On-tissue derivatization of lipopolysaccharide for detection of lipid A using MALDI-MSI

Abstract

We developed a method to directly detect and map the Gram-negative bacterial virulence factor lipid A derived from lipopolysaccharide (LPS) by coupling acid hydrolysis with matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI). As the structure of lipid A (endotoxin) determines the innate immune outcome during infection, the ability to map its location within an infected organ or animal is needed to understand localized inflammatory responses that results during host–pathogen interactions. We previously demonstrated detection of free lipid A from infected tissue; however detection of lipid A derived from intact (smooth) LPS from host–pathogen MSI studies, proved elusive. Here, we detected LPS-derived lipid A from the Gram-negative pathogens, Escherichia coli (Ec, m/z 1797) and Pseudomonas aeruginosa (Pa, m/z 1446) using on-tissue acid hydrolysis to cleave the glycosidic linkage between the polysaccharide (core and O-antigen) and lipid A moieties of LPS. Using accurate mass methods, the ion corresponding to the major Ec and Pa lipid A species (m/z 1797 and 1446, respectively) were unambiguously discriminated from complex tissue substrates. Further, we evaluated potential delocalization and signal loss of other tissue lipids and found no evidence for either, making this LPS-to-Lipid A-MSI (LLA-MSI) method, compatible with simultaneous host–pathogen lipid imaging following acid hydrolysis. This spatially sensitive technique is the first step in mapping host-influenced de novo lipid A modifications, such as those associated with antimicrobial resistance phenotypes, during Gram-negative bacterial infection and will advance our understanding of the host–pathogen interface.

INTRODUCTION

Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) has been widely used for the generation of molecular maps created from tissue sections, including biomolecules and pharmaceuticals.(1–4) There are many advantages of the technique, such as label-free detection and targeting of the molecules by adjusting the experimental and instrumental conditions, and these topics have been extensively reviewed.(5) The spatially resolved molecular information from tissue can be further exploited to study the distribution of chemical species with diagnostic potential and for surgical decision-making.(6–8) Finally, this technique is increasingly employed as a tool to survey the host–pathogen interaction to determine the tissue level response to infection for novel therapeutic target identification.(9,10) We previously identified optimal MALDI detection conditions for lipid A, both in vitro and in vivo, demonstrating the feasibility of lipid A mapping in tissue.(10,11)

Lipopolysaccharide (LPS) comprises the outer leaflet of the outer membrane of Gram-negative bacteria and is characterized by three distinct moieties, from proximal to distal: lipid A, core, and O-antigen.(12,13) Lipid A is the membrane-forming moiety anchoring LPS within the membrane and is the most structurally conserved region of LPS, though lipid A modification is readily observed.(140 Core and O-antigen are composed of increasingly diverse carbohydrates.(15) Lipid A, also known as endotoxin, is the Toll-like receptor 4 complex (TLR4)-activating component of LPS and triggers the early innate immune response to defend against Gram-negative bacterial infection.(13) Escherichia coli (Ec) synthesizes a bisphosphorylated, hexa-acylated lipid A with the highest known TLR4-activation.(16) Most Gram-negative bacteria have unique lipid A structures that can also be modified in vivo to alter host innate immune responses, and resistance to antibiotics and antimicrobial peptides.(17,18) The pivotal observation that Pseudomonas aeruginosa (Pa) exhibits specific structural modifications associated with airway infections of patients with cystic fibrosis (CF) highlights the need to observe lipid A structure in vivo.(19–21) The host response to bacterial infection influences lipid A structure; thus, mapping lipid A to a specific location and tissue response is necessary to understand the host–pathogen interaction.(22) Previously, to analyze the structure of lipid A required isolation and analysis ex vivo, thereby losing any localized structural modifications that would be present in vivo.

Low-input extraction methods for lipid A derivation from full-length (also called smooth) LPS have been reported and are commonly used to investigate lipid A structural modification. (23,24) The targeted outcome of these techniques relies on cleavage of the glycosidic linkage between the first core sugar, 3-deoxy-D-manno-octulsonic acid (KDO) and the nonreducing glucosamine of the lipid A diglucosamine backbone prior to lipid A analysis using mass spectrometry (MS).(25) The most widely employed method for low-input extraction was pioneered by the Caroff group, using a mixture of isobutyric acid and ammonium hydroxide to both solvate and facilitate hydrolysis of lipid A.(26) Using this method, mass spectra of lipid A can be achieved from less than one colony of bacteria (~10⁸ bacterial cells ~10⁹ bacterial cells). Further studies demonstrated the potential for lipid A structure to differentiate many clinically relevant bacterial species as a potent rapid diagnostic tool, especially as lipid A relates directly to antimicrobial susceptibility patterns.(24,27) While the translational uses of lipid A extracted from cultured bacteria are promising, in vitro growth on solid or in liquid media has the potential to change lipid A phenotype and, thus, may not best represent its physiological state during infection. Measuring and characterizing lipid A from a zero-passage sample (no ex vivo expansion culture) hinges on overcoming the following technical hurdles: driving the detection limit to a minimal number of bacterial cells to obviate growth artifacts and optimizing the reaction conditions to detect lipid A from complex biological substrates while simultaneously maximizing its structural interpretation. To overcome the first limitation, we modified the Fast Lipid Analysis Technique (FLAT) for low input lipid A detection from LPS within a bacterial colony smeared from a toothpick, without centrifugation or lyophilization steps.(28) Due to the simplicity of the FLAT method for lipid A hydrolysis from LPS on a surface, we coupled it to tissue substrates for MALDI-MSI. This work combines the tenets of low-input extraction with detection from complex substrates to characterize lipid A from LPS directly on infected tissue sections.

We previously demonstrated that lipid A can be readily detected from infected tissue using a model system that makes a unique lipid A without core and O-antigen (“free” lipid A). (10) Spleens from mice infected with Francisella novicida showed a unique ion (m/z 1665.1), consistent with free lipid A using MSI. This study demonstrated that lipid A, when not bound to core and O-antigen, is readily detected in negative ion mode using the matrix norharmane (NRM) from bacterially infected tissue under the same MALDI-MSI preparatory conditions as phospholipids.(10) Other Gram-negative bacteria, such as the clinically relevant organisms, Ec and Pa do not transport free lipid A to the bacterial outer membrane. Rather, these organisms produce smooth LPS that is difficult to analyze by mass spectrometric techniques as fully intact molecules. In order to observe lipid A on-tissue from model infections, it is necessary to cleave it from full-length LPS. In this study, we demonstrate an effective method for on-tissue derivatization of LPS to observe lipid A directly on complex tissue substrates using MALDI-MSI, termed LPS-to-Lipid A-MSI (LLA-MSI).

EXPERIMENTAL

The FLAT lipid A hydrolysis method was modified for high-throughput and is described as follows.(28) Briefly, buffered citrate solution (0.2 M citric acid, 0.1 M trisodium citrate) was applied to a target surface containing spotted bacterial solution or thin tissue sections using a SunCollect matrix sprayer device (SunChrom, Napa, CA) on-tissue as follows: the sample was incubated in a humidified, closed glass chamber for 30 min at 110 °C. After heating, the chamber was removed from the oven and samples were washed with water (total volume 1 mL) by running water down the length of the slide with a pipet and let it dry in the atmosphere.

Two bacterial species producing LPS were used in these studies: Ec ATCC 25922 and Pa ATCC 31482 (sourced from the American Type Culture Collection, ATCC, Manassas, VA). For positive control samples modeling an infection of rodent tissue, a diluted overnight shaking liquid culture (see above) of Ec or Pa optical density (OD) 600 value: 1.0 ± 0.2, approximately 10 colony forming units (cfu)/mL of bacteria) was prepared in 2% porcine gelatin and used to inflate mouse lungs. The lung inflation procedure has been previously described in detail.(29) Frozen tissues were mounted to the cryostat stage using cold water. Twelve-micron kidney sections were cut using a Leica cryostat (Buffalo Grove, IL, U.S.A.), mounted onto a cold glass slide, thaw mounted, and incubated at 37 °C until visibly dry (approximately 2 min). Thirteen-micron sections were prepared from gelatin inflated lung, unless otherwise noted. Spot analysis of on-tissue acid hydrolysis experiments were performed on uninfected, unfixed, frozen mouse kidney, and lung tissues. A total of 1 μL of different concentrations of bacteria (Ec OD₆₀₀: 1.7 ± 0.2, 1.0 ± 0.1, 0.5 ± 0.1, and 0.3 ± 0.1; Pa OD₆₀₀: 1.7 ± 0.2, 1.0 ± 0.1, 0.5 ± 0.1, and 0.3 ± 0.1) were deposited on the tissue sections, to be used for determination of limit of detection (LOD) on the tissue sections. Subsequently, LLA-MSI was performed as described above.

For spot image analysis of on tissue substrates, matrices were applied with a SunCollect MALDI Sprayer. Spot analysis on tissue to determine LOD and tissue MSI data were collected on an Autoflex Speed and a Microflex LRF (Bruker Daltonics, Billerica, MA), respectively. In all cases, the matrix was prepared by dissolving 10 mg of NRM in 1 mL of chloroform/ MeOH (2:1, v/v) for spot analysis or pneumatic sprayer. Expanded materials, bacterial growth conditions, data collection, and instrumentation information are available in the Supporting Information.

RESULTS AND DISCUSSION

As smooth LPS is large and heterogeneous, a novel extraction and/or hydrolysis method was developed for the direct analysis of LPS-borne lipid A within tissues. For this analysis, the FLAT method previously used for low input lipid A detection directly from bacterial colonies was modified for use directly on unfixed tissue sections (Scheme 1, Supporting Information) and termed LLA-MSI. Briefly, tissue sections (infected, uninfected, or prepared with a control bacterial suspension) were coated with extraction solution, briefly incubated in a humidified chamber for 30 min at 110 °C, and dried. Lipid A was then detected in the negative ion mode with a NRM matrix, as previously described.(29) Both bacterial species used in these studies synthesize full-length LPS. Figure 1 illustrates a lipid A ion map from MALDI-MSI experiments using mouse kidney and as a model tissue for process validation (Ec, Figure 1a; Pa, Figure 1c). Ec produces a signature lipid A that results in a MALDI-generated ion at m/z 1797 and has been extensively described.(30) Four concentrations of Ec suspension were spotted directly on tissue sections: OD₆₀₀ 1.7 ± 0.2, 1.0 ± 0.1, 0.5 ± 0.1, and 0.3 ± 0.1 (OD₆₀₀ = 1.0 yields approximately 10⁹ CFU/mL) corresponding to region(s) of interest (ROIs) 1, 2, 3, and 4, respectively, in Figure 1a. The unspotted area as a tissue control is designated as ROI 5. The red circles on Figure 1 highlight the location of the solution deposited on the surface. The distribution of the ion m/z 1797 ± 1 was consistent with Ec lipid A (Figures 1e and S1) following on-tissue hydrolysis. We subsequently determined the limit of detection (LOD) for Ec lipid A as 5 × 10⁵ CFU deposited on tissue (1 μL of a 5 × 10⁸ CFU/mL stock suspension).

Figure 1.

MALDI ion map and spectra of bacterial culture spotted on mouse kidney tissue. (a, b) Ec and (c, d) Pa. (a) Labels 1, 2, 3, and 4 have an OD₆₀₀ value of 1.7, 1.0, 0.5, and 0.3, respectively. Label 5 is no deposition of Ec on tissue. (b) Average spectra for (a) with Ec lipid A m/z 1797 (e). (c) Labels 1 and 5, 2 and 6, and 3 have an OD₆₀₀ value of 1.7, 1.0, and 0.1, respectively. Labels 7 and 8 have an OD₆₀₀ value 0.5 and 0.25, respectively. Label 4 is no deposition of Pa on the tissue. Ion map at m/z 1446 (f) represents on-tissue (4–8) derivatization or from ITO surface (1–3). (d) Average spectra for (c). (a–d) MALDI-TOF MSI, normalized, TIC (total ion current).

Subsequently, we validated LLA-MSI using mouse lung tissue and the human opportunistic pathogen, Pa. Lipid A structural modifications are extensively described in primary clinical isolates of Pa from patients with CF and modulate with disease severity. Five different concentrations of Pa suspensions were deposited onto the tissue and neighboring ITO surface (Figure 1c). ROI 4 contains no bacteria and is the control. The major lipid A ion resulting from in vitro culture of Pa is m/z 1446 (Figures 1f and S1) and has been described previously.(31) The ion m/z 1446 was detected in the spotted Pa controls and with decreasing cell density (Figure 1c, labels 5– 8) in relationship to the amount deposited on tissue (Figure 1d). Comparing the relative abundance of m/z 1446 between the ITO surface and tissue surface, the ITO surface yielded the lowest visualization threshold (OD₆₀₀ 0.10 ± 0.03 spotted, Figure 1c). However, the ion is not observed above baseline on tissue at higher concentrations of spotted bacteria (OD₆₀₀, 0.25 ± 0.05, ROI 8). This may be a result of the ion-suppressive effect of tissue, explaining the loss of one to 2 orders of magnitude in ion abundance.(32) Consistent with Ec lipid A on-tissue, the LOD of Pa lipid A from on-tissue derivatization was approximately 5 × 10⁸ CFU/mL, or (Figure 1d) corresponding to 5 × 10⁵ deposited CFU in 1 μL. These results indicated that LLA-MSI could be applied to various types of tissue sections with multiple Gram-negative bacterial species to detect lipid A from smooth LPS with diversified lipid A structures. We did not observe significant delocalization of lipid A signal from the deposited area. Further, we compared the endogenous lipid distribution in uninfected kidney tissue before and after heat treatment, as shown in Figure S2. The endogenous ion distributions interrogated in mouse kidney showed no appreciable change in localization or abundance after LLA-MSI and we found no tissue deterioration (presented in Supporting Information).

Recently, we developed an optimized method for lung inflation compatible with downstream MSI experiments that allowed us to preserve intact morphology of lung tissue.(29) We applied the method to generate Gram-negative bacteria-loaded lungs as a controlled model of infected tissue. Bacterial pellets (Ec or Pa) were resuspended with 2% porcine gelatin in water to a final concentration of 10⁹ CFU/mL and subsequently used to inflate wildtype mouse lungs. Figure S3a,c illustrates the low spatial resolution detection of Ec and Pa lipid A on lung tissue, respectively. There is an intense peak at m/z 1797 that is consistent with Ec lipid A, as shown in Figure S3b (average spectrum extracted from Figure S3a). In contrast to the relatively uniform lipid A population produced by Ec, Pa produced multiple lipid A peaks, consistent with in vitro data. In the lung tissue preloaded with Pa, we observed intense peaks at m/z 1446 and 1462, as expected from the experiments using spotted bacteria (Figure 1). The characteristic cardiolipin from Pa was also observed at m/z 1404; the ion originated from cardiolipin because Pa does not produce a lipidVIA that is structurally consistent with the canonical enteric lipidVIA that yields the ion m/z 1404. There is a neighboring ion (m/z 1448) that is the base peak in both averaged mass spectra and that is consistent with cardiolipin in gelatin inflated mouse lung tissue (33) without bacterial preparation (Figure S3d).

We anticipated complexity in this region of the spectrum as it was possible that the peak at m/z 1446 originated from a mixture of lipid A ([C₇₀H₁₃₂N₂O₂₄P₂-H]⁻, m/z 1445.85670) and cardiolipin (CL 72:9, [C₈₁H₁₄₀O₁₇P₂-H]⁻, m/z 1445.94930). We sought to baseline resolve this region to fully determine the identities of this overlapping region of the spectrum. Indeed, the MALDI-TOF MSI peak observed in Figure S3d was a mixture of the two peaks, Pa lipid A and CL 72:9, because when high resolving mass spectrometry (HRMS; resolving power: ~60000 at 1445) was used for MALDI-MSI on tissue sections from Pa-loaded gelatin-inflated mouse lungs (Figure 2), we observed peak separation. To prepare the positive control tissue, 1.5 mL of Pa (10⁹ CFU/mL) solution was used to inflate a mouse lung. First, we investigated the endogenous lipid signature to confirm the absence of lipid A and presence of CL 72:9 in the spectrum using a tissue section that was not acid-hydrolyzed (Figure 2a). As expected, no peak was consistent with Pa lipid A at m/z 1445.8567; rather, only cardiolipin was observed at m/z 1445.9521, a 1.9 ppm mass error difference compared with the theoretical value at m/z 1445.94930 ([CL 72:9-H]−).

Figure 2.

Differentiation of the Pa lipid A ion from neighboring cardiolipin by accurate mass assignment on-tissue. (a) Spectrum of summed area in box (inset) showing low signal for Pa lipid A without LLA-MSI treatment. (b) Spectrum of summed area in red box (inset) showing Pa lipid A ion (m/z 1445.8624) with LLA-MSI, separated from neighboring cardiolipin peak (m/z 1445.9521). Tissue thicknesses 25 and 50 μm for (a) and (b), respectively. Chemical formulas given for both Pa lipid A and cardiolipin with respective mass errors. (a, b) MALDI-IT-Orbitrap MSI, negative ion mode, NRM matrix, normalized TIC, and raster width 400 μm.

In Pa-loaded gelatin-inflated mouse lungs that were treated using the LLA-MSI method, we observed both lipid A and cardiolipin ions (Figure 2b). In the treated tissue section, we observed the ion m/z 1445.8624 and identified it as Pa lipid A (mass error 3.9 ppm). We can conclude that our LLA-MSI method worked well in various tissue models. It is expected that the homogeneous bacterial solution will uniformly occupy the airspace of the whole lung volume using this preparation method, thus, a wide raster width was used for these studies. Therefore, there was no significant localization of bacteria in certain regions as would be expected in microcolony growth of a true in vivo infection. In order to demonstrate a maximized difference in peak patterns to separate the Pa lipid A from cardiolipin by HRMS, we employed two different thicknesses of the tissue sample (25 and 50 μm) in Figure 2a and b, respectively. We also evaluated sections of the same 25 μm thickness of treated tissue and clearly showed separation of lipid A and cardiolipin (Figure S5), as seen in Figure 2b with 50 μm thick sections. These experiments demonstrated that Pa lipid A is present only after LLA-MSI and is distinct from neighboring cardiolipin using an accurate mass method.

Conclusions

In this study, we demonstrated an effective method that allowed detection of lipid A from Gram-negative bacteria within the model tissue. As far as we know, this is the first method to detect and map lipid A through an on-tissue derivatization process from hydrolysis of smooth LPS. By on-tissue spot analysis using MALDI-MSI, the LOD was established at 5 × 10⁵ CFU/μL. Using Pa-loaded positive control tissues in the target organ (lung), we observed lipid A ions at m/z 1446 and 1462, establishing that multiple lipid A structures can be observed and are similar to in vitro profiles. Further, we demonstrated that endogenous lipids are neither delocalized nor lost in the LLA-MSI method, meaning that host–pathogen interaction imaging can be performed on a single section in the negative ion mode, and the data will contain usable host/pathogen lipid data as well as lipid A. This method could become an important tool for studying lipid A structural variation during infection, which is especially relevant in models of human disease, such as Pa infection of the lungs of patients with CF.