Recent advances in the mass spectrometric profiling of bacterial lipids

Abstract

Exploring the lipids of bacteria presents a predicament that may not be broadly recognized in a field dominated by the biology and biochemistry of eukaryotic— and especially, mammalian— lipids. Bacteria make multifarious metabolites that contain fatty acyl chains of unusual length and unsaturation attached to assorted headgroups, including sugars and fatty alcohols. Lipid profiling approaches developed for eukaryotic lipids often fail to detect, resolve, or identify bacterial lipids due to their wide range of polarities (including very hydrophobic species) and diverse positional and stereochemical variations. Global lipid profiling, or lipidomics, of bacteria has thus developed as a separate mission with methodological and scientific considerations tailored to the biology of these organisms. In this review we summarize findings primarily from the last three years that exemplify recent advances and continuing challenges to learning about bacterial lipids.

Introduction

The structural diversity of lipids and their non-templated synthesis provide significant challenges for their detection, identification, and quantification. Mass spectrometry has become a method of choice for meeting many of these challenges, thanks to advances in instrumentation and methodology. This has led to the rise of lipidomics as a category of metabolomics. The infrastructure for lipidomics—methods, tools, and databases—has arisen mostly around profiling eukaryotic lipids. Many of the questions surrounding lipid membrane composition and biosynthesis are the same for bacteria, but what answers we have so far make it clear that bacteria have divergent pathways and enzymes. a [1,2]. Bacteria-specific classes and modifications include the widely studied lipopolysaccharide component of gram-negative outer membranes and multiple modes of fatty acyl unsaturation, including methyl branching and cyclopropanation. These distinctions necessitate developing bacterial lipidomics as a distinct pursuit essential to tackling these problems—to discovering new metabolites, (re)assigning enzyme functions, and characterizing metabolic responses. In this review we highlight studies, especially those published on bacterial pathogens in the last three years, that exemplify exciting advances and persistent challenges in exploring bacterial lipids and their biological roles.

Innovations in Lipid Profiling Methodology

Recent developments in profiling bacterial lipids have concerned adapting approaches originally directed at other organisms or analytical research areas. Particularly relevant innovations concern separation and dissociation, towards lipid (sub)class discrimination and structural determination, and spatial resolution, towards resolving subcellular distribution (Figure 1). Improvements in preparation and dissociation are particularly relevant to resolving isobaric species. One example, which we also discuss further below, is the challenge of distinguishing bacterial lipids that bear mixtures of short-chain fatty acids (SCFAs) and isobaric branched-chain fatty acids (BCFAs), with implications for membrane fluidity based on the ratios of these modifications. Below we discuss studies that focus on methods; in later sections we detail their application to particular biological questions.

Figure 1. Methodological innovations recently adapted to the identification and spatial resolution of bacterial lipids.

Hydrophilic interaction lipid chromatography (HILIC) combined with ion mobility separation (IMS) has improved separation by lipid class and alkyl chain length and modification. High-energy UV photodissociation has enabled in situ determination of the position of unsaturation (e.g., cyclopropanation) in lipid/fatty acid chains. In secondary ion mass spectrometry (SIMS) “soft” ion sources like C₆₀ beams have allowed molecular depth profiling in addition to lateral ion imaging of individual bacterial cells at micron resolution. Figure created using BioRender.com

Identification: Improvements in separation and dissociation

In ion mobility separation (IMS), ions in the gas phase are distinguished by size, shape, and charge via their characteristic drift time through a charged tube. The key IMS parameter is the collisional cross section (CCS), which relates the applied voltage to the drift time of a given ion. IMS has been used heavily in commercial analytical applications and has recently gained greater traction in lipid analysis due to the speed of separation (on the range of ms), technically straightforward integration with existing methods (e.g., in combination with liquid chromatography), and potential to resolve isobaric or co-eluting species.

IMS in lipidomics has been covered in detail elsewhere, including recently in this journal [2,3]; here and in later sections we focus on applications to eubacteria. The Xu and Hines labs in particular have pioneered IMS for bacterial lipids primarily via integration with hydrophilic interaction lipid chromatography (HILIC) [4–6]. While common reverse-phase chromatography separates lipids based primarily on alkyl chain hydrophobicity, HILIC separates by headgroup polarity, leading to discrimination by lipid class. In applying HILIC-IM-MS to E. coli lipids, Hines and Xu provided additional CCSs for higher-order lipid species such as triacylated phospholipids and cardiolipins, thus providing additional standard parameters for more comprehensive bacterial lipid profiling via multi-dimensional retention time-drift time-accurate mass analysis [5]. In comparing strains defective for phosphatidylglycerol (PG) or cardiolipin (CL) synthesis genes, they noted specific decreases in 33:1 and 35:1 phosphatidaylethanolamine (PE) that contain presumed cyclopropanated 17:1 and 19:1 fatty acids. These acyl chain-specific changes particular to PE demonstrate the detailed structural discrimination afforded by their hybrid method. Most recently, Freeman et al. have developed a reverse-phase IM hybrid method to resolve intact lipids from Staphylococcus aureus containing straight-chain vs. branched-chain fatty acids, importantly, with higher sensitivity than fragmentation approaches and without the loss of information introduced by hydrolysis [7]. Obstacles to the broader adoption of IMS are common to metabolomics in general: The need for more validation with standards and for methods and databases tailored to (bacterial) lipids. Recent efforts to address the latter include the creation of multi-dimensional libraries that combine CCS, retention time, accurate mass, and fragmentation data [8].

Structural characterization of lipids by mass spectrometry is another significant ongoing challenge. De novo structural determination of novel lipids typically requires laborious isolation of amounts sufficient for multi-dimensional NMR. Improvements in MS instrumentation have enabled the detection of low-abundance species; additional methods for in-line dissociation would facilitate the structural definition of these ions of interest. Multi-dimensional MS using widely available collision methods reveal head group and acyl chain composition, but these low-energy methods often cannot provide stereochemical or positional information. Biological information about, for example, lipid structural changes in response to environmental cues, is therefore lost.

Recent reviews detail efforts to fill this gap via higher-energy dissociation techniques. Here we summarize newer efforts to analyze intact lipids (i.e., without prior decomposition or chemical modification). The Brodbelt group and others have explored ultraviolet photodissociation (UVPD), with recent validation using 213 nm irradiation based on a new commercial UVPD ion trap spectrometer (OrbiTrap) [9]. For bacterial lipids, UVPD alone and in combination with collisional methods have uncovered the positions of cyclopropanation in E. coli glycerophospholipids and mycobacterial mycolic acids [10]. UVPD yielded double cross-ring cleavage and a resulting characteristic doublet separated by 14 Da (Figure 2). Comparison between species revealed greater variation in the position of the cyclopropane along the fatty acid chain in Mycobacterium tuberculosis vs. M. bovis. Notably, UVPD also enabled de novo structural characterization of intact lipooligosaccharides from a drug-resistant strain of the opportunistic pathogen Acinetobacter baumanii [11]. Not only were the acyl modifications of the lipid A substructures determined, but the components and connectivity for the outer core of the oligosaccharide substructure was deduced, revealing a linear chain of one HexNAc and two Hex saccharides.

Figure 2. Ultraviolet dissociation of α-mycolic acid from mycobacteria reveals the positions of the cyclopropane modifications.

A negative mode 213 nm UVPD mass spectrum with corresponding fragment ion map highlighting the characteristic doublet resulting from cleavage across the cyclopropane ring. Figure reproduced with permission from [10].

Although clearly promising, UVPD and other high-energy collisional methods yield complex signatures that require laborious spectral analysis; in addition, definitive assignment requires resolution of isobaric species via integration with existing separation protocols, as was recently done to compare the positions of unsaturation in E. coli and A. baumannii glycerophospholipids [12]. Overall, there remains a major need for appropriate workflows to promote wider application.

Localization: Improvements in imaging mass spectrometry

Determining the (subcellular) localization of lipids in bacteria is of high interest for two major reasons: To answer questions about (1) outer membrane biogenesis, which includes transport of chemically indistinguishable molecules (vs. biosynthesis, which involves stepwise chemical transformations), and (2) the transfer and transformation of lipids between bacteria and their environment (e.g., bacterial pathogens and their hosts). The ability to profile lipids as a function of subcellular localization would provide unprecedented insight into cell envelope biogenesis— and cell growth in general— and into metabolic responses to environmental cues relevant to pathogenesis, antibiotic tolerance, and other biomedically related phenomena.

Spatially resolved MS, such as imaging matrix-assisted laser desorption ionization (MALDI), desorption electrospray ionization (DESI), and secondary ion mass spectrometry (SIMS) techniques have been widely applied to cellular populations such as bacterial colonies and films. Recent reviews cover the range of imaging MS approaches and their biological applications [13,14]. In general these methods provide lateral information via scanning ionization beams. SIMs, for example, analyzes secondary ions generated by directing a focused high-energy primary ion source at a surface under vacuum. For lipid profiling applications, two important considerations are the lateral resolution and yield of intact molecular secondary ions diagnostic for lipids. These factors are determined by the primary ion source, which can range from gaseous elements (e.g., ⁴⁰Ar⁺, Xe⁺) to liquid metal (e.g., Au₃⁺, Bi₃⁺) to molecules (e.g., SF₆, C₆₀); applications of all three have proved productive for lipid imaging in cells and tissues [14].

Spatially resolved work in bacteria has concerned primarily targeted detection of hydrophilic metabolites and xenobiotics, with the exception of secreted rhamnolipids in biofilms [15–19]. Cluster ion beam sources, specifically C₆₀ fullerene, have also enabled dynamic secondary ion mass spectrometry, also known as molecular depth profiling (Figure 1). The cluster beam effectively desorbs molecular ions while limiting sub-surface damage, thereby allowing ablation of a top layer and accurate interrogation of the underlying layer (in contrast to the fragment ions generated by more tightly focused, but higher energy, atomic beams). A cluster beam thus can be used to generate subcellular information even from multicellular populations.

An early endeavor in this area used TOF-SIMS to contrast the depth profiles of two secondary metabolite antibiotics in Streptomyces coelicolor cellular aggregates at 10–100 nm depth resolution [15]. A more recent study leveraged the advantages of the C₆₀ cluster beam and reported ~200–300 nm resolution in both lateral and depth dimensions for the detection of ampicillin and tetracycline within single treated E. coli cells [17]. These exploratory studies focused on antibiotics and did not attempt to detect native metabolites. However, these methods could hopefully be turned towards profiling applications, as exemplified by the recent work of Li, Balan, and Vertes [20]. While at far lower depth resolution (~100 uM), this study provided preliminary assignment of metabolites, including 40 lipids belonging to major sub-classes of PE, PG, phosphatidylserine (PS) and free FA, from live colonies on agar and from cell pellets.

Recent Advances in Understanding Bacterial Biology Enabled by Lipidomics

Lipids are critical mediators of bacterial pathogenicity [21]. Bacterial lipids can trigger host immune responses, but also enable immune evasion, as reviewed in [22–24]. Lipids are dynamic in structure and relative abundance as bacteria adapt to their environment: Altered lipid metabolism can promote membrane integrity and help bacteria resist host- or xenobiotic-mediated stress [25,26]. Below we review how lipidomics has forwarded our increasingly detailed understanding of lipid identity and composition in bacteria.

Discovery of novel metabolites and the reassignment of lipid biosynthetic pathways

Lipopolysaccharide (LPS) has long dominated the study of lipids in gram negative bacterial virulence. However, recent profiling of Salmonella enterica, which causes typhoid, revealed two abundant lipids derived from trehalose, 6-phosphatidyltrehalose (PT) and 6,6′-diphosphatidyltrehalose (diPT) (Figure 3A) [27]. This discovery was enabled in part by a ground-breaking normal phase LC-MS method originally developed to profile mycobacterial lipids [28]. Quantitative comparison among clinical strains showed that these novel trehalose lipids predominate in pathogenic serovars. Moreover, DiPT activated immune cells via Mincle, a surface receptor that is also activated by a related mycobacterial lipid, trehalose dimycolate. While trehalose dimycolate is a recognized virulence factor in pathogenic mycobacteria, this is the first study to identify 6,6’-modified trehaloses outside of the Corynebacterineae. The association of (di)PT lipids with pathogenic serovars and their activation of Mincle suggest a novel role for lipids other than LPS in Salmonella pathogenesis. Indeed, the presence of (di)PTs provides a possible explanation for the failure of anti-LPS therapies in typhoid and a possible association of these lipids with symptomatic presentation. In cell-based assays, DiPT activates macrophages via the immune receptor Mincle and is thus a potential immune adjuvant. The discovery of these trehalose lipids in S. typhi also led to the reassignment of a ClsB homologue previously associated with cardiolipin synthesis. Lipidomic profiling of null mutants showed that clsB is instead involved in (di)PT biosynthesis. Since the predicted substrates of ClsB would be PG and thus not clearly related to possible precursors for these novel trehalose metabolites, unbiased chemotyping by lipidomics was essential to assigning S. typhi ClsB to this pathway.

Figure 3. Novel bacterial trehalose lipids discovered by global lipidomics.

A) Normal-phase LC-MS profiling enabled the detection and identification of trehalose lipids in Salmonella typhi [20]. The cyclopropyl chain length position was not determined directly, but corresponds with known structures. The stereochemistry is also unknown. B) The sensitivity of MS detection led to the recognition of novel modifications to trehalose mycolate lipids in Corynebacterium glutamicum [22]. An additional 6′ straight-chain fatty acyl modification (Acyl; red) was found on hydroxy and acetyl trehalose monocorynemycolates (hTMCM and AcTMCM). The acetyl modification (Ac; blue) had previously been detected only on trehalose with one mycolate, but in Klatt et al. was also found also on hydroxy trehalose dicorynemycolate (hTDCM). Various lengths for both the straight-chain fatty acyl and corynemycolate chains were detected; representative C32:0/16:0 structures are shown.

Even well-studied trehalose mycolate lipids have been the source of recent surprises thanks to lipid profiling. Global lipidomics of Corynebacterium glutamicum identified new subclasses of trehalose corynomycolates [29]. Identification of these low abundance subclasses were possible only via the sensitivity and in situ structural determination afforded by multi-stage mass spectrometry. These new minor components sport an additional acyl group or acetyl modification (Figure 3B); the latter is a hypothesized means of regulating the transport of intermediates across the inner membrane. The detection of these unexpected derivatives suggest that biosynthesis is more complicated, with possible additional regulatory mechanisms or dead-end modifications.

Another major outcome was the localization of lipid classes to the inner or outer membranes of C. gutamicum by comparing successive extracts of whole cells, using water-saturated 1-butanol as a selective solvent for outer membrane lipids. In this study targeted quantitative analysis revealed expected enrichment of biosynthetic intermediates in the inner membrane fraction and final products in the outer membrane, thereby validating the approach. A mutant lacking the inner membrane trehalose corynomycolate transporter TmaT revealed a surprising depletion of phosphatidylglycerols and accumulation of triacylglycerols in the inner membrane fraction compared to the wild type. These data further support a link between triacylglycerol degradation and mycolic acid biosynthesis, as previously described in closely related mycobacteria. Localizing lipids to specific subcellular locations is integral to understanding lipid biosynthesis and transport, but has proved technically challenging. This work in C. glutamicum provides important validation and reference data for future studies.

Lipidomics was recently integral to uncovering a biosynthetic anomaly in several Streptococcus species [30]. Lipoteichoic acids (LTA) are a class of lipids located at the outer leaflet of the plasma membrane of gram positive bacteria and comprise five types (I-V). Streptococcus mitis in particular presented a conundrum, as it has type I LTA biosynthetic machinery, but apparently produces only type IV LTA. The sensitivity of unbiased mass spectrometric profiling revealed intermediates for both type I and type IV LTAs that were undetectable by previous thin-layer chromatography or antibody-based approaches. Lipid profiling also revealed that although type I LTA is derived from phosphatidylglycerol as expected, synthesis in S. mitis does not depend on the canonical type I LTA synthase encoded by ltaS. Overall, this work uncovered novel LtaS-independent synthesis of type I LTA in S. mitis and two other Streptococcal species.

Comparative profiling of genetically modified strains has also been deployed in an effort to understand altered biology. MTBVAC is a live attenuated tuberculosis vaccine candidate derived from M. tuberculosis and containing deletions in the phoP and fadD26 genes, which function in lipid biosynthesis and metabolism. A recent study aimed to identify global metabolic alterations arising from these modifications on the hypothesis that such changes are related to attenuation and will inform the identification of vaccination biomarkers [31]. Uniquely, this work deployed reverse-phase and HILIC separations separately to increase coverage of mycobacterial lipids, which encompass a particularly challenging breadth of polarities. The authors speculated that elevated levels of phosphatidylinositol mannoside lipids in MTBVAC vs. the wild type may underlie the sustained immunogenicity of MTBVAC. However, 70% of the detected molecular events could not be assigned to known metabolites, underscoring a crucial and enduring obstacle to maximizing bacterial lipidomic data: the high proportion of uncharacterized species.

Profiling of MTBVAC may inform our understanding of how changes in this strain affect host responses. On the other hand, profiling has also tackled the converse: How the host affects bacterial lipids? Comparison of Staphylococcus aureus lipids by HILIC-IM-MS after culturing in laboratory medium vs. human serum revealed that serum lipids enter the cell envelope and are metabolized by the bacteria [32]. Notably, confirming membrane incorporation rather than nonspecific association of serum lipids was a laborious but crucial validation and an important consideration for any studies examining the impact of host lipids on bacterial membrane composition. Overall, S. aureus metabolized free oleic acid and cholesterol esters from human serum into straight chain unsaturated fatty acids and esterified them into phospholipids. Growth on human serum also stimulated an increase in unsaturation in cardiolipin; tandem MS/MS provided specific information on the position and chain length of the acyl modifications. The authors speculate that these class-specific changes in fatty acid composition could be important for immune evasion and antibiotic resistance in vivo.

Towards understanding how the bacterial lipidome responds to the environment

Examining bacterial responses to host-related stresses is an important tool in understanding host-microbe interactions and bacterial pathogenesis. Changes in RNA transcripts and proteins in response to environmental cues or genetic alterations have been extensively documented thanks to advances in sample preparation and instrumentation that made high-throughput studies technically and economically feasible. Analogous screens to track changes in the lipidome are stymied by the prospect of innumerable solvent extractions and lengthy LC-MS methods on the order of tens of minutes.

A recent report in E. coli improved the outlook for high-throughput lipidomics by streamlining extraction and profiling: They applied a one-step extraction in glass-coated plates and a fast 4-min HILIC LC method to 113 single-gene overexpression and deletion mutants for lipid metabolism enzymes [33]. Notable findings included the particularly distinct lipidome of the cyclopropane-fatty-acyl-phospholipid synthase cfa deletion strain. While the decrease in fatty acid cyclopropanation in this strain was expected, changes in abundance of entire lipid classes was more surprising, especially the accumulation of cardiolipin. This suggests an unexpected preference of the CL synthase for straight-chain vs. cyclopropanated fatty acids. More generally, these results exemplify the unexpectedly subtle and specific changes to the lipidome that mass spectrometry reveals through lipid class-specific information. In this example and others that follow, changes to fatty acid composition do not necessarily affect all lipid classes equally, with implications for metabolic and membrane adaptations to environmental cues.

This same study also examined lipidomic alterations in response to environmental changes, specifically the effects of short-chain alcohols, as relevant to E. coli engineered to produce and export these industrially useful starting materials. Across their E. coli mutant library, Cfa-overexpressing strains had the most distinct lipidome in response to culturing on alcohols. PCA analysis also suggested a correlation between alcohol chain length and phospholipid carbon number. The authors created a model to calculate lipid species abundance based on these outstanding factors, including Cfa expression level and the chain lengths of both alcohol and lipid. This work hints at the predictive capabilities that comprehensive lipid datasets may enable, but also the investment in analytical tools and models that will be required to capture the full depth of the newly acquired information.

An intriguing alternative to streamlining MS analysis is to use a streamlined organism. Chwatsek et al. took this approach in studying the gram negative bacterium Methylobacterium extorquens, which has only 25 phospholipid species (vs. ~100 in E. coli) and is thus amenable to profiling by shotgun lipidomics [34]. Among other notable observations, phosphatidycholine and phosphatidylethanolamine showed the greatest variability in acyl chain length and saturation in response to any stress, whereas these same attributes changed very little for phosphatidylglycerol. Such class-specific alterations have also been observed in yeast [35] and thus further emphasize that lipid class-specific acyl chain remodeling is a general strategy for stress adaptation in prokaryotes and eukaryotes.

Revealing relationships between the bacterial lipidome and antibiotic resistance

Whole-genome sequencing of antibiotic-resistant strains and characterization of resistance-associated mutations in lipid biosynthetic genes have confirmed a role for lipids in modulating drug sensitivity. While mutations in particular biosynthetic or regulatory pathways have implicated specific lipid classes, the potential broader effects of resistance mutations on lipid metabolism and thus a more complete understanding of resistance mechanisms is only now beginning to emerge thanks to lipid profiling of resistant strains.

Daptomycin is an amphiphilic lipopeptide used against gram positive bacterial infections, but emerging resistance poses a major threat to its clinical use. Hines et al. reported multidimensional HILIC-IM-MS profiling that revealed alterations across the lipidome in three daptomycin susceptible-resistant pairs of E. faecalis, S. aureus and C. striatum [4]. This study exemplifies the use of ion mobility to resolve lipids, providing a faster alternative to traditional LC methods: 160 lipids species were correlated with their unique CCSs in a single <10-minute run. Notably, mutation of the lipid biosynthetic enzyme pgsA correlated with decreases in the levels of PGs, CLs, and amino acid-modified PGs, and increases in precursores of PG such as glycolipids and phosphatidic acids. These changes were concomitant with the acquisition of daptomycin resistance in both S. aureus and C. striatum, despite their quite distinct cell envelope architectures. The authors propose that buildup of neutral lipids (digalactosyldiacylglycerol (DGDGs), phosphatidylinositols (PIs), and glucuronosyldiacylglycero (GlcADGs)) resulting from pgsA loss of function contributes to the antibiotic resistance phenotype. Echoing a theme of this review, alterations in these lipid classes were fatty acid-specific, pointing to more subtle metabolic shifts than previously recognized and suggesting potential biomarkers for identifying specific resistance mechanisms. A follow-up study extended this work to an examination of how membrane PG levels correlate with antibiotic cross resistance in methicillin-resistant strains of S. aureus [36].

Several analogous studies have investigated lipid changes associated with resistance to another lipopeptide antibiotic class, the polymyxins. Mutations in the sensor kinase gene pmrB are well correlated with modified LPS and resistance to both polymyxin B and colistin in multiple bacterial species, but the likely broad effects of disrupting the PmrB regulatory pathway on bacterial metabolism have not been detailed. Indeed, in addition to predicted alterations in LPS, Han et al. observed unexpected global downregulation of phospholipids, fatty acids, and acyl-coenzyme A in polymyxin B-resistant pmrB mutants of Pseudomonas aeruginosa [37]. They propose that these broad changes also contribute to resistance, although their interesting observation is still in need of further investigation. While resistance in P. aeruginosa is often related to changes to LPS, LPS is not essential in A. baumanii and polymyxin resistance can involve not only complete loss of LPS, but polymyxin dependence, such that these strains cannot be cultured in the absence of antibiotic. Zhu et al. separately profiled A. baumanii outer and inner membranes and discovered a greater abundance of phosphatidylglycerol in the outer membrane of polymyxin-addicted strains [38]. Both studies on polymyxin resistance suggest that in addition to changes in LPS, perturbations to lipid metabolism and localization also affect polymyxin interactions with the outer membrane and are thus additional mechanisms of antibiotic resistance and dependence.

Conclusions

Recent findings in bacterial lipidomics emphasize the unprecedented structural detail, range of detection, and lipidome-wide quantitative comparisons that are enabled by mass spectrometric profiling. In particular the detection of lipid class-specific changes in fatty acid composition in response to genetic and environmental perturbations is sure to motivate the investigation of pathway-specific lipid metabolism and consequent changes to cellular membrane properties.

As with many -omics methods, the rate of data acquisition threatens to overwhelm our ability to mine the information meaningfully. Especially in bacteria, the number of unique ions detected far exceeds the number of known metabolites. Validation is often absent even for well characterized lipids because commercial standards are not available. A concerted effort is also required to standardize methods, assemble available parameters into accessible databases (including expanding existing resources like LIPID MAPS [39]), and develop tools to analyze complex datasets. Investment in such resources will determine whether lipidomics achieves a general level of adoption and thus scientific impact equivalent to proteomics. The future of spatially resolved lipidomics is even more unclear. Most advances are motivated by targeted applications (such as the uptake and distribution of xenobiotics), but expansion into global profiling will be helped by aforementioned improvements in validation and analysis tools. We look forward to further advances in both methodology and knowledge that addresses questions in bacterial lipids and hope that perspectives like this review will inform and inspire new entrants into the field.