Genomic and computational approaches to dissect the mechanisms of STAT3's universal and cell type-specific functions.

STAT3 is the quintessential pleiotropic transcription factor with many biological roles throughout development as well as in multiple adult tissues. Its functional heterogeneity is encoded in the range of genome-wide binding patterns that specify different regulatory networks in distinct cell types. However, STAT3 does not display remarkable DNA binding preferences that may help correlate specific motifs with individual biological functions or cell types. Therefore, achieving a detailed understanding of the regulatory mechanisms that endow STAT3 (or any other pleiotropic transcription factor) with such a rainbow of functions is not only a central problem in biology but also a fiendishly difficult one. Here we describe key genomic and computational approaches that have shed light into this question, and present the two current models of STAT3 binding (universal and cell type-specific). We also discuss the role that the local epigenetic environment plays in the selection of STAT3 binding sites.

The Problem of (Understanding) Transcription Factor Pleiotropy

Transcription factors (TFs) bind to short DNA sequences where they combine with other co-factors to regulate the expression of target genes in specific epigenetic and nuclear contexts. Some of the most dramatic effects a TF can have are those related to cellular differentiation. For instance, MyoD alone is capable of trans-differentiating fibroblasts to myoblasts, and a combination of only four TFs (OCT4, SOX2, KLF4, and c-MYC) is sufficient to reprogram terminally differentiated fibroblasts into iPS cells that display an embryonic stem cell-like phenotype. Fibroblasts can also be transformed into tripotent neural precursor cells with a defined cocktail of TFs (BRN2, SOX2, and FOXG1). Clearly a TF’s ability to recognize some sort of “code” contained in the DNA region it binds to is essential for the successful execution of gene expression programs. However, the analysis of the DNA sequence preferences of TFs using a variety of high-throughput methods have underlined the general binding degeneracy of TF families,- indicating that there is no simple relationship between the DNA sequence a TF binds to and the biological program it executes. Therefore, mechanisms other than the mere binding of a TF to DNA must be involved in determining the cell type-specific functions of TFs. Complex models describing the interplay between groups of TFs, as well as the influence of the epigenetic environment on transcription, have been proposed to explain the functional specificity of TFs. However, none of these models are entirely satisfactory or comprehensive, and no broadly applicable rules have been described to explain the mechanisms whereby TFs discriminate and select specific binding sites genome-wide to perform specific functions.

The diversity of cells and tissues where a TF is expressed can serve as an approximation to infer its functional diversity. We may thus classify TFs into one of three categories: (1) TFs whose expression is restricted to a single cell-type (e.g., Oct4 [Pou5f1] is primarily expressed in embryonic stem cells); (2) TFs restricted to a single germ lineage, such as the SoxB1 subfamily of TFs (ectodermal) and STAT4 (primarily mesodermal); and (3) TFs that are ubiquitously expressed, such as most members of the STAT family of TFs (Fig. 1). It is this latter category of widely expressed TFs that presents the greatest intellectual challenge as they normally display a large variety of functions (pleiotropy), including opposing functions in distinct cell types, and sometimes within the same cell type too.

Figure 1. The functional diversity of a TF can be inferred from its expression pattern throughout the body. For instance, the expression of Oct4 (Pou5f1) is primarily restricted to ESCs, whereas most STAT family members are widely expressed in multiple mouse tissues except STAT4, which is primarily mesodermal. Sox2 and Sox3, like STAT4, are also restricted to a single developmental lineage, the ectoderm. RNA-seq data was extracted from the GEO database accessions: GSE20851, GSE20898, GSE29209, GSE29278, GSE31530, GSE33024, GSE34550, GSE36026, GSE39524, GSE39656, GSE39756, GSE40350, GSE40463, GSE42207, GSE42443, and GSE42880.

In this review we explore the general problem of understanding the regulatory mechanisms of multi-functional TFs by taking STAT3 as a prime example of a pleiotropic TF. Recent work that integrates high-throughput genomics and detailed computational analyses has shed new light into the mechanisms employed by STAT3 to perform completely different biological functions in distinct cell types. These models are specific to STAT3, but the concepts and tools involved in the analyses are applicable to other TFs, thus opening the door to a more thorough mechanistic understanding of TFs with complex functions.

The Dazzling Functional Diversity of STAT3

STAT3 is constitutively expressed and its genetic deletion is embryonic lethal, possibly due to an essential role in maintaining pluripotency. This early severe phenotype therefore masks many other functional defects associated with the loss of the STAT3 gene and which have been painstakingly teased apart using cell type-specific knockouts and carefully designed cell culture experiments. For instance, the roles of STAT3 in metabolism are many and varied, including links to obesity and glucose tolerance, as shown by the deletion of STAT3 in the neural system, hypothalamus, pancreatic islets, and adipose cells where it leads to a general impairment of metabolism- due to the critical role that STAT3 plays downstream of leptin signaling. Stat3S727A/− mice, which have reduced levels of STAT3 transcriptional activity, present growth retardation and a reduction in thymocyte numbers as well as defects in insulin growth factor-1 (IGF-1) and growth hormone (GH). This suggests that STAT3 might be indirectly responsible for promoting their expression, and hints at a positive feedback loop controlling growth as both IGF-1 and GH can activate STAT3.,

STAT3 is also essential in many developmental pathways. For example, in keratinocytes STAT3 is required for hair production, cell migration, and wound repair, whereas the genetic ablation of STAT3 in cardiomyocytes results in cardiac dysfunction as well as an increased sensitivity to inflammatory stimuli. Moreover, a liver-specific knockout resulted in the suppression of liver regeneration after a partial hepatectomy. The hematopoietic system is perhaps where STAT3 functions have been most intensely studied, although its role in hematopoietic progenitors remains controversial. The genetic deletion of STAT3 in blood progenitors leads to a higher rate of myeloid cell production, particularly macrophages and neutrophils. In this context, STAT3 is likely controlling proliferation and homeostatic control. STAT3 is also required later in hematopoiesis in Flt3L-dependent dendritic cell development, as well as for controlling the apoptosis of conventional dendritic cells. In B lymphocytes, STAT3 is required for the differentiation of IgG B cells; whereas in cytokine-stimulated (IL-6, IL-21, and IL-23) CD4+ T cells STAT3 specifies the differentiation of Th17 cells by binding at multiple sites in the vicinity of key genes, including Il17a and Il17f., STAT3 is also important for the immune-suppressive Treg cells through a direct action on Foxp3 expression.

Besides its involvement in development, STAT3 plays central roles in cellular responses to environmental stimuli. On the whole-organ level, an interferon-inducible conditional STAT3 knockout in liver cells results in a loss of correct inflammatory function in the liver. The stimulation of myeloid cells with IL-10 activates STAT3 and has an anti-inflammatory effect,- whereas STAT3 synergizes with glucocorticoid receptor signaling to regulate inflammation in the pituitary. Likewise, in fibroblasts IL-6 synergizes with IL-17A/NFκB signaling to engage an inflammatory response., Opposing effects of STAT3 have also been described within the same cell type: in DCs, the IL-6-mediated pro-inflammatory activity of STAT3 appears to be rapidly induced, followed by a decline, whereas the IL-10-mediated anti-inflammatory response is encoded in the sustained activity of STAT3. This change in duration of STAT3 activity is controlled by SOCS3, which binds to the phospho-Y759 residue of the IL-6 receptor (specifically the gp130 subunit) to inhibit downstream signaling through STAT3, and so blocks IL-6 activity. However SOCS3 does not bind to any analogous residue in the IL-10 receptor,, and so IL-10 can continue to sustain the activation of STAT3. In addition to the opposing effects of IL-6 and IL-10, complicating this picture further is the action of IL-21 on conventional DCs (a subtype of DCs) where it activates STAT3 to engage apoptosis. These observations in DCs highlight the complexity of STAT3 function: three different cytokines signal through STAT3 to produce three entirely distinct cellular responses in the same cell type.

Given its vast functional diversity it is not surprising that mutations in STAT3 are responsible for a plethora of diseases. These include the Hyper-IgE syndrome, a condition characterized by recurrent colds, pneumonia, eczema, scoliosis, and extreme elevation of IgE. The underlying cause of the Hyper-IgE syndrome remains unclear, and the phenotype is complex with various abnormalities throughout the immune system, particularly in T cell development as these patients present impaired T follicular helper cells and lack both CD4+ and CD8+ T memory cells. Interestingly many of the mutations identified in Hyper-IgE syndrome patients are located in the DNA binding domain of STAT3, and they impair but do not abolish STAT3’s DNA-binding ability. In contrast, gain-of-function mutations in STAT3 are rare but they do occur and are linked with cancer. For instance, several different mutations in STAT3 (primarily within the SH2 domain) are associated with inflammatory hepatocellular adenoma and activate transcription in the absence of a cytokine. STAT3 is associated with a wide array of cancers, especially for its role in promoting inflammation within and around the tumor as many tumors show excessive STAT3-mediated inflammation in combination with NFκB signaling., Although the exact relationship between STAT3 and the various types of cancer continues to be actively researched (“Is STAT3 a cause or a consequence?”), at the very least in the case of pancreatic ductal adenocarcinoma STAT3 is essential for both tumor initiation and progression where it cooperates with SOX2 to transform foregut basal progenitor cells. Moreover, STAT3 has been widely implicated in inflammatory diseases, such as Crohn disease: the IL-10 knockout mouse is the prototypical mouse model for Crohn disease and when STAT3 is knocked out during hematopoiesis the resulting mice display Crohn disease-like symptoms., Finally, an epithelial cell-specific knockout mouse was developed to study Sjögren syndrome, a systemic autoimmune disease where immune cells target exocrine glands, and which is characterized by the lack of Nfkbiz, a universal STAT3 target gene. The involvement of STAT3 in so many pathologies makes it a promising target molecule for anti-cancer treatment despite the general difficulty of inhibiting TFs with small molecules. Decoy oligonucleotides that specifically inhibit STAT3 may hold promise for the treatment of cancer.

In addition to STAT3’s role as a cytokine-activated TF, STAT3 has also been found inside mitochondria in multiple cell types. Here STAT3 does not bind mitochondrial DNA as might be expected, but instead forms a complex with GRIM-19 to bind to the electron transport chain, modulate reactive oxygen species production and confer a protective effect against ischemia., Additionally STAT3 binds to Stathmin to directly regulate microtubule dynamics in migrating T cells. These fascinating roles of STAT3 are independent of its ability to bind DNA and highlight the functional diversity of STAT3 to act not just as a TF but also as an adaptor protein.

STAT3 as a Model TF to Dissect the Mechanisms Regulating Cell Type-Specific Functions

Besides its large number of documented functions, one important aspect of STAT3 that makes it an ideal system for dissecting the working mechanisms of pleiotropic TFs is its two-step mode of activation upon cytokine stimulation, which results in directly measurable effects. Therefore, the activity of STAT3 can be controlled as a natural switch, and with careful manipulation of the culture environment its activity can be regulated. This is a very important advantage over many other TFs where elaborate experiments using artificial switches need to be performed to modulate their activity.

The DNA-Binding Specificity of STAT3 Cannot Be Correlated with Specific Functions or Cell Types

STAT3 is known to exist in two isoforms, α and β. The former represents the standard version of the molecule, whereas the β isoform is characterized by a C-terminal truncation of the transactivation domain. The β isoform was originally suspected to have dominant-negative effects, but it is actually capable of rescuing STAT3 α embryonic function, and mice survive until birth. However, STAT3 α knockout mice die rapidly after birth, indicating that the post-natal functions of STAT3 require the α isoform. STAT3 α knockout mice also show a complex difference in effecting an inflammatory response in tissue-specific knockouts. Crucially the DNA-binding domain of STAT3 β is identical to that of the α isoform, and can activate the same sets of target genes in most situations.

It has long been suspected that STAT3 regulates different target genes in distinct cell types as previously illustrated for a limited number of genes (summarized by Levy and Lee). The application of ChIP sequencing (ChIP-seq) has underlined the highly divergent cell type-specific binding patterns of STAT3 in various cell types, which in turn reflect the distinct functions of STAT3 in the body. Cell type-specific binding patterns have also been described for other TFs and to different degrees. For instance, out of ~20 000 SOX2 binding sites in embryonic stem cells (ESCs) and neural progenitor cells, only 1200 such sites overlap, while in a more extreme example, only three SMAD3 binding sites overlap between pro-B cells, ESCs and myotubes. Most members of the STAT family bind to a sequence known as the “GAS motif” (“TTCCnGGAA”), a notable exception being STAT6 which has a preference for “TTCCnnGGAA”. The analysis of thousands of in vivo STAT3-binding sites from ChIP-seq studies performed in ESCs, AtT-20 cells (pituitary-like), CD4+ T cells, and macrophages showed that variations of the GAS motif alone are incapable of identifying specific functions in these cell types. Moreover, these divergent binding preferences are indistinguishable from those of other STAT family members,, a characteristic shared with many other TF families. Therefore, alternative STAT3 binding sequences (particularly TGCnnnGAA and TTAnnnGAA) probably represent surrogate methods for STAT3 recruitment. Additional complexity comes from the ability of STAT3 to heterodimerize with other STAT family members, thereby displaying preferences for distinct consensus binding sites, or presenting novel surfaces for co-factors to bind to. Indeed, it has long been known that STAT3 and STAT5 can form heterodimers with a binding preference for distinct DNA sequences, and the heterodimerization of STATs has been suggested to explain the vast functional diversity of STAT family members. Unfortunately no systematic analysis has yet been performed on heterodimeric STAT complexes, while other examples exist where two STATs engage in competitive binding. For example, STAT3 and STAT5 have opposing roles in Th17 cells and mutually compete for the same binding sites. So, if variations of the canonical GAS motif cannot broadly account for the functional specificity of STAT3 in distinct cell types, what are the mechanisms responsible for STAT3’s widely divergent binding patterns and functions in these cells?

STAT3 Has Two Modes of Binding: Universal and Cell Type-Specific

Besides the widely divergent genomic binding patterns of STAT3 in ESCs, AtT-20 cells, CD4+ T cells, and macrophages, our analysis also unveiled a “shared overlap” of 35 non-random in vivo binding sites in these four cell types. What makes this set of 35 (evolutionarily conserved) binding sites so relevant is that the genes that STAT3 appears to be regulating are essential for STAT3 signaling. Therefore we describe two distinct modes of STAT3 binding: one that is universal to all cells (“universal” or “cell type-independent”), and various cell type-specific binding modes whereby STAT3 mediates diverse functions in distinct cells by operating cell type-specific regulatory networks.

Within the universal core of 35 STAT3 binding sites we found that STAT3 binds to its own transcription start site (TSS) in all four cell types to promote its own transcription (Fig. 2A), and that it also regulates genes important for functions downstream of itself. For instance, STAT3 is recruited to SOCS3 (Fig. 2B), which blocks the IL-6 pro-inflammatory response in macrophages by binding to the IL-6 receptor. SOCS3 similarly moderates the pro-inflammatory response in DCs and is also upregulated by STAT3 in mammary cells. In ESCs, SOCS3 regulates LIF/STAT3 signaling and is thought to be a STAT3 target in many other settings. Additionally, STAT3 binds to Bcl3 in all four cell types: in interleukin 10-stimulated macrophages Bcl3 suppresses TNF-α expression, whereas in a myeloma cell line Bcl3 has a pro-apoptotic effect downstream of IL-6. The protein tyrosine phosphatase Ptpn1, a well-known negative regulator of JAK-STAT signaling, is likewise regulated by STAT3. STAT3 is also recruited to several other TF genes, including Nfkbiz, Junb, Fos, Irf2, and Bcl6. These findings suggest that the universal core of 35 STAT3 binding sites dictates the self-regulation of STAT3 signaling by: (1) perpetuating STAT3’s transcription; (2) functioning as a master regulator of other TFs working downstream of STAT3; (3) stimulating the transcription of cytoplasmic enzymes that control STAT3’s activity; and (4) ensuring a robust cellular division program and the maintenance of a stable cell type.

Figure 2. STAT3 binds to a universal core of 35 binding sites to regulate a specific gene set that engages a self-regulatory loop for STAT3 signaling. The examples shown here include: (A) STAT3 binding to its own promoter in all cell types examined; and (B) STAT3 being recruited to the SOCS3 promoter. STAT3-binding data sets were obtained from the following ChIP-seq libraries: GSE27161, GSE37235, GSE21669, GSE19198, GSE11431, and GSE31531.

The Reconstruction of Transcriptional Regulatory Modules Sheds Light on STAT3’s Regulatory Mechanisms

The vast majority of STAT3-binding sites are cell type-specific, and therefore it is reasonable to assume that distinct regulatory networks are responsible for the various functions of STAT3. One attractive model to explain these cell type-specific binding patterns is the local assembly of groups of TFs and co-factors around STAT3 to form distinct transcriptional regulatory modules (TRMs) that provide cell type-specific activity. This model has been shown to work in several biological systems, including ESCs, where the TRM centers on the core TFs Pou5f1 (OCT4), SOX2, and NANOG but also contains KLF4, SMAD1, ESRRB, and STAT3. In macrophages, PU.1 partners with CEBPa, CEBPg, and AP-1, while in B cells PU.1 forms a regulatory network with E2A, EBF1, and FOXO1 to define a core TRM that determines the B cell phenotype.,

The identification of TRMs is a difficult task since extensive prior information is required to prioritize candidates for experimental validation. A typical cell may express several hundreds of the ~1400 sequence-specific TFs, all of which can potentially form stable interactions with important co-factors such as histone epigenetic regulators (this is particularly true in the enhanceosome and TF-collective models). We have developed a computational method for the systematic identification of TRMs that works by integrating TF genome-wide locations (from ChIP-seq experiments) with analysis of TF motif enrichment, cell type-specific expression data and protein-protein interactions. We applied our method, called rTRM, to identify the TRMs that endow STAT3 with both universal and cell type-specific functions. We found that in the universal STAT3 TRM, STAT3 forms a specific network with MYC and E2F1, and experimentally showed that E2F1 not only co-localizes with STAT3 in macrophages in over a dozen genomic sites but that it is also prebound at these sites in the absence of cytokine stimulation. This suggests a temporal regulatory code that might help STAT3 locate specific binding sites in the genome (Fig. 3). The TRMs reconstructed for the cell type-specific binding patterns of STAT3 suggest that STAT3 specifically combines with different factors to perform different functions in distinct cell types (Fig. 3).

Figure 3. STAT3 combines with specific factors to regulate universal and cell type-specific functions. STAT3 always recognizes the GAS motif in DNA (or slight variations thereof), but co-operates with other TFs and co-factors in what are called transcriptional regulatory modules (TRMs). TRMs include both cell type-specific and general TFs, and are largely responsible for the various functions of STAT3 in different cell types.

STAT3 itself is the subject of a large number of post-translational modifications, and specific residues have been implicated in regulating STAT3 homodimer formation as well as in selecting the partner TFs and therefore the TRMs that STAT3 assembles into. For example, in addition to activatory phosphorylations, particularly Y705 and S727, the p300-mediated acetylation of K685 is required to form stable homodimers and for DNMT1 recruitment. Acetylation of K49 and K87 are required for STAT3 to interact with p300., K140 is also methylated and appears to negatively regulate a specific subset of genes, presumably by influencing the TRM that assembles around STAT3. Finally, STAT3 and NFκB have a close relationship, particularly in cancer. Here, unphosphorylated STAT3 can bind to p65/RelA and is recruited to NFκB elements in the DNA to activate gene expression in the absence of cytokine signals.,

The exact role of these post-translational modifications in the context of specific cell types will be an important field of future study. Unfortunately, genome-wide ChIP-seq experiments to detect STAT3 binding have so far used either a pan-STAT3 antibody,,,, a combination of pan-STAT3 and antiphospho-Y705 antibodies, or an antiphospho-S727 STAT3 antibody. As such the genome-wide roles of STAT3 acetylation, methylation and other post-translational modifications remain enigmatic.

Reconstructing TRMs is extremely valuable when access to experimental data sets is limited, or when no regulatory proteins have been identified. One specific advantage of rTRM over other methods is that rTRM integrates protein-protein interaction data sets, which not only provide an additional filtering step but also encapsulate the well-known biological fact that proteins need to physically interact with other proteins to perform specific tasks. Moreover, since protein-protein interfaces evolve faster than protein folds, it is not surprising that the combinatorial potential of proteins into specific modules (TRMs) underlies the functional repertoire displayed by pleiotropic TFs like STAT3.

The Role of the Local Epigenetic Environment in Regulating the Binding and Activity of STAT3

Computational predictions in the mouse genome have thrown a bewildering 1.3 million putative STAT3 binding sites. Still, using a conservative approach by considering only canonical STAT3 binding sites (TTCCnGGAA) yields ~130 000 putative binding sites, a figure much larger than the thousands of genomic sites that STAT3 occupies in vivo as informed by ChIP-seq experiments.,,,,, Our view is that STAT3 is recruited to sites that are not only in an “open chromatin” state, but which possibly are also “primed” specifically for STAT3. For instance we have reported that E2F1 is already bound at genomic sites of future STAT3 recruitment, both in untreated and in IL-10 stimulated macrophages. This observation and the rapid activation of many STAT3 target genes (sometimes taking as little as tens of minutes to maximal mRNA production) suggest that STAT3 is not a pioneer factor in the same way as the FoxA family members, but instead is taking advantage of pre-defined cell type-specific modules already present at specific enhancers. However, in other biological systems STAT3 can also modulate the local epigenetic environment directly by recruiting the histone acetyl transferase p300 and thus promote gene expression, or through SIN3A, a transcriptional repressor that recruits histone deacetylases to remove acetyl groups from histones. STAT3 can also recruit the DNA methyltransferase DNMT1 to bring about DNA methylation and so silence gene expression., For these reasons, we conclude that it would be fair to categorize STAT3 as an “opportunistic pioneer” TF, i.e., one that initially takes advantage of pre-existing enhancers, but which is also capable of establishing its own transcriptional network by modifying the local chromatin environment with the help of accessory enzymes.

The ability of STAT3 to bind to pre-existing enhancers is reminiscent of key genomic studies of the macrophage pro-inflammatory response where it has been shown that the macrophage developmental factor PU.1 remains bound in activated macrophages to provide a spectrum of possible transcriptional responses. Lipopolysaccharide (LPS) stimulation of macrophages leads to the recruitment of p300 to sites where PU.1 is already pre-bound to activate pro-inflammatory genes, indicating that PU.1 is thus maintaining the local chromatin in a permissive state to allow the rapid induction of gene expression. However, this model has recently been challenged by the identification of the new class of “latent enhancers”, which lack any epigenetic marks typical of enhancers (particularly H3K4me1, H3K4me3, and H3K27ac; specific combinations of histone epigenetic marks are better predictors of enhancers than individual marks) before activation and only recruit lineage-specific TFs in the presence of a triggering stimulus. One possible explanation for the working mechanism of latent enhancers is that early-responding genes are pre-bound by cell type-determining TFs such as PU.1, and that upon LPS activation NFκB binds to these sites to rapidly induce gene transcription. In a second series of events, NFκB may establish its own regulatory network by remodeling the cell’s epigenetic landscape and activating a distinct set of genes. It will be fascinating to see if this mechanism also applies to STAT3 and PU.1 is actually marking sites for subsequent STAT3 occupancy during IL-10 stimulation. Indeed, preliminary work in our laboratory suggests that in macrophages PU.1 co-localizes with STAT3 in ~25% (400/1723) of the STAT3 binding sites in macrophages.,,

Concluding Remarks and Perspectives

Understanding the multiple regulatory mechanisms of a pleiotropic TF like STAT3, and how these generate the vast spectrum of functions across the body is a problem of byzantine complexity. Although many models have been proposed to explain the cell type-specific binding of TFs, no single, general, model can yet explain how the many functions of pleiotropic TFs are accomplished. Here we have reviewed the case of STAT3, whose genomic binding patterns are predominantly cell type-specific. However, STAT3 also binds to a small, non-random, set of 35 binding sites shared across multiple cell types and which appear to encode a mode of auto-regulation for STAT3 signaling. Therefore we propose two distinct modes for STAT3 binding: one that is universal to all cells, and various cell type-specific binding modes whereby STAT3 mediates diverse functions in distinct cells by operating cell type-specific regulatory networks. The models described here (Fig. 3) are specific to STAT3, but the concepts and tools involved in the analyses can be applied to dissect the regulatory mechanisms of other TFs, thus opening the door to a more thorough mechanistic understanding of TFs with complex functions.

As more ChIP-seq data accumulate for specific TFs in multiple cell types, coupled with functional genomic analysis, it is likely that a set of models will emerge to explain the regulatory mechanisms of pleiotropic TFs. STAT3 is a fascinating TF to study as well as an excellent model system, not only because it works as a natural biological switch but also because of its involvement in such a wide array of essential biological processes. Furthermore, since STAT3 is implicated in many diseases, particularly cancer and inflammation, achieving a detailed mechanistic understanding of how STAT3 regulates its transcriptional programs will have important consequences for the design of disease-specific therapies.