Cell signaling research is challenged with the question of how ligand specificity can emerge when different pathways share the same core components (Kholodenko, 2006
). In the current study, we combined computational modeling and experiments to provide insight into this question. The models bring together extensive previous experimental data with our own new data to unveil that ligand-specific pc-Fos responses are brought about by a precise, spatially-distributed control system that involves a cascade of CFLs interlinked with transcriptional negative feedback loops. Owing to the time lag between transcription initiation and translation, this cytoplasmic-signal-to-protein-expression CFL structure acts as an “AND-gate” to convert the sustained vs. transient cytoplasmic
ppERK temporal profiles into the all-or-none pc-Fos responses. Negative transcriptional feedback not only causes the similar c-fos
expression durations for EGF and HRG, but also endows the pc-Fos response with robustness to parameter perturbations. The “inner” CFL involving RSK makes pc-Fos robust to noise in the ppERK input.
In this work we developed a mechanistic model and a core model, which have complementary properties. The mechanistic model allows us to ascribe observed behavior to precise biochemical mechanisms, aiming to create an in silico replica of cellular networks. Mechanistic biochemical models are directly tested against experiments, but these models must be refined continuously to keep pace with the constantly increasing detailed knowledge of molecular mechanisms. The current study, in which we refined our initial model following the results of siRNA, double-ligand pulse and CHX experiments, exemplifies this continuous refinement. Nevertheless, mechanistic models have large potential to facilitate understanding complex signaling networks. However, when the detailed mechanistic knowledge is lacking, it is desirable to employ simple, core models. Core models do not have excessive numbers of species and parameters, but capture and explain the key features that control the system behavior. Our core model serves just this purpose; when our data showed the limitations of the current knowledge, the core model helped us comprehend the emergent properties of the c-Fos expression network.
The biological significance of the CFL-regulated pc-Fos response is that a robust switch-like activation of transcription factors will lead to drastically different subsequent waves of gene expression, and consequently different phenotypes. The CFL structure also allows the cell to turn off gene expression rapidly as soon as the input signal is lost, while buffering the cell against unwarranted gene expression in response to spurious inputs or noise (). In addition to these cytoplasmic-signal-to-protein-expression CFLs, active nuclear ERK, RSK and c-fos mRNA generate the nuclear-signal-to-mRNA CFL that operates on a shorter time scale (). These fast and slow CFLs are organized in a “cascade” structure, where the faster, “inner” loop (ppERK-pRSK-c-fos mRNA) operates within the context of the slower, “outer” loop (ppERK-c-fos mRNA-pc-Fos protein). Because it takes time to propagate the disturbances in cytoplasmic ppERK through the inner loop before they reach c-fos mRNA, the inner loop filters fast ppERK input noise. When this cascade CFL structure is combined with the transcriptional negative feedback loops, which make the system robust to parameter perturbation, the overall network acquires even greater noise reduction capabilities ().
Regulatory motifs in the c-Fos expression network and emerging differential, long-term transcription factor expression
Why does the cell employ dusp and additional c-fos repressor(s) to downregulate the c-fos mRNA response when in principle the dusp response alone should be adequate for this task? One reason is that functional redundancy leads to robustness against system failures resulting from breakdown of any single component. This is a universally desirable feature that conceivably may have been selected for during evolution. Another, less obvious reason arises from the double-ligand pulse experiments, which show that an unidentified fos repressor makes MCF-7 cells refractory to further ligand stimulation in terms of c-fos expression. Thus, expression of this additional repressor converts cells into a different state, in which they no longer respond to ligands. Since HRG stimulation causes MCF-7 cell differentiation, the c-fos repressor may play a key role in ensuring that the cells follow the differentiation pathway despite the potential presence of other signals.
The opposing cell-fate decisions caused by EGF and HRG (proliferation versus differentiation) should be underlined by distinct gene expression patterns. We suggest that the quantitative
differences in c-fos
mRNA expression at the immediate early gene level are translated into robust qualitative
differences for later waves of gene expression changes. Differences in expression of immediate early transcription factors such as c-fos
would have a large impact on successive gene expression waves, if these factors are hubs in the regulatory network. As network hubs have many interaction partners and the DEF domain is critical for the all-or-none pc-Fos response (Murphy et al., 2002
), we looked at the number of interaction partners for transcription factors with and without DEF domains. We indeed found that transcription factors with a DEF domain had a larger mean number of interaction partners (23.1) than non-DEF domain containing factors (15). For DEF domain containing transcription factors known to be HRG-induced immediate early responders in MCF-7 cells, the mean number of interaction partners (44) was even larger (Fig. S5
and Nagashima et al., 2007
). Our hypothesis is further supported by previously published gene expression responses to HRG and EGF over longer time periods (; Nagashima et al., 2007
). At early times (45 minutes) nearly all the transcription factors that are differentially expressed in response to EGF and HRG are shared. However, as time progresses the overlap between these two sets decreases dramatically. We hypothesize that in large part this is due to HRG-induced, pc-Fos protein controlled gene expression. We propose that this quantitative-to-qualitative gene expression control principle may be general to mammalian signal transduction systems that induce distinct cell fates. Thus, we suggest that the integral negative feedback-embedded, cascade CFL structure that controls the initial, robust switch-like pc-Fos response is critical for control of cell fate decision processes.