Lipidomic Analysis of Dynamic Eicosanoid Responses during the Induction and Resolution of Lyme Arthritis*

Eicosanoids and other bioactive lipid mediators are indispensable regulators of biological processes, as demonstrated by the numerous inflammatory diseases resulting from their dysregulation, including cancer, hyperalgesia, atherosclerosis, and arthritis. Despite their importance, a robust strategy comparable with gene or protein array technology for comprehensively analyzing the eicosanoid metabolome has not been forthcoming. We have developed liquid chromatography-tandem mass spectrometry methodology that quantitatively and comprehensively analyzes the eicosanoid metabolome and utilized this approach to characterize eicosanoid production during experimental Lyme arthritis in mice infected with the bacterium Borrelia burgdorferi. Eicosanoids were extracted throughout infection from the joints of Lyme arthritis-resistant and -susceptible mice and subjected to lipidomic profiling. We identified temporal and quantitative differences between these mouse strains in the production of eicosanoids, which correlated with differences in arthritis development. The eicosanoid biosynthetic enzyme cyclooxygenase (COX)-2 has been implicated in the regulation of Lyme arthritis pathology, and subsequent lipidomic profiling of B. burgdorferi-infected COX-2−/− mice identified reductions not only in COX-2 products but, surprisingly, also significant off-target reductions in 5-lipoxygenase metabolites. Our results demonstrate the utility of a comprehensive lipidomic approach for identifying potential contributors to disease pathology and may facilitate the development of more precisely targeted treatment strategies.

Systems biology has advanced exponentially during the past decade, producing a wealth of data relating to disease susceptibility and pathology. Because of its immense scope and complex chemical composition, metabolomics has lagged behind its genomics and proteomics counterparts, hampering the advancement of sequential systems biology as a cohesive approach to disease investigation and drug discovery (1). The unique challenges faced by metabolomics researchers are evident in the emerging field of lipidomics, the analysis and characterization of the enormous and diverse set of lipid molecules. For this reason, the Lipid Metabolites and Pathway Strategy (LIPID MAPS) consortium was created to provide an intellectual infrastructure for the advancement of lipidomics (2). To expedite the elevation of lipidomics to a level comparable with that of genomics and proteomics, efforts have included the establishment of a comprehensive lipid classification system (3), the adoption of liquid chromatography-tandem mass spectrometry (LC-MS/MS) 5 as the preferred methodology for lipid analysis (2), and the development of a publicly accessible data base of lipid mass spectra and standards (see Lipidomics Gateway web site). These integrated components provide a springboard for the investigation of potential interactions between large classes of lipid signaling molecules and their impact on the induction and modulation of disease pathology.
Eicosanoids comprise a diverse class of over one hundred bioactive lipid mediators derived from the metabolism of polyunsaturated fatty acids by cyclooxygenase (COX) (4,5), lipoxygenase (LOX) (4 -6), and cytochrome P450 (CYP) (4,7) as well as other non-enzymatic pathways (4). As a class, these molecules act as spatially localized hormone-like signaling molecules that have a wide range of effects, and under non-disease conditions contribute to the maintenance of homeostasis in almost every organ system (5). However, it is because of their participation in inflammatory diseases that eicosanoids have been the target of intense study. For example, cyclooxygenases can generate pro-inflammatory prostaglandins (PG), which are responsible for many of the hallmark signs of inflammation such as heat, redness, swelling and pain, whereas cytochrome P450s create epoxyeicosatrienoic acids (EET), which may affect local inflammation by promoting vasodilatation, and the lipoxygenases, which generate molecules with diverse signaling effects such as pro-inflammatory leukotrienes (LT) or anti-inflammatory/pro-resolution lipoxins (LX), resolvins (Rv), and protectins (PD). Metabolic and non-enzymatic breakdown products of these lipids can have significantly altered bioactivity, further complicating our understanding of these intricate pathways.
The currently accepted model of inflammation proposes that both the induction of inflammation and its eventual resolution and return to homeostasis are not passive events but instead are coordinated, orderly processes actively signaled by specific protein and lipid molecules (6). The inflammatory response is the first step in the immune response and is characterized by an influx of innate immune cells, the production of soluble proinflammatory mediators, and tissue destruction. Conversely, resolution sees an influx of anti-inflammatory macrophages, pro-resolution lipid production, and tissue remodeling and repair. Although the roles of some individual eicosanoids in inflammation have been well studied, the overall biological significance of how these signals act in vivo to properly initiate and resolve inflammation has yet to be fully elucidated. Studies specifically addressing pathway interactions have investigated the connections between PGE 2 and LX signaling in several animal models of autoimmunity (8,9), and Schmelzer et al. (10) have identified in vivo changes in prostaglandin signaling following modulation of endogenous EETs. Although these studies have examined isolated sectors of the eicosanoid metabolome, no reported study has comprehensively examined the temporal production of these metabolites. Our understanding of eicosanoid involvement in regulating the response to infection is even more fragmented and has been focused primarily on a few select metabolites (11,12). Thus, a complete metabolomics approach on a scale comparable with current gene microarray technology is essential to achieve a thorough understanding of the inflammatory and immune responses and to facilitate the design of more effective disease therapies.
To date, lipidomics studies have been capable of measuring, at most, only ϳ40 eicosanoids, fewer than half of the known eicosanoid species, with a per sample run time of over 50 min (13,14). Additionally, extraction methods have been limited to primarily soft tissue samples such as lung or brain, excluding the possibility of lipidomic analysis of more complex tissues such as arthritic joints. These constraints create significant impediments toward the general applicability of previous methodology and require considerable technical development when applying them to other biological systems. For this reason, we have developed a protocol for complete eicosanoid extraction from complex tissue combined with a comprehensive LC-MS/MS methodology with a run time of 25 min/sample to identify and quantify 104 unique lipid species, which encompasses about 90% of the known eicosanoid metabolome (4). These advances allow this technique to be robustly applied to diverse biological samples at a systems biology level.
To demonstrate the applicability of such an approach, we utilized this strategy to investigate the role of eicosanoids in regulating the development and resolution of arthritis pathology in mice infected with Borrelia burgdorferi, the causative agent of Lyme disease. Lyme disease is the most common vector-borne disease in the Northern hemisphere and presents diverse pathologies, the most common of which is arthritis (15). Lyme arthritis manifests in about 60 -80% of individuals not treated with antibiotics and is characterized by cellular infil-trate consisting of primarily neutrophils and macrophages, a pathology that is recapitulated in the murine model (16). The resolution of arthritis is thought to be dependent upon the clearance of B. burgdorferi from the joints by specific antibodies. Genomic profiling of mouse strains that are resistant (C57BL/6, DBA/2) or susceptible (C3H/He) to the development of Lyme arthritis has yielded intriguing data (17) but has generated more questions than answers as to the determining factor(s) for arthritis development. Although previous studies indicate that eicosanoids may play a role in regulating Lyme arthritis pathology, it is unknown which eicosanoids are involved or what their specific roles might be in this pathology (11).
Here we present the first comprehensive lipidomic analysis of an in vivo inflammatory response, from initiation to resolution. Analysis of joints from Lyme arthritis-resistant and -susceptible mouse strains revealed significant differences in basal and infection-induced eicosanoid production between the two strains. We further demonstrate the utility of this approach by performing lipidomic analysis on joints from COX-2-deficient (COX-2 Ϫ/Ϫ ) animals during infection with B. burgdorferi at the peak of the inflammatory response. Lipidomics of COX-2 Ϫ/Ϫ joints exhibited the expected decrease in COX-2 products but also revealed surprising decreases in multiple products from the 5-LOX pathway. These data establish the robustness of this protocol, as well as its potential to uncover unexpected regulatory pathways.

EXPERIMENTAL PROCEDURES
Materials-LC grade solvents were purchased from EMD Biosciences (San Diego, CA). Synergy C18 reverse phase high pressure liquid chromatography (HPLC) column and Strata-X solid phase extraction columns were purchased from Phenomenex (Torrance, CA). Eicosanoids were purchased from Cayman Chemicals (Ann Arbor, MI) and Biomol (Plymouth Meeting, PA). With the exception of dihomo-PGF 2␣ , primary standards for dihomo (adrenic acid-derived)-prostaglandins were not commercially available; these metabolites were analyzed qualitatively using previously characterized extracts from RAW264.7 murine macrophage cells prepared as described (18).
Animals-Female C3H/HeJ (C3H) and DBA/2J (DBA) mice, 4 -6 weeks of age, were purchased from The Jackson Laboratory (Bar Harbor, ME). COX-2 heterozygous mice (B6;129S7-Ptgs2 tm1Jed ) were purchased as breeders from The Jackson Laboratory and backcrossed onto the C3H genetic background for 10 generations. Heterozygous mice were then intercrossed to produce knock-out and wild-type littermates. Animals were housed in a specific pathogen-free facility and given sterile food and water ad libitum. All studies were conducted in accordance with the Animal Care and Use Committee of the University of Missouri.
Bacteria and Infections-A virulent, low passage, clonal isolate of the B. burgdorferi N40 strain was used for all infections. Frozen stocks were placed in 7.5 ml of Barbour, Stoenner, Kelly (BSK) II medium (Sigma-Aldrich) with 6% rabbit serum (Sigma-Aldrich) and grown to log phase by incubation for 5-6 days at 32°C. Spirochetes were enumerated using dark field micros-copy and a Petroff-Hausser counting chamber (Hausser Scientific, Horsham, PA). Spirochete dilutions were made in sterile BSK II medium, and mice were inoculated in both hind footpads with 5 ϫ 10 4 B. burgdorferi organisms in 50 l of medium for a final inoculum of 1 ϫ 10 5 , a concentration that reliably produces arthritis in susceptible animals (19). Sham-infected control animals were injected with 50 l of BSK II medium/ hind footpad.
Arthritis Severity Scores-To determine arthritis severity scores, a histologic analysis of the left tibiotarsal joint from each mouse was performed following sacrifice. The tibiotarsal joint was excised, fixed in 10% buffered zinc-formalin, and embedded in paraffin, and 5-m sections were stained with hematoxylin and eosin. The sections were evaluated in a blind manner by two independent observers and were assessed for arthritis severity on a scale of 0 to 4. A score of 0 indicated normal tissue, 1, 2, and 3 indicated minimal, mild, and moderate inflammation, respectively, and 4 indicated severe arthritis. The pathology present in histologic sections was characterized by neutrophil and monocyte infiltration into the joints, tendons, and ligament sheaths; hypertrophy and hyperplasia of the synovium; and fibrin exudates. The extent of the observed inflammatory changes formed the basis for the arthritis severity scores.
Determination of Antibody Levels-B. burgdorferi-specific IgM and IgG levels in the sera of infected animals were determined by enzyme-linked immunosorbent assay as described previously (11).
Determination of B. burgdorferi Loads-All tissues were snap-frozen in liquid nitrogen immediately after collection and then stored at Ϫ80°C until extraction. Tissues were homogenized in TRIzol (Invitrogen), and DNA was extracted according to the manufacturer's instructions. Multiplex quantitative PCR for determination of B. burgdorferi loads was performed using TaqMan Universal PCR Master Mix (Applied Biosystems, Foster City, CA) as described previously (19). Bacterial loads are expressed as copies of B. burgdorferi flagellin per 1000 copies of mouse nidogen.
Eicosanoid Extraction from Tibiotarsal Tissue-Tissues were snap-frozen in liquid nitrogen and stored at Ϫ80°C until extraction. The extraction protocol is a modification of Chen et al. (20). For extraction, each frozen tibiotarsal joint was wrapped in foil and placed in liquid nitrogen. The joint was then pulverized and the resulting frozen powder placed immediately in 3 ml of ice-cold 50% ethanol and then weighed. This protocol resulted in a nominal loss of tissue; the weight was later used for the normalization of eicosanoid levels in each sample. 10 l of antioxidant mixture (0.2 mg/ml butylated hydroxytoluene, 0.2 mg/ml EDTA, 2 mg/ml triphenylphosphine, 2 mg/ml indomethacin in a solution of 2:1:1 methanol:ethanol:H 2 O) was added to each sample, and all tubes were kept on ice until incubation at Ϫ20°C for 72 h. Samples were then centrifuged at 3500 ϫ g for 30 min, and the clear ethanolic supernatant was removed to a new tube and dried under nitrogen gas. Samples were reconstituted with 2 ml of 10% MeOH and supplemented with 50 l of internal standard containing the following deuterated eicosanoids ( ) 9,10-diHOME, and (d 4 ) 12,13-diHOME. Samples were sonicated for 30 s with a probe sonicator and purified by solid phase extraction as described previously (21,22). Prior to LC-MS/MS analysis, samples were evaporated using a SpeedVac and reconstituted in 50 l of LC solvent A (water-acetonitrile-acetic acid (70:30: 0.02, v/v/v)).
Liquid Chromatography and Mass Spectrometry-Eicosanoids were separated by reverse-phase LC on a Synergy C18 column (2.1 ϫ 250 mm) at a flow rate of 300 l/min at 50°C. The column was equilibrated in solvent A (water-acetonitrileacetic acid (70:30:0.02; v/v/v)), and 40 l of sample (80%) was injected using a 50-l injection loop and eluted with 0% solvent B (acetonitrile-isopropyl alcohol (50:50, v/v)) between 0 and 1 min. Solvent B was increased in a linear gradient to 25% solvent B until 3 min, to 45% until 11 min, to 60% until 13 min, to 75% until 18 min, and to 90% until 18.5 min. Solvent B was held at 90% until 20 min, dropped to 0% by 21 min, and held until 25 min. Eicosanoids were analyzed using a tandem quadrupole mass spectrometer (ABI 4000 Q-Trap, Applied Biosystems) via multiple reaction monitoring (MRM) in negative ion mode. The electrospray voltage was Ϫ4.5 kV, and the turbo ion spray source temperature was 525°C. Collisional activation of eicosanoid precursor ions used nitrogen as a collision gas. Eicosanoids were measured using precursor3product (MRM) pairs, and the MS analysis was divided into six periods ( Fig. 1, supplemental Table 1). The duty cycle for each period ranged between 390 and 840 ms, with an average time of 565 ms. The declustering potential and collision energy for each eicosanoid was optimized for maximal signal using flow injection mass spectrometry. Eicosanoids were identified in samples by matching their MRM signal and LC retention time with those of a pure standard. Extraction and ionization efficiencies were determined for the internal standards by comparing the ion intensity of the samples with those of the extraction controls. The range of efficiency for a typical experiment was between 20 and 75%, with an average of around 50%.
Quantitative Eicosanoid Analysis-Quantitative eicosanoid determination was performed by the stable isotope dilution method (21,22). Deuterated internal standards (2.5 ng) were added to each sample to account for extraction efficiency and mass spectrometry ion suppression. To calculate the amount of eicosanoid in a sample, the ratio of the natural eicosanoid peak area to the deuterated eicosanoid peak area (termed ''eicosanoid ratio'') was determined for each sample. The eicosanoid ratio, proportional to ng of eicosanoid, was converted using the standard curve, 2.5 ng of each internal (deuterated) eicosanoid standard mixed with the following amounts of natural eicosanoid (nondeuterated) primary standard: 0.1, 0.3, 1, 3, 10, 30, and 100 ng. The hepoxilin standards that are commercially available from Biomol each contain a mixture of two diasteriomers, which are not distinguishable by this methodology and are identified only as hepoxilin (HX) A 3 and HXB 3 , respectively. For PD 1 and ⌬15t-PD 1 , the isomer 10S,17S-diHDoHE was used as the quantitative primary standard; for 11␤ PGE 2 , the isomer PGE 2 was used. Performing linear regression using Y ϭ mX, where (Y ϭ eicosanoid ratio, X ϭ ng of eicosanoid, m ϭ slope), the eicosanoid ratio was multiplied by (1/m) to determine the ng eicosanoid in a sample. The deuterated internal standards contained a small but detectable level of natural, undeuterated eicosanoid; to account for this, extraction controls were performed in quadruplicate and the ng of eicosanoid subtracted from each sample. The final values were normalized to the original weight of the tissue and expressed as ng of eicosanoid/mg of tibiotarsal tissue.
Heat Maps-In addition to quantitative analysis, data were expressed as a heat map to identify relative increases and decreases in metabolites over time. Similar to the quantitative analysis, the extraction eicosanoid ratio was subtracted from the sample ratio and normalized to the original weight of the tissue; this value was then normalized to the corresponding value at day 0. The (1/m) value determined from the quantita-tive curve cancels out of relative fold calculations. One advantage to eliminating (1/m) is that, for eicosanoids with a limited availability, analysis can be performed in the absence of full quantitation.
Statistics-Results are expressed as the means Ϯ S.E. Comparisons were made using one-way analysis of variance followed by a multiple comparison procedure, or t test, and p values Յ 0.05 were considered significant.

Murine Model of Lyme Arthritis-Upon infection with
B. burgdorferi, arthritis-susceptible C3H/HeJ (C3H) mice developed a severe inflammatory response as measured by arthritis severity scores, which peaked at 2 weeks post-infection and then spontaneously resolved. Resistant DBA/2J (DBA) mice developed only a very mild inflammation ( Fig. 2A). Bacterial loads in the joint tended to mirror the course of pathology (Fig. 2B), peaking just before the arthritic response. However, inoculum size does not affect susceptibility or resistance to arthritis development in most mouse strains (23). The difference in histopathology was most evident at the peak of arthritis, the development of which is dependent upon neutrophil recruitment into the infected joint ( Fig. 2C, day 10) (19), whereas macrophages predominated during the resolution phase (Fig. 2C, day 35) (16). Resolution of arthritis is attributed to B. burgdorferi-specific IgM-and IgG-mediated (Fig. 2D) clearance of bacteria from infected joints (24). Overall, DBA mice demonstrated less cellular infiltration into the joints with almost no differences in the bacterial loads and antibody responses, demonstrating that Lyme arthritis susceptibility is genetically regulated and independent of the numbers of bacteria in the joints (25).
Lipidomic Analysis of Infected Joints-Tibiotarsal (ankle) joints from B. burgdorferi-infected C3H and DBA mice were collected at various time points post-infection and analyzed for eicosanoid levels during the progression of arthritis (Fig. 3). The majority of joint mass is bone, making it difficult to extract the low levels of eicosanoids from the surrounding tissue comprised of infiltrating cells, muscle, synovium, intrarticular fluid, and cellular exudate. Joints were kept frozen with liquid nitrogen and then pulverized. The resultant powder was placed in cold 50% ethanol containing an antioxidant mixture to prevent lipid degradation. After 72 h of incubation at Ϫ20°C, samples were clarified by centrifugation and supernatants dried completely under nitrogen gas. Repeated analysis found dramatically decreased eicosanoid levels with Ϫ20°C incubation times less than 72 h and little to no increase in lipid concentrations with incubation times greater than 72 h. Samples were reconstituted with 10% methanol containing 21 deuterated internal standards representative of products from all major eicosanoid biosynthetic pathways. Subsequently, the samples were purified by solid phase extraction and stored in methanol for analysis (21,22).
Eicosanoids were analyzed using liquid chromatography coupled with tandem mass spectrometry. Upon injection, eicosanoids were eluted using a 25-min reverse phase liquid chromatography gradient capable of separating the majority of known eicosanoids. Following LC separation, each eicosanoid was selected for identification using MRM, whereby the mass spectrometer identified a given molecule by its parent mass and a characteristic fragment ion generated in the second quadrupole. In combination with the LC retention time, this technique can distinguish large numbers of eicosanoids in a single analysis (Fig. 1, supplemental Table 1). The mass spectrometer duty cycle limits the number of MRM transitions that can effectively be run simultaneously, and thus they were separated into six different temporal analysis periods. The average detection limit using this method was 10.5 pg, with more than 80% of the metabolites in this method detectable at this level. Eicosanoids were measured using the stable isotope dilution method, which assesses each metabolite as the ratio of signal from the eico-sanoid to that of the isotopically distinguishable internal standard (22). Using stable isotope dilution quantitation helps diminish the potential error due to extraction, purification, and matrix effects that undoubtedly arise from tissue samples.
In total, 49 or 47 unique molecular species were detected in joints from C3H or DBA mice, respectively, including metabolites derived from cyclooxygenase, lipoxygenase, cytochrome P450, and non-enzymatic and degradative pathways. In contrast, only 41 were detected in sham-infected joints. Significant temporal changes in the COX, LOX, and CYP pathways were identified in both C3H and DBA mice during the course of infection (Fig. 3). In C3H mice, the majority of these eicosanoids were generally elevated between days 3 and 17 postinfection. Arthritis-resistant DBA mice generated a similar but muted metabolic profile, concentrated temporally in days 7 and 10. Non-enzymatically generated isoprostane eicosanoids increased dramatically in C3H mice, indicating massive changes in oxidative stress, which correlated with the severity of arthritis in this strain. Sham-infected C3H mice showed no significant changes in eicosanoid production in the ankle joint (supplemental Fig. S1). Selected metabolites from each of the eicosanoid classes are described herein, and their proposed contribution to the inflammatory response is suggested and subsequently used to develop a model of eicosanoid dynamics during the induction and resolution of inflammation in response to B. burgdorferi infection. The complete quantitative data set has been made available (supplemental Tables 2-4).
COX and CYP Metabolites-Previous studies examining models of rheumatoid arthritis and other models of autoimmunity have defined PGE 2 and PGD 2 as classic pro-inflammatory COX products that induce and regulate inflammatory responses, as well as some aspects of adaptive immunity, such as T cell activation and cytokine expression (26,27). During experimental Lyme arthritis, PGE 2 and PGD 2 levels increased in C3H mice beginning at day 3 and returned to basal levels by day 17 (Fig. 4, A and B), preceding both bacterial clearance from the joint and the production of B. burgdorferi-specific IgG antibodies. Arthritis-resistant DBA mice produced high levels of PGE 2 from days 7 through 10; however, PGD 2 was not significantly increased during infection. PGD 2 can be hydrolyzed to 15d PGJ 2 , which purportedly exerts anti-inflammatory effects by inhibiting NF-B and activating peroxisome proliferatoractivated receptor ␥ (28,29). However, work demonstrating this effect was performed primarily in vitro, and thus its role in vivo remains controversial (30). Joints from both strains contained significant levels of 15-d PGJ 2 at d7, as well as at d3 in C3H animals (Fig. 4C). PGE 2 is cleared from the extracellular milieu by enzymatic oxidation to form the less bioactive metabolite 15k-PGE 2 . Production of this metabolite followed a pattern similar to that of its precursor and was detected at high levels from d3 through d21 post-infection in both DBA and C3H mice (Fig. 4D). These data suggest that PGE 2 and PGD 2 may promote the onset of the inflammatory response, possibly by inducing the production of neutrophil-chemotactic molecules such as CXCL-1 and CXCL-2, the production of which are responsive to PG levels and which have previously been implicated in the development of Lyme arthritis (19,31,32). Additionally, these eicosanoids may affect the adaptive immune response by regulating the activation of antigen-presenting cells such as dendritic cells and macrophages (27,33).
EETs are produced primarily by the CYP2C and CYP2J epoxygenases and modulate inflammation by promoting vaso-dilatation, angiogenesis, and cellular proliferation (34), and some have been shown to inhibit NF-B activity in vitro (35,36). Soluble epoxide hydrolase terminates EET signaling by hydrolysis to biologically inactive dihydroxyeicosatrienoic  Table 2). EETs also increased in DBA mice at days 7 and 10, returning to basal levels subsequent to bacterial clearance from the joints (Fig. 4F). Production levels were similar between C3H and DBA mice for all of the EETs except for 5,6-EET, which was present at much higher levels relative to both C3H 5,6-EET levels and the other three EETs in DBA joints. The ratio of any given EET/DHET pair remained constant through the time of infection in both C3H and DBA mice (supplemental Fig. S2, A and  B), suggesting that soluble epoxide hydrolase activity was unaffected by B. burgdorferi infection. However, whereas 5,6-EET, 8,9-EET, and 14,15-EET were 6-, 12-, and 4-fold, respectively, more prevalent than their corresponding DHET, 11,12-EET was 4-fold less abundant than 11,12-DHET (Fig. 4G). Node et al. (35) showed that 11,12-EET specifically prevents leukocyte recruitment by decreasing NF-B-regulated vascular cell adhesion molecule (VCAM) expression in a model of vascular inflammation, whereas similar effects have been demonstrated with all four EETs using an ex vivo model of blood vessel laminar flow (36). It is unlikely that soluble epoxide hydrolase selectively metabolizes 11,12-EET, because the rank order of degradation rate is 14,15-EET Ͼ 11,12-EET ϭ 8,9-EET (37, 38). However, any 11,12-EET interacting with NF-B would be covalently bound and undetected by our assay, leading to the appearance of increased metabolism. This suggests that EETs, specifically 11,12-EET, might play an important role during infection-induced inflammation by counterbalancing pro-inflammatory signals. Lipoxygenase Metabolites-Mammalian LOX can be classified according to enzymatic activity (5-LOX, 8-LOX, 12-LOX, 15-LOX) or phylogenetic relatedness (4, 39). 5-LOX activity is attributed to a single, highly conserved protein that creates a diverse range of pro-and anti-inflammatory signals. 5-LOX catalyzes the rate-limiting step for the synthesis of leukotrienes, important products and regulators of innate immune cell function and powerful neutrophil chemoattractant signals (40). It also facilitates the formation of 5-HETE from arachidonic acid, which can be further oxidized by neutrophils to 5-oxoETE, which has been shown to be highly chemotactic for eosinophils (41). In B. burgdorferi-infected C3H mice, neutrophil influx corresponded to changes in both of these pathways. Similar to the prostaglandins, 5-oxoETE levels increased in C3H mice starting at day 3 and returned to basal levels by day 21 (Fig. 5A). DBA mice also demonstrated significantly increased 5-oxoETE levels at d7 and d10 post-infection, which then quickly dropped to near basal levels by d14. The levels of LTE 4 were dramatically affected during the immune response in C3H mice (Fig. 5B) and were significantly increased at all but two time points (d21 and d24 post-infection). Remarkably, no cysteinyl leukotrienes were detected at significantly elevated levels in DBA joints. Thus, the production of 5-LOX metabolites corresponded with the intensity of neutrophil recruitment to the site of B. burgdorferi infection.
The different murine paralogues of 12-LOX and 15-LOX (12/ 15-LOX) mediate the formation of 12-hydroperoxyeicosatetraenoic acid, whereupon cellular peroxide levels and sequence-specific protein activities determine its further metabolism. The leukocyte 12/15-LOX protein is highly expressed in murine cells of the monocyte lineage, especially resident macrophages, and its expression correlates with that of a diverse array of chemokines and cytokines (42,43). In a model of acute sterile inflammation, mice lacking this specific 12/15-LOX gene exhibited increased monocyte/macrophage influx, and these cells expressed a cytokine profile reminiscent of anti-inflammatory cells, including lower levels of macrophage chemoattractant protein-1 (MCP-1; CCL2) and interleukin-17 (43). 12-Hydroperoxyeicosatetraenoic acid can be reduced to 12-HETE, which promotes monocyte chemotaxis in vitro by inducing MCP-1 expression (44) and in vivo induces monocyte adhesion through epithelial intercellular adhesion molecule (ICAM) up-regulation in a murine model of atherosclerosis (45). Alternatively, 12-hydroperoxyeicosatetraenoic acid can be isomerized to form biologically inert HXB 3 or the biologically active HXA 3 , which elicits neutrophil chemotaxis without superoxide production or granule release in vitro (46). In vivo results are equivocal; administration of HXA 3 inhibits fibrosis and suppresses inflammatory macrophage recruitment in a model of bleomycin-induced lung injury (47); however, HXA 3 is required for the recruitment of neutrophils in some models of mucosal infection (48). B. burgdorferi -infected C3H mice produced HXA 3 at significantly increased levels between days 3 and 14, whereas DBA production of HXA 3 in DBA strains spiked at days 10 and 28 post-infection (Fig. 5C). Hepoxilin B 3 was detected at lower levels in both strains during the same period (supplemental Tables 2 and 3). 12-HETE levels steadily rose in C3H joints during the induction phase of inflammation, peaking at day 14 and then returning to base-line levels by d21 post-infection (Fig. 5D). Similar to HXA 3 , 12-HETE peaked at d10 and d28 in DBA mice. Our results for 12-HETE and HXA 3 imply that during a sustained, long term model of an inflammatory response, such as infection, these molecules may perform dual roles by recruiting cells to clear bacteria from the site of infection and also counterbalancing pro-inflammatory signals produced at the same time.
LX, Rv, and PD comprise a class of lipid mediators produced by the action of a transcellular enzyme cascade of LOXs and epoxygenases that are differentially expressed between heterogeneous combinations of neutrophils, mac-rophages, platelets, or endothelial cells (6,49). Therapeutically administered protectins have antiinflammatory properties, such as the reduction of neutrophil influx, and pro-resolution properties, including the recruitment of monocytes, the activation of antiinflammatory macrophages, and the phagocytosis of cellular and microbial debris (6,50). Although LXA 4 concentrations have been determined in serum by enzymelinked immunosorbent assay in a model of parasite infection (51), lipoxins have not previously been measured by LC-MS/MS from tissue during infection. Additionally, neither Rv nor PD have been measured in tissue following infection. Protectins were identified in B. burgdorferi-infected joints from both mouse strains. PD 1 increased 5-10-fold from days 7 to 17 in C3H joints, whereas DBA mice did not demonstrate a significant increase in PD 1 levels during B. burgdorferi infection (Fig. 5E). The less biologically active protectin isomers, ⌬15t-PD 1 and 10S,17S-diHDoHE, were also generated at higher concentrations in C3H than in DBA joints in response to infection (supplemental Fig. S3, A and B). PD 1 was 4-fold less abundant than 10S,17S-diHDoHE; however, it has been shown to block neutrophil influx with greater potency in vivo, which suggests that PD 1 may have a physiologically important role in controlling the response to B. burgdorferi infection (52). Additionally, at various time points post-infection, C3H mice demonstrated significantly higher levels of 17-HDoHE, a stable form of the Rv and PD biosynthetic intermediate 17-HpDoHE (Fig. 5F) (53). At d14, 17-HDoHE was produced at Ͼ1-fold levels in C3H joints versus DBA; however, this situation was reversed at d28, when DBA mice produced significantly greater levels than C3H mice. Collectively, these data suggest that, although they posses the ability to produce proresolution and anti-inflammatory molecules, C3H mice may have a functional defect in the metabolism of precursor molecules and a higher threshold for the response to this group of metabolites, resulting in less effective control of the inflammatory response to infection.  Linoleic Acid Metabolites-Linoleic acid is an 18-carbon polyunsaturated fatty acid containing only two double bonds, limiting its utility as a substrate for eicosanoid biosynthetic enzymes in comparison with arachidonic acid. CYP epoxygenation of linoleic acid forms 9,10-and 12,13-epoxyoctadecamonoenoic acid, commonly classified as leukotoxins, which are generated from activated neutrophils and macrophages and can induce cell death (54). The toxic effects of these metabolites are thought to be mediated by their conversion into the corresponding diols, 9,10-and 12,13-diHOME, by soluble epoxide hydrolase (55). Analogous to arachidonic acid-derived HETEs, both CYP and LOX can hydroxylate linoleic acid to form HODEs, which can subsequently be oxidized to form oxooctadecadienoic acids (oxoODEs). These metabolites have been shown to be potent activators of peroxisome proliferator-activated receptor ␥ (56).
In contrast to arachidonic acid, the CYP metabolites of linoleic acid metabolites did not demonstrate dramatic temporal changes in response to B. burgdorferi infection (supplemental Tables 2 and 3). Whereas 12-HETE levels steadily rose over the first 14 days of infection in C3H mice, high levels of 9-HODE and 13-HODE were produced throughout infection. In general, linoleic acid metabolites demonstrated more variability between samples than those from arachidonic acid, making it difficult to discern a temporal pattern for their production. Interestingly, in both mouse strains the amount of a given linoleic acid metabolite was typically 4 -20 times higher than its corresponding arachidonic acid metabolite (Fig. 6). This may be due to the comparatively higher amounts of linoleic acid found in mammalian tissues (57). However, given their potentially important bioactivity, these data indicate that this class of metabolites warrants consideration and further investigation.
Arthritis-susceptible Mice Have an Altered Basal Eicosanoid Profile-Although the peak levels of many COX, LOX, and CYP metabolites were comparable between the two strains, several eicosanoids in DBA joints exhibited basal production levels above those seen in C3H mice. In light of the fact that many of these lipids are considered ''pro-inflammatory,'' it seemed counterintuitive for base-line levels of these molecules to be higher in the Lyme arthritis-resistant mice, which exhibited a milder inflammatory response as compared with the severe arthritis demonstrated in C3H mice. In some cases, this necessitated a severalfold increase in production above that seen in DBA joints in order to reach a similar tissue lipid concentration. For instance, pro-inflammatory PGE 2 was present in C3H joints at 50% of the basal levels found in the resistant DBA strain (Fig. 7A). Similarly, the anti-inflammatory EETs were all produced at significantly lower basal levels in C3H versus DBA joints (Fig. 7B). Both pro-inflammatory HXA 3 and anti-inflammatory 12-HETE basal levels were 1-2-fold higher in DBA than in C3H mice (Fig. 7C). Taken alone, decreased levels of these lipids suggest there should be less inflammation in C3H mice.
However, C3H mice also demonstrated high basal levels of LTE 4 (Fig. 7D) and little to no basal production of the proresolving protectins and RvD 1 (Fig. 7E). Although RvD 1 did not change significantly during the course of infection (supplemental Tables 2 and 3), differences in basal pro-resolution metabolite levels could alter the inflammatory tone, priming C3H mice toward a stronger inflammatory response. These surprising data suggest that the lipids produced by the microenvironment at homeostasis may be just as important as lipid production throughout the response to infection. Overall, the basal levels of several eicosanoids observed in DBA mice were higher than in the C3H mice, suggesting a mechanism for the more controlled response to B. burgdorferi infection in the DBA strain.
Lipid Metabolic Consequences of COX-2 Deficiency-The above findings illustrate the complexity of the inflammatory response to infection and suggest the potential importance of products from all of the major eicosanoid biosynthetic pathways. To demonstrate the applicability of this approach to the investigation of disease pathogenesis, we utilized this lipidomic protocol to determine eicosanoid levels in the context of COX-2 deficiency. We choose to focus on COX-2 for several reasons. In the clinical setting, patients diagnosed with rheumatoid or Lyme arthritis are typically treated with nonsteroidal anti-inflammatory drugs (NSAIDs) or COX-2-selective inhibitors such as celecoxib (Celebrex TM , Pfizer) (15). Numerous studies in animal models of rheumatoid arthritis have illustrated the importance of the COX-2 enzyme in autoimmune arthritis development (58,59). However, very little is known about the role of eicosanoids in infectious arthritides. Previously we described the development of Lyme arthritis in C3H and DBA mice lacking COX-2 activity (11). Although COX-2 deficiency resulted in attenuated ankle swelling and lower PGE 2 concentrations in the joint, the development of arthritis pathology was unaltered. Importantly, however, mice lacking COX-2 failed to resolve their inflammatory response and demonstrated prolonged arthritis pathology, as characterized by sustained neutrophil influx into the joints. These data were contrary to results in models of autoimmune arthritis, in which a lack of COX-2 activity prevented or ameliorated arthritis (58,59). These data hinted at the complexity of eicosanoid involvement in regulating the immune response and illustrated the need for in-depth lipidomic analysis.
COX-2 Ϫ/Ϫ and wild-type (WT) littermate controls were infected with B. burgdorferi, and lipidomic analysis was performed on ankle joints at d0 and d10 post-infection. This phase of the inflammatory response was chosen because basal levels and early events during infection appear to be the key to orchestrating the subsequent immune response and the outcome of arthritis pathology (11). Global eicosanoid profiles from WT and COX-2 Ϫ/Ϫ C3H mice identified a number of off-target changes ( Fig. 8A and supplemental Table 5). As expected, PGE 2 and PGD 2 levels were significantly reduced in COX-2 knock- out mice; likewise, the amounts of the corresponding breakdown metabolites were also decreased (Fig. 8B). The loss of COX activity had no significant effect on the CYP-derived EETs or the 12-LOX metabolites HXA 3 and 12-HETE. Surprisingly, elimination of COX-2 activity resulted in a concomitant drop in pro-inflammatory 5-LOX products LTC 4 , LTE 4 , and 5-HETE. Although at this early time point there was an equal level of infiltrating neutrophils, a key cellular source of LTs, between WT and COX-2 Ϫ/Ϫ the levels of 5-LOX activity dropped dramatically in the tibiotarsal joints of the COX-2 Ϫ/Ϫ mice. Thus, during the response to B. burgdorferi infection, arachidonic acid shunting into other eicosanoid pathways may not play a dominant role, in direct contrast to results found with ex vivo cell culture systems (60) or in vivo molecular stimuli such as bacterial lipopolysaccharide (61). For these reasons, the investigation of inflammatory signaling in infectious disease models, such as experimental Lyme arthritis, coupled with lipidomic analysis may redefine the role that different eicosanoid cascades play on a systemic level during an immune response.

DISCUSSION
Although steps have been taken to increase the number of eicosanoids included in lipidomic-style screens, full scale lipidome analyses have been lacking. Previous attempts at eicosanoid lipidomics have focused on singular pathways or phases of the inflammatory response (10,14,62). Although studies of this type increase our knowledge of the inherent potential of certain molecules, they provide an incomplete picture and give the impression that some lipids function in isolation. In reality, they are likely governed by up-stream mediators and, in turn, regulate other metabolites to produce the overall outcome of the inflammatory and immune responses.
Herein we have described a protocol for lipidomic analysis of nearly the entire known eicosanoid lipidome and applied this method to the investigation of infection with a clinically relevant pathogen, B. burgdorferi. By initially utilizing two mouse strains with differential susceptibility to the development of Lyme arthritis, analysis of the inflammatory response found production of not only ''the usual suspects,'' such as PGE 2 and PGD 2 , but also discovered the production of eicosanoids not previously described during a normal immune response, including CYP and 12-LOX products 11,12-EET and PD 1 , respectively. Lipidomic characterization of Lyme arthritis-resistant and -susceptible strains also revealed several unexpected findings. The high basal production of many pro-and anti-inflammatory metabolites in DBA relative to C3H joints implied that the initial setting in which an inflammatory response takes place may be crucial for the eventual disease outcome. Additional analysis found that, although C3H joints contained lower levels of some anti-inflammatory and pro-resolution molecules, they were able to produce high concentrations of precursor molecules, indicating a defect not in the initiation of this particular metabolic cascade but in its completion and intensity.
Analysis of joints from Lyme arthritis-susceptible COX-2 Ϫ/Ϫ animals at early time points post-infection dramatically demonstrated the consequences of perturbing eicosanoid biosynthetic pathways. Previously, inhibition of COX enzymes was expected, at most, to increase production of LOX metabolites due to the possibility of substrate shunting. It is therefore of FIGURE 9. Lipid mediators regulate all phases of B. burgdorferi infection. Pro-inflammatory COX metabolites PGE 2 and PGD 2 are produced early to initiate various aspects of the innate and acquired immune response, whereas 15d-PGJ 2 is produced as an anti-inflammatory counterbalance. 12-LO-derived HXA 3 participates in neutrophil recruitment as 12-HETE modulates monocyte recruitment preceding antibody-mediated Borrelia clearance from the joints. DHA-derived protectins collaborate in the control and subsequent resolution of inflammation via the regulation of both leukocyte recruitment and macrophage phagocytosis of apoptotic cells. Throughout infection, EETs control both vascular permeability and the recruitment of mononuclear cells. Decreased production of these and other lipid mediators is indicative of the successful resolution of the inflammatory response and the return to homeostasis. great interest that our lipidomic analysis revealed significant decreases in not only COX but 5-LOX products, an effect not previously described in response to the perturbation of the COX-2 pathway.
Based upon our lipidomic analysis of Lyme arthritis-resistant and -susceptible wild-type strains and subsequent analysis of arthritis-susceptible COX-2 Ϫ/Ϫ mice, and using in vitro and in vivo knowledge of eicosanoid function, we propose the following model to describe the progression of Lyme arthritis (Fig. 9). Early during infection, prostaglandins, leukotrienes, and HXA 3 may cooperate to induce the release of additional pro-inflammatory factors, facilitating the recruitment of neutrophils into the joint and the generation of antibodies to combat B. burgdorferi infection. Concomitantly, 15d PGJ 2 and EETs would counterbalance this response by activating peroxisome proliferatoractivated receptor ␥ and inactivating NF-B, and these are aided later by DHA-derived, pro-resolution protectins. Corresponding to the peak bacterial levels, 12-HETE production peaks, possibly inducing the recruitment of anti-inflammatory macrophages. Once bacteria and neutrophils have been cleared from the site of infection, many of these molecules slowly return to basal levels, indicating a return to homeostasis. Results utilizing COX-2 Ϫ/Ϫ mice further suggested that, although COX metabolites are produced early during infection, COX, and perhaps LOX, products may also be necessary for the subsequent induction of resolution. Undoubtedly, future studies will refine this model; however, it serves as an excellent starting point for probing the complex manner in which eicosanoids collaborate to coordinate an inflammatory immune response.
Susceptibility or resistance to experimental Lyme arthritis does not necessarily translate to susceptibility in other disease models. Interestingly, DBA/2 mice are more susceptible to the development of atherosclerosis and Pseudomonas aeruginosainduced lung injury than are C3H/He mice (63,64). Tissuespecific cytokine profiles have identified a few differences in protein production; however, lipidomic analysis may reveal tissue-to-tissue differences in the production of eicosanoids that regulate cytokines and chemokines. Just as vascular permeability in the joint differs from that in the skin (65), eicosanoid production likely varies as well. Thus, it will be of interest in the future to investigate basal and induced lipid production in different tissues to determine susceptibility to tissue-specific diseases.
A comprehensive eicosanoid analysis as described herein has a number of attractive clinical applications. In the case of Lyme arthritis, the interruption of specific eicosanoid signals significantly affected the immune response by modulating the recruitment of leukocytes to the infected tissue and their activation. Although interrupting COX signaling reduced one of the cardinal signs of inflammation, swelling, it did not prevent the development of arthritis and had the unintended consequence of impeding resolution by creating global changes in the eicosanoid profile. Studying eicosanoid signaling as a complete system can identify such unforeseen molecular changes. The robustness of this technique lends potential for its easy application beyond the scope of Lyme arthritis to other tissues and diseases. In addition to investigating the mechanisms behind a given disease pathology, this technology could easily be translated into a diagnostic tool for assessing individual patients. The identification of differential eicosanoid production among individual tissue types may open the gate for the development of more disease-specific tissue-targeted therapies and would facilitate the progress of individualized medicine, resulting in more effective treatments with a lower incidence of side effects. Already, genetic screening and microarray technology have philosophically influenced our clinical understanding and treatment of disease (66). A lipidomic platform on a par with this technology has the potential to make an equally profound and complimentary impact. The power of the comprehensive eicosanoid metabolomic analysis described herein lies in its capacity to investigate a naturally progressing inflammation and its potential to translate to clinical diagnosis and treatment of disease.