Transcription is a critical step in the expression of all gene products, and is coordinately regulated to induce broad changes in the cellular state. Eukaryotic gene transcription follows an elaborate sequence of events involving modification enzymes, transcription factors (TFs), co-factors and RNA polymerase (1–3
). Constructing a comprehensive model of gene transcription that incorporates these various biological processes holds the potential to decipher systems-level behavior in the cell (4
). A crucial component of transcriptional control relies on sequence-specific binding of TF proteins to short DNA sites in the relative vicinity of the target gene. However, an effective interaction between the TF and the gene's regulatory elements is critically mediated by other cellular processes and signaling pathways.
In response to various stimuli, cell signaling pathways relay information to the nucleus and alter the transcriptome, often via post-translational modification (PTM) of the TF proteins (6–10
). Numerous types of chemical modifications of TF proteins have been documented, including phosphorylation (11
), acetylation (12
) and methylation (14
). A classic example of PTM-mediated transcriptional regulation involves the TF CREB, which requires phosphorylation of serine at position 133 in order to promote transcription. This serine residue is targeted by multiple signaling pathways, and coordinates different transcriptional programs depending on other modified residues (8
). In this way, PTM-dependent TFs act as ‘molecular switchboards’ mapping upstream signals to gene transcripts and ultimately coordinating complex cellular responses to internal and external stimuli (7
For many TFs, the full cohort of regulatory PTMs and the modifying enzymes responsible for catalyzing their addition and removal are not known. However, new experimental techniques (15–17
) now provide additional clues for this level of regulation. Given the importance of PTMs in determining TF activity and the eventual control of gene transcription, it is imperative that models of transcriptional regulatory networks incorporate PTMs and the mediating modification enzymes.
Most approaches to infer transcriptional regulatory networks consider only regulatory interactions, or ‘network edges’, between TFs and target genes, and do not include the modulators of these TF–gene interactions, such as modification enzymes [see (4
) for recent reviews and (19–27
) for specific examples]. Although a few computational methods have been developed to infer modulators of TF–gene interactions (28–34
), none of these methods infer the target genes and upstream modifiers of a TF concurrently, nor do they integrate heterogeneous data sources.
Here we propose the first principled computational model of gene transcription that explicitly incorporates interactions between modifying enzymes and TFs, thus extending the prevalent view of TF–gene connectivity to modifier–TF–gene connectivity. Our method, called ‘Modification-dependent Network-based Transcriptional Estimator’ (MONSTER), combines expression compendia with other data sources indicative of physical interactions to simultaneously infer the target genes and upstream modifiers of each TF. We first use simulated data sets to demonstrate that our computational model and the parameter estimation procedure are robust against noise from a variety of sources. Next, we use a well-studied stress–response regulatory network in the model system Saccharomyces cerevisiae to demonstrate the accuracy of MONSTER on experimental data.
Finally, we apply MONSTER to investigate the STAT1-mediated regulatory network in human B cells. B cells play a critical role in adaptive immune response, and dysregulation of B cell networks can lead to a number of human diseases including autoimmune disorders (35
), leukemias (36
) and lymphomas (37
). A highly pleiotropic TF, STAT1 is a critical mediator of B cell development and function and is subject to complex post-translational regulation (38–41
). MONSTER predicts a module of STAT1 target genes and modifying enzymes active in B cells, which is well-supported by the STAT1 literature, and includes novel hypotheses about the role of STAT1 in specific signaling pathways.