A Sensitive GC-MS Method for Quantitation of Lipid A Backbone Components and Terminal Phosphate Modifications

Abstract

Lipid A, the hydrophobic anchor of lipopolysaccharide (LPS) present in the outer membrane of Gram-negative bacteria, serves as a target for cationic antimicrobial peptides, such as polymyxins. Membrane stress from polymyxins results in activation of two-component regulatory systems that produce lipid A modifying enzymes. These enzymes add neutral moieties, such as aminoarabinose (AraN) and ethanolamine (EtN) to lipid A terminal phosphates that mask the phosphate’s negative charge and inhibit electrostatic interaction with the cationic polymyxins. Currently, these modifications may be detected by MALDI-TOF MS; however, this analysis is only semiquantitative. Herein we describe a GC-MS method to quantitate lipid A backbone hydrolyzed into its individual moieties, and derivatized via methoximation followed by silylation prior to analysis via GC-MS. Changes in AraN and EtN quantity were characterized using a variety of regulatory mutants of Salmonella, revealing differences that were not detected using MALDI-TOF MS analysis. Additionally, an increase in the abundance of AraN and EtN modifications were observed when resistant Enterobacter and Escherichia coli strains were grown in the presence of colistin (polymyxin E). Lastly, increased levels of Pi were found in bisphosphorylated lipid A compared to monophosphorylated lipid A samples. Because lipid A modifications serve as indicators of polymyxin resistance in Gram-negative bacteria, this method provides the capacity to monitor polymyxin resistance by quantification of lipid A modification using GC-MS.

Graphical Abstract

INTRODUCTION

Gram-negative bacteria contain an inner and outer membrane, whereby the outer leaflet of the outer membrane is composed mainly of lipopolysaccharide (LPS)¹. Lipid A is the membrane anchor of LPS and serves as a pathogen-associated molecular pattern (PAMP) sensed by the eukaryotic innate immune system, in addition to being a barrier between the bacterial cell and extracellular milieu.^2,3 The structure of lipid A normally consists of a diglucosamine backbone with amide and ester linked acyl chains, as well as two terminal phosphate moieties (Figure 1). The phosphate moieties contribute a net negative charge on the bacterial surface allowing polymyxins, which are cationic cyclic antimicrobials, to electrostatically bind and disrupt the outer membrane resulting in cell death.⁴ In response to polymyxins or other environmental stresses, Gram-negative bacteria can activate two-component regulatory systems (TCSs) that facilitate modification of the terminal phosphates with neutral moieties that mask the negative charge and prevent electrostatic interactions with the positively charged polymyxins.^4,5 These TCSs are well studied in Salmonella, whereby the two main systems involved in lipid A remodeling are PhoPQ and PmrAB.^5–8 Membrane stress activates PhoPQ leading to production of PagL, a 3-position deacylase, and PagP, a palmitoyl transferase, both localized in the outer membrane. PhoPQ activation also increases the level of PmrD that stabilizes the active state of a second TCS, PmrAB, resulting in production of EptA and ArnT (also known as PmrK). The enzymes EptA and ArnT modify the terminal phosphates of lipid A with phosphoethanolamine (pEtN) and aminoarabinose (AraN), respectively (Figure 1, S1).^9–11 Addition of pEtN and AraN has been shown to confer antimicrobial resistance in various Gram-negative bacteria.^12,13 Recently a growing family of mobile colistin resistance (MCR) enzymes, analogous to EptA, have been discovered mainly in E. coli; however, mcr plasmids may be transferred between Gram-negative bacteria, resulting in functional expression of a phosphoethanolamine transferase that enables antibiotic resistance to polymyxins, such as colistin (polymyxin E).⁴

Figure 1.

Basic structure of lipid A. Most lipid A structures consist of a diglucosamine backbone with two terminal phosphates. These phosphates can be modified with aminoarabinose (AraN) and phosphoethanolamine (pEtN), especially in the presence of positively charged antimicrobials.

Currently, lipid A modifications are detected using MALDI-TOF MS; however, this type of analysis is semiquantitative. While quantitation of lipid A acyl chain distribution via gas chromatography through fatty acid methyl ester (FAME) nalysis is well-established,^14–18 methods for analysis of the lipid A backbone components and subsequent modifications are limited. In this study, we employed a derivatization scheme to convert the nonfatty acid lipid A components from polar molecules to nonpolar molecules suitable for analysis in the gas phase. We used silylation to derivatize and quantify the lipid A backbone components, glucosamine (GlcN) and inorganic phosphate (Pi), as well as the terminal phosphate modifications, ethanolamine (EtN) and aminoarabinose (AraN). Silylation alone resulted in difficult to interpret chromatograms due to the various isomeric forms of sugar molecules.^19–23 Methoximation prior to silylation, which converts the individual sugar molecules to methoxyimine open chain forms, resulted in a singular chromatographic peak for each sugar molecule.^20–24 This two-step derivatization scheme is commonly applied in GC-MS-based metabolomics studies,^21–25 but issues observed with the approach include the need for anhydrous conditions and the long delay time between batch derivatization and injection into the GC-MS instrument.^26,27 We overcame these issues by employing just-in-time derivatization, which utilizes an automated autosampler to deliver all derivatization reagents and inject samples as needed. This allows samples to remain capped throughout the derivatization process in order to preserve anhydrous conditions and avoids long delay times because samples are injected immediately upon finishing derivatization.

In this study, lipid A samples were isolated from various Gram-negative bacterial backgrounds and hydrolyzed prior to derivatization using an automated autosampler to avoid batch effects and increase reproducibility and sensitivity. We found increased levels of EtN and AraN (~1.5- to 3-fold) in Salmonella regulatory mutants that overexpress the PhoPQ and PmrAB two-component regulatory systems. Additionally, increased EtN and AraN levels were present in resistant isolates of Gram-negative bacteria when grown in the presence of colistin, a cationic antimicrobial peptide that targets the bacterial membrane via electrostatic interaction. Differential phosphate levels were also found when comparing bis- and monophosphorylated lipid A samples. Overall, this GC-MS-based assay can simultaneously quantitate the lipid A backbone components, GlcN and phosphate, in addition to the terminal phosphate modifications, EtN and AraN, all from a single sample.

EXPERIMENTAL SECTION

Bacterial Strains and Growth Conditions.

Bacterial strains (Table S1) included Salmonella enterica serovar Typhimurium regulatory mutants obtained from Samuel I. Miller (University of Washington, Seattle, WA). E. coli BORT (O18ac:K1:H7) was purchased from ATCC. Bacterial cultures (50 mL) of the strains were grown in lysogeny broth (LB) supplemented with 1 mM MgCl₂ at 37 °C with aeration and harvested in the stationary phase. Antibiotic resistant strains of Escherichia coli and Enterobacter were obtained as clinical isolates and were grown as described above in the presence or absence of 2 μg/mL colistin.

LPS Purification and Lipid A Isolation.

Lipid A Isolation for EtN, AraN, and GlcN Analysis.

Prior to LPS extraction, lipid A backbone modifications were assessed via reflectron negative ion mode MALDI-TOF MS using the fast lipid analysis technique (FLAT) extraction protocol.²⁸ LPS was then extracted using the TRI reagent (Millipore Sigma, Burlington, MA) via a rapid small-scale method as described.²⁹ Mild acid hydrolysis was then performed to hydrolyze the LPS into lipid A as described.²⁹

Lipid A Isolation for Phosphate Analysis.

To avoid phosphate carryover into GC-MS analysis from the TRI reagent, E. coli BORT LPS was isolated using the double hot phenol method as described³⁰ and derived into lipid A via mild acid hydrolysis. MPLA and 3D PHAD were purchased from Avanti Polar Lipids.

Trifluoroacetic acid (TFA) Hydrolysis of Lipid A Samples.

TFA hydrolysis was performed to liberate the individual moieties of the lipid A backbone. This hydrolysis method breaks the ester linkage connecting phosphate and ethanolamine in phosphoethanolamine-modified lipid A. Thus, EtN was quantified as a surrogate for phosphoethanolamine. Two identical lipid A samples were prepared to avoid degradation of AraN under TFA hydrolysis as reported previously.³¹ Each lipid A sample was divided into two 500 μg aliquots and dried by lyophilization within a glass vial. Each vial received 10 μL of 1 mM internal standard (2-amino-2-methyl-1,3-propanediol), which is of completely synthetic origin and therefore guaranteed to not be present in any biologic sample. Then 100 μL of 1 M TFA was added, and samples were heated at 90 °C. For the analysis of AraN, one vial was removed from heat after 30 min, whereas for analysis of all other analytes, the other vial was removed after 24 h. After heating, cationic replacement of TFA salts with HCl was performed to minimize column degradation. Briefly, 10 μL of 1 M HCl was added to the sample, vortexed, and incubated at room temperature for 5 min. The sample was then diluted to 1 mL with endotoxin-free water, flash frozen, and lyophilized. This process was repeated once more. The samples were stored under desiccation until ready for derivatization.

Standard Solutions.

d-Glucosamine hydrochloride (GlcN) (≥99%), ethanolamine hydrochloride (EtN) (≥99%), sodium phosphate monobasic (Pi) (≥99%), and 2-amino-2-methyl-1,3-propanediol (internal standard, IS) (≥99%) were all purchased from Sigma-Aldrich. Amino-arabinose (4-amino-4-deoxy-l-arabinose, AraN) was chemically synthesized as described.³²

Standard stocks were prepared at 1 mM using LC-MS grade Water (Sigma-Aldrich) and stored at −20 °C until use. Standard solutions containing 10 μL of 1 mM IS were prepared directly in autosampler vials, flash frozen, lyophilized, and stored capped under desiccation until ready for derivatization.

Derivatization Reagents.

Pyridine (≥99.9%) and methoxyamine hydrochloride (MeOX) (LiChropur grade) were purchased from Millipore Sigma. N-Methyl-N-(trimethylsilyl)-trifluoroacetamide with 1% trimethylsilyl chloride (MSTFA with 1% TMCS) in 1 mL ampules was purchased from Fischer Scientific, Waltham, MA. MeOX powder was resuspended in pyridine at 20 mg/mL in 8 mm amber GC vials, capped immediately, and vortexed until completely dissolved. MSTFA with 1% TMCS ampules were opened and transferred to 8 mm amber GC vials and capped immediately.

Derivatization.

Silylation is moisture sensitive;²⁷ therefore, to execute derivatization under anhydrous conditions, all samples remained capped throughout the derivatization process, with reagents being added using a liquid syringe.

Batch Derivatization.

Due to the long wait times before analysis, efforts to avoid sample degradation consisted of purging capped sample vials with nitrogen gas. To do so, a nitrogen gas balloon connected directly to a 20 mL syringe with a 28-gauge needle was plunged into the vial cap along with an empty 28-gauge needle and incubated for 5 min to displace any atmospheric air that may remain within the vial. Sample vials containing lyophilized standard or biologic hydrolyzed lipid A sample then received 50 μL of 20 mg/mL MeOX reagent and were incubated at 70 °C for 30 min followed by 50 μL of MSTFA with 1% TMCS and incubation at 50 °C for 1 h. Samples were then transferred to autosampler vials containing 150 μL glass inserts and analyzed via GC-MS as described below in the chromatography and analysis section.

Automated Derivatization.

Just-in-time derivatization was executed using a PAL 6000Plus autosampler (Shimadzu) that uses GCMS RealTime analysis software (Shimadzu). Sample vials were positioned accordingly on the autosampler along with reagent vials containing the 20 mg/mL MeOX in pyridine and MSTFA with 1% TMCS reagents. Samples received 50 μL of MeOX reagent and were incubated at 70 °C for 30 min under heated agitation (180 rpm) followed by 50 μL of MSTFA with 1% TMCS reagent and incubation at 50 °C for 1 h under heated agitation (180 rpm). Derivatization reagents were added using a 50 μL smart syringe. Upon completion of derivatization, samples were rested for 10 s at room temperature then injected into the GC-MS using a 10 μL smart syringe and analyzed via GC-MS as described below in the chromatography and analysis section. Derivatization was staggered so that samples received the exact same treatment whereby injection occurred upon completion of derivatization avoiding the long delay times that occur when performing batch derivatization.

Chromatography and Analysis.

GC-MS separation and detection was carried out on a Shimadzu GC-MS QP2010 Ultra equipped with a DB-5 ms (30 m × 0.25 mm i.d. × 0.25 μm film thickness, 5% phenyl–95% dimethylpolysiloxane) analytical column from Agilent. The GC-MS instrument was equipped with a split/splitless inlet. Inert splitless liners with a single taper were used. For analysis, 1 μL of sample was injected in splitless mode (220 °C, splitless time 0.5 min, septum purge flow 3 mL/min). The following GC temperature program was used: 70 °C (hold 1 min), 15 °C/min ramp to 135 °C, 2 °C/min ramp to 146 °C, 15 °C/min ramp to 200 °C, 2 °C/min ramp to 210 °C, 15 °C/min ramp to 300 °C. Hold at 300 °C for 5 min. Ultrahigh purity helium (UHP300) was used as a carrier gas maintained at a constant velocity of 40 cm/s. The transfer line to the MS was set to 250 °C, filament source to 250 °C, and quadrupole temperature to 150 °C. The electron ionization (EI) source was operated at 70 eV. Analytes were monitored using selected ion monitoring. For data evaluation, GCMS Postrun analysis software version 4.52 (Shimadzu) was utilized.

RESULTS AND DISCUSSION

Derivatization of Sugar Molecules of Interest.

Silylation is a common derivatization technique;³³ however, it poses a unique problem in that a given sugar molecule will generate several chromatographic peaks due to its multiple isomeric forms.^19–23 These isomeric forms arise from the presence of anomeric carbons in the ring structure, as well as the presence of both ring and open chain forms of the sugar molecule (Figure S2A). Multiple chromatographic peaks for a single analyte will lead to complicated chromatograms and reduced sensitivity making silylation alone a less favorable derivatization technique. We overcame this problem by using a two-step derivatization technique of methoximation prior to silylation. Because methoximation converts ketone and aldehyde groups into methoxyimine groups, sugar molecules are converted into solely the methoximated open chain form and give rise to a singular derivative after silylation (Figure S2B).^20–24 Initially, we sought to optimize the methoximation reaction using a GlcN standard, the sugar molecule that makes up the lipid A backbone. After evaluating various time and temperatures, we determined that temperature was the main driving factor for formation of the methoxyamine open chain derivative (Figure S3A,B). It was found that at 70 °C, the methoximation reaction was completed after 30 min, whereas at 37 °C the reaction did not approach completion until 90 min (Figure S3C).

A second complicating factor arises when silylation is used to derivatize molecules that contain a variety of functional groups with varying reactivity of their molecular bonds. In our assay, several analytes of interest (GlcN, AraN, and the IS) contained both hydroxyl and amino groups, with the hydroxyl groups being much more reactive toward silylation than the amino groups.^23,34 Additionally, the amino groups contain two protons, which can lead to either one or two N Si bond – formations.²³ Altogether, this can lead to the formation of multiple derivatives for a given silylated analyte, so it is necessary to determine the optimal silylation conditions which result in the greatest chromatographic peak height.

We subsequently optimized our assay to determine the derivatization conditions that resulted in the greatest chromatographic peak height for the molecules present in lipid A. For this analysis, we utilized two different methoximation conditions, 37 °C for 2 h and 70 °C for 30 min, along with various times and temperatures of silylation to determine the optimal combination of methoximation and silylation conditions. Under the various conditions, we evaluated the chromatographic peak height of GlcN and EtN standards as representatives for the sugar molecules and small molecules, respectively. Chromatographic peak height for the two analytes remained constant across the varying conditions (Figure S4) as reported previously when optimizing silylation conditions.²³ Because metabolomic studies using GC-MS Article 02: Untitled commonly employ silylation in the range of 40–50 °C for 1 h^20–24 and our results showed little difference between silylation at 37 and 70 °C, silylation at 50 °C for 1 h was chosen for our derivatization scheme.

Batch Processing versus Automated Processing of Samples.

Silylation of complex molecules is often achieved within 1–2 h when heated above 40 °C, as determined by chromatographic peaks present in a splitless injection. However, it is known that silylation of sugars is often incomplete within this period even when these chromatographic peaks are observed.²⁶ Therefore, a major drawback of performing silylation in large batches is that the optimized derivatization time and temperature is distorted by long wait times on the autosampler prior to analysis. While the first sample is injected immediately after the optimized silylation conditions are applied, the subsequent samples are left on the autosampler to further undergo silylation at room temperature before injection, which could be several hours later depending on the batch size and analysis time. To overcome this hurdle, we employed just-in-time derivatization using an automated autosampler that performs derivatization in staggered intervals, whereby samples are injected immediately after completion of derivatization.

The application of just-in-time derivatization had a significant impact on two chromatographic factors: (1) reproducibility of the IS and (2) chromatographic peak signal intensity of the sugar molecules. Regarding reproducibility of the IS, we found that silylation resulted in either two or three trimethylsilyl (TMS) bond formations, resulting in chromatographic peaks for both the 2TMS and 3TMS derivatives. It is known that the affinity for silylation is much higher for hydroxyl groups than for amino groups; therefore, because our IS contained two hydroxyl groups and an amino group, we found that using batch derivatization resulted in a shift in IS peak profile from the 2TMS derivative comprising only silylated hydroxyl groups to the 3TMS derivative comprising an additional silylated amino group as samples were injected at increasing time points from when they were initially derivatized (Figure S5A). However, when using just-in-time derivatization, we found that the IS peak profile remained constant with a larger peak for the 2TMS derivative as compared to the 3TMS derivative (Figure S5B) and that the standard deviation in the combined peak height of the 2TMS and 3TMS IS peaks was reduced by 6-fold (Figure S5C). This suggested that sample injection immediately after derivatization resulted in consistent signals for both the 2TMS and 3TMS IS derivatives, whereas long delay times before injection caused a shift from the 2TMS to 3TMS derivative.

Additionally, it was found that the wait time between derivatization and injection greatly affected the chromatographic peak intensity of sugar molecules as well. Using a GlcN standard, we found that reducing the wait time between derivatization and injection from 5 min to 30 s increased GlcN peak height by 11-fold (Figure S6). This may be due to the instability of the silylated amino group, because Si–N bonds Altogether, these data are more labile than Si–O bonds.²³ showed that use of just-in-time derivatization results in increased precision and in some cases increased sensitivity, when using silylation to derivatize complex molecules containing more than one functional group with varying affinity for silylation.

Chromatographic Separation and Detection.

Because it was unknown that neighboring peaks in biologic samples might overlap with our analytes of interest, chromatographic separation was optimized to include a slow ramp of 2 °C/min to encompass the analyte retention times to ensure adequate separation. To identify characteristic fragment ions for each analyte, full profile EI spectra for individual standards were generated by scanning m/z 35–500 (Figure S7). Characteristic fragment ions from the full profile EI spectra were identified and used for selected ion monitoring (SIM). Under the derivatization conditions, all analytes produced a singular derivative except for the IS, which produced a 2TMS and 3TMS derivative. The 2TMS derivative produced a characteristic fragment ion at an m/z of 146 whereas the 3TMS derivative at an m/z of 218. Thus, these two values were used to monitor the presence of the IS peaks. For the nonsugar molecules, EtN and Pi, m/z values of 174 and 299 were used to monitor their presence, respectively. For the sugar molecules, GlcN and AraN, several characteristic fragment ions were observed from the full profile EI spectra. The ions with the greatest abundance were used to detect their presence, therefore m/z values of 159 and 204 were used to monitor the presence of GlcN and AraN, respectively. While monitoring m/z 204 in biologic samples, a neighboring peak was close in retention time to the AraN peak, even when a ramp of 2 °C/min was applied to enhance separation. To avoid any obscuring of peak area, peak height was used for quantification, because the neighboring peak did not affect the maximal peak height for the AraN analyte. For each analyte, an additional characteristic fragment ion was monitored as a qualifier to ensure proper identification (Table 1).

Table 1.

Chromatographic Detection of Analytes of Interest

analyte	quantifier ion(m/z)	qualifier ion(m/z)	LoD
EtN	174	262	100 nM
Pi	299	314	100 nM
GlcN	159	319	1 μM
AraN	204	160	1 μM

For all analytes, standard curves were generated with 2 > 0.99). The limit of detection (LoD) excellent linearity (R for small molecules, EtN and Pi, was better than that of the sugars, likely due to the instability of derivatized sugar molecules and the fact that derivatization of sugars is known to be incomplete. EtN and Pi both had LoD values in the nanomolar range, whereas the LoD for GlcN and AraN was in the micromolar range (Table 1).

Hydrolysis Conditions To Liberate Analytes from Lipid A.

Because we sought to analyze individual components from a single lipid A sample, we utilized acid hydrolysis to release the molecules of interest from the lipid A structure prior to GC-MS analysis. A prior study by Kalhorn et al. reported the optimized TFA-mediated hydrolysis conditions to liberate the individual components of lipid A.³¹ Using those same conditions, we confirmed that hydrolysis of AraN required 30 min, whereas a longer hydrolysis time of 24 h results in its degradation (Figure S8A). Additionally, we confirmed that hydrolysis of the GlcN backbone requires a long 24-h period at an elevated temperature (Figure S8B). This is likely due to the strength of the β(1–6) glycosidic linkage connecting the diglucosamine lipid A backbone. Additionally, it was confirmed that all other analytes, including the IS, were stable under the 24-h TFA hydrolysis period as well. We found that addition of the IS before TFA hydrolysis compared to after hydrolysis resulted in increased peak reproducibility (Figure S9). Thus, for lipid A sample analysis, each sample was divided into two vials to undergo TFA hydrolysis, derivatization, and GC-MS analysis. One sample underwent TFA hydrolysis for 30 min and a second for 24 h to analyze AraN and all the other analytes, respectively.

Analysis of EtN and AraN Levels in Salmonella Regulatory Mutants.

To demonstrate the utility of this assay to quantitate lipid A terminal phosphate modifications, we assessed EtN and AraN levels in Salmonella regulatory mutants³¹ with modified PhoPQ or PmrAB TCS systems. These systems either constitutively expressed or contained null mutations in essential TCS genes. Currently, detection of lipid A modification uses MALDI-TOF MS which can be achieved from either a colony or bacterial pellet using a method called fast lipid analysis technique (FLAT).²⁸ However, the short-coming of this method is that it is qualitative and not quantitative; thus, only the presence or absence of lipid A modifications can be assigned. In this study, each bacterial strain/mutant was initially screened by FLAT to determine which lipid A modifications could be readily detected. The expected lipid A modifications⁵ were observed (Figure S10A), whereby downstream effectors, such as PagL and PagP from PhoPQ activation, resulted in deacylation and acylation (m/z 1570 and 2035, respectively) and EptA and ArnT from PmrAB activation resulted in pEtN and AraN addition (m/z 1920 and 1928, respectively). When two essential biosynthetic genes required for AraN addition (pmrE and pmrF) were inactivated, this modification was no longer observed, while pEtN modification remained present (Figure S10B).

Using the GC-MS assay described herein, we quantified differences in EtN and AraN levels across the various Salmonella regulatory mutants. Quantification of EtN revealed a basal level of EtN addition in the WT strain (something not observed via MALDI-TOF MS) along with a 1.6-fold increase in EtN modification upon PhoPQ activation. PmrAB activation led to a significant increase in EtN and deletion of the genes responsible for AraN addition appeared to alter the elevated levels of EtN. Quantification of AraN revealed a basal level of AraN addition in the WT strain (something not observed via MALDI-TOF MS). PhoPQ activation led to a 3.5-fold increase in AraN addition, while deletion of PhoP abolished AraN addition. This was expected because PhoPQ activation in turn stabilizes the activated state of PmrAB, which is responsible for executing AraN addition. PmrAB activation led to a 2.8-fold increase in AraN addition, whereas deletion of the genes responsible for ArnT production ablated AraN addition (Figure 2). A large standard deviation in AraN levels in the PmrA^c strain led to the average fold increase in AraN levels from the WT being lower in this strain compared to PhoP^c, suggesting that constant PmrAB activation can lead to highly variable levels of AraN addition. These data not only provide quantitative values to the trends observed via MALDI-TOF MS but also revealed lipid A modification in samples where no observable ion was detected via MALDI-TOF MS. Altogether, this provides a new avenue for investigating sensitive changes in lipid A modifications that may not be observable by the current approaches.

Figure 2.

Analysis of EtN and AraN levels in Salmonella regulatory mutants. (A) Representative chromatograms of the various Salmonella regulatory mutants hydrolyzed lipid A samples after derivatization and GC-MS analysis using selected ion monitoring (SIM) at the respective m/z for EtN, AraN, and GlcN. (B) Quantification of the levels of EtN and AraN for the various Salmonella regulatory mutants shown as a ratio to GlcN (n = 3). Statistical significance was determined by a two-way ANOVA: P values are denoted as *, **, and *** for values ≤0.05, ≤ 0.01, and ≤0.0001, respectively.

To validate the reproducibility of this assay, we quantified lipid A samples extracted from the WT, PhoP^c, and PmrA^c strains on three separate occasions as representatives of samples with low, medium, and high levels of EtN and AraN (Table 2). The standard deviation of the sugar analytes, AraN and GlcN, were higher than that of EtN. For AraN, this may be attributed to variable levels of this modification across biological replicates whereas for GlcN, this may be due to altered levels of lipid A present in each lyophilized sample prior to hydrolysis and derivatization. This emphasized the importance of utilizing the ratio of EtN or AraN levels to GlcN levels for the final analysis of lipid A modification as shown in Figure 2. By doing so we ensured that the quantified EtN and AraN levels were correlated to the level of modification of the intact lipid A molecule because the GlcN molecules originate from the lipid A diglucosamine backbone and serve to standardize the exact amount of lipid A that was present in the analysis.

Table 2.

Repeatability in Measurements of Independent Lipid A Samples

sample	EtN (μM)	AraN (μM)	GlcN (μM)
S. typhimurium WT	11.69 ± 3.22	7.66 ± 7.171	48.45 ± 16.81
S. typhimurium PhoPc	20.55 ± 1.98	35.90 ± 12.19	58.73 ± 16.46
S. typhimurium PmrAc	32.05 ± 1.07	27.86 ± 22.03	51.88 ± 3.63

Analysis of EtN and AraN Levels in Antibiotic Resistant Strains of Gram-Negative Bacteria.

The utility of this assay to quantitate lipid A terminal phosphate modifications, EtN and AraN, proves particularly useful regarding antimicrobial resistance. In this study, we utilized two Gram-negative bacterial isolates: 1) an E. coli isolate that contained the mcr-1 plasmid known to confer polymyxin resistance by addition of pEtN and 2) an antibiotic resistant Enterobacter isolate that induces AraN addition in the presence of colistin. Currently, we are only able to assign the presence or absence of these modifications by MALDI-TOF MS;^12,35 however, this assay enabled quantitation of these modifications when grown in the absence or presence of colistin.

While MALDI-TOF MS analysis revealed the presence of lipid A modifications (Figure S11), we found that when grown with colistin, levels of EtN increased by 7-fold for the E. coli mcr-1-expressing isolate and AraN levels went from undetectable to detectable in the Enterobacter isolate (Figure 3). This provides further utility for this assay, warranting future studies that investigate the correlation between the level of lipid A terminal phosphate modification and the degree of antibiotic resistance to polymyxins.

Figure 3.

Analysis of EtN and AraN levels in antimicrobial resistant isolates. (A) Representative chromatograms of the E. coli mcr-1 and Enterobacter isolates hydrolyzed lipid A samples after derivatization and GC-MS analysis using selected ion monitoring (SIM) at the respective m/z for EtN, AraN, and GlcN. Quantification of the levels of EtN and AraN for the (B) E. coli mcr-1 isolate and (C) Enterobacter isolate when grown in the absence or presence of colistin shown as a ratio to GlcN (n = 3).

Analysis of Differentially Phosphorylated Lipid A Samples.

Lipid A structures are highly variable across Gram-negative bacteria and can contain one, two, or three phosphate groups attached to the glucosamine backbone.¹⁰ Bis-phosphorylated lipid A structures are the most widely recognized; however, discovery of mono- and triphosphorylated lipid A structures revealed specific enzymes that can modify the backbone phosphate moieties. For instance, lipid A triphosphorylation by an undecaprenyl phosphotransferase, LpxT, adds an additional phosphate group shown to impact polymyxin resistance.³⁶ Alternatively, dephosphorylation of lipid A via phosphatases LpxF and LpxE has also been implicated in polymyxin resistance in Gram-negative bacteria.^37,38 Introducing one or a combination of these enzymes into Gram-negative bacteria can effectively reprogram the lipid A biosynthetic pathway to generate lipid A structural variants within a strain of interest.^39–41 One such application in an avirulent strain of Yersinia pestis led to the discovery of lipid A-based adjuvants that resemble the FDA-approved mono-phosphoryl lipid A (MPLA) molecule used as an adjuvant in vaccines, such as cervarix.^30,42

We observed derivatized phosphate in our biological samples, stemming from the terminal phosphates on the lipid A. This spurred an investigation into whether this assay could discriminate against differentially phosphorylated lipid A samples. To do so, phosphate levels from a bis-phosphorylated E. coli lipid A sample (Figure 4A) were compared with those of monophosphorylated lipid A samples (Figure 4B), MPLA and PHAD (a synthetic analogue of MPLA), which are two lipid A-based vaccine adjuvants.^42–47 We found that those with monophosphorylated lipid A structures had lower levels of phosphate compared to the bis-phosphorylated lipid A sample from E. coli (Figure 4C,D). This suggested that this assay can serve not only as a tool for analyzing terminal phosphate modifications but also to assess the level of lipid A phosphorylation itself as well. While thin layer chromatography (TLC) and MALDI-TOF MS are currently used to monitor lipid A phosphate addition or removal,³⁶ this assay provides another avenue for future studies that investigate the discrimination between lipid A mono-, bis-, and triphosphorylation.

Figure 4.

Analysis of Pi levels in differentially phosphorylated lipid A samples. The lipid A structures of (A) E. coli and (B) MPLA and PHAD (a synthetic analogue of MPLA). (C) Representative chromatograms of the hydrolyzed lipid A samples after derivatization and GC-MS analysis using selected ion monitoring (SIM) at m/z 299 for Pi analysis. (D) Quantification of the levels of Pi using the described GC-MS assay (n = 3).

CONCLUSIONS

This study describes a newly developed GC-MS method that can quantitively determine lipid A terminal phosphate modifications such as EtN and AraN, which are known correlates of polymyxin resistance in Gram-negative bacteria. As antibiotic resistance among Gram-negative bacteria increases, lipid A serves as a valuable target for the development of therapeutics, particularly those that can prevent modification of lipid A with EtN and AraN. Until now, development of such therapeutics has been hindered by the lack of tools to quantitate such modifications. This is the first study to utilize GC-MS for quantitation, which is significant because another common method of lipid A structural analysis is by FAME analysis which is performed via gas chromatography. The ability to determine lipid A fatty acid composition via FAME analysis as well as terminal phosphate modification using a single GC-MS instrument now provides that capacity to gain complete lipid A structural characterization in an efficient manner. Because lipid A serves a variety of functions, such as vaccine adjuvants^42–47 and diagnostic markers for bacterial infections,⁴⁸ this assay provides a new opportunity for its characterization. Additionally, this assay will aid in the development of novel antimicrobial compounds that target lipid A modification by providing precise quantitation of the EtN and AraN levels in their presence. Ultimately, this newly developed GC-MS method enables quantitative characterization of the lipid A backbone characteristics, from phosphate level to AraN or EtN modification.