Recent studies have largely focused on understanding how topological properties and network organization characterize complex biological systems. However, the dynamic nature of the nodes and its role in systems behavior on a genomic-scale has remained unclear. In this study, we used a new method called vertex sort to classify DNA-binding TFs in the yeast regulatory network into seven hierarchical levels, which naturally clustered into three mutually exclusive hierarchical layers that were named as top, core, and bottom. By integrating several genome-scale datasets onto the inferred hierarchy, we found that TFs have static and dynamic properties that are similar within a layer and different across layers. These findings indicate that the network topology and the nodal (TF) dynamics are tightly linked.
We did not find the inherent hierarchy to be strictly linear or pyramidal as previously suggested (Yu and Gerstein, 2006
). Instead, we found that the central skeleton appears to have a feed-forward structure (). The core layer is enriched in TFs that regulate many genes and the bottom layer contains TFs that regulate few genes (), further suggesting that the hierarchy is non-pyramidal. In addition, the fact that the core-layer TFs regulate or co-regulate 87% of all target genes, and are involved in over 90% of all feed-forward loops suggest that core-layer TFs are the power-centers of the regulatory network, which is consistent with previous observations (Maslov and Sneppen, 2005
; Farkas et al, 2006
). Our observation of a multi-level hierarchy with a strongly connected component (core-layer; SCC) is consistent with what has been discussed recently by Bar-Yam et al (2009)
, where they suggested that the complexity of the transcriptional network in cellular systems is somewhere in between a strictly hierarchical ‘autocratic' structure (with multiple hierarchical levels and no SCC) and a highly interconnected ‘democratic' structure (in which a few master-regulator TFs regulate the set of other TFs that mutually regulate each other, effectively forming a two-level hierarchy).
Overlaying large-scale genomic datasets on transcript abundance, transcript half-life, translation efficiency, protein abundance, protein half-lives, and protein and transcription noise on the inferred hierarchical structure showed that the dynamics of TFs in the regulatory network is not random. Rather, we find that TFs in distinct hierarchical layers of the network have similar dynamic properties, indicating that the network topology and the nodal (TF) dynamics at the mRNA and the protein level are tightly linked. Although the presence of a hierarchical structure in the yeast regulatory network is of interest in itself, our finding that the TFs possess inherent characteristics that encapsulate their dynamic roles in systems behavior is noteworthy and unexpected.
Our observations that transcript half-lives of TFs from the three layers are comparable (), but the top-layer TFs are present in relatively higher abundance at the protein level () and have a much longer protein half-life () when compared with that of core- and bottom-layer TFs suggest that post-translational regulation has an important role in ensuring the availability of right amount of each TF within the cell. The need for the presence of top-layer TFs to relay faithful signals down the transcriptional cascade and their involvement in many biological processes () could explain why top-layer TFs need to be relatively abundant than the core- and bottom-layer TFs. These findings are consistent with what has been proposed by Farkas et al (2006)
who suggested a model in which regulatory cascades originating from distinct fractions of the regulatory network control robust integrated responses to complex stimuli.
The observation that top-layer TFs show a relatively higher variability in protein abundance between individuals in a clonal population of cells () suggests that such a behavior may confer a selective advantage to individuals, as this permits at least some members in a population to respond effectively to changing conditions by triggering relevant transcriptional cascades (Spudich and Koshland, 1976
; McAdams and Arkin, 1999
; Rao et al, 2002
; Kaern et al, 2005
; Raser and O'Shea, 2005
; Blake et al, 2006
; Ramsey et al, 2006
; Samoilov et al, 2006
; Acar et al, 2008
; Heath et al, 2008
; Lopez-Maury et al, 2008
; Raj and van Oudenaarden, 2008
; Shahrezaei and Swain, 2008b
). For instance, ABF1, which is a multifunctional TF present in the top layer, is an abundant protein whose levels are noisy in a clonal population of cells. However, the activity of ABF1 depends on the availability of its co-activators (e.g., CDC6) and on its phosphorylation state, which is known to be regulated by several kinases (e.g., casein kinase 2) or phosphatases (Silve et al, 1992
; Upton et al, 1995
). The relatively higher noise in the abundance of ABF1 might ensure that, at least, some members in a population would respond rapidly to changing environments (i.e., when co-activators or kinases are activated in response to the altered external stimulus). We propose that high variability in the expression of key TFs, whose TGs might contribute to phenotypic variation, might be a general strategy to facilitate adaptation to diverse environments (see for a model). This does not exclude the possibility that variation in the protein expression levels of specific target genes (independent of the variation in the levels of their regulating TFs) might dictate cell-fate outcomes in a post-transcriptional or post-translational manner. Hence, a detailed investigation that integrates multiple types of networks with data on cell-to-cell variation in the levels of transcripts and proteins might elucidate the contribution from TF-dependent and TF-independent modes for adaptability of cells to changing environments.
Figure 6 A schematic model describing the conceptual framework of differential utilization of the same underlying regulatory network by distinct members of a genetically identical cell population. (A) A toy regulatory network showing two regulatory pathways, which (more ...)
Further, our observation that the protein levels of the core-layer TFs (the power-centers) and bottom-layer TFs are inherently tightly regulated suggests that such a tight regulation, along with other regulatory mechanisms such as post-translational modifications or physical interactions with other proteins, might act as a filter to minimize noise propagation down the hierarchy due to any ‘inadvertently' triggered response. In other words, tight regulation of the core- and bottom-layer TFs by rapid degradation (i.e., shorter protein half-life; ) would ensure that such TFs are present only in low levels under normal conditions. Their presence in relatively lower levels might facilitate minimization of noise propagation because sufficient levels of TFs may not be present to trigger an appropriate response when transient signals ‘inadvertently' activate them. Therefore, we suggest that the tight regulation of protein levels of the core- and bottom-level TFs might ensure fidelity and robustness in a regulatory cascade.
Taken together, our findings suggest that (i) the higher variability in abundance of top-layer TFs compared with core- and bottom-layer TFs in distinct members of a clonal cell population might permit differential utilization of the same underlying network (see ) and (ii) the tight regulation of core- and bottom-layer TFs might contribute to fidelity in gene expression. Thus, the interplay between the dynamics of individual nodes and the topology of the regulatory network would make the underlying network robust and permit at least some members in a population to effectively adapt to (or survive in) changing environments.
Our findings have implications in synthetic biology experiments aimed at engineering gene regulatory circuits (Becskei and Serrano, 2000
; Elowitz and Leibler, 2000
; Gardner et al, 2000
). In particular, the dynamics of TFs in terms of their abundance, half-life, and noise cannot be ignored as modulating these attributes could affect the outcome of a regulatory cascade. The proposed conceptual framework (see ) from our findings serves as a general model and also has important implications for a number of apparently different but related phenomena as outlined below.
Bacterial persistence or adaptive resistance
This is a phenomenon where a fraction of a genetically homogeneous microbial population survives upon exposure to stress, such as antibiotics (Balaban et al, 2004
; Levin, 2004
; Dhar and McKinney, 2007
). Although key regulatory proteins, which facilitate random phenotypic switching, have been implicated in this phenomenon (Dhar and McKinney, 2007
), we suggest that the altered dynamics and stochastic expression of certain regulatory proteins may alter the susceptibility of an individual by facilitating differential utilization of the underlying network efficiently (). This might permit alteration of the phenotype (i.e., the switching in susceptibility states) in an otherwise genetically identical population of cells. Indeed, this suggestion is supported by recent studies which have implicated TFs such as PhoU (Li and Zhang, 2007
) and the two component signal transduction system DpiAB (Miller et al, 2004
) in this phenomenon.
Differential cell-fate outcome in response to the same uniform stimulus
Factional survival or cell-death in clonal cell populations upon drug treatment is a well-known phenomenon in certain diseases such as cancer. Two important studies recently showed that this could be achieved by variability in the expression dynamics of key regulatory proteins that determine cell fate such as death or survival. In one study (Cohen et al, 2008
), it was shown that the dynamics of the regulatory proteins, which either dictate cell death or survival, varied widely between individual cancer cells. In the other study (Spencer et al, 2009
), it was shown that naturally occurring differences in the levels or states of proteins regulating apoptosis are the primary causes of cell-to-cell variability in the timing and probability of death in individual members of the population upon induction of apoptosis. We suggest that this dynamic variability in expression level of key regulatory proteins might permit differential sampling (i.e., the survival network or the apoptotic network) of the same underlying regulatory network (governing all cells) by different members in a clonal population (see ). This differential network utilization might result in divergent cell-fate outcomes among different individuals in an otherwise identical cell population.
Phenotypic variability in fluctuating environments
When organisms experience fluctuating environments, individuals of the same population may exhibit very different phenotypes. This may be achieved either by sensing followed by response or through the generation of diversity by random switching between different states (Kussell and Leibler, 2005
; Acar et al, 2008
). Although variation in expression levels of certain regulatory proteins has been implicated in this phenomenon, we suggest that such variation in expression levels may allow stochastic switching between phenotypes by poising certain individuals or permitting the sampling of relevant parts of the same underlying network (see ). This suggestion is supported by recent studies. Upon fluctuating nutrient starvation, some yeast cells sporulated, whereas others underwent delayed sporulation (Nachman et al, 2007
). This variability was shown to be governed by the variation in the production rate of the meiotic master regulator Ime1p and its gradual increase over time. In another set of studies, by analyzing the probabilistic and transient differentiation of Bacillus subtilis
cells into the state of competence (Iber, 2006
; Suel et al, 2006
; Maamar et al, 2007
), it was shown that variation in the expression level of ComK determines the frequency of the differentiation events. The increased variability in expression level of these key regulatory proteins, in turn, might permit different individuals in the same population to activate relevant parts (i.e., the sporulation sub-network or the competence sub-network) of the same underlying global transcriptional network to exhibit phenotypic variability (i.e., to sporulate or not; to exhibit a competence or vegetative state).
Cellular differentiation and development
During the course of stem-cell differentiation and development, it has been well established that maintaining the right balance of combinations of key TFs and appropriate signaling environment in space and time dictates (a) lineage specification of the progenitor cells (e.g., myeloid lineage commitment from hematopoietic stem cells; see (Iwasaki and Akashi, 2007
)) and (b) formation of distinct spatial patterns of cell-types during organ development (e.g., cell fate specification in neural development; see (Guillemot, 2007
)). In this context, the inherent dynamics of TFs, such as cell-to-cell variation in expression levels, might have an important role in development, stem cell maintenence and differentiation (Arias and Hayward, 2006
; Huang et al, 2007
; Silva and Smith, 2008
). Although dedicated circuits that filter noise in expression may be required for certain processes to be robust and reproducible (Arias and Hayward, 2006
; Barkai and Shilo, 2007
), we suggest that distinct TF dynamics, as dictated by the position in the hierarchy of the transriptional network, might provide the flexibility for individuals in a population to initiate distinct response and sample distinct networks that permit lineage commitment when the appropriate signaling environment is experienced (see ). The implications of our findings assume significance particularly in light of the recent discovery that transcriptome-wide noise of specific TFs controls lineage choice during stem cell differentiation (Chang et al, 2008
; Kalmar et al, 2009
In conclusion, investigating the dynamics (e.g., cell-to-cell variability in abundance, half-life of transcripts and proteins, etc) of individual nodes in relation to the network organization in other biological networks such as the protein interaction network, kinase-substrate regulatory network, metabolic network, and understanding their impact on each other will be important to obtain a better understanding of how (i) cells adapt to changing environments, (ii) different phenotypic outcomes are mediated in clonal cell populations and (iii) mutations that disrupt the dynamics of key regulatory proteins may influence disease conditions.