Prostaglandin E2 promotes intestinal inflammation via inhibiting microbiota-dependent regulatory T cells

PGE2 inhibits Tregs and promotes intestinal inflammation through actions on mononuclear phagocytes and the gut microbiota.


INTRODUCTION
Inflammatory bowel disease (IBD) is a chronic inflammatory disorder of the intestine that causes abdominal pain, diarrhea, bleeding, and increased risk of intestinal cancer. There are two main subtypes of IBD, i.e., Crohn's disease (CD) and ulcerative colitis. Multiple factors including lifestyle (e.g., smoking, diet, medication, and psychological state), environmental risk factors (e.g., infections and air pollution), genetic and epigenetic alterations, and host immune functions can potentially trigger the development and progression of IBD (1)(2). The gut microbiota plays a critical role in maintaining health of the host. Dysfunction of this symbiosis may result in development of various human diseases such as IBD, metabolic syndrome, infections, allergy, and cancer (3)(4). Interplay between the host and gut microbiota controls intestinal homeostasis and inflammatory responses through mechanisms that involve modulation of gut-resident regulatory T cells (T regs ), which express the transcription factor, forkhead box P3 (Foxp3) (5)(6)(7)(8). Dysregulation of intes-tinal T regs is implicated in the pathogenesis of IBD (9). Microbial antigens, metabolites [e.g., vitamins and short-chain fatty acids (SCFAs)], and signaling molecules released during tissue damage (e.g., alarmins) contribute to the induction of distinct intestinal T reg subsets, which play critical roles in intestinal homeostasis and control mucosal inflammation (10)(11)(12)(13). However, the mechanisms that negatively regulate microbiota-T reg cross-talk for IBD pathogenesis are incompletely studied.
Prostaglandins (PGs) are bioactive lipid mediators that are generated from arachidonic acid via cyclooxygenases (COXs) and specific PG synthases (14). The PG family has five members including PGE 2 , PGD 2 , PGF 2 , PGI 2 , and thromboxane A 2 (TXA 2 ). PGs signal in an autocrine and/or paracrine manner through their distinct G protein-coupled receptors including receptors EP1 to EP4, PGD 2 receptors DP1 and DP2, PGF 2 receptor FP, PGI 2 receptor IP, and TXA 2 receptor TP. PGE 2 is present in most tissues at biologically functional nanomolar levels in the steady state, and its levels are increased at the sites of inflammation (14)(15)(16). Nonsteroidal antiinflammatory drugs (NSAIDs), such as aspirin and indomethacin, are widely used to reduce pain, fever, and inflammation by inhibiting COX activities and therefore decreasing PG production. However, NSAIDs are generally avoided for individuals who have gut conditions due to the gastrointestinal adverse effects (17). This is because PGE 2 plays critical roles in maintaining the gut epithelium, protecting against acute damage, and facilitating regeneration after injury through actions on various cell types including macrophages, epithelial, stromal, and innate lymphoid cells (18)(19)(20)(21).
Genome-wide association studies have revealed that polymorphisms in the PTGER4 gene (encoding human PGE 2 receptor EP4) are associated with overexpression of EP4 and a more severe disease phenotype in patients with IBD (22)(23)(24). Moreover, variants in the PTGER4 gene exert a significant association with CD, in third place among all susceptible genetic loci after variants in NOD2 and IL23R

Increase of intestinal T regs by COX inhibition
To test whether endogenous PGs regulate intestinal T regs in the steady state, we administered WT mice with indomethacin in drinking water and analyzed T regs in various organs including colons, small intestines, mesenteric lymph nodes (mLNs), and spleens. Indomethacin treatment increased the accumulation of Foxp3 + T regs in all of these tissues with greater effects in the intestinal lamina propria (LP) (by ~2-fold in the colon and small intestine) than that in the mLN and spleen (both by ~1.4-fold) (Fig. 2, A and B). Furthermore, indomethacin significantly increased mean fluorescence intensity (MFI) of Foxp3 among Foxp3 + T regs in the colon (Fig. 2B), suggesting that inhibition of endogenous PG biosynthesis not only increases intestinal T reg frequencies but also enhances Foxp3 expression at the single-cell level. We also administered mice with indomethacin at lower doses of 1 to 2 mg/kg of body weight per day, which is the equivalent of ~6 to 12 mg/day for adults weighing 75 kg, levels known not to induce intestinal damage in human. Similarly, low doses of indomethacin still increased intestinal Foxp3 + T regs (fig. S2).
It has been recently reported that a subpopulation of intestinal T regs that express the transcription factor retinoid-related orphan receptor gamma t (RORt), namely, RORt + Foxp3 + T regs , inhibit intestinal inflammation with greater suppressive potential than RORt − Foxp3 + T regs (6,7,31). We therefore examined the effects of endogenous PGs on RORt + Foxp3 + T regs and found that administration of indomethacin markedly increased the percentages of the RORt + Foxp3 + subpopulation among Foxp3 + T regs in the colon, but not in the mLN or spleen (Fig. 2, C and D). Furthermore, indomethacin also boosted absolute numbers of colonic RORt + Foxp3 + T regs , but not the numbers of RORt − Foxp3 + T regs or RORt + Foxp3 − T H 17 cells (Fig. 2E). These results indicate that inhibition of endogenous PGs increases the accumulation of intestinal T regs .

Reversion of COX inhibition-dependent increase in intestinal T regs by EP4 agonism
As indomethacin inhibits all PG production, we next examined whether PGE 2 and its receptors control intestinal T regs . We coadministered mice with indomethacin together with a selective EP2 agonist (Butaprost) or selective EP4 agonists (L-902,688). We found that increase in colonic T regs by indomethacin was prevented by the EP4 agonist in colons and mLNs, while the EP2 agonist only reduced T regs in mLNs (Fig. 2, A and B), indicating that PGE 2 suppresses colonic T reg accumulation mainly through EP4. Similarly, coadministration of EP4 agonist notably down-regulated RORt + Foxp3 + T regs in the colon and likely in the small intestine, but not in the mLN or spleen (Fig. 2, C and D). Furthermore, selective activation of EP4 also decreased absolute numbers of colonic RORt + Foxp3 + T regs , but not RORt − Foxp3 + T regs or RORt + Foxp3 − T H 17 cells (Fig. 2E). These results indicate that activation of PGE 2 -EP4 signaling overturns NSAID-dependent augmentation of T regs with the greatest potency in the intestine.
To further examine whether exogenous activation of EP4 suppresses colonic T regs , we injected the EP4 agonist alone into naïve WT C57BL/6 mice without coadministration of indomethacin. Unlike activation of EP4 in indomethacin-treated mice where all endogenous PG production was inhibited (Fig. 2), administration of EP4 agonist into naïve WT mice where all PG signaling pathways were intact had few effects on colonic T reg accumulation, albeit a trend for reducing the proportion of RORt + Helio − T regs ( fig. S3). This may be due to the already high levels of PGE 2 and other EP4 ligands (e.g., PGE 1 ) in naïve intestines (Fig. 1A). There is another possibility that blockade of endogenous PGs by indomethacin disrupted the intestinal epithelial line (19), which leads to attachment and translocation of more invasive commensal microbes and in turn increases the accumulation of colonic T regs . Coadministration of EP4 agonist prevented indomethacin-dependent epithelial damage and attachment of invasive microbes, resulting in reduction of colonic T reg accumulation. In the naïve intestine without disruption of the epithelial integrity (i.e., without COX inhibition), however,

of 16
EP4 agonism has thus no influence on T reg accumulation in the colon. Nevertheless, these results collectively indicate that PGE 2 -EP4 signaling has a potency to inhibit intestinal T reg accumulation.

Involvement of the gut microbiota in PGE 2 suppression of intestinal T regs
To examine whether PGE 2 -EP4 signaling inhibits intestinal T regs in the steady state via direct or indirect actions on T cells, we generated Lck-Cre-driven EP4 conditional knockout mice by crossing EP4-flox mice (32) to Lck-Cre mice (i.e., Lck Cre EP4 fl/fl mice) to delete EP4 expression in Lck-expressing T cells. Lck Cre EP4 fl/fl and control mice had comparable colonic T reg accumulation in the steady state (Fig. 3, A and B), suggesting that PGE 2 inhibits intestinal T reg accumulation in the steady state independent of EP4 signaling in T cells.
The gut microbiota is crucial for the development of intestinal T regs , especially the RORt + Foxp3 + T reg subset (6,7,31). We therefore investigated whether the gut microbiota is involved in PGE 2 -dependent control of intestinal T regs . We analyzed colonic T regs from WT mice RNA sequencing (RNA-seq) data of normal C57BL/6 mouse small intestines (n = 3) were retrieved from the GEO dataset GSE97371. (C) Gene expression of PG receptors in human sigmoid (n = 149) and transverse (n = 104) colon biopsy samples of healthy individuals. RNA-seq data were downloaded from the Genotype-Tissue Expression project database and analyzed using Python 3.7.0. RPKM, reads per kilobase of transcript. Each scatter dot plot in bar graphs represents data from one mouse (A and B) or individual (C). Data shown as means ± SD (A and B) or presented as violin bars with scatter plots (C) are analyzed by analysis of variance (ANOVA) with post hoc Holm-Sidak's multiple comparisons test. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. n.s., not significant.
in which the gut microbiota had been depleted by antibiotics, or from MyD88/TRIF double knockout mice deficient in both myeloid differentiation primary response 88 (MyD88) and TRI domain containing adaptor-inducing interferon-beta (TRIF) that were unable to sense microbial signals. Inhibition of endogenous PGE 2 by indomethacin increased both the frequencies and numbers of colonic T regs (especially the subpopulation of RORt + Foxp3 + T regs ) but not RORt + Foxp3 − T H 17 cells in WT mice that have been treated with vehicle control (Fig. 3C). In contrast, there was no increased accumulation of colonic T regs by indomethacin in antibiotic-treated WT mice nor in those mice with dual deficiency of MyD88 and TRIF (Fig. 3C). Similarly, activation of EP4 reduced indomethacin-dependent increase in colonic RORt + Foxp3 + T regs in vehicle-treated, rather than antibiotic-treated, mice (Fig. 3C). These results suggest involvement of the commensal microbiota in the PGE 2 -dependent control of intestinal T regs .

Alteration of the gut microbiota by PGE 2 -EP4 signaling
Use of NSAIDs has been reported to induce changes in the gut microbiota composition in humans and rodents (33)(34)(35). To examine whether PGE 2 -EP4 signaling modulates the gut microbiota, we collected (E) Absolute numbers of RORt + Foxp3 + T regs , RORt − Foxp3 + T regs , RORt + Foxp3 − T H 17 cells in colon LP (cLP). Each scatter dot plot in bar graphs represents data from one mouse. Data shown as means ± SD are pooled from four independent experiments and analyzed by ANOVA with post hoc Holm-Sidak's multiple comparisons test (B, D, and E). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. n.s., not significant.
cecal contents from mice that had been treated with indomethacin or that had been cotreated with indomethacin and an EP4 agonist and performed 16S ribosomal RNA (rRNA) gene metabarcoding to study microbiota composition. Principal components analysis (PCA) suggested that there were no differences in overall -diversity of gut microbiota signatures measured by unweighted UniFrac distances among the three groups (Fig. 4A). The analysis of -diversity indices (i.e., richness and evenness) showed that the three groups had also comparable observed operational taxonomic units (OTUs), Chao1 index, Shannon diversity, and Inverse Simpson (InvSimpson) indices (Fig. 4B). However, there were trends showing that the indomethacin + EP4 agonisttreated group had higher observed OTUs and Chao1 index (Fig. 4B).
We then asked whether PGE 2 -EP4 signaling alters specific bacterial communities. We found that treatment with indomethacin increased the abundance of the Firmicutes phylum and reduced the abundance of the Bacteroidetes phylum, and the changes in phylum-level microbiota composition by indomethacin were slightly reversed by cotreatment with the EP4 agonist (Fig. 4C). Indomethacin increased, but EP4 agonist reduced, several SCFA-producing bacteria belonging to the Muribaculaceae family or the Clostridium cluster XIVa (e.g., Lachnospiraceae and Ruminococcaceae) (Fig. 4D) (5).
SCFA-producing bacteria such as Clostridia play critical roles in intestinal T reg induction and accumulation. Furthermore, Anaeroplasma bactoclasticum, which promotes expression of immune-regulatory transforming growth factor- in the gut (36), was also up-regulated by indomethacin and reduced by the EP4 agonist (Fig. 4D). To validate the 16S rRNA gene sequencing results, we used real-time quantitative polymerase chain reaction (qPCR) to detect gene expression of SCFA-producing bacteria in mice from independent cohorts. As confirmed, activation of EP4 significantly reduced the phylum Firmicutes and increased the phylum Bacteroidetes (Fig. 4E). In agreement with the 16S RNA sequencing (RNA-seq) results, EP4 activation notably decreased the amounts of Clostridia including total Clostridium cluster XIVa, Clostridium sp., Clostridium coccoides, ASF500 (Ruminococcaceae), and ASF360 (Lactobacillus sp.) (Fig. 4E). In addition, several aggressive microbial species such as species of the genera Rikenella and Escherichia were also increased by indomethacin but reduced by coadministration of EP4 agonist (Fig. 4D). Rikenella and Escherichia are pathogenic strains that can trigger mucosal inflammation and formation of mucus lesions (37). Therefore, besides SCFA-producing gut microbes, these pathogenic bacteria may also contribute to NSAID/PGE 2 -dependent (C) Percentages and numbers of RORt + Foxp3 + T regs , RORt − Foxp3 + T regs , and RORt + Foxp3 − T H 17 cells in colon LP of WT and MyD88/TRIF double knockout (DKO) mice that were treated with vehicle, indomethacin, or indomethacin plus EP4 agonist, L-902,688 (n = 7 to 18). Antibiotics (ABX) were administered to WT mice in drinking water for 2 weeks started 1 week before receiving indomethacin or vehicle. Each scatter dot plot in bar graphs represents data from one mouse. Data shown as means ± SD are pooled from two (B) or five (C) experiments and analyzed by ANOVA with post hoc Holm-Sidak's multiple comparisons test. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. n.s., not significant. . Amounts of SCFAs were detected using gas chromatography and normalized to the vehicle group. Each scatter dot plot represents data from one mouse (A, B, and D to F). Data plotted in box and whiskers bar graphs are pooled from five experiments and analyzed by nonparametric Kruskal-Wallis test with post hoc Dunn's multiple comparisons test (E and F). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. modulation of gut epithelial injury and accumulation of T regs in the intestine.
To examine whether PGE 2 modulates SCFA production, we measured SCFA levels in cecal contents. Indomethacin treatment significantly increased the levels of caproic acids and branched SCFAs (isobutyric and isovaleric acids) in extracts of cecal contents (Fig. 4F). Indomethacin also had a trend to enhance valeric acid (Fig. 4F). Coadministration of EP4 agonist specifically reduced the levels of caproic acid (Fig. 4F). The levels of caproic and valeric acids were found significantly decreased in feces from patients with active CD compared to inactive CD or healthy controls (38,39). Clostridia are producers of caproic acid and branched SCFAs including isobutyric and isovaleric acids (40,41). Branched SCFAs, similar to conventional SCFAs modulate the host immune response (42). These corroboratory results thus suggest that PGE 2 -EP4 signaling reduces intestinal T reg accumulation, at least partially, via reducing SCFA-producing microbiota.

Modulation of mucosal T regs and intestinal inflammation by EP4-modified gut microbiota
To investigate whether PGE 2 -modified microbiota modulates intestinal T reg responses and inflammation, we adoptively transferred cecal microbiota obtained from WT C57BL/6 mice that had been treated with indomethacin, indomethacin plus EP4 agonist, or vehicle control into recipient WT C57BL/6 mice (Fig. 5A). Recipient mice were pretreated with antibiotics in drinking water for 2 weeks before receiving transplantation of microbiota and then received normal drinking water for another 10 days, followed by euthanasia to analyze colonic immune cell responses at the steady state. Some recipient mice received normal drinking water for 8 days after stopping antibiotic treatment, followed by receiving dextran sulfate sodium (DSS) in drinking water for an additional 6 days (Fig. 5A). Compared to mice that had received cecal microbiota from vehicle-treated mice, mice transplanted with cecal microbiota from indomethacin-treated mice had increased T regs and Foxp3 expression at single-cell levels in the steady state (Fig. 5B). This increased expression was associated with prevention of body weight loss, reduced disease activity index (DAI), and increased colon length under DSSinduced inflammatory conditions in mice that had received cecal microbiota from indomethacin-treated mice (Fig. 5, C and D). In contrast, mice transplanted with cecal microbiota from mice that have been pretreated with indomethacin plus EP4 agonist had a trend to reduce colon T regs and Foxp3 expression compared to mice received cecal microbiota from mice that have been pretreated with indomethacin (Fig. 5B). Furthermore, the severity of colitis in mice transplanted with cecal microbiota from mice that have been pretreated with both indomethacin and EP4 agonist was similar to mice that had received cecal microbiota from vehicle-treated mice but was significantly greater than that in mice received cecal microbiota from indomethacintreated mice (Fig. 5, C to E). Histological analysis showed near normal proximal and distal colonic mucosa, or only scattered mild inflammatory changes, in mice transplanted with cecal microbiota from indomethacin-treated mice (Fig. 5F). There was widespread and variably severe mucosal ulceration with patches of almost complete loss of the crypt epithelium, with marked infiltration of fibrotic mucosal tissue by both acute and chronic inflammatory cells in mice that had received cecal microbiota from vehicle-treated mice, with more severe inflammatory and ulcerative changes in the distal colon compared with the proximal colon. In contrast, there was a less severe pattern of inflammation with more variable mild to moderate inflammatory cell infiltration with only patchy partial loss of mucosal crypt epithelium in those mice transplanted with cecal microbiota from indomethacin-and EP4 agonist-treated mice, again with greater inflammatory changes in the distal colon compared with the proximal colon (Fig. 5F). This was associated with reduced T cell infiltration to colon LP in mice that have received cecal microbiota from indomethacin-treated mice compared to the other two groups (Fig. 5G). Together, these results suggest that PGE 2 -EP4 signaling-modified gut microbiota contributes to inhibition of mucosal T regs and exacerbation of intestinal inflammation.

Increased colonic T reg accumulation in MNP-specific EP4-deficient mice
MNPs are critical to mediate the microbiota-dependent generation of intestinal Foxp3 + T regs (43)(44)(45). To further study the interplay between PGE 2 and the gut microbiota in the control of intestinal T regs , we examined MNPs in the colon. Comparing to treatment with vehicle control, administration of indomethacin increased both the frequency and number of colonic CD11c + MHC II + CD11b + MNPs, and this was again inverted by coadministration of the EP4 agonist (Fig. 6A). CD11c + MHC II + CD11b − MNPs were not affected by either indomethacin or EP4 activation (Fig. 6A). The effects of indomethacin and EP4 agonist on colonic CD11c + MHC II + CD11b + MNPs were invisible in colons of antibiotic-treated mice or in MyD88/TRIF double-deficient mice (Fig. 6A). Moreover, transfer of gut microbiota from mice that had been treated with indomethacin increased colonic CD11c + MHC II + CD11b + MNPs in host mice, and this was again reduced by transfer of gut microbiota from mice that had been cotreated with indomethacin and EP4 agonist (Fig. 6B). These results indicate that PGE 2 -EP4 signaling suppresses intestinal MNPs through modulating the gut microbiota.
We further asked whether EP4 signaling in MNPs suppresses colonic T regs . To address this question, we crossed EP4-flox mice to CD11c-Cre mice to generate MNP-specific EP4-deficient mice. Inactivation of EP4 signaling in MNPs increased both percentages and absolute numbers of Foxp3 + T regs in the colon (Fig. 6, F and G). and numbers (bottom) of colon LP CD11c + MHC II + CD11b + and CD11c + MHC II + CD11b − MNPs from C57BL/6 WT and MyD88/TRIF double knockout mice treated with vehicle, indomethacin, or indomethacin plus EP4 agonist L-902,688 (n = 7 to 18). Antibiotics were administered to WT mice in drinking water for 2 weeks started 1 week before receiving indomethacin or vehicle. (B) Percentages (top) and numbers (bottom) of colon LP CD11c + MHC II + CD11b + and CD11c + MHC II + CD11b − MNPs in mice that were treated with antibiotics for 2 weeks, followed by transfer of cecal microbiota harvested from mice that had been administered with vehicle (Veh-CMT), indomethacin (Indo-CMT), or indomethacin plus EP4 agonist (Indo/ EP4 ago-CMT) (n = 5 to 6). (C and D) Representative flow cytometry plots (C) and percentages (D) of colon LP CD11c + MHC II + CD11b + and CD11c + MHC II + CD11b − MNPs at colon LP of CD11b-DTR mice administered with diphtheria toxin (DT) or phosphate-buffered saline (PBS) (n = 6 each group). CD11b geometric gMFI of CD11c + MHC II + CD11b + MNPs are also shown (D, right). (E) Numbers of colon LP CD45 + CD3 + CD4 + Foxp3 + T regs in CD11b-DTR mice treated with DT or PBS together with vehicle control, indomethacin, or indomethacin plus EP4 agonist (n = 6 to 12). (F) Representative flow cytometry plots of colonic Foxp3 + T regs gated on live CD45 + CD3 + CD4 + T cells and RORt + or Helios + T regs in CD11c Cre EP4 fl/fl (n = 6) or control EP4 fl/fl mice (n = 5). (G) Percentage and numbers of Foxp3 + T regs . (H) Percentage and numbers of RORt + Helios − , RORt − Helios − or Helios + T regs . Each scatter dot plot in bar graphs represents data from one mouse. Data shown as means ± SD are pooled from five (A), three (D and E), or one (B, G, and H) independent experiments and analyzed by ANOVA with post hoc Holm-Sidak's multiple comparisons test (A, B, and E) or two-tailed unpaired Student's t test (D, G, and H). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. n.s., not significant. Subsequent analysis revealed that EP4 deficiency in MNPs specifically boosted colonic RORt + Helios − T regs and had few, if not inhibitory, effects on the RORt − Helios − T reg subset (Fig. 6, F and H). Together, these results indicate that EP4 signaling in MNPs is responsible for PGE 2 inhibition of colonic RORt + T regs .

Impaired type I IFN signaling responsible for PGE 2 suppression of intestinal MNPs and T regs
MNPs mediate intestinal T reg development and expansion by producing soluble mediators such as type I IFNs (46,47). The IFN-/ receptor (IFNAR) signaling pathway has been shown to be required for T reg development and function, especially under stress conditions or in a competitive environment (48). To test whether PGE 2 regulates type I IFN signaling in intestinal CD11c + MHC II + CD11b + MNPs, we cultured bone morrow-derived dendritic cells (BMDCs) without (medium only) or with cecal microbial products (CMPs; which might include pathogen-or microbe-associated molecular patterns including gut microbial metabolites) obtained from mice that had been treated with vehicle control, indomethacin, or indomethacin plus EP4 agonist. We then measured mRNA expression of type I IFN signaling pathway genes in BMDCs by real-time qPCR. BMDCs cultured with CMPs from control mice were characterized by higher expression of type I IFN (e.g., Ifnb) and the downstream genes (e.g., Irf7 and Isg15) compared to that cultured with medium only (Fig. 7A). Expression of Irf7 and Isg15 was further increased in BMDCs cultured with CMPs obtained from mice that had been treated with indomethacin compared to that cultured with CMPs obtained from control mice (Fig. 7A). In contrast, mRNA expression levels of all these genes were markedly down-regulated in BMDCs that had been cultured with CMPs obtained from mice that have been cotreated with indomethacin and EP4 agonist (Fig. 7A). Moreover, colonic CD11c + MHC II + MNPs sorted from indomethacintreated mice expressed higher levels of type I IFN signaling pathway genes (e.g., Ifnb, Irf7, and Isg15) compared to MNPs sorted from colons of vehicle-treated mice (Fig. 7B). Colonic MNPs in mice treated with indomethacin had higher levels of phosphorylated signal transducers and activators of transcription 1 (STAT1) than that in vehicle-treated mice (Fig. 7C). These results suggest that modulation of gut microbial products by PGE 2 -EP4 signaling inhibits type I IFN production and signaling in MNPs. Similarly, colonic T regs from indomethacin-treated mice also had more T regs expressing phosphorylated STAT1 compared to vehicle-treated mice (Fig. 7D).
We next examined whether type I IFN signaling is required for PGE 2 control of intestinal T regs using IFNAR  chain (Ifnar)-deficient mice. Again, indomethacin increased colonic CD11c + MHC II + CD11b + MNPs and RORt + Foxp3 + T regs , and this was prevented by EP4 agonist in WT mice (Fig. 7 E and F). However, neither indomethacin nor EP4 agonist had effects on colonic CD11c + MHC II + CD11b + MNPs and RORt + Foxp3 + T regs in Ifnar-deficient mice (Fig. 7, E and F). These results thus suggest that the PGE 2 -modified microbiota suppresses intestinal T regs by down-regulating type I IFN signaling via MNPs.

DISCUSSION
In this report, we have demonstrated that PGE 2 , the well-known mediator of inflammation, negatively controls intestinal T reg responses. Endogenous PGE 2 -EP4 signaling alters the gut microbial community, e.g., reducing SCFA-producing bacteria, which in turn down-regulates MNP production of type I IFNs, leading to repression of intestinal T reg accumulation and augmentation of intestinal inflammation (Fig. 7G). In contrast, blockade of PGE 2 signaling (e.g., by NSAIDs) increases beneficial microbes and also some invasive microbes, likely as a consequence of damage to the epithelium, which augments MNP production of IFNs, resulting in expansion of mucosal T regs and limitation of intestinal inflammation.
There are several potential mechanisms underpinning EP4dependent inhibition of intestinal T regs as PGE 2 could target multiple enteric cell types including epithelial cells, T cells, and MNPs. Mice with EP4 deficiency in T cells had comparable numbers of intestinal T regs to EP4-sufficient mice, excluding a role for EP4 signaling in T cells. Specific deletion of EP4 in epithelial cells (e.g., driven by Villin-cre) impaired colon homeostasis, increased epithelial cell apoptosis and immune cell infiltration, and exacerbated DSS-induced mucosal inflammation (49). Although T reg accumulation in the colon in epithelial EP4-deficient mice has not been assessed, gene expression of Foxp3 that is exclusively expressed on T regs in colon tissue is comparable between epithelial-specific EP4 deficient and control mice (49). Thus, endogenous PGE 2 is unlikely to act through epithelial cells for T reg inhibition in the naïve intestine. However, use of NSAID (e.g., indomethacin) may destroy the epithelium, leading to translocation of commensal microbes (19). Attachment of some invasive commensal microbes (e.g., Helicobactor hepaticus) to the epithelium is believed to favorably increase T reg accumulation in the gut (50). Specific activation of EP4 agonism may thus contribute to reversing NSAID-dependent increase in intestinal T regs by preventing NSAID-dependent epithelial injury. This is, at somewhat extend, supported by the finding that EP4 agonist alone (without indomethacin) had few effects on colonic T reg accumulation ( fig. S3). Therefore, a possibility that indomethacin increases colonic T regs , at least partly, through epithelial injury and attachment of the invasive microbes cannot be excluded.
Blockade of PGE 2 biosynthesis by NSAIDs alters the gut microbiota by expanding the beneficial SCFA-producing microbes and increasing SCFA secretion in the intestine, which was reversed by the EP4 agonist. Given an increased ratio of Firmicutes to Bacteroidetes has been reported to be related to anti-inflammatory potential under various inflammatory conditions (3), our results showing that PGE 2 -EP4 signaling reduced Firmicutes but increased Bacteroidetes at the phylum level may have implications in modulation of intestinal inflammation. Our gut microbiota transfer experiments demonstrated that EP4-modified commensal microbes efficiently suppressed intestinal T regs and mediated intestinal inflammation. Furthermore, our findings on the manipulation of gut microbiota in mice are in keeping with observations in humans showing that ingestion of antipyretic analgesics (i.e., NSAIDs or paracetamol that also can inhibit PG synthesis) increased the abundance of beneficial gut commensals such as Verrucomicrobia and SCFAproducing bacteria such as Butyrivibrio and Clostridiaceae (35). A recent report on in vitro culture of commensal bacteria suggested that most NSAIDs, including indomethacin and aspirin, were unlikely to affect bacterial growth on various microbial strains (51). PGE 2 was reported to promote the growth of some bacterial strains such as Escherichia coli (52). Therefore, further studies are needed to decipher whether PGE 2 acts directly on the gut microbiota or indirectly on host cells to regulate the growth, survival, or function of specific gut commensal bacterial strains.
Another possible mechanism for EP4 suppression of colonic T regs is through a direct action on MNPs. Manipulation of PGE 2 -EP4 signaling (i.e., using COX inhibitors or EP4 agonist) alters the numbers of intestinal CD11c + CD11b + MHC II + MNPs, and depletion of CD11b + MNPs prevented PGE 2 -dependent regulation of intestinal T regs . CCR2-expressing monocyte-derived MNPs are critical for intestinal T H 17 cell responses (53), but these migratory MNPs are unlikely to be involved in PGE 2 -dependent regulation of intestinal T reg responses, indicating a role for gut resident MNPs. Notably, deletion of EP4 in CD11c + MNPs increased accumulation of colonic T regs , especially the RORt + subset, indicating an essential role for EP4 signaling in MNPs. Thus, PGE 2 -modified gut microbi-ota controls MNP production of type I IFNs, which in turn mediates intestinal T regs and MNPs themselves. This is consistent with a recent report showing that gut commensals stimulated CD11b + dendritic cells to produce IFN-, which augmented the proliferation of intestinal T regs (46), although IFNAR signaling in T regs was reported to impair T reg suppressive function (54). PGE 2 -EP4 signaling more specifically suppresses RORt + T regs than the RORt − T reg subpopulation. Comparing to RORt − Foxp3 + T regs , RORt + Foxp3 + T regs express higher levels of mucosal resident T reg -associated genes such as Ffar2 (31) that encodes the SCFA receptor G protein-coupled receptor 43 (GPR43). The SCFA-GPR43 axis is critical for the development, recruitment, and expansion of colonic T regs (10). IFNAR-sufficient T regs similarly express greater Ffar2 and Rorc genes than IFNAR-deficient T regs (54). Therefore, higher levels of SCFA and type I IFN in the intestine of indomethacintreated mice may enhance the accumulation of GPR43-expressing RORt + T regs compared to that in the mice that were treated with vehicle or indomethacin plus EP4 agonist. This may be, at least partially, responsible for a specific effect of the EP4-IFN axis on RORt + T reg subpopulation.
Overactivation of the PGE 2 pathway including PGE 2 biosynthesis and its receptor signaling is an outstanding marker under most, if not all, human inflammatory conditions such as multiple sclerosis, rheumatoid arthritis, IBD, and cancers (14). Together with our previous findings (26,27), results from this current study imply PGE 2 as a common inflammatory mediator for immune inflammation by balancing T cell responses, i.e., enhancing T reg but limiting effector T cell responses. This is further supported by a recent study demonstrating that PGE 2 exacerbates tumor necrosis factor (TNF)induced inflammatory responses in human intestinal epithelial cells from patients with IBD who are resistant to TNF inhibitor therapy (55). Moreover, studies from multiple groups including ourselves have shown that lack of PGE 2 -EP4 signaling in T cells reduced both chemical-triggered acute and naïve T cell transfer-induced chronic intestinal inflammation, associated with reduction of inflammatory T H 1 and/or T H 17 cell responses (27,56). Thus, PGE 2 mediates intestinal inflammation possibly through actions on both adaptive T cells and innate MNPs, and at least in the latter scenario, the gut microbiota is involved.
Collectively, we have defined a role for PGE 2 in control of intestinal T reg accumulation, which involves MNPs and the gut microbiota, leading to facilitation of mucosal inflammation. Our findings provide an explanation for human genetic findings of the association between PTGER4 (EP4) gene polymorphisms and IBD susceptibility and suggest a potential therapeutic strategy for treating intestinal inflammation by targeting the PGE 2 -EP4-microbiota-MNP-T reg cascade.

Mice
Rag1 −/− , MyD88 −/− TRIF −/− , Ifnar −/− , CD11b-DTR, Ccr2 −/− , Lck Cre EP4 fl/fl , and WT C57BL/6 mice were bred and maintained under specific pathogen-free conditions in accredited animal facilities at the University of Edinburgh. To generate MNP-specific EP4-deficient mice, we crossed EP4-floxed mice to CD11c-Cre-GFP mice (JAX 007567). WT mice were bred in our own animal facilities or purchased from Harlan, UK. Age (>7 weeks old) and sex-matched mice were used. Mice were randomly allocated into different groups and analyzed individually. No mice were excluded from the analysis except exclusions due to technical errors in preparation of intestinal LP leukocytes. All experiments were conducted in accordance with the U.K. Animals (Scientific Procedures) Act of 1986 with local ethical approval from the University of Edinburgh Animal Welfare and Ethical Review Body.

Cecal microbiota transplantation, colitis, and treatments with small molecular compounds
Recipient specific pathogen-free (SPF) WT C57BL/6 mice were pretreated with antibiotics for 2 weeks and rest for 1 day before receiving fresh cecal microbiota collected from SPF WT C57BL/6 mice that had been treated with vehicle, indomethacin (5 mg/kg per day in drinking water), or indomethacin plus EP4 agonist L-902,688 (10 g per mouse per day via daily intraperitoneal injection) for 5 days. After antibiotics treatment, some recipient mice were given normal drinking water for another 9 days before euthanasia for analysis of colon immune cells in the steady state, while other recipient mice were administered with normal drinking water for 7 days, followed by 2.5% (w/v) of DSS (molecular weight, 36 to 50 kDa; MP Biochemicals) in drinking water for six consecutive days to induce colonic inflammation. The DAI was scored by the following system: body weight, 0 (no or <1% weight loss compared to day 0 body weight), 1 (1 to 5% weight loss), 2 (5 to 10% weight loss), 3 (10 to 20% weight loss), and 4 (>20% weight loss); bleeding, 0 (no bleeding), 1 (blood present in/on feces), 2 (visible blood in rectum), and 4 (visible blood on fur); stool consistency, 0 (well-formed/normal stool), 1 (pasty/semiformed stool), 2 (pasty stool with some blood), 3 (diarrhea that does not adhere to anus), and 4 (diarrhea that does adhere to anus); general appearance, 0 (normal), 1 (piloerection only), 2 (piloerection and lethargy), and 4 (atoxic, motionless and sunken eyes). Mice were immediately culled when body weight loss was greater than 25% or the total colitis score is 12 or higher. Indomethacin (5 mg/kg per day or indicated doses) and vehicle (0.5% EtOH) were administrated through drinking water that was refreshed every 2 to 3 days. Butaprost (10 g per injection; Abcam), L-902,688 (10 g per injection; Cayman Chemical), and control (0.5% EtOH in PBS) were used by daily intraperitoneal injections. Antibiotics (containing ampicillin, gentamycin, metronidazole, neomycin, and vancomycin, each 0.5 mg/ml) and sucralose (4 mg/ml) were used in drinking water. DT (25 ng/g of bogy weight; Sigma-Aldrich) was injected intraperitoneally every 48 hours.

Histology
Intestine samples were fixed with 10% neutral buffered formalin solution (Sigma-Aldrich) and embedded in paraffin, and 5-m sections were used for staining with hematoxylin and eosin.

BMDC differentiation and stimulation
Bone marrow cells from femurs were cultured at 4 × 10 5 cells/ml of culture medium with granulocyte-macrophage colony-stimulating factor (GM-CSF) (20 ng/ml) in complete RPMI 1640 to induce the differentiation of BMDCs. RPMI 1640 with GM-CSF was refreshed every 3 days. On day 9, BMDCs were harvested by collecting the nonadherent cells after gently swirling the plate and restimulated for 6 hours with RPMI 1640 alone or cecal microbial products (CMPs) obtained from mice that received various treatments as indicated in related figure legends. CMPs were prepared by dissolving cecal contents with sterile PBS (100 mg/ml) with vigorous vertex followed by centrifugation at 10,000 rpm for 2 min and filtered through a 0.22-m Millex-GP filter unit. The CMP solution was then stored at −20°C and diluted by 40 times using PBS(−) for use in experiments. Cells were cultured at 37°C with 5% CO 2 .

Oxylipin analysis
Small intestine and colon samples were weighed and homogenized with ceramic beads in 1 ml of antioxidation buffer containing 100 M diethylenetriamine pentaacetic acid and 100 M butylated hydroxytoluene in PBS using a Bead Ruptor Elite for 2 × 30 s intervals at 6 m/s, under cooled nitrogen gas (4°C). Samples were spiked with 2.1 to 2.9 ng of PGE 2 -d 4 , PGD 2 -d 4 , PGF 2 -d 4 , and TXB 2 -d 4 standards (Cayman Chemical) before homogenization. Lipids were extracted by adding a 1.25-ml solvent mixture (1 M acetic acid/ isopropanol/hexane; 2:20:30, v/v/v) to 0.5-ml supernatants in a glass extraction vial and vortexed for 30 s. A total of 1.25 ml of hexane was added to samples, and after vortexing for 30 s, tubes were centrifuged (500g for 5 min at 4°C) to recover lipids in the upper hexane layer (aqueous phase), which was transferred to a clean tube. Aqueous samples were reextracted as above by addition of 1.25 ml of hexane, and upper layers were combined. Lipid extraction from the lower aqueous layer was then completed according to the Bligh and Dyer technique using sequential additions of methanol, chloroform, and water, and the lower layer was recovered following centrifugation as above and combined with the upper layers from the first stage of extraction. Solvent was dried under vacuum, and lipid extract was reconstituted in 200 l of high-performance liquid chromatography grade methanol. Lipids were separated by liquid chromatography using a gradient of 30 to 100% B over 20 min (A, water:Mob B 95:5 + 0.1% acetic acid; B, acetonitrile:methanol -80:15 + 0.1% acetic acid) on an Eclipse Plus C18 Column (Agilent) and analyzed on a Sciex QTRAP 6500 liquid chromatographytandem mass spectrometry system. Source conditions are as follows: temperature at 475°C, IS-4500, GS1 60, GS2 60, and CUR 35. Lipids were detecting using multiple reaction monitoring with the following parent to daughter ion transitions: PGD 1  . Peaks were only selected when their intensity exceeded three times above the baseline noise. The ratio of analyte peak areas to internal standard was taken, and lipids were quantified using a standard curve made up and run at the same time as the samples. Each oxylipin was then standardized per milligram of colon tissue.

SCFA profiling
The levels of SCFA (acetic acid, propionic acid, butyric acid, valeric acid, caproic acid, heptanoic acid, and caprylic acid) and branched SCFA (isobutyric acid and isovaleric acid) in cecal contents were detected as described previously (57). Briefly, the SCFA and branched SCFA (isobutyric acid and isovaleric acid) were extracted from acidified slurries three times in total using diethyl ether. Extracts were analyzed using gas chromatography (Agilent 7890A) with flame ionization detector, as described previously (cite paper above). Each of the SCFA was quantified against calibration curves plotted using authentic external standards [acetic acid (174.8 mM), propionic acid (133.4 mM), butyric acid (107.3 mM), valeric acid (89.2 mM), caproic acid (77.4 mM), heptanoic acid (69.8 mM), caprylic acid (58.5 mM), isobutyric acid (106.5 mM), and isovaleric acid (86.9 mM) all stored in 2 M NaOH and using 2-ethylbutyric acid (73.2 mM) as internal standard. Concentration of SCFA in cecal contents were calculated as micromoles per gram and normalized to the vehicle control group.

16S rRNA gene sequencing
Aliquots for sequencing of the 16S rRNA gene were first amplified with the V3-V4 region primers 341F (5′-CCTACGGGAGGCAG-CAG-3′) and 518R (5′-ATTACCGCGGCTGCTGG-3′) as described previously (64). A reagent-only control (DNA extraction kit blank) and a mock bacterial community (HM-782D, BEI Resources, American Type Culture Collection, Manassas, VA) were also prepared in the same manner. A single library pool was compiled using equimolar concentrations of DNA as measured using a fluorometric assay (Qubit dsDNA Broad-Range Assay Kit, Invitrogen, UK). The Illumina MiSeq platform (Illumina, CA) was used for sequencing (Edinburgh Genomics, UK), using V2 chemistry and producing 250-base pair paired-end reads. Using the mock bacterial community data, the sequencing error rate was calculated as 0.01%. The raw sequence reads, with primers removed, are publicly available via the National Center for Biotechnology Information Sequence Read Archive under accession number PRJNA564944.

16S rRNA gene sequencing data analysis
For the raw 16S RNA data in FASTQ format, amplicon primers were first removed to prevent false-positive detection of chimeras using the cutadapt plugin (65). Ten thousand sequences were sampled at random, and the qualities at each base position were examined for determining the parameters of the denoising process. The paired-end reads were further trimmed, filtered, denoised, and merged using the DADA2 plugin (66) through the Wales supercomputer portal. A naïve Bayes classifier within QIIME2 v2019.10 (67) was trained against the Silva v132 database (https://arb-silva.de/) on the amplified region. In addition, a machine learning Python library scikit-learn (68) was used to classify OTUs on the basis of 100% sequence identity. A total of seven taxonomic levels were used for 16S rRNA datasets. For measuring diversity, we generated de novo phylogenetic trees through multiple sequence alignment, masking, tree building, and rooting using the multiple alignment using fast Fourier transform (MAFFT) program (69). Percent abundance of taxa was determined by calculating it as a proportion of the total read count across all samples. For instance, the percent abundance of the ith taxa in the jth sample is computed as where n is the total number of samples and R represents read counts of each taxa.
The OTU table was rarefied across samples to the 90% of the lowest sample depth with a random seed set as 123 to eliminate the bias caused by the different sample sizes. For the overall bacterial community within samples, -diversity estimators including the observed species, the Chao1 index, the Shannon diversity, and the InvSimpson index were calculated using phyloseq (70). The -diversity estimates were compared between three groups using nonparametric Kruskal-Wallis test along with Dunn's multiple comparison correction. -Diversity between samples was computed using the unweighted UniFrac distances (71).

Statistical analysis
All data were expressed as means ± SD except that in Fig. 7A, where the data were expressed as means ± SEM as indicated in the figure legends. Statistical significance between two groups was examined by unpaired Student's t test, while the analysis of variance (ANOVA) with post hoc Holm-Sidak's multiple comparisons test was used to evaluate multiple groups. Statistical work was performed using Prism 8 software (GraphPad), and P < 0.05 was considered as significance.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/7/7/eabd7954/DC1 View/request a protocol for this paper from Bio-protocol.