PGE2 induced in and released by dying cells functions as an inhibitory DAMP

Significance

Inflammation is an integral part of the immune responses for the protection of the host to harmful stimuli. However, because inflammation can also underlie the initiation and/or exacerbation of a variety of diseases, it must be tightly regulated. Upon cell death, self-derived molecules, called damage-associated molecular patterns (DAMPs), are released to activate inflammatory responses. It has been unknown whether inhibitory molecules are also released to balance the magnitude of the inflammatory responses. In this study, biochemical and in vivo data demonstrate that prostaglandin E2 is a DAMP that negatively regulates cell death-induced inflammatory responses. These findings reveal an unprecedented facet in the regulation of inflammation via the combined activity of activating and inhibiting DAMPs, which may have clinical implications.

Abstract

Cellular components released into the external milieu as a result of cell death and sensed by the body are generally termed damage-associated molecular patterns (DAMPs). Although DAMPs are conventionally thought to be protective to the host by evoking inflammatory responses important for immunity and wound repair, there is the prevailing notion that dysregulated release of DAMPs can also underlie or exacerbate disease development. However, the critical issue for how resultant DAMP-mediated responses are regulated has heretofore not been fully addressed. In the present study, we identify prostaglandin E2 (PGE2) as a DAMP that negatively regulates immune responses. We show that the production of PGE2 is augmented under cell death-inducing conditions via the transcriptional induction of the cyclooxygenase 2 (COX2) gene and that cell-released PGE2 suppresses the expression of genes associated with inflammation, thereby limiting the cell’s immunostimulatory activities. Consistent with this, inhibition of the PGE2 synthesis pathway potentiates the inflammation induced by dying cells. We also provide in vivo evidence for a protective role of PGE2 released upon acetaminophen-induced liver injury as well as a pathogenic role for PGE2 during tumor cell growth. Our study places this classically known lipid mediator in an unprecedented context—that is, an inhibitory DAMP vis-à-vis activating DAMPs, which may have translational implications for designing more effective therapeutic regimens for inflammation-associated diseases.

Inflammation is a complex biological response and integral aspect of the immune system for the protection of host tissues to harmful stimuli, such as pathogens, damaged cells, or irritants. However, inflammation can be a two-edged sword. Dysregulated and excessive inflammatory responses are linked to the initiation and/or exacerbation of a variety of diseases (1, 2). A multitude of stimuli that can evoke an inflammatory response have been identified. Whether external or internal in origin, these have been categorized into either pathogen-associated molecular patterns (PAMPs) or damage-associated molecular patterns (DAMPs), respectively (3–5). Thus, bacteria-derived lipopolysaccharide (LPS) and virus-derived nucleic acids typically represent PAMPs, which essentially function under infectious conditions (3–5). On the other hand, self-derived high-mobility group box 1 (HMGB1), heat-shock proteins (HSPs), adenosine triphosphate (ATP), and uric acid crystal are among the well-known DAMPs, which are endogenous danger signals capable of triggering inflammatory responses under noninfections or sterile conditions (5, 6).

DAMPs, each of which usually carries a well-defined intracellular function, are released or exposed following an injury or a variety of stresses that typically induce cell death, and they elicit signals via innate receptors such as Toll-like receptors (TLRs) and distinct classes of cytosolic receptors for the evocation of inflammatory responses (5, 7). Thus, a fine-tuning of the inflammatory responses is essential to accomplish the ideal outcome of inflammation in promoting wound healing and tissue repairs. Of note, the study on DAMPs has gained much attention in the context of inflammation-associated diseases. Accumulating evidence indicates that DAMPs, continuously released by dying cells, account for persistence of inflammation with structural and functional alterations of tissues that are collectively referred to as chronic inflammatory processes and that such DAMP-mediated inflammation may underlie the development or exacerbation of numerous, seemingly unrelated diseases such as cancer, autoimmunity, and neurodegenerative diseases (8, 9). As such, an interesting issue raised is whether DAMP(s) can also bring to bear inhibition of inflammatory responses.

In the present study, we initially identified in supernatants of dying cells an inhibitory molecule of inflammatory responses. Further characterization, purification, and mass spectrometry analysis revealed the molecule to be prostaglandin E2 (PGE2), a classically known lipid mediator (10). PGE2 is an eicosanoid lipid mediator that can exert various effects including cell proliferation, stem cell expansion, and immunosuppression (11). PGE2 is the main product of two cyclooxygenases, COX-1 and COX-2, of which COX-1 is constitutively expressed and COX-2 is induced by various stimuli such as cytokines and various PAMPs (12, 13). PGE2 is the most abundant eicosanoid lipid in the inflammatory environment, and its receptors EP2 and EP4 and their downstream signaling to cAMP are proposed for the suppression of inflammation, although the precise mechanism is unclear (10, 12, 14–16). COX-2 is also overexpressed in various forms of cancer, and PGE2 produced by COX-2 has been implicated in the exacerbation of cancer (11, 13). Nevertheless, PGE2 released from dying cells and its physiological relevance have not been described previously.

We show that PGE2 is induced and released in various forms of cell death and serves as an immunosuppressive DAMP in the context of sterile inflammation and tumor growth in vivo. On the basis of our results, we propose that PGE2 functions as an inhibitory DAMP (iDAMP) vis-à-vis activating DAMPs (aDAMPs). We discuss the significance of these findings in the regulation of inflammatory responses and translational implications for inflammation-associated diseases.

Results

Inhibition of LPS-Induced Inflammatory Cytokine Production by Dead Cells.

We first examined the potential immunomodulatory activity of a dead cell-derived supernatant on macrophage cell line RAW 264.7. Necrosis was induced in mouse Lewis lung cell carcinoma (3LL) cells or embryonic fibroblasts (MEFs) by a freeze-and-thaw method (17), and each of the cell supernatants was examined for its immunostimulatory activities and effect on LPS-stimulated induction of TNF-α in RAW 264.7 cells. As shown in Fig. 1A, necrotic cell supernatants did not strongly induce TNF-α production in this experimental setting. Interestingly, these supernatants inhibited LPS-induced TNF-α production in a dose-dependent manner, albeit with different magnitudes from one another (Fig. 1B). This suppression was also observed in in vitro cultured dendritic cells (DCs) stimulated by LPS (Fig. S1A). The suppressive effect was also observed with supernatants of the necrotic SL4 cells, a mouse colon carcinoma cell line, and HeLa cells, a human cervical cancer cell line (Fig. S1B).

Fig. 1.

Suppression of LPS-induced cytokine expression in the presence of a supernatant of dead cells or PGE2. (A) RAW cells (5 × 10⁴ cells) were treated either with an increasing volume of the supernatant of necrotic 3LL cells or MEFs (the volume of the supernatant is equivalent to 5 × 10⁴, 3 × 10⁵, or 1.5 × 10⁶ cells) or with LPS (20 ng/mL) for 24 h. TNF-α protein levels in the cultured media were determined by ELISA. (B) RAW cells were stimulated with LPS and cocultured with the same volumes of supernatant as described in A to examine their combined effect on TNF-α production. (C) TNF-α protein levels determined by ELISA in cultures of RAW cells stimulated with LPS and necrotic supernatant from 3LL cells (equivalent to 3 × 10⁵ cells) following treatment with DNaseI or RNaseA for 24 h. (D) RAW cells (1.5 × 10⁵ cells) were stimulated with LPS (5 ng/mL) with or without a necrotic supernatant from 3LL cells (9 × 10⁵ cells) pretreated with proteinase K or trypsin. After 2 h of stimulation, Tnf mRNA levels were determined by quantitative RT-PCR (qRT-PCR). (E) As described in D except that the lipid-extracted fraction from the necrotic supernatant of 3LL cells (9 × 10⁵ cells) was added to LPS-stimulated cells in lieu of protease treatment. (F) Lipids in the fractionated supernatant (nos.17–21) in Fig. S1D were analyzed by mass spectrometry. The quantity of each lipid was expressed as a proportion to a corresponding standard sample. (G) RAW cells were stimulated with LPS in the absence (PBS) or presence of PGE2 (50 nM or 500 nM) for 2 h or in the presence of 6-keto-PGF1α. The induction levels for Tnf mRNA were then determined. NC; necrotic cell supernatant.

Fig. S1.

Suppression of LPS-induced cytokine gene expression by necrotic cell supernatant treatment. (A) Conventional DCs (cDCs) (5 × 10⁴ cells) were stimulated with the necrotic supernatant of 3LL cells (3 × 10⁵ cells) and LPS (20 ng/mL). After 24 h of stimulation, TNF-α protein levels were determined by ELISA. (B) RAW 264.7 cells (5 × 10⁴ cells) were stimulated by LPS (20 ng/mL) in the presence or absence of the necrotic supernatant of 3LL, SL4, or MEFs or HeLa cells (the volume of the supernatant is equivalent to 5 × 10⁴, 3 × 10⁵, or 1.5 × 10⁶ cells). After 24 h of stimulation, TNF-α protein levels were determined by ELISA. (C) RAW 264.7 cells (1.5 × 10⁵ cells) were stimulated by LPS (5 ng/mL) in the presence or absence of the necrotic supernatant of 3LL cells (9 × 10⁵ cells). After 2 h of stimulation, levels of Tnf and Ifnb1 mRNAs were determined by qRT-PCR. (D) The supernatant from necrotic 3LL cells was separated by size exclusion chromatography. Then, RAW 264.7 cells (5 × 10⁴ cells) were simulated by LPS (20 ng/mL) with addition of each fraction for 24 h. TNF-α protein levels were then determined by ELISA. (E) PGE2 levels of each fraction (D) determined by ELISA. (F) RAW cells (1.5 × 10⁵ cells) were stimulated with LPS (5 ng/mL) in the absence (denoted PBS) or presence of pure PGE2 (50 nM or 500 nM) for 2 h (Left) or in the presence of 6-keto-PGF1α (50 nM or 500 nM; Right). The induction levels for Ifnb1, Ccl3, and Cxcl10 mRNA were then determined by qRT-PCR. All data are shown as means ± SD of triplicate determinants. (G) RAW cells (1.5 × 10⁵ cells) were cultured in the absence (denoted PBS) or presence of pure PGE2 (50 nM or 500 nM) for the indicated time. LDH levels were then determined. (H) PGE2 levels in the necrotic supernatant of 3LL, SL4, MEFs, and HeLa cells were determined by ELISA. (I) Peritoneal macrophages (1.5 × 10⁵ cells) were primed with LPS (1 μg/mL) for 4 h and either left untreated or treated with ATP (5 mM) for 1 h to induce pyroptosis. PGE2 levels in the culture medium were determined by ELISA. (J) 3LL cells (1.5 × 10⁵ cells) were treated with cisplatin (50 μM) or etoposide (50 μM) for the indicated time periods to induce apoptosis. PGE2 levels in the culture medium were determined by ELISA. (K) 3LL cells (1.5 × 10⁵ cells) were subjected to freeze-and-thaw cycles (1, 3, and 5 cycles). The Ptgs1, Ptgs2, and Ptges mRNA levels were then determined by qRT-PCR. (L) The 3LL cells were subjected to the freeze-and-thaw method in the absence (denoted DMSO) or the presence of Ceefourin 1 (1 μM or 10 μM). PGE2 levels in the necrotic supernatant were determined by ELISA. All data are shown as means ± SD of triplicate determinants. NC, necrotic cell supernatant.

The supernatant of the necrotic 3LL cells also suppressed LPS-induced mRNA expression for Tnf and Ifnb1, suggesting the transcriptional inhibition of these genes (Fig. S1C). One interpretation of these data is that necrotic cells release molecules that suppress activities of the dead cell-derived immunostimulatory DAMP molecules under these experimental settings (see the next section).

Characterization of the Immunosuppressive Molecule(s).

To characterize further the nature of the immunosuppression, we first treated 3LL necrotic cell supernatants with DNase I or RNase A and observed no effect on the suppressive activity on the LPS-induced production of TNF-α in RAW 264.7 cells (Fig. 1C). Further, neither proteinase K nor trypsin treatment showed any effect on the suppression (Fig. 1D). Because these results indicate the immunosuppressive molecule is neither a nucleic acid nor a protein, we next extracted the lipid fraction from the necrotic supernatant by the Bligh–Dyer method. Interestingly, addition of the lipid cellular fraction reproduced the suppressive effect of the necrotic cell supernatant (Fig. 1E).

Lipid mediators typically have a much lower molecular weight than proteins do, and indeed, size exclusion chromatography revealed that the immunosuppressive activity is found in the less than 6 kDa molecular-weight fraction (Fig. S1D). We also found in these fractions abundant PGE2 and 6-keto-PGF1α by mass spectrometry analysis (Fig. 1F) and PGE2 by ELISA (Fig. S1E). We then observed that synthetic PGE2, but not 6-keto-PGF1α, suppressed LPS-induced Tnf and Ifnb1 mRNA expression (Fig. 1G and Fig. S1F), an observation consistent with previous reports (10, 12, 14). In addition, the suppressive effect of PGE2 was also observed for the induction of Ccl3 and Cxcl10 mRNA (Fig. S1F). Because PGE2 did not affect the viability of RAW 264.7 cells (Fig. S1G), the mRNA suppression observed here is likely due to PGE2-mediated signaling (see Discussion).

Perhaps expectedly, PGE2 was also found in the supernatants of various necrotic cells (Fig. S1H), and its levels were correlated with the magnitude of TNF-α suppression (Fig. S1B). We next asked whether PGE2 is released by other forms of cell death. PGE2 was massively released by peritoneal macrophages after stimulation by ATP and LPS, a combination that induces pyroptosis (Fig. S1I and ref. 18). Further, a similar observation was made when 3LL cells underwent apoptosis by the treatment of cisplatin or etoposide (Fig. S1J). These results suggest that the release of PGE2 is a common feature of various forms of cell death.

Of the two cyclooxygenase enzymes that catalyze the synthesis of PGE2, COX-2 is selectively induced under inflammatory conditions (12, 13). We then questioned whether the Ptgs2 gene is induced under cell death conditions to promote PGE2 release by dead cells. Interestingly, Ptgs2 mRNA expression levels were significantly increased in 3LL cells during freeze-thaw treatment (Fig. S1K), suggesting that PGE2 released by dead cells is, at least in part, a consequence of Ptgs2 gene induction during the process of cell death. We also examined whether the dead cell’s release of PGE2 involves multidrug resistance-associated protein 4 (MRP4) (19) by treating 3LL cells with the MRP4 inhibitor Ceefourin 1. However, Ceefourin 1 did not significantly affect PGE2 release from dead 3LL cells (Fig. S1L), supporting the notion that the release is due to a passive leakage as a DAMP during cell death.

Immunosuppressive Function of PGE2 in Innate Immune Signaling.

We asked whether PGE2 suppresses inflammatory responses evoked by other pattern recognition receptor (PRR) ligands by stimulating myeloid cells with various ligands for TLRs and cytosolic sensors. We found that PGE2 suppressed Tnf mRNA induced by poly I:C (via TLR3), 5′pppRNA (via RLR), or B-DNA (via cGAS)-treated peritoneal macrophages (Fig. S2A). Further, PGE2 also suppressed Ifnb1 mRNA induced by poly I:C (Fig. S2B) in these cells and the induction of Tnf and Ifnb1 mRNA by CpG-A ODN (via TLR9) or R837 (via TLR7) stimulation of plasmacytoid DCs (Fig. S2 C and D). On the other hand, unlike the case for LPS stimulation, PGE2 did not suppress the induction of Ifnb1 mRNA by 5′pppRNA or B-DNA (Fig. S2B). These results collectively suggest that PGE2 suppresses Tnf gene expression induced upon the stimulation of TLRs or cytosolic sensors, whereas the suppression of the Ifnb1 gene by PGE2 occurs for TLR stimulation but not for the stimulation of cytosolic sensors.

Fig. S2.

Suppression of PRR-mediated Tnf and Ifnb1 mRNA induction by PGE2 treatment. (A) Peritoneal macrophages (1 × 10⁵ cells) were stimulated by LPS (20 ng/mL), poly-I:C (100 μg/mL), B-DNA (5 μg/mL), or 5′pppRNA (1 μg/mL) in the presence or absence of PGE2 (50 nM or 500 nM) for 4 h. Tnf mRNA levels were determined by qRT-PCR. (B) Same as A but Ifnb1 mRNA levels were determined. (C) Plasmacytoid DCs (pDCs) were stimulated by CpG-A ODN (3 μM) or by R837 (5 μg/mL) in the presence or absence of PGE2 (50 nM or 500 nM). After 4 h of stimulation, Tnf mRNA levels were determined by qRT-PCR. (D) Same as C but Ifnb1 mRNA levels were measured. All data are shown as means ± SD of triplicate determinants.

The observations showing that dead cell-derived PGE2 inhibits the LPS-mediated induction of Tnf and Ifnb1 mRNA also prompted us to study the underlying mechanism, which has been poorly understood (12, 14–16). Because the induction of TNF-α by LPS requires activation of NF-κB and MAPK pathways (3, 4), we next examined the effect of PGE2 on the LPS-mediated activation of canonical NF-κB and MAPK in RAW 264.7 cells. As shown in Fig. 2A, PGE2 treatment did not affect LPS-mediated phosphorylation levels of IκB-α, ERK, and JNK. PGE2 treatment did not affect the transcriptional activation of NF-κB, as monitored by an NF-κB–driven luciferase reporter assay (Fig. 2B) and by ChIP assay for the binding of p65 to the Tnf promoter (Fig. 2C). These results therefore suggest that PGE2 may suppress Tnf mRNA induction by yet unknown mechanism (see Fig. 2D).

Fig. 2.

Effect of PGE2 on the LPS-induced signaling pathways. (A) RAW cells (5 × 10⁶ cells) were stimulated with LPS (20 ng/mL) in combination with PGE2 (500 nM). Whole-cell lysates were subjected to immunoblot analysis. (B) RAW cells (1 × 10⁵ cells) transiently transfected with an NF-κB luciferase reporter construct were stimulated with LPS (5 ng/mL) with or without PGE2 (50 nM or 500 nM). Luciferase activity was measured and expressed as units relative to the unstimulated control. (C) RAW cells (1 × 10⁷ cells) were stimulated with LPS (5 ng/mL) with or without PGE2 (50 nM) for 1 h, and the cell lysates were subjected to ChIP assay using an anti-p65 antibody and primers amplifying the NF-κB sites within the promoter of the Tnf gene. (D) RAW 264.7 cells (1 × 10⁵ cells) were pretreated with CHX (20 μg/mL) for 1 h and then stimulated with LPS in the presence or absence of PGE2 for 2 h. Then, the induction levels of Tnf mRNA were determined by qRT-PCR. (E) Same as D except that the induction levels of Ifnb1 mRNA were examined.

We also examined the effect of PGE2 on the LPS-induced activation of IRF3 by monitoring its phosphorylation, a hallmark of IRF3 activation and essential for the induction of the Ifnb1 gene in RAW 264.7 cells (3, 4, 20). As shown in Fig. 2A, IRF3 phosphorylation remained low in the presence of PGE2. The data indicate that PGE2 interferes with the LPS–TLR4–TRIF signaling pathway (3, 4) that activates IRF3 for Ifnb1 gene induction. Interestingly, when the cells were treated by cycloheximide (CHX) before LPS stimulation, the Tnf and Ifnb1 mRNA induction levels remained essentially unaffected by PGE2 (Fig. 2 D and E). These results suggest that the cytokine gene suppression by PGE2 is contingent on de novo synthesis of a protein, the identification of which will require further investigation. It also remains to be examined whether similar inhibitory mechanisms are also operative for other PRR signaling pathways.

Enhanced Immunostimulatory Activities of Necrotic Cells by Depletion of PGE2.

A prevailing notion during cell death is that multiple DAMPs are released and activate inflammatory responses (5, 7, 8). However, under our experimental settings, which are essentially identical to those used by others, immune responses are not robust (Fig. 1A and ref. 21). Thus, we envisaged the following scenario: DAMPs that have the potential to evoke inflammatory responses are suppressed by the induction and release of PGE2 by dying cells. To test this concept experimentally, we asked whether inhibition of PGE2 production would convert the necrotic cells to more potent cells in the evocation of inflammatory responses.

We first pretreated 3LL cells with indomethacin, an inhibitor of COX-1 and COX-2 enzymes, and then examined the immunostimulatory activity triggered by the supernatant of indomethacin-treated necrotic cells. PGE2 release was expectedly suppressed in the supernatant of necrosis-induced 3LL cells by the indomethacin treatment (Fig. S3), and concomitantly, the induction of Tnf mRNA in peritoneal macrophages was greater compared with untreated cells (Fig. 3A). Similar observations were made with treated extracts from other cell lines (Fig. 3 B and C). These data support the view that PGE2 functions as an iDAMP that counteracts the inflammatory responses evoked by aDAMPs.

Fig. S3.

Inhibition of PGE2 production in necrotic 3LL cells by indomethacin treatment. The PGE2 level of the necrotic supernatant of 3LL cells either pretreated by mock or indomethacin was determined by ELISA. All data are shown as means ± SD of triplicate determinants.

Fig. 3.

Enhancement of immunostimulatory activity of necrotic cells by the inhibition of PGE2 synthesis. (A) Peritoneal macrophages (1.5 × 10⁵ cells) were stimulated with a necrotic supernatant (9 × 10⁵ cells) of 3LL cells pretreated by indomethacin (Indo.) or without treatment (DMSO). After 2 h of stimulation, the Tnf mRNA levels were determined by qRT-PCR. (B) Same as A except that SL4 cells were used for the necrotic supernatant. (C) Same as A except that MEFs were used for the necrotic supernatant. (D) Peritoneal macrophages were stimulated with a necrotic supernatant as described in A. After 2 h of stimulation, total RNA was extracted and subjected to microarray analysis. A heat map is depicted in the Left panel. Representative genes whose mRNA expression levels are augmented by indomethacin treatment and may be involved in the regulation of oncogenesis are shown in the Right panel.

To obtain a more comprehensive view of the interplay between aDAMP(s) and PGE2, we performed a microarray analysis in which peritoneal macrophages were stimulated with supernatants of mock-treated or indomethacin-treated SL4 necrotic cells. Interestingly, we found that genes that possess antitumor properties were up-regulated by indomethacin-treated supernatants, compared with PBS and mock-treated supernatants (Fig. 3D). For example, Ccl2 and Clec5a, along with Tnf, are markers of M1-type macrophages, which are known to be tumoricidal (22, 23), and Fpr1 is critical to the efficient induction of antitumor immunity in DCs (24). Furthermore, several tumor suppressor genes, such as Grhl1 and Mir133b, are up-regulated by indomethacin-treated supernatants (25). These results suggest that PGE2, functioning as an iDAMP, might modulate macrophage’s function and suppress the antitumor effects invoked by aDAMPs (see the next section).

Role of PGE2 as an iDAMP in Vivo.

The above observations prompted us to examine the in vivo relevance of PGE2 as an iDAMP in two experimental systems: acetaminophen-induced hepatotoxicity and tumor cell growth. Peritoneal injection of acetaminophen induces massive necrosis of hepatocytes and evokes TNF-α production mediated by aDAMPs (26). Because PGE2 is produced by COX-2 in the liver (27), we hypothesized that this necrotic liver cell-derived PGE2 would function as iDAMP to counteract the activity of aDAMPs by suppressing inflammatory responses. We therefore examined the effect of celecoxib, an inhibitor of COX-2 that is clinically used to treat pain or inflammation. As shown in Fig. 4A, administration of celecoxib resulted in elevated levels of the serum TNF-α in this liver necrosis model. Expectedly, this was accompanied by elevated levels of alanine transaminase (ALT) and aspartate transaminase (AST), the hallmarks of liver damage (Fig. S4A). Thus, these results support the view that PGE2 functions as iDAMP to suppress the aDAMP-mediated liver inflammation to cause liver damage.

Fig. 4.

Potential role of PGE2 as an iDAMP in vivo. (A) C57BL/6 mice were injected with acetaminophen intraperitoneally with or without simultaneous oral administration of celecoxib (n = 4 each). After 24 h, serum TNF-α protein levels were determined. (B) SL4 cells (1 × 10⁶ cells) were cultured under hypoxic (1% O₂) or normoxic condition (20% O₂) for 24 h, and then PGE2 levels were determined. At this time point, about 90% of the cells underwent cell death in the hypoxic condition, whereas more than 90% of the cells were alive under the normoxic condition. (C) Cell numbers of parental SL4 cells, SL4–COX2i, SL4–COX2d1, and SL4–COX2d2 grown in vitro. Notably, PGE2 did not affect the viability of SL4 cells at our experimental setting. (D) SL4–COX2i, SL4–COX2d1, or SL4–COX2d2 cells (1 × 10⁶ cells) were transplanted into C57BL/6 mice (n = 6 each) by s.c. injection, and the tumor volume was measured. Asterisks show statistical significance (P < 0.05) between SL4–COX2i and SL4–COX2d clones. (E) SL4–COX2i or SL4–COX2d1 cells (2 × 10⁵ cells) were injected into C57BL/6 mice intravenously. After 17 d of injection, lungs were subjected to H&E staining. Original magnification, 40×. (F) SL4–COX2i or SL4–COX2d1 cells (1 × 10⁶ cells) were transplanted into C57BL/6 mice by s.c. injection (n = 6 each). After 12 d of transplantation, single-cell suspensions were prepared from the transplanted tumors and subjected to flow cytometry analysis. Expression of CD206 on CD45⁺CD11b⁺F4/80⁺ tumor-infiltrating macrophages (Left) and mean fluorescent intensity (MFI) of CD206 (Right) are shown. Asterisks, P < 0.01. APAP, acetaminophen.

Fig. S4.

Preparation of Ptgs2-deficient SL4 cells. (A) C57BL/6 mice were administered celecoxib (100 mg/kg) orally with a near-simultaneous i.p. injection of acetaminophen (500 mg/kg) (n = 4). After 24 h, the serum ALT and AST levels were determined. (B) Ptgs2-deficient SL4 cells were generated by the CRISPR/Cas9 system. Also shown is the expression status of Ptgs2 mRNA in parental SL4, SL4–COX2i, SL4–COX2d1, and SL4–COX2d2 cells. Total RNA was extracted from those cells and subjected to qRT-PCR analysis. (C) SL4–COX2i and SL4–COX2d cells were cultured under hypoxia (1% O₂, 5% CO₂). After 24 h, PGE2 levels of the culture supernatant were determined by ELISA. At this time point, about 90% of the cells underwent cell death. (D) Parental SL4 and SL4–COX2i cells (1 × 10⁶ cells) were transplanted into C57BL/6 mice by s.c. injection and evaluated for their growth at the indicated time (n = 6 each). (E) SL4–COX2i or SL4–COX2d1 cells (1 × 10⁶ cells) were transplanted into C57BL/6 mice by s.c. injection (n = 6 each). After 12 d of transplantation, single-cell suspensions were prepared from the tumor and subjected to flow cytometry analysis. The proportion of CD45⁺ cells among whole cells or that of DCs (F4/80⁻CD11c⁺ cells), neutrophils (CD11b⁺Gr1⁺ cells), macrophages (CD11b⁺F4/80⁺ cells), NK cells (CD3ε⁻NK1.1⁺ cells), B cells (CD3ε⁻B220⁺ cells), CD4⁺T cells (CD3ε⁺CD4⁺ cells), and CD8⁺T cells (CD3ε⁺CD8⁺ cells) among CD45⁺ cells are shown. (F) Bone-marrow macrophages (M0) were treated with LPS (50 ng/mL) plus IFN-γ (100 ng/mL) or IL-4 (20 ng/mL) for 7 h to differentiate them into M1 or M2 macrophages, respectively. These cells were concomitantly incubated with either the mock- or indomethacin-treated necrotic supernatant of SL4 cells. The Tnf, Il12b, and Arg1 mRNA levels were determined by qRT-PCR. Data are shown as means ± SD of triplicate determinants unless otherwise specified. APAP, acetaminophen; Indo., indomethacin; NC, necrotic cells; N.S., not significant.

Necrosis is a well-known feature that commonly occurs during in vivo tumor growth (28). Typically, because newly formed blood vessels for tumors are aberrant and have poor blood flow, tumors become hypoxic and many tumor cells undergo cell death by necrosis (29). As such, hypoxia is an event intrinsic to tumor growth, wherein cancer cells undergo genetic and adaptive changes to survive in the hypoxic environment, thereby acquiring a more malignant phenotype (29). In this context, tumor cells often show elevated COX-2 expression (30, 31), suggesting the possibility that the COX-2–mediated production and release of PGE2 by hypoxia-induced necrotic tumor cells may serve as an iDAMP, thereby affecting the growth of live tumor cells in the in vivo tumor microenvironment. To experimentally address this issue, we first examined PGE2 production by hypoxia-induced necrosis of SL4 cells in vitro and found that these cells produce PGE2 at a level about sevenfold higher than that of the cells grown in normoxic conditions (Fig. 4B).

We then created COX-2–deficient SL4 cells by a CRISPR/Cas9-mediated gene deletion system (Fig. S4B). Two independently obtained clones lacking the functional Ptgs2 gene, designated as SL4–Cox2d1 and SL4–Cox2d2, were grown in vitro, along with a control clone, SL4–Cox2i, in which the Ptgs2 gene remained intact, and the parental SL4 cells. As expected, PGE2 production levels in the supernatant of the hypoxia-treated SL4–Cox2d1 and SL4–ox2d2 cells were more than fivefold lower than those from the SL4–Cox2i and parental cells (Fig. S4C). Perhaps not surprisingly, these cells all showed similar growth rates in vitro (Fig. 4C). These cells were then s.c. injected into C57BL/6 mice, and their growth was examined. COX-2–deficient SL4–Cox2d1 and SL4–Cox2d2 cells showed marked growth retardation compared with SL4–Cox2i cells (Fig. 4D), which showed a growth rate similar to the parental SL4 cells (Fig. S4D). Further, a similar trend was also found in the lung metastasis model (Fig. 4E). To gain insights into these observations, we isolated immune cells in the tumor-surrounding area of these mice and subjected these cells to flow cytometry analysis. We found no overt differences in terms of the population of T cells, B cells, natural killer (NK) cells, macrophages, neutrophils, and DCs (Fig. S4E). Of note, however, we found that the CD206-expressing macrophages, consistent with an M2 phenotype, are more abundant in the parental SL4 tumor-bearing mice compared with those mice bearing the mutant SL4 cells lacking COX-2 (Fig. 4F).

These results are consistent with the widely accepted notion that PGE2 skews macrophages toward M2-type macrophages that promote tumor growth in the tumor microenvironment (32). The results also support the scenario that PGE2 release under hypoxic conditions in vivo is a newly identified mechanism of tumor immune evasion, wherein dying tumor cells assist the growth of surviving tumor cells by modifying the tumor microenvironment. As a means to verify this view, we also examined the effect of dying cells on macrophage polarization by the following in vitro assay. Bone marrow-derived cells were first cultured by M-CSF to generate M0-type macrophages, and subsequently, these cells were further cultured in the presence of either LPS + IFN-γ or IL-4, a well-known protocol for M1 or M2 macrophage polarization, respectively, to examine the effect of dead cell-derived PGE2 on macrophage polarization. Interestingly, we found that the necrotic supernatant of SL4 cells potentiated M2 polarization, whereas that of indomethacin-treated SL4 cells had the opposite effect (Fig. S4F and Discussion).

Discussion

In recent years, DAMPs have garnered much attention for their role as endogenous danger signals that alert the innate immune system and evoke inflammatory responses typically caused by cell death due to tissue injury or stress (5, 7, 8). Thus, DAMPs are important for maintaining homeostasis of the host; however, excessive and/or sustained DAMP-induced signaling may underlie the development and/or exacerbation of many diseases such as autoimmunity and cancer (8, 33). Given this potential for harm, negative regulatory mechanisms of DAMP-mediated inflammation have undoubtedly evolved. For example, aDAMPs were reported to recruit CD24–Siglec G/10 and Siglec G/10-associated phosphatases, such as SHP1, to repress TLR-mediated or Nod-like receptor-mediated signaling (34). In the present study, however, we show that PGE2 released upon cell death intrinsically functions to negatively regulate inflammatory responses activated by DAMPs.

Of the numerous DAMPs identified to date, including HMGB1, HSPs, and the products of purine metabolism (e.g., ATP and uric acid), all function as activators of inflammation (5, 6). That PGE2, when released by dead cells, inhibits inflammation identifies it as a previously unidentified class, which we refer to as iDAMP vis-à-vis immune aDAMPs. Consistent with this role, the inhibition of the PGE2 synthesis pathway in various cell types renders them more potent to evoke inflammatory responses upon cell death in vitro (Fig. 3 A–C). Further, evidence that PGE2 released by dead cells protects against acetaminophen-induced hepatotoxicity and enhances tumor cell growth in vivo also lends support to our hypothesis (Fig. 4 A, D, and E).

To our knowledge, this study reveals the first example of the interplay between aDAMPs and iDAMPs in the regulation of inflammatory responses. The significance of this is that although ∼10⁵ human cells in our body undergo cell death every second, the vast majority are believed to be silent with respect to immune activation due to their removal by phagocytosis and others (35). Our data, which illustrate the enhanced immunostimulatory activity in the absence of PGE2 that otherwise functions as an iDAMP, may be related to the typical immune-incompetent nature of dying cells. Our observations made in acetaminophen-induced liver inflammation support this view (Fig. 4A). Also consistent with the evidence, mice defective in PGE2 synthesis develop severe peritonitis (36), although a firm connection between the PGE2’s function as iDAMP and these inflammatory disorders requires formal demonstration. In the clinic, acetaminophen is often used in the treatment of inflammatory diseases. However, in light of our present findings, new caution may be needed for the use of this and other COX inhibitors.

This study also raises the question of which PGE2 receptor is responsible for the function of PGE2 as a dead cell-derived iDAMP. In this context, there are previous reports showing that PGE2 suppresses LPS-induced gene expression via EP2 and EP4, which are expressed on macrophages (10, 12, 14). As such, we think it is most likely that the immunosuppressive function of PGE2 as an iDAMP is also mediated by these receptors, although we cannot rigorously exclude the involvement of other receptors. How does PGE2 signaling inhibit inflammatory responses? That PGE2 can suppress LPS-induced inflammatory cytokine and type I IFN secretion is known, however the mechanism of these suppressive effects has been controversial (12, 14). One study showed suppression of the NF-κB pathway by cAMP, a downstream effecter molecule of PGE2 signaling, whereas another argued that the NF-κB pathway is unaffected by cAMP signaling (15, 16). We found no evidence that the NF-κB pathway is affected by PGE2 signaling (Fig. 2 A–C), but de novo protein synthesis is required for the suppression of TNF-α production by PGE2 (Fig. 2D). The suppression of Ifnb1 gene induction by PGE2 may also involve a similar mechanism (Fig. 2E). Obviously, identification of a target molecule, presumably induced by PGE2 signaling, that mediated direct suppression of these pathways may be an interesting course of research.

An additional ramification of our study is the hitherto unknown facet of PGE2 as an iDAMP in dying tumor cells for promoting growth of live tumor cells. There is emerging recognition that immune system evasion is as a hallmark of cancer progression. PGE2 synthesis and signaling pathways were shown to favor tumor growth in vivo (11, 13, 37). In this context, tumor cell death and release of iDAMPs, due either to hypoxia, loss of cell-cycle control, and so forth, would be predicted to alter the tumor microenvironment to favor immune suppression and tumor survival. Our findings on the reduced growth in vivo of SL4 tumor cells lacking the Ptgs2 gene offer yet an unidentified mechanism by which tumor cells favor their growth and survival (Fig. 4 D and E). One may argue that tumor cell death is beneficial for surviving tumor cells as a kind of “altruistic suicide” by the release of PGE2 as an iDAMP. Thus, our present findings may offer an unprecedented rationale for clinical trials evaluating COX-2 inhibitors as chemoprevention and therapeutic agents (38, 39).

There may be multiple mechanisms that underlie the promotion of tumor cell growth by dead cell-derived PGE2. One is our findings that dead tumor cell-derived PGE2 suppresses the gene induction of cytokines critical for coping with tumor growth (Fig. S1C and refs. 40, 41), which would prevent direct attack by these cytokines. SL4 cells, for instance, are known to undergo apoptosis in the presence of TNF-α (42). Another mechanism triggered by PGE2 and supported by our data is a conditioning of the tumor microenvironment in which macrophages are skewed toward a protumor M2 phenotype (43). Conversely, a microarray analysis showed the induction of several genes by an indomethacin-treated necrotic cell supernatant, genes such as M1 macrophage signature genes and tumor suppressor genes that have antitumor effects (Fig. 3D and refs. 24, 25). Although further study is required, such genes would contribute to the suppression of tumor growth. Also, our data do not exclude additional mechanisms of regulation, which may warrant further investigation.

In conclusion, we revealed that PGE2 is an iDAMP counteracting the function of aDAMPs. Our findings suggest that PGE2 may be a promising target, particularly in instances where massive cell death is associated. Further, given that many elements of the hypoxia–response pathway are candidates for therapeutic intervention, our results reveal an unprecedented facet of PGE2 interference—further support for the possible use of COX inhibitors for cancer therapy (38, 39). It remains to be seen whether additional iDAMPs exist. Given our data showing the enhanced immunostimulatory activity of indomethacin-treated cells, this system may be useful in identifying the nature of aDAMPs.

Materials and Methods

Induction of necrosis was performed as described previously (17, 21). For the preparation of a PGE2-depleted necrotic supernatant, cells were incubated with indomethacin (10 μM) overnight and subjected to a freeze-and-thaw method in the presence of indomethacin (10 μM). For induction of apoptosis, cells were treated with etoposide (50 μM) or cisplatin (50 μM). Peritoneal macrophages were primed with LPS (1 μg/mL) for 4 h and treated by ATP (5 mM) for 1 h to induce pyroptosis. All animal experiments were done in accordance with guidelines of The University of Tokyo and were approved by the animal research committee of The University of Tokyo.

Additional information can be found in SI Materials and Methods.

SI Materials and Methods

Mice.

C57BL/6 mice were purchased from CLEA Japan Inc. All animal experiments were done in accordance with guidelines of the University of Tokyo.

Reagents and Cell.

ATP, poly(I:C), R837, and 5′-ppp-RNA were purchased from Invivogen. LPS (O55:B4), B-DNA, CHX, etoposide, and cisplatin were purchased from Sigma-Aldrich. CpG-A ODN and other oligonucleotides were purchased from FASMAC. PGE2, 6-keto-PGF1α, celecoxib, and indomethacin were purchased from Cayman Chemical. Ceefourin 1 was purchased from Abcam. Peritoneal macrophages were prepared as described previously (44). Briefly, mice were injected intraperitoneally with 2 mL of 4% (wt/vol) thioglycollate (DIFCO) solution. The peritoneal cavity was washed by PBS to collect peritoneal exudate cells 4 d after the thioglycollate injection, and the cells were incubated on a Petri dish for 2 h in RPMI supplemented with 10% (vol/vol) FCS. After the incubation, cells were washed with RPMI medium and used for each assay. Preparation of MEFs and plasmacytoid DCs differentiated from bone marrow cells was described previously (45). RAW264.7 cells, mouse melanoma cell line B16F1 cells, 3LL cells, colon carcinoma cell line SL4 cells, and human cervix adenocarcinoma cell line HeLa cells were obtained and maintained as described previously. Hypoxic exposure [1% (vol/vol) O₂, 5% (vol/vol) CO₂] was performed in a MCO-5MUV incubator (Panasonic).

ELISA.

Production of TNF-α (R&D Systems) and PGE2 (Cayman chemical) in the culture supernatants was quantified by ELISA kits according to the manufacturer’s protocol.

qRT-PCR Analysis.

Total RNA from tissues or cells was extracted using RNAiso (TaKaRa Bio) or NucleoSpin RNA II (MACHEREY NAGEL) and was reverse-transcribed with PrimeScript RT Master Mix (TaKaRa Bio). qRT-PCR was performed on LightCycler 480 using the SYBR Green PCR Master Mix (Roche Life Science), and values were normalized to the expression of Gapdh mRNA. Primer sequences are as follows: Gapdh forward 5′-ctcatgaccacagtccatgc-3′; Gapdh reverse 5′-cacattgggggtaggaacac-3′; Tnf forward 5′-tcataccaggagaaagtcaacctc-3′; Tnf reverse 5′-gtatatgggctcataccagggttt-3′; Ifnb1 forward 5′-ACGCCTGGATGGTGGTCCGA-3′; Ifnb1 reverse 5′-TGCCTGCAACCACCACTCATTCT-3′; Ptgs1 forward 5′-CACAACACTTCACCCACCAG-3′; Ptgs1 reverse 5′-AAGAGCCGCAGGTGATACTG-3′; Ptgs2 forward 5′-GAGTGGGGTGATGAGCAACT-3′; Ptgs2 reverse 5′-AAGTGGTAACCGCTCAGGTG-3′; Ptges forward 5′-TCGCCTGGATACATTTCCTC-3′; Ptges reverse 5′-CCATGGAGAAACAGGAGAAC-3′; Arg1 forward 5′-gcaacctgtgtcctttctcc-3′; Arg1 reverse 5′-gcaagccaatgtacacgatg-3′; Il12b forward 5′-GACACGCCTGAAGAAGATGAC-3′; Il12b reverse 5′-TAGTCCCTTTGGTCCAGTGTG-3′; Ccl3 forward 5′-CTCTGCAACCAAGTCTTCTCAGCG-3′; Ccl3 reverse 5′-TCAGGAAAATGACACCTGGCTGGG-3′; Cxcl10 forward 5′-ACTGCATCCATATCGATGAC-3′; and Cxcl10 reverse 5′-TTCATCGTGGCAATGATCTC-3′. All experiments were performed at least three times in triplicate.

DNase, RNase, and Proteinase Treatment.

A supernatant from necrotic cells was treated with DNaseI solution (0.5 U/μL; TaKaRa Bio), RNaseA (0.25 mg/mL; MACHEREY-NAGEL), or PBS and was incubated at 37 °C for 30 min. For proteinase K or trypsin treatment, the necrotic supernatant was treated with proteinase K (100 μg/mL; TaKaRa Bio) or trypsin (0.01%; GIBCO) at 37 °C for 1 h. The protease K-treated sample was then incubated with p-Amidinophenylmethylsulfonylfluoride (APMSF) (5 mM; Nacalai Tesque) on ice for 20 min to inactivate proteinase K.

Extraction of Lipid.

Extraction of lipid from the necrotic supernatant of 3LL cells was performed as described previously (46). Briefly, 1 mL of chloroform and 2 mL of methanol were mixed with 0.8 mL of necrotic supernatant and were incubated for 5 min at room temperature. We mixed 1 mL of chloroform and 1 mL of PBS with the above solution. The mixture was then centrifuged at 1,000 × g for 2 min. The lower layer was extracted, dried up, dissolved in DMSO, and used as total lipid.

Gel Filtration Chromatography and Mass Spectrometry Analysis.

Gel filtration chromatography was performed on a Superdex 75 10/300 GL column by a ÄKTA purifier (GE Healthcare). The column was equilibrated with 2 column volumes of PBS, and the supernatant from necrotic cells (1 mL) was loaded on the column. Eluent drops (18 drops) were continuously collected into wells of a 48-well plate using AC-5700P MicroCollector (ATTO). The fractions of 17–21 in Fig. S1D were subjected to mass spectrometry analysis (Chemicals Evaluation and Research Institute).

Measurement of LDH.

RAW 264.7 cells (1.5 × 10⁵ cells) were seeded in a 48-well plate and cultured overnight. These cells were then treated with DMSO or PGE2 (50 nM or 500 nM) and cultured for the indicated time. Lactate dehydrogenase (LDH) levels in cultured media were determined by Cytotoxicity Detection Kit Plus LDH (Roche), according to the manufacturer’s protocol.

Immunoblot Analysis.

Immunoblot analysis was performed as described previously (45). Antibodies for JNK (9252S), p-JNK (9251S), ERK (4695S), p-ERK (4370S), p-IκBα (2859S), and p-IRF3 (4947S) were purchased from Cell Signaling. Antibodies for β-actin (AC15) and IRF3 (A303-384A) were purchased from Sigma and BETHYL, respectively. β-actin was used as a loading control.

Reporter Assay.

RAW 264.7 cells (5 × 10⁶) seeded on a 10-cm Petri dish (BD Biosciences) were transiently cotransfected with X-tremeGENE 9 DNA transfection reagent (Roche Life Science) and 5 μg firefly luciferase reporter plasmid for NF-κB (47). After 24 h of the transfection, the cells were reseeded on a 24-well plate and stimulated by LPS (5 ng/mL) without or with PGE2 (50 nM and 500 nM). After 2 or 4 h of incubation, cells were harvested with lysis buffer, and luciferase activity was measured with a PicaGene Dual kit (TOYO B-Net Co.).

ChIP–qPCR.

ChIP was performed with a Magna ChIP G Chromatin Immunoprecipitation kit (Millipore) and anti-p65 antibody (sc-372; Santa Cruz Biotechnology), according to the manufacturer’s protocol. Specific primers for the Tnf promoter region were 5′-TTGCCACAGAATCCTGGTGG-3′ (forward) and 5′-ATGTGGAGGAAGCGGTAGTG-3′ (reverse). ChIP signals were quantified by qPCR analysis with LightCycler480 and SYBR Green PCR Master Mix.

Microarray Analysis.

Peritoneal macrophages were stimulated with a mock- or indomethacin-treated necrotic supernatant of SL4 cells for 2 h. Total RNA was extracted from cocultured peritoneal macrophages and analyzed by GeneChip Mouse Genome 430 2.0 Array (Affymetrix). Differentially expressed genes between the two samples (>1.5-fold up or down) were subjected to a cluster analysis by R version 2.15.1 (R Foundation for Statistical Computing).

Acetaminophen-Induced Hepatitis.

Acetaminophen intoxication was induced in overnight-starved male mice by i.p. injection of 500 mg/kg acetaminophen (Sigma-Aldrich). Mice were administered 100 mg/kg of celecoxib (Cayman chemical) or ethanol at the same time. After 24 h of the acetaminophen injection, serum was collected from the tail vein and subjected to ELISA for TNF-α. ALT and AST levels were examined by TA-LN kainos assay (KAINOS).

CRISPR/Cas9.

Ptgs2^−/− SL4 cells were generated by CRISPR/Cas9-mediated genome engineering using the CRISPR design tool (www.genome-engineering.org). The genomic sequence of the Ptgs2 gene (5′-GCGGACTCCACGTGACGTAG-3′ and 5′-CAATAGTCTTAATGGCTTAC-3′) was targeted. Oligonucleotides corresponding to these guide sequences were cloned into the BbsI site of pSpCas9(BB)-2A-GFP (PX458) (Addgene), a bicistronic expression vector encoding both Cas9 and the single guide RNA. The expression vector construct was transiently transfected in SL4 cells, and then GFP-expressing cells were sorted by FACSAria (BD Biosciences).

S.c. Tumor Growth Assay.

Mice were injected s.c. with 1 × 10⁶ SL4 cells. Tumor volume was evaluated using the equation, volume = πab²/6, where a and b are the lengths of the major and minor axes, respectively.

Lung Metastasis Assay.

Mice were i.v. injected with 2 × 10⁵ of SL4 cells. After 17 d of injection, lungs were subjected to histlogical analysis, assessed by microscopy of sections stained with hematoxylin and eosin.

Flow Cytometry Analysis.

Single-cell suspensions were prepared from the tumor. These cells were stained with the following antibodies: FITC-conjugated anti-CD11c, Alexa647-conjugated anti-CD206, PECy7-conjugated anti-Gr1, PerCPCy5.5-conjugated anti-F4/80, PE-conjugated anti-CD11b, and Pacific Blue-conjugated anti-CD45.2 for the myeloid lineage; and FITC-conjugated anti-NK1.1, APC-conjugated anti-B220, PECy7-conjugated anti-CD4, PerCPCy5.5-conjugated anti-CD8, PE-conjugated anti-CD3ε, and Pacific Blue-conjugated anti-CD45.2 for the lymphoid lineage. DCs were defined as CD45⁺F4/80⁻CD11c⁺ cells. Neutrophils were defined as CD45⁺Gr1⁺CD11b⁺ cells. Macrophages were defined as CD45⁺CD11b⁺F4/80⁺ cells. NK cells were defined as CD45⁺NK1.1⁺CD3ε^– cells. B cells were defined as CD45⁺B220⁺ NK1.1⁻ cells, and T cells were defined as CD45⁺CD3ε⁺NK1.1⁻ cells. Antibodies were purchased from BioLegend. The cells were then analyzed using LSRII Fortessa flow cytometer (BD Biosciences).

Macrophage Differentiation Analysis.

Bone-marrow cells were prepared and cultured with M-CSF (Peprotech) as described previously (44). Cells (2.5 × 10⁵ cells) were seeded in 48-well plates overnight and then treated as follows: for M1 macrophage differentiation, LPS (50 ng/mL) and IFN-γ (100 ng/mL); for M2 differentiation, IL-4 (20 ng/mL). The necrotic supernatant of SL4 cells was added concomitantly. Seven hours after the treatment, cells were harvested and total RNA was extracted.

Statistical Analysis.

Data are expressed as mean ± SD. Student’s t test was performed, and the difference was considered to be statistically significant at P < 0.05.