Stochastic signalling rewires the interaction map of a multiple feedback network during yeast evolution

GAL genes enhance their own transcription via the transcription factor Gal4p. With synthetic circuits and stochastic simulation, we show that the evolution of the feedback system does not rely on tuning the strength of the Gal4p – promoter interaction to change its activation rate but on the adjustment of bursting kinetics in stochastic gene expression.

HFSP Long-Term Fellow Chieh Hsu and colleagues
authored on Thu, 15 March 2012

The baker’s yeast, Saccharomyces cerevisiae, utilizes the sugar galactose as a carbon source with the GAL gene family. Gal2p transports galactose into the cell. Gal1p is an  enzyme which catabolizes galactose.The GAL genes, with the exception of GAL4, are regulated by a common transcription activator, Gal4p, which is inhibited by Gal80p. The binding of Gal80p and galactose bound Gal3p or Gal1p supresses the inhibitory effect of Gal80p and therefore activates the GAL genes. GAL1 and GAL3 are paralogs which, after the whole genome duplication, evolved from the GAL1/3 gene encoding a bifunctional protein with both enzymatic and network regulatory activities. Both the protein coding region and the regulatory region of the paralogs diverged asymmetrically: Gal3p lost its enzymatic function completely while Gal1p retained the ability, although weakened, to transduce the galactose signal to the network; GAL3 retained only one Gal4p binding site; while GAL1 retained all four. 

Figure: Stochastic bursting behavior triggers a positive feedback system differently. a) The synthetic circuits. The gene produces rtTA which activates its own production and the rtTA binding to the GAL based promoter is controllable by adjusting the concentration of the inducer, doxycycline. The rtTA level is amplified and read out by a gene with rtTA binding sites controlling a fluorescent protein, GFP. b) Basal expression of GAL2 promoter triggers the feedback circuit faster than that of GAL1. Amplification of GAL1 basal expression results in a long-tailed distribution of GFP signal in a population while in the case of GAL2, the GFP distribution is more centered and closer to Gaussian distribution (right panel). c) A cartoon depicts how the stochastic burst nature of the basal transcription affects the positive feedback loop with the same average expression level.

In the GAL network, Gal2p and Gal3p enclose positivie feedback loops for increasing the cellular galactose concentration and enhancing Gal4p binding to the promoter, respectively. We show that Gal1p also conveys a positive effect on the feedback system predominately, despite it having two oppsite effects on the system, reducing intracellular galactose level and trasmitting galactose signal to increase Gal4p dependent gene expression. By comparing other GAL genes which have Gal4p binding sites in the promoter region, we tried to find the distinctive feature of the promoters in the three genes, GAL1, GAL2, and GAL3, of which the parallel postive feedback loops consist. Both GAL1 and GAL2 have multiple Gal4p binding sites and broad dynamic range (the range between maximum and minimum gene expression levels upon induction), while GAL3 has a single site and narrow dynamic range.

We analyzed how the two major determinants – the basal expression and the number of activator binding sites – affect a positive feedback loop. We considered the simplest positive feedback: a transcription activator promoting its own expression. The binding strength is regulated by the concentration of a given small molecule inducer. For computational simulation, we applied the Gillespie algorithm, where the gene activation/ inactivation and molecular production/decay are treated as individual stochastic events. We found that both elevating the basal expression and increasing the number of activator binding sites of the promoter, which enhances the affinity of the activator, result in rapid system activation.

Interestingly, with the same, or even slightly less, basal expression levels, GAL2 core promoter activates the system faster than that of GAL1, when we employed the synthetic strategy to construct the simple positive feedback in the yeast cells. GAL3 basal expression, when reduced to the similar level as GAL1 by glucose, also triggers the system more rapidly. Gene expression is discrete and the production of mRNA bursts pulses with inactive periods in between. The promoter which fires more frequently with small amplitude can have the same expression level as the one with lower burst frequency yet higher amplitude. However, in a population, the two promoters lead to different distribution of cellular mRNA and protein molecules. Taking together the simulation and synthetic biology results, we conclude that high-frequency-small-burst-sized basal expression is favorable for the rapid activation of a positive feedback system. In addition, as shown experimentally, increasing the number of binding sites in the promoter does not promote the feedback system’s activity (in measurement of “ON” cell proportion, see the original article), although it does enhance the overall expression level. It has been suggested previously that increasing the binding site number might introduce a refractory period of a promoter at the beginning of its activation. Based on the simulation results, this refractory period can make the activity of feedback system with the multiple-binding-site promoter indistinguishable from the one with the single-site promoter.

Our results indicate that knowing only the gene expression level is not sufficient to predict the behavior of a feedback system. Stochastic nature of network components affects the system significantly. This effect can override the direct influence of network wiring and enables even a weakly expressed gene to activate the feedback loop efficiently. During evolution, as in the GAL network, both expression level and the bursting nature of the core promoters may be tuned to achieve a rapid metabolite adaptation with minimal sensory molecule level.


Stochastic signalling rewires the interaction map of a multiple feedback network during yeast evolution, Hsu C., Scherrer S., Buetti-Dinh A., Ratna P., Pizzolato J., Jaquet V. and Becskei A., Nature Communications 3:682 doi: 10.1038/ncomms1687 (2012).

Pubmed link

Nature Communications article