Genomic, Lipidomic and Metabolomic Analysis of Cyclooxygenase-null Cells: Eicosanoid Storm, Cross Talk, and Compensation by COX-1

The constitutively-expressed cyclooxygenase 1 (COX-1) and the inducible COX-2 are both involved in the conversion of arachidonic acid (AA) to prostaglandins (PGs). However, the functional roles of COX-1 at the cellular level remain unclear. We hypothesized that by comparing differential gene expression and eicosanoid metabolism in lung fibroblasts from wild-type (WT) mice and COX-2-/- or COX-1-/- mice may help address the functional roles of COX-1 in inflammation and other cellular functions. Compared to WT, the number of specifically-induced transcripts were altered descendingly as follows: COX-2-/- > COX-1-/- > WT + IL-1β. COX-1-/- or COX-2-/- cells shared about 50% of the induced transcripts with WT cells treated with IL-1β, respectively. An interactive “anti-inflammatory, proinflammatory, and redox-activated” signature in the protein–protein interactome map was observed in COX-2-/- cells. The augmented COX-1 mRNA (in COX-2-/- cells) was associated with the upregulation of mRNAs for glutathione S-transferase (GST), superoxide dismutase (SOD), NAD(P)H dehydrogenase quinone 1 (NQO1), aryl hydrocarbon receptor (AhR), peroxiredoxin, phospholipase, prostacyclin synthase, and prostaglandin E synthase, resulting in a significant increase in the levels of PGE2, PGD2, leukotriene B4 (LTB4), PGF1α, thromboxane B2 (TXB2), and PGF2α. The COX-1 plays a dominant role in shifting AA toward the LTB4 pathway and anti-inflammatory activities. Compared to WT, the upregulated COX-1 mRNA in COX-2-/- cells generated an “eicosanoid storm”. The genomic characteristics of COX-2-/- is similar to that of proinflammatory cells as observed in IL-1β induced WT cells. COX-1-/- and COX-2-/- cells exhibited compensation of various eicosanoids at the genomic and metabolic levels.


Introduction
Eicosanoids are lipid mediators of inflammation [1]. Cytosolic phospholipase A 2 (cPLA 2 ) cleaves some membrane lipids to generate 20-carbon arachidonic acid (AA), which can be channeled toward eicosanoid synthesis [1]. The generation of AA by cPLA 2 is a rate-limiting step for biosynthesis of eicosanoids, which include prostaglandins (PGs), thromboxanes (TXs), and leukotrienes (LTs). PGs and TXs, which are collectively referred as prostanoids, are generated by the two isoforms of cyclooxygenases (COX), COX-1 and COX-2, whereas LTs are generated by 5-lipoxygenase (5-LOX) [1,2]. Although both COX enzymes generate PGs, COX-1 is constitutively expressed in most normal cells at the basal level and is considered to regulate a number of housekeeping functions such as vascular hemostasis, renal blood flow, and glomerular function [1,2]. On the other hand, COX-2, which is inducible in nature, is mostly expressed at sites of inflammation and cancer [1,2]. Expression of COX-2 can be triggered by cytokines, growth factors, and other proinflammatory stimuli [2]. Cell-and/or tissue-specific gene expression of COX-1 and COX-2 has been reported in different organs and tumors [3][4][5]. Moreover, COX-2 overexpression induces tumors in COX-2 transgenic mice, whereas the COX-2 inhibitors coxibs induce oxidative stress and reduce polyps associated with colon cancer [2,6,7].
Fibroblasts have a wide distribution throughout the body. Owing to their ability to secrete extracellular matrix (e.g., collagen) and release proinflammatory mediators [8], fibroblasts are involved in various pathophysiological conditions such as arthritis, cancer, lung fibrosis, wound healing, and stem cell maturation [8]. Fibroblastic COX-1 and COX-2 activity [8] can be modulated by various receptors such as PGE receptors (EP receptors) and Toll-like receptors (TLRs) [1,2,8]. These receptors may influence eicosanoid synthesis, growth, chemotaxis, matrix, and matrix metalloprotease synthesis [1,2]. We therefore chose to use fibroblasts over macrophages for the present study, in line with our previous investigations on the regulation of COX and PGs in peripheral blood leukocytes, chondrocytes (differentiated fibroblasts), and synovial fibroblasts in human arthritis [9][10][11]. We expect our systems approach (genomic, metabolomic, protein-interatomic, and proteomic) from the same cells will allow identification of common and distinct functions of the two closely related COX isozymes [12,13] in sterile inflammation, innate immune response, and homeostasis in the cells.
The uncontrolled production or defective expression of COX-1 or COX-2 has been recognized as a health risk [14,15]. Similarly, nonsteroidal anti-inflammatory drugs (NSAIDs) and COX-2 inhibitors (coxibs) also exhibited health risk in large-scale population studies [16,17]. The Food and Drug Administration (FDA) of the United States has recommended that coxibs be avoided in individuals with an elevated risk of cardiovascular disease (CVD) and in patients with established CVD [16,17], since the drugs increased the risk of ischemic CVD, heart failure, increased blood pressure, and cardiac arrhythmia [2,16,17]. One explanation of these adverse events is the differential accumulation of TXs, LTs, and PGs in the presence of coxibs and/or decreased level of prostacyclins [2,[6][7][8][9][10][11] and/or ''non-PG effects" of COX inhibitors [18][19][20]. Inhibition of COX-2 by coxibs in human arthritis not only curbs COX-2-mediated PGs but also shifts prostanoid synthesis toward the COX-1-mediated pathway [3,10]. Unlike COX-2, the participation of COX-1 has been subverted, for its involvement in inflammation [1,2]. For example, the anti-inflammatory effects of a highly-selective coxib were evident only when administered at doses that are inhibitory for COX-1, indicating a substantial contribution of COX-1 in inflammation [2].
In the present study, we report the alternation in gene expression and end products of lipid synthesis for COX-1, COX-2, and 5-LOX pathways in COX-1 -/-(COX-1-ablated) and COX-2 -/-(COX-2-ablated) cells. We demonstrate the dominant role of COX-1 at the genomic, redox, and metabolomic levels in COX-2-null cells. In addition, gene array data also agree with the metabolic activity of eicosanoids and redox changes in COX-2 -/cells.
Mouse fibroblast cells were obtained from WT cells (WT), COX-1 -/-, and COX-2 -/mice. In addition, WT cells stimulated with 10 ng/ml IL-1b served as a positive control to procure an ''IL-1b inflammatory signature" at the genomic level. All the cells were subjected to gene expression arrays [26][27][28] and as shown in Figure 1A. We implemented bioinformatics analysis to identify specific transcripts that were upregulated by 1.75fold or downregulated by 0.5-fold in COX-1 -/-, COX-2 -/-, and IL-1b stimulated WT cells, respectively, when compared to basal gene expression in WT cells. Hierarchical clustering of gene transcripts showed altered gene expression in COX-1 -/and COX-2 -/cells as compared to WT cells ( Figure 1B; Figure S1 for high resolution and gene annotations). These also include COX-1 and genes involved in eicosanoid synthesis, inflammation, homeostasis, and cell cycle in COX-2 -/cells.

Gene expression arrays of eicosanoid metabolism
To understand the role of COX-1, we focused on the alterations in gene expression in COX-2 -/cells. Sixty-five transcripts were identified to be involved in eicosanoid, lipid, and redox metabolism pathways. These transcripts were grouped into different eicosanoid pathways as shown in the heat map of differentially-expressed transcripts from WT, WT + IL-1b, COX-1 -/-, and COX-2 -/cells ( Figure 2). More than 25% of these 65 transcripts that were upregulated in IL-1b stimulated WT cells were also spontaneously augmented in COX-2 -/cells.
The genes of PLA 2 -group [29], including Pla 2 g4, Pla 2 g6, and Pla 2 g7, were highly expressed in COX-2 -/followed by COX-1 -/-, IL-1b + WT, and WT cells ( Figure 2 and Table S2). Compared to WT and COX-1 -/cells, increased expression of genes encoding PGE synthase (Ptges) and PGI 2 (as called prostacyclin) synthase (Ptgis) was detected in COX-2 -/cells. key enzymes in the synthesis of various LTs [30,31]. There was an increase in genes encoding various GST isoforms, including GST theta 1 (Gstt1), GST alpha 4 (Gsta4), GST kappa 1 (Gstk1), and microsomal GST 1 and 2 (Mgst1, Mgst2), in COX-2 -/cells as compared to WT, WT + IL-1b, and COX-1 -/cells. Similarly, increased expression was also revealed for genes encoding superoxide dismutase (Sod), peroxiredoxin (Prdx), and proteins involved in the The differentially-expressed transcripts were clustered into different pathways. Gene expression is represented as fold change in relation to WT cells from low to high in color gradient (from green to red). Red indicates a higher expression, while green indicates a lower expression. The average baseline fold change of all the three housekeeping transcripts was 1.0 ± 0.3.

Network analysis
The Database Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) contains information on about 5.2 million proteins in 1133 species from experimental data, computational prediction methods, and public text collections. Data are weighted and integrated with a confidence score calculated for all protein interactions. STRING allows merging of gene expression arrays and metabolic activity at the level of protein-protein interactome [13,34]. Using the edge information present in the STRING database, protein-protein interaction map ( Figure 3) was derived from the gene expression data in this study. In WT cells, there exist interactions of different proteins in the eicosanoid pathways ( Figure 3A), representing the background signal of WT cells. The map shows the interactions among the functional genes/proteins associated with 5-, 12-, and 15-LOXs, COX-1 and COX-2, and lipid metabolism. Figure 3 showed an increase in gene/ protein activities in the following order of magnitude: COX-2 -/-> COX-1 -/-= IL-1b + WT > WT cells. The differences in redox-related protein were more pronounced in COX-2 -/cells ( Figure 3D). For example, expression of NAD(P)H dehydrogenase quinone 1 (NQO1) and its receptor aryl hydrocarbon receptor (AhR) was increased in COX-2 -/as compared to WT, COX-1 -/-, and IL-1b-stimulated cells. NQO1 is induced during cellular stress, detoxification, tumors, tardive dyskinesia, and hematotoxicity [35,36]. AhR has been shown to regulate xenobiotic-metabolizing enzymes such as cytochrome P450, which is involved in eicosanoid metabolism [36]. Taken together, these interaction maps showed that unlike WT and COX-1 -/cells, COX-2 -/cells exhibited a distinct proteinprotein interaction signature.

Lipidomic studies and strategy
Accumulation of PGE 2 and PLA 2 in COX-1 -/and COX-2 -/cells We integrated the omics analysis as shown in Figure S4 [37]. We next selected stable end-metabolites (from WT and COX mutants), which can be quantitatively assayed and may also represent pathways involved in eicosanoid synthesis. The expression levels of genes encoding isoforms of PLA 2 of the groups VII (Pla 2 g7), IVA (Pla 2 g4), and VI (Pla 2 g6) were spontaneously upregulated in the following order: COX-2 -/-> COX-1 -/-= IL-1b + WT cells ( Figure 2 and Table S2). Basal levels cPLA 2 were not detected by western blot analysis in unstimulated fibroblasts in this and other studies [3,29]. Nevertheless, COX-1 -/and COX-2 -/cells (and WT + IL-1b cells [3,38]) showed upregulation of cPLA 2 as determined by Western blot analysis ( Figure 4). These data not only confirmed previous observations [29] with respect to the functional role of cPLA 2 , but also identified new classes of PLA 2 : the lipoprotein-associated phospholipase A 2 (Lp-PLA 2 ) encoded by Pla 2 g7, which is associated with eicosanoid metabolism in these mutant fibroblasts. Consistent with the upregulated expression of cPLA 2 in COX-ablated cells, the levels of PGE 2 were significantly increased (3-4-folds) in COX-1 -/and COX-2 -/cells, although it is present at a basal level in WT cells ( Figure 4A). Similar observations were made after normalizing the amount of PGE 2 with the cellular DNA content ( Figure 4B). We obtained the similar results as those reported by Ballou et al. [3]. The increased levels of PGE 2 was accompanied by increases in the expression of Ptgs1, Ptgs2, Acsl4, Elovl6, Fads1, Plscr1, Plscr2, Ptgis, and Ptges, which are involved in lipid metabolism ( Figure 2 and Table S2). These results suggest an increase in the synthesis and compensation of the dominant PG (e.g., PGE 2 ) with other enzymes in lipid metabolism by COX-1 and COX-2 pathways in the absence of either COX isoforms.

Regulation of PGE 2 in COX-1-and COX-2-ablated cells
We then examined if the expression of COX-1 and COX-2 was sensitive to inhibitors (indomethacin) and activators (IL-1b and/or AA) of COX-1 and COX-2 in WT, COX-1 -/-, and COX-2 -/cells ( Table 1). We found that indomethacin significantly inhibited PGE 2 accumulation in COX-1 or COX-2ablated cells, confirming that both enzymes were independently involved in the generation of PGE 2 [1,2]. On the other hand, we also tested the effects of the endogenous activator AA ( Figure 5) and found that AA significantly increased the levels of PGE 2 in WT and COX-1 -/cells, but not in COX-2 -/cells. These experiments show the importance of endogenous AA as a rate limiting step for regulation of COX-1 and COX-2 pathways [1][2][3]39].
Analysis of prostanoids in COX-1 -/and COX-2 -/cells PGE 2 is one of the dominant prostanoids in COX-ablated fibroblasts as observed in other cell types [1,2]. We examined the production of PGD 2 , TXB 2 , 6-keto-PGF 1a , and PGF 2a , which has not been examined previously in these COX mutants ( Figure 6). The amount of PGD 2 was significantly increased in the COX-1 -/and COX-2 -/cells as compared to that in WT cells. Compared to WT cells (around 170-180 pg/ml), the amount of TXB 2 , a stable and an inactive metabolite of TXA 2 , was significantly increased in COX-2 -/but not COX-1 -/cells. Different from PGD 2 and TXB 2 , PGF 1a was identified in low concentrations in WT cells, but increased significantly in both COX-1 -/and COX-2 -/cells. On the other hand, PGF 2a levels were relatively high in WT cells but significantly increased in COX-2 -/cells, while only marginal increase was detected in COX-1 -/cells. In addition, there was a 1-5-fold increase in several prostanoids (like PGE 2 ) in the presence of exogenous AA (data not shown). In summary, as compared to WT cells, the levels of TXB 2 and PGF 2a were significantly higher in COX-2 -/cells but not COX-1 -/cells. There was a compensation of basal levels of PGE 2 , PGD 2 , and PGF 1a in COX-1 -/and COX-2 -/cells due to increased activity of cPLA 2 and high levels of common substrate: AA in both mutants. This leads to an imbalance in the basal levels of prostanoids in the mutants (by the COX-1 and COX-2 pathways) as compared to WT cells.

Cross-talk between COX and 5-LOX pathways in COX-2 -/cells
The shift from PGE 2 to LTB 4 in cytokine-activated cells in the presence of coxibs has been reported in human arthritisaffected tissues and cells [9,40]. In the present study, we compared the contributions of COX-1 and COX-2 in the 5-LOX pathways with and without any stimulation (Figure 7). COX-2 -/and COX-1 -/cells showed a spontaneous release of LTB 4 , although it was much higher in COX-2 -/cells. In addition, extra AA or IL-1b significantly increased the levels of LTB 4 accumulation in COX-1 -/cells, which is not the case for COX-2 -/cells. In summary, we demonstrated the differential and preferential contribution of LTB4 synthesis by the COX-1 pathway over the COX-2 pathway.

Discussion
The present study describes the common and dissimilar functional interactions of the COX isozymes at the cellular level. Previous pharmacogenomics analysis showed inhibition of PGE 2 by coxib in human arthritis-affected blood cells and cartilage, resulted in altered gene expression [9][10][11]. Coxibs or statins with different chemical structures and similar pharmacological targets, elicited distinct side effects, changes in protein profile and gene expression [1,2,[16][17][18][19][41][42][43]. It should be noted that altered gene expression incited by coxibs may not necessarily account for the inhibition of COXs and PGs, since some effects of coxib treatment cannot be reversed by the addition of exogenous prostanoids [18][19][20]42,43]. Furthermore, the differential physiological actions of coxibs for curbing pain and inhibiting colon cancer are concentration dependent [1,2].

Role of COX-1 and COX-2 in inflammation and inflammation resolution
IL-1b and other members of the growing superfamily of IL-1 are involved in sterile inflammation under various pathophysiological conditions [21][22][23][24][25]. Approximately 50% of transcripts modulated in COX-1-or COX-2-ablated cells were also observed in the ''IL-1b inflammatory signature". Since inflammasome and COX-2 pathways are induced by IL-1b, the Gene expression is represented as fold change in relation to WT cells from low to high in color gradient (from green to red; fold change = 0-2.5). Fold change is set as 1 for WT. The highest intensity red color in this scale indicates that fold change P2.5-fold that of basal levels.
intricate roles of COX-2 during sterile inflammation and innate immunity were not surprising [23,24]. However, the inflammatory response exhibited by the upregulated COX-1 expression (in the absence of COX-2 activity) was certainly perplexing. These results further support the previously overlooked role of COX-1 within and beyond eicosanoid metabolism and inflammation. Others and we have shown that low levels of PGE 2 exhibit proinflammatory activity by activating NFjB pathway [44]. However, prolonged activation of COX by IL-1b and high levels of PGE 2 foster inflammation resolution by inhibiting the NFjB pathway [45,46]. Prolonged and a constitutive upregulation of PGE 2 (in COX-ablated cells and IL-1b-stimulated WT) triggered the upregulation (by 2-3fold) of a gene (Nfkbiz), which inhibits NFjB functions [44].

Role of PLA 2 isoforms and COX-1 in inflammation
In the current studies, the differential expression of various PLA 2 isoforms in COX-1 -/and COX-2 -/cells were induced by the exclusive COX-2 or COX-1 pathways, respectively.  Note: * indicates significant difference in the amount of PGE 2 (mean ± SEM) spontaneously released by cells treated with indomethacin (5 lg/ml) in relation to the respective untreated control cells as determined by Student's t-test, n = 3 (P < 0.05).
Our studies suggest that COX-1 upregulates gene expression and metabolic activity, which results in a significant increase in the synthesis of prostanoids and redox-related activity in COX-2 -/cells. It should be noted that COX-2 -/fibroblasts showed a nearly fourfold increase in COX-1 gene expression as compared to WT cells. Although COX-1 is regarded as a ''constitutive enzyme" [1][2][3], COX-1 gene expression is upregulated in the colon and ovarian cancers [53,54]. Estradiol also stimulates gene expression of PLA 2 and COX-1 in endothelial cells [55]. Von Moltke et al. studied the role of COX-1 during systematic inflammation induced by flagellin in mice [56]. They showed that COX-1-derived products drive the initial phase of the inflammatory process, whereas COX-2 upregulation followed a few hours later [56]. These observations highlight the new role of COX-1-generated eicosanoids and physiological consequences in vivo and also support the hypothesis that the basal activity of COX-1, when placed in a pivotal position, can mount an inflammatory response with an ''eicosanoid storm" to compensate for the multifunctional eicosanoids, lipids, and redox-related mediators.  was evaluated by ELISA. Significant alteration in the amount of LTB 4 (P < 0.05) between COX-1 -/and COX-2 -/cells is indicated with # , while difference between the treated and untreated samples in the same cell groups is indicated with * . All data are expressed as mean ± SEM (n = 3). Significance levels were calculated using Student's t-test.

Regulation of metabolites by
eicosapentaenoic acid in addition to AA [37]. Another noteworthy difference between COX-1 -/and COX-2 -/fibroblasts is their ability to synthesize different quantities and forms of PGs, TXs, LTs, and lipids. Changes in levels of eicosanoids may regulate each other. For example, an increased production of PGD 2 in human arthritis-affected tissues increases the accumulation of PGE 2 , PGF 1a , PGF 2a , and TXB 2 but inhibits LTB 4 [57]. The expression of COX-1 can have far-reaching effects on lipid metabolism. For example, Ma et al. reported significant changes in the levels of phosphatidylserine, triacylglycerol, and cholesterol, which alter the cholesterol-tophospholipid ratio in COX-2 -/mice and may impart resistance to neuroinflammation [58]. Thus increasing the activity of COX-1 (in COX-2 -/cells) not only regulates eicosanoid synthesis but also augments lipid metabolism as described in Figure 2. An increased activity of COX-1 increased the levels of LTB 4 in COX-2 -/cells that were comparable to those induced by IL-1b-or AA-induced COX-1 -/cells. Similar effects were observed in human osteoarthritis-affected cartilage that spontaneously released PGE 2 and LTB 4 in ex vivo models [9]. The proinflammatory activity of LTB 4 is well documented in fibroblasts. Expression of TNFa and IL-1b is increased by LTB 4 but inhibited by LTB 4 inhibitors, such as MK886 and bestatin [1,2].

Increase in redox molecules in COX-ablated cells
The increase in GSTs in COX-2 -/cells was higher than that in COX-1 -/cells, followed by that in WT cells. GSTs contribute to the detoxification of xenobiotics and endogenous peroxided lipids, and biosynthesis of LTs [30][31][32]. Peroxiredoxins function as antioxidant enzymes that also control inflammation-induced peroxide levels by removing H 2 O 2 [59]. Mice lacking peroxiredoxin develop severe hemolytic anemia and hematopoietic cancers, and have shortened lifespan [59]. Microsomal PGE synthase-1 (mPGES-1) requires glutathione as a co-factor [37]. One of the most notable observations is the increase in gene expression of NQO1 and AHR. AhR-mediated NQO1 gene expression is increased by a variety of antioxidants, tumor promoters, and H 2 O 2 [35,36]. COX-2 -/cells showed an increase in the pentose phosphate pathway (HMP shunt, Amin AR unpublished data). The HMP shunt is a metabolic redox sensor and modulates gene expression during an anti-oxidant response [60]. These types of cellular events that lead to oxidative stress   Tables S1 and S2. and ''Warburg effect" [60] are interlinked with the regulation of COXs, NFjB activation during inflammation induces intracellular ROS [46,47,60]. COX-1 peroxidase activity has been reported to serve as an intracellular signal leading to NFjB activation [47,60]. The possibility that the products of COX-1 and COX-2 may also be sensors and/or inducers for such imbalanced redox activity is compelling. The current understanding of COX-1 in inflammation and homeostasis is constantly evolving. These studies advocate that COX-1 may participate in regulating sterile inflammation, which is more latent and subtle. There are significant collaborations and functional redundancies between COX-1 and COX-2. For example, the essential and collaborative role of both COX-1 and COX-2 has been observed in ear inflammation and arthritis models of mice [1,61]. In addition, in the K/BxN serum-transfer model of arthritis, COX-1-derived PGs, in particular PG1 2 , contributed remarkably to initiating and prolonging the pathology [1]. Similarly, both COX isoforms contributed to PG production in the carrageenan model of inflammation, depending on the type of stimulus and tissuespecific expression of COXs [1]. The coordination and collaboration of the COXs remain well balanced, especially in vivo, in an inflammatory process, but may depend upon the availability of the COX isoform, time course, and intensity of the inflammation. It is apparent that if either COX isoform is functionally unavailable, the other becomes pivotal and attempts to compensate the events with a different intensity and/or tissue specificity [9], as observed in COX-1 and COX-2 knockout mice and fibroblasts [3][4][5][6][7][8][9][10]14,15,39]. Figure 8 summarizes the changes observed in WT, COX-2 -/-, COX-1 -/cells based on genomics and metabolomics studies described in this study These investigations paint a broader picture of eicosanoid synthesis and signaling, gene expression, and lipid metabolomics, which can identify and connect unforeseen functions of COX-1 and COX-2.

Conclusion
This study describes the functional role of COX-1 and COX-2 at the cellular, metabolic, and genomic levels. Dysregulation of COX-1 or COX-2 results in a burst of gene expression within and beyond eicosanoid and redox metabolism in fibroblasts. The changes in basal gene expression of upregulated COX-1 in COX-2 -/cells exhibited a mixture of pro-and antiinflammatory response, as well as an activated redox signature at the genomic level. The basal levels of COX-1 > COX-2 are involved in channeling LTs from COX to 5-LOX pathway. There is compensation of eicosanoids and redox metabolism by COX-1 and COX-2 and vice versa. This systems approach allows the integration of genomic, bioinformatic, and lipidomic datasets. It opens new areas of biomarkers and COX regulation beyond eicosanoid metabolism, which would help to develop safer next generation of coxibs by taking into consideration the board impact of COX genes on other metabolic pathways.

Labeling and hybridization of microarray gene chips
Total RNA was isolated and labeled with the ENZO BioArray High-yield RNA transcript labeling kit (Affymetrix, Santa Clara, CA) and then hybridized using U74Av2 Murine Genome Array (Affymetrix) as described [26]. Genes representing 12,492 transcripts from the A chip and those annotated till 2012 were identified as present calls as per the manufacturers' recommendations.

Normalization of the microarray data and bioinformatics analysis
The Affymetrix microarray image files for each sample were imported into the Affymetrix Expression Console Software package (www.affymetrix.com) to calculate the robust multiarray analysis-normalized gene expression signals for 12,488 mouse gene probe sets on the U74Av2 arrays. The binary logarithm signal values were then exported into a MS Excel spreadsheet for data analysis. The log transformed expression values ranged from 6.2 to 15.4. Since the median standard deviation of the logged signal values across all probe sets for the replicate samples was small (<0.03), the fold change (FC) of the gene expression was computed by simply averaging the logged signal values for the replicate samples for comparison with the logged expression values for a single WT control sample. Our functional genomics analysis focused primarily on genes with FC P1.75-fold (upregulation) and 60.5-fold (downregulation), in relation to WT [26][27][28]. The normalization was verified using four housekeeping genes encoding glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and b actin, as well as ribosomal proteins L30 (RPL30) and S13 (RPS13), which showed similar expression in all samples [64]. The heatmaps were constructed using Gitools [65] with gene expression in the WT taken as the baseline for each transcript.

Hierarchical clustering
The hierarchical clustering of median expression value for every gene was calculated and subtracted from the absolute expression of each gene in each aforementioned experimental condition (median centering of genes). Using the tool ''Genesis" and the median-centered gene expression matrix, the hierarchical cluster of genes (Pearson's correlation distance and average linkage) was determined and the annotated data were presented as heatmaps.

Interaction network
Interactions among proteins encoded by the selected group of eicosanoid metabolism genes were presented in a network model. Edge information among the nodes was extracted using the online tool STRING (v9.1) [66] and the interaction network was visualized using Cytoscape (v 3.1) [67].

Effects of AA on eicosanoid synthesis in COX-ablated cells
Cells were seeded in triplicate at a concentration of 10,000 cells/cm 2 in 6-or 24-well plates. After treatment with 0.5 lM AA for 15 min, the cells were washed and replenished with fresh medium. The levels of eicosanoids were monitored since then and the experiments were terminated between 18 and 24 h before the fibroblasts start the exponential growth. Release of eicosanoids into the medium was subsequently estimated using ELISA or RIA as previously reported [8,9,63].

Statistical analysis
The statistical analyses were performed using GraphPad Software (v1.14) (San Diego, CA). Student's t-test was used to analyze the data unless otherwise specified. Data from each experiment are represented as mean ± standard deviation.
Differences between the mean values of the control and experimental groups are considered significant with P < 0.05.