Spatial lipidomics reveals biased phospholipid remodeling in acute Pseudomonas lung infection

Summary Pseudomonas aeruginosa (Pa) is a pathogen causing chronic pulmonary infections in patients with cystic fibrosis (CF). Manipulation of lipids is an important feature of Pa infection and on a tissue-level scale is poorly understood. Using a mouse model of acute Pa pulmonary infection, we explored the whole-lung phospholipid response using mass spectrometry imaging (MSI) and spatial lipidomics. Using a histology-driven analysis, we isolated airways and parenchyma from both mock- and Pa-infected lungs and used systems biology tools to identify enriched metabolic pathways from the differential phospholipid identities. Infection was associated with a set of 26 ions, with 11 unique to parenchyma and 6 unique to airways. Acyl remodeling was differentially enriched in infected parenchyma as the predominant biological function. These functions correlated with markers of polymorphonuclear (PMN) cell influx, a defining feature of the lung response to Pa infection, implicating enzymes active in phospholipid remodeling.


INTRODUCTION
Mass spectrometry imaging (MSI) is a rapidly expanding method of -omic level profiling that preserves spatial distribution and is now commonly used to profile host-pathogen interactions.We used a combination of MSI modalities to resolve the differential spatial lipidomic [1][2][3] profile in mouse lungs in response to acute infection by Pseudomonas aeruginosa (Pa).Infection by Pa is a lifelong complication in patients with cystic fibrosis (CF) [4][5][6] and with the emergence of multi-and extensively drug-resistant Pa strains, 7 identifying new mechanisms to control the infection are of critical need.We previously demonstrated the ability of MSI to describe a lethal pathogenic mechanism in a mouse model of Tularemia using differential lipid maps from mouse spleen, establishing that MSI can illuminate lethal pathogenic mechanisms of bacterial infection from lipids alone. 8Further, in that model, the anion-forming phospholipids (PLs) were sufficient to inform pathogenic mechanisms. 8In this work, we focused on the lipid ions detected in negative ion mode by MALDI-TOF mass spectrometry.We analyzed the infected lung samples using MALDI-TOF at 50 mm spatial resolution to illustrate the spatial elements of the infection.In parallel, we used a data-dependent acquisition (DDA)-MSI method to generate lipidome-per-pixel images to structurally identify the lipids within the infected lungs. 3Together, these data formed the basis for a systems biology type analysis 9 identifying dysregulation of the lipid profile as a central component of the mouse pulmonary response to Pa infection.
Lipids are the major constituents of biological membranes and are influenced by intracellular and extracellular conditions.A growing number of studies show a role for the dysregulation of host lipids in CF airway inflammation, lipid metabolism, and disease.In CF, patients have increased production of prostaglandin E2 metabolites and thromboxane B2 and its metabolites at all ages, suggesting a role for PL-derived metabolites in CF airway pathogenesis. 10Additionally, the Pa virulence factor, cystic fibrosis transmembrane conductance regulator (CFTR) inhibitory factor (Cif), promotes sustained airway inflammation by reducing host pro-resolving lipid mediators. 11Cif hydrolyzes epithelialderived 14,15-epoxyeicosatrienoic acid, disrupting transcellular production of the pro-resolving lipid 15-epi lipoxin A 4 (15-epi LXA 4 ) by neutrophils.Clinical data from CF patients revealed that Cif abundance correlated with increased inflammation, decreased 15-epi LXA 4 , and reduced pulmonary function.Finally, a growing body of research suggests a significant role for the recycling of polyunsaturated fatty acid (PUFA)-containing phospholipids, linking specific phospholipids to ferroptosis and inflammation. 12he Lands cycle 13,14 is a route for membrane phospholipid recycling (Graphical Abstract).6][17] When a fatty acid is removed from a phospholipid, what remains in a lysophospholipid that on its own has diverse downstream roles.However, to maintain the membrane phospholipid pool, the lysophospholipid can be re-acylated by an acyltransferase family enzyme-reconstituting an intact phospholipid.While four different processes are involved in the Lands cycle (based on various dependencies for acyl-coenzyme ll OPEN ACCESS A), the two main functions are de-acylation and re-acylation.This process diversifies the cellular phospholipid pool, the result of which can impact membrane-dependent functions.Due to an historical absence of tools to spatially resolve this process in situ, little is known about the role of the Lands cycle and dysregulation thereof in the context of a CF lung infection.The Lands cycle enzymes are highly correlated with neutrophil function, 15,18 and dysregulation of this pathway is consistent with the type of immune response observed in this model.To further extend the translational aspect of these results, ibuprofen is effective in relieving symptoms in young patients with CF, with the mechanism of ibuprofen reducing production of prostaglandins downstream of the Lands cycle. 19These studies highlight a critical role for lipid manipulation during Pa infection and represent druggable targets.
CF lung disease is characterized by CFTR dysfunction, decreased mucociliary clearance, and chronic inflammation.Bacterial infections cause much of the damage seen in CF patients' airways.However, a substantial portion of the lung damage is caused by the activation of the host innate immune system that accompanies infection, specifically neutrophil activation. 20Few animal models of CF airway infection recapitulate this neutrophil predominance and dysfunction due to differences in CFTR expression between humans and mice.Neutrophils primarily drive this chronic inflammation through the release of pro-inflammatory mediators, myeloperoxidase, and neutrophil elastase (NE). 21In the absence of pathogenic bacteria, airway mucus plugging and hypoxia promote interleukin-8 (IL-8) and leukotriene B 4 (LTB 4 ) release from respiratory epithelial cells, increasing neutrophil migration to the lung.In the lung, neutrophils exacerbate the inflammatory milieu, while further stimulating epithelial cells to secrete more IL-8 and LTB 4 , recruiting more neutrophils to the lung. 22,23Once individuals with CF are colonized with Pa, this pro-inflammatory cycle worsens, leading to permanent lung damage and respiratory decline.Although the CF lung has increased neutrophil levels at the time of initial infection, CFTR-deficient neutrophils are less effective at bacterial clearance through numerous mechanisms. 24Additionally, CF neutrophils are more likely to undergo NETosis, the formation of neutrophil extracellular traps (NETs).NETosis is a mechanism of pro-inflammatory cell death where chromosomal DNA and NE are extruded from the neutrophil. 24hile NETosis may be an effective way of infection control in other disease states, in CF it is ineffective and further worsens chronic inflammation and lung damage.An improved depth of cellular and molecular characterization of this host-pathogen interaction is necessary to better understand the mechanisms involved, and ultimately, to develop an improved bactericidal response from the host by targeting host processes.
In this study, we used an acute Pa infection model in wild-type mice to demonstrate the spatially dependent differential lipid responses in the lungs and identify new host-directed therapeutic targets.

Differential ion fingerprint in infected lungs
We used histological features to instruct segmentation of 2 regions of interest (ROIs) from each of the datasets: airway (AW) and parenchyma (PCM) from Pa-infected or mock-infected mouse lungs (Figure S1).Slides were stripped of matrix following MSI data capture, stained with H&E, optically scanned at 20x, and the tissue architecture was used to determine AW and PCM areas (Figure 1).The resulting ROIs contained 638 pixels (+/À 41) for AW and 1794 pixels (+/À 101) in the PCM.Average mass spectra for each ROI were exported, recalibrated, averaged (by group, mock-or Pa-infected), and normalized, and ion intensities were compared between groups (Figure 2) resulting in a pattern of 26 ions enriched in the Pa-infected AW and PCM components, non-overlapping with any mock-infected region.In the AW ROI, 6 ions were associated with Pa infection and no ions uniquely associated with mock infection or overlapping both conditions.The PCM ROI showed 11 ions enriched in the Pa-infection condition compared to 5 associated only with mock infection.A total of 9 ions were associated with Pa-infected AW and PCM.These data demonstrated a pattern of enrichment of specific ions with Pa-infected AW and PCM.

Phospholipid fingerprint of infected lungs
Using the ion fingerprints associated with each functional ROI and experimental group, we assigned putative identities using a combination of: i) on-tissue fragmentation using TOF/TOF, ii) accurate mass and ion trap MS/MS fragmentation obtained using DDA-imaging, and iii) previous curation in bacterial infection in mouse models.When ion abundance thresholds could not be met for satisfactory MS/MS identification, putative identities were inferred from bulk extracts using the most abundant isomer that could be identified using the criteria defined previously.Phospholipids were the predominant class identified (Table 1) with nearly half (8 of 17) confirmed as polyunsaturated.Representative data from two ions from the triplicate MALDI-TOF dataset are given in Figure S2.While phosphatidylethanolamine (PE) and phosphatidylglycerol (PG) are the predominant classes of phospholipids found in the Pa membrane, 25 these lipids are not exclusive to the bacteria and were included in the analysis.However, the sum composition 34:1 (carbons:unsaturations) is the most abundant composition usage used by Pa, in vitro, and it is possible that their enrichment in the Pa-infected PCM lung ROI is directly attributable to the bacteria. 25Extensive use of PUFA, notably arachidonic acid (C20:4), an omega-6 fatty acid, was found in the Pa-infected PCM, along with evidence for preferential incorporation of oleic (C18:1) and linoleic (C18:2, also omega-6) acids.Extensive incorporation of PUFAs is not found in lipid extracts of Pa.

Lands cycle implicated in infected lungs
Lipid fingerprints (Figure 2) were assigned metabolite identities using the human metabolite database identifiers, 26 with any ambiguities parsed as described in the STAR Methods (Table 1).Not all differential ions could be confidently assigned and were omitted.A metabolite-protein orthologous interaction network was constructed using the human KEGG knowledgebase for each of the ROI types (AW and PCM; Figures S3 and S4).For Pa-infected AW, we observed two interaction nodes: phosphatidylserine (PS) and phosphatidic acid (PA) indicating potential protein interactors (Figures S4B and S4C; Table S1).In the Pa-infected PCM, both the PS and PA nodes from the AW network were found and a third, PE node was found indicating potential protein interactors (Figure S4A; Table S1).We assembled the lists of protein interactors suggested by the network analysis into generalized enzymatic function with any known ligand specificity (Table S1).Next, we queried the lists of protein interactors generated from the metabolite-protein interaction network for enrichment of generalized pathways.Two functions predominated: lysolipid acyltransferases and sn2-position phospholipases (Table S2).These two classes of enzymes implicate active involvement of the Lands cycle and are an active component of neutrophil responses; neutrophils are the primary cellular infiltrate in Pa-infected lungs.By immunohistochemistry (IHC), we observed a robust neutrophilic influx in the PCM of Pa-infected lungs compared to mock-infected controls (Figure 3).Further, in PCM regions of dense cellular influx, neutrophils are co-localized with abundant Pa antigen.Notably, both Pa organisms are present as dense microcolonies in these regions along with Pa antigen-positive areas lacking extensive intact organisms indicating incomplete neutrophil-mediated killing at 48 h.These results highlight a key role of lipid immunometabolism in the neutrophil-mediated response to Pa infection.

DISCUSSION
Chronic pulmonary infections by Pa are a major cause of morbidity in patients with CF and chronic obstructive pulmonary disease.There is a gap in knowledge surrounding the fundamental host response to Pa infection in the lung from the perspective of lipids and metabolites.Here, we used an acute Pa infection in wild-type mouse lungs to determine whether histologically segmented lipidomic data could identify novel patterns of lipid metabolism to advance into studies of chronic Pa infection in CF models.
In this study, we focused on the role of phospholipids detected from tissue in the negative ion mode only.Our previous data from infected spleens demonstrated that negative ion mode phospholipids were sufficient to illustrate innate immune inflammation mechanisms.Though, given the important role in the lung of phospholipids that preferentially ionize in positive ion mode (ex: phosphatidylcholine, PC) as a major surfactant component, and the increasing links between lysoPCs and host response, future studies may benefit from including both positively and negatively ionized lipids.An important caveat of the DDA-MSI method is that precursor ion selection and the subsequent fragment scans are iterative across the entire tissue section.Therefore, the final ion identities are not necessarily sensitive to the location in which they were detected.It is possible that this contributes to some inconsistency in ion identification assigned in the TOF dataset as AW, but the identity used could have stemmed from an isobaric or isomeric precursor in the PCM, and vice versa.As a mitigating factor, the bulk of the spectra (based on area) in the DDA-MSI run would be in the PCM segment meaning the majority of identities would stem from the major region of interest.Finally, lacking stereospecific and double-bond positional isomer resolution in these experiments, we chose to include both possible stereoidentities in the metabolic network analysis, though many of the arrangements are improbable based on commonly observed stereospecificities and this will have an impact on the proposed enzyme involvement.We limited double-bond isomers to 9Z in cases where multiple positions were present in the metabolite library for monounsaturated acyl chains.Including multiple identities for a single ion interpretation could bias toward acyl remodeling by overrepresenting acyl possibilities, a possible limitation to this approach.This limitation is somewhat mitigated by the lack of acyl remodeling activities implicated in the AW segments, where the same metabolite identity and network analysis strategy was used.If the multiple assignments for a single ion biased the pathway analysis, then the acyl remodeling biases should be present in both AW and PCM, but it was a significantly enriched and differential feature of the PCM lipid profile, only.
The lung tissues evaluated here have a complex early immune response and the concomitant cellular and molecular changes of the host response are largely responsible for the differential lipid profiles during infection.However, Pa organisms are also present and replicating in abundance (evidenced by IHC results) and the total lipid response reflects some amount of bacterial lipids.The predominant acyl arrangement in lipid extracts from clinical and laboratory-adapted Pa strains is 34:1 with a high prevalence observed as the 16:0_18:1 molecular lipid species.The 34:1 composition is found on PE and PG head groups in lipid extracts from Pa with other minor head groups involved at far lower  abundances. 25Notably, the 34:1 sum composition is generally absent from the differential lipid profiles of Pa-infected AW and PCM, lending strong support to the interpretation that the bulk of the differential lipid response is due to the immune response rather than replicating Pa organisms.Ultrahigh resolution MSI 27 could address these issues in future studies, as practical and commonplace spatial resolution moves closer the cellular size of a single bacterium.Finally, an important next step will be to translate these results from acute infection to a model system that better reflects the features of the CF lung, for example the bENaC mouse model 28 using mucoid clinical isolates of Pa to recapitulate important aspects of Pa-infected human lungs from patients with CF.
Here, we demonstrate a dysregulated lipid response to pulmonary infection with Pa in an acute infection model.The differential lipid profiles are spatially informed and demonstrate the differences between the lipid response to infection in the AW compared to the PCM.A pathway analysis of the enriched functions implicated by the differential lipid profiles supports a strong role for acyl remodeling as a central characteristic of the lung immune response to Pa infection.These results strongly highlight the need to move away from bulk extractions to characterize -omic level changes in histologically and functionally complex organs.Finally, enzymes of the Lands cycle are supported as potential targets for investigation of new host-directed therapies for Pa infection.

Limitations of the study
Several limitations are present in this work.An acute pulmonary infection model does not accurately reflect all of the complexities of a chronic infection.These results only address an acute response and are the starting place for future work in chronic pulmonary infection models.Further, we could not separate the bacterial lipids from host lipids in this study and bacterial lipid changes could contribute, at least in part, to the overall differential lipid fingerprint.The neutrophil response was not correlated or directly overlaid with specific lipids, rather, the regional neutrophil responses (ex: parenchyma) were compared to the differential parenchymal lipid identities and this could limit the linkage to neutrophil-based mechanisms.In this study, we used an orthologous lipid:protein interaction network based on known human interactions and any mouse-specific activities would be underrepresented.While we identified acyl compositions in this work, finer details such as double-bond isomers and stereospecificities were not assigned.For this reason, we included both plausible sn-configurations while seeding the network analysis and that could contribute to the determination of acyl remodeling function.Finally, these studies were all carried out with a laboratory-adapted strain of Pa and in wild-type mice.Future work should focus on Pa clinical isolates and mouse strains that mimic the unique metabolic environment and immune responses of the CF lung.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Figure 1 .
Figure 1.Representative images and reference histology from tissue segmentation (A) Projection of polyunsaturated fatty acid (PUFA)-containing phospholipids, ions m/z 745.503 and 857.518 in both mock-and Pa-infected lungs from DDA-MSI experiment.Ion identities as given ([M-H]-for both).Representative of hundreds of lipid ions in dataset.Normalized, total ion current (TIC), scale bar 0-100% rel.intensity.Negative ion mode, MALDI-ELITE, 80 mm.(B) H&E reference stain from post-MSI tissues analyzed in (A).(C) Representative airway (AW, asterisk) and parenchyma (PCM, unmarked) regions manually segmented for histology-driven differential lipid detection.Arrowhead: dense polymorphonuclear (PMN) cellular infiltrate typical in the PCM of Pa-infected lungs.

Figure 2 .
Figure 2. Regional analysis of differential lipids and identification of enriched pathway functions in Pa-infected PCM (A) Venn diagram showing total unique ions describing mock-or Pa-infected segments.(B) Differential and significant pathway enrichment in Pa-infected PCM compared to AW segment.Statistical analysis based on MALDI-TOF triplicate dataset, negative ion mode, 50 mm spatial resolution.
differential assignments, associated with Pa infection (AW and PCM).Assignments: differential ions identified in triplicate MALDI-TOF dataset.Ions referenced to DDA-imaging dataset # and/or confirmed with MS/MS fragment data*.

Figure 3 .
Figure 3. Dense infiltration of infected lungs by neutrophils co-localized with Pa bacteria Triplicate lungs from mock-(left) and Pa-infected (right) serial sections from MALDI-TOF dataset stained with anti-Ly6G [clone 1A8] (emerald green) and anti-Pa (red) with light Mayer's hematoxylin counterstain.Capture areas representative of overall tissue section and shown from an 8x field view in a predominantly PCM area.Scale bar: 300 mm.

Table 1 .
Lipid identities comprising differential fingerprint