Coactivation of GR and NFKB alters the repertoire of their binding sites and target genes

Abstract

Glucocorticoid receptor (GR) exerts anti-inflammatory action in part by antagonizing proinflammatory transcription factors such as the nuclear factor kappa-b (NFKB). Here, we assess the crosstalk of activated GR and RELA (p65, major NFKB component) by global identification of their binding sites and target genes. We show that coactivation of GR and p65 alters the repertoire of regulated genes and results in their association with novel sites in a mutually dependent manner. These novel sites predominantly cluster with p65 target genes that are antagonized by activated GR and vice versa. Our data show that coactivation of GR and NFKB alters signaling pathways that are regulated by each factor separately and provide insight into the networks underlying the GR and NFKB crosstalk.

Full text

Research
Coactivation of GR and NFKB alters the repertoire
of their binding sites and target genes
Nagesha A.S. Rao,
1,4
Melysia T. McCalman,
1,4
Panagiotis Moulos,
2,4
Kees-Jan Francoijs,
1
Aristotelis Chatziioannou,
2
Fragiskos N. Kolisis,
2
Michael N. Alexis,
3
Dimitra J. Mitsiou,
3,5
and Hendrik G. Stunnenberg
1,5
1
Department of Molecular Biology, Radboud University, 6500 HB Nijmegen, The Netherlands;
2
Metabolic Engineering and
Bioinformatics Group, Institute of Biological Research and Biotechnology, National Hellenic Research Foundation, 11635 Athens,
Greece;
3
Molecular Endocrinology Programme, Institute of Biological Research and Biotechnology, National Hellenic Research
Foundation, 11635 Athens, Greece
Glucocorticoid receptor (GR) exerts anti-inflammatory action in part by antagonizing proinflammatory transcription
factors such as the nuclear factor kappa-b (NFKB). Here, we assess the crosstalk of activated GR and RELA (p65, major
NFKB component) by global identification of their binding sites and target genes. We show that coactivation of GR and
p65 alters the repertoire of regulated genes and results in their association with novel sites in a mutually dependent
manner. These novel sites predominantly cluster with p65 target genes that are antagonized by activated GR and vice
versa. Our data show that coactivation of GR and NFKB alters signaling pathways that are regulated by each factor
separately and provide insight into the networks underlying the GR and NFKB crosstalk.
[Supplemental material is available for this article.]
Glucocorticoids (GCs) are essential steroid hormones that regulate
a variety of physiological processes including development, apo-
ptosis, metabolism, and homeostasis. The biological actions of
GCs are mediated through the ubiquitously expressed glucocor-
ticoid receptor (GR), a ligand-activated transcription factor that
belongs to the nuclear receptor superfamily. Unliganded GR re-
sides in the cytoplasm as an inactive complex that dissociates upon
hormone binding, and activated GR translocates to the nucleus to
regulate transcription of its target genes (Schaaf and Cidlowski
2002; Pratt and Toft 2003; Nicolaides et al. 2010). GCs exert es-
sential immunosuppressive and anti-inflammatory actions and
have been widely used as drugs to treat immune and inflammatory
disorders. Using single-gene approaches, several modes of action
for GC’s anti-inflammatory properties have been proposed (Necela
and Cidlowski 2004; De Bosscher and Haegeman 2009; Coutinho
and Chapman 2010). Direct binding of activated GR on GREs
(glucocorticoid responsive elements; classical model) and in-
teraction with NFKB and AP1 (nonclassical model) are the main
mechanisms of regulation associated with glucocorticoid-mediated
transactivation and transrepression (Yamamoto 1985; Konig et al.
1992; Beato et al. 1995; Gottlicher et al. 1998; De Bosscher et al.
2003; Necela and Cidlowski 2004).
NFKB is a family of constitutively expressed transcription
factors that impact many biological processes such as cell growth,
proliferation, development, and inflammatory and immune re-
sponses. Inactive NFKB, a dimer of the p50 and p65 (or other family
members), remains in the cytosol due to its association with the
inhibitory proteins of the NFKBI family (NFKBIA) (Vallabhapurapu
and Karin 2009). In response to diverse internal and external in-
flammatory stimuli, such as the proinflammatory cytokine tumor
necrosis factor alpha (TNF), NFKBIA is phosphorylated and rapidly
degraded, releasing the NFKB dimer, which translocates to the
nucleus. Activated NFKB binds to kB response elements (NFKB RE)
and regulates expression of genes encoding various proteins such
as proinflammatory cytokines, chemokines, receptors, and adhe-
sion molecules (Barnes 1997; Barnes and Karin 1997; Baeuerle
1998; Hayden and Ghosh 2008).
Given that NFKB is a key mediator of immune and in-
flammatory responses and that GR exerts anti-inflammatory func-
tions, the crosstalk between GR and NFKB signaling is of particular
importance and has been the major focus of research for many years
(Van Bogaert et al. 2010). The most extensively studied case
has been the transrepression of NFKB and AP1 by GR. The in-
hibitory effect of GR is postulated to be largely due to recruitment
of GR, via protein–protein interaction by DNA-bound NFKB or AP1
(tethering model) (Jonat et al. 1990; Cato and Wade 1996; McEwan
et al. 1997; Karin 1998; De Bosscher et al. 2003). GR and p65 or
JUN, an AP1 subunit, are suggested to physically interact and
mutually antagonize each other’s transcriptional activity (Konig
et al. 1992; Ray and Prefontaine 1994; Gottlicher et al. 1998;
Adcock et al. 1999). In addition, other mechanisms have been
suggested for the anti-inflammatory effects of GR, such as modu-
lation of chromatin environment (Ito et al. 2000; Tsaprouni et al.
2002; Beck et al. 2008) and competition for a limiting amount of
cofactors, such as the acetyltransferases CREBBP and EP300 (Kamei
et al. 1996). In principle, multiple layers of regulation seem to be
involved in the crosstalk of GR and NFKB or AP1; however, the
mechanism and extent of crosstalk has remained unresolved.
In the present study we set out to decipher the global GR and
NFKB interaction on chromatin and to identify their targets genes.
We mapped theGR- and p65-binding sites on a genome-wide scale
upon activation of GR and NFKB separately or upon coactivation.
In parallel, we determined RNA Pol II (RNAPII) occupancy as a di-
rect measure of the transcriptional activity. We show that GC and
NFKB signaling pathways are significantly rearranged following
4
These authors contributed equally to this work.
5
Corresponding authors.
E-mail h.stunnenberg@ncmls.ru.nl.
E-mail dmitsiou@eie.gr.
Article published online before print. Article, supplemental material, and pub-
lication date are at http://www.genome.org/cgi/doi/10.1101/gr.118042.110.
1404 Genome Research
www.genome.org
21:1404–1416 Ó2011 by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/11; www.genome.org

coactivation. By pairing distinct genomic binding patterns of GR
and p65 with changes in transcriptional events, we provide an
extensive insight into GR and NFKB crosstalk.
Results
Transcriptional read-out of GR and NFKB activation
We determined the direct global transcriptional changes induced
upon activation of GR or NFKB pathways individually and upon
coactivation by performing RNAPII ChIP-seq (see Methods). We
used the well-studied HeLa B2 cells that express endogenous GR
and p65 proteins and respond properly to glucocorticoids and
proinflammatory cytokines (such as TNF). An overview of the
number of tags that were sequenced and mapped in our experi-
ments is presented in Supplemental Table 1. RNAPII occupancy
throughout the gene body has been used as a direct read-out of
transcriptional activity in a number of previous studies (Nielsen
et al. 2008; Sultan et al. 2008; Welborenet al. 2009). We found 561
genes with significantly altered RNAPII occupancy upon treatment
with triamcinolone acetonide (TA), a synthetic GR ligand (Fig. 1A).
Furthermore, 1045 genes responded to treatment with the in-
flammatory cytokine TNF and 1003 genes showed altered RNAPII
occupancy upon coactivation (TA +TNF). The RNAPII data were
validated by assessing the mRNA levels of 44 randomly selected
genes with altered RNAPII occupancy using RT–qPCR (Fig. 1B). The
mRNA levels of the TA-responded genes strongly correlated with
the RNAPII occupancy (top). However, we observed reduced cor-
relation for genes responded to TA +TNF treatment (bottom) and
poor correlation for TNF-responded genes (middle), most probably
because genes involved in the inflammatory process respond
rapidly to treatment and their mRNAs are short-lived (Hamilton
et al. 2010), and thus not suitable for assessing transcriptional ac-
tivity. In line, we found that the RNAPII occupancy correlated
better with primary transcript (primRNA) levels than with the
mRNA levels for 15 TNF-responded genes (Supplemental Fig. 1A).
Therefore, the RNAPII occupancy is a more reliable read-out of
transcription and is used as a measure for regulated genes in the
present study. Collectively, 1932 genes showed significantly al-
tered RNAPII occupancy in at least one condition tested. Com-
parison of the repertoires of regulated genes showed that 43% of
the TA +TNF-regulated genes responded only when both ligands
are added (Fig. 2A), indicating that GR and NFKB crosstalk alters
signaling pathways that are regulated by each factor separately.
Analysis of the profiles of regulated genes revealed six distinct
clusters that were assigned to biological functions based on GO
analysis (Fig. 2B; Table 1; Supplemental Fig. 1B; Supplemental
Table 2). A detailed description of the six clusters is provided in the
Supplementary Notes (description of Supplemental Fig. 1B). Con-
cerning the effect of coactivation as compared with that of activa-
tion of GR and NFKB separately, four of the identified clusters are of
particular interest: Genes in clusters 2 and 6 appear to comprise
mainly genes up-regulated by TA that are either not affected (cluster
2) or suppressed by TNF (cluster 6) upon coactivation (TA +TNF).
Cluster 4 includes genes up-regulated by TNF, 20% of which are
significantly suppressed by TA upon TA +TNF treatment (proin-
flammatory genes), and cluster 5 contains genes synergistically up-
regulated by TA and TNF. Intriguingly, both clusters 2 and 4 comprise
genes involved in MAPK signaling pathways: Cluster 2 includes
inhibitors of the MAPK signaling (such as DUSP1), which mediate
at least part of the anti-inflammatory effects of Gcs, whereas cluster
4 contains genes that are involved in the activation of MAPK (such
as MAP3K8,TNF,TGFB,TGFBR, and FAS) and downstream effec-
tors (such as NFKB1 and NFKB2), indicating an active MAPK
pathway involved in inflammatory crosstalk.
Taken together, our data indicate that coactivation of GR and
NFKB alters expression of genes regulated by activation of each
factor separately, thus leading to changes of signaling pathways
regulated by TA or TNF
Genome-wide binding of activated GR and NFKB
We next assessed the chromatin-binding profile of GR and p65 on
a genome-wide scale. We performed ChIP–seq to identify the GR-
binding sites using chromatin from cells treated with TA or DMSO
(see Methods). The number of mapped reads is presented in Sup-
plemental Table 1. Analysis of our data revealed that the depth of
sequencing was sufficient for the goals of our study (see legend to
Supplemental Table 1) and further validatedthe quality of our data
sets (Supplemental Fig. 2A,B). Peak calling identified 8306 GR-
binding sites (P<10
5
,FDR<0.05) upon treatment with TA,
Figure 1. Genes with altered RNAPII occupancy upon TA and/or TNF
treatment. (A) The number of genes with altered RNAPII occupancy (in-
creased and decreased) in response to TA, TNF, and TA +TNF treatment,
as indicated. (B) RNAPII occupancy vs mRNA levels. Scatter plots depict-
ing the Pearson correlation (r) between changes in mRNA levels (y-axis)
and in RNAPII occupancy (x-axis) for randomly selected regulated genes.
Changes in mRNA and RNAPII occupancy were expressed as fold in-
duction of the indicated treatment over DMSO (in log
2
scale).
Genome Research 1405
www.genome.org
Global GR and NFKB crosstalk

including sites in the promoters of known GR target genes such as
PER1 (Fig. 3A). To confirm the specificity of our observations we
developed HeLa B2 GR_KD cells that express very low levels of GR
(described in Supplemental Fig. 2C–E). ChIP–qPCR experiments
using independent biological replicates from WT and GR_KD cells
validated our data (24/25 randomly selected sites) (Fig. 3B; Sup-
plemental Fig. 3A). Similar results were obtained using either the
monoclonal antibody used in ChIP–seq or a polyclonal antibody
raised against a different epitope of human GR (data not shown).
It should be noted that peaks were detected in the control
(GR_DMSO); however, these sites could not be confirmed as hor-
mone-independent GR-binding sites by ChIP–qPCR using the
distinct polyclonal antibody (data not shown). Location analysis
revealed that about half of the binding sites are intragenic (48%),
a significant percentage are at >25 kb from transcription start sites
(24%) and only a minor part in promoters (7%) (Fig. 3C), in agree-
ment with recent studies for GR (Reddy et al. 2009), ER (Carroll
et al. 2006; Lin et al. 2007; Welboren et al. 2009), and PPARg:RXR
(Nielsen et al. 2008). Our data set partially overlaps with a genome-
wide GR profile in human A549 lung epithelial carcinoma cells
(Fig. 3D; Reddy et al. 2009). The observed difference is likely to re-
flect the cell-type specificity of GR signaling with the overlapping
Figure 2. Cluster analysis of genes regulated upon TA and/or TNF treatment. (A) Comparison of the genes regulated by TA, TNF, and TA +TNF. Venn
diagram represents the overlap between the genes regulated upon each treatment. (B) Cluster analysis of the regulated genes. Using k-means clustering,
the 1932 collectively regulated genes were clustered into six distinct clusters based on their changes in RNAPII occupancy upon treatment with TA, TNF,
and TA +TNF, relative to DMSO, as depicted by the heatmap. The number of genes in each cluster is indicated. The cluster profiles are as follows: (Cluster
1) largely unaffected by TA and down-regulated by TNF and to a higher extent by TA +TNF; (Cluster 2) up-regulated by TA and TA +TNF, and down-
regulated by TNF; (Cluster 3) up-regulated by TNF and down-regulated by TA and TA +TNF; (Cluster 4) largely unaffected by TA and strongly up-
regulated in the presence of TNF (a subset of these genes is down-regulated by TA +TNF); (Cluster 5) mildly up-regulated by TA or TNF and strongly up-
regulated by TA +TNF; (Cluster 6) up-regulated by TA and down-regulated by TNF and to a lesser extent by TA +TNF. The main characteristics of genes in
clusters 2, 4, 5, and 6 are indicated at right.
Rao et al.
1406 Genome Research
www.genome.org

binding sites being most probably common sites across different
cell types, whereas nonoverlapping sites represent cell-type-specific
regulatory elements. The ligand specificity (TA vs. Dex), may also
play a significant role, given that the availability of intracellular
steroids, and thus their potency, is altered by transporters that
act in a ligand- and cell-type-specific manner in mammalian cells
(Kralli and Yamamoto 1996). Antibody-dependent variation and
sample handling could also account for the observed differences.
Next, we extended our analysis to identify the global p65-
binding profile upon treatment with TNF. ChIP-seq data revealed
the presence of p65-binding sites in the promoter regions of
known p65 target genes such as CCL2 (Fig. 3E). Peak calling
revealed 12,552 TNF-induced p65-binding sites (P<10
5
, FDR <
0.05) that were validated by ChIP–qPCR
using WT and p65_KD cells (20/20 ran-
domly selected sites) (Fig. 3F; Supple-
mental Fig. 3B; HeLa B2 p65_KD cells
expressing very low levels of p65 are de-
scribed in Supplemental Fig. 2F,G). Loca-
tion analysis revealed that the p65-bind-
ing sites are mainly intragenic (46%) and
only 7% are located in promoters (Fig.
3G), in agreement with previous studies
(Lim et al. 2007). Our p65 data set showed
limited overlap with a genome-wide p65
profile identified by ChIP–PET in THP-1
human monocytic cells (Fig. 3H; Lim
et al. 2007). The high accuracy, sensitiv-
ity, and sequence depth of ChIP–seq
probably allowed for the identification
of transient or weak binding sites as well
as high-affinity sites in our data set. The
observed difference is also likely due to
the cell lines and ligands (TNF vs. LPS), as
well as to sample handling.
To gain insight into the molecular
crosstalk between GR and NFKB at the
level of binding to chromatin, we map-
ped both GR- and p65-binding sites upon
coactivation (TA +TNF). Peak calling re-
vealed 8696 GR- and 12,713 p65-binding
sites (P<10
5
, FDR <0.05) (Fig. 4A) that
were validated by ChIP–qPCR (Supple-
mental Fig. 3A,B; 25/25 and 20/20 ran-
domly selected GR and p65 sites, respec-
tively). Figure 4A summarizes the GR- and
p65-binding sites detected upon TA, TNF,
and TA +TNF treatment using the re-
spective DSMO ChIP–seq data as control.
To link the identified binding sites with
the regulated genes, we calculated the
median distance between the transcrip-
tion start sites (TSS) of the regulated genes
and the closest binding site, thus assign-
ing one nearest peak to each regulated gene.
Interestingly, promoter proximal binding
of GR or p65 correlates well with up-reg-
ulation of their targets genes, whereas
down-regulated genes are often associated
with distal GR and p65 binding (Fig. 4B;
Supplemental Fig. 3C), in agreement with
similar findings for GR sites in human
A549 cells (Reddy et al. 2009).
The above data provide the global profiles of GR and NFKB
chromatin binding sites and show that coactivation does not strongly
affect the number of binding sites identified upon single activation.
Motif analysis
DenovomotifsearchineachsetofGR-andp65-bindingsitesfrom
different treatments identified significantly enriched DNA motifs
(Fig. 4C; Supplemental Fig. 3D), which were used to scan each peak
set. Analysis of scan results showed that GR and NFKB response ele-
ments (GRE and NFKB RE) are the most prevalent motifs identified in
the 25%–30% or GR- and p65-binding sites, respectively (Fig. 4D).
Table 1. GO analysis for the genes of each of the six clusters (with the respective P-values)
and examples of genes for the indicated GO terms
Cluster 1 P-value Genes
Cell cycle P=2.80310
4
BUB; BUB1B; CCNA2; CCNB1; PTTG1
Propanoate metabolism P=5.94310
4
MLYCD; ACACB; ACSS2
Reductive carboxylate
cycle (CO2 fixation)
P=1.40310
3
ACLY; ACSS2
Cluster 2 P-value Genes
MAPK signaling pathway P=1.80310
5
DUSP1; DUSP4; AKT1; FLNA; MAPK8IP3;
MKNK2; HSPB1; ARRB1; MYC; GADD45B;
PDGFB; MAP2K2; CACNB3; CACNA1H;
MAP3K6
Focal adhesion P=8.46310
6
AKT1; FLNA; ITGA3; ITGA5; LAMA5; PDGFB;
PPP1CA; CCND1; ACTG1; VASP; VEGFA;
ZYX; ITGA10
Insulin signaling pathway P=9.22310
4
AKT1; FLOT2; MKNK2; PPP1CA; MAP2K2;
PYGB; TSC2; SOCS3
Chronic myeloid leukemia P=8.41310
4
CTBP1; AKT1; MYC; MAP2K2; CCND1; RUNX1
mTOR signaling pathway P=7.97310
5
AKT1; DDIT4; STK11; TSC2; VEGFA; ULK1
Urea cycle and metabolism
of amino groups
P=4.56310
4
CKB; NAGS; GAMTp; PYCRL
Adipocytokine signaling
pathway
P=4.18310
3
AKT1; RXRA; RXRB; STK11; SOCS3
Cluster 3 P-value Genes
Ribosome P=1.24310
6
RPL7A; RPL9; RPL11; RPL21; RPS13; RPS18
Biosynthesis of steroids P=2.84310
5
FDFT1; HMGCR; IDI1
Terpenoid biosynthesis P=1.85310
4
FDFT1; IDI1
Cluster 4 P-value Genes
Cytokine-cytokine receptor
interaction
P=8.29310
14
EGFR; CXCL2; IFNAR2; IFNGR2; FAS; IL1RAP;
IL4R; IL6; IL6ST; IL12RB2; IL15 ; LIFR; CCL2;
CCL20; BMP2;TGFB1; TGFBR1;TNF; TNFRSF10B
MAPK signaling pathway P=1.46310
6
MAP3K8; GADD45A; EGFR; FAS; NFKB1; NFKB2;
PLA2G4A; MAP2K3; TGFB1; TGFBR1;
TNF; DUSP16
Apoptosis P=8.47310
10
BIRC2; FAS; IL1RAP; IRAK2; NFKB1; NFKB2;
NFKBIA; BID; TNF;
Jak-STAT signaling pathway P=2.08310
6
IFNAR2; IFNGR2; IL4R; IL6; IL6ST; IL12RB2;
IL15; JAK1; LIFR; TNFRSF10B
Toll-like receptor signaling
pathway
P=4.00310
7
IFNAR2; IL6; NFKB1; NFKB2; NFKBIA; MAP2K3;
TNF; IKBKE
Cluster 5 P-value Genes
HIV-I Nef: negative effector
of Fas and TNF
P=1.03310
4
BIRC3; PSEN1; TRAF2; CFLAR
TNFR1/TNFR2 signaling P=7.90310
4
BIRC3; TRAF2; TNFAIP3;
Glycerophospholipid
metabolism
P=3.24310
3
CDIPT; ACHE; PPAP2C
Wnt signaling pathway P=4.91310
3
WNT4; PPP2CB; PSEN1; NKD2
Cluster 6 P-value Genes
Axon guidance P=8.78310
3
SEMA6C; DPYSL2; PLXNA2; RGS3; SEMA3F;
SEMA3B
PPAR signaling pathway P=2.97310
3
SLC27A5;CPT1A;FABP3; ACADM;SLC27A1
Glutathione metabolism P=3.14310
4
G6PD; ANPEP; GSS; GSTT1; IDH2
Phosphatidylinositol
signaling system
P=3.38310
3
PIB5PA; INPPL1; INPP5K; PIK3R2; CALM3
Hypoxia and p53 in the
Cardiovascular system
P=3.86310
3
CDKN1A;DNAJB1; BAX
Global GR and NFKB crosstalk
Genome Research 1407
www.genome.org

Interestingly, the AP1 response element (AP1 RE) is prominently
present in both GR and p65 data sets under all conditions, thus
corroborating and extending the important role of AP1 in gluco-
corticoid and TNF signaling pathways. We also found statistically
significant enrichment of response elements for AP2, SP1, CEBPA,
RORA1, TEAD1, and RREB1, in agreement with published data
(Phuc Le et al. 2005; So et al. 2007); however, these motifs do not
show differential occurrence between different categories of bind-
ing sites (Fig. 4D).
Our data show the presence of GRE
and NFKB RE only to a fraction of GR- and
p65-binding sites and suggest that other
factors/mechanisms can facilitate their
association with chromatin at the sites
that do not contain the respective motifs.
Coactivation alters the GR- and
NFKB-binding site repertoires
and gene-expression profiles
Studies at the single gene level have pro-
vided evidence for a direct physical in-
teraction between GR and inflammatory
regulators such as NFKB and AP1 (Konig
et al. 1992; Gottlicher et al. 1998); how-
ever, genome-wide assessment of the over-
lap and crosstalk between GR and NFKB has
not been reported. Our analysis revealed
that the majority of GR-binding sites oc-
cupied upon induction with TA +TNF
were also occupied upon treatment with
TA alone (Fig. 5A, maintained sites). Fur-
thermore, 643 GR sites were lost upon
coactivation with TNF. However, most of
them were weak sites and ChIP–qPCR
experiments for their validation were in-
conclusive (data not shown). Loss of
these sites may also be the result of inef-
ficient GR ChIP due to interaction of GR
with other factors or modifications in-
duced by coactivation, thus masking the
epitope recognized by the antibody used
in this study.
Importantly, a fraction of GR sites
(12%) is detected only upon coactivation
(gained sites) (Fig. 5A), and could be val-
idated by ChIP–qPCR experiments (18/19
randomly selected sites; Supplemental
Fig. 4A). Tag density maps for lost, main-
tained, and gained GR sites clearly show
that our ChIP–seq data analysis is com-
prehensive, and there is a clear distinction
between the different categories of bind-
ing sites (Supplemental Fig. 4B). Given
that the frequency of NFKB RE and AP1
RE is significantly increased in the gained
sites (Supplemental Fig. 4C), we looked for
GR and p65 co-occupancy at these sites.
Interestingly, we identified that a large
fraction of gained GR sites were co-oc-
cupied by p65, as illustrated by the tag
density map showing the p65 occupancy
around these gained GR sites (Fig. 5B, left) and the presented ex-
ample (Fig. 5E). ChIP–reChIP–qPCR experiments validated their
co-occupancy (15/19 randomly selected sites) (Supplemental Fig.
4D). The NFKB RE and AP1 RE occurred at the highest frequency in
these gained GR sites co-occupied by p65 (Fig. 5C). It is therefore
likely that TNF-induced binding of NFKB (or for that matter, AP1)
facilitates GR entry at the gained sites. In line with this hypothesis,
p65 KD abolished binding of GR (and p65) at the gained GR sites
co-occupied by p65, as shown by the box plots from GR and p65
Figure 3. (Legend on next page)
Rao et al.
1408 Genome Research
www.genome.org

ChIP_seq in p65_KD cells (Fig. 5D) and illustrated for the ZBTB20
gene (Fig. 5E). These data support the recruitment of GR by acti-
vated chromatin-bound p65 at the gained GR sites (Fig. 5F). GR
and p65 ChIP–qPCR experiments in WT and p65_KD cells vali-
dated the results (Supplemental Fig. 4E). A fraction of gained GR
sites without p65 co-occupancy was also detected (the p65 occu-
pancy pattern around these gained GR sites upon treatment with
TA +TNF is similar to that of DMSO-treated cells) (Fig. 5B, right).
These sites show increased occurrence of GRE and AP1 RE (Sup-
plemental Fig. 5A). Strikingly, ChIP-seq using WTand p65_KD cells
showed that binding of GR at these sites was still dependent on the
presence of p65 (Supplemental Fig. 5B). Inefficient ChIP of p65 due
to epitope masking could account for its absence from these gained
sites.
A similar analysis was performed for the p65-binding sites.
The majority of p65 sites are maintained upon single treatment or
coactivation (Fig. 6A). Most of the 1075 p65 sites lost upon coac-
tivation with TA and TNF were weak binding sites (data not
shown), and at present we cannot exclude the possibility of in-
efficient ChIP of p65 at these sites upon cotreatment. Interestingly,
10% of the p65 sites were identified only upon coactivation
(gained sites) (Fig. 6A) and validated by ChIP–qPCR (15/19 ran-
domly selected sites, Supplemental Fig. 6A). Tag density maps de-
pict the three categories of p65 sites (Supplemental Fig. 6B). Since
these sites are highly enriched for GRE (Supplemental Fig. 6C), we
searched for GR binding within the gained p65 sites and defined
sites with, as well as without, GR co-occupancy (high and low GR
occupancy around these gained p65 sites, respectively) (Fig. 6B,E).
ChIP–reChIP–qPCR experiments confirmed the co-occupancy of
both p65 and GR (11/12 randomly selected sites, Supplemental
Fig. 6D). Motif analysis revealed the predominant occurrence of
a GRE in the co-occupied gained sites but not in the sites without
GR co-occupancy (Fig. 6C ; Supplemental Fig. 5D). Importantly,
GR and p65 ChIP-seq in GR_KD cells showed reduced binding of
GR as well as p65 in the co-occupied sites (Fig. 6D,E), thus con-
firming the recruitment of p65 by activated chromatin-bound GR
at the gained p65 sites (Fig. 6F). These ChIP–seq data were vali-
dated by GR and p65 ChIP–qPCR experiments (Supplemental Fig.
6E). Binding of p65 at the gained sites without GR co-occupancy
was also dependent on the presence of GR (Supplemental Fig. 5E).
GR ChIP–qPCR with the monoclonal antibody used for ChIP–seq
and a polyclonal antibody raised against a different epitope of GR
did not reveal GR co-occupancy at these gained p65 sites (data not
shown). However, at present we cannot exclude the possibility of
inefficient GR ChIP at these sites due to epitope masking.
Comparative analysis of maintained GR and p65 sites showed
that a substantial number of sites are common for GR and p65
(Supplemental Fig. 7A–C). Comparison of ChIP–seq data from WT,
GR_KD, and p65_KD cells revealed that binding of GR or p65 on
the maintained sites does not depend on the presence of the other
factor (Supplemental Fig. 7D). These data show that GR and p65
binds to the maintained sites independently of each other, and
therefore validate the effect of KDs on the gained sites as specific.
Our data suggest that TNF-induced binding of p65 facilitates
GR recruitment at the gained GR sites (Fig. 5F). Similarly, TA-in-
duced binding of GR facilitates binding of p65 at the gained p65
sites (Fig. 6F). To assess the involvement or consequence of GR and
p65-binding sites that are gained upon coactivation in the regu-
lation of clusters of genes defined by RNAPII occupancy (Fig. 2B),
we assigned these gained sites to the genes of each cluster and then
assessed the peak enrichment (see Methods). Interestingly, gained
GR sites co-occupied by p65 are enriched in the cluster containing
proinflammatory genes (up-regulated by TNF), some of which are
suppressed by TA (Fig. 5G, cluster 4), thus suggesting that recruitment
of G R b y p65 mediat es, a least in part, the anti-inflammatory
function of GR. Assignment of the gained p65 sites co-occupied by
GR uncovered enrichment to clusters containing GR targets that are
partially suppressed by additional TNF treatment (Fig. 6G, cluster 6).
These data indicate that recruited p65 antagonizes GR function on
these genes. In contrast, the high enrichment of gained p65 sites to
GR targets that are largely unaffected by additional TNF treatment
(cluster 2) indicates that recruitment of p65 alone is insufficient
and that other factors and/or conditions are required for cross-
regulation to occur. The gained GR and p65 sites not co-occupied
by p65 and GR, respectively, are enriched in all four clusters of
regulated genes tested (2, 4, 5, and 6) but, in general, to a lesser
extent as compared with the sites co-occupied by both factors
(Supplemental Fig. 5C,F). The maintained sites co-occupied by GR
and p65 were significantly enriched to clusters containing proin-
flammatory genes, some of which are down-regulated by TA
(cluster 4) or genes strongly up-regulated by TA and TNF synergism
(cluster 5) (Supplemental Fig. 7E).
Taken together, our data provide evidence that the genome-
wide GR- and p65-binding repertoires are significantly affected by
the combination of stimuli and support
the notion that GR and p65 are recruited
by each other (and probably other factors
such as AP1) to new binding sites gained
upon coactivation in a mutually dependent
manner. Gained GR sites are enriched in
the cluster containing proinflammatory
genes, part of which are suppressed by TA,
while gained p65 sites are enriched in the
cluster of TA targets suppressed by TNF.
Discussion
To unravel the global crosstalk between
GR and NFKB we applied ChIP-seq and
mapped GR- and p65-binding sites and
changes in RNAPII occupancy upon acti-
vation of GR and NFKB alone or in com-
bination. We identified a large number
of GR and p65-binding sites as well as
Figure 3. Genome-wide GR- and p65-binding sites upon TA or TNF treatment. (A,E) RNAPII and GR
(A) or p65 ChIP-seq data (E) illustrate RNAPII occupancy and GR and p65 binding in the promoter and/
or within the gene body of target genes. Strong binding sites were detected upon TA (green) and TNF
treatment (pink), whereas residual binding was observed in the absence of ligand (gray). Data were
viewed in the University of California at Santa Cruz (UCSC) Genome Browser. The maximum number of
overlapping tags, representing peak height, is indicated on the y-axis. (B,F) Validation of GR-binding
sites (B) and p65-binding sites (F). Strip plots of GR ChIP–qPCR data for 25 randomly selected GR sites
from WT and GR_KD cells (B) and p65 ChIP–qPCR data for 20 randomly selected p65 sites from WT and
p65_KD cells (F) treated as indicated. Results are expressed as a percentage of input chromatin. Binding
sites with fold induction (relative to DMSO) >2 are regarded as validated. Establishment of GR_KD and
p65_KD cells are described in Supplemental Figure 2. (C,G) Genomic location analysis of GR-binding
sites (C) and p65-binding sites (G). GR sites identified upon TA treatment (C) and p65 sites identified
upon TNF treatment (G) were divided into the following categories based on their position relative to
the nearest gene: distant (>25 Kb), within 5 Kb upstream, within 5 kb downstream, within 5–25 kb
upstream, within 5–25 kb downstream and intragenic (exon, intron, and within 0–5 kb up- and
downstream). The Pink Thing (http://pinkthing.cmbi.ru.nl) was used for the analysis. (D) Comparison of
genome-wide GR profiling data. Venn diagram of the overlap of GR sites identified in the present study
(8306 sites, ChIP-seq using HeLa B2 cells treated with TA for 4 h) with those identified in Reddy et al.
(2009) (4392 sites, ChIP-seq using A549 cells treated with Dex for 1 h). (H) Comparison of genome-wide
p65 profiling data. Venn diagram of the overlap of p65 sites identified in the present study (12,552 site s,
ChIP-seq using HeLa B2 cells treated with TNF for 1 h) with those identified in Lim et al. (2007) (5858
sites, ChIP-PET using THP-1 cells treated with LPS for 1 h).
Global GR and NFKB crosstalk
Genome Research 1409
www.genome.org

differentially regulated genes. By pairing the binding-site data and
the gene expression data, we provide a global overview of the GR
and NFKB crosstalk.
Comparison of the regulated gene repertoires determined by
RNAPII ChIP–seq revealed that coactivation of both GR and p65
strongly affects gene expression, as half of the TA +TNF-affected
genes do not respond to either TA or TNF alone. This extensive and
highly significant reprogramming of gene regulation appeared to be
orchestrated to a largeextent by the changes in GR-and p65-binding
patterns. The striking discordance between the number of regulated
genes and the GR- and p65-binding sites suggests that multiple sites
are involved in the regulation of a single gene and/or that binding
of a transcription factor is not sufficient to drive gene expression.
Comprehensive analysis of GR- and p65-binding sites re-
vealed that GRE and NFKB RE are the predominantly occurring
motifs, respectively. Nevertheless, only 25%–30% of the binding
sites contain the respective motifs (our data, and Martone et al.
2003; Lim et al. 2007; Reddy et al. 2009), indicating that other
(transcription) factors can facilitate the association of GR and p65
with chromatin. The high occurrence of GREs in gained p65-
binding sites occupied only upon coactivation and, conversely,
NFKB RE in gained GR sites, underscores the extent of GR and p65
mutual recruitment at these sites. In addition, the occurrence of
the AP1 RE in the 20%–25% of GR- and p65-binding sites suggests
that AP1 is likely to mediate part of the GR and p65 binding to
chromatin. This is further supported by the binding of FOS in GR-
binding sites containing AP1 REs (our unpublished observations).
Comparison of GR- and p65-binding site repertoires revealed
significant changes upon coactivation of GR and p65. Although
coactivation does not affect the majority of binding sites, it in-
duces binding at a considerable number of new sites. The de-
pendence of GR and p65 on each other for binding at these gained
sites was confirmed and extended by ChIP–seq data from GR_KD
and p65_KD cells and uncovered the functional and probably
physical interaction of GR and p65 on chromatin on a genome-
wide scale. The sites that are bound by GR only upon coactivation
and that are co-occupied by p65 are enriched in the cluster con-
taining proinflammatory genes, part of which are suppressed by
liganded GR (cluster 4). A typical example is the inflammatory
cytokine CCL2 gene that is induced by p65 and is effectively sup-
pressed by GR upon coactivation. This p65-mediated recruitment
of GR at inflammatory genes that are down-regulated upon co-
activation corroborates the previously proposed mechanisms for
the anti-inflammatory action of glucocorticoids, and extends it
as a widely used mechanism of cross-regulation. Similarly, re-
cruitment of p65 to sites occupied by GR in clusters 2 and 6 sug-
gests that the GR and p65 crosstalk works in a bidirectional way,
i.e., p65 is also targeted to GR on glucocorticoid activated genes
and down-regulates them, as seen for genes of cluster 6. Our data
also show binding of GR and p65 at common sites (maintained)
that are not dependent or influenced by each other. These loci
contain juxtaposed cis-acting elements involved in the regulation
of shared target genes enriched in clusters 4 and 5 (as exemplified
by trans-activation of the anti-inflammatory NFKBIA gene for
cluster 4 and TNFAIP3 gene for cluster 5).
The recruitment of GR and p65 to gained sites that are not
occupied by the other factor, but require its presence, was rather
unexpected. Our data suggest that activated GR (or p65) may be
required to induce chromatin modifications that will create a more
accessible environment for binding of the other factor upon co-
activation. In this case, GR (or p65) itself is likely to bind at a
distant site and exerts its effect by chromatin looping. Alterna-
tively, GR (or p65) could bind transiently on chromatin to po-
tentiate binding of the other factor. In line with this, molecular
chaperones (such as the p23, and to a lesser extent, the HSP90)
have been shown to localize to genomic response elements and to
promote disassembly of GR and NFKB transcriptional regulatory
complexes, thus regulating responsesto signaling changes (Freeman
Figure 4. Analysis of genome-wide GR- and p65-binding sites upon TA and/or TNF treatment. (A) The number of GR- and p65-binding sites identified
upon TA, TNF, and TA +TNF treatment using the respective DMSO ChIP-seq data as control. (B) Median distance (in Kb) of GR- and p65-binding sites to
the TSS of the nearest gene regulated by TA, TNF, and TA +TNF, as indicated. (C) Response elements for GR, NFKB, and AP1 identified as the predominant
motifs in the GR and p65 binding sites. Motifs were identified by a de novo motif search and visualize d using WebLogo. (D) Motif occurrence within the GR
and p65 binding sites. The bar graph shows the percentage of GR and p65 sites containing the indicated motifs.
Rao et al.
1410 Genome Research
www.genome.org

and Yamamoto 2002). Although the absence of GR and p65 co-
occupancy at the gained p65 sites was confirmed by using two
antibodies against GR, at present we cannot exclude the possibility
of inefficient GR and p65 ChIP at gained p65 and GR sites, re-
spectively, because of ‘‘epitope masking’’ due to protein–protein
interactions/modifications induced by coactivation.
A major part of the GR- and p65-binding sites are located at
long distances from the annotated genes. GR (or p65) bound at
Figure 5. Gained GR-binding sites and their correlation to gene expression profile. (A) Profile of GR-binding sites. Venn diagram of the overlap of GR sites
detected upon treatment with TA or TA +TNF. (B) Tag density maps depicting the pattern of p65 occupancy around (peak mode 62.5 kb) gained GR sites.
Color density indicates the level of p65 occupancy (square rootof tag density; see scale below)in a 250-bp window. The position ofthe example presented in
E(ZBTB20)isindicated.(C) Motif occurrence within gained GR sites co-occupied by p65. The bar graph shows the percentage of sites containing the
indicated motifs. (D) Boxplots ofGR and p65 tag countsdistributed under peak locations (log
2
scale) of gained GR sites co-occupied by p65, upon TA +TNF
treatment,in WT cells and the respective tagdistributions under the samelocations in p65_KD cells. (E)GR and p65 ChIP-seq data illustrate binding ofGR and
p65 at gained GR sites co-occupied by p65, detected upon the indicated treatments, in WT and p65_KD cells. Data were viewed in the UCSC Genome
Browser. The maximum number of overlapping tags, representing peak height, is indicated on the y-axis. (F) Model of GR and p65 interaction at gained GR
sites co-occupied by p65. (G) Enrichment of gained GR sites co-occupied by p65 that are assigned to the genes of each cluster. Peak enrichment fold ratio
represents the ratio of the nonrandom enrichment (ratio of assigned binding sites to the total number of genes in each cluster) to the random enrichment
(parallel enrichment calculation for 100 random sets of nonregulated genes, performed to assess the statistical significance of the peak enrichment analysis).
Global GR and NFKB crosstalk
Genome Research 1411
www.genome.org

distal sites is likely to be involved in long-range chromatin in-
teractions with GR (or p65) bound at other sites and probably other
factors, thus forming loop structures. Such loop structures/hubs
have been globally described for the ERa-binding sites that interact
with each other and function as ‘‘anchor’’ regions; most genes
close to these hubs are up-regulated in an ERa-dependent manner
(Fullwood et al. 2009). In line with this observation, our data also
show that binding sites occurring close to the TSS correlate well
Figure 6. Gained p65-binding sites and their correlation to gene expression profile. (A) Profile of p65 binding sites. Venn diagram of the overlap of p65 sites
detected upon treatment with TNF or TA +TNF. (B) Tag density maps depicting the pattern of GR occupancy around (peak mode 62.5 kb) gained p65 sites.
Color density indicates the level of GR occupancy (square root of tag density; see scale below) in a 250-bp window. The position of the example presented in
E(ZFAND5)isindicated.(C) Motif occurrence within gained p65 sites co-occupied by GR. The bar graph shows the percentage of sites containing the indicated
motifs. (D) Boxplots of GR and p65 ta g counts distributed under peak locations (log
2
scale) of gained p65 sites co-occupied by GR, upon TA +TNF treatment, in
WT cells, and the respective tag distributions under the same locations in GR_KD cells. (E) p65 and GR ChIP-seq data illustrate binding of p65 andGR at gained
p65 sitesco-occupied by GR, detected upon the indicatedtreatments,in WT and p65_KDcells. Data wereviewed in the UCSC GenomeBrowser. Themaximum
number of overlapping tags, representing peak height, is indicated on the y-axis.(F) Model of GR and p65 interactionat gained p65 sitesco-occupiedby GR. (G)
Enrichment of gained p65 sites co-occupied by GR that are assigned to the genes of each cluster. Peak enrichment fold ratio is as described in Figure 5G.
Rao et al.
1412 Genome Research
www.genome.org

with up-regulated genes. At the single gene level, binding of NFKB
at a distal site was shown to regulate CCL2 (also known as MCP-1)
gene transcription by enabling binding of SP1 at a promoter-
proximal site in an EP300-dependent manner (Teferedegne et al.
2006). Similarly, GR has been shown to activate the CIZ1-LCN2
locus by loop structure formation (Hakim et al. 2009). Chromatin
looping may also enable the GR and p65 co-occupancy at the sites
gained upon coactivation. Such chromatin loops/hubs have been
proposed to serve as sites for recruitment of transcription factors
that enable coordinated transcriptional regulation in a stimulus-
and cell-specific manner.
In conclusion, our results provide a novel genome-wide
footprint of GR and NFKB crosstalk and substantially contribute to
the understanding of the networks underlying the glucocorticoid
and inflammatory signaling pathways.
Methods
Cell culture, Western blot, and luciferase assay
HeLa B2 cells were maintained in Dulbecco’s modified Eagle me-
dium (DMEM) supplemented with 10% fetal calf serum (FCS) at
37°Cin5%CO
2
. HeLa B2 are HeLa cells stably transfected with the
glucocorticoid-responsive mouse mammary tumor virus long ter-
minal repeat–luciferase reporter vector (MMTV–Luc) (Hollenberg
and Evans 1988), together with the neomycin resistance gene,
using the calcium phosphate coprecipitation method. Eighteen
hours after transfection, cells were washed with phosphate-buff-
ered saline (PBS), fed with fresh medium, and 24 h later refed with
medium containing geneticin (0.5 mg/mL of medium). Cells were
fed with fresh geneticin-containing medium every 2–3 d, and
colonies were isolated 3 wk later and tested for luciferase activity in
the presence or the absence of 1 mM TA (T6501, Sigma-Aldrich). For
GR analysis, cells were cultured in DMEM supplemented with 10%
charcoal stripped FCS for 48–72 h before subsequent treatment
and/or harvesting.
For Western blot analysis, whole-cell extracts (20–30 mgof
protein) were analyzed by SDS-PAGE, followed by immunoblot-
ting using antibodies against GR (Siriani et al. 2003) (MAb-NRhGR-
050, Diagenode), p65 (sc-372, Santa Cruz Biotechnologies), and
TBP (MAb-002-100, Diagenode). Proteins were visualized using
ECL (GE healthcare).
For the luciferase assay, cells were treated with either DMSO or
1mM TA for 16 h. Cells were harvested in Glo Lysis Buffer, and
luciferase activity was measured using a Steady Glo Luciferase As-
say system (Promega) in a Safire 2 plate reader. Luminescence
readings were normalized with the protein content of the samples.
mRNA analysis
Total RNA was harvested using Qiagen RNeasy kit with on-column
DNase treatment according to the manufacturer’s protocol (Qiagen).
A total of 1 mg of RNA was reverse transcribed to cDNA using iScript
cDNA Synthesis kit (BIO-RAD) according to manufacturer’s pro-
tocol. Intron-spanning primers were used to determine mature
mRNA levels using RT–qPCR. Primers spanning intron/exon bound-
aries were used to quantify primary RNA transcripts. Expression
levels were normalized to GAPDH or HPRT levels. The primer pairs
used in this study are listed in Supplemental Table 3.
Lentiviral shRNA knockdown
shRNA lentiviral vector stock (pKLO.1-puro—MISSION shRNA,
Sigma) containing shRNAi oligo DNA directed either against GR or
p65 was produced in Human Embryonic Kidney 293T cells by
transfection of the shRNA construct together with pHR8.2R and
pCMV-VSV-G helper constructs using lipofectamine (Sigma).
Next, HeLa B2 cells (1.5 310
6
in a 10-cm dish) were transduced
with 3 mL of the virus containing supernatant supplemented with
8mg/mL Polybrene (Sigma-Aldrich). Twenty-four hours after the
first transduction, a second round of transduction was preformed
similar to the first. The medium was refreshed 1 d post-infection
and selection was performed using 6 mg/mL of puromycin (Sigma-
Aldrich) for three passages. The mRNA and protein levels of GR and
p65 were assessed in the KD (and WT) cells by RT–qPCR and
Western blot, respectively.
ChIP, ChIP–reChIP (sequential ChIP), qPCR,
and deep sequencing
HeLa B2 cells were cultured in DMEM supplemented with 10%
charcoal-stripped FCS for 72 h and subsequently treated with ei-
ther DMSO or 1 mM TA for 4 h with or without an additional
treatment with 10 ng/mL TNF (T0157, Sigma-Aldrich) for the last 1
h. Chromatin was harvested and single ChIP followed by qPCR
was preformed according to standard protocol (Denissov et al.
2007) with minor adjustments: Following a wash with buffer C
(150 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 50 mM HEPES at pH
7.6), the cells were resuspended in ChIP-incubation buffer at
a concentration of 15 310
6
cells/mL and sheared using a Branson-
250 sonicator (power 5.5, 12 beats, 10 sec per beat, 40-sec interval).
Sonicated chromatin, equivalent to 3.5 310
6
cells, was incubated
with the relevant antibody overnight at 4°C. Antibodies against
human GR (MAb-NRhGR-050), RNAPII (AC-055-100, Diagenode),
and p65 (sc-372) were used. ChIP–reChIP was preformed as de-
scribed (Akkers et al. 2009) with minor adjustments: Chromatin
was eluted from the beads and diluted in ChIP incubation buffer
without SDS (5% Triton X-100, 0.75 M NaCl, 5 mM EDTA, 2.5 mM
EGTA, 100 mM HEPES) to adjust the final SDS concentration to
0.15%, and then subjected to a second round of immunoprecipi-
tation. ChIPed DNA was analyzed by real-time qPCR with specific
primers (Biolegio, listed in Supplemental Table 3)using the 2x SYBR
Geen mix (BIO-RAD) in a MyiQ thermocycler (BIO-RAD). Primers
amplifying exon 2 of myoglobin(MB) were used as negative control.
In addition, ChIPed DNA was prepared for sequencing and se-
quenced according to the manufacturer’s instructions (Illumina)
essentially as described (Nielsen et al. 2008; Welboren et al. 2009;
Martens et al. 2010). Shortly, adaptor sequences were linked to
generated ChIPed DNA, the library was sized selected (200–250 bp),
and amplified by PCR. Subsequent sequencing was carried out on
a Genome Analyzer (Illumina). Validation of GR- and p65-binding
sites was performed with ChIP–qPCR experiments using the same
antibodies as the ones used for ChIP–seq. In addition, a polyclonal
antibody raised against amino acids 77–120 of hGR (DJ Mitsiou and
MN Alexis, unpubl.) was usedto validate GR sites and to confirm the
absence of GR from the gained p65 sites without co-occupancy.
Read alignment and normalization
The image files generated by the Genome Analyzer were proces-
sed to extract DNA sequence data and the 32-bp tags were un-
ambiguously mapped to the human genome (NCBI hg18) using
the eland aligner, allowing, at most, 1 nt mismatch. The 32-bp
sequence reads were directionally extended to 200 bp, corre-
sponding to the length of the original fragments used for se-
quencing plus the ligated adapters. For each base pair in the ge-
nome the number of overlapping sequence reads was determined,
averaged over a 10-bp window, and visualized in the University of
California Santa Cruz Genome Browser (http://genome.ucsc.edu).
All ChIP–seq raw data files have been submitted to the GEO da-
Global GR and NFKB crosstalk
Genome Research 1413
www.genome.org

tabase (accession no. GSE24518). As previously discussed (Zhang
et al. 2008), large differences in sequencing depth may lead to an
increase of FDR and result in false-positive peaks. In order to
compensate for differences in sequencing depth and mapping ef-
ficiency among ChIP–seq samples at different conditions, the total
number of tags of each sample was equalized by uniformly re-
moving tags relatively to the sample with the lower number of tags
(Nielsen et al. 2008). After equalization, the tracks contained the
same number of sequence reads and can be compared quantita-
tively. The RNAPII, GR, and p65 data sets were normalized as in-
dependent groups.
Peak detection and clustering
Detection of putative GR- and p65-binding sites was performed
using MACS (Zhang et al. 2008) with P<10
5
for GR and p65
normalized tracks at FDR level <0.05. DMSO tracks of GR and
p65 ChIP–seq were used as controls for the detection of GR and
p65 peaks, respectively. For each transcription factor, peak loca-
tions identified by MACS upon single activation (e.g., GR_TA)
and coactivation (e.g., GR_TA +TNF) were combined in a common
pool, and sequence tags were counted under each peak location of
the pool independent of treatment. Next, the following Rc analysis
was applied (peaks from GR and p65 were not combined):
Rc=log2
tkT1
ðÞ

tkT2
ðÞ
=Þ;ð
where 
tkdenotes the average number of tags under a specific ge-
nomic region (in this case, under each peak), averaged per kbp, and
T
1
,T
2
are the treatments for which samples are sequenced (for GR,
T
1
corresponds to TA +TNF and T
2
to TA, while for p65, T
1
corre-
sponds to TA +TNF and T
2
to TNF). Larger values of R
c
indicate
binding enrichment, while smaller values indicate binding de-
pletion upon T
1
. Gained/lost peaks for each transcription factor
were defined as peaks falling away from the median of R
c
distri-
bution 623MAD. Final peaks for each treatment were defined by
combining peaks called by MACS and peaks identified by R
c
analysis, excluding redundancies, as follows:
GR_TA =GR_TAMACS [GR_TA +TNFRc

n
GR_TA +TNFMACS \GR_TAMACS

GR_TA +TNF =GR_TA +TNFMACS [GR_TARc

n
GR_TAMACS \GR_TA +TNFMACS

p65_TNF =p65_TNFMACS [p65_TA +TNFRc

n
p65_TA +TNFMACS \p65_TNFMACS

p65_TA +TNF =p65_TA +TNFMACS [p65_TNFRc

n
p65_TNFMACS \p65_TA +TNFMACS

,
where [denotes the set union operator, \denotes the set inter-
section operator, and \ denotes the set difference operator. In ad-
dition, MACS denotes peaks called by MACS and R
c
peaks defined
by R
c
analysis.
Finally, PinkThing (http://pinkthing.cmbi.ru.nl/cgi-bin/
index50.pl [F Nielsen, M Kooyman, HG Stunnenberg, and M
Huynen, in prep.]) was used to determine the location distribution
of identified peaks in relation to the distance to the closest genes.
RNAPII analysis (clustering, gene annotation,
distance analysis)
To identify differentially expressed genes, the number of RNAPII
sequence tags was counted for all NCBI hg18 (July 2007) annotated
human genes (from +500 to end of gene) accessed from Ensemble
(release 50; July 2008), excluding genes smaller than 750 bp. Next,
sequence tags within gene bodies were retrieved and averaged per
750 bp, and the log
2
fold ratio between average tags in treated and
DMSO samples was calculated. Genes presenting a ratio away from
the distribution median 6MADforanyoftheTA,TNF,andTA+
TNF conditions were classified as differentially expressed. An addi-
tional filter based on an exponential model, which empirically de-
termines thresholds on the minimum sum of tags in DMSO and
treatment samples required for a gene to be called expressed, in re-
lation to the gene body length was applied. In order to identify
groups of genes presenting similar expression patterns among the
TA, T NF, an d TA +TNF treatments, k-means clustering was per-
formed using the Pearson correlation distance. The optimal number
of clusters (six) was determined using the Gap statistic (Tibshirani
et al. 2001) coupled with an iterative procedure implemented in
Gene ARMADA software (Chatziioannou et al. 2009). Over-repre-
sented GO categories for each cluster of RNAPII-regulated genes
were detected using WebGestalt (Zhang et al. 2005). The median
distance from the summit of GR- or p65-binding sites to the closest
regulated gene TSS was calculated using the web base tool ‘‘fetch
closest feature’’ of Galaxy (Blankenberg et al. 2007).
Peak assignment and enrichment analysis
In order to associate the putative GR- and p65-binding sites with
regulated genes for each condition, the distance of peaks falling
within 610 Kb relative to the TSS of each regulated gene was
calculated so as to obtain a first distribution of nearby peaks to
each gene (one peak is assumed to be assigned to multiple genes
and vice versa). The final distribution of assigned peaks was
obtained by applying the hypergeometric test to each peak lo-
cated near regulated genes. This method detected peaks that are
close to groups of regulated genes in an over-represented manner,
relative to the whole genome that was used as background. To
determine whether a peak set was enriched to certain clusters or
not, the ratio of assigned peaks to each cluster, to the total number
of genes in that cluster was used to determine a ‘‘peak per gene
index’’ for each peak set and for each condition, according to the
following formula:
ETi
Cj
=log2
#Tipeaks assigned to Cj
Cj

!
;
where T
i
is the treatment (i=GR_TA, GR_TA +TNF, p65_TNF,
p65_TA +TNF), C
j
is the cluster of genes (j=1...6) and | C
j
| rep-
resents the cluster cardinality, or the number of genes in cluster j.
In order to assess whether the enrichment scores occur by chance,
the same peak sets were assigned to random sets of genes spanning
the whole genome, devised to clusters of the same cardinality as
the original ones. This procedure was repeated for n=100 times,
and statistical significance was assessed using as statistical mea-
surement a modification of the estimation of the Achieved Signifi-
cance Level for bootstrap (Efron and Tibshirani 1993):
^
pTi
Cj
=
#ETi
Cj;random #ETi
Cj;nonrandom
no
n;
that is, the ratio of the number of times random enrichment is
found greater than or equal to the nonrandom enrichment, to the
number of repetitions. A cluster was enriched for a peak set if ^
pwas
<0.01. Supplemental Table 4 summarizes the assigned GR and p65
peaks.
Rao et al.
1414 Genome Research
www.genome.org

Motif search
The coordinates of peak maxima for each peak set were extended
by 200 bp on each side to create 400-bp genomic regions. De novo
motif search was performed using a combination of widely used
software coupled with motif quality control, clustering, and sta-
tistical significance metrics (van Heeringen and Veenstra 2011) in
each set of peaks for each treatment and for varying motif lengths
(6–25 bp). The top scoring clustered motifs were curated and
converted to Position Weight Matrices (PWMs), which were com-
pared against the JASPAR v3 database using STAMP (Mahony and
Benos 2007). The data-driven PWMs were subsequently used to
scan all peak subsets for the occurrence of the respective motifs as
previously described (Smeenk et al. 2008). A specific sequence
background model was constructed for each scan using random
nonbound sequences.
Data access
All ChIP-seq raw data files have been submitted to the NCBI Gene
Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/
geo/) under accession no. GSE24518.
Acknowledgments
We thank Simon van Heeringen and Arie B. Brinkman from the
Department of Molecular Biology, Radboud University, Nijmegen,
for generously providing some data analysis tools. This project is
supported by the TI-Pharma (Netherlands, grant T2-101-1) and the
European Union (MTKD-CT-2004-509836: ‘‘MACROGENEX’’).
References
Adcock IM, Nasuhara Y, Stevens DA, Barnes PJ. 1999. Ligand-induced
differentiation of glucocorticoid receptor (GR) trans-repression and
transactivation: preferential targetting of NF-kappaB and lack of I-kappaB
involvement. Br J Pharmacol 127: 1003–1011.
Akkers RC, van Heeringen SJ, Jacobi UG, Janssen-Megens EM, Francoijs KJ,
Stunnenberg HG, Veenstra GJ. 2009. A hierarchy of H3K4me3 and
H3K27me3 acquisition in spatial gene regulation in Xenopus embryos.
Dev Cell 17: 425–434.
Baeuerle PA. 1998. Pro-inflammatory signaling: last pieces in the NF-kappaB
puzzle? Curr Biol 8: R19–R22.
Barnes PJ. 1997. Nuclear factor-kappa B. Int J Biochem Cell Biol 29: 867–870.
Barnes PJ, Karin M. 1997. Nuclear factor-kappaB: a pivotal transcription
factor in chronic inflammatory diseases. N Engl J Med 336: 1066–1071.
Beato M, Herrlich P, Schutz G. 1995. Steroid hormone receptors: many
actors in search of a plot. Cell 83: 851–857.
Beck IM, Vanden Berghe W, Vermeulen L, Bougarne N, Vander Cruyssen B,
Haegeman G, De Bosscher K. 2008. Altered subcellular distribution of
MSK1 induced by glucocorticoids contributes to NF-kappaB inhibition.
EMBO J 27: 1682–1693.
Blankenberg D, Taylor J, Schenck I, He J, Zhang Y, Ghent M, Veeraraghavan
N, Albert I, Miller W, Makova KD, et al. 2007. A framework for
collaborative analysis of ENCODE data: making large-scale analyses
biologist-friendly. Genome Res 17: 960–964.
Carroll JS, Meyer CA, Song J, Li W, Geistlinger TR, Eeckhoute J, Brodsky AS,
Keeton EK, Fertuck KC, Hall GF, et al. 2006. Genome-wide analysis of
estrogen receptor binding sites. Nat Genet 38: 1289–1297.
Cato AC, Wade E. 1996. Molecular mechanisms of anti-inflammatory action
of glucocorticoids. Bioessays 18: 371–378.
Chatziioannou A, Moulos P, Kolisis FN. 2009. Gene ARMADA: an integrated
multi-analysis platform for microarray data implemented in MATLAB.
BMC Bioinformatics 10: 354. doi: 10.1186/1471-2105-10-354.
Coutinho AE, Chapman KE. 2010. The anti-inflammatory and
immunosuppressive effects of glucocorticoids, recent developments and
mechanistic insights. Mol Cell Endocrinol 1: 2–13.
De Bosscher K, Haegeman G. 2009. Minireview: latest perspectives on
antiinflammatory actions of glucocorticoids. Mol Endocrinol 23: 281–291.
De Bosscher K, Vanden Berghe W, Haegeman G. 2003. The interplay
between the glucocorticoid receptor and nuclear factor-kappaB or
activator protein-1: molecular mechanisms for gene repression. Endocr
Rev 24: 488–522.
Denissov S, van Driel M, Voit R, Hekkelman M, Hulsen T, Hernandez N,
Grummt I, Wehrens R, Stunnenberg H. 2007. Identification of novel
functional TBP-binding sites and general factor repertoires. EMBO J 26:
944–954.
Efron B, Tibshirani R. 1993. An introduction to the bootstrap. Chapman & Hall/
CRC, Boca Raton, FL.
Freeman BC, Yamamoto KR. 2002. Disassembly of transcriptional regulatory
complexes by molecular chaperones. Science 296: 2232–2235.
Fullwood MJ, Liu MH, Pan YF, Liu J, Xu H, Mohamed YB, Orlov YL, Velkov S,
Ho A, Mei PH, et al. 2009. An oestrogen-receptor-alpha-bound human
chromatin interactome. Nature 462: 58–64.
Gottlicher M, Heck S, Herrlich P.1998. Transcriptional cross-talk, the second
mode of steroid hormone receptor action. J Mol Med 76: 480–489.
Hakim O, John S, Ling JQ, Biddie SC, Hoffman AR, Hager GL. 2009.
Glucocorticoid receptor activation of the Ciz1-Lcn2 locus by long range
interactions. J Biol Chem 284: 6048–6052.
Hamilton T, Novotny M, Pavicic PJ Jr., Herjan T, Hartupee J, Sun D, Zhao C,
Datta S. 2010. Diversity in post-transcriptional control of neutrophil
chemoattractant cytokine gene expression. Cytokine 52: 116–122.
Hayden MS, Ghosh S. 2008. Shared principles in NF-kappaB signaling. Cell
132: 344–362.
Hollenberg SM, Evans RM. 1988. Multiple and cooperative trans-activation
domains of the human glucocorticoid receptor. Cell 55: 899–906.
Ito K, Barnes PJ, Adcock IM. 2000. Glucocorticoid receptor recruitment of
histone deacetylase 2 inhibits interleukin-1beta-induced histone H4
acetylation on lysines 8 and 12. Mol Cell Biol 20: 6891–6903.
Jonat C, Rahmsdorf HJ, Park KK, Cato AC, Gebel S, Ponta H, Herrlich P.1990.
Antitumor promotion and antiinflammation: down-modulation of AP-
1 (Fos/Jun) activity by glucocorticoid hormone. Cell 62: 1189–1204.
Kamei Y, Xu L, Heinzel T, Torchia J, Kurokawa R, Gloss B, Lin SC, Heyman
RA, Rose DW, Glass CK, et al. 1996. A CBP integrator complex mediates
transcriptional activation and AP-1 inhibition by nuclear receptors. Cell
85: 403–414.
Karin M. 1998. New twists in gene regulation by glucocorticoid receptor: is
DNA binding dispensable? Cell 93: 487–490.
Konig H, Ponta H, Rahmsdorf HJ, Herrlich P. 1992. Interference between
pathway-specific transcription factors: glucocorticoids antagonize
phorbol ester-induced AP-1 activity without altering AP-1 site
occupation in vivo. EMBO J 11: 2241–2246.
Kralli A, Yamamoto KR. 1996. An FK506-sensitive transporter selectively
decreases intracellular levels and potency of steroid hormones. J Biol
Chem 271: 17152–17156.
Lim CA, Yao F, Wong JJ, George J, Xu H, Chiu KP, Sung WK, Lipovich L, Vega
VB, Chen J, et al. 2007. Genome-wide mapping of RELA(p65) binding
identifies E2F1 as a transcriptional activator recruited by NF-kappaB
upon TLR4 activation. Mol Cell 27: 622–635.
Lin CY, Vega VB, Thomsen JS, Zhang T, Kong SL, Xie M, Chiu KP, Lipovich L,
Barnett DH, Stossi F, et al. 2007. Whole-genome cartography of estrogen
receptor alpha binding sites. PLoS Genet 3: e87. doi: 10.1371/
journal.pgen.0030087.
Mahony S, Benos PV. 2007. STAMP: a web tool for exploring DNA-binding
motif similarities. Nucleic Acids Res 35: W253–W258.
Martens JH, Brinkman AB, Simmer F, Francoijs KJ, Nebbioso A, Ferrara F,
Altucci L, Stunnenberg HG. 2010. PML-RARalpha/RXR alters the
epigenetic landscape in acute promyelocytic leukemia. Cancer Cell 17:
173–185.
Martone R, Euskirchen G, Bertone P, Hartman S, Royce TE, Luscombe NM,
Rinn JL, Nelson FK, Miller P, Gerstein M, et al. 2003. Distribution of NF-
kappaB-binding sites across human chromosome 22. Proc Natl Acad Sci
100: 12247–12252.
McEwan IJ, Wright AP, Gustafsson JA. 1997. Mechanism of gene expression
by the glucocorticoid receptor: role of protein-protein interactions.
Bioessays 19: 153–160.
Necela BM, Cidlowski JA. 2004. Mechanisms of glucocorticoid receptor
action in noninflammatory and inflammatory cells. Proc Am Thorac Soc
1: 239–246.
Nicolaides NC, Galata Z, Kino T, Chrousos GP, Charmandari E. 2010. The
human glucocorticoid receptor: molecular basis of biologic function.
Steroids 75: 1–12.
Nielsen R, Pedersen TA, Hagenbeek D, Moulos P, Siersbaek R, Megens E,
Denissov S, Borgesen M, Francoijs KJ, Mandrup S, et al. 2008. Genome-
wide profiling of PPARgamma:RXR and RNA polymerase II occupancy
reveals temporal activation of distinct metabolic pathways and changes
in RXR dimer composition during adipogenesis. Genes Dev 22: 2953–
2967.
Phuc Le P, Friedman JR, Schug J, Brestelli JE, Parker JB, Bochkis IM, Kaestner
KH. 2005. Glucocorticoid receptor-dependent gene regulatory
networks. PLoS Genet 1: e16. doi: 10.1371/journal.pgen.0010016.
Pratt WB, Toft DO. 2003. Regulation of signaling protein function and
trafficking by the hsp90/hsp70-based chaperone machinery. Exp Biol
Med 228: 111–133.
Global GR and NFKB crosstalk
Genome Research 1415
www.genome.org

Ray A, Prefontaine KE. 1994. Physical association and functional
antagonism between the p65 subunit of transcription factor NF-kappa B
and the glucocorticoid receptor. Proc Natl Acad Sci 91: 752–756.
Reddy TE, Pauli F, Sprouse RO, Neff NF, Newberry KM, GarabedianMJ, Myers
RM. 2009. Genomic determination of the glucocorticoid responsereveals
unexpected mechanisms of gene regulation. Genome Res 19: 2163–2171.
Schaaf MJ, Cidlowski JA. 2002. Molecular mechanisms of glucocorticoid
action and resistance. J Steroid Biochem Mol Biol 83: 37–48.
Siriani D, Mitsiou DJ, Alexis MN. 2003. Overexpressed glucocorticoid
receptor negatively regulates gene expression under conditions that
favour accumulation of non-hormone-binding forms of the receptor. J
Steroid Biochem Mol Biol 84: 171–180.
Smeenk L, van Heeringen SJ, Koeppel M, van Driel MA, Bartels SJ, Akkers RC,
Denissov S, Stunnenberg HG, Lohrum M. 2008. Characterization of
genome-wide p53-binding sites upon stress response. Nucleic Acids Res
36: 3639–3654.
So AY, Chaivorapol C, Bolton EC, Li H, Yamamoto KR. 2007. Determinants
of cell- and gene-specific transcriptional regulation by the
glucocorticoid receptor. PLoS Genet 3: e94. doi: 10.1371/
journal.pgen.0030094.
Sultan M, Schulz MH, Richard H, Magen A, Klingenhoff A, Scherf M, Seifert
M, Borodina T, Soldatov A, Parkhomchuk D, et al. 2008. A global view of
gene activity and alternative splicing by deep sequencing of the human
transcriptome. Science 321: 956–960.
Teferedegne B, Green MR, Guo Z, Boss JM. 2006. Mechanism of action of
a distal NF-kappaB-dependent enhancer. Mol Cell Biol 26: 5759–5770.
Tibshirani R, Walther G, Hastie T. 2001. Estimating the number of clusters in
a data set via the gap statistic. JR Stat Soc 63: 411–423.
Tsaprouni LG, Ito K, Punchard N, Adcock IM. 2002. Triamcinolone
acetonide and dexamethasome suppress TNF-alpha-induced histone H4
acetylation on lysine residues 8 and 12 in mononuclear cells. Ann N Y
Acad Sci 973: 481–483.
Vallabhapurapu S, Karin M. 2009. Regulation and function of NF-kappaB
transcription factors in the immune system. Annu Rev Immunol 27: 693–733.
Van Bogaert T, De Bosscher K, Libert C. 2010. Crosstalk between TNF and
glucocorticoid receptor signaling pathways. Cytokine Growth Factor Rev
21: 275–286.
van Heeringen SJ, Veenstra GJ. 2011. GimmeMotifs: a de novo motif
prediction pipeline for ChIP-sequencing experiments. Bioinformatics 27:
270–271.
Welboren WJ, van Driel MA, Janssen-Megens EM, van Heeringen SJ, Sweep
FC, Span PN, Stunnenberg HG. 2009. ChIP-Seq of ERalpha and RNA
polymerase II defines genes differentially responding to ligands. EMBO J
28: 1418–1428.
Yamamoto KR. 1985. Steroid receptor regulated transcription of specific
genes and gene networks. Annu Rev Genet 19: 209–252.
Zhang B, Kirov S, Snoddy J. 2005. WebGestalt: an integrated system for
exploring gene sets in various biological contexts. Nucleic Acids Res 33:
W741–W748.
Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum
C, Myers RM, Brown M, Li W, et al. 2008. Model-based analysis of ChIP-
Seq (MACS). Genome Biol 9: R137. doi: 10.1186/gb-2008-9-9-r137.
Received November 18, 2010; accepted in revised form June 28, 2011.
Rao et al.
1416 Genome Research
www.genome.org