Prostaglandin E2 suppresses the differentiation of retinoic acid–producing dendritic cells in mice and humans

To investigate the development of RA-producing DCs, we initially screened a panel of factors for their ability to induce the expression of the RALDH enzymes that convert retinal to RA. To measure RALDH activity, we used the commercially available substrate aldefluor (ALD), which is converted into a fluorescent product by RALDH enzymes, allowing the identification of RALDH-expressing cells by flow cytometry. Consistent with previous studies (Yokota et al., 2009; Guilliams et al., 2010), we found that although the spleen was devoid of ALD fluorescent cells, a subset of CD11c⁺ cells in both the mLNs (∼23%) and skin-draining pLNs (∼9%) converted the ALD substrate and fluoresced, indicating functional RALDH enzymes (Fig. 1 a). To confirm that ALD fluorescence correlated with RALDH activity and RA production, we sorted mesenteric CD11c⁺ DCs into ALD-positive and -negative subsets (Fig. 1 b) and found that only the ALD⁺ DCs expressed the RALDH2 transcript (Fig. 1 c). Second, although both subsets stimulated robust peptide-specific T cell proliferation in vitro (Fig. S1 a), the ability to drive CCR9 expression upon T cell priming was enriched and restricted to ALD⁺ DCs (Fig. 1 d). Thus, ALD fluorescence appears to mark the subset of DCs that express RALDH2 and have a unique ability to prime CCR9 expression, which is consistent with the synthesis of RA (for convenience herein we shall refer to the ALD^+/− cells as RALDH⁺ and RALDH⁻). Consistent with a previous study (Guilliams et al., 2010), the RALDH⁺ CD11c⁺ cells from the mLN were positive for CD103 and heterogenous for CD11b, whereas the RALDH⁺ DCs isolated from the skin-draining pLNs were CD103⁻CD11b⁺, suggesting that RALDH activity was common to differing DC subsets in vivo (Fig. S1 b).

To examine what factors can induce RALDH expression during DC development, we differentiated DCs from BM precursors with FLT3-L in the presence of a panel of immune factors. Although FLT3-L can drive the differentiation of CD11c⁺ DCs from BM precursors, FLT3-L alone does not induce RALDH expression (Fig. 1 e). Consistent with an earlier study (Yokota et al., 2009), the addition of anyone of GM-CSF, RA, IL-4, IL-13, and, to a lesser degree, thymic stromal lymphopoietin (TSLP) resulted in the emergence of RALDH⁺ DCs, suggesting that each of these factors can induce the expression of RALDH during DC development (Fig. 1 e). To examine whether GM-CSF would also induce RALDH activity in differentiated lymphoid DCs, we sorted RALDH⁻ CD11c⁺ cells from the mLN, skin draining pLN, and spleen and cultured them with GM-CSF for 48 h before measuring RALDH activity. As seen in Fig. 1 f, a large fraction of sorted RALDH⁻ DCs from all organs up-regulated the RALDH enzyme in response to GM-CSF, an activity which appears to be enhanced by LPS. Together, this data shows that multiple factors can induce RALDH expression in vitro and that DCs that do not express RALDH in vivo have the potential to do so in response to these positive signals.

Given that RALDH activity is restricted to a minor subset of DCs in vivo, it was surprising that such a large proportion of DCs could be induced to express RALDH in vitro by a wide panel of factors including GM-CSF. We therefore postulated that RALDH expression may be actively suppressed in vivo by negative regulators. The stroma is emerging as major source of molecules that modulate DC activity, and we wondered whether stromal-derived factors might also be involved in regulating DC RALDH expression. To this end, we differentiated DCs from BM precursors with GM-CSF in the presence of supernatants (SNs) derived from a panel of stromal or lymphocyte cell lines. Again, culturing BM with GM-CSF alone drives the differentiation of RALDH-expressing DCs, which account for ∼25% of CD11c⁺ cells (Fig. 2 a). However, when BM-derived DCs (BM-DCs) were differentiated with GM-CSF in the presence of SN taken from a primary skin stromal line or the 3T3 fibroblast line, RALDH⁺ DCs failed to develop (Fig. 2 a). This suppressive activity was restricted to SN purified from stromal cell lines, as SN from the thymoma EL-4 did not impair the development of RALDH-expressing DCs at any concentration (Fig. 2, a and b). Consistent with the loss of RALDH⁺ activity, CD11c⁺ DCs differentiated in the presence of skin SN expressed substantially less RALDH2 mRNA, suggesting that RALDH inhibition occurs at a transcriptional level (Fig. 2 c). Importantly, although CD11c⁺ DCs cultured with skin SN do not develop RALDH expression, the expression of MHC class II and CD86 by CD11c⁺ DCs was equivalent to that of control DCs (Fig. S2 a). Furthermore, skin SN–conditioned DCs produced equivalent or enhanced levels of IL-12 upon TLR stimulation, suggesting that the loss of RALDH activity was not a consequence of general dysfunction (Fig. S2 b). Consistent with this, when pulsed with the influenza NP_366–374 peptide, both control and skin SN–conditioned DCs could drive the proliferation of the CD8⁺ NP_366–374–specific F5 T cells in vitro, as measured by the sequential loss of CFSE fluorescence (Fig. 2 d). However, only control BM-DCs induced CCR9 expression on dividing T cells, which is consistent with their intact RALDH expression. Indeed, although skin SN–conditioned DCs had lost the ability to induce CCR9 expression, they had an enhanced ability to drive P-selectin ligand (PSL) expression, a lectin involved in tethering to nonintestinal endothelial (Fig. 2 d). These phenotypic differences correlated with distinct homing potential in vivo. When transferred into congenic recipients at equivalent frequencies, although both T cell populations were found at equal numbers in the spleen, T cells primed by control DCs homed more effectively to the small intestine epithelial, whereas T cells primed by skin SN–conditioned DCs had an enhanced capacity to traffic into inflamed skin (Fig. 2 e). In summary, this data shows that the stroma can produce soluble factors that suppress RALDH expression during DC differentiation, limiting the development DCs capable of priming gut-homing T cells.

We next sought to identify the soluble factor produced by skin stroma that inhibits DC RALDH expression. The ability to suppress RALDH expression was maintained in skin SN that had been heated or digested with trypsin, indicating that the active factor was unlikely to be a protein or a peptide (Fig. 3 a). We next tested whether a lipid may be responsible for RALDH inhibition. To this end, we grew skin stroma lines in the presence of lipid inhibitors to block the production of ceramides (fumonisin B1; He et al., 2006), prostaglandins (indomethacin; Nakata et al., 1981), or glycosphingolipids (N-butyldeoxynojirimycin [NBDJ]; Andersson et al., 2000), thus generating skin SN depleted of these various lipid species. As seen in Fig. 3 a, although skin SN depleted of ceramides or sphingolipid retained maximal suppressive activity, SN from indomethacin-treated skin stroma had a substantially reduced ability to suppress the development of RALDH-expressing DCs. Indomethacin inhibits the cyclooxygenase (COX) enzymes, which convert arachidonic acid to prostaglandin H, which is then converted into one of five active metabolites: PGD2, PGE2, PGFα, PGI, or thromboxane (Katler and Weissmann, 1977; Fig. 3 b). We asked whether these prostaglandins could rescue the suppressive activity of indomethacin-treated skin SN. As seen in Fig. 3 c, the addition of exogenous PGE2, but not of the other prostaglandins, restored the suppressive ability of indomethacin-treated skin stroma SN. Consistent with PGE2 as the active factor, endogenous PGE2 was detected in the skin SN (at ∼2 ng/ml or 5–10 nM), but not in SN, from indomethacin-treated skin stroma (Fig. 3 d). Intriguingly, exogenous PGE2 alone, in the absence of skin SN, had a reduced suppressive activity (Fig. S3), suggesting that the skin SN enhances the delivery or responsiveness to PGE2. Nonetheless, this data suggests that stromal-derived PGE2 inhibits the differentiation of RALDH-expressing DCs.

To further examine whether PGE2 signaling negatively regulated the expression of the RALDH enzymes, we asked whether mice deficient in PGE2 receptors had aberrant RALDH activity. PGE2 signals through four receptors: EP-1, -2, -3, or -4. These are G-protein–coupled receptors that have distinct downstream signaling events (Sugimoto et al., 2000; Sugimoto and Narumiya, 2007). We used genetically engineered mice that lack each individual EP receptor to examine whether a PGE2 receptor deficiency enhanced RALDH expression (Segi et al., 1998; Ushikubi et al., 1998; Hizaki et al., 1999). To this end, we differentiated DCs with GM-CSF from BM isolated from either WT or EP-KO mice in the presence or absence of skin SN before measuring RALDH activity. As seen in Fig. 4, compared with their WT 129 controls, BM from EP-4–deficient mice gave rise to a significantly enhanced frequency of RALDH⁺ DCs when cultured with GM-CSF alone (Fig. 4, a and b). Second, although the skin SN completely abolished the development of RALDH⁺ DCs from WT or EP-1, -2, or -3 KO BM, a population of RALDH⁺ DCs did emerge from EP-4 KO BM cultured with skin SN (Fig. 4, a and b). EP-4 deficiency did not alter other aspects of BM-DC development, in terms of DC frequency or phenotype, with equivalent levels of CD86, CD11b, or CD103 expression by EP-4 KO and WT BM-DCs (Fig. S4 a). Thus, the absence of EP-4 signaling appeared to enhance DC RALDH expression selectively. Consistent with PGE2 signaling via EP-4 to inhibit RALDH expression, the EP-4–specific agonist ONO-AE1-329 (Sugimoto and Narumiya, 2007; Jones et al., 2009), but not EP-1, -2, or -3 agonists, inhibited DC RALDH expression similar to PGE2 (Fig. 4 c). However, it is important to note that in vivo, EP-4 KO mice only showed a statistically significant increase in the frequency of RALDH⁺ DCs in the pLNs but not the mLN or spleen (Fig. S4). Given this modest phenotype in vivo, it is probable that some redundancy between other PGE2 receptors, notably EP-2, which, like EP-4, enhances cAMP (Sugimoto et al., 2000), or alternate factors may exist in vivo.

Given our findings in vitro, we wished to examine whether reducing PGE2 synthesis would increase the frequency of RA-producing DCs in vivo. Splenic DCs do not express RALDH and consequently fail to induce the expression of the gut-homing receptor CCR9 during T cell priming (Johansson-Lindbom et al., 2003; Iwata et al., 2004). Given that PGE2 is present within the spleen in the steady state (Fig. 5 a), we asked whether limiting PGE2 systemically—through the i.v. injection of indomethacin—would result in the emergence of RALDH-expressing DCs in the spleen that would in turn drive CCR9 expression during T cell activation. To this end, we transferred CFSE-labeled CD8⁺ F5 T cells into C57BL/6 recipients and, 1 d later, elicited priming within the spleen through the i.v. injection of the NP_366–374 peptide with LPS in combination with indomethacin treatment. 60 h later, we analyzed RALDH activity by splenic CD11c⁺ DCs and the expression of CCR9 on dividing F5 T cells. As seen in Fig. 5, indomethacin administration reduces splenic PGE2 levels (Fig. 5 a) and results in a four to fivefold increase in RALDH-expressing splenic DCs (Fig. 5 b). Critically, in comparison with untreated mice where dividing F5 T cells do not express CCR9, a large fraction (∼43%) of dividing T cells up-regulate CCR9 in indomethacin-treated mice, which is consistent with the enhanced RALDH expression by local splenic DCs (Fig. 5 c). Thus, inhibiting prostaglandin synthesis allows the emergence of RALDH-expressing DCs in nonintestinal sites such as the spleen, which in turn prime gut-homing T cells at this site.

We next wished to examine whether PGE2 similarly regulated RALDH expression in human DCs. To this end, we cultured CD14⁺ monocytes with GM-CSF and IL-4 in the presence or absence of prostaglandins and asked: (a) does GM-CSF/IL-4 induce RALDH expression in human mo-DCs; and (b) does PGE2 inhibit this induction? As seen in Fig. 6, although monocytes do not express RALDH2 mRNA, differentiation with GM-CSF and IL-4–induced RALDH expression by mo-DCs as measured by both ALD staining and RT-PCR (Fig. 6, a and b). Consistent with our previous experiments, PGE2, but none of the other prostaglandins, inhibited RALDH expression at a protein and mRNA level, suggesting that PGE2 suppressed RALDH transcription in human mo-DCs (Fig. 6, a and b). We next asked which PGE2 receptor was signaling RALDH inhibition in human mo-DCs. Using a panel of EP-specific antagonists, we found that AH6809, an antagonist for EP-1 and EP-2 (Jones et al., 2009), enhanced the frequency of RALDH-positive DCs in the absence of exogenous PGE2, suggesting that some level of endogenous PGE2 was limiting RALDH expression in culture (Fig. 6 c). Second, AH6809 blocked the complete suppression of RALDH⁺ DCs after the addition of exogenous PGE2. Given that the EP-1–specific antagonist SC1990 (Jones et al., 2009) had no effect, AH6809 is presumably acting by blocking EP-2 signaling. Surprisingly, the EP-4 antagonist AH23848 also had no effect on RALDH expression. We extended this analysis by using EP-specific agonists to ask which agonists could recapitulate PGE2 activity. It is of note that these experiments were performed in the presence of indomethacin to limit endogenous PGE2 production. As seen in Fig. 6 d, agonists for EP-1 or EP-3 showed negligible activity, whereas EP-4–specific agonists (AE1-329) partially limited the development of RALDH⁺ DCs (Sugimoto and Narumiya, 2007). Consistent with the antagonist data, the EP-2–specific agonist (AE1-259) was substantially more potent, completely inhibiting RALDH expression with an activity equivalent to PGE2 (Fig. 6 d). Thus, similar to what we observe in mice, PGE2 inhibits RALDH transcription during DC differentiation. However, in contrast to mouse BM-DCs, this activity appears to be mediated predominantly through PGE2–EP-2 signaling. Given this distinction, we compared the PGE2 receptor expression by mouse BM-DCs and human mo-DCs sorted into RALDH⁺ and RALDH⁻ subsets (Fig. 6, e and f). Consistent with their sensitivity to EP-2 signaling, human monocytes and both RALDH-negative and -positive mo-DCs express EP-2 in addition to EP-4 (Fig. 6 e). However, in the case of mouse BM-DCs, although both DC subsets expressed EP-4, EP-2 expression was restricted to RALDH-negative DCs (Fig. 6 f). Therefore, although human mo-DCs express both EP-2 and EP-4 and are highly sensitive to EP-2-PGE2 signaling, mouse RALDH⁺ BM-DCs have diminished EP-2 receptor expression and are consequently refractory to EP-2–specific agonists.

Our findings demonstrate that PGE2 inhibits RALDH transcription signaling via either EP-2 (in human mo-DCs) or EP-4 (for mouse BM-DCs). Both EP-2 and EP-4 are Gαs protein–coupled receptors that increase intracellular cAMP (Sugimoto et al., 2000). This suggests that the suppression of RALDH expression is likely to be downstream of an increase in intracellular cAMP. Consistent with this, the addition of the dibutyryl cAMP analogue (dB-cAMP) during mo-DC differentiation similarly suppressed the development of RALDH⁺ DCs in vitro (Fig. S5). How would increased cAMP inhibit RALDH expression? Elevated cAMP regulates gene expression via the phosphorylation of downstream nuclear transcription factors of the cAMP response element (CRE) binding protein (CREB)/CRE modulator (CREM)/activating transcription factor 1 (ATF-1) family (Borrelli et al., 1992; Mayr and Montminy, 2001). These proteins bind at conserved CRE binding sites (TGACGTCA), initiating the transcription of CRE-responsive genes (Borrelli et al., 1992; Mayr and Montminy, 2001). However, cAMP can also give rise to proteins that repress the transcription of CRE-responsive genes. Notably, cAMP can drive the expression of the ICER, a truncated CREM protein that lacks a trans-activation domain and thereby represses the transcription of genes whose promoters contain CRE-binding sites (Foulkes et al., 1991; Foulkes and Sassone-Corsi, 1992; Molina et al., 1993). Given that RALDH2 transcription is repressed by PGE2, and that the RALDH2 promoter contains multiple CRE-binding sites (Wang et al., 2001), we speculated that PGE2 might repress RALDH2 transcription through the induction of ICER. Consistent with this possibility, PGE2 induced ICER mRNA and protein expression in both CD11c⁺ mouse BM-DCs and human mo-DCs in vitro (Fig. 7, a–d). To test whether ICER regulates RALDH expression, we differentiated DCs with GM-CSF from BM of either WT or CREM/ICER-deficient mice and compared their RALDH activity. Consistent with ICER inhibiting RALDH transcription, CREM/ICER KO BM-DCs had significantly higher levels of RALDH activity compared with their WT counterparts (Fig. 7 e). Given that ICER inhibits DC RALDH expression in vitro, we next asked whether the subset of DCs that express RALDH in vivo do so in part by an ability to avoid ICER expression. To this end, we measured ICER expression directly ex vivo comparing RALDH⁺ and RALDH⁻ DC subsets sorted from mLNs and skin-draining pLNs (Fig. 7 f). Intriguingly, we found a reciprocal expression between ICER and RALDH2, with little or no ICER mRNA being detected in RALDH⁺ DCs, whereas ICER was expressed by RALDH⁻ DCs isolated from both mLN and skin-draining pLNs (Fig. 7 f). Furthermore, ICER expression coincided with EP-2, with both being expressed exclusively by RALDH⁻ DCs. This is consistent with our observation for BM-DCs, again indicating that RALDH⁺ DCs do not express EP-2. These patterns of expression are consistent with the possibility that those DCs that express the RALDH enzymes in vivo do so, in part, by limiting ICER expression, possibly by virtue of reduced PGE–EP-2 signaling.

The production of RA by DCs is critical for driving the development of gut-tropic immune responses (Iwata et al., 2004; Coombes et al., 2007; Jaensson et al., 2008). However, the factors that regulate RA synthesis by DCs remain poorly defined. Consistent with others (Yokota et al., 2009), we show that in reductionist in vitro settings, multiple factors, most notably GM-CSF, can induce RALDH expression in both mouse and human DCs. Indeed GM-CSF was able to induce RALDH activity even by RALDH-negative lymphoid DCs stimulated ex vivo. Despite this observation that multiple factors—many of which are widely available in vivo—can elicit RALDH expression in vitro, the capacity to produce RA is highly restricted to defined subsets of DCs in vivo. We therefore speculated that in vivo, negative regulators of RALDH might act to limit the systemic emergence of RA-producing DCs. Consistent with this possibility we discovered that PGE2 blocked RALDH expression induced by GM-CSF, repressing the transcription of the RALDH enzymes in both mouse and human DCs. Consistent with an inhibitory role, blocking PGE2 signaling via receptor antagonists or using BM precursors that were deficient in the PGE2 receptor subtype 4 (EP-4) gave rise to an enhanced frequency of RALDH⁺ DCs in vitro, whereas reducing PGE2 levels in vivo resulted in the emergence of RA-producing DCs in the spleen that were capable of priming CCR9⁺ gut-homing T cells at this site. These findings are consistent with an earlier observation in madine-darby canine kidney cells (Napoli, 1993) and suggest that PGE2 is a negative regulator of RALDH expression in DCs that effectively limits the emergence of RA-producing DCs.

PGE2 appears to inhibit RALDH expression at the transcriptional level. Of the four PGE2 receptors (EP-1–4), PGE2 signaled predominantly through EP-2 to inhibit RALDH transcription in human mo-DCs and EP-4 in mouse BM-DCs. Both EP-2 and EP-4 are G_αs protein–coupled receptors whose engagement results in adenyl cyclase release and an increase in intracellular cAMP (Sugimoto et al., 2000). Given this, in addition to the fact that the cAMP analogue similarly suppressed RALDH, it is likely that the inhibition of RALDH expression is downstream of cAMP. Elevated cAMP leads to the phosphorylation of nuclear transcription factors of the CREB/CREM/ATF-1 family (Mayr and Montminy, 2001). When phosphorylated, these can recruit cofactors that in complex bind at conserved CRE binding sites (TGACGTCA) and thereby initiate the transcription of CRE-responsive genes (Borrelli et al., 1992; Mayr and Montminy, 2001). However, CREB/CREM/ATF-1 will also drive the expression of the ICER that is transcribed from the intronic CRE-P2 promoter of CREM. ICER is a truncated protein that lacks the trans-activation domain of CREM but retains the DNA-binding motif and thereby acts as a repressor of CRE-responsive genes (Foulkes et al., 1991; Foulkes and Sassone-Corsi, 1992; Molina et al., 1993). Given that the RALDH2 promoter contains multiple CRE binding sites (Wang et al., 2001), the induction of ICER proteins has the potential to repress RALDH2 transcription. Consistent with this, our findings that BM-DCs from mice lacking ICER had significantly enhanced RALDH activity suggest that ICER actively limits RALDH expression during DC development. Given that PGE2 stimulates ICER expression in mouse and human DCs, we propose that PGE2 most likely represses the transcription of RALDH via the induction of ICER. Second, our finding that ICER was absent or expressed at substantially reduced levels by RALDH⁺ DCs in vivo suggests that RA-producing DCs express the RALDH enzyme, in part as a result of their unique ability to limit or avoid ICER expression. Collectively, these results provide a mechanistic insight into how PGE2 signaling would drive the transcriptional repression of RALDH.

Why those RA-producing DCs that emerge in vivo do not express ICER and how they escape PGE-mediated RALDH inhibition remains an outstanding question. One possibility is that these DCs may develop in microenvironments devoid of PGE2. However, PGE2 is present constitutively in the small intestine at levels similar to that found in the skin and spleen (Fig. S6). Furthermore, it is well established that DCs, including those found in the mLNs, produce PGE2 (Broere et al., 2009). For these reasons, we feel it likely that RALDH⁺ DCs are exposed to PGE2 in vivo. Alternatively, PGE2 signaling may fail to induce ICER expression in this DC subset, thus leaving the transcription of RALDH unhindered. A third possibility is that RALDH⁺ DCs are themselves refractory to PGE2. Although this issue warrants further studies, it is intriguing that mouse RALDH⁺ DCs that emerge both in vitro and in vivo do not express the EP-2 receptor, whereas their RALDH-negative counterparts do. Although EP-4 signaling does drive RALDH inhibition in mouse BM-DCs, in the case of human monocytes (which express both EP-2 and -4), PGE2–EP-2 signaling was clearly the dominant pathway for RALDH inhibition. Blocking signaling through EP-2, but not EP-4, abrogated PGE2-mediated RALDH inhibition, whereas EP-2 agonists were substantially more potent in comparison with their EP-4 counterparts on human mo-DCs. The enhanced activity of EP-2 may relate to PGE2–EP-2 signaling, providing a more sustained signal compared with EP-4, as the latter is internalized after stimulation, whereas EP-2 does not undergo agonist induced desensitization (Nishigaki et al., 1996; Desai et al., 2000). Given that EP-2 is the dominant receptor for signaling RALDH inhibition in human mo-DCs, coupled with the finding that RALDH⁺ DCs are devoid of EP-2, it is tempting to speculate that RALDH-expressing DCs arise from a unique precursor that does not express EP-2 and are thus refractory to PGE2. Although this is possible, it is important to note that mice lacking the EP-2 receptor did not show an increase in RALDH⁺ DCs in vivo. Thus, the absence of EP-2 does not alone explain the development of RALDH⁺ DCs, potentially as the result of a level of redundancy between EP-2 and -4 or additional inhibitory factors that also drive the phosphorylation of CREB proteins. Consequently, the exact mechanism that enables RA-producing DCs to escape PGE2-mediated RALDH inhibition remains unclear and is an area of ongoing research.

Early inflammatory molecules, such as IL-1 and TNF, induce the expression of COX-2 and thereby increase local PGE2 production (Alaaeddine et al., 1999; Dinarello, 2002). As such, DCs differentiating at the site of inflammation, including inflammatory monocytes which infiltrate inflamed tissues, are likely to be exposed to high concentrations of PGE2. With this in mind, it is interesting that inflammatory mo-DCs infiltrating the GALT during periods of colitis do not acquire RALDH activity (Siddiqui et al., 2010). Instead, these inflammatory mo-DCs produce high levels of IL-23, a cytokine which is enhanced by PGE2 (Khayrullina et al., 2008), and appear to augment Th-17 responses that in turn exacerbate intestinal pathology (Siddiqui et al., 2010). Our findings, which establish a link between PGE2 and RALDH expression, may help to explain why inflammatory mo-DCs fail to acquire the capacity to produce RA. Given that RA-producing DCs potently inhibit Th-17 differentiation and instead favor the differentiation of FOXP-3 regulatory T cells (Mucida et al., 2007), defining the processes which underpin the loss of RA-producing DCs during periods of chronic inflammation may have important implications in understanding the onset immunopathology.

Our results clearly demonstrate that PGE2 treatment modulates how DCs program homing receptor expression during T cell priming. Reducing PGE2 levels in vivo enhanced the ability of DCs to drive CCR9 expression upon T cell activation, whereas DCs differentiated in the presence of exogenous PGE2 lost the capacity to induce CCR9, instead showing an enhanced ability to elicit the expression of nonintestinal homing receptors, such as PSL, upon activated T cells. Given that PGE2 is commonly used in clinical trials to mature mo-DCs used for immunotherapy (Jonuleit et al., 1997; Schadendorf et al., 2006), our findings have immediate implications within this setting. PGE2 treatment would reduce RALDH activity and thereby enhance the capacity of DCs to activate T cells capable of migrating to nonintestinal sites, such as the skin, and thus would be highly appropriate when eliciting T cells response to cancers such melanoma. In contrast, blocking PGE2 signaling would enhance RALDH expression and, thus, the ability of DCs to prime CCR9⁺ gut-homing T cells that would be most effective against intestinal cancers. Second, our findings that indomethacin treatment promoted the systemic priming of CCR9⁺ T cells raises the possibility that the administration of COX inhibitors in combination with vaccines may be a means to elicit gut-tropic immune responses after parenteral inoculations. Given that most vaccines are delivered via subcutaneous or intramuscular injections, routes which do not typically elicit gut-homing T cells, this approach maybe beneficial when vaccinating against pathogens, such as HIV, that replicate within the intestinal tract. Thus, we suggest that manipulating DCs’ exposure to PGE2 may be a means to enhance the efficacy of T cell immunity via controlling the tissue specificity of such responses.

In summary, we show in this paper that although multiple factors can elicit the expression of RALDH during DC differentiation, PGE2 provides a dominant inhibitory signal that represses the transcription of RALDH. The development of RA-producing DCs may therefore reflect a balance between the ability to respond to positive stimuli, such as GM-CSF which drives RALDH expression, while ignoring or subverting the inhibitory stimuli such as PGE2 that elicit transcriptional repression. Given the central role that RA-producing DCs play in the induction of gut-specific immunity, we believe that modulating PGE2 signaling during DC differentiation may therefore be a powerful means to manipulate the tissue tropism of immune responses.

C57BL/6, F5, and B6.SJL-Ptprc^aPep3^b/BoyJ (B6Ly5.1) mice were maintained at the Biological Services Unit at John Radcliffe Hospital (University of Oxford). F5 TCR transgenic mice express a TCR specific for the H2-D^b-restricted A/NT/60/68 influenza nucleoprotein peptide NP68_366–374-ASNENMDAM (Mamalaki et al., 1993). Mice with deletions in one of Ptger1 (EP-1), Ptger2 (EP-2), Ptger3 (EP-3), or Ptger4 (EP-4) were generated as previously described (Segi et al., 1998; Ushikubi et al., 1998; Hizaki et al., 1999). With the exception of EP-4^−/−, these mice were backcrossed at least eight times onto the C57BL/6 background. Ptger4^−/− mice do not survive on the C57BL/6 background as a result of patent ductus arteriosus (Segi et al., 1998) and were backcrossed on a mixed background of 129/Ola × C57BL/6. Mice were provided by S. Narumiya (Kyoto University, Kyoto, Japan), and breeding colonies were maintained by M. Belvisi at Imperial College London (London, England, UK). CREM/ICER-deficient mice were generated as previously described (Blendy et al., 1996; Conti et al., 2004). Mutant and WTs are F1 hybrids (129SVEV:C57BL/6) obtained by crossing heterozygous CREM/ICER +/− 129SVEV N12 × CREM/ICER+/− C57BL/6 N15. Mice were provided by J. Blendy (University of Pennsylvania School of Medicine, Philadelphia, PA) and colonies were maintained at the Department of Pharmacology (University of Pennsylvania School of Medicine). All mice were used under institutional guidelines with the authority of a UK Home Office project license.

Recombinant mouse G-CSF, M-CSF, TGF-β, IL-4, IL-6, IL-7, IL-10, IL-12, and IL-13 and human FLT3-L and GM-CSF were purchased from PeproTech. Mouse TSLP and the PGE2 ELISA kit were obtained from R&D Systems and CPG₁₈₂₆ from InvivoGen. Ciglitazone, all-trans RA, Escherichia coli LPS, trypsin-coated agarose beads, SC19220, and AH23848 were obtained from Sigma-Aldrich. PGD, PGF, PGI, thromboxane, and AH6809 were obtained from Cayman Chemical and PGE2 was from the John Radcliffe Hospital Pharmacy (Oxford, London, UK). The EP-specific agonists ONO-D1-004 (EP-1), ONO-AE1-259 (EP-2), ONO-AE-248 (EP-3), and ONO-AE1-329 (EP-4) were a gift from the ONO Pharmaceutical Company. Mouse GM-CSF and human IL-4 were produced in house. IL-12p40 ELISA was performed with capture (C15.6) and detection (C17.8) antibodies purchased from eBioscience.

Primary skin stromal lines were generated from adult C57BL/6 mice by digesting full-thickness skin in 1 mg/ml of type II collagenase (Worthington Biochemical Corporation) and 10 µg/ml DNase (Sigma-Aldrich) and grown in complete DME (DME supplemented with 1 mM sodium pyruvate, 25 mM Hepes, MEM-NEA, 50 µM 2-mercaptoethanol, l-glutamine, and antibiotics) with 10% FCS (DME-10) until confluent adherent monolayers developed. 3T3 fibroblasts were maintained in DME-10 and the EL-4 thymoma in complete RPMI with 10% FCS (RP-10). To generate cell line SNs, cells were grown until 50–70% confluent and media was replaced with fresh RP-10. Cells were cultured for a further 2 d before 0.22 µM SN was harvested and filtered for use. In some instances, cells were grown for the final 2 d in the presence of 1 µM fumonisin B1 (Sigma-Aldrich), 1 µM NBDJ (Sigma-Aldrich), or 1 µM indomethacin (Sigma-Aldrich), or SN was subsequently heated (80°C for 20 min) and digested with 5–10 U/ml trypsin-agarose beads for 2 h at 37°C (Sigma-Aldrich). The following formulae was used to compare the relative inhibitory activity of treated versus untreated SN: [(without SN^a – treated SN^a)/(without SN^a – untreated SN^a)] × 100 = % relative inhibition by treated SN, where a = % ALD⁺ of CD11c⁺ cells.

Antibodies for flow cytometry were as follows: anti-CD11c APC (N418), anti-TCR APC (H57-597), anti-CD3 eFluor 450 (17A2), anti-CD45.2 APC (104), anti-CD103 biotin (M290), anti–CD11b-APC (M1/70), anti-CD8α APC (53.67), anti-Ia/Ie FITC (M5/114), anti-CD86 PE (GL-1), rat-IgG2a PE isotype control, and anti–human Fc PE (eBioscience). Anti-CCR9 PE (242503) and the P-selectin/human Fc chimeric protein were obtained from R&D Systems. Anti–human CD86 APC (2331) was obtained from BD. ALD staining (STEMCELL Technologies) was used as per the manufacturer’s instructions. 10 µg/ml propidium iodide (Sigma-Aldrich) was added immediately before analysis on FACSCalibur or Cyan (BD) flow cytometers. Data were analyzed with FlowJo software (Tree Star).

CD11c⁺ DCs were isolated from tissues essentially as previously described (Vremec and Shortman, 1997). In brief, organs were digested in 1 mg/ml of type II collagenase (Worthington Biochemical Corporation) with 10 µg/ml DNase (Sigma-Aldrich) for 20 min at room temperature before the addition of 10 µM EDTA for a final 5 min. In some instances, DCs were subsequently enriched with anti-CD11c microbeads (Miltenyi Biotec) as per the manufacturer’s instructions.

BM was isolated from the femur and tibia of mice and cultured in 6-well plates (10⁶ cells/well). For FLT3-L DCs, cells were cultured with 100 ng/ml of human recombinant FLT3-L in RP-10 with GM-CSF, G-CSF, M-CSF, TSLP, TGF-β, IL-4, IL-6, IL-7, IL-10, IL-12, IL-13, or PGE2 at 20 ng/ml, or RA and 100 µM Ciglitizone. After 3 d, 0.25 ng/ml LPS was added and cells were harvested the next day. For GM-CSF–cultured BM-DCs, cells were cultured with 20 ng/ml GM-CSF in RP10 alone or with SN at 75% of culture volume (unless otherwise stated). After 3 d, 0.25 ng/ml LPS was added and cells were harvested the next day. For in vitro proliferation assays, DCs were subsequently enriched with anti-CD11c or anti–MHC class II microbeads (Miltenyi Biotec). Human mo-DCs were generated essentially as previously described (Sallusto and Lanzavecchia, 1994). In brief, CD14⁺ monocytes were purified from buffy coats with CD14⁺ microbeads (Miltenyi Biotec) and cultured in 6-well plates (10⁶ cells/well) in RP-10 with 50 ng/ml of human recombinant GM-CSF (PeproTech) and IL-4 (5%). After 3 d, 1 ng/ml LPS was added and cells were harvested the next day. In some instances, 1 µM indomethacin was added at the beginning of culture to block the endogenous prostaglandin.

RNA was extracted with RNA easy kit (QIAGEN) as per manufacturer’s instructions. cDNA was synthesized from 20 ng RNA with oligo-dT primers (Invitrogen) and Superscript Reverse transcription (Invitrogen). Primers and conditions are listed in Table S1. ICER sense primer corresponds to the 5′ untranslated region of ICER and the anti-sense to bp 496–516 of CREM (Tetradis et al., 1998).

For quantitative (q) PCR, GAPDH and RALDH2 mRNA levels were determined with primers/probe (Roche; Table S1) using TaqMan Gene Expression assays (Applied Biosystems) with TaqMan Gene Expression Master Mix (Applied Biosystems), according to the manufacturer’s instructions. Changes in gene expression relative to the endogenous control GAPDH were calculated using the formula 2-^ΔCT. Fold changes in expression relative to control DCs were calculated according to the 2-^ΔΔCT method.

CD8⁺ T cells were enriched from splenocytes of F5 mice using a CD8⁺ negative isolation kit (Dynal; Invitrogen) and labeled with 1 µM CFSE for 10 min at 37°C. DCs were purified with anti-CD11c or MHC class II microbeads beads (Miltenyi Biotec) according to the manufacturer’s instructions, and pulsed with 0.1 µg/ml NP_366–374-peptide for 45 min at 37°C. 25 × 10³ DCs were plated into 96-well flat-bottom plates with 75 × 10³ CFSE-labeled CD8⁺ F5 T cells. After 4 d, cells were stained with anti-TCR and anti-CCR9 or -PSL antibodies with appropriate secondary antibodies and analyzed by flow cytometry.

F5 T cells activated in vitro by NP_366–374 peptide-pulsed control or SN-conditioned DCs were labeled with 0.1 µM CFSE or 20 µM CTO, pooled at a 1:1 ratio, and ∼2 × 10⁶ total donor cells were transferred i.v. into B6.Ly5.1 mice that had been sensitized with oxazolone and challenged on the ear as previously described (Johnson et al., 2006). After 16 h, recipient mice were sacrificed, perfused, and single cell suspension made from the spleen, ear, and intestine as previously described (Weigmann et al., 2007). Cells were stained with anti–mouse CD45.2 for analysis by flow cytometry.

CD8⁺ T cells were enriched from splenocytes of F5 mice, labeled with 1 µM CFSE, and 5 × 10⁵ cells were transferred into C57BL/6 recipients by i.v. tail vein injections. 1 d later, mice were injected i.v. with 50 µg NP_366–374–peptide with 10 µg LPS. After 60 h, spleens were harvested, digested with collagenase, and stained with ALD and CD11c, as previously described, or stained with anti-CD8 and anti–CCR9 antibodies for analysis by flow cytometry.

Cell pellets were lysed in 1% Igepal, 0.5% sodium deoxycholate, and 0.1% SDS with inhibitors (1 mM phenylmethylsulfonyl fluoride, 50 mM sodium fluoride, 2 mM sodium orthovanadate, and complete protease inhibitor [Roche]). Lysate was separated on 15% polyacrylamide gel via SDS-PAGE, transferred to a PVDC membrane (Hybond; GE Healthcare), and blocked with PBS, 5% BSA, and 0.1% Tween-20. Anti–CREM-1 antibody (X-12; 1:500; Santa Cruz Biotechnology, Inc.) was added in blocking buffer overnight at 4°C and detected with horseradish peroxidase–conjugated secondary antibody before developing using Super Signal chemiluminescent reagent (Thermo Fisher Scientific).

Data were analyzed using a two-tailed Student’s t test.

Fig. S1 shows that both ALD-positive and -negative subsets stimulated robust peptide-specific T cell proliferation in vitro and that RALDH activity is common to differing DC subsets in vivo. Fig. S2 shows that CD11c⁺ DCs have equivalent expression of MHC class II and CD86 to control DCs and that skin SN-conditioned DCs produced equivalent or enhanced levels of IL-12 upon TLR stimulation. Fig. S3 shows that exogenous PGE2 alone, in the absence of skin SN, has reduced suppressive activity. Fig. S4 shows that EP-4 deficiency did not change the frequency or phenotype of DCs during BM-DC development and appeared to enhance DC RALDH expression selectively. In vivo, EP-4 KO mice showed a statistically significant increase in the frequency of RALDH⁺ DCs in the pLNs, but not in mLNs or the spleen. Fig. S5 shows that, when added during mo-DC differentiation, the cAMP analogue dibutyryl-cAMP suppressed the development of RALDH⁺ DCs in vivo. Fig. S6 shows that PGE2 is present constitutively in the small intestine at levels similar to those found in the skin and the spleen. Table S1 lists the primers and conditions used in PCR.

We would like to thank ONO Pharmaceuticals Co. (Osaka Japan) for the provision of the EP-specific agonists ONO-D1-004, ONO-AE1-259, ONO-AE-248, and ONO-AE1-329. We are also very grateful to Mark Birrell and Maria Belvisi (Imperial College, London, UK) for generously providing EP-KO mice and Julie Blendy (University of Pennsylvania School of Medicine) for generously providing the CREM/ICER-null mice. We would also like to thank Ellen Pure and Irene Crichton (Wistar Institute, Pennsylvania) for their support, Craig Waugh for flow cytometric sorting, and Giorgio Napolitani and Carmen De Santo for critical reading and discussion.

A. Stock was supported by a CJ Martin Fellowship from the NHMRC (Australia). This work was supported by Cancer Research UK (C399/A2291) and the UK Medical Research Council.

The authors have no competing financial interests to declare.