An Atlas of Combinatorial Transcriptional Regulation in Mouse and Man

Abstract

Tissue specificity is enabled by spatial and temporal patterns of gene expression which in turn are driven by transcriptional regulatory networks ([Naef and Huelsken, 2005] and [Zhang et al., 2004]). Such networks involve assemblies of control proteins, such as DNA-binding transcription factors (TFs) connected to the sets of promoters of genes they induce or repress (Tan et al., 2008b). Typically, TFs do not act independently but form complexes with other TFs, chromatin modifiers, and cofactor proteins, which bind together and assemble upon the regulatory regions of DNA to affect transcription (Fedorova and Zink, 2008). Mapping the combinatorial interactions among TFs would represent a significant leap forward in our understanding of how tissue specificity is determined.

In recent years, a variety of genome-scale technologies have been introduced which allow mammalian transcriptional regulatory networks to be investigated at high resolution and depth. Many such studies have inferred transcriptional networks through mRNA expression profiling combined with genome-wide active promoter mapping and promoter motif analysis (e.g., Suzuki et al., 2009). These data have been supplemented with fluorescence-activated cell sorting (FACS) (Shachaf et al., 2008) or reverse transcriptase quantitative polymerase chain reaction (qRT-PCR) ([Roach et al., 2007] and [Wen et al., 1998]).

Another technology that has revolutionized the study of transcriptional networks is chromatin immunoprecipitation (ChIP), which when coupled with microarrays or high-throughput sequencing (Johnson et al., 2007), enables genome-wide measurements of TF binding locations in vivo. A complementary approach is the protein binding microarray (PBM) (Berger et al., 2008), which rapidly characterizes the complete DNA sequence repertoire bound by a TF in vitro. ChIP and PBMs have been applied to map transcriptional networks in a variety of human cell types, including stem cells ([Cole et al., 2008] and [Lee et al., 2006]) and lymphocytes ([Marson et al., 2007] and [Schreiber et al., 2006]), and to characterize the binding motifs of many mammalian TF families (Berger et al., 2008).

Although these studies have led to the construction of very large models of transcriptional networks, they are based on experiments that largely treat each TF in isolation. For instance, ChIP-chip measures binding locations for one TF at a time, although separate profiles for several TFs can be later combined into networks (Mathur et al., 2008). However, it is well known that the transcriptional output of a gene is due to the joint activity of many TFs whose binding and activation are highly interdependent. This cooperation is often mediated by direct physical contact between two or more TFs, forming homodimers, heterodimers, or larger transcriptional complexes. In fact, it has been estimated that approximately 75% of all metazoan TFs heterodimerize with other factors (Walhout, 2006). Newman and Keating used protein arrays to reveal a network of several hundred domain interactions among the bZIP TF family alone (Grigoryan et al., 2009). Other studies have successfully assembled large networks of protein interactions using technologies such as coimmunoprecipitation and two-hybrid screening (Park et al., 2005 D. Park, S. Lee, D. Bolser, M. Schroeder, M. Lappe, D. Oh and J. Bhak, Comparative interactomics analysis of protein family interaction networks using PSIMAP (protein structural interactome map), Bioinformatics 21 (2005), pp. 3234–3240. View Record in Scopus | Cited By in Scopus (17)[Park et al., 2005] and [Yu et al., 2008]), but to date these have not been systematically applied to map networks of transcription factors. Thus, a clear and immediate task is to map which combinations of TFs act together and how these combinations lead to modes of regulation that are not evident when each factor is considered separately.

Toward this goal, we have pursued an integrative approach to systematica