Macrophage responses to lipopolysaccharide are modulated by a feedback loop involving prostaglandin E2, dual specificity phosphatase 1 and tristetraprolin

In many different cell types, pro-inflammatory agonists induce the expression of cyclooxygenase 2 (COX-2), an enzyme that catalyzes rate-limiting steps in the conversion of arachidonic acid to a variety of lipid signaling molecules, including prostaglandin E2 (PGE2). PGE2 has key roles in many early inflammatory events, such as the changes of vascular function that promote or facilitate leukocyte recruitment to sites of inflammation. Depending on context, it also exerts many important anti-inflammatory effects, for example increasing the expression of the anti-inflammatory cytokine interleukin 10 (IL-10), and decreasing that of the pro-inflammatory cytokine tumor necrosis factor (TNF). The tight control of both biosynthesis of, and cellular responses to, PGE2 are critical for the precise orchestration of the initiation and resolution of inflammatory responses. Here we describe evidence of a negative feedback loop, in which PGE2 augments the expression of dual specificity phosphatase 1, impairs the activity of mitogen-activated protein kinase p38, increases the activity of the mRNA-destabilizing factor tristetraprolin, and thereby inhibits the expression of COX-2. The same feedback mechanism contributes to PGE2-mediated suppression of TNF release. Engagement of the DUSP1-TTP regulatory axis by PGE2 is likely to contribute to the switch between initiation and resolution phases of inflammation.

In many different cell types, pro-inflammatory agonists induce the expression of cyclooxygenase 2 (COX-2), an enzyme that catalyzes rate-limiting steps in the conversion of arachidonic acid to a variety of lipid signaling molecules, including prostaglandin E 2 (PGE 2 ). PGE 2 has key roles in many early inflammatory events, such as the changes of vascular function that promote or facilitate leukocyte recruitment to sites of inflammation. Depending on context, it also exerts many important anti-inflammatory effects, for example increasing the expression of the anti-inflammatory cytokine interleukin 10 (IL-10), and decreasing that of the pro-inflammatory cytokine tumor necrosis factor (TNF). The tight control of both biosynthesis of, and cellular responses to, PGE 2 are critical for the precise orchestration of the initiation and resolution of inflammatory responses. Here we describe evidence of a negative feedback loop, in which PGE 2 augments the expression of dual specificity phosphatase 1, impairs the activity of mitogen-activated protein kinase p38, increases the activity of the mRNA-destabilizing factor tristetraprolin, and thereby inhibits the expression of COX-2. The same feedback mechanism contributes to PGE 2 -mediated suppression of TNF release. Engagement of the DUSP1-TTP regulatory axis by PGE 2 is likely to contribute to the switch between initiation and resolution phases of inflammation.
The cyclooxygenase enzymes COX-1 and COX-2, encoded by the genes Ptgs1 and Ptgs2, catalyze rate-limiting steps in the synthesis of various prostanoid signaling molecules from the lipid precursor arachidonic acid 1, 2 . COX-1 is constitutively expressed by many cells. In contrast, COX-2 is expressed at low levels by the majority of cells, but transiently induced in response to growth factors, stresses or pro-inflammatory stimuli. Prostaglandin E 2 (PGE 2 ) is the major downstream product of COX-2-mediated arachidonic acid metabolism in many cells. PGE 2 increases blood flow, vascular permeability and nociception, thereby contributing to all four of the cardinal signs of inflammation; redness, swelling, heat and pain. The pro-inflammatory actions of PGE 2 underlie the clinical usefulness of non-steroidal anti-inflammatory drugs (NSAIDs), which inhibit both COX-1 and COX-2 2 . However, constitutive COX-1-mediated prostaglandin synthesis in the gastric mucosa helps to maintain the integrity of this vulnerable tissue 1 , which accounts for the increased incidence of gastric ulcers amongst patients using NSAIDs for prolonged periods. The rationale for the generation of COX-2-selective inhibitors was based on the assumption that COX-1 functions principally as a homeostatic enzyme, whereas COX-2 functions principally as a pro-inflammatory mediator. Selective inhibitors of COX-2 were predicted to exert anti-inflammatory effects whilst sparing gastric side effects 3 .
As we are reminded by the costly failure of the COX-2-selective inhibitor Rofecoxib (Vioxx), biology is rarely so straightforward or convenient 4 . COX-1-dependent synthesis of thromboxane by platelets promotes vasoconstriction and platelet aggregation. These prothrombotic actions are opposed by prostaglandin I 2 (prostacyclin).
The significantly increased risk of myocardial infarction and stroke in patients taking Vioxx eventually led to the withdrawal of this drug from the market. The basis of elevated cardiovascular risk is still not fully understood, but has been ascribed to an imbalance between COX-1-mediated pro-thrombotic and COX-2 mediated anti-thrombotic influences. Another confounding aspect of prostaglandin biology is that the actions of PGE 2 are not invariably pro-inflammatory; nor are the effects of COX-2-selective inhibitors invariably anti-inflammatory. PGE 2 has been shown to enhance expression of IL-10, inhibit the expression of TNF and other inflammatory mediators, and promote the differentiation of macrophages towards an alternatively-activated, anti-inflammatory M2 phenotype [5][6][7][8][9][10] . In rheumatoid synovial explant cultures or peripheral blood-derived monocytes, NSAIDs increased the expression of TNF 11 . Prior in vivo exposure to COX-2-selective inhibitors also primed human peripheral blood monocytes and mouse peritoneal macrophages for increased expression of TNF in response to an LPS challenge 11,12 .
Cellular responses to PGE 2 are mediated by four G protein-coupled receptors named EP1-EP4, which are the products of the genes Ptger1-Ptger4 13,14 . These receptors differ in their affinity for PGE 2 and in the signal transduction pathways that they engage. Both EP2 and EP4 are Gα s -linked and activate adenylyl cyclase to elevate intracellular cAMP levels. EP4 has additionally been shown to signal via PI3K. EP1, which is coupled to Gα q , signals via phospholipase C to induce calcium flux. EP3 exists in a number of distinct forms arising from differential splicing of the Ptger3 transcript, and appears to be promiscuous in terms of its signaling pathway engagement. There is clear potential for cell-specific programing of responses to PGE 2 via modulation of the expression of the four receptors or their variants. Anti-inflammatory actions of PGE 2 have generally been ascribed to EP4 and/or EP2 5, 6, 15-17 , however molecular mechanisms remain unclear. The increased expression of the anti-inflammatory cytokine IL-10 does not provide an explanation, as PGE 2 can still inhibit macrophage expression of TNF in the absence of IL-10 18,19 .
The inflammation-induced biosynthesis of PGE 2 is regulated largely at the level of Ptgs2 gene expression. The typical transient pattern of expression of Ptgs2 mRNA is only partly explained by transcriptional activation involving nuclear factor κB (NF-κB) and other transcription factors 20 . Efficient expression also requires the stabilization of Ptgs2 mRNA via the mitogen-activated protein kinase (MAPK) p38 signaling pathway, and conversely, MAPK p38 inhibitors accelerate the degradation of Ptgs2 mRNA 21,22 . This post-transcriptional regulation is mediated by an adenosine/uridine-rich element (ARE) immediately 3′ to the Ptgs2 translation termination codon. When inserted into a stable reporter mRNA, the Ptgs2 ARE confers rapid decay that is mediated by shortening of the protective poly-(A) tail (deadenylation), and can be prevented by activation of MAPK p38 [23][24][25] . This sequence element is therefore similar to MAPK p38-responsive mRNA destabilizing elements present in pro-inflammatory mRNAs such as Tnf, Csf2, Cxcl1, Il6 and many others 26 .
The mouse Zfp36 gene enodes the ARE-binding protein tristetraprolin (TTP) 27,28 . In Zfp36−/− macrophages lacking TTP protein, Ptgs2 mRNA was highly stable and could not be destabilized by a MAPK p38 inhibitor 29 . TTP binds to AREs in the 3′ untranslated regions of target transcripts and recruits a complex of deadenylase enzymes, which catalyzes shortening of the poly-(A) tail, usually as a prelude to the rapid destruction of the mRNA body. The mRNA-destabilizing activity of TTP is regulated by a phosphorylation switch 30 . Pro-inflammatory agonists and cell stresses activate MAPK p38, which in turn phosphorylates and activates MK2 (MAPK-activated protein kinase 2). MK2 phosphorylates serines 52 and 178 of TTP, resulting in the recruitment of 14-3-3 adaptor proteins, impairment of the interaction between TTP and the deadenylase complex, and consequent stabilization of target mRNAs [31][32][33] . Protein phosphatase 2A (PP2A) catalyzes the removal of these two phosphate groups and the activation of TTP. Therefore a dynamic equilibrium exists between forms of TTP that are phosphorylated or unphosphorylated at serines 52 and 178, and this equilibrium favors the stabilization of TTP-regulated mRNAs under conditions of strong MAPK p38 activity. Coupling between MAPK p38 activity and the stability of pro-inflammatory mRNAs contributes to the precise orchestration of the on and off phases of inflammatory responses [34][35][36] .
The activity of MAPK p38, and hence the phosphorylation state of TTP, is regulated by a negative feedback loop involving dual specificity phosphatase 1 (DUSP1) 37 . Pro-inflammatory stimuli induce MAPK p38-dependent expression of DUSP1, which then dephosphorylates and inactivates MAPK p38 to enforce the off-phase of the inflammatory response. Due to the failure of this feedback mechanism, Dusp1−/− mice and cells are prone to respond excessively to pro-inflammatory agonists, and over-express many inflammatory mediators, including COX-2 38,39 . The over-expression of COX-2 by Dusp1−/− macrophages is not fully understood; both transcriptional and post-transcriptional mechanisms have been proposed 40,41 . Here, we investigate the functional relationship between DUSP1, MAPK p38, TTP and COX-2. First, we use a number of genetically modified mouse strains to demonstrate that DUSP1 and MAPK p38 control the expression of COX-2 by regulating the phosphorylation of TTP at serines 52 and 178. Next we turn to the effect on macrophages of the major COX-2 product, PGE 2 . We show that PGE 2 enhances the expression of DUSP1 and thereby downregulates the expression of COX-2, creating another auto-regulatory feedback mechanism. PGE 2 also acts via the receptor EP4 to inhibit macrophage expression of TNF, in a manner that is at least partially dependent on both DUSP1 and the modulation of TTP's phosphorylation state. Despite being an authentic TTP target, IL-6 escapes negative regulation by PGE 2 . The influence of PGE 2 on Dusp1 gene expression creates a potent mechanism for context-dependent and gene-specific modulation of inflammatory responses.

Results
Ptgs2 gene expression is negatively regulated by TTP. A previous publication described high resolution mapping of TTP binding sites in the mouse macrophage transcriptome 35 . Publicly accessible data from this study (http://ttp-atlas.univie.ac.at) revealed strong binding of TTP to the 3′ UTR of Ptgs2 mRNA, which was restricted to a region immediately downstream of the open reading frame, containing a cluster of six AUUUA motifs (Fig. 1a). This region mediated regulation of Ptgs2 mRNA stability by the MAPK p38 signaling pathway, and was recognized by TTP in vitro 24,25 . Other putative TTP binding sites in the Ptgs2 3′ UTR 42 appear to be recognized by TTP poorly or not at all in mouse macrophages. RNA immunoprecipitation experiments confirmed binding of TTP to Ptgs2 mRNA in the mouse macrophage cell line RAW264.7 (Fig. 1b). We recently used homologous recombination to generate a novel mouse strain (known as Zfp36aa/aa), in which serines 52 and 178 of endogenous TTP protein were substituted by non-phosphorylatable alanine residues 43 . The mutant form of TTP could not be inactivated by MK2-mediated phosphorylation, and functioned as a constitutive mRNA destabilizing factor, decreasing the expression of several inflammatory mediators in vitro and in vivo. Zfp36aa/aa mice were strongly resistant to experimental endotoxemia and arthritis 43,44 . Zfp36aa/aa bone marrow macrophages (BMMs) under-expressed Ptgs2 mRNA at a steady-state level (Fig. 1c), particularly at later time points. The decrease in The primary transcript is illustrated below the graph, with coding exonic sequences as grey bars and non-coding exonic sequences as black bars. Positions of AUUUA motifs are represented above the transcript, and the sequence immediately 3′ to the stop codon is expanded below. Figure adapted from TTP-atlas data (http://ttp-atlas.univie.ac.at) 35 . (b) RNA immunoprecipitation was performed on whole cell lysates of RAW264.7 cells, untreated or stimulated with LPS for 2 h, using a TTP antiserum or pre-immune control (PI). Gapdh, Tnf and Ptgs2 mRNAs were measured by quantitiative PCR, and fold enrichment was calculated relative to the PI control. A representative of three similar experiments is shown. Error bars indicate SD of triplicate measurements. (c) Zfp36+/+ and Zfp36aa/aa BMMs were stimulated with 10 ng/ml LPS for the times indicated, and Ptgs2 mRNA was measured by qPCR. The graph represents mean ± SEM of three independent BMM cultures of each genotype. ***p < 0.005; Holm-Sidak method for multiple comparison. (d) Zfp36+/+ and Zfp36aa/aa BMMs were stimulated with 10 ng/ml LPS for 4 h then actinomycin D was added and Ptgs2 mRNA measured by qPCR at the intervals indicated. The graph shows mean ± SEM of three independent BMM cultures of each genotype. (e) Zfp36+/+ and Zfp36aa/aa BMMs were stimulated with LPS for the times indicated and COX-2 protein was detected by western blotting. Representative of three repeats. (f) Zfp36+/+ and Zfp36aa/aa BMMs were stimulated with LPS for the times indicated and PGE 2 in the tissue culture supernatant was measured by ELISA. The graph represents mean ± SEM of three independent BMM cultures of each genotype. ***p < 0.005; Holm-Sidak method for multiple comparison. steady-state Ptgs2 expression was accompanied by an increase in its rate of degradation (Fig. 1d). COX-2 protein was expressed at low levels in LPS-treated Zfp36aa/aa BMMs (Fig. 1e). The LPS-induced release of PGE 2 was significantly diminished in Zfp36aa/aa BMMs (Fig. 1f). Heterozygous Zfp36+/aa BMMs also under-expressed COX-2 protein (Supplemental Figure 1), consistent with our previous description of the non-phosphorylatable TTP mutant as a dominant inhibitor of inflammatory gene expression 43 . These findings confirm that Ptgs2 mRNA is an authentic, direct target of negative regulation by TTP.

DUSP1 regulates Ptgs2 gene expression by modulating TTP phosphorylation. Steady state levels
of Ptgs2 mRNA were elevated in LPS-treated Dusp1−/− BMMs compared with identically treated Dusp1+/+ controls (Fig. 2a), accompanied by an increase in Ptgs2 mRNA stability (Fig. 2b). The expression of COX-2 protein was increased in LPS-treated Dusp1−/− BMMs (Fig. 2c), as was the release of PGE 2 (Fig. 2d). We hypothesized that deletion of the Dusp1 gene increases the expression of Ptgs2 via enhanced phosphorylation and inactivation of TTP. To test this hypothesis, expression of Ptgs2 mRNA, COX-2 protein and PGE 2 was compared in wild type, Dusp1−/−, Zfp36aa/aa and Dusp1−/−: Zfp36aa/aa BMMs. At 1 h, Dusp1 gene disruption increased the expression of Ptgs2 mRNA, but targeted mutation of the Zfp36 locus was without effect (Fig. 3a). At 4 h, Ptgs2 mRNA continued to be over-expressed by Dusp1−/− BMMs, but in contrast was under-expressed by Zfp36aa/aa BMMs. Most importantly, Dusp1−/−: Zfp36aa/aa BMMs also under-expressed Ptgs2 mRNA at 4 h. A similar pattern was observed at the level of COX-2 protein (Fig. 3b) and PGE 2 biosynthesis (Fig. 3c). To investigate whether the same mechanism was relevant in vivo, mice of the same four genotypes were challenged by intraperitoneal injection of LPS, and three hours later Ptgs2 mRNA expression was measured in spleen, an organ which plays a critical role in systemic responses to endotoxin (Fig. 3c). The LPS challenge strongly enhanced splenic Ptgs2 expression in mice of all four genotypes. Higher Ptgs2 expression was seen in Dusp1−/− than in wild type control mice, but significantly lower expression was seen in both Zfp36aa/aa and Dusp1−/−: Zfp36aa/aa mice. Therefore, both in vitro and in vivo, disruption of the Dusp1 gene and dysregulation of MAPK signaling enhances Ptgs2 gene expression in a manner dependent on the phosphorylation of TTP at serines 52 and 178. The graph represents mean ± SEM of three independent BMM cultures of each genotype. ***p < 0.005; Holm-Sidak method for multiple comparison. PGE 2 modulates the phosphorylation of MAPK p38, the expression of DUSP1, COX-2, TNF and IL-6. We have performed two independent microarray experiments in primary mouse BMMs and one in primary human monocyte-derived macrophages, investigating LPS-induced changes of gene expression. The first mouse microarray experiment has been described 37 and deposited at the Gene Expression Omnibus (GSE68449), whilst the others are being prepared for submission. All three arrays have been extensively validated. According to all three experiments, both mouse and human primary macrophages expressed mRNAs encoding the prostaglandin receptors EP2 and EP4, whereas mRNAs encoding the other two members of this receptor family, EP1 and EP3, were essentially undetectable (Fig. 4a,b). In macrophages of both species, expression of Ptger4/PTGER4 mRNA was increased by LPS, whereas expression of Ptger2/PTGER2 mRNA was decreased or unaffected. Expression of EP4 in BMMs was confirmed by flow cytometry (Fig. 4c), whilst EP2 protein could not be detected using available reagents. LPS induced a modest increase of EP4 protein levels, which was statistically significant at 2 h (Fig. 4c,d).
Although PGE 2 is commonly thought of as a pro-inflammatory signaling molecule 2 , it is also known to exert anti-inflammatory effects in myeloid and other cells via activation of the cAMP pathway by EP2 and/or EP4 6,16,17,45 . Expression of the Dusp1 gene is regulated by cAMP via CREB binding sites in the proximal promoter [46][47][48][49][50] . Induction of DUSP1 expression by PGE 2 was recently demonstrated in airway smooth muscle cells 51 . We therefore investigated whether PGE 2 could modulate the expression of DUSP1 and the activity of MAPK signaling pathways in BMMs. On its own, PGE 2 weakly increased the expression of Dusp1 mRNA in BMMs (Fig. 4e), although we were not able to detect PGE 2 -induced DUSP1 protein or changes in MAPK p38 phosphorylation. LPS rapidly and transiently increased Dusp1 mRNA. Combined treatment with LPS and PGE 2 resulted in cooperative enhancement of Dusp1 mRNA levels, particularly at the peak of expression, 1 h after the stimulus. Very similar cooperative regulation of DUSP1 mRNA by LPS and PGE 2 was observed in primary human monocyte-derived macrophages (Fig. 4f).  . Exogenous PGE 2 modulates Dusp1 gene expression and MAPK p38 signaling. Expression of the mouse prostaglandin receptor genes Ptger1-4 (a) or the corresponding human genes PTGER 1-4 (b) was measured by microarray as described in Fig. 3. Graphs represent RMA ± SEM from three independent mouse or human macrophage cultures. n.s., not significant; *p < 0.05; **p < 0.01; ***p < 0.005; ANOVA. (c) Representative flow cytometry of EP4 expression in BMMs treated with LPS for 0 or 4 h. (d) MFI of EP4 expression was measured at the indicated times after stimulation of BMMs with 10 ng/ml LPS. The graph represents mean ± SEM of three independent wild type BMM cultures. *p < 0.05; ANOVA. (e) BMMs were treated for the indicated times with LPS (10 ng/ml), PGE 2 (1 nM) or both. Dusp1 mRNA was measured by qPCR, and plotted as fold increase compared to untreated controls. The graph shows mean fold increase ± SEM from three independent BMM cultures. **p < 0.01; ***p < 0.005; ANOVA. (f) Primary human monocytederived macrophages were treated, and DUSP1 mRNA was measured, as in (e). ***p < 0.005; ANOVA. (g) Dusp1+/+ and Dusp1−/− BMMs were treated with LPS (10 ng/ml) with or without PGE 2 (1 nM) for the times indicated. DUSP1, phospho-p38, COX-2 and β-actin proteins were detected by western blotting. Representative of four experimental repeats.
In Dusp1+/+ BMMs the LPS-induced expression of DUSP1 protein was transient, but was sustained by addition of PGE 2 (Fig. 4g). The specificity of the antibody is confirmed by the failure to detect corresponding bands in Dusp1−/− BMMs. In parallel with prolonged expression of DUSP1, the inactivation of MAPK p38 was accelerated by PGE 2 . In three independent experiments PGE 2 decreased LPS-induced MAPK p38 phosphorylation by a factor of 0.48 ± 0.06 and increased DUSP1 levels by a factor of 1.80 ± 0.26 (mean fold changes ± SEM) at the 2 h time point. At the 4 h time point phosphorylated MAPK p38 was readily detected in BMMs treated with LPS alone but not in those treated with LPS + PGE 2 . LPS-induced expression of COX-2 protein was also inhibited by addition of PGE 2 . In Dusp1−/− BMMs, the activation of MAPK p38 in response to LPS was prolonged and the expression of COX-2 protein was enhanced. Neither of these responses was affected by addition of PGE 2 . Hence the expression of COX-2 is regulated by a negative feedback loop that is mediated by its major catalytic product PGE 2 and dependent on the expression of DUSP1.
In both Dusp1+/+ and Dusp1−/− BMMs, exogenous PGE 2 inhibited the expression of TNF (Fig. 5a). These inhibitory effects were statistically significant at 4 and 8 h after the addition of LPS, and greater in magnitude in Dusp1+/+ than Dusp1−/− BMMs. Differential sensitivity of Dusp1+/+ and Dusp1−/− BMMs to anti-inflammatory effects of PGE 2 was confirmed in dose-response experiments (Fig. 5b). The anti-inflammatory effect was selective, since the expression of IL-6 was not decreased by PGE 2 in Dusp1+/+ macrophages (Fig. 5c). In Dusp1−/− BMMs, IL-6 expression was increased by the highest concentration of PGE 2 . Zfp36aa/aa BMMs were relatively insensitive to inhibitory effects of PGE 2 on the release of TNF (Fig. 5d), confirming that PGE 2 modulates TNF expression in part by influencing the phosphorylation of TTP. A selective antagonist of EP2 did not influence the inhibition of TNF biosynthesis by PGE 2 (Fig. 5e, columns 4-7). A selective EP4 antagonist dose-dependently reversed the suppression of TNF expression by exogenous PGE 2 (Fig. 5e, columns 8-11) but did not, on its own, increase TNF production (Fig. 5e, column 12). This suggests that EP4 is the major mediator of the anti-inflammatory effects of PGE 2 in this context. IL-10 is a potent inhibitor of TNF 52 , and was previously shown to be induced by PGE 2 18, 53, 54 . We therefore considered whether anti-inflammatory effects of PGE 2 could be mediated by increased expression of IL-10. Although PGE 2 increased IL-10 levels, in our hands the effect was modest in magnitude, and achieved statistical significance only in Dusp1−/− BMMs (Supplemental Figure 2). Dusp1−/− BMMs strongly over-expressed IL-10, but also over-expressed TNF. They also demonstrated increased IL-10 expression in response to PGE 2 , but were relatively insensitive to inhibitory effects of PGE 2 on TNF release. Therefore, TNF release appears not to be influenced by variations in endogenous IL-10 levels, possibly because of the different kinetics of expression of the two genes. As others have concluded 18,19 , IL-10 cannot account for anti-inflammatory actions of PGE 2 in this context.

The inhibitory effect of exogenous PGE 2 is strongly time-dependent. If LPS-induced secretion
of PGE 2 exerts negative feedback via EP4 to limit the expression of inflammatory mediators, the inhibition of COX-2 function and endogenous PGE 2 synthesis would be expected to increase TNF release. To test this, we first confirmed that LPS-induced release of PGE 2 was effectively eliminated by a selective inhibitor of COX-2 enzymatic activity, NS398 (Fig. 6a). The effect of NS398 on LPS-induced release of TNF was then tested in Dusp1−/− and Dusp1+/+ BMMs. Neither genotype of BMM displayed any change of TNF expression in the presence of a concentration of NS398 sufficient to inhibit LPS-induced PGE 2 release (Fig. 6b). Differences of timing could explain why the expression of TNF is sensitive to exogenously added but insensitive to endogenously produced PGE 2 . The accumulation of PGE 2 is a gradual process requiring both de novo gene expression and several catalytic steps, such that PGE 2 levels in cell culture medium become elevated only after 4 h (Figs 1f and 2d). In contrast the activation of Dusp1 gene expression is rapid and transient, and cooperative regulation by LPS and PGE 2 is short-lived (Fig. 4e). Therefore, in cells exposed to synchronous stimulation, most of the important regulatory processes are likely to have occurred before endogenously generated PGE 2 levels are high enough to have an impact. To test this concept, BMMs were stimulated for 4 h with LPS, without addition of exogenous PGE 2 , or with addition 1 h before the LPS stimulus, at the same time as the stimulus, 1 h or 2 h after the stimulus. PGE 2 was able to inhibit TNF biosynthesis only if added before or at the same time as the LPS stimulus (Fig. 6c). Although the majority of LPS-induced TNF production occurred after 1 h (Fig. 5a), addition of PGE 2 at 1 h or later had no significant inhibitory effect.

Discussion
At early time points in the response to LPS, expression of the Ptgs2 gene was enhanced by Dusp1 deletion but unaffected by gain of function mutation of Zfp36 (Fig. 3a, 1 hour). This suggests the existence of a mechanism of regulation of Ptgs2 expression by MAPKs, independent of TTP phosphorylation. For example it is possible that enhanced early MAPK p38 activity in Dusp1−/− BMMs contributes to transcriptional activation of the Ptgs2 gene via NF-κB or other transcription factors 55 . However, it is clear that this mechanism dwindles in importance as the time course progresses and the contribution of the DUSP1-TTP axis dominates the expression of Ptgs2 mRNA both in vitro (Fig. 3a, 4 hour) and in vivo (Fig. 3d), the cumulative expression of COX-2 protein (Fig. 3b) and the production of PGE 2 (Fig. 3c). We could find no evidence that the targeted mutation of TTP had any impact on NF-κB activity or transcription of the target genes that we examined 43 . In contrast both deletion of the Dusp1 gene and targeted mutation of the Zfp36 gene clearly influenced the stability of Ptgs2 mRNA (Figs 1d and 2b), as well as that of other TTP targets 37,43 . These findings place the Ptgs2 gene alongside Tnf, Cxcl1, Cxcl2, Ifnb1, Il1b and several others, as targets of the DUSP1-TTP regulatory axis, which tightly couples mRNA stability to the activity of the MAPK p38 signaling pathway 30,34,35,37,43,56 . Increased expression of the Ptgs2 gene may contribute to the innate immune pathology of both Dusp1−/− and Zfp36−/− mice 27,38 , whilst decreased expression of Ptgs2 may play a role in the resistance of Zfp36aa/aa to experimental endotoxemia and arthritis 43,44 .
The second half of our study focuses on possible consequences of altered PGE 2 release by LPS-activated macrophages. Both mouse and human primary macrophages expressed the adenylyl cyclase-coupled PGE 2 receptors EP2 and EP4 and displayed cooperative regulation of Dusp1/DUSP1 gene expression by LPS and exogenous PGE 2 . In wild type murine macrophages, exogenous PGE 2 accelerated the decline of MAPK p38 activity and reduced the expression of COX-2 protein. In Dusp1−/− macrophages neither of these effects was seen (Fig. 4). Our findings reveal a negative feedback loop that is illustrated schematically in Fig. 7. PGE 2 down-regulates the expression of COX-2 by increasing the expression of DUSP1, decreasing the activity of MAPK p38 and enhancing the function of TTP. This contrasts with the MAPK p38-dependent positive feedback regulation of COX-2 expression by PGE 2 that has been described by others 45,57,58 . We found no evidence for activation of MAPK p38 or induction of COX-2 expression by PGE 2 in primary murine macrophages. The reason for the discrepancy is not clear. (a) Dusp1+/+ and Dusp1−/− BMMs were treated with 10 ng/ml LPS for the indicated times in the absence or presence of 1 nM PGE 2 , and TNF was measured by ELISA. The graph shows mean ± SEM of three independent BMM cultures of each genotype. The numbers above the graph indicate percentage inhibition of TNF release by PGE 2 . ***p < 0.005; Holm-Sidak method for multiple comparison. (b) Dusp1+/+ and Dusp1−/− BMMs were treated with 10 ng/ml LPS for 4 h in the presence of increasing concentrations of PGE 2 , and TNF was measured by ELISA. The graph on the left illustrates absolute quantities of TNF. In the right hand graph, the quantity of TNF in the absence of PGE 2 is normalized to 100% for each genotype to illustrate differences in the effects of PGE 2 . Mean absolute or relative quantities of TNF from six independent BMM cultures of each genotype are plotted, ±SEM. **p < 0.01; ***p < 0.005; Holm-Sidak method for multiple comparison. (c) As (b), except that IL-6 was measured. ***p < 0.005. (d) As (b), except that the comparison was between Zfp36+/+ and Zfp36aa/aa BMMs, and n = 3. *p < 0.05; ***p < 0.005; Holm-Sidak method for multiple comparison. (e) Wild type BMMs were treated with 10 ng/ml LPS in the absence or presence of 1 nM PGE 2 and the indicated concentrations of PF04418948 ((a) specific EP2 antagonist) or ONO-SE3-208 (a) specific EP4 antagonist). TNF was measured by ELISA. The graph represents mean TNF concentration ± SEM from three independent BMM cultures. All statistical comparisons are against the third column, in which BMMs received both LPS and PGE 2 . n.s., not statistically significant; *p < 0.05; ***p < 0.005; ANOVA.
Exogenous PGE 2 strongly inhibited the LPS-induced expression of TNF by primary mouse macrophages (Fig. 5). The PGE 2 receptor EP4 contributed to this anti-inflammatory effect, whereas EP1 or EP3 appeared not to be expressed, and we could not find evidence for involvement of EP2. We are currently investigating whether LPS-induced up-regulation of EP4 sensitizes macrophages to anti-inflammatory effects of PGE 2 . Importantly, Dusp1 gene disruption and targeted mutation of the Zfp36 (TTP) gene had opposite effects on the expression of TNF, increasing or decreasing it, respectively ( Fig. 5b and d, left panels). However, both genetic modifications rendered macrophages similarly insensitive to PGE 2 -mediated suppression of TNF synthesis ( Fig. 5b and d, right  panels). These results establish a detailed, novel molecular mechanism for anti-inflammatory effects of PGE 2 . IL-10 is a potent anti-inflammatory cytokine 52,59 . IL-6 is not merely pro-inflammatory, but also has important functions in resolution and tissue remodeling 60 . Both Il6 and Il10 genes are well-documented TTP targets 27 , yet they escaped negative regulation by PGE 2 . In fact, both were up-regulated by PGE 2 in macrophages lacking DUSP1 (Figs 5c and 2). Significantly, both genes are positively regulated by cAMP signaling and have been shown to be induced by PGE 2 18, 54, 61-65 . This may be an adaptation that permits their continued expression under conditions favoring the suppression of TNF. According to this model (illustrated for Il6 in Fig. 7), PGE 2 exerts opposing, direct positive and indirect negative effects on gene expression. It positively regulates expression at the transcriptional level via cAMP signaling, and negatively regulates expression post-transcriptionally via induction of DUSP1, inhibition of MAPK p38 and enhancement of TTP function. In wild type macrophages these effects are balanced. In the absence of DUSP1, the negative regulation is lost and the positive regulation unmasked.
Macrophage TNF biosynthesis was inhibited by exogenously added PGE 2 but insensitive to endogenously produced PGE 2 , most likely due to a time delay in LPS-induced PGE 2 biosynthesis (Fig. 6). This discrepancy highlights the artificial nature of in vitro experiments, in which relatively pure cell populations are exposed to a synchronous stimulus. In vivo, as cells migrate to a site of infection or tissue damage, they will encounter a milieu that is continually changing as the inflammatory response peaks and resolves. We suggest that the local biosynthesis of PGE 2 helps to shape this program by modulating the responses of cells arriving at different stages. A monocyte or macrophage arriving early and encountering LPS in the absence of PGE 2 will generate a response very different from that of a late-arriving cell that encounters LPS in the presence of high concentrations of PGE 2 . In this sense, PGE 2 may provide temporal context for responses to pro-inflammatory stimuli, assisting in transitions between initiation and resolution phases. This mode of action of PGE 2 may help to explain why cyclooxygenase inhibition sometimes does not have straightforwardly anti-inflammatory consequences, or can even be toxic to resolution 11,12,[66][67][68] .

Methods
Animals and experimental procedures. All mice were maintained at the Biomedical Services Unit of the University of Birmingham. Animal care and experimental procedures were performed according to Home Office gidelines under PPL 40/8003, and approved by the University of Birmingham Local Ethical Review Committee.
In vivo LPS challenge was performed by intraperitoneal injection of PBS or 5 mg/kg LPS. After three hours all mice were humanely killed. Spleens were recovered and homogenized in RLT buffer (Qiagen) prior to isolation of RNA as described below.
Cell culture. Mice between 6 and 12 weeks of age were humanely culled, bone marrow flushed from femurs, and bone marrow-derived macrophages (BMMs) obtained by culture for 7 days in RPMI1640 containing 10% heat-inactivated FCS and 100 ng/ml M-CSF. Prior to experimentation, macrophage purity was assessed by flow cytometry. Routinely >95% of cells were F4/80+ at the end of the 7 day culture period. BMMs were harvested Note that the effects of PGE 2 -mediated feedback are delayed because of the gradual accumulation of PGE 2 , and therefore operate in a paracrine rather than autocrine fashion. The figure ilustrates possible regulatory connections, and is not intended to capture all the temporal complexities of feedback control. by scraping, seeded at a density of 10 6 /ml in appropriate culture vessels and rested overnight in the absence of M-CSF before being stimulated. Primary human macrophages were generated from peripheral blood monocytes of healthy donors as previously described 52 , by culture for 5-7 days in RPMI1640 containing 10% heat-inactivated FCS and 100 ng/ml M-CSF. After this time, cells were harvested by scraping, seeded at a density of 10 6 /ml and rested overnight in the absence of M-CSF before being stimulated. Measurement of mRNA. RNA was isolated from primary human or mouse macrophages using QIAshredder columns and RNeasy kits (Qiagen). cDNA was generated using iScript cDNA Synthesis kits (Bio-Rad). Gene expression was measured by quantitative PCR with a Roche Light-Cycler 480 Mark II, using custom-designed primers (Eurofins) and SYBR Premix Ex Taq (Takara). Relative expression was calculated using th ΔΔCt method with B2m or B2M mRNA for normalization.
For microarray analysis of gene expression in primary mouse macrophages, RNA was prepared as above and processed as described 37 . For microarray analysis of gene expression in human monocyte-derived macrophages, RNA was prepared as above, cleaned and concentrated then applied to Affymetrix HuGene 1.0 ST arrays. Data analysis was essentially as described 37 , using two-way mixed model ANOVA in Partek Genomics Suite version 6.6. One of the two mouse microarray experiments has been submitted to the Gene Expression Omnibus (GSE68449) at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/). The second mouse BMM microarray and the human monocyte-derived macrophage microarray experiments are more fully described by manuscripts currently in preparation, and the data will be submitted to GEO in full. In the interim, these datasets are available from the author on reasonable request. RNA Immunoprecipitation. 2 × 10 7 RAW264.7 cells were left untreated or stimulated with 10 ng/ml LPS for 1 h, then harvested by scraping, washed twice with ice-cold PBS and lysed by repeated freeze-thawing in 1 ml of ice cold polysome lysis buffer (100 mM KCl, 10 mM HEPES [pH 7.0], 5 mM MGCl 2 , 0.5% Nonidet P-40, 1 mM DTT, 100 U/ml RNase inhibitor, protease and phosphatase inhibitor cocktails [Roche]). Immunoprecipitations and mRNA measurements were essentially as described 43 , except that a pre-immune (PI) rabbit serum was used as the immunoprecipitation control, and quantitative PCR was used to derive fold enrichment of various mRNAs in the anti-TTP immunoprecipitates compared with the PI controls (calculated as 2 −ΔCt ).
Measurement of proteins and PGE 2 . Intracellular proteins were detected by western blotting using reagents listed above. Secreted IL-6, TNF and PGE 2 were detected by sandwich ELISA using commercial kits, according to manufacturers' instructions. For detection of cell surface EP4, BMMs were fixed and permeabilized using Cytofix/Cytoperm solution according to manufacturer's instructions (BD Biosciences) and subjected to flow cytometry using an APC-coupled antibody, Cyan flow cytometer (Beckman Coulter) and FloJo software (TreeStar Inc).