Introduction

One of the most significant network motifs (Milo et al, 2002; Shen-Orr et al, 2002) in transcription regulation networks is the feed-forward loop (FFL). This motif was first defined in Escherichia coli (Shen-Orr et al, 2002), and then found in diverse organisms including Saccharomyces cerevisiae (Lee et al, 2002; Milo et al, 2002), Bacillus subtilis (Eichenberger et al, 2004), Caenorhabditis elegans (Mangan et al, 2003) and humans (Mangan et al, 2003; Odom et al, 2004). In the FFL, transcription factor X activates a second transcription factor Y, and both activate the output gene Z (Figure 1A). There are eight types of FFLs, characterized by the signs of the transcription interactions (repression or activation) (Mangan and Alon, 2003). One of the most abundant FFL types, called the type-1 coherent FFL (Mangan and Alon, 2003; Ma et al, 2004), has three positive regulations.

Figure 1

(A) The type-1 coherent FFL (Mangan and Alon, 2003). In many cases, Y regulates its own production as shown. (B) The SUM-FFL in the flagella class 2 regulation network. X is flhDC, Y is fliA and Z is the fliLMNOPQR operon (termed fliL) and other class 2 operons. In this circuit, the activator X regulates Y, X and Y act additively to activate the output gene Z. The input Sx is the production rate of X (or, more generally, a stimulus that activates X). The input Sy regulates the activity of Y. In the flagella system, Y positively regulates its own production. (C) A more detailed view of the flagella network and the basal-body checkpoint. The flhDC promoter is controlled by several transcription factors responsive to environmental stress and starvation. The class 2 genes encode the structural proteins that make up the basal bodies. FliA is involved in a positive feedback loop called the basal body checkpoint. In this loop, the activity of FliA as a transcription factor is inhibited by binding the protein FlgM (dashed -| sign indicating inhibition). FlgM is exported out of the cell once the first active basal bodies are formed, by a specific transport mechanism that exports FlgM through the basal bodies (dashed inhibition symbol between the basal body and FlgM). Thus, FliA helps activate genes that produce basal bodies, which export the inhibitor FlgM out of the cell, relieving the inhibition of FliA.

Full figure and legend (27K)Figures & Tables index

In order to understand the function of the FFL, one needs to specify the input function that integrates the effects of X and Y on gene Z. Previous experimental work characterized the function of the FFL in the ara system of E. coli. This FFL has an AND input function, in which both X and Y are needed to activate Z (Mangan et al, 2003). The AND-FFL showed a delay in Z expression following step activation of X, and no delay following deactivation of X. Here, we focus on the case where either X or Y is sufficient to activate Z (similar to an OR gate). We choose an experimental system in which X and Y act additively to regulate Z. That is, the input function at the Z promoter sums over the two inputs. We term this motif the SUM-FFL. One of the simplest ways to implement an additive input function is to provide a gene with two different promoters, each responding to one of the inputs. Such multiple promoters are indeed found in many gene systems.

We previously modeled the FFL with OR-gate logic (Mangan and Alon, 2003), which can be considered as a Boolean approximation to a SUM input function. The mathematical models suggested that a coherent FFL with an OR gate can carry out an information-processing function termed 'sign-sensitive delay': The output Z responds rapidly when the level of X increases, whereas Z responds only at a delay once X levels decrease (Mangan and Alon, 2003). The delay is due to the presence of Y. After X is deactivated, it takes time for Y levels to decrease sufficiently to de-activate Z. Thus, this gene circuit can protect against transient deactivation, because Z production can proceed even if X activity is briefly lost. If the activator Y was removed from the circuit, Z would respond rapidly both to increases and decreases in X activity and the protection function would be lost.

As in the case of OR-FFL, a SUM-FFL can show a delay following OFF steps of X activity. This contrasts with the AND-gate FFL, which shows delay following ON but not OFF steps (Mangan and Alon, 2003; Mangan et al, 2003). Models of the SUM-FFL show a delay for a wide range of biochemical parameters, such as the production and degradation rates of the proteins, or the activation coefficients of the genes (Kalir and Alon, 2004). The length of the delay can be tuned by changing these parameters (Mangan and Alon, 2003). Positive feedback of Y on itself (Figure 1A) can further increase the delay time by slowing the reduction in Y levels following deactivation of X (see a simple mathematical analysis of positive auto-regulation in the appendix). In general, negative auto-regulation can speed responses, whereas positive auto-regulation slows response time (Savageau, 1974; Rosenfeld et al, 2002).

The above-mentioned theoretical treatment of the FFL deals with the interactions of three genes in isolation. In reality, this circuit is embedded in a network of interactions. It is therefore crucial to experimentally test the dynamical behavior of this motif in living cells. For this purpose, we consider a well-characterized gene-regulation system, the flagella biosynthesis network of the bacterium E. coli (Aldridge and Hughes, 2002).

When growth conditions become mildly unfavorable, E. coli produces several rotating flagella and swims away. The genes that make up the flagella motor are regulated by a SUM-FFL (Kalir and Alon, 2004) (Figure 1B). The master flagella activator X (FlhDC) activates a second activator Y (FliA). The activators X and Y function additively to activate the genes Z that build the flagella motor (Z represents the flagella class 2 genes arranged in operons such as fliLMNOPQR, here termed fliL).

The concentration and activity of the two regulators X and Y is affected by signals, termed Sx and Sy (Figure 1B). These inputs to this system are as follows: the rate of production of X is controlled by factors that respond to environmental signals such as glucose starvation (CRP) (Silverman and Simon, 1974), heat shock (dnaKJ and GrpE) (Shi et al, 1992), osmotic stress (ompR) (Shin and Park, 1995), low-PH (H-NS) (Soutourina et al, 1999) and cell density (QseBC) (Sperandio et al, 2002). Flagella in the best studied E. coli strains are activated at late exponential growth phase, and deactivated in stationary phase (Amsler et al, 1993) as well as under high salt concentration and alcohol molecules (Shi et al, 1993).

The activity of Y is controlled by the signal Sy, a checkpoint that monitors the production of flagellar motors (basal bodies). The transcriptional activity of Y is inhibited by binding a protein inhibitor (FlgM, an anti- factor; Kutsukake and Iino, 1994). The inhibitor FlgM is exported out of the cells by completed basal bodies (Hughes et al, 1993; Karlinsey et al, 2000). Thus, when the first basal bodies are completed, FlgM is exported, relieving the inhibition of Y so that it begins to activate downstream genes Z (Figure 1C). Note that the presence of Sy (that is, the absence of FlgM) is required for the delay function of the SUM-FFL.

Here, we studied the effects of the SUM-FFL on the dynamics of the flagella gene expression using high-resolution measurements from living cells. We find that the SUM-FFL can generate a delay in the turn-OFF dynamics of the system, a delay that is dependent on the presence of Y. The delay is on a time-scale similar to that required for assembly of a flagellum.

Results

To study the dynamics of the SUM-FFL in the flagella system, we constructed E. coli cells in which X is under control of an inducible promoter. In these cells, the production of X (FlhDC) can be turned ON or OFF by means of a chemical inducer (L-arabinose) added externally to the cells (Kalir and Alon, 2004). The rate of Z (FliL) production from these cells was monitored in real time by means of a green-fluorescent protein (GFP) fused to a copy of the DNA regulatory region of gene Z (the fliL promoter) on a low-copy plasmid (Kalir et al, 2001). In order to measure GFP fluorescence, which corresponds to Z promoter activity, the cells were grown in an automated flourimeter during ON and OFF steps of X production (Kalir and Alon, 2004). As a control, we compared the dynamics to cells in which the gene for Y (fliA) was deleted.

To study turn-ON of gene expression, we added an inducer to the cells to initiate the production of X. We find that Z shows rapid production following an ON step of X production (Figure 2A). Cells deleted for Y showed about a 50% lower maximal Z expression. To study the response time, we normalized the fluorescence per cell signal to its maximal value. We find that cells deleted for Y show a rapid production of Z, similar to cells wild type for Y (Figure 2A). To study turn-OFF of gene expression, we shifted cells growing with inducer for 3 h to a medium without inducer (and with saturating anti-inducer D-fucose; Wilcox, 1974). We find that the deactivation of Z occurred at a delay of about 60-80 min compared to a cell in which Y is deleted (Figure 2B). Thus, the SUM-FFL displays a sign-sensitive delay, with a delay following OFF but not ON steps of X production.

Figure 2

Experimental dynamics of Z (fliL) expression in a strain containing Y (fliA, strain U306+pJM45+pJM35, RP437flhD, ) and a strain deleted for Y (U307+pJM45+pJM35, RP437flhDfliA, ) (strains and plasmids were described in Kalir and Alon, 2004).(A) Production of X (FlhDC) regulated by the araBAD promoter on a low-copy plasmid was controlled by an inducer externally added to the cells (L-arabinose). The anti-inducer D-fucose allowed deactivation of X expression. (B) Dynamics of Z expression following induction of X. Cells were grown in defined glycerol medium as described (Kalir and Alon, 2004) with saturating inducer (2 mM arabinose), and Z production rate was monitored using GFP controlled by the Z promoter. GFP fluorescence divided by cell density (OD), normalized to a maximum of one, is shown. (C) Dynamics following turn-OFF of X production. Cells were grown with inducer for 3 h and then shifted to medium with no inducer and saturating anti-inducer (50 mM D-fucose).

Full figure and legend (240K)Figures & Tables index

The sign-sensitive delay also occurred in experiments in which X was deactivated following 4 or 5 h of induction (data not shown). During the activation phase of X, the basal body checkpoint appears to be activated so that Y can be active (as seen by the fact that genes regulated by Y and not X, such as the class 3 operon fliC, are activated). A much shorter delay occurred when X was induced for only 2 h before inactivation (data not shown). A short period of FlhDC induction may not allow Y levels to accumulate, or might not be sufficient to form functional basal bodies, and therefore the action of Y would be blocked. This agrees with the theoretical prediction that the delay should depend on the presence of Sy.

We also studied the dynamics of fliL expression in wild-type RP437 cells in which flhDC is under control of its native promoter on the chromosome (that is, in which X is not produced by an inducible promoter as in Figure 2A). In these cells, flagella expression is turned OFF when cells approach the end of exponential growth (Amsler et al, 1993; Kalir et al, 2001; Kalir and Alon, 2004). We compared wild-type cells to cells in which the fliA gene is deleted (fliA). The maximal promoter activity (Kalir and Alon, 2004) of the fliL promoter was lower by about 30% in the fliA strain. To study the response time, we normalized the promoter-activity dynamics by the maximal promoter activity. We find that the wild-type and fliA strains show similar normalized turn-ON dynamics (Figure 3). The turn-OFF of the wild-type cells is delayed with respect to the strains missing fliA (Figure 3). Thus, in the wild-type context, Y appears to prolong the production of Z in the turn-OFF phase of the dynamics.

Figure 3

Experimental dynamics of Z (fliL) expression in wild-type cells (U16+pJM35, RP437 ), and in cells deleted for fliA (U309+pJM35, RP437fliA ). Cells were grown in defined glycerol medium (Kalir and Alon, 2004) and Z production rate was monitored using GFP controlled by the Z promoter. Promoter activity, defined (Kalir and Alon, 2004) as the rate of change of GFP flourescence divided by cell density (OD), normalized to a maximum of one, is shown.

Full figure and legend (138K)Figures & Tables index

Discussion

We find that the SUM-FFL in the flagella system displays sign-sensitive delay, with a significant delay following OFF steps of X production. This qualitatively agrees with theoretical predictions (Mangan and Alon, 2003) for the function of this FFL.

Why is a mechanism needed that prolongs flagella gene expression after the master regulator X is deactivated? The production of the flagella master activator X in wild-type cells is governed by multiple environmental inputs, such as carbon starvation, temperature, osmotic stress and cell density (Figure 1C) (Shi et al, 1993; Aldridge and Hughes, 2002). These factors fluctuate in the environment, especially if the cell swims from place to place. The present results suggest that the SUM-FFL makes the flagella system insensitive to brief periods in which X is deactivated. It allows the flagella system to turn-OFF only when the proper conditions are sensed for a lengthy period of time. The time-scale of the delay generated by the SUM-FFL, about 60-80 min under the present conditions, is comparable to the time needed to complete a flagellum, on the order of 1-2 h (Aizawa and Kubori, 1998; Kalir et al, 2001) (about 1-2 cell generations).

Network motifs appear to allow a qualitative understanding of the dynamics of gene expression in the simple systems studied so far. For example, they allow an understanding of the dynamics of the B. subtilis sporulation network, which is made of several cascaded FFLs, based on the features of each individual FFL (Eichenberger et al, 2004). In the flagella system, the SUM-FFL forms the backbone of the regulatory system, but it does not act in isolation. Rather, it is embedded within the network as part of at least two additional positive feedback loops: auto-regulation of FliA and the basal-body checkpoint (Figure 1C). Despite being embedded in a larger circuit, the flagella SUM-FFL performs sign-sensitive delay as predicted from theoretical analysis (Mangan and Alon, 2003) of the isolated motif. This raises the hope that motifs are, at least in some systems, wired into networks in such a way that allows understanding of the networks dynamics based on the behavior of each individual motif.

Network motifs can serve as qualitative models for the system dynamics, but they cannot be considered as fully detailed models. This is because a detailed model would require many additional subtle mechanisms and interactions, of which many are probably currently uncharacterized. For example, when comparing deletion mutants to wild-type cells, we may be changing not only the connections in the motif, but also other cell components. Thus, the present results should be tested with the quantitative blueprint models of mutant cells, similar to those established for the wild-type flagella system (Kalir and Alon, 2004).

The FFL network motif appears in the transcriptional wiring of organisms from bacteria to humans (Milo et al, 2002, 2004; Mangan et al, 2003; Odom et al, 2004). More generally, it is a basic building block of biological information-processing networks that range from the scale of molecules to the scale of connections between cells (Milo et al, 2002, 2004), such as the network of synaptic connections between neurons in C. elegans. In neuronal networks, the FFL motif seems not to result merely from spatial effects in which neighboring neurons tend to synapse to each other (White et al, 1986): such spatial effects would also produce three-neuron feedback loops, but such feedback loops are found to be rare (Itzkovitz and Alon, 2005). It is possible that the SUM-FFL can act as a delay element also in these networks, protecting the output from transient deactivation.

The present experimental study adds the SUM-FFL to previously studied motifs such as the AND-FFL that can act as a sign-sensitive delay element (Mangan et al, 2003), the incoherent FFL that can generate pulses and speed responses (Mangan and Alon, 2003; Basu et al, 2004), the single-input module that can generate 'just-when-needed' temporal gene expression programs (Laub et al, 2000; Ronen et al, 2002; Shen-Orr et al, 2002; McAdams and Shapiro, 2003; Zaslaver et al, 2004), negative auto-regulation that can speed response times (Savageau, 1974; Rosenfeld et al, 2002) and decrease the variability of steady-state expression (Becskei and Serrano, 2000), and hybrid feedback loops that can generate oscillations (Goldbeter, 2002; Lahav et al, 2004; Nelson et al, 2004). It would be important to characterize and study additional motifs, in order to approach the goal of a complete dictionary of basic circuit elements and their functions (Elowitz and Leibler, 2000; Gardner et al, 2000; Batchelor and Goulian, 2003; Rosenfeld and Alon, 2003; Voigt et al, 2004; Wall et al, 2004).