs. S8, S9) 93 . The expression of pvfB and pvfC genes in animal pathogen Pseudomonas entomophila L48 and the plant pathogen Pseudomonas syringae pv. syringae UMAF0158 were previously shown to lead to production of a family of PNOs, namely PNO B (12, m/z 181.135), PNO A (13, m/z 197.126), and dPNO (14, m/z 199.145) 93 . The pvfB and pvfC genes are homologous to the Burkholderia genes hamC and hamD present in the biosynthetic gene cluster of fragin. Disruption of pvfA-D cluster in animal pathogen Pseudomonas entomophila L48 and the plant pathogen Pseudomonas syringae pv. syringae UMAF0158 was shown to significantly reduce the virulence of these strains 93 . Subsequent studies found that the pvf gene cluster is involved in synthesis of a signaling molecule which regulates production of small molecule virulence factors such as monalysin in P. entomophilia and mangotoxin in P. syringae [94][95][96] . Both dPNO and PNO B appeared in the feature-based molecular network, while PNO A did not due to low abundance. However, a node corresponding to PNO A was observed when data was analyzed with a classical molecular network (Fig. 5b). Two additional nodes in this network were putatively annotated as 2-isopropyl-3-methoxypyrazine (m/z 153.102, 15) and 2,5-diisopropylpyrazine (m/z 165.138, 16) based upon available literature of bacterially produced pyrazines (Fig. 5b, Supplementary Fig. S9) 97 . The MS 2 spectra of these related molecules did not have fragment ions in common, and so molecular networking methods alone failed to highlight the structural   www.nature.com/scientificreports/ relatedness of these three compounds. Nevertheless, CANOPUS enabled us to independently predict that each of these compounds were pyrazines, ultimately leading to their annotation. These observations reveal that the ham gene cluster that is responsible for fragin biosynthesis also leads to production of PNOs in B. cenocepacia. PNO production was also induced by the addition of trimethoprim in both the SCFM2 and LB culture media. Thus, unlike the siderophores described above which showed mediadependent induction by trimethoprim, fragin and PNOs are similarly produced in the presence of antibiotic trimethoprim in both LB and SCFM2. The production of fragin and pyrazines including PNOs in the presence of trimethoprim in both SCFM2 and LB media highlights that antibiotics can serve as signaling molecules capable of significantly modulating expression of the genes encoding virulence factors across multiple nutritional environments.
Differential detection of lipids in LB and SCFM2. The overall lipid composition of bacterial cells has been reported to be influenced by several local environmental factors, including pH, nutrient availability, oxygen levels, temperature, and buildup of metabolic waste products 98 . Several clusters with hits to lipids in the GNPS spectral database were detected at higher levels in bacterial extracts grown in SCFM2, including hopanoids,  www.nature.com/scientificreports/ phytomonic acid, monosaturated monoacylglycerols (MGs), and phosphatidylethanolamines (PEs), described below (Fig. 6). Annotation of the hopanoid cluster was performed by first searching for candidate features with m/z calculated for known bacterial hopanoids. Identified candidate features were then verified by comparing experimental MS 2 spectra to spectra previously published in literature 99,100 . Once annotations were supported through spectral matching, these features were used to further propagate annotations to connected nodes within the molecular network based on mass differences. Hopanoids are pentacyclic triterpenoids frequently found in bacterial membranes 101,102 . These polycyclic lipids are structurally analogous to eukaryotic sterols and are thought to have similar functions in regulating fluidity, permeability, and stabilization of bacterial membranes. Hopanoid biosynthesis has been previously reported in B. cenocepacia, where they are involved in resistance to low pH and antibiotics while also being important for swimming and swarming motility 101,102 . A metabolite feature with m/z 708.540 is annotated as bacteriohopanetetrol (BHT) cyclitol ether (17) and was detected only in extracts of SCFM2 (Fig. 6a, Supplementary Fig. S10a) 99,102 . Production of BHT cyclitol ether has been previously reported in B. cenocepacia, and our predicted annotation was supported by comparing experimental MS 2 spectra with previously published spectra available in literature 99,102 . In addition, we annotated the node with m/z 706.526   www.nature.com/scientificreports/ as unsaturated BHT cyclitol ether (18) (Fig. 6a). Both Δ 6 and Δ 11 monosaturated BHT analogs have been previously characterized in bacteria 100 . Although our mass spectrometry analysis alone is insufficient to pinpoint the location of this unsaturation, we have putatively annotated this feature as bacteriohop-6-enetetrol cyclitol ether (18) since unsaturation at this location has been reported in B. cepacia strains 102,103 . This feature was detected in all three B. cenocepacia strains when grown in SCFM2 medium with the highest intensity observed in the K56-2 strain, but when cultured in LB it was only detected in J2315 cultures at low levels. Annotation of another feature with m/z 724.535 is consistent with a gain of a hydroxyl group from BHT cyclitol ether (17). This feature was exclusively detected in K56-2 strains grown in SCFM2 medium and the associated MS 2 spectra indicates this feature is likely bacteriohopanepentol (BHP) cyclitol ether (19) (Supplementary Fig. S10b). BHP derivatives have been fully characterized only in Acetobacter spp., Azotobacter vinelandii, and Nostoc spp., although our observation is supported by prior evidence suggesting BHP derivatives are produced by B. cepacia strains as well [103][104][105][106] . The feature with m/z 722.520 was detected in all three B. cenocepacia strains (largely in K56-2 samples) grown in SCFM2 medium, which is presumably bacteriohop-6-enepentol cyclitol ether (20) although the location of unsaturation cannot be unambiguously determined as mentioned above. Next, a feature with m/z 750.551 was detected exclusively in SCFM2 containing cultures with a mass difference of 42.011 (C 2 H 2 O) from BHT cyclitol ether representative of acetylation. The final feature in the hopanoid cluster with m/z of 748.543 is annotated as unsaturated analog of the acetylated BHT cyclitol ether (m/z 750.551), with the unsaturation most likely occurring at the C6 position. Another metabolite in the lipid family is annotated as phytomonic acid (21) based on the spectral match to the MS 2 spectrum in the GNPS database. This annotation was confirmed using a commercial analytical standard (Fig. 6b, Supplementary Fig. S11a). Phytomonic acid was detected in both LB and SCFM2 media, although consistently higher levels were detected in SCFM2 medium for all three strains ( Supplementary Fig. S11b). Production of cyclopropane acids have been previously reported in B. multivorans 107 . Cyclopropane fatty acids such as phytomonic acid are suggested to regulate membrane fluidity and stability and have been shown to increase extracellular survival in acidic or under conditions of high osmolarity 108,109 . These lipids also are major components of the membranes of intracellular pathogens such as Brucella abortus and Mycobacterium tuberculosis 108 (Fig. 6c, Supplementary Fig. S12). For both 19:1 MG (22) and 17:1 MG (24), two unique metabolic features were detected with identical masses and MS 2 patterns but slightly different retention times, possibly corresponding to distinct cis/trans isomers. Burkholderia spp. are known to produce the lipase LipA, which yields monoacylglycerols as a product during degradation of di-and tri-acylglycerides 110,111 . In B. cenocepacia, production of the LipA lipase is induced as part of the CepIR quorum sensing system 86,112 . Due to the ability of monoacylglycerols to destabilize bacterial cell membranes, several have been reported to demonstrate antimicrobial activities which vary based on chain length and degree of unsaturation 113,114 .

25-
Finally, several phosphatidylethanolamines (PEs) were detected in this study, all consistently detected under similar conditions. In Gram-negative bacteria, the inner leaflet of the outer membrane is made up of phospholipids, of which PEs are the major component 115 (Fig. 6d) 116 . Production of 2-OH-PEs has been observed in B. cepacia, and is reported to increase during stress observed under growth in high temperature 117 . Differential metabolism of the antibiotic trimethoprim between LB and SCFM2. Several compounds that were differentially detected between growth in SCFM2 and LB were found to be related to trimethoprim as evidenced by MS 2 spectral similarity. As previously described, MS2LDA was used to discover metabolites that contained trimethoprim as substructure. Thus, being higher in molecular weight than trimethoprim, these metabolites represent biochemical transformations of trimethoprim itself by B. cenocepacia bacteria 43 . In this study, MS2LDA Mass2Motif 543 was annotated as a trimethoprim substructure. We compared the presence of these metabolites in the extracts of bacteria cultured in LB and SCFM2 (Fig. 7). Metabolism of trimethoprim was carried out by only the pigmented strains J2315 and C5424 and not by the non-pigmented K56-2 strain, as reported previously 43 . We generated a volcano plot of molecules containing the trimethoprim motif to visualize compounds which were differentially detected across the two media conditions (Fig. 7a). This analysis revealed that the majority of the trimethoprim metabolites showed significantly higher production in SCFM2 as compared to LB. Structural characterization of these metabolites will provide insight into the pathways used by pigmented Bcc strains to metabolize xenobiotic compounds like the antibiotic trimethoprim and will facilitate future studies exploring how biotransformation will impact antibacterial activity.

Conclusion
It is well established that environmental conditions are major drivers of secondary metabolite production in bacteria. Therefore, selecting an appropriate culture media for in vitro bacterial growth is crucial for designing metabolomic studies relevant to the biological system of interest. In this study, we characterized the metabolomic profiles of three clinical B. cenocepacia isolates in SCFM2 and LB media. Sublethal concentrations of www.nature.com/scientificreports/ trimethoprim has previously been shown to induce secondary metabolite production in Burkholderia spp. Thus, culturing was also performed with and without this antibiotic to understand how trimethoprim-induced metabolomic responses vary under CF relevant environmental conditions 42,43 . We demonstrate considerable metabolic variability between SCFM2 and LB media. In particular, we report that growth in SCFM2 medium upregulates production of pyochelin-type siderophores in the presence of trimethoprim while downregulating production of ornibactin siderophores compared to LB medium. We also note that AHL quorum sensing signals are differentially produced between the two media, with C13-AHLs exclusively detected in SCFM2 in the presence of antibiotic trimethoprim. Thus, trimethoprim induces both the production of pyochelin-type siderophores and a specific AHL signal in SCFM2 medium and not in LB. Moreover, we observed that metabolism of trimethoprim itself is significantly upregulated in SCFM2 medium compared to LB. Finally, we show that several lipid families (including hopanoids, cyclopropane fatty acids, monoacylglycerols, and phosphatidylethanolamines) exhibit increased production in SCFM2 compared to LB. While further work is needed to fully understand the biochemical mechanisms underlying these findings, this study provides insight into variable production of secondary metabolite production by Burkholderia cenocepacia spp. and is important for delineating personalized metabotypes of different strains of the same species in an in vitro model of CF. Even though metabolomics approaches have been significantly advanced in the last decade enabling detection of low-abundance metabolites, it remains a challenge to directly detect and identify a diversity of pathogen-specific chemical signals in clinical samples such as human sputum since host biomass typically vastly exceeds microbial biomass. Thus, interrogation of pathogenic strains isolated from infection sites in simplified model systems capable of inducing chemical signals observed during infection serves as an excellent discovery tool. Characterizing metabolomic profiles of B. cenocepacia strains isolated from infection sites and cultured in SCFM2 medium in the presence of antibiotics can become a valuable approach to guide appropriate use of combination treatments for effective pathogen clearance.

Methods
Bacterial strains. Burkholderia cenocepacia C5424, K56-2, and J2315 clinical isolates were used for culturing during this study 59,118,119 . These strains belong to the highly transmissible and epidemic ET12 lineage, wherein the J2315 and K56-2 strains are clonally related. As reported, the C5424 and J2315 strains produced observable amounts of the pigment pyomelanin whereas the K56-2 strain did not 46 .
Media formulation and growth conditions. SCFM2 medium was prepared as previously described by Turner and co-workers 31 . All culturing was performed following previously established methods 43 . Briefly, B. cenocepacia C5424, J2315, and K56-2 frozen glycerol stocks were streaked onto LB agar plates, and then incubated overnight at 37 °C. Subsequently, 5 mL of LB media was inoculated using colonies from the overnight incubated plates, which was then incubated overnight at 37 °C while shaking at 250 rpm. www.nature.com/scientificreports/ was added to whole cultures and mixed every 30 min for 2 h before centrifuging at 2000×g for 3 min. The organic layer (top) was aspirated off with a glass pipette and transferred to a scintillation vial before drying in vacuo with a centrifugal evaporator. For solid-phase extractions, cultures were centrifuged at 10,000×g for 10 min and decanted to remove supernatant fluid. SPE columns were initially washed with 10 mL of 100% MeCN, subsequently equilibrated with water (10 mL) and then loaded with the culture supernatant. Analytes were then sequentially eluted from the column using 5 mL each of 20%, 50% and 100% MeCN. These three fractions were pooled together and dried in vacuo with a centrifugal evaporator.
Ultra-high-performance liquid chromatography tandem mass spectrometry (UHPLC-MS/MS) data acquisition. Dried (1) were performed using a Waters Corporation Cortecs UPLC T3 column (2.1 × 150 mm, 1.6 µm particle size) coupled to a high-resolution accurate mass Orbitrap ID-X tribrid mass spectrometer. The chromatographic method for sample analysis involved elution with 100% water 0.1% formic acid (mobile phase A) and MeCN and 0.1% formic acid (mobile phase B) using the following gradient program: 0 min 95% A; 0.5 min 95% A; 6 min 0% A (curve 7); 9.4 min 0% A; 9.5 min 95% A; 11 min 95% A. The flow rate was set at 0.4 mL/min. The column temperature was set to 40 °C, and the injection volume was 1 µL. The Orbitrap ID-X is a tribrid spectrometer that utilizes quadrupole isolation with dual detectors, an orbitrap and an ion trap, with a maximum resolving power of 500,000 full width at half maximum (FWHM) at m/z 200 and mass accuracy of < 1 ppm. The heated electrospray ionization (HESI) source was operated at a vaporizer temperature of 275 °C, a spray voltage of 3.5 kV, and sheath, auxiliary, and sweep gas flows of 40, 8, and 1 in arbitrary units, respectively. The instrument acquired full MS data in the 100-1000 m/z range in positive ionization mode at 30,000 resolution. MS 3 data was collected with a MS 1 isolation window of 0.8 mz, HCD activation of 30% ± 50%, a MS 2 isolation window of 1.8 mz, MS 2 HCD activation of 40%, and product ion detection in the orbitrap at 30,000 resolution. Data processing, feature-based molecular networking, and feature annotation. The LC-MS/ MS data presented in this manuscript is deposited in the online repository MassIVE and publicly available (MSV000087793). The Compass DataAnalysis software was used to convert all the raw spectral files (.d format) to centroided, lock-mass corrected format (.mzXML) for downstream analyses. The converted spectral files were uploaded to MZmine2 (v.2.5.1) for feature detection 44 . Data files were batch processed and filtered using a MS 1 mass detection signal threshold of 5.0 × 10 2 counts. The following parameters were applied: (i) chromatogram builder (minimum time span: 0.05 min; minimum intensity of the highest data point in the chromatogram: 1.5 × 10 3 ; m/z tolerance: 10 ppm); (ii) chromatogram deconvolution (local minimum search, m/z range for MS 2 scan pairing: 0.025 Da; retention time range for MS 2 scan pairing: 0.2 min); (iii) isotopic peaks grouper (m/z tolerance: 10 ppm; retention time tolerance absolute: 0.1 min; maximum charge: + 3; monotonic shape: true; representative isotope: most intense); (iv) join aligner (m/z tolerance: 10 ppm; retention time tolerance: 0.1 min); (v) peak finder (intensity tolerance: 10%; retention time tolerance (absolute): 0.1; m/z tolerance: 10 ppm). A table with ion intensities for each feature was exported (.csv format) for statistical analyses and the "Export for SIRIUS" module was used to generate an .mgf file for batch analysis with SIRUS 4. Additionally, the "Export for GNPS" module was used to convert and export the feature quantification table (.csv format) and the corresponding list of MS 2 spectra linked to MS 1 features (.mgf format) needed to generate a feature-based molecular network. The feature quantitation table and .mgf file were submitted to the Global Natural Products Molecular Networking (GNPS) platform along with a metadata file (.txt format) containing sample information, and a feature-based molecular network was created 45 . The molecular network and parameters used can be accessed using the following link: https:// gnps. ucsd. edu/ Prote oSAFe/ status. jsp? task= a80b0 5b26a 1747c fa6ca f7f94 46edd 75. Briefly, the data was filtered by removing all MS/MS peaks within ± 17 Da of the precursor m/z. A parent mass tolerance of 0.01 Da and a MS/MS fragment ion tolerance of 0.05 Da were applied to create consensus spectra. The network was created with the edges filtered to have a cosine score above 0.7 and at least 4 matched peaks, www.nature.com/scientificreports/ and edges connecting two nodes were set to be kept in the network if each of the nodes appeared in each other's respective top 10 most similar nodes. All the mass spectra in the generated network were queried against the spectral libraries available on GNPS. The matched experimental and library spectra were set to have a similarity score above 0.7 and more than 4 matched peaks. This network was exported to Cytoscape for visualization, and nodes appearing in media or solvent blanks were removed for clarity (unless otherwise indicated) 120 . Boxplots were constructed for features of interest using the Plotter Dashboard (v.0.4) available on the GNPS platform. The UpSet plot was generated with the UpSetR package in R 48 . Metabolomic features of interest that remained unknown after library searching were further annotated by first searching through literature for molecules which are known to be produced by Burkholderia spp. and developing an in-house database. Our database was curated by inserting structures manipulated using MarvinSketch (v.20.9.0, ChemAxon Ltd.) into a spreadsheet installed with JChem for Excel (v.20.8.0.62, ChemAxon Ltd.). This database was used to identify candidate annotations for features, and these annotations were confirmed by manually inspecting MS 2 spectra and either comparing against published MS 2 spectra (when available) or against the MS 2 spectra of commercial analytical standards. We then applied the MS2LDA workflow to recognize patterns of common MS 2 fragments and neutral losses ("Mass2Motifs") corresponding to molecular substructures in our dataset 121 . This MS2LDA analysis is available at the following link: https:// gnps. ucsd. edu/ Prote oSAFe/ status. jsp? task= 62862 2833d 00460 48dba 64e65 ced54 2c. In addition, we performed batch analysis on the dataset using SIRIUS 4 integrated with CSI:FingerID and CANOPUS 91,92,[122][123][124][125] . SIRIUS 4 (v.4.0.1) was employed (using the default settings for a qTOF instrument) to predict molecular formulas for unknown features and develop fragmentation trees for manual annotation of MS 2 spectra 122-124 . CSI:FingerID was then utilized to predict molecular properties of unknown features, which were then queried against molecular properties predicted for compounds in all available molecular databases 125 . This in silico tool led to a ranked list of predicted structures, even when published MS 2 spectra were not available for these structures. Next, we deployed CANOPUS to classify features into molecular families using ClassyFire, providing biological insight in the absence of structural annotations 91,92 . MASST searching was used to find metabolomic datasets collected by other researchers containing a match to the metabolic feature annotated as enantiothiazostatin (1) 67 . This MASST job and the parameters used are available at: https:// gnps. ucsd. edu/ Prote oSAFe/ result. jsp? task= c5d7b 5cc74 59443 286fe d543e 5140c a4& view= view_ all_ datas ets_ match ed.
To determine differences in trimethoprim metabolism between B. cenocepacia strains grown in SCFM2 and LB media, metabolic features containing a trimethoprim substructure (appearing as Mass2Motif 543 in the MS2LDA analysis) were extracted from the feature quantification table (.csv format) generated by MZmine2. These features were then subjected to further statistical analysis on the MetaboAnalyst web server 126 . After uploading the peak intensity table for trimethoprim metabolites, data was log transformed and Pareto scaling was applied. The volcano plot comparing trimethoprim against non-trimethoprim cultures were then created using a fold-change threshold of 2.0 and an FDR adjusted p value threshold of 0.05, as calculated by a t-test.
ICP-MS analysis of LB and SCFM2 media. ICP-MS was performed on aliquots of LB and SCFM2 media on a Perkin Elmer NexION 2000 with a S10 Autosampler at the Emory University Mass Spectrometry Center. Samples were diluted 1:10 and 1:100 with Type I deionized water before aspirating into the plasma using the NexION 2000 Peristaltic pump. A kinetic energy discrimination method with helium flow set to 5 arbitrary units was used to detect the 57 Fe isotope using a pulse counting detection method. Calibration was done from 0.1 to 1000 ppb. Samples were analyzed in technical triplicate for each media type. To determine the amount of iron in each media sample, the mean and standard deviation of all concentrations measured across both dilutions was calculated and converted from ppb to micromolar units.

Data availability
The data obtained in this study has been deposited with the Mass Spectrometry Interactive Virtual Environment (MassIVE) with the identifier MSV000087793 and is accessible at ftp:// massi ve. ucsd. edu/ MSV00 00877 93/. www.nature.com/scientificreports/