Research resource: transcriptome profiling of genes regulated by RXR and its permissive and nonpermissive partners in differentiating monocyte-derived dendritic cells.

Retinoid X receptors (RXRs) are heterodimerization partners for many nuclear receptors and also act as homodimers. Heterodimers formed by RXR and a nonpermissive partner, e.g. retinoic acid receptor (RAR) and vitamin D receptor (VDR), can be activated only by the agonist of the partner receptor. In contrast, heterodimers that contain permissive partners, e.g. liver X receptor (LXR) and peroxisome proliferator-activated receptor (PPAR), can be activated by agonists for either the partner receptor or RXR, raising the possibility of pleiotropic RXR signaling. However, it is not known to what extent the receptor's activation results in triggering mechanisms dependent or independent of permissive heterodimers. In this study, we systematically and quantitatively characterized all probable RXR-signaling pathways in differentiating human monocyte-derived dendritic cells (Mo-DCs). Using pharmacological, microarray and quantitative RT-PCR techniques, we identified and characterized gene sets regulated by RXR agonists (LG100268 and 9-cis retinoic acid) and agonists for LXRs, PPARs, RARα, and VDR. Our results demonstrated that permissiveness was partially impaired in Mo-DCs, because a large number of genes regulated by PPAR or LXR agonists was not affected by RXR-specific agonists or was regulated to a lesser extent. As expected, we found that RXR agonists regulated only small portions of RARα or VDR targets. Importantly, we could identify and characterize PPAR- and LXR-independent pathways in Mo-DCs most likely mediated by RXR homodimers. These data suggested that RXR signaling in Mo-DCs was mediated via multiple permissive heterodimers and also by mechanism(s) independent of permissive heterodimers, and it was controlled in a cell-type and gene-specific manner.

The retinoid X receptor (RXR) is an enigmatic member of the nuclear receptor superfamily and is a true manifestation of the paradigm of reverse endocrinology. First, the receptor was cloned (1), and later, 9-cis retinoic acid (9cisRA) was identified as its presumed endogenous ligand (2). Additional potential natural agonists (docosahexaenoic acid and phytanic acid) (3,4) have also been described, and several RXR-specific agonists, termed rexinoids (5,6,7,8,9), have been synthesized, allowing pharmacological mapping of RXR signaling. The three mammalian RXR isotypes, RXRα, RXRβ, and RXRγ (10,11,12,13,14), encoded by distinct genes exhibit different tissue distribution (reviewed in Ref. 15). All isotypes can form homodimers and heterodimers with a large number of partners, including vitamin D receptor (VDR), retinoic acid receptor (RAR), thyroid hormone receptor (TR), liver X receptor (LXR), peroxisome proliferator-activated receptor (PPAR), farnesoid X receptor (FXR), constitutive active/androstane receptor (CAR), pregnane X receptor (PXR), and two members of nuclear receptor subfamily 4, group A (NR4A), Nur-related protein 1 (Nurr1) and growth factor-inducible immediate early gene nur/77-like receptor (Nur77) (reviewed in Refs. 15,16,17,18). Apart from the members of NR4A, which do not have an apparent and distinct hydrophobic pocket for ligand binding, RXR partners are activated by lipid-soluble molecules and control diverse biological processes via transcriptional regulation (16,17,18). Ligand-activated RXR partners carry out transcriptional regulation via various mechanisms (17,18,19), and the requirement of RXR seems to be different for gene activation and repression. In the case of transactivation, RXR serves as an obligate partner and is required for high affinity binding of most RXR partners to their cognate hormone responsive element. Mechanisms whereby agonists for RXR partners, especially LXRs and PPARs, inhibit gene expression are not completely understood (reviewed in Ref. 19), and in most cases, RXR as a partner is not required.

RXR partners can be classified into functionally distinct permissive and nonpermissive groups (15). RXR heterodimers that contain permissive partners, including PPAR, LXR, and FXR, can be activated by RXR agonists even in the absence of the ligand of the partner. Importantly, when both partners are activated, they act in an additive or synergistic manner. In contrast, heterodimers formed by RXR and a nonpermissive partner (e.g. RAR, TR, and VDR) cannot be activated by an RXR agonist but only the agonist of the partner receptor (15). This phenomenon is referred as “subordination” and is due to the fact that nonpermissive partners inhibit RXR activation (20,21,22,23). Remarkably, in the presence of its ligands, RAR “becomes” permissive, and RAR and RXR agonists together have greater effect than the RAR agonist alone, therefore RAR-RXR is called a conditional (permissive) heterodimer. In cotransfection assays, RXR agonists do not exhibit an additive effect together with agonists for VDR and TR. The differences between permissive and nonpermissive heterodimers were demonstrated by in vitro differentiation studies as well. For example, 3T3-L1 preadipocytes and NB4 myeloid cells were differentiated in the presence of PPARγ agonist rosiglitazone (RSG), and RAR agonist TTNPB (Arotinoid acid, 4-[(E)-2-(5,6,7,8-Tetrahydro-5,5,8,8-tetramethyl-2-naphthalenyl)-1-propenyl] benzoic acid), respectively (24,25). The RXR-specific agonist LG100268 (LG268) has the capacity to promote the differentiation of 3T3-L1 preadipocytes to mature adipocytes, whereas it cannot promote the differentiation of NB4 cells (24,25).

In our previous works, we systematically analyzed the contribution of PPARγ, RAR, VDR, and LXR to the gene expression and immunophenotype of primary human dendritic cells (DCs) (26,27,28,29,30). DCs are initiators and regulators of adaptive immunity as professional antigen presenting cells, and they are also involved in innate immunity (31,32). Functions of DCs are affected by exogenous signals (e.g. microbial products and cytokines, respectively), which determine their fate and lead to subtype specification. Nuclear hormone receptors have been implicated in these processes (reviewed in Ref. 33). Similarly to its partners, activation of RXR has also a profound effect on DC differentiation, apoptosis, immunogenicity, and T-cell activation capacity of DCs (34,35,36). To identify gene sets, pathways, and functions affected by PPARγ, RAR, VDR, and LXR, we used mainly monocyte-derived DCs (Mo-DCs). Mo-DCs, a well-characterized in vitro model of DCs, are obtained by culturing CD14+ human primary monocytes in the presence of granulocyte-macrophage colony-stimulating factor (GM-CSF) and IL-4 (37).

In this study we used this cell type and performed global gene expression analyses using Affymetrix DNA microarrays followed by real-time quantitative RT-PCR (RT-qPCR) validation to systematically characterize and compare the gene expression changes resulted by liganding of RXR or its partners. Mo-DCs offer several advantages for such global analysis. First, Mo-DCs express high level of RXRα, and ligand activation of RXR results in phenotypic and functional changes (34,35,36). Second, at least four ligand-inducible permissive RXR partners (PPARγ, PPARδ and LXRα, LXRβ) are present and active in Mo-DCs, allowing the investigation of pleiotropic effects of RXR agonists. The expression profile of all other potential RXR partners could be determined by using a microarray-based approach. Last, human Mo-DCs are a homogenous primary nondividing cell population, an ideal subject of transcriptome analyses.

In this study, we demonstrate that RXR agonists cannot replicate the effects of agonists for permissive partners, because permissiveness is partially impaired in Mo-DCs. Our analyses also showed that the changes initiated by RXR agonists go beyond partial activation of permissive partners, and there are unique pathway(s) initiated.

Results

RXR agonists regulate DC phenotype and activate various pathways in Mo-DCs

First, we determined which RXR isotypes are expressed in Mo-DCs and other DC types. Using Affymetrix microarray data of human Langerhans cells, dermal DCs, blood CD1c+ DCs along with Mo-DCs (28,38,39), we found that RXRα is the dominant isotype in all tested DC types (Fig. 1, A and B). RXRα is expressed at high level in Mo-DCs compared with other nuclear receptors and belonged to the top 4% intensities of all probe sets (Fig. 1A). RXRα is also expressed at the protein level, and it is localized in the nuclei of Mo-DCs as determined by immunohistochemistry (Fig. 1C).

Figure 1

RXRα is highly expressed and transcriptionally active in Mo-DC. A, The expression of genes coding RXRα, RXRβ, and RXRγ in human Mo-DCs was compared with other nuclear receptors (NRs) and to all probe sets of Affymetrix microarrays. Normalized signal intensities of probe sets specific for RXR isotypes are shown. If more than one probe set represent a certain gene, the probe set having the highest signal intensity is shown. B, RXRα is the dominant isotype in all tested DC types as determined by microarray. C, The RXRα protein (red fluorescence) is localized in the nuclei [visualized by 4′,6-diamidino-2-phenylindole (DAPI)] of Mo-DCs identified as DCSign positive cells (green cell membrane fluorescence). Original magnification, ×40. D, Cell surface expression of the indicated DC markers are regulated by RXR agonists, 9cisRA, and rexinoid LG268 determined by flow cytometry after 5 d of ligand treatment. E, Induction of direct target genes of LXR, PPAR, and RAR by agonists for RXR (9cis RA and LG268), LXRα/β (GW3965), and PPARγ (RSG) was detected by RT-qPCR. Monocytes were conditioned in the presence of IL-4 and GM-CSF, agonists were added 18 h after plating. Values are expressed as mean of technical triplicates ± sd of the mean. ABCG1, ATP-binding cassette, subfamily G, member 1.

Next, we tested the capacity of a natural RXR ligand, 9cisRA, and a prototypic rexinoid, LG268, in the regulation of differentiation markers of Mo-DCs. Monocytes were isolated from buffy coats of healthy blood donors and differentiated in the presence of IL-4 and GM-CSF for 5 d in the presence of RXR agonists. Consistent with a previous report (36), we found that CD1a and CD80 are down-modulated, whereas CD86 and CD1d are up-regulated in the presence of RXR agonists as assessed by flow cytometry (Fig. 1D). We also investigated the capacity of RXR agonists on the transcriptional regulation of pyruvate dehydrogenase kinase, isozyme 4 (PDK4), ATP-binding cassette, subfamily G, member 1 (ABCG1), and transglutaminase 2 (TGM2) by RT-qPCR. PDK4 and ABCG1 are direct target genes of PPARs and LXRs, respectively (40,41,42), whereas TGM2 is regulated in mice directly by both RAR-RXR heterodimer and RXR homodimers (43,44). Due to the fact that PPARγ and LXRα expression levels are very low in monocytes (26,30), we treated Mo-DCs 18 h after plating with vehicle, RXR agonists, or agonist for PPARγ (RSG) and LXRα/β (GW3965) for 3, 6, 12, or 24 h. We found that both 9cisRA and LG268 have the capacity to induce the expression of these genes as early as 3 h (Fig. 1E). These data suggested that many RXR-dimers could be activated by RXR agonists in Mo-DCs.

Expression profiles of RXR partners in Mo-DC

To determine the full spectrum of RXR heterodimeric partners (15), we evaluated the expression pattern of all potential RXR partners during DC differentiation (Fig. 2). Our Affymetrix DNA microarray data, consistent with our previous reports (26,30), demonstrated that LXRα, PPARγ, and PPARδ are highly expressed in Mo-DCs, and their expression levels are increased during monocyte to DC differentiation. In contrast, the mRNA level of LXRβ is decreased. The nonpermissive partners VDR and RARα are also expressed at high levels. Based on our microarray data, PPARα, PXR, CAR, FXR, RARβ, TRβ, RARγ, and TRα are not expressed (absent flags for all probe sets at all time points) or expressed at very low level. Nur77 and Nurr1, two members of NR4A subfamily, are expressed in monocytes at high levels, but during the first day of differentiation, their expression levels are decreased dramatically. Obviously, mRNA levels are not always correlated with protein levels. It is possible that mRNA is not translated into functional protein and vice versa, proteins with long half-life may be present despite the reduced level of mRNA. The mRNA levels of Nurr1 and Nur77 are high in monocytes, so it is possible that proteins are still present despite the fact that mRNA levels are down-regulated. However, mRNAs of xenobiotic receptors are not detectable at any time points during the differentiation program, suggesting that these proteins are not present in Mo-DCs.

Figure 2

Expression patterns of genes coding RXRs and RXR partners during monocyte to DC differentiation. Human monocytes were isolated and cultured in the presence of IL-4 and GM-CSF for 24 h and for 5 d. RNA samples were isolated from monocytes (0.0) from differentiating DCs (24.0) and immature DCs (120.0). Microarray experiment was performed, and expression patterns of RXR partners were evaluated. Per chip normalized signal intensities of probe sets specific for indicated nuclear receptors are shown. If more than 1 probe set represent a certain gene, the probe set having the highest signal intensity is shown. Raw signal intensities were normalized to the median of all transcripts (per chip normalization). RXR partners were considered to be expressed on median or high level if normalized expression levels were above 10 (dashed line). The genes that were given low significant attribute (absent flag) for all time points were marked as absent (A). CAR, Constitutive active/androstane receptor.

Considering all these pieces of information, we concluded that LG268 and 9cisRA could activate the following dimers in Mo-DCs: RXR homodimer, PPARγ-RXR, PPARδ-RXR, LXRα-RXR and LXRβ-RXR, and might act on Nurr1-RXR and Nurr77-RXR. 9cisRA can bind to both units of RARα-RXR, whereas LG268 very likely activates this heterodimer only in the presence of endogenous RAR agonist(s).

Gene sets regulated by LG268 and 9cisRA and agonists for LXRα/β, PPARγ, and PPARδ only partially overlap

We used Affymetrix HG133 Plus 2.0 microarrays for monitoring transcriptional changes globally. We aimed at using ligands long enough to detect either up- or down-regulation but for a relatively short period of time to reduce the secondary effects. Therefore, we treated the differentiating cells with agonists for 12 h (Fig. 3A). This experimental design allowed us to determine the transcriptional changes when PPARs and LXRs are expressed at high level. For activation of RXR and its partners, we used the following agonists: LG268 (RXR agonist) and 9cisRA (RAR and RXR agonist), GW3965 (LXRα,β panagonist) (45), RSG (PPARγ), and GW1516 (PPARδ) (Fig. 3A). We processed RNA obtained from three healthy donors and hybridized them to Affymetrix microarrays. The microarray data were imported to GeneSpring software, and genes regulated at least 2-fold up or down were identified as described in Materials and Methods. We identified a gene set, which was regulated by both 9cisRA and LG268 (Fig. 3B). These genes are very likely bona fide RXR-regulated targets (referred as “LG268 and 9cisRA” in Fig. 3). This gene set contains 339 probe sets, representing 226 annotated genes, and is shown on an area-proportional Venn diagram in Fig. 3B (46). Annotated genes were determined by using Panther Classification system (47). We also identified probe sets that were regulated by agonists for LXRα/β, PPARγ, and PPARδ. Altogether, agonists for these receptors could regulate 802 probe sets representing 553 annotated genes (Fig. 3C) (referred as “GW3965 and/or RSG and/or GW1516” in Fig. 3).

Figure 3

Gene sets regulated by LXRα/β, PPARγ, PPARδ, and RXR agonists only partially overlap in Mo-DCs. A, Experimental setup for Affymetrix microarray analysis. Eighteen hours after isolation of human monocyte from healthy donors (n = 3), cells were treated for 12 h with LG268 (RXR agonist) and 9cisRA (agonist for RAR and RXR), GW3965 (LXRα/β panagonist), RSG (PPARγ agonist), and GW1516 (PPARδ agonist). Cells were cultured in presence of IL-4 and GM-CSF from zero time point. B, Area-proportional Venn diagram shows the overlap between probe sets regulated by RXR agonists. The intersection of gene sets regulated by RXR agonists is labeled as “LG268 AND 9cisRA.” C, Overlap between probe sets regulated the agonist specific for LXRα/β (GW3965), PPARγ (RSG), and PPARδ (GW1516) is shown on an area-proportional Venn diagram. The union of these genes is labeled as “GW3965 AND/OR RSG AND/OR GW1516.” D, Area-proportional Venn diagram shows the overlap between gene sets regulated by both RXR agonists and by “GW3965 AND/OR RSG AND/OR GW1516.” Lists of genes/probe sets representing different sets of Venn diagram (intersection and two complement sets) are shown in heat maps. Color intensities reflect the ratios of signal intensities as shown.

In this way, we could define a gene set regulated by both RXR agonists (“LG268 and 9cisRA”) and another one regulated by at least one agonist for RXR permissive partners (“GW3965 and/or RSG and/or GW1516”). In the next step, we compared these two gene sets to identify overlapping as well as nonoverlapping gene sets (Fig. 3D). We found that the intersection of these gene sets contained 201 probe sets, whereas 601 and 138 probe sets were not regulated by RXR agonists or by agonists for LXRs and PPARs, respectively. Remarkably, this quantitative analysis demonstrated that 75% of genes regulated by at least one agonist for RXR permissive partners was not regulated by both RXR agonists. It should be noted that several of them were regulated either by 9cisRA or LG268 but not by both (Fig. 3D, heat map on the left). Conversely, a significant fraction (40%) of the “LG268 and 9cisRA” set was not regulated by LXR and PPAR agonists. Taken together, our most surprising finding was the size of complement sets. Therefore, in our detailed analyses, we aimed at addressing why RXR and its permissive partners cannot completely replicate the effects of each other.

RXR agonists are less effective than their LXR and PPAR counterparts in the regulation of direct LXR and PPAR target genes in Mo-DCs

First, we focused on the gene set that was regulated by LXR and PPAR agonists but was not affected (or more precisely, was regulated by <2-fold) by RXR agonists (Fig. 3D). Approximately half of these genes was down-regulated by LXR and PPAR agonists (Fig. 3D, heat map on the left). Mechanisms whereby these agonists inhibit gene expression are not completely understood, and this is especially true for RXR agonists. In many cases, LXR and PPAR act as monomers, therefore it is very likely that RXR and its partners down-regulate significantly different subsets of genes. Remarkably, many genes up-regulated by LXR and PPAR agonists, especially genes exhibiting small (2- to 3-) fold inductions, were also not affected by RXR agonists. We found that 83% of probe sets that were regulated by none RXR agonists were up-regulated by less then 3-fold upon RSG treatment. This number was 60% in the case of genes regulated by both RXR agonists (Supplemental Fig. 1, published on The Endocrine Society’s Journals Online web site at http://mend.endojournals.org). Because these differences were also observed in the case of genes regulated by GW3965 and GW1516 (data not shown), we concluded that PPARδ-RXR and LXR-RXR signaling was also impaired in Mo-DCs. We also tested the regulation of a series of direct LXR and PPAR target genes, with characterized response element (Fig. 4). Based on microarray data (Fig. 4, A and B) and dose response curves (Fig. 4, C and D), we made the following observations. First, the efficacy of 9cisRA and LG268 is much lower in most cases than those of GW3965, RSG, and GW1516. The only exception was ANGPTL4. Second, in the case of LXR direct targets, the ratio between the efficacies of RXR agonists and GW3965 were very similar in the investigated genes; there were no significant gene-to-gene variations. In contrast, PPAR direct targets showed large gene-to-gene variations, e.g. between PDK4 and CD36. Third, the two RXR agonists were usually similarly efficient in most cases with the notable exception of CD36 and 10 μm 9cisRA on the regulation of ABCA1. This latter observation may be explained by the fact that in mouse primary macrophages, ABCA1 was regulated by RAR agonists (48). Last, the response of LG268 usually saturated between 10 and 100 nm in the case of LXR target genes, and a very high dose (10 μm) LG268 was less effective than 1 μm (Supplemental Fig. 2). Collectively, analyses of the regulation of direct LXR and PPAR target genes demonstrated that permissive heterodimers are “asymmetric” in Mo-DCs in the aspect of transcriptional regulation.

Figure 4

RXR agonists are less effective than their LXR and PPAR counterparts in the regulation of direct LXR and PPAR target genes. A and B, The fold changes induced by LG268, 9cisRA, GW3965, RSG, and GW1516 were determined by Affymetrix microarrays. Bar graphs show the effect of agonists on LXR direct target genes (A) and PPAR direct target genes (B). Fold inductions were calculated from mean values of biological triplicates for agonist-treated and vehicle-treated samples. C and D, Dose response curves of selected LXR (C) and PPAR (D) direct target genes measured by RT-qPCR. Differentiating DCs were treated with various concentrations of RXR, LXR, or PPAR agonists 18 h after plating. Cells were harvested 12 h thereafter. One representative experiment of three performed is shown. Values are expressed as mean of technical triplicates ± sd of the mean. ABCG1, ATP-binding cassette, subfamily G, member 1.

Both ligands are required in phorbol myristate acetate (PMA)-treated THP-1 cells but not in Mo-DCs for maximal activation of fatty acid-binding protein 4 (FABP4)

In cotransfection assays, both PPARγ and RXR ligands were required for maximal activation (7). Due to the fact that RXR agonists are less effective than RSG in Mo-DCs in the regulation of direct PPAR target genes, we were curious whether both ligands are required for maximal transcriptional activation in Mo-DCs. Therefore, we added RSG for 12 h at increasing concentrations alone or along with a fixed concentration of LG268 and measured the mRNA level of PPAR target gene FABP4 that exhibited especially low response upon RXR activation (Fig. 4D). Remarkably, we found that activation of RXR is not required for the maximal response, but LG268 inhibited the effect of RSG (Fig. 5A). This was an unexpected observation and raised the question whether this phenomenon is specific for Mo-DCs. We tested the regulation of FABP4 by LG268 and RSG in other myeloid cells, monocyte-derived macrophages, and THP-1 cell line cultured in the presence of 20 nm PMA. This latter cell type is considered as a model for monocyte-derived macrophages (49). We found that FABP4 was regulated in monocyte-derived macrophages similarly to Mo-DCs, and LG268 inhibited the effect of RSG (data not shown). In contrast, the response with both ligands was synergistic in PMA-treated THP-1 cells (Fig. 5B). These data suggested that RXR signaling was different in various myeloid cells. Taken together, we concluded that RXR agonists cannot regulate many genes affected by LXR and PPAR agonists mainly because two reasons: first, agonists for RXR and its partners very likely use different pathways to down-regulate genes, and second, in the case of up-regulation, RXR permissiveness is partially impaired.

Figure 5

Regulation of FABP4 by agonists for PPARγ and RXR in (A) differentiating Mo-DC and (B) PMA-treated THP1 cells. Cells were harvested 12 h after ligand treatment, and mRNA level of FABP4 was determined by RT-qPCR. Increasing concentration of LG268 or RSG alone or increasing amount of RSG with a fixed concentration of LG268 (100 nm) were used. PMA was added at 20 nm 3 h before ligand treatment. One representative experiment of three performed is shown. Values are expressed as mean of technical triplicates ± sd of the mean.

RXR agonists regulate only a fraction of genes affected by RARα and VDR agonists

We and others demonstrated that nonpermissive partners, RARα and VDR, are expressed in Mo-DCs (Fig. 2) and regulate the immunephenotype of this cell type (reviewed in Ref. 33). Previously, we performed microarray analysis to identify targets of VDR (29) and RARα (Széles L. and L. Nagy unpublished data) in differentiating Mo-DCs. We used these datasets and compared them with gene sets regulated by RXR agonists. We were especially interested in two questions. First, we aimed at evaluating the capacity of RXR agonists in the transcriptional regulation of genes that were affected by RARα and VDR agonists. Although RAR and VDR are considered nonpermissive partners, under certain circumstances, direct targets of VDR-RXR can be influenced/induced by RXR agonists (50), and RAR-RXR can be activated from the RXR side (conditional permissiveness). Second, we were curious whether RARα agonist AM580, and/or VDR agonist 1,25-dihydroxyvitamin D3 (1,25-vitD) can regulate “LG268 and 9cisRA” genes that were not regulated by LXR and PPAR agonists (cluster 2 in Fig. 6A).

Figure 6

Comprehensive analysis of transcriptional changes induced by agonists for RXR and its partners in Mo-DCs. A, Genes regulated by LG268 and 9cisRA were classified into two clusters based on their response to agonists for permissive partners, LXR and PPAR. Cluster 2 contains genes that are not regulated by LXR and/or PPAR agonists. B and C, Probe sets regulated by agonists of nonpermissive partners: RARα (AM580) and VDR (1,25-vitD) were also identified and compared with the entire list of RXR-regulated probe sets as well as to cluster 2. Area-proportional Venn diagrams show the overlap between the indicated probe sets. Note the proportion (65%) of probe sets of cluster 2 that are regulated by RARα agonist AM580. D, Comparisons of the effect of agonists of RXR partners and RXR-specific LG268. RXR partners were activated by GW3965 (LXRα/β), RSG (PPARγ), GW1516 (PPARδ), AM580 (RARα), and 1,25-vitD (VDR). Twenty genes exhibiting the highest fold-induction values were identified for each agonist by microarray analysis. Fold inductions were calculated from mean values of biological triplicates of agonist-treated vs. vehicle-treated microarray samples.

To address these questions, we compared the probe sets regulated by agonists for nonpermissive partners RARα and VDR, and the RXR-regulated probe sets. We found that RXR agonists could regulate only 8% of probe sets regulated by 1,25-vitD (129 out of 1635), and LG268 was not effective even in the case of most highly up-regulated genes (Fig. 6D and Supplemental Fig. 3). Although many genes that were highly induced by AM580 were also affected by LG268 (Fig. 6D), RXR agonists could regulate only 14% of probe sets regulated by AM580 (153 out of 1097). Remarkably, comparison of probe sets regulated by agonists for RARα and VDR to cluster 2 also demonstrated that AM580 (and 1,25-vitD to a smaller extent) could regulate large proportion of probe sets of cluster 2. Collectively, our analyses on the agonists for nonpemissive partners showed that VDR targets were regulated as one would expect: highly induced by 1,25-vitD and slightly or none by RXR agonists. Although the majority of AM580-regulated genes were also not affected by LG268, we could identify several genes, which were up-regulated by both AM580 and LG268.

RARα agonist AM580 and RXR agonists up-regulate many genes that are not induced by agonists for LXR and PPARs

We could identify genes that were regulated by agonists for RXR and RAR but were not affected by LXR and PPAR agonists (Fig. 6, A and C). We selected six genes from cluster 2 and validated our microarray results by RT-qPCR (Fig. 7A). The six genes included TGM2, ASB2, and CD1d, which have been described as RAR-RXR direct target genes in mice and/or human (44,51,52,53). Although RAR is a nonpermissive partner, RAR and RXR agonists together have greater effect than the RAR agonist alone (54,55). We aimed at evaluating whether RXR agonists could regulate these genes, because the endogenous RAR agonist is present in Mo-DCs or other mechanism(s) might be responsible for this phenomenon. We cannot exclude that RAR-RXR (activated by AM580) and “partner X”-RXR (activated by LG268) can regulate common targets (Fig. 7B). In this way, the two agonists are both effective but trigger different dimers. According to the expression patterns of RXR and its partners (Fig. 2), the partner X could be RXR itself, or a member of the NR4A subfamily. Although the mRNA level of Nurr1 and Nur77 are very low after 24 h differentiation, it is possible that the proteins are still present.

Figure 7

RARα agonist AM580 up-regulates many genes that are induced by RXR agonists but not by agonists for LXR or PPARs. A, Six genes from cluster 2 were chosen for RT-qPCR validation. Monocytes differentiating into DCs were treated with RXR, LXR, PPARs, or RARα agonists 18 h after plating. Cells were harvested 12 h thereafter. B, Schematic illustration shows that RXR agonist can regulate transcriptional targets of RAR-RXR via different mechanisms. Note that these genes were not regulated by agonists for LXRs and PPARs. HRE, Hormone responsive element. C, Concentration of ATRA and 9cisRA were determined by an LC-MS method in differentiating DCs (diff.DC; 30 h) and immature DCs (IDC; 5 d). Chromatograms of one representative experiment with external 9cisRA and ATRA standards are shown. E, LG268 and 9cisRA regulate most selected genes in the late phase of DC differentiation as determined by RT-qPCR. Cells were treated with LG268, 9cisRA, or 1 μm XCT0135908 (XCT; Nurr1-RXR-selective RXR agonist) at d 4.5 for 12 h. One representative experiment of three performed is shown. Values are expressed as mean of technical triplicates ± sd of the mean (A and D).

We first investigated the presence of RAR agonists in differentiating Mo-DCs and determined the intracellular all-trans retinoic acid (ATRA) and 9cisRA concentrations using a sensitive and specific liquid chromatography-mass spectrometry (LC-MS) method (27,56). We cultured DCs as previously described, and 18 h after plating, we treated the cells with LG268 for additional 12 h. The estimated ATRA concentration was 0.68 ± 0.032 ng/g cell pellet (equivalent to ∼2.2 nm), whereas the 9cisRA level was under our detection limit (Fig. 7C). We assumed that this amount of ATRA might be sufficient to regulate gene expression and “sensitize” RAR-RXR for RXR activation.

We concluded that using a condition when both ATRA concentration and Nurr1 and Nur77 level are lower could help in defining the most likely scenario. The expression profile of RXR partners (Fig. 2) and our previous study documented that at d 5, in the late phase of DC differentiation, the concentration of intracellular ATRA as well as the level of Nurr1 and Nur77 are considerably lower (27). Therefore, we measured 9cisRA and ATRA in immature DCs after 5 d of differentiation. We found that ATRA concentration is six times lower (Fig. 7C), whereas 9cisRA remained undetectable.

We repeated our experiment in the late phase of differentiation using various RXR agonists (Fig. 7D). We found that LG268 retained its capacity to induce the expression of five out of six genes in the late phase of differentiation, whereas 9cisRA could up-regulate all selected genes. As expected, Nurr1-RXR specific RXR agonist XCT0135908 was not effective in most cases. Taken together, this part of our analysis documented that activation of RXR could also regulate gene expression in a permissive heterodimer-independent manner, in a phase when ATRA and 9cisRA levels are very low, most likely via RXR homodimers. Further studies on promoters of individual genes could reveal whether and how RXR homodimers regulate these genes.

Discussion

At least one RXR isotype can be detected in almost every tissue (15), indicating the importance of these receptors in cellular signaling. RXRs play a dual role: they function as obligate dimerization partners for many nuclear receptors and also as ligand-activated receptors. The receptor activity of RXR was documented in jellyfish (57) and seems to be retained during phylogenesis: a series of studies in mice and Xenopus laevis demonstrated that RXR signaling is active in vertebrates in vivo (3,58,59,60,61,62). In this study, we used Mo-DCs as a case study to systematically and quantitatively characterize the transcriptional programs regulated by agonists for RXR and its partners to get a complete account on its impact on gene expression. According to our knowledge, there is not another complete study in which all aspects of RXR signaling have been accounted for in a single in vivo relevant cell type.

RXR permissiveness: a central issue in RXR biology

Although pioneering studies using cotransfection assays and EMSA were crucial in understanding the complexity of RXR biology, these techniques have their own limitations, including that conclusions are drawn based on analyses of one or a few genes and response elements. These methods are also not suitable for investigation of cell-type specific transcriptional regulation by nuclear receptors. Moreover, RXR permissiveness was defined and studied by cotransfection assays mainly using direct transactivation as a model, but LXR and PPAR regulate transcription via several other mechanisms (reviewed in Ref. 19), including various forms of transcriptional repression. Therefore, gene expression data obtained by measuring transcript levels of individual genes or large gene sets in different tissues are necessary to draw general conclusions and to define permissiveness on a global scale. In cotransfection assays, the components (e.g. receptors, ligands, and response elements) are well defined, in this way, “RXR permissiveness” refers to molecular mechanism, e.g. permissive heterodimers respond to ligands that bind to either receptor subunit. In contrast, one must be aware of the complexity of RXR biology when drawing conclusions from gene expression data. For example, if both RSG and LG268 can induce the transcription of a certain gene, one cannot conclude that these ligands necessarily affect the respective subunits of PPARγ-RXR for multiple reasons. First, RXR homodimers can regulate PPAR target genes (62). Second, the subsets of regulated genes by agonists for LXRα/β, PPARγ, and PPARδ overlap (Fig. 3C). Therefore, the transcription of such genes can be altered parallel via three different heterodimers by rexinoids. Third, PPAR and RXR agonists can regulate genes, by employing distinct mechanisms. In Mo-DCs, TGM2 is regulated by agonists for RAR, RXR, and PPARγ (Fig. 1E) (27). RAR-RXR and RXR homodimer regulate TGM2 very likely directly (at 3 h, primary effect) (44), whereas RSG via the induction of the synthesis of ATRA (after 36–48 h, secondary effect) (27). For these reasons, if a gene is regulated by both ligands at the transcriptional level, there is some uncertainty whether the two agonists act on subunits of the same heterodimer. Because of these reasons, the term “pharmacological permissiveness” would more accurately describe the phenomenon that RXR agonists can replicate the effect of permissive partner.

Our study mainly conducted with synthetic, receptor-specific ligands. However, using synthetic nuclear receptor ligands often raises the question whether the endogenous agonists have similar effects in vivo. On the one hand, we cannot exclude the possibility that endogenous LXR and PPAR agonists can exhibit more-or-less different characteristics in vivo. For example, PPARs bind a broad range of natural ligands with different affinities to the receptor. Therefore, it is possible, that an RXR agonist can be as effective as a less-potent PPAR agonist. On the other hand, rexinoids are essential tools to investigate permissiveness, because natural RXR agonists are not selective and induce other nuclear receptors as well. However, in our study, the natural RXR agonist 9cisRA and rexinoid LG268 exhibited similar characteristics (e.g. impaired RXR-permissiveness), suggesting that this phenomenon is not ligand specific.

RXR permissiveness is relative and determined by partner-, cell type-, and gene-specific mechanisms

Our global transcriptome analyses provided support for the distinction of permissive vs. nonpermissive partners. We found that RXR agonists regulated most LXR and PPAR direct target genes (Fig. 4) and highly induced genes (Fig. 6D), although not as effectively as agonists of LXR and PPAR. In contrast, RXR agonists regulated only small portions of RARα or VDR targets (Fig. 6). Although RXR agonists could regulate some VDR and RARα targets (Fig. 6B), these data would not necessarily mean that VDR-RXR or RAR-RXR are permissive in Mo-DCs. Our detailed analysis on a subset of genes (Fig. 7), as well as previous studies on response elements bound by RXRs (44,63,64), suggest that RXR homodimers might regulate RAR targets.

We found that LXR and PPARs behave as permissive partners, but permissiveness is impaired in various aspects in Mo-DCs. For example, we demonstrated that RXR agonists less effectively regulated the expression of many PPARγ target genes as compared with RSG. Moreover, the maximal activation of the prototypic PPARγ target gene FABP4 did not require both ligands, whereas LG268 antagonized the effect of RSG (Fig. 5A). A similar phenomenon, antagonism by RXR agonists, was observed in the case of FXR-RXR target gene, BSEP (65).

The fact that RXR permissiveness can be limited in certain cell types was documented earlier (66,67). Using Zucker diabetic fatty rats, the effects of PPARγ-specific RSG and RXR-specific LG268 on metabolic gene expression were compared in white adipose tissue, skeletal muscle, and liver (67). Remarkably, in adipose tissue, LG268 could not replicate the effect of RSG, therefore the authors concluded that PPARγ-RXR is “nonpermissive” in adipose tissue. In another study, rexinoid bexarotene induced lipogenic LXR target genes, exerting a permissive effect on LXR-RXR (66). In contrast, the same compound failed to regulate LXR genes implicated in cholesterol homeostasis. This latter study and our observations on PPAR targets (Fig. 4B) support the concept of gene-selective permissivity. Remarkably, although the mechanisms for prevention of RXR receptor function by its nonpermissive partners are characterized (20,21,22,23), the molecular mechanisms explaining the tissue-specific activation of heterodimers are not investigated in detail. However, the cofactor pool available in a certain cell type is very likely the most important determinant of such activities. Schulman et al. (24) demonstrated that RSG recruits cAMP response element-binding protein to the PPARγ-RXR complex, whereas LG268 recruits steroid receptor coactivator-1. In this way, the pool of cofactors may allow efficient activation for both RXR and its permissive partners in certain cell types (e.g. in fibroblasts as it is suggested by cotransfection assays) but may be suboptimal for RXR signaling in other tissues, such as Mo-DCs, mature adipose tissue, or liver.

The value of a global analysis of RXR signaling

Microarray analyses are suitable to identify potential target genes of transcription factors, including nuclear receptors. RXR is unique among nuclear receptors, because one must consider the activity of RXR and RXR partners together to evaluate many aspects of RXR signaling and to address to what extent the receptor’s activation results in triggering various heterodimers and RXR homodimer. Using global microarray analysis, we were able to identify four attributes of RXR signaling that cannot be identified by investigation of individual genes alone. First, as we discussed above, we could demonstrate that RXR permissiveness is impaired in Mo-DCs, and dozens of LXR and PPAR targets were regulated to a lesser extent by RXR agonists compared with LXR and PPAR agonists. Second, our global analysis also showed that there is significant overlap between targets of RXR partners: LXR and PPARs as well as VDR and RAR (Figs. 3 and 6). Due to the fact that heterodimers of these receptors bind to distinct response elements, our data suggest that these heterodimers may occupy complex response elements, use common response elements, or there are significant cross talks between these receptors. Third, agonists for RXR partners could activate the majority of RXR targets (Fig. 6). Many genes regulated by RXR agonists were not affected by permissive partners (cluster 2 in Fig. 6), but approximately 75% of genes in cluster 2 subset were regulated by RARα or VDR agonists. In this way, only 36 probe sets (10 up-regulated and 26 down-regulated) representing 30 annotated genes were not affected by at least one agonist for RXR partners. It is well established that RXR homodimer as a functional transcriptional unit exists and plays an overlapping function with PPAR-RXR (62) and probably with other heterodimers (44,68). However, the existence of a pathway regulated exclusively via RXR homodimers/monomers is still an unexplored area. Our study establishes that it represents a smaller fraction of gene expression changes with a bias toward repression. Last, our analysis showed that more than 40% of genes regulated by RXR agonists were down-modulated. There are several mechanisms proposed for ligand-dependent repression by RXR partners (19), but we still have very limited knowledge about how RXR can down-regulate gene expression and whether RXR can function as a monomer in repression. Our study shows that these questions are valid and relevant and need to be answered using the combination of global as well as gene-specific analyses. We believe that studies like this in other cell types would be useful and can lead to a more complete picture and a better understanding of RXR signaling and can also help to determine the therapeutic utility of rexinoids in various tissues and diseases.

Materials and Methods

Cell culture and DC generation

CD14+ monocytes were obtained from platelet-free buffy coats from healthy donors by Ficoll gradient centrifugation followed by immunomagnetic cell separation with anti-CD14-conjugated microbeads (VarioMACS Separation System; Miltenyi Biotec, Bergisch Gladbach, Germany). Monocytes were cultured in multiwell culture plates or tissue flasks at a density of 106 cells/ml in RPMI 1640 (Sigma-Aldrich, St. Louis, MO) supplemented with 10% fetal bovine serum (Invitrogen, Carlsbad, CA), 2 mm glutamine, 800 U/ml GM-CSF (Leucomax, Gentaur Molecular Products, Brussels, Belgium), 500 U/ml IL-4 (PeproTech EC, London, UK), and penicillin/streptomycin (Sigma-Aldrich). IL-4 and GM-CSF was replenished at d 3. To obtain monocyte-derived macrophages, cells were conditioned similarly, but M-CSF (Leucomax) was used instead of GM-CSF. Langerhans cells, dermal DCs, and blood CD1c+ DCs were isolated as described previously (38,39). THP-1 cells were obtained from American Type Culture Collection (Manassas, VA) and cultured in RPMI 1640 supplemented with 10% fetal bovine serum, 2 mm glutamine, and penicillin/streptomycin. PMA was added to THP-1 cells at 20 nm 3 h before ligand treatment.

Ligands

Cells were treated with vehicle (1:1 ethanol-dimethylsulfoxide) or with the following ligands: 1 μm 9cisRA (BASF, Ludwigshafen, Germany), 100 nm LG268 (a gift from R. Heyman; Ligand Pharmaceuticals, San Diego, CA), 1 μm GW3965 (GlaxoSmithKline, Research Triangle Park, NC), 1 μm RSG (Alexis Biochemicals, San Diego, CA), 1 μm GW1516 (GlaxoSmithKline), 1 μm XCT0135908 (a gift from Thomas Perlmann), 100 nm AM580 (Biomol-Enzo Life Sciences, Plymouth Meeting, PA), and 10 nm 1,25-vitD (Biomol-Enzo Life Sciences). For dose response analyses, ligands were used at indicated concentrations.

Microarray analysis: sample preparation, labeling, and hybridization

Monocytes differentiating into DCs were treated with agonists for RXR and its partners or vehicle 18 h after plating (experiment with RXR and permissive partners) or 14 h after plating (experiment with nonpermissive partners). Cells were harvested 12 h thereafter. Total RNA from 6 × 106 cells was isolated using the RNeasy kit (QIAGEN, Hilden, Germany). Experiments were performed in biological triplicates representing samples from different donors. Further processing and labeling, hybridization to Human Genome U133 Plus 2.0 Arrays (Affymetrix, Santa Clara, CA) and scanning were carried out at Clinical Genomics Center, University of Debrecen, and the European Molecular Biology Laboratory, Heidelberg, Germany. Microarray data have been deposited into the Gene Expression Omnibus database under accession no. GSE23618 (DC subtypes), GSE8658 (Mo-DC differentiation), and GSE23073 (differentiating Mo-DC treated with nuclear receptor ligands).

Microarray data analysis

Image files were imported to GeneSpring 7.3 (Agilent, Santa Clara, CA). Raw signal intensities were normalized per chip (to 50th percentile). We removed probe sets that failed to reach a raw signal intensity of 125 (approximately the median of the signal intensity values of all probe sets) at least in three out of 18 samples (experiment with RXR and permissive partners) or three out of nine samples (experiment with nonpermissive partners). To identify probe sets regulated by at least 2-fold by agonists, we compared the mean of triplicates of vehicle-treated vs. agonist-treated samples. In this study, we used less strict criteria than in our previous studies (28,29), because we wanted to avoid not only the false positive, but also the false negative results. Filtering on expression values and fold inductions, we could reduce the number of the false negative data that could appear as false positive result in different sections of Venn diagrams. However, we draw our conclusion only if RT-qPCR validations supported our microarray data. For heat map visualization of signal intensities, each probe set was normalized to the signal intensities of vehicle controls (fold change).

Real-time quantitative RT-PCR

RT-qPCR was carried out as described earlier (26,29) using TaqMan probes (Applied Biosystems, Foster City, CA). Gene expression was quantified by the comparative threshold cycle method and normalized to cyclophilin A expression. All experiments were conducted as biological triplicates. Values are expressed as mean ± sd of the mean.

Immunohistochemistry

Double immunofluorescence (IF) staining was made on serial sections of cell blocks obtained from the pelleted Mo-DCs, followed by formalin fixation and paraffin embedding. Double IF staining was made as described earlier (69) using mouse monoclonal antibodies (mAbs) to RXRα (Perseus Proteomics, Tokyo, Japan) and DCSign (BD Pharmingen, San Diego, CA) by means of a sequential immunolabeling procedure. In brief, after a 1-h incubation of sections with mAb to RXRα and washings in Tris-buffered salt solution (pH 7.4) at room temperature, we applied antimouse secondary IgG coupled to horse radish peroxidase-polymer (Dako, Glostrup, Denmark) and a tyramide-Texas Red solution as the final fluorochrome for the demonstration of RXRα (red nuclear fluorescence if positive). After washing, a mAb to DCSign-biotinylated anti mouse IgG (Fab) complex was added to the sections for an additional 1-h incubation followed by a streptavidin-FITC development for 30 min. After washing, the nuclear counterstain was made with 4′,6-diamidino-2-phenylindole containing the cover medium (Vector Laboratories, Burlingame, CA). To check the specificity, IF labelings were made separately for both antibodies, and negative controls were also included for each IF staining. Negative controls for RXRα and DCSign were consisted of slides with isotype-matched irrelevant mAb IgG raised against Saccharomyces (Dako) in place of the primary antibody.

Flow cytometry

Cell staining was performed using phycoerythrin (PE)-conjugated mAbs. Labeled Abs for flow cytometry included anti-CD80-PE, CD86-PE, CD1a-PE, CD1d-PE, and isotype-matched controls (BD Pharmingen). The cells were assessed for fluorescence intensity using FACSCalibur cytometer (BD Biosciences, San Jose, CA). Data analysis was performed using Cellquest software (BD Biosciences).

Determination of ATRA and 9cisRA concentration

Concentrations of ATRA and 9cisRA were measured in cell pellets by our LC-MS method described in detail previously (27,56). In brief, 50–100 mg of cell pellet was diluted with a 300 μl volume of isopropanol, and the extracts were dried in a concentrator (Eppendorf 5301) at 30 C. The dried extracts were resuspended with 60 μl of methanol, diluted with 40 μl of 60 mm aqueous ammonium acetate solution, transferred into the autosampler, and the whole volume was subsequently analyzed with an LC-MS system. The LC-MS system consisted of a Waters 2695XE separation module, MS-MS detector (Micromass Quattro Ultima PT; Waters, Milford, MA), including an APCI ionising option (Ion sabre APCI; Waters). The HPLC system and the mass spectrometer, including chromatogramm procession and evaluation, were controlled by the MassLynx 4.1 software (Waters).