Anti-inflammatory Chromatinscape Suggests Alternative Mechanisms of Glucocorticoid Receptor Action

Summary

Despite the widespread use of glucocorticoids (GCs), their anti-inflammatory effects are not understood mechanistically. Numerous investigations have examined the effects of glucocorticoid receptor (GR) activation prior to inflammatory challenges. However, clinical situations are emulated by a GC intervention initiated in the midst of rampant inflammatory responses. To characterize the effects of a late GC treatment, we profiled macrophage transcriptional and chromatinscapes with Dexamethasone (Dex) treatment before or after stimulation by lipopolysaccharide (LPS). The late activation of GR had a similar gene expression profile as from GR pre-activation, while ameliorating the disruption of metabolic genes. Chromatin occupancy of GR was not predictive of Dex-regulated gene expression, contradicting the ‘trans-repression by tethering’ model. Rather, GR activation resulted in genome-wide blockade of NF-κB interaction with chromatin and directly induced inhibitors of NF-κB and AP-1. Our investigation using GC treatments with clinically relevant timing highlights mechanisms underlying GR actions for modulating the ‘inflamed epigenome’.

eTOC text

Although glucocorticoids are widely used anti-inflammatory drugs, relevant mechanisms are unclear. Oh et al. monitored the epigenomic landscape of macrophages, and found that the gene-inducing activity of GR is crucial for boosting inhibitors of inflammatory factors. This cautions the idea that GR ligands selectively promoting trans-repression should improve therapeutic outcome.

Introduction

Macrophages are critical effector cells of the innate immune system and they can take on diverse homeostatic or pathological roles through context-dependent differentiation (Gosselin et al., 2014; Wynn et al., 2013). Macrophages have a direct role in numerous tissue-damaging inflammatory conditions including Alzheimer’s disease, atherosclerosis, inflammatory bowel disease, and stroke. Uncontrolled macrophage activation can lead to secretion of pro-inflammatory cytokines such as TNF-α and IL-1β, induction of nitric oxide (NO) synthase, further in vivo recruitment of other immune cells, and/or activation of adaptive immune cells via antigen presentation or cytokine-mediated polarization. Even for diseases where macrophages may not have a causal role, therapeutic interventions impact macrophages because of their ubiquitous presence throughout the body.

Glucocorticoids (GCs) are the most widely prescribed anti-inflammatory drugs to treat numerous inflammatory conditions ranging from asthma to psoriasis (Gerstein et al., 2012). GCs exert their therapeutic effects primarily through the glucocorticoid receptor (GR). The generally accepted mechanism is that GR antagonizes chromatin interactions of transcription factors such as NF-κB by tethering to these DNA-bound pro-inflammatory factors and recruiting co-repressors, resulting in subsequent trans-repression of TNF-α and other inflammatory genes (Reichardt et al., 2001). Global effects of a synthetic GR ligand Dexamethasone (Dex) have been studied in cells treated with Dex prior to or together with an inflammatory stimulus (Rao et al., 2011; Uhlenhaut et al., 2013). However, the clinical use of GCs is to treat inflammatory conditions whereby the GC targets pathologically active immune cells, not quiescent immune cells.

Hence, missing from the extensive literature on GCs and GR biology is a characterization of the genomic action of GR in immune cells that have already received an inflammatory signal. Because there is clear evidence that the chromatinscape is a dynamic entity that is shaped by cell differentiation and external stimuli (Ghisletti et al., 2010; Gosselin et al., 2014; He et al., 2012; Heinz et al., 2010; John et al., 2011; Siersbaek et al., 2011; Thurman et al., 2012), such a gap in our knowledge raises several questions. Are the transcription factors in a race for gene regulation, wherein the first transcription factors (TFs) to arrive on the regulatory site dominate over the late-arriving TFs? Does the repertoire of GR binding sites depend on the timing of activation with respect to the onset of inflammation? If so, how is the binding repertoire related to GR action? To address these unresolved questions, we examined the epigenomic and transcriptomic responses of primary macrophages, varying the order of GC treatment with respect to the onset of an inflammatory signal.

Results

Primary macrophages treated with Dex before or after an inflammatory stimulus share a global transcriptional profile

We investigated three different courses of lipopolysaccharide (LPS) responses in mouse bone marrow derived macrophages (BMDMs): a reference time course after LPS stimulation alone without Dex, another time course treated early with Dex (before LPS stimulation), and another course treated late with Dex (3 hours after LPS) (Figure 1A). RNA-seq was performed in biological duplicates to profile the transcriptome at 4 hours and 10 hours after LPS stimulation, except for the late Dex-treated course where sample collection is feasible only for the 10 hour LPS time point. We assessed how the expression of LPS-regulated genes is altered by Dex, treated early or late with respect to the onset of LPS stimulation. Both early and late treatments of Dex reversed the LPS-induced transcriptional regulation of a large number of genes, with the early Dex treatment producing higher magnitude effects (Figure 1B, Figure S1). Since global scatter plots only allow pairwise comparisons of individual conditions, we performed an unsupervised cluster analysis to discover distinct expression patterns across all the conditions. We utilized a robust variant of the widely used K-means algorithm, varying the number of clusters, and applied the clustering method to the set of LPS-induced genes (1,820 in total according to our criteria; see Methods) (Figure 2A) (Supplementary Table S1). The analysis revealed three major clusters of LPS-induced genes (Clusters 1–3). Cluster 1 contained genes whose expression is reduced by Dex, largely independent of the timing of Dex. Genes in cluster 2 and 3 were less affected by Dex and distinguishable by the magnitude of LPS-induced fold change in expression.

Figure 1. Transcriptomic profiling of LPS responses in macrophages with varied timing of Dex treatment (Dex pre-treatment versus late-treatment).

(A) Experimental design for BMDM treatment and sample collection for RNA-seq. The timeline is with respect to the onset of LPS stimulation (t = 0). (B) Genome-wide expression patterns from RNA-seq data. Scatter plots show mean log2 (FPKM + 0.1) values from biological duplicates. The red and blue points mark highly induced and repressed genes, respectively, i.e. genes with fold change greater than 4 for at least one post-LPS time point. The pink and skyblue points mark modestly induced and repressed genes, respectively, i.e. those with fold change between 2 and 4 for at least one post-LPS time point. The black points show the remaining genes. Green dashed line indicates the diagonal. The color coding of genes is the same for all the plots to keep track of the LPS-regulated genes in plots comparing Dex-treated and Dex-untreated samples. See also Figure S1.

Figure 2. Classification of genes based on sensitivity to Dex and the treatment timing.

(A) Heatmap of three gene classes from RNA-seq obtained by a robust K-means clustering method, partitions around the medoids (see Methods). The largest cluster (“Transrepression by GR”) is shown on top. The other two clusters are further distinguished as highly LPS-induced genes and moderately LPS-induced genes, both of which show little effects from Dex treatments (“Dex-insensitive”). The color scale displays log2 FPKM ratios with respect to the LPS 0h values. Values from biological duplicates are shown in separate columns for each sample. (B) Gene ontology of genes sensitive to the timing of Dex treatment. For the lists of genes, see Supplementary Tables S1–S3. See also Figure S2.

Select genes are differentially regulated by Dex based on the timing of treatment

We noted that the cluster analysis did not detect a class of genes whose expression differed between the two Dex treatments of interest in our study. Unsupervised cluster analysis can miss subtle patterns in a small number of genes if they are obscured by other large-amplitude effects. Therefore, we focused on the two conditions, directly comparing the early and late Dex treatments. 102 genes (referred to as the class “lower in late Dex”) were indeed found to have a higher expression in the early vs late Dex (Supplementary Table S2), potentially corresponding to genes that Dex may suppress more efficiently when given with a clinically relevant timing. On the other hand, 319 genes were expressed at lower amounts in the early vs late (referred to as “lower in early Dex”) (Supplementary Table S3). The differential expression of selected genes from each class was confirmed by RT-qPCR (Figure S2). Gene ontology analysis revealed a striking separation of these two classes: the “lower in late Dex” class was enriched for genes regulating various aspects of metabolism, while the “lower in early Dex” class showed enrichment for immune genes including those encoding cytokines, chemokines, and their receptors (Figure 2B, Supplementary Tables S3–S4).

The “lower in late Dex” class contained several genes with essential roles in inflammatory conditions, including Cx3cr1, Selp, Cd55, Cxcl13, Clec5a, and Pstpip1. However, their functional categories did not show up in our gene ontogeny (GO) analysis, because these genes were outnumbered by the metabolism-related genes in our GO analysis. Despite the limited number, there is abundant evidence that these genes are mechanistically implicated in inflammatory diseases. For example, Cx3cr1 (encoding fractalkine receptor) is a useful surface marker of tissue resident macrophages distinguishing them from infiltrating monocyte-derived macrophages (Mizutani et al., 2012). P-selectin, encoded by Selp, mediates neutrophil adhesion to endothelium (Woollard et al., 2008) and has been reported to exacerbate inflammatory vascular diseases (Kisucka et al., 2009). CXCL13 is a chemokine that recruits T and B lymphocytes to inflamed tissues, which can promote the formation of ectopic germinal centers (Hui et al., 2015; Weiss et al., 2016). Clec5a encodes C-type lectin domain family 5, member A (also known as myeloid DNAX activation protein 12-associating lectin-1), which is a cell surface receptor specifically expressed in TNF-α producing inflammatory macrophages (Gonzalez-Dominguez et al., 2015). Activation of CLEC5A, such as by direct binding of dengue virus, leads to TNF-α and NO production in myeloid cells, which induces lethal shock (Chen et al., 2008; Cheung et al., 2011). Several mutations in Pstpip1 have been implicated in the pathogenesis of autoinflammatory pyogenic arthritis, pyoderma gangrenosum, and acne (PAPA) (Smith et al., 2010; Wise et al., 2002; Zeeli et al., 2015), and the mutant proteins bind to components of the inflammasome, causing constitutive assembly and activation of the complex (Shoham et al., 2003; Yu et al., 2007). In summary, our analysis reveals a small cohort of highly inflammatory genes among those that are apparently controlled more effectively by a late Dex treatment (in the LPS-activated macrophages) in comparison to Dex given early.

GR activation globally inhibits genomic occupancy of NF-κB

To understand the gene regulatory outcomes in terms of molecular events occurring in chromatin, we mapped the occupancy of RelA, the transcriptionally active subunit of NF-κB classical dimer in BMDMs. The various treatment schedules were essentially the same as for RNA-seq, while the chromatin samples were collected at slightly earlier time points (0, 3, 8 hours after LPS) (Figure S3A). The binding patterns of NF-κB in response to LPS showed expected stimulus-dependent recruitment to promoters and distal binding sites (Figure 3A–C, Figure S3B). Peak binding of RelA was observed three hours after LPS (4,533 binding sites in total), presumably because the nuclear influx and binding of NF-κB is synchronous immediately after LPS stimulation. This is also consistent with the heterogeneous nuclear occupancy of NF-κB in the late stage of LPS response across the population at the single cell level (Lee et al., 2009; Sung et al., 2014b).

Figure 3. Dex treatments globally inhibit genomic occupancy of RelA subunit of NF-κB.

See Figure S3A for experimental scheme of BMDM treatment and chromatin sample collection. (A–C) Genome browser shots of RelA ChIP-seq on representative genes from the clusters: Tnf (cluster 2) (A), Ccr7 (cluster 3) (B), Sfpi1 (cluster 1) (C). Data tracks show ChIP density normalized for sequencing depth. Genomic coordinates in mm9. (D) Scatter plots of ChIP-seq density showing global loss of RelA binding in Dex-treated BMDMs. (E) Occurrences of RelA binding near genes grouped by the clusters from RNA-seq analysis. Violin plots show the number of 3h RelA ChIP-seq sites within 20 kb of TSS for genes in the indicated clusters or subsets. “LPS-induced_1” corresponds to the first cluster of genes from the top of heatmap in Figure 2A, etc. The thick box inside each violin indicates the top 25 percentile and bottom 25 percentile marks, and the median is labeled by a white dot. *, **, *** indicate statistically significant differences between the two groups with Welch two-sample t-test p values of 1.3×10⁻⁷, 5.9×10⁻¹¹, 2. ×10⁻¹⁴, respectively. See also Figures S3 and S4.

According to the widely discussed model of GR trans-repression, GR binds in close proximity or tethered to chromatin-bound NF-κB or AP-1 and interferes with their activity (Chinenov et al., 2012; Reichardt et al., 2001). The genome-wide occupancy of NF-κB would then remain largely unchanged after GR activation. Contrary to this expectation, the RelA ChIP-seq analysis of Dex-treated BMDMs showed a global reduction of NF-κB binding in the genome, regardless of the timing of Dex treatment (Figure 3D, Figure S3B). Moreover, the disruption of RelA binding was observed not only for Dex-repressed genes but also for Dex-insensitive genes.

The observed widespread decrease of NF-κB binding could be due to disruption of chromatin interaction or lower concentrations of NF-κB in the nucleus. To distinguish between the two possibilities, we performed live cell imaging of EGFP-RelA RAW264.7 reporter clonal cell line (Sung et al., 2014b) under identical treatment conditions. Applying previously developed quantitative image analyses, we generated single cell real time dynamics data and obtained similar nuclear RelA dynamics. There was only a slight reduction of nuclear RelA occupancy in Dex pre-treated cells (Figure S4 A–C). This is consistent with the most dramatic reduction of RelA ChIP-seq density from Dex pre-treatment of BMDM. However, there were little or no defect in terms of upstream signaling activities or nuclear RelA amounts in Dex-treated BMDMs as observed by immunoblot (Figure S4 D–E), which supports that signaling strength or nuclear abundance do not explain the lack of RelA chromatin occupancy in Dex treated BMDMs. A technical caveat that might have contributed to the reduced RelA ChIP-seq intensity is potential epitope inaccessibility in RelA when bound to GR. However, this seems unlikely, because the polyclonal antibody targets the C-terminus of RelA and the GR binding is through the N-terminal Rel homology domain (McKay and Cidlowski, 2000).

When the distributions of NF-κB binding sites near LPS-induced genes were analyzed for each gene cluster, the Dex-transrepressed gene cluster contained genes that were normally bound by NF-κB at fewer sites (Figure 3E, “LPS-induced_1”). The observation suggests that these genes are readily perturbed by GR, perhaps because of their critical dependence on the relatively small number of enhancers through which NF-κB exerts its regulatory actions. It is possible that the genes with numerous RelA sites may still be subject to NF-κB regulation even when GR disables a few of their NF-κB responsive enhancers.

Genomic occupancy of GR is greatly enhanced in LPS-activated macrophages, regardless of gene regulatory outcome

Given the shared and differential features of gene regulation from the differently timed Dex treatment, we asked whether they can be explained by GR binding patterns. GR ChIP-seq was performed for two samples: BMDMs treated with Dex for 3 hours (without LPS) and BMDMs activated by LPS with late Dex treatment (3 hours after LPS) (Figure S3A). The comparison of ChIP-seq profiles from Dex alone versus late-treatment after LPS revealed that the late treatment of Dex produced a dramatically enhanced binding of GR (Figure 4A–C), even though most Dex-sensitive genes were strongly repressed by the early Dex treatment. When GR was activated after LPS stimulation, it was able to bind to nearly eight thousand new sites in addition to about a thousand sites bound by GR when activated before LPS. The vast majority of GR sites from Dex alone were also bound by GR when it was activated after LPS (1,191/1,391) (Figure 4C and Figure S6).

Figure 4. GR binding is greatly enhanced in LPS-activated BMDMs.

(A–B) Genome browser shots of RelA and GR ChIP-seq on Serpine1 (A) and Ccr7 (B) loci. Boxed regions show highly increased GR binding co-localized with LPS-induced RelA binding. Data tracks show ChIP density normalized for sequencing depth. Genomic coordinates in mm9. (C) Comparison of the two GR binding repertoires in BMDMs in the absence or presence of LPS. The Venn diagram is drawn with areas proportional to the sizes of subsets shown. (D) Distance from de novo GR binding sites or from pre-existing (before LPS) GR sites to LPS 3h RelA ChIP-seq sites. The two sets of GR binding sites have significantly different distributions (Kolmogorov-Smirnov two-sample test p value = 2.2 × 10⁻¹⁶). (E) Occurrences of GR binding near genes grouped by the clusters from RNA-seq analysis. Violin plots show the number of post-LPS GR ChIP-seq sites within 20 kb of TSS for genes in the indicated clusters or subsets. The violin plots were generated in the same manner as in Figure 3E. The “Dex-regulated” set consists of genes which were expressed independently of LPS but differentially regulated by Dex. See also Figures S3, S5, and S6.

The result that GR binds to such a small number of sites (1,391 in the absence of LPS) with mostly weak ChIP signal three hours after Dex treatment was unexpected, because Dex is a ligand that binds and activates GR directly without going through any signal transduction pathways. This finding is in stark contrast to previous reports where strong GR binding is observed often within one hour in many cell types (John et al., 2011; Reddy et al., 2009). Since we obtained genome-wide GR ChIP-seq data at one time point following Dex treatment, we cannot exclude the possibility that peak GR binding is achieved at earlier times. However, that scenario seems unlikely because we have confirmed by ChIP-qPCR that no GR binding is detected for earlier time points after Dex treatment alone at a GR target locus Lcn2 in BMDMs (Figure S6A).

We then sought to understand how LPS-activated macrophages gain the additional GR binding sites. A motif discovery analysis was performed using the MEME suite of tools to identify features of DNA sequences at the sites that GR could not bind with Dex alone but required LPS for binding. A motif for PU.1, a transcription factor essential for myeloid differentiation, as well as the expected canonical GR response element (GRE) motif, was enriched in sites that were bound by GR regardless of LPS. In particular, a slightly degenerate form of this motif was found in virtually all GR sites that were gained after LPS stimulation. Our differential motif analysis showed that the GR binding sites which appeared after LPS stimulation were preferentially enriched for a NF-κB (RelA) motif (Figure S6B). These findings are consistent with the view that GR mainly targets the enhancers shaped by the lineage factor PU.1 in resting macrophages and this limited repertoire of GR binding sites is expanded through loading mechanisms assisted by LPS-activated factors such as NF-κB and AP-1 (Biddie et al., 2011; Miranda et al., 2013; Voss et al., 2011). RelA may not be solely responsible for direct recruitment of GR to these de novo sites, as only 17% of the de novo GR sites were co-localized with RelA binding sites (observed from ChIP-seq at 3h after LPS stimulation). However, the de novo GR sites were preferentially located in the vicinity of RelA binding sites (median distance: 35 kb, Figure 4D). These results suggest that GR binding might be enhanced after NF-κB and other LPS-activated factors alter the chromatinscape not only directly at potential GR binding sites but also possibly via a long-range mechanism involving higher-order structures.

Next we asked whether the GR binding sites found specifically near Dex-sensitive genes conformed to this globally enhanced occupancy or exhibit distinct behaviors that explain the effects of GR on these LPS-induced targets. To address this question, we employed two approaches. First, we compared the occurrences of GR binding sites near the gene classes that we obtained from RNA-seq (Figure 4E). Even though the first class (“LPS-induced_1”) was the most sensitive to Dex, these genes tend to harbor about the same number of, or even fewer, GR binding sites in comparison to those in the other classes. The second method was to look for a subtle relationship without grouping genes or sites using an algorithm. We integrated the Dex-related expression data values from RNA-seq and GR ChIP-seq signal intensity of nearby binding sites, and did not observe any relationship between GR binding and its gene regulatory effects (Figure S6C). For example, both Serpine1 and Ccr7 were more effectively suppressed by Dex pre-treatment versus late treatment, but showed minimal GR binding with Dex alone and greatly enhanced GR binding in BMDMs activated by LPS and subsequently treated with Dex (Figure 4 A, B). From these results, we conclude that the genomic distribution of GR binding sites does not explain the gene-specific regulatory activity of GR in macrophages.

GR activation results in modest local modulation of chromatin accessibility

The binding patterns described above underscore the idea that the genomic occupancy of NF-κB and GR may not fully report the changes impressed upon the macrophage epigenome by Dex and LPS. Thus we probed chromatin accessibility, a proxy for regulatory activity, by performing DNase-seq on BMDMs eight hours after LPS under the three conditions: without Dex, pre-treated with Dex, and late-treated with Dex. After confirming the reproducibility of DNase-seq intensity profiles across biological replicates (Figure S7A), we compared chromatin accessibility across the conditions for DHSs near LPS-induced genes in the clusters identified earlier. Klf6 belongs to the Dex-sensitive cluster 1 with numerous promoter-proximal and distal DHSs. Most of these DHSs showed reduced accessibility to some extent in Dex-treated BMDMs (Figure 5A, solid boxes). However, even the Dex-insensitive DHSs coincided with GR binding near Klf6 and many other loci (Figure 5A–D, dashed boxes), ruling out a simple correlation between GR occupancy and Dex effects upon chromatin. In fact, we found no globally noticeable Dex-dependent changes in chromatin accessibility at GR bound regions (Figure S7B). Moreover, relatively Dex-resistant genes Il12b (cluster 2) and Clec5a (cluster 3) also had both Dex-suppressed DHSs and Dex-insensitive DHSs (Figure 5C–D).

Figure 5. Chromatin accessibility near Dex-sensitive genes after Dex treatments.

(A–D) Genome browser shots of DNase-seq and RelA and GR ChIP-seq at Klf6 (A), Tnf (B), Lta (B), Ltb (B), Il12b (C), Clec5a (D). The gene names have been color-coded to indicate their gene cluster membership in Figure 2. Boxed regions show whether the DNase density is decreased in Dex-treated conditions, as indicated at the bottom of browser shots. Data tracks show ChIP or DNase density normalized for sequencing depth. Genomic coordinates in mm9. (E) DNase density in BMDMs pre-treated (left) or late-treated (right) with Dex versus DNase density from Dex untreated BMDMs (8h LPS alone). Colored points mark distal (> 1 kb from TSS) DHSs within 10 kb of LPS-induced genes. Logratios of RNA-seq FPKM values are color-coded as indicated in the scale bar. The ratios FPKM(LPS 10h) / FPKM(LPS 10h, Dex pre-treated) (left) and FPKM(LPS 10h) / FPKM(LPS 10h, Dex late-treated) (right) are shown to match the DNase conditions. DNase density data are representative of two independent experiments. See also Figures S5 and S7.

To determine whether there exists any link between gene regulation and chromatin behavior, we compared genome-wide DNase-seq intensity profiles of Dex pre-treated [or late-treated] versus Dex untreated BMDMs. Overall the chromatin accessibility was largely unchanged, especially for promoter DHSs (Figure 5E, black points). However, when (non-promoter) DHSs within 10kb of LPS-induced genes were labeled according to the Dex effects upon the expression of these genes, a subtle pattern emerged: DHSs near Dex-suppressed genes have reduced accessibility from Dex treatment (Figure 5E, red to orange points). On the other hand, distal DHSs near Dex-enhanced genes had increased accessibility from Dex treatment (Figure 5E, yellow to green points). Notably, this genome-wide pattern was shared by both Dex pre- and Dex late-treated BMDMs (Figure 6). These results indicate that, despite the variable distribution of DHSs and their heterogeneous behaviors in Dex-treated BMDMs, the regulatory sites near the strongly Dex-regulated genes tend to exhibit the corresponding regulation of chromatin accessibility in the same direction.

Figure 6. Chromatin accessibility is similarly modulated by Dex pre- and late-treatments.

(A) Genome browser shots of Ccl2-Ccl7-Ccl11-Ccl12-Ccl8 locus. (B) Genome browser shots of Glul and Wee1 loci. Data tracks show density normalized for sequencing depth. Genomic coordinates in mm9. DHSs whose intensity change after Dex treatment are marked with colored boxes as indicated. (C) Violin plots of Dex-specific changes in DNase-seq intensity at DHSs within 10 kb of genes in the indicated groups. White dots mark median values. Thick bars end at 25 percentile and 75 percentile values. Dex-induced genes are defined as those whose expression is at least 2-fold increased in Dex pre-treated BMDMs (LPS 0h or LPS 4h). Asterisks mark statistically significant difference of the group with respect to any of the other two groups (Welch two-sample t-test p < 2.2×10⁻¹⁶). See also Figure S7.

GR trans-activates inhibitors of inflammatory factors

The lack of correlation between GR binding and chromatin accessibility (Figure 5A–D and Figure S7B) motivated us to consider secondary non-chromatin mechanisms which may contribute to the anti-inflammatory effects of GC. In addition to suppressing numerous inflammation-associated genes, GR transcriptionally activates many genes which encode negative regulators of inflammatory factors (Hubner et al., 2015). Among the highly Dex-induced genes from our RNA-seq analysis were such examples as Nfkbia (encoding IκBα), Tnfaip3 (encoding A20), Dusp1 and Dusp6 (encoding dual specificity phosphatases, DUSPs), Tsd22d3 (encoding glucocorticoid-induced leucine zipper, GILZ). The negative feedback regulator IκBα of NF-κB binds directly to NF-κB dimers and depletes them from the nucleus by constant nucleocytoplasmic shuttling and preferential localization in the cytoplasm. A20 is a dual-function ubiquitin-editing enzyme. On the one hand, its de-ubiquitinase domain removes activating K63-linked ubiquitin chains from signaling adaptor proteins such as TRAF6 and RIP1, upstream of the Inhibitor of κB kinase (IKK) complex. On the other hand, its E3 ligase domain adds the degradation-inducing K48-linked ubiquitin chains. Thus, A20 potently suppresses NF-κB activity by blocking the ubiquitin-mediated signaling and by promoting the degradation of key factors (Wertz et al., 2004). DUSPs counteract upon mitogen-activated protein kinase (MAPK) signaling cascades, thereby limiting AP-1 action (Bhattacharyya et al., 2007). GILZ can bind NF-κB directly and inhibit its activity (Ayroldi et al., 2001). Most of the genes encoding these factors are known direct targets of GR. Indeed, our GR ChIP-seq analysis of BMDMs reveals that these genes have binding sites which GR can efficiently target even without the chromatin alterations induced by LPS (Figure 7), in contrast to the vast majority of GR binding sites which show attenuated occupancy without LPS. Altogether, these data suggest that GR transcriptionally activates powerful negative regulators which target several distinct nodes within the inflammatory signaling cascade, regardless of the changing chromatinscape (over the course of an inflammatory challenge).

Figure 7. Binding of GR to anti-inflammatory target genes does not require LPS stimulation.

(A–D) Genome browser shots of GR and RelA ChIP-seq at genes encoding negative regulators of NF-κB and AP-1 signaling: Nfkbia, Tnfaip3, Dusp1, Tsc22d3. Data tracks show ChIP density normalized for sequencing depth. Genomic coordinates in mm9.

Discussion

We have characterized the transcriptomic and epigenomic effects of a late GC treatment by contrasting them against a GC pre-treatment of LPS-stimulated macrophages. It was important to examine the late GC effects because clinical GCs are administered to inflamed tissues as anti-inflammatory interventions. Because a late GC treatment makes GR act upon an inflammatory epigenome, we hypothesized that GR binding may be altered in comparison to a GC pre-treatment where GR encounters a basal chromatin. The binding repertoire of GR has previously been reported to change in HeLa cells in response to stimulation with TNF-α (Rao et al., 2011) and in BMDMs after LPS stimulation (Uhlenhaut et al., 2013). However, these studies did not examine the immediate genomic binding of GR in the activated versus resting cells, as in this study. In particular, the BMDM study used an overnight pre-treatment of Dex (Uhlenhaut et al., 2013), which makes it difficult to draw direct comparisons to our data on GR binding immediately after Dex activation.

We have characterized the primary GR regulome in cells pre-activated with an inflammatory stimulus. GR can only bind to a surprisingly small number of primary target sites in the basal macrophage epigenome and that this limited repertoire dramatically expands if GR is ligand-activated into the ‘inflamed chromatinscape’. An increase of GR occupancy by Dex and LPS co-treatment (versus Dex alone) has been reported previously on a few target sites in BMDMs (Chinenov et al., 2012). We found that this enlarged GR binding repertoire did not correlate with chromatin modulation or gene regulation. In particular, the enhanced occupancy of GR in LPS-stimulated BMDMs did not correspond to more potent gene repression. Our result argues against the proposed mechanisms of GR gene repression which involve GR tethering to DNA-bound NF-κB (Reichardt et al., 2001), direct binding to negative GR responsive elements (nGREs) (Surjit et al., 2011) or recruitment of co-repressors such as GR-interacting protein (GRIP)1 (Chinenov et al., 2012; Uhlenhaut et al., 2013). It is possible that these processes may occur at some genomic loci in some contexts. However, they are unlikely to be prevalent mechanisms, because a Dex pre-treatment results in very limited GR genomic occupancy in macrophages (which would be expected from either situation) but produces a gene repression program as efficient as a late treatment.

In contrast to GR binding, the genomic occupancy pattern of NF-κB subunit RelA correlated with the observed gene repression in Dex-treated BMDMs. RelA binding was almost entirely abolished in Dex pre-treated cells where GR translocated into the nucleus before NF-κB. In contrast, a previous study using a higher concentration of Dex and an overnight pre-treatment reported a large number of RelA sites in BMDMs (Uhlenhaut et al., 2013). Since the RelA ChIP intensity from LPS alone was not included in the same experiment, it is difficult to draw any conclusion about Dex-specific changes in RelA binding from that study. According to our direct comparison of NF-κB binding time courses with or without Dex, prevention of RelA binding might be a more likely mechanism for the reduced NF-κB genomic occupancy than an active displacement by GR, although proof is beyond the scope of this study. We should note that prevention of NF-κB (re-)binding to chromatin can produce the observed reduction of occupancy even in Dex late-treated BMDMs, because occupancy measured by ChIP is a snapshot from dynamic transcription factor interactions with chromatin (Hager et al., 2009; Poorey et al., 2013; Sung et al., 2014a; Voss et al., 2011). The disruption of chromatin-bound RelA by GR is not inconceivable, though, since GR has been reported to bind the DNA binding Rel homology domain of RelA directly (McKay and Cidlowski, 2000). It will be interesting in the future to determine whether the residual ChIP-seq signal in Dex late-treated BMDMs represents a long-lasting NF-κB occupancy and whether GR can indeed displace RelA from chromatin in living cells. We have considered the possibility that an increased amount of IκBα protein induced by GR in Dex pre-treated cells may block the nuclear translocation of NF-κB. This has been ruled out from likely scenarios, because our live cell imaging analysis of the fluorescent reporter macrophages shows that the nuclear influx of RelA immediately following LPS stimulation is unaffected by Dex pre-treatment.

The differentially timed Dex treatments allowed us to assess whether GR and NF-κB are in a ‘race for regulation’, wherein factors with antagonistic gene programming functions compete for chromatin occupancy and whichever factor arrives at target chromatin first exerts dominant regulatory action. This idea was not supported by our data. Dex pre-treatment activated GR before inflammatory factors, but the small number of GR binding sites before LPS stimulation could not possibly cover all genomic sites that would later be occupied by NF-κB. Moreover, in Dex late-treated cells, GR translocated to the nucleus and encountered chromatinscape already interacting with inflammatory transcription factors, and yet GR could now bind to even more sites. These gained GR binding sites had no particular relationship with respect to the lost RelA binding sites, as virtually all RelA binding sites in the genome were disengaged by Dex treatment. Taken together, these data argue against competition of GR and NF-κB for chromatin binding.

Our data suggest that the so-called “dissociated ligands” (Gerstein et al., 2012) of GR may not produce the desired therapeutic properties. The rationale behind the search for dissociated ligands was to selectively disable the transactivation by the GR dimer (which was thought to induce side effects of GCs) but to retain the GR monomer tethering to inflammatory transcription factors. First, recent data have largely dismissed the dichotomy of molecular actions between GR monomers versus dimers (Presman et al., 2014). In fact, a tetramer form of DNA-bound GR has been observed in live cells (Presman et al., 2016). There is currently little evidence that the anti-inflammatory action of GR comes mostly from the monomer form. Importantly, the ligands that weaken the ability of GR to activate transcription of target genes are likely to impair the induction of the potent inhibitors (such as IκBα and DUSPs) of inflammatory pathways, thereby producing significantly diminished anti-inflammatory effects. Therefore, the function of GR as a transcriptional activator is probably as important for the anti-inflammatory effects of GCs as the widely invoked trans-suppression of NF-κB and AP-1 target genes. Our findings shed light on the multi-faceted GR actions upon the ‘inflamed epigenome’ which underlie clinically relevant GC interventions.

STAR Methods

Contact for reagent and resource sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Myong-Hee Sung (sungm@mail.nih.gov).

Experimental model and subject details

Primary cell cultures

Mice were maintained in specific-pathogen-free conditions and all procedures were performed according to guidelines and study protocols approved by the Institutional Animal Care and Use Committee, National Institute of Allergy and Infectious Diseases, National Institutes of Health (NIH). BMDMs from wild type C57BL/6 mice were prepared by in vitro differentiation for 6 days in complete Dulbecco’s modified Eagle’s medium (DMEM, 10% (v/v) fetal bovine serum (FBS)) at 37°C and 5% CO₂ supplemented with 60 ng/ml recombinant mouse M-CSF (R&D). Cells were treated at various time points with 100 nM Dex (Sigma) and/or 10 ng/ml LPS (Alexis Biochemicals, Salmonella minnesota R595 TLRgrade, ALX-581-008-L002). For each assay, cells were preincubated in Hank’s balanced salt solution containing 2% FBS (v/v) at 37°C and 5% CO₂ for 2–3 hours before the application of LPS.

Method details

RNA isolation and real time quantitative RT-PCR

Total RNA was isolated from approximately 1x10⁶ cells by using an RNAeasy Mini Kit (Qiagen). Real-time qRT-PCR was performed as previously described (Hakim et al., 2013). cDNA was synthesized from the extracted 300ng RNA with the iScript cDNA Synthesis Kit (Bio-Rad, 170-8891). qRT-PCR assays were performed with IQ SYBR Green PCR Supermix with an ABI 7900 HT, according to the manufacturer’s protocol. Gene expression in each sample was normalized to that of β-actin and fold-changes relative to basal expression were calculated with the 2^−ΔΔCt method. Each experiment was performed with three independent repeats. Error bars represent SD of the biological replicates. The following forward and reverse primers spanning consecutive exons were designed to amplify the transcript.

Gene	Forward	Reverse
Cx3cr1	CATCTTCCTGTCCGTCTTCTAC	GGTAGATGTCAGTGATGCTCTT
Selp	TCTGGTTCAGTGCTTTGATCTC	ACAGAACACCCGTGAGTTATTC
Cd55	GTCGAGCCACGAAACATTCT	CCACAATAGTACCAACTGGGAAA
Cxcl13	TGTTGTCGGTCTAAACATCATAGA	GGCACGAGGATTCACACATA
Clec5a	TCGCTGCACCGAATATCTTATC	CGAGATGATCATGTGCCAGTT
Nos2	TGTTCTCAGCCCAACAATACA	GTCCAGGGATTCTGGAACATT
Ccl24	CTTGCTGCACGTCCTTTATTT	TATGGCCCTTCTTGGTGATG
Mmp13	GTGCCTGATGTGGGTGAATA	CAGAATGGGACATATCAGGAGTATAG
Clec7a	TCAGCACTCAAGACATCCATAAA	AGCAACCACTACTACCACAAAG
Il12b	CATCTGCTGCTCCACAAGAA	TGAACCGTCCGGAGTAATTTG
Beta-actin	GCTTCTAGGCGGACTGTTACTGA	GCGCAAGTTAGGTTTTGTCAAA

RNA-seq

RNA was isolated as above from each sample treated with LPS and/or Dex according to the experimental design. Sequencing libraries were generated using Illumina TruSeq V3 protocol and subject to 101 bp paired-end sequencing on Illumina HiSeq2000. Sample quality, library complexity, and alignment statistics were checked using an established pipeline at the NCI Center for Cancer Research Sequencing Facility. All RNA-seq analyses have been performed with two biological replicates. The sequencing reads were aligned against the reference mouse genome mm9 and ensemble v70 transcripts using TopHat (v. 2.0.8). The alignment statistics were greater than 93% for all samples. All short reads were assembled with Cufflinks (v. 2.0.2) with -G option to quantitate against annotated reference transcripts.

FPKM data were imported into R for further analyses. All computational analyses were performed on log2 (FPKM + 0.1) values. Transcripts shorter than 200 bp were removed before subsequent analyses. edgeR was used to calculate false discovery rate (FDR) and assess differential gene expression based on FDR < 0.05. LPS-induced genes were defined as the transcripts whose replicate-averaged expression increases by at least 2-fold in at least one time point after LPS stimulation versus basal. For cluster analysis, partitions around the medoids (PAM) algorithm was used after gene-specific centering (to normalize with respect to the basal). Clustering was applied to the average of the biological duplicate expression data for 0h, 4h, 10h LPS, and Dex-pretreated 0h, 4h, 10h LPS samples. Different numbers of clusters were tested to determine the final number of clusters which represent major patterns in the data. Cluster heatmaps were generated displaying all the RNA-seq duplicate data. Genes differentially expressed in the Dex pre- versus late-treated BMDMs (at 10h LPS) were obtained by requiring at least 2-fold difference in the replicate-averaged expression. Gene ontology analysis was performed using the web analysis tool at http://central.biomart.org/enrichment with MGI IDs. Top enriched ontology classes were obtained using the tool VLAD v1.6.0 at http://proto.informatics.jax.org/prototypes/vlad/.

Chromatin immunoprecipitation (ChIP)

Cells were treated with 10 ng/ml LPS and/or 100 nM Dex for appropriate durations as specified in the experimental design. ChIP assays were performed as described previously (Miranda et al., 2013) with modifications. The following antibodies were used: a mouse monoclonal antibody against GR (MA1-510, Thermo Fisher Scientific) and a rabbit polyclonal anti-RelA antibody (Ab7970, Abcam). For RelA ChIP, BMDMs were subject to crosslinking with 1% formaldehyde for 10 minutes at room temperature. For GR ChIP, we used a dual crosslinking method (Uhlenhaut et al., 2013) to detect both tethered and directly bound chromatin. BMDMs were subject to crosslinking with 2mM disuccinimidyl glutarate (DSG) for 30 min followed by 1% formaldehyde for 10 min at room temperature. After crosslinking, chromatin was isolated from about 4 × 10⁷ cells, which allows 2 to 3 different immunoprecipitations per sample at a time, and then processed as follows. The lysis buffer to shear the chromatin contained 0.5% SDS, 10 mM EDTA (pH 8), 50 mM tris-HCl (pH 8), and proteinase inhibitor cocktail. Sonication was performed to shear the chromatin to generate DNA fragments with a size range of 400 to 500 base pairs. The sheared chromatin samples were diluted 1:5 in dilution buffer (0.01% SDS, 1.1% Triton-X, 1.2 mM EDTA (pH 8), 20 mM tris-HCl (pH 8), 167 mM NaCl, and proteinase inhibitor cocktail). For each ChIP, 200 μg chromatin DNA was precipitated with antibody coated bead complexes (Dynabeads Protein A for RelA ChIP; anti-IgG paramagnetic beads for GR ChIP, Invitrogen).

For ChIP-seq, sequencing libraries were generated using Illumina TruSeq V3 protocol and subject to 51 bp single-end sequencing on Illumina HiSeq2000. Sample quality and library complexity were checked using an established pipeline at the NCI Center for Cancer Research Sequencing Facility. All ChIP-seq analyses have been performed with two or three biological replicates. The ChIP-qPCR assays were performed using biological triplicates. The ChIP products were subjected to qPCR analysis with an IQ SYBR Green Supermix and a MyIQ singlecolor, real-time PCR detection system (Bio-Rad) according to the manufacturers’ instructions. DNA mixtures purified from aliquots of each chromatin sample were also subjected to qPCR analysis as input samples, and the results were presented as the ChIP-qPCR measurements normalized to their respective input measurements. The ChIP-qPCR primer sequences are shown below.

Gene	Forward	Reverse
Lcn2	TCACCCTGTGCCAGGACCAA	TGGGGAAGGGTGAGCAAGCT
Per1	GGGACCCCCTTCCTCCTAAC	AGCGCACTAGGGAACATCGT
Tnf	AGAGGCTGAGACATAGGCAC	CTTTCACTCACTGGCCCAAG
Rela	GTGGGAGGGGCGTAACTATT	CCACTATGCCAGAAGGAGGA
Il6, promoter	TCTTTCCCATCAAGACATGCTC	ATGTGACGTCGTTTAGCATCG
Il6, 20kb upstream	TGGACAAGATAGAAGGTGGAAGT	TGTGTCTTGGGCTCTGTAGA
Il12b	GTCCCTCCAGTTTCACCAGT	TACCCCACTGTCATGATGCA
Dusp1	AGTTCCAAAAGAAAGGGGAAAG	CATTGATGTCAAGGCTAGCAAA
Negative control	AGGCCTGCACTACCAAACAC	TAATGCCCTTGCAGAAGACC

DNase-seq

DNase-seq was performed as previously described (John et al., 2011; Morris et al., 2014). After treatment with Dex and/or LPS, nuclei of BMDMs were isolated, and chromatin were digested with 80 U/ml DNase I (Sigma) for 3 min at 37 °C and incubated at 55 °C overnight with 10 μg/ml of RNaseA (Invitrogen) and Proteinase K (Ambion). Digested DNA fragments were purified using phenol-chloroform and enriched by size selection over sucrose gradient. DNA fragments between 100 bp and 500 bp were precipitated and dissolved in nuclease free H₂O. Two biological replicate samples per treatment group were prepared separately for high throughput sequencing. Sequencing libraries were generated using Illumina TruSeq V3 protocol and subject to 51 bp single-end sequencing on Illumina HiSeq2000. Sample quality and library complexity were checked using an established pipeline at the NCI Center for Cancer Research Sequencing Facility. All DNase-seq assays have been performed with two or three biological replicates.

Sequencing data analysis

All sequencing data files were processed and reads were aligned to the mm9 reference genome, using a standardized procedure established for Illumina HiSeq2000 by the NCI Center Cancer Research Sequencing Facility, NIH. The uniquely mapped reads were analyzed to identify regions termed “hotspots” using DNase2Hotspots (http://sourceforge.net/projects/dnase2hotspots)(Baek and Sung, 2016; Baek et al., 2012). All the ChIP-seq and DNase-seq hotspots were compiled into a master set of genomic regions for integrative analysis. The region boundaries were determined as the union of all overlapping hotspots. For quantifying the signal intensity (termed density) of ChIP-seq or DNase-seq over each region in the master set, we used “AveD”, the count of overlapping fragments averaged over the region (Baek and Sung, 2016; Baek et al., 2012). For the GR dual crosslinking ChIP-seq data, we applied a density-based threshold to remove small noisy hotspots before calculating the number of binding sites. The midpoint between the two Gaussian distributions of maximum density values was chosen to be the threshold. For calculating the distances between GR and RelA binding sites, GR hotspots were AveD-thresholded at 50 percentile observed for the late Dex treatment, and RelA were AveD-thresholded at 50 percentile observed for 3h LPS. The density-based thresholding removed small low-confidence hotspots that may be unreliable.

Area-proportional Venn diagrams showing factor binding sites were drawn using EulerAPE v3 available at http://www.eulerdiagrams.org/eulerAPE (Micallef and Rodgers, 2014).

Motif discovery analysis was performed using MEME and Tomtom (Bailey et al., 2015). For querying against a database of known TF binding specificity, we used CisBP (Weirauch et al., 2014). The number of ChIP sites for each enrichment analysis was restricted to 1000 for GR ChIP-seq sites gained after LPS stimulation. For constant GR ChIP sites, all 1191 sites were used. Motif discovery was also performed with the small number of GR binding sites (193) only observed in BMDMs without LPS (treated with Dex alone), but no good matches to known motifs were identified.

Live cell imaging and quantification

Time lapse microscopy of live RAW264.7 reporter clone was performed as described (Sung et al., 2014b) using a Zeiss LSM 880 equipped with a controlled incubation chamber (37° C, humidified 5% CO₂). Cells were treated with LPS with or without Dex pre- or late-treatment onstage through separate injection tubing for LPS and Dex. The timing of Dex treatments was identical to that for the ChIP-seq experimental design. Time lapse images were acquired at 8 minute intervals with two sequential frames for EGFP and mCherry for each stage position with 0.3% of power for both 488nm and 561nm laser lines, using Plan-Apochromat 40× oil objective (1.4 NA), zoom 1, 512 by 512 pixels, no averaging, maximally open pinhole, “Definite Focus” autofocus, and the GaAsP spectral detector. The 16 bit raw TIFF imaging data were analyzed and quantified using previously developed custom MATLAB codes (Sung et al., 2014b) with slight modifications to match the current acquisition parameters.

Western blotting

Whole-cell extracts and nuclear extracts from 1×10⁷ BMDM cells per condition were prepared in RIPA cell lysis buffer (Millipore, 20-188) and NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Sci 78833), respectively. 40 μg of protein were denatured with 2× SDS-PAGE sample buffer (Quality Biological, 351-082-661) at 85 °C for 5 minutes. 25 μl of protein was resolved with an 8 to 16% Tris-Glycine SDS Gel (Invitrogen, XP08162)/Tris-Glycine Running Buffer System (Invitrogen, LC2675) and transferred to nitrocellulose membranes. The membranes were analyzed by Western blotting according to a standard protocol. The antibodies used are as follows: rabbit anti-p65 (Santa Cruz, sc-372), rabbit anti-IκBα (Santa Cruz, sc-371), rabbit anti-STAT3 (Santa Cruz, sc-482), rabbit anti-p38 MAPK (Cell Signaling, 9212L), rabbit anti-phospho p38 MAPK (Cell Signaling, 4511S), rabbit anti-ERK2 (Santa Cruz, sc-154), rabbit anti–Rho-GDI (Sigma, R3025), rabbit anti-phospho ERK1/2 (Cell Signaling, 9101), rabbit anti-hnRNP (Santa Cruz, sc-28726), and HRP-conjugated ECL anti-rabbit immunoglobulin G (GE Healthcare).

Quantification and statistical analysis

Software, quantification, and computational methods are described in the relevant method sections above.

Data and software availability

All data have been deposited in the GEO database at https://www.ncbi.nlm.nih.gov/geo/ in the SuperSeries with accession number GSE93739.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-p65	Santa Cruz Biotechnology	Cat# sc-372; RRID: AB_632037
Rabbit polyclonal anti-IκBα	Santa Cruz Biotechnology	Ca# sc-371; RRID: AB_2235952
Rabbit polyclonal anti-STAT3	Santa Cruz Biotechnology	Cat# sc-482 RRID: AB_632440
Rabbit polyclonal anti-p38 MAPK	Cell Signaling Technology	Cat# 9212L; RRID: AB_330713
Rabbit monoclonal anti-phospho p38 MAPK	Cell Signaling Technology	Cat# 4511S;RRID: AB_2139682
Rabbit polyclonal anti-ERK2	Santa Cruz Biotechnology	Cat# sc-154; RRID: AB_2141292
Rabbit polyclonal anti–Rho-GDI	Sigma-Aldrich	Cat# R3025 RRID: AB_261317
Rabbit anti-phospho ERK1/2	Cell Signaling Technology	Cat# 9101, RRID: AB_331646
Rabbit polyclonal anti-hnRNPL	Santa Cruz Biotechnology	Cat# sc-28726 RRID: AB_2264302
Donkey anti-Rabbit IgG	GE Healthcare	Cat# NA934; RRID: AB_772206
Mouse monoclonal anti-GR	Thermo Fisher Scientific	Cat# MA1-510 RRID: AB_325427
Rabbit polyclonal anti-NFkB p65	Abcam	Cat# ab7970 RRID: AB_306184
Bacterial and Virus Strains
N/A
Biological Samples
N/A
Chemicals, Peptides, and Recombinant Proteins
LPS from Salmonella Minnesota R595	Enzo Life Sciences, Inc	Cat# ALX-581-008-L002
Recombinant mouse MCSF protein	R & D	Cat# 416-ML
DNaseI	Sigma-Aldrich	Cat# D4527
37% Formaldehyde	Sigma-Aldrich	Cat# F1635
Disuccinimidyl Glutarate (DSG)	Thermo Fisher Scientific	Cat# 20593
Dexamethasone	Sigma	Cat# D4902
Critical Commercial Assays
RNAeasy Mini Kit	Qiagen	Cat# 74106
iScript cDNA Synthesis Kit	Bio-Rad	Cat# 170-8891
IQ SYBR PCR Supermix	Bio-Rad	Cat# 170-8886
Deposited Data
All RNA-seq, ChIP-seq, and DNase-seq data	GEO	GSE93739
Experimental Models: Cell Lines
RAW264.7 dual reporter clone	Iain Fraser	N/A
Experimental Models: Organisms/Strains
Mouse: C57BL/6	Jackson Laboratory
Oligonucleotides
Primers for Real-time qRT-PCR
Cx3cr1 Forward: CATCTTCCTGTCCGTCTTCTAC Reverse: GGTAGATGTCAGTGATGCTCTT	This paper
Selp Forward: TCTGGTTCAGTGCTTTGATCTC Reverse:ACAGAACACCCGTGAGTTATTC	This paper
Cd55 Forward: GTCGAGCCACGAAACATTCT Reverse: CCACAATAGTACCAACTGGGAAA	This paper
Cxcl13 Forward:TGTTGTCGGTCTAAACATCATAGA Reverse: GGCACGAGGATTCACACATA	This paper
Clec5a Forward TCGCTGCACCGAATATCTTATC Reverse CGAGATGATCATGTGCCAGTT	This paper
Nos2 Forward TGTTCTCAGCCCAACAATACA Reverse TCCAGGGATTCTGGAACATT	This paper
Ccl24 Forward CTTGCTGCACGTCCTTTATTT Reverse TATGGCCCTTCTTGGTGATG	This paper
Mmp13 Forward: GTGCCTGATGTGGGTGAATA Reverse :CAGAATGGGACATATCAGGAGTATAG	This paper
Clec7a Forward: TCAGCACTCAAGACATCCATAAA Reverse: AGCAACCACTACTACCACAAAG	This paper
Il12b Forward: CATCTGCTGCTCCACAAGAA Reverse: TGAACCGTCCGGAGTAATTTG	This paper
Beta actin Forward: GCTTCTAGGCGGACTGTTACTGA Reverse: GCGCAAGTTAGGTTTTGTCAAA	This paper
Primers for ChIP qPCR
Lcn2 Forward: TCACCCTGTGCCAGGACCAA Reverse: TGGGGAAGGGTGAGCAAGCT	This paper
Per1 Forward: GGGACCCCCTTCCTCCTAAC Reverse: AGCGCACTAGGGAACATCGT	This paper
Tnf Forward: AGAGGCTGAGACATAGGCAC Reverse: CTTTCACTCACTGGCCCAAG	Sung et al., 2014b
Rela Forward: GTGGGAGGGGCGTAACTATT Reverse: CCACTATGCCAGAAGGAGGA	Sung et al., 2014b
Il6,promoter Forward :TCTTTCCCATCAAGACATGCTC Reverse: ATGTGACGTCGTTTAGCATCG	This paper
Il6, 20kb upstream Forward: TGGACAAGATAGAAGGTGGAAGT Reverse: TGTGTCTTGGGCTCTGTAGA	This paper
Il12b Forward: GTCCCTCCAGTTTCACCAGT Reverse: TACCCCACTGTCATGATGCA	This paper
Dusp1 Forward: AGTTCCAAAAGAAAGGGGAAAG Reverse: CATTGATGTCAAGGCTAGCAAA	This paper
Negative Control Forward: AGGCCTGCACTACCAAACAC Reverse: TAATGCCCTTGCAGAAGACC	Sung et al., 2014b
Recombinant DNA
Software and Algorithms
TopHat v. 2.0.8
Cufflinks v. 2.0.2
R, various versions in 2014–2017	The R Project	https://www.r-project.org/
Bioconductor, various versions in 2014–2017	Bioconductor developers	https://www.bioconductor.org/
Biomart enrichment analysis	Biomart	http://central.biomart.org/enrichment
VLAD v. 1.6.0	The Jackson Laboratory	http://proto.informatics.jax.org/prototypes/vlad
DNase2Hotspots	Songjoon Baek and Myong-Hee Sung	http://sourceforge.net/projects/dnase2hotspots
EulerAPE v.3	Micallef and Rodgers, 2014	http://www.eulerdiagrams.org/eulerAPE
MEME suite	Bailey et al., 2015	http://meme.nbcr.net
CisBP database	Weirauch et al., 2014	http://cisbp.ccbr.utoronto.ca
MATLAB	Mathworks	N/A

Supplementary Material

3. Supplemental Excel Table S1. The list and description of genes that are significantly induced by LPS.

The ‘class’ column indicates cluster membership regarding the three clusters in Figure 2. (Related to Figure 2)

4

Supplemental Excel Table S2. The list and description of genes in the “lower in late Dex” group. (Related to Figure 2)

5

Supplemental Excel Table S3. The list and description of genes in the “lower in early Dex” group. (Related to Figure 2)

6

Supplemental Excel Table S4. Ontology enrichment analysis on the “lower in late Dex” genes. (Related to Figure 2)

7

Supplemental Excel Table S5. Ontology enrichment analysis on the “lower in early Dex” genes. (Related to Figure 2)

Highlights.

• We delineated the effects of glucocorticoids given after an inflammatory signal.

• GR inhibits chromatin occupancy, but not nuclear residence, of NF-κB.

• GR binding has modest effects on chromatin accessibility in macrophages.

• GR activates negative regulators of NF-κB and AP-1 in resting or LPS-induced cells.