Novel in vivo targets of ΔNp63 in keratinocytes identified by a modified chromatin immunoprecipitation approach

The transcription factor p63, which belongs to a family of genes that also include p53 and p73, plays an important role in the transcriptional regulation of many biological processes including development, differentiation and apoptosis[1, 2]. Interestingly, p63 exhibits a restricted spatio-temporal expression pattern with high levels reported in epithelial cells. Indeed, the function of p63 has been predominantly examined in stratified epithelium in many organs such as the skin, mammary glands, prostate etc. Both gain-of-function and loss-of-function studies have clearly demonstrated that p63 is a critical master regulator of the epithelial differentiation program[1, 3]. This is quite evident in the dramatic phenotype of the p63 knockout mice, which lack stratified epithelium and their derivatives in multiple tissues and organs[4, 5].

The biological function of p63 is mediated by several isoforms derived from distinct transcripts that are generated from a complex genomic structure. The p63 gene gives rise to two major transcript variants through the use of distinct promoters, which are located far apart from each other[6]. The proximal promoter located upstream of exon1 directs the expression of transcripts that encode for an amino terminal transactivation domain (TA), whereas an internal promoter embedded within the third intron controls the expression of transcripts that lack this domain (ΔN). In addition, both the TA and ΔN transcripts are differentially spliced at the 3' end to generate proteins with unique C-termini that are designated as α, β and γ isoforms of p63.

All isoforms of p63 share a DNA-binding and an oligomerization domain, which shows sequence conservation with p53. Hence, these proteins are capable of sequence specific DNA-binding to p53 response elements and related sequences. Although the ΔNp63 isoforms were initially thought to function by exerting dominant negative effects on TAp63, it is increasingly becoming clear that the ΔNp63 proteins also mediate direct transcriptional activation and repressor activities on target genes [7–10]. In view of the fact that there are high levels of ΔNp63 protein but not TAp63 in many epithelial cells and that the ΔNp63 isoform is the only form of p63 present in lower organisms, it is thought that ΔNp63 may be the primary mediator of the biological function of the p63 gene.

Since p63 is a transcription factor, it is likely that it governs the various cellular processes and developmental decisions by regulating specific target genes. Although p63 has been shown to regulate some well-characterized p53 responsive genes, it is becoming increasingly evident that there exists a unique set of p63 target genes[11]. This notion is further strengthened by the fact that p63 has a distinct functional role in development and that the DNA-binding activity of these two proteins exhibits clear differences[1, 12, 13]. It is also possible that both p53 and p63 can regulate common target genes as exemplified by maspin, IGFBP-3 and PERP, which though initially were thought to be regulated by p53 are now clearly proven to also be p63 targets [14–17]. To identify additional p63 targets, some laboratories have utilized an experimental model system where cells lacking p63, such as Saos2 or HEK 293, were assessed for global alterations in gene expression by microarray analysis under conditions where p63 is over expressed[12, 18, 19]. Although these and similar other studies have unearthed some interesting potential p63 targets, identifying the exact location of the cis-elements that mediate the transcriptional effects of p63 have remained elusive[20]. Hence, a direct examination of p63-responsive elements in epithelial cells, which express high levels of p63 under physiological conditions, is warranted.

Keratinocytes express high levels of ΔNp63 and have provided a useful model system to dissect the role of p63 and to study its target genes[10]. These studies have been facilitated by chromatin immunoprecipitation (ChIP) experiments to confirm known p63 binding sites within the promoters of selected downstream targets in vivo. Although ChIP can be also a powerful tool to identify previously unknown targets of transcription factors, technical limitations have often precluded such an experimental strategy. In this paper, we report our initial results with an improved genomic ChIP approach to isolate and characterize 62 potential genomic ΔNp63 binding sites in keratinocytes. These sites are associated with various genes involved in a wide variety of cellular processes including transcription, signaling and metabolism. As proof of principle we have confirmed the association of ΔNp63 with a subset of these targets both in vitro and in vivo. Finally, we demonstrate that the expression levels of the potential ΔNp63 target genes located close to these genomic binding locations are affected when p63 levels are lowered. Collectively, our studies highlight the robustness of our experimental strategy to unearth novel targets for ΔNp63 and thus can be extended to other transcription factors.

Identifying target genes is critical in understanding the mechanism by which the transcription factor p63 regulates the intricate process of epithelial development and differentiation. Towards this end, several laboratories have undertaken a target discovery approach that involves manipulating the activity of p63 followed by gene expression analyses[12, 18, 23, 24]. Although such genome-wide microarray analysis has been valuable and has led to the identification of several transcripts that are potentially turned on and off by p63, these experiments have several limitations. First, it is difficult to ascertain whether a responsive gene is a direct or an indirect target of p63. Second, in experiments based on over expression of p63, it is likely that the response of some genes is driven by the exaggerated non-physiological concentration of this transcription factor. Third, the selected cell types that have often been utilized for these studies do not express endogenous p63 and thus may not represent the normal cellular milieu under which p63 operates. Finally, such studies a priori do not provide any information on the location of the p63-responsive elements that control the expression of target genes. Search for such regulatory elements is further hampered by the degenerate nature of the p63 DNA-binding sequence. Indeed, bioinformatics approaches often identify an unrealistically large number of potential p63 binding sites without distinguishing those that are functionally relevant. Hence, despite the identification of many p63 target genes, it has been difficult to ascertain whether p63 is involved in direct regulation of those genes by binding to their promoter and/or enhancer elements.

In contrast, here we have undertaken a direct approach of ChIP in combination with sequencing and mapping of isolated genomic DNAs to locate binding sites for ΔNp63 and its potential target genes in vivo. Indeed, such ChIP strategies have been successfully utilized to not only determine whether a candidate genomic site is occupied by a specific transcription factor but to also identify transcription factor targets. However, in such cases it has been technically challenging to separate out the specific regulatory DNA segment that is bound to the transcription factor from the vast excess of non-specifically precipitated DNA. To overcome these challenges, we have fine-tuned the ChIP procedure and modified it to successfully identify in vivo ΔNp63 target genes from HaCaT cells, which express high levels of ΔNp63, the major isoform of p63 protein in keratinocytes. First, since the immunoprecipitated DNA is of limited amount, we utilized a ligation-mediated PCR technique, in which the ChIP products were ligated to linkers and amplified by PCR using the linker sequence. This allowed us to obtain sufficient material for subsequent purification and cloning. Second, to eliminate non-specific DNA fragments prior to cloning the ChIPed DNA, we purified the PCR-amplified fragments by incubating them with agarose beads containing GST-ΔNp63α. This step ensures that fragments that contain DNA binding-sites for ΔNp63 are selectively enriched and can be separated from non-specific DNA. The potential caveats to such a purification step are that some DNA fragments containing genuine ΔNp63-targets might fail to bind to recombinant GST-ΔNp63α or that non-specifically precipitated DNA fragments can fortuitously bind to GST-ΔNp63α under the in vitro conditions of the assay. In addition, genomic segments to which ΔNp63 are recruited indirectly for example through protein-protein interactions are likely to be under-represented in this experimental scheme. However, our results suggest that the benefits of utilizing such an enrichment step clearly outweigh the potential drawbacks of the overwhelming background of non-specific genomic immunoprecipitated DNA.

One of the interesting findings that emerge from our studies is that the vast majority of these ΔNp63-binding sites are not localized to the proximal promoters of the target genes. This is in agreement with recent genome-wide target identification studies on other transcription factors such as Sp1, p53 and Myc, which demonstrate that a majority of the binding sites are not in the proximal promoter but rather scattered[25]. Interestingly, a large subset of the ΔNp63 targets were not close to any known gene; these could represent potential long range enhancers or unknown genes that have not been annotated or non-coding RNAs. Taken together, our study reinforces the need to not limit examination of the proximal promoter region only when searching for direct targets of transcription factors.

Our improved ChIP-based screening strategy allowed us to identify a large number of ΔNp63-response elements and their corresponding targets. Here we have examined a subset of the ΔNp63 target genes in detail. Our data show that the majority of the candidate target genes are co-expressed with ΔNp63 in HaCaT cells and that the specific segment isolated from the cloning of DNA immunoprecipitated is indeed occupied by ΔNp63 as determined by independent ChIP experiments. We also demonstrate that these segments harbor at least one p63 response element that can bind to both recombinant ΔNp63 and ΔNp63 present in HaCaT nuclear extract and that these segments can be transcriptionally activated by ΔNp63α in transient transfection assays. Finally, reduction in the levels of ΔNp63 in HaCaT cells led to significant decrease in transcript levels for a majority of the p63 target further validating the relevance of p63 in regulating these genes and confirming that the eleven genes fulfill the criteria to be direct transcriptional targets of p63. Only two genes, MXD3 and B4GALT1 were de-repressed even though in transient transfection experiments the DNA fragments immunoprecipitated by ΔNp63 corresponding to both MXD3 and B4GALT1 were positively activated. The fact that the majority of the targets identified by our approach are activated by ΔNp63 may reflect a bias towards selection of a high affinity p63-binding site due to the incorporation of an additional purification step introduced in the cloning procedure. However it is clear from numerous studies that ΔNp63 also represses transcription (as evident by our results for MXD3 and B4GALT1). This might potentially be mediated through interaction of ΔNp63 with non canonical response sites. Collectively, our data support the growing consensus that the ΔNp63 isoform is primarily involved in transcriptional activation of target genes rather than merely acting as a dominant negative that opposes the function of TAp63.

The diverse nature of the targets identified in our study reinforces the notion that p63 plays a role in complex biological pathways that affect a wide variety of cellular processes. As a critical regulator of keratinocyte development and differentiation, it is not difficult to imagine p63 as a focal point in the transcriptional network and cascades. This role of p63 as master regulator is supported by a large number of potential targets identified in this study that are transcription factors such as MXD3 and STAT6. Similarly our identification of NOTCH2NL as a ΔNp63 target gene contributes to the growing list of Notch family members and effectors of this pathway that are regulated by p63[19, 26]. The p63 targets uncovered in our study include genes previously linked to p63, such as DDR1. Discoidin domain receptor 1 (DDR1) is a unique receptor tyrosine kinase activated by various types of collagens and is known to play a role in epithelial cell attachment, migration, survival, and proliferation[27, 28]. Upregulation of DDR1 by p63 has been observed in several cases by gain of function studies; our ChIP analysis now clearly demonstrates that DDR1 is a direct transcriptional target[12, 29]. What is also clear from surveying the list of ΔNp63 targets is that there are predicted genes with unknown functions. Regulation of these unknown genes by ΔNp63 may be one underlying mechanism by which p63 may mediate some of its myriad biological activity.

As this manuscript was in preparation, two independent studies were published in which genome-wide p63 binding sites were determined by ChIP-Chip technology. In the first study, the Mantovani group performed ChIP-on-Chip experiments on HaCaT cells using two different platforms; the 12K CpG islands and 12K promoter arrays that led to the identification of ~200 target loci [30]. In the second study, Yang et al utilized genome wide tiled microarrays covering the entire nonrepetitive human genome, that lead to identification of ~5800 target sites for p63[31]. These studies were performed in ME180 cervical carcinoma cell line with the 4A4 anti-p63 antibody that recognizes all p63 isoforms. A closer look at these data and that of ours presented in this paper show that the data sets generated from these studies display significant overlap but also distinct patterns. Indeed twenty-five percent of the ΔNp63 target genes that we report in this work have also been shown to be putative p63 target genes based on these two studies (Table 1 and 3). The differences in the data sets probably result from experimental noise associated with the ChIP-ChIP experiments, differences in cell culture conditions such as cell type and/or proliferation state and the distinct antibodies utilized in these studies. The presence of different p63 isoforms and their unique properties surely adds to this complexity.

An interesting point to be considered is that for many of the p63 targets, it is likely that there are more than one p63-response elements. These elements are thus scattered in the proximal promoter, in the intragenic region and at a distance far from both the 5' and 3' end of the gene itself. Pertinent to this notion, we find a p63-response element in the third intron of the DDR1 gene, whereas the data from Yang et al., point to p63-response element located 5' of the gene [31]. It is likely that both these sites are bona-fide p63 binding sites. The presence of multiple p63-response elements reflects the importance of p63 in regulating DDR1 gene expression. This situation is similar to a well-characterized p63 target, the K14 gene, where p63 binding sites are present both in the proximal promoter and a distal enhancer[10]. Because each experimental condition captures only a snapshot of the entire p63 directed transcriptome, it is important that additional studies are performed to take into account the dynamic cellular environment in which p63 operates. Our approach described in this paper validates the use of ChIP coupled with an enrichment strategy to identify transcriptional targets in vivo and demonstrates the feasibility of such an approach that can be applied on a large scale. The various complementary strategies should provide a starting point to dissect out the network of relevant p63 targets and offer a strong basis for the elucidation of the gene regulatory pathways that are controlled by p63 in keratinocytes and other cell types.