Using a global map of p53 transcription factor binding sites in the human genome that was generated by the ChIP-seq method [19
], we searched for p53 binding at locations adjacent to miRNAs that had been shown by expression profiling to be differentially expressed in heart failure [6
]. At least one p53 binding site was located within 3,000 bp upstream or downstream of mir-15b, mir-21 and mir-125b. We tested and found that the expression of mir-21, but not mir-15b or mir-125b, was responsive to p53 activation by doxorubicin (Additional file 2
). Similarly, mir-21 was upregulated by hypoxia, which is another stimulus known to activate p53 in cardiac cells [11
mir-21 belongs to a conserved miRNA family with single recognizable orthologs in many different invertebrate species [20
]. A previous gene structure study of mir-21 identified a promoter sequence (miPPR21) in a highly conserved region approximately 2.5 kb upstream of the putative p53-binding site (which we called 'GIS') [21
] (Figure ). Aside from mir-21 itself and miPPR21, GIS is the only other region of significant sequence conservation in this genomic region (Figure ; Additional file 3
). However, analysis of GIS revealed not a p53 consensus motif, but a consensus motif for κB binding (Additional file 3
). This motif is consistent with that reported in another genome-wide analysis of ChIP mapping NF-κB/RELA binding sites [22
Figure 1 Genomic structure of mir-21. Location of a previously described promoter (miPPR-21) , our putative regulatory region (GIS) , a H3K4me3 binding site as determined by previous ChIP-seq , and a STAT3 binding site according to Loffler et al. (more ...)
In the stressed myocardium, mir-21 is significantly upregulated in cardiac fibroblasts and is responsible for fibroblast growth factor secretion as well as for the extent of interstitial fibrosis in heart failure via its effect on its target gene, Spry1
]. Moreover, the therapeutic benefit of inhibiting mir-21 in heart failure was also demonstrated. We therefore focused our attention on mir-21 expression in cardiac fibroblasts and found that, as with hypoxia, the hypoxia-mimetic DFX, which effectively activates p53 in vitro
], also upregulated mir-21 in primary rat cardiac fibroblasts (Figure ). It was also recently shown that NF-κB signaling is critical for the response to hypoxia [24
] because hypoxia may directly induce NF-κB activation through a complex sequence of signals involving decreased prolyl hydroxylase-mediated prolyl hydroxylation of IKKβ leading to phosphorylation-dependent degradation of the endogenous NF-κB inhibitor, IκBα, and nuclear translocation of NF-κB [25
]. Consistent with this and other data [26
], we found that DFX induced NF-κB/RELA nuclear accumulation and this was significantly inhibited by the cell-permeable NF-κB inactivator quinazoline [27
] (1 μM NFI; Figure ). Quinazoline (6-amino-4-(4-phenoxyphenylethylamino)) specifically inhibits NF-kB activation and nuclear translocation [28
]. Correspondingly, NFI significantly inhibited DFX-induced mir-21 upregulation (Figure ). We also noted that DFX induced p53 nuclear accumulation as predicted but mir-21 levels were effectively inhibited by NFI, despite unchanged levels of nuclear p53 following DFX+NFI treatment (Figure ). These data suggested that NF-κB was the primary mediator of mir-21 induction by DFX and/or p53 induction of mir-21 required activation of NF-κB.
Figure 2 p53 and NF-κB cooperate to induce mir-21. (a) Primary neonatal rat cardiac fibroblasts were treated with or without DFX and the NF-κB inactivator (NFI; 1 μM quinazoline) and mir-21 was quantified using the TaqMan miRNA assay. (more ...)
Next we tested the activity of the putative p53-binding site GIS by cloning it upstream of firefly luciferase and examining reporter gene expression. Supporting the hypothesis that p53 requires and cooperates with NF-κB/RELA, p53 alone did not upregulate luciferase activity, whereas p53 significantly augmented the activity that was induced by NF-κB/RELA (Figure ). As before, inactivation of NF-κB by NFI abrogated GIS-driven gene expression. Mutation or deletion of the κB-consensus motif in this regulatory sequence reduced p53-RELA-mediated luciferase reporter gene expression by 50% and 30%, respectively (Figure ). The previously described mir-21 promoter (miPPPR21) approximately 2.5 kb upstream of GIS was shown to respond through conserved AP1 and PU.1 binding sites [30
]. Neither p53 nor NF-κB/RELA upregulated expression of the reporter construct based on this promoter (miPPPR21-luciferase; Additional file 4
), indicating that p53/NF-κB regulated mir-21 expression through GIS but not miPPPR21.
To determine the necessity for NF-κB/RELA in mir-21 induction by DFX or p53, we incubated RelA-/- MEF cells with or without DFX and detected no change in mir-21 levels (Figure ), despite DFX-induced activation of p53 as shown by an increase in p53 target gene expression (MDM2 and BAX) (Figure ) and an increase in reporter activity using a luciferase construct driven by 13 p53-binding sites (PG13-luciferase, data not shown). Importantly, RelA-/- MEF cells reconstituted with ectopic RelA showed rescue of DFX induced mir-21 upregulation (Figure ).
Our results raise the possibility that RELA and p53 interact with the putative regulatory region GIS. Thus, we performed ChIP using anti-RELA and anti-p53 antibodies and found that the GIS region was occupied by both RELA and p53 in vivo
(Figure ). Once again, NFI disrupted the GIS-p53 association, indicating that p53 binding required RELA (Figure ). To determine whether RELA and p53 co-exist in a single molecular complex, we first performed co-immunoprecipitation assays and found an interaction between endogenous RELA and p53 proteins that was disrupted by NFI (Figure ). The p53-RELA interaction was direct and dependent on the carboxy-terminal transactivation domain of RELA because a purified recombinant GST fusion protein of the RELA carboxy-terminal domain, but not the amino terminus DNA binding domain, was sufficient to interact with p53 (Additional file 5
). Next we performed a sequential ChIP assay (re-ChIP) in which we initially performed ChIP with a p53 antibody, released the immunoprecipitated chromatin and then performed another ChIP using a RELA antibody. GIS was significantly enriched by p53-RELA re-ChIP and this association was disrupted by NFI, indicating that RELA and p53 were simultaneously residing at the GIS genomic location (Figure ). Furthermore, we performed oligonucleotide pulldown where the genomic sequence of GIS was synthesized, biotinylated, immobilized onto streptavidin-coated beads, and incubated with protein lysates from cells that had been treated with or without DFX. The GIS oligonucleotide, but not a scrambled control, effectively pulled down p53 and RELA in DFX-treated cells (Figure ). Similarly, we found that the GIS oligonucleotide also pulled down the NF-κB subunit p50 (Figure ) but not p52 (data not shown), suggesting that the p53-RELA complex included this subunit of NF-κB.
Figure 3 p53 and NF-κB form a complex and occupy the putative GIS regulatory region simultaneously. (a) ChIP was performed on cardiac fibroblasts with or without DFX using antibodies against either p53 or RELA. Results show fold enrichment of real-time (more ...)
The presence of a κB motif instead of a p53 consensus sequence on GIS prompted us to consider if p53 was behaving as a co-factor and if the p53-GIS interaction was indirect and independent of the p53 DNA binding domain. We therefore performed another oligonucleotide-pull-down using lysates from p53-deficient cells (Soas2) pre-transfected with vector only, wild-type p53 or p53 bearing a mutation in the DNA binding domain (p53R175H). Both wild-type and mutant p53 associated with the GIS oligonucleotide (Figure ). Consistent with this, we also found that the p53-RELA interaction was independent of the p53 DNA binding domain (Figure ); moreover, mutant p53 was potentially capable of upregulating mir-21 expression (Figure ). Taken together, our data support the conclusion that p53 does not contribute to the GIS binding interface but instead behaves as a co-factor in this molecular complex and utilizes RELA for its association with GIS to transactivate mir-21 expression.
Figure 4 NF-κB forms a complex at the GIS regulatory region with both wild-type p53 and p53 with a DNA-binding domain mutation. (a) p53-deficient Soas2 cells were transfected with wild-type p53 (p53WT), DNA binding domain mutant p53 (p53R175H) or vector (more ...)
Sites of active chromatin at regulatory sequences are associated with the characteristic Histone-3 mark of lysine-4 tri-methylation (H3K4me3) [31
]. We therefore performed ChIP using a specific H3K4me3 antibody and detected a marked enrichment of GIS compared to miPPR21 (Additional file 6
). The association of the GIS genomic location with H3K4me3 has also been mapped by others in a genome-wide ChIP scan using human embryonic stem cells [32
Since levels of mir-21 are significantly elevated in dilated human hearts and murine hearts with decompensated hypertrophy [6
], and Thum et al.
] recently validated the therapeutic value of targeting mir-21 in a mouse model of heart failure, we undertook further analysis using human left ventricular tissues from patients who had undergone cardiac transplantation for end-stage dilated cardiomyopathy and age-matched normal control left ventricular tissues from individuals involved in road traffic accidents (Additional file 1
). Despite different etiologies of heart failure (such as ischemic and non-ischemic), end-stage cardiomyopathy is collectively characterized by disease processes and molecular pathways such as apoptosis, dysregulated calcium signaling, decompensated contractility, G-protein coupled receptor down-regulation, maladaptive angiogenesis and fibrosis. Hence, as predicted for our heterogeneous series of end-stage cardiomyopathic hearts, we found significant nuclear accumulation of RELA in both myocytes and non-myocytes from cardiomyopathic hearts compared to control (Figure ). Significant p53 activation was detectable only in non-myocytes (Figure ). As a functionally significant output of the piggyback mechanism, we found that both p53 and RELA were simultaneously resident at the GIS site, and this association was significantly enriched in cardiomyopathic hearts compared to normal controls (Figure ).
Figure 5 The p53-NF-κB complex is present at the GIS regulatory region in human dilated cardiomyopathic hearts. (a-c) Human left ventricular tissue sections immunostained for NF-κB/RELA showed that NF-κB/RELA (in brown) was predominantly (more ...)
We predicted that although different mechanisms may determine p53-RELA complex formation and its chromatin association, the specificity for this complex at some genomic locations such as mir-21 GIS may be assisted by additional factors. In order to investigate this, we performed parallel high-throughput sequencing with eight human cardiac sequential chromatin immunoprecipitates (four diseased and four controls; Additional file 1
). We identified 26,628 genomic locations in normal hearts and 33,578 in diseased hearts (model fold = 100) aligned to the reference human genome (Gene Expression Omnibus [GES21356]). Among these, 12,311 tag locations, excluding repetitive elements, were unique to disease and had significant conservation across species (Additional files 7
, and listed in Additional file 9
). Of note, only 3% (381 out of 12,311) were identical to a previous global ChIP for RELA [22
], although in the latter, ChIP was generated using a non-cardiac cell line and a stimulus unrelated to hypoxia. Including a location adjacent to mir-21, 1,344 out of 12,311 (10.9%) were identified to contain the bona fide
κB consensus motif. This observation suggested that a diverse range of p53-RELA complexes may be involved in its chromatin association and most appear to be independent of the κB motif. Nonetheless, using CEAS [17
], we analyzed these 1,344 genomic locations and the previous global RELA ChIP [22
] in parallel. Several transcription factor motifs were overrepresented and common to both our subset of locations and the global RELA ChIP, except for STAT1, STAT3, STAT5 and STAT6 (Additional files 10
), with the STAT3 motif being the most prominent. The JAK/STAT3 pathway is particularly important for the secretory function and survival of cardiac fibroblasts [33
]. Moreover, in multiple myeloma cancer cell lines, mir-21 expression is STAT3-mediated and two conserved STAT3 binding sites lie upstream of mir-21 [34
]. We therefore examined whether the p53-RELA piggyback mechanism was STAT3-dependent. By using structure-based virtual screening, the cell-permeable compound S3I-201 was previously identified to bind to the STAT3 Src homology 2 (SH2) domain, and inhibit STAT3 dimerization, phosphorylation and DNA-binding [35
]. We used S3I-201 and found that STAT3 inhibition disrupted p53-RELA- GIS association (Figure ), and inhibited mir-21 upregulation (Figure ), without altering p53 and RELA nuclear abundance (Figure ). Moreover, p53-RELA remained in complex, although STAT3 inhibition had blocked this complex from interacting with GIS and disrupted the interaction between STAT3 and p53-RELA complex (Figure ). Using Stat3-/-
MEF cells, we found that STAT3 deficiency blocked DFX-induced mir-21 but this was recovered in Stat3-/-
MEF cells that were reconstituted with wild-type Stat3
(Figure ), further demonstrating that STAT3 is required for p53-RELA-mediated mir-21 gene expression.
Figure 6 p53-NF-κB mediated mir-21 expression is dependent on STAT3. (a) Cardiac fibroblasts were treated with DFX with or without an inhibitor of STAT3 DNA binding (S3I-201), and cell lysates were incubated with streptavidin-coated beads on which biotinylated (more ...)