• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Jul 5, 2005; 102(27): 9517–9522.
Published online Jun 22, 2005. doi:  10.1073/pnas.0500345102
PMCID: PMC1172236

Hysteresis in a synthetic mammalian gene network


Bistable and hysteretic switches, enabling cells to adopt multiple internal expression states in response to a single external input signal, have a pivotal impact on biological systems, ranging from cell-fate decisions to cell-cycle control. We have designed a synthetic hysteretic mammalian transcription network. A positive feedback loop, consisting of a transgene and transactivator (TA) cotranscribed by TA's cognate promoter, is repressed by constitutive expression of a macrolide-dependent transcriptional silencer, whose activity is modulated by the macrolide antibiotic erythromycin. The antibiotic concentration, at which a quasi-discontinuous switch of transgene expression occurs, depends on the history of the synthetic transcription circuitry. If the network components are imbalanced, a graded rather than a quasi-discontinuous signal integration takes place. These findings are consistent with a mathematical model. Synthetic gene networks, which are able to emulate natural gene expression behavior, may foster progress in future gene therapy and tissue engineering initiatives.

Keywords: synthetic biology, synthetic gene networks, bistability, erythromycin, feedback-loop

Hysteresis is a frequent phenomenon with implications for both everyday life as well as life sciences. Traffic jams occur when the car density exceeds a certain threshold value; return to free-flow traffic requires the car density to drop beyond a jam-triggering level (1). In biology, hysteresis plays a role in both macroscopic and microscopic events. Growth of an insect population above a certain threshold value will result in a plague of insects, the control of which requires reduction of the population size well below this threshold level (2). Furthermore, chemotactic behavior of prokaryotes resulting from direction changes of flagellar rotation follow a hysteretic switching mode (3). On a microscopic level, thermal hysteresis proteins, evolved to promote low-temperature survival of different species, lower the freezing point of H2O without significantly altering its melting characteristics (4). Hysteretic feedback control phenomena also manage glucose vs. lactose utilization preference in Escherichia coli (5) and ensure unidirectional cell-cycle progression in eukaryotes (6, 7).

Reductionism in the form of “-omics”-type disciplines has resulted in encyclopedic information on individual molecules and their functions. However, most biologic functions result from an interplay of several network components known as “modules.” New biological disciplines, including modular (8) and network biology (9), evolved to elucidate the general principles governing structure–function phenomena of such modules. Across disciplines, networks are known to be assembled from individual network modules or motifs. Feed-forward loops and bi-fans, known to the digital electronics community as generic circuitry modules, also operate in biological systems to control processes ranging from transcription networks to food webs (10). The system's scope of information flow across a cell's molecular networks compares in complexity and architecture with computer chips and the Internet (9).

Whenever analytic data-based network diagrams are drawn to explain complex cellular phenomena, there is always a degree of uncertainty as to whether all of the components required to elicit a certain effect have been included. To minimize such doubt, the new synthetic biology discipline makes use of another approach to understand the molecular crosstalk of cellular networks: emulation of desired signal integration by de novo construction and analysis of small transgene regulation modules (11). This strategy is compatible with mathematical modeling, because all components of synthetic networks are well characterized. Furthermore, synthetic networks can be rewired by adding or removing components or circuits to gain further insight into molecular signal integration. Synthetic biology has achieved spectacular success, including construction of epigenetic toggle switches and oscillators (1214) in E. coli and epigenetic transgene expression imprinting in mammalian cells (15).

Networks containing positive feedback control enable two-level expression states with hysteretic (16, 17) or bistable (18, 19) switching characteristics. Bistable expression switches quasi-discontinuously flip from OFF to ON and from ON to OFF states once a controlling stimulus reaches a specific threshold (20) (Fig. 1A). However, hysteresis requires a bigger signal to switch from OFF to ON compared with the ON-to-OFF switch. Therefore, hysteretic expression readout depends on the integration of current as well as historic input signals (Fig. 1B). The most extreme form of hysteresis is the epigenetic toggle switch, which imprints two stable expression states, even in the absence of switch-triggering signals. Endogenous and synthetic multistable expression systems have been scrutinized in E. coli (5, 14), which revealed that positive feedback alone was insufficient for multistability. Hysteretic networks with imbalanced expression of network components produced a graded dose-response (5), akin to systems devoid of any positive feedback loop (20) (Fig. 1C).

Fig. 1.
Stimulus vs. response profiles for generic bistable (A), hysteretic (B), and graded (C) gene expression switches and schematic representation of the model hysteretic gene network (D). (A) The most basic bistable expression switch enables a quasi-discontinuous ...

Capitalizing on the interconnection of tetracycline- and macrolide-responsive transgene control modalities (21, 22), we designed and characterized a synthetic mammalian gene circuitry showing hysteretic signal integration. A tetracycline-responsive positive feedback loop, communicating with a constitutive macrolide-dependent transrepressor expression unit, impinges on transcription-modulation of a desired transgene. As predicted in silico, the transcription readout of a well-balanced hysteretic network is either ON or OFF, depending on the presence or absence of regulating macrolide antibiotics. At macrolide concentrations around the switching dose, the network shows hysteretic expression behavior characterized by higher macrolide concentrations required for OFF-to-ON than ON-to-OFF expression changes. Because the expression output of our synthetic hysteretic network is a function of its expression history, it represents an example of artificial memory imprinted on a transgene's expression status in mammalian cells. Synthetic gene networks with hysteretic transgene expression output may be considered to be an important milestone on the path to providing prosthetic networks for gene therapy and tissue engineering in the future.

Materials and Methods

Plasmid Construction. pBP228 (tetO7-ETR8-PhCMVmin-SEAP-IRES-tTA-pA; SEAP, human placental secreted alkaline phosphatase) was constructed by excising tTA from pSAM200 (23) by HpaI/BamHI and cloning it into the SwaI/BglII sites of pBP187 [tetO7-ETR8-PhCMVmin-SEAP-IRES-pA; IRES, internal ribosome entry site of polioviral origin (24)].

Cell Culture, Transfection, and Stable Cell Line Construction. Chinese hamster ovary (CHO) cells CHO-WW198 (W. Weber, personal communication) and CHO-HYST were cultured in FMX-8 medium (Cell Culture Technologies, Zurich) supplemented with 5% FCS (PAA Laboratories, Linz, Austria; Catalog No. A15-022, Lot No. A01129-242) and appropriate selective antibiotics zeocin (100 μg/ml) (CHO-WW198, CHO-HYST) and/or blasticidin (5 μg/ml) (CHO-WW198). CHO-WW198 were transfected at 50% confluence by using FuGENE6 (Roche Diagnostics) according to the manufacturer's protocol by cotransfecting pBP228 and the zeocin resistance-conferring plasmid pZeoSV2 (Invitrogen) at a molar ratio of 15:1 into CHO-WW198. This pBP228:pZeoSV2 ratio increases the frequency of selected zeocin-resistant CHO-WW198 derivatives also containing cotransfected pBP228-encoded expression units. Separation of hysteretic and resistance gene expression units minimizes undesired deregulation of the antibiotic-responsive hybrid promoter by the constitutive zeocin-driving promoter without compromising the selection process (25). After a 2-week selection period (trypsinizing and reseeding after 7 days) in medium containing zeocin and blasticidin, hysteretic SEAP expression (26) of the mixed transgenic population was confirmed as outlined for individual cell clones below. The mixed cell population was subsequently diluted to five cells per ml and seeded into 96-well plates (200 μl per well). After a 2-week incubation period in 96-well plates, supernatants of 50 clones were analyzed for SEAP expression. Positive clones were split into two wells containing zeocin and blasticidin selection medium. One of those wells was supplemented with erythromycin (EM) to assess the transgene regulation performance of individual clones. All 50 clones showed regulated SEAP expression, and CHO-HYST44, CHO-HYST42, and CHO-HYST7 were chosen for further analysis.

SEAP Assay. SEAP production was quantified in cell culture supernatants as described in ref. 26. The SEAP detection limit is 400 nanounits/liter (27).

Quantitative RT-PCR. Total RNA was isolated with the NucleoSpin RNA II kit (Macherey-Nagel, Oensingen, Switzerland) from 106 CHO-HYST7/42/44 cells according to the manufacturer's protocol. Two micrograms of total RNA was reverse transcribed into cDNA in a 100-μl reaction by using TaqMan reverse transcription reagents (Applied Biosystems). Relative quantification of tTA mRNA in CHO-HYST clones was performed in a 25-μl reaction combining the ddCT method, a tTA-specific transcript assay (Applied Biosystems; forward primer, 5′-AAGTGGGTCCGCGTACAG-3′; reverse primer, 5′-AGCAGGCCCTCGATGGTA-3′; probe, 5′-ACCCGTAATTGTTTTTCG-3′), 100 ng of cDNA, and Applied Biosystems 7500 real-time PCR equipment (Applied Biosystems) following suppliers' instructions. Standardized quantification of endogenous 18s-RNA levels ensured appropriate internal controls (Applied Biosystems, product ID Hs99999901_s1).

Regulating Antibiotics. EM (Fluka) was prepared as stock solutions of 1, 0.1, and 0.01 mg/ml in ethanol and used at final concentrations of 0, 125, 250, 500, 750, 1,000, 1,500, and 2,000 ng/ml.

Modeling. berkeley madonna software (www.berkeleymadonna.com) was used to solve model equations and generate parameter plots, and mathematica 5.0 (Wolfram Research, Champaign, IL) was applied to design simulations shown in Fig. 2.

Fig. 2.
Graphic plot of the left- and right-hand side of Eq. 4. For a given α (which indirectly represents transrepressor concentration [TR]), the fate of the hysteretic system is determined by [TA]0. For [TA]0 > a, the final [TA] is b; for [TA] ...


Qualitative Model. The synthetic hysteretic gene network has the following setup (Fig. 1D): A transactivator (TA) activates its own transcription by binding to a chimeric promoter (Phybrid), which also can be repressed by a specific constitutively expressed [constitutive promoter (Pconst)-driven] transrepressor (TR) whose activity is modulated by a clinically licensed antibiotic. TA as well as TR bind to their specific palindromic operator within Phybrid in a cooperative manner. In the absence of antibiotics, transactivator and transrepressor bind to their operator. In such a situation, the transrepressor action dominates over transactivator function (24). The behavior of the hysteretic gene network can be described by using the following dimensionless equation:

equation M1

where [TA] is the concentration of the transactivator TA, p is the strength of Phybrid, [TR] is the concentration of active (not inactivated by antibiotic) transrepressor molecules, and deg is the degradation rate of TA. The term [TA]2/(1 + [TA]2) describes TA-mediated induction of positive feedback control. The exponent 2 reflects cooperative binding of TA dimers to their palindromic binding sites. The maximum intracellular [TA] is subject to physiologic constraints, which are considered by the term (1 - [TA]/2.5) to limit [TA] to an arbitrary unit of 2.5. The term 1/(1 + [TR]2) reflects the repression of the hybrid promoter by TR. TA degradation depends on its degradation rate deg. Below, we define [TA]0 as the initial [TA], [TA] as the state at which TA expression is OFF, and [TA] as the state at which TA expression is ON.

To analyze the geometric structure of the model, the nullclines of Eq. 1 were determined as follows. First, d[TA]/dt was set to 0, from which follows:

equation M2

The parameters deg, p, and (1 + [TR]2) were lumped together to create a new parameter α = p/(deg × (1 + [TR]2)), resulting in

equation M3

Eq. 3 has a stable fixed point at [TA] = 0. The intuitive explanation is that no positive feedback expression can be induced in the absence of TA. The other fixed points of Eq. 3 result from the solutions of

equation M4

These fixed points can be found graphically by plotting the left- and the right-hand sides of Eq. 4 and by seeking for the intersections. Fig. 2 shows that there are zero, one, or two intersections, depending on the value of α. Assume an α for which there are two intersections a and b. As α decreases, the two intersections a and b approach each other and eventually coalesce in a so-called saddle-node bifurcation (28) when the line intersects the curve tangentially. If α decreases further, there are no intersections and therefore no additional solutions apart from [TA] = 0. To determine the stability, we take into account that [TA] = 0 is stable (positive feedback expression fails to start in the absence of transactivator). Along the [TA]-axis, the type of stability alternates (28). Therefore, a must be unstable and b must be stable. These findings are logic if the architecture of the network is recalled. For [TA]0 = 0, there is no transactivator that could initiate positive feedback expression. At point a, the TA degradation and production are equal. However, if [TA]0 decreases slightly, the resulting positive feedback expression is weaker and [TA] drops to [TA]. If [TA]0 increases slightly, TA production increases dramatically and [TA] increases to the new fixed point b. The unstable point a represents a threshold [TA]0, below which [TA] drops to [TA] and above which [TA] jumps to the [TA] at b. In summary, the fate of the system is determined by [TA]0. [TA] is achieved for [TA]0 > a. Fig. 2 also reveals an important feature of the system. Hysteresis exclusively occurs if there are one unstable and two stable fixed points. This result is only the case if the curve intersects the line twice as in Fig. 2 A and B. The basis for two intersections is a curve approaching 0 for very low and high [TA] with a maximum as shown in Fig. 2. The shape of this curve is due to cooperative binding of the transactivator as described by the term [TA]2/(1 + [TA]2). If positive feedback control were described by a Michaelis–Menten-like enzyme kinetics equation of the [TA]/(1 + [TA]) type, the right-hand side of Eq. 4, (1/(1 + [TA]) × (1 - [TA]/2.5) × α), would be hyperbola-like and thus intersect the line only once. As a result, the system would have only one stable point. All [TA]0 > 0 would impinge on this stable point, and no hysteretic effect would be observed. Thus, cooperative binding of the TA is a prerequisite for hysteresis.

The qualitative behavior of the model remains stable within a two-orders-of-magnitude range of promoter strength and degradation rate values.

Semiquantitative Model. To determine whether the network behaves hysteretically, [TA]0 must be adjustable to different initial values. In the model, it is sufficient to change α numerically. In an in vivo situation, [TA]0 is modulated by controlling the active [TR] using the antibiotic AB. To obtain a semiquantitative model, inactivation of TR by AB must be taken into account. Parameter α was defined as α = p/(deg × (1 + TR2)). If AB is bound to TR, then TR is inactive. The fraction θ of a fixed [TR] bound to a varying [AB] can be described as

equation M5

where K = 1/[AB]1/2 is the inverse of the concentration, at which half of the transrepressors are bound to AB. Based on experimental data, [AB]1/2 was estimated to be 200 (ng/ml) for macrolide-responsive expression systems (22). The active TR concentration (with an arbitrary maximum of 10, which exceeds the maximum [TA] of 2.5 due to expression from a constitutive promoter) as a function of the antibiotic concentration is therefore

equation M6

By feeding TR([AB]) back into Eq. 1, a model with a quantitative antibiotic input concentration (ng/ml) is obtained. Yet, the model expression output is still in arbitrary units.

equation M7
equation M8

Promoter strength p was set to 1 per time unit, and TA's degradation rate was set to 0.01 (29). The mathematical model based on these parameter values predicts a final [TA] of zero for inputs of <100 ng/ml AB. If the system operates at [TA], it starts switching from the ON to the OFF state at a concentration of ≈500 ng/ml AB. The antibiotic concentration at which the output of the network switches from OFF to ON depends on [TA]0↓. In vivo, [TA] will decrease to values slightly above zero due to leaky TA expression. Because the basal TA expression cannot be precisely determined, the model can only predict the emergence of hysteresis but not at which [AB] a network initially set to [TA]0↓ will jump to [TA] (Fig. 3).

Fig. 3.
Parameter plot showing the dependence of the final [TA] on [AB] for different [TA]0. [TA]0 is crucial to determine the [AB] required to switch the system from OFF to ON. The numbers above the curve indicate for which [TA]0 the final [TA] switches from ...

In Vivo Implementation of the Synthetic Hysteretic Gene Network in CHO Cells. The in vivo implementation of the synthetic hysteretic gene network was achieved as follows (Fig. 4). The tetracycline-dependent transactivator (tTA) (21) induces a hybrid promoter (tetO7-ETR8-PhCMVmin) (24) driving its own as well as SEAP expression by means of a positive feedback loop. The macrolide-dependent transrepressor (E-KRAB; ref. 22) represses tetO7-ETR8-PhCMVmin-driven transcription in an EM-responsive manner. This prototype hysteretic network was engineered on two plasmids (pBP228 and pWW198; Fig. 4) to prevent interference resulting from E-KRAB-mediated autotrans-silencing.

Fig. 4.
Molecular configuration of the hysteretic synthetic mammalian gene network. The tetracycline-dependent transactivator tTA, a fusion of the E. coli TetR repressor and the Herpes simplex VP16 transactivation domain (TetR–VP16), binds to its cognate ...

To characterize the synthetic hysteretic gene network, a double-transgenic cell line CHO-HYST, stably harboring pWW198 (PSV40-E-KRAB-pA) and pBP228, was constructed. For our experiments, we chose clone CHO-HYST44, whose SEAP expression was repressed by E-KRAB to undetectable levels in the absence of EM. In addition to CHO-HYST44, we selected clones CHO-HYST42 and CHO-HYST7, the maximum SEAP expression levels of which were 6- and 25-fold higher compared with CHO-HYST44 and could be repressed 23 ± 2-fold and 10 ± 1-fold in the presence of EM, respectively.

Because tTA and SEAP are expressed cocistronically, [tTA]0↑ and [tTA]0↓ correlates to SEAP0↑ and SEAP0↓. CHO-HYST44 was cultivated in the presence and absence of 1 μg/ml EM for 3 days to achieve SEAP and SEAP. A period of 3 days is beyond the time required to switch SEAP expression from OFF to ON or from ON to OFF (22). Cells were then trypsinized, washed twice for 20 min (to remove residual EM), and reseeded at a density of 40,000 cells per well of a 24-well plate and cultivated at different EM concentrations ([EM]). As shown in Fig. 5A, the population with an EM-free cultivation history (resulting in SEAP0↓) required 1,000 ng/ml EM to completely switch SEAP expression from OFF to ON. However, CHO-HYST44 populations with an EM-containing cultivation history (resulting in SEAP0↑) switched SEAP expression OFF when [EM] dropped to <500 ng/ml. The response of the hysteretic gene network to different [EM] is imprinted by the population's EM cultivation history. Because SEAP expression levels for [EM] of >1,000 ng/ml are identical, a temporal effect resulting from residual SEAP in the secretory pathway preswitching cannot account for the differences in the OFF-to-ON vs. ON-to-OFF expression level of the synthetic hysteretic gene network. Such a temporal effect would have resulted in higher SEAP expression levels for cells preset in the ON state, irrespective of the [EM] after the population was split. The SEAP expression state was fully reversible and could repeatedly be switched from OFF to ON or ON to OFF (Fig. 5B).

Fig. 5.
Validation of hysteretic behavior (A) and reversibility of switching (B) of CHO-HYST44. (A) CHO-HYST44 populations, double transgenic for pWW198 and pBP228, were cultivated in the presence (+) and absence (-) of EM for 3 days to set SEAP expression to ...

Fitting the Model to Experimental Data. As mentioned above, [EM] required to switch the hysteretic gene network from OFF to ON depends on [tTA]0↓ after cultivation in EM-free medium. Expression data from CHO-HYST44 cultivations revealed that an [EM] of 1,000 ng/ml was required to completely switch SEAP expression from the OFF to ON state. To determine which [tTA]0↓ would result in an OFF-to-ON expression switch at 1,000 ng/ml, a variety of parameter plots correlating [tTA]final and [EM] for different [tTA]0 were generated (Fig. 3). We found that [tTA]0↓ up to 55-fold lower than [tTA]0↑ resulted in an OFF-to-ON switch at 1,000 ng/ml EM (Fig. 5A). Whenever [tTA]0↓ is >55 times lower compared with [tTA]0↑, OFF-to-ON switches occur at higher [EM]. This simulation was confirmed by RT-PCR-based tTA-transcript quantification in CHO-HYST44 ([tTA]0↓/[tTA]0↓, 1 ± 0.1; [tTA]0↑/[tTA]0↓, 54 ± 13).

Imbalanced Building Blocks. To determine whether cells harboring a hysteretic gene network with imbalanced expression of individual network components exhibit graded rather than hysteretic target gene expression, we subjected both CHO-HYST42 (intermediate-level SEAP expression; [tTA]) and CHO-HYST7 (high-level SEAP expression; [tTA]) to the same cultivation procedure that resulted in the data shown for CHO-HYST44 (low-level SEAP expression; [tTA]) in Fig. 5A. The relative [tTA] in CHO-HYST7/42/44 as deduced from their SEAP expression levels was confirmed by quantitative tTA transcript-specific RT-PCR (CHO-HYST44, 1 ± 0.2; CHO-HYST42, 2.2 ± 0.4; CHO-HYST7, 5.4 ± 1; all tTA mRNA levels were normalized to the ones of CHO-HYST44.). Although CHO-HYST42 exhibited residual [EM]-imprinted hysteresis revealed by significantly different SEAP production levels within an [EM] of 125–500 ng/ml, CHO-HYST42's dose-response characteristics had considerably shifted from a bistable to a graded response profile (Fig. 6A). In comparison with CHO-HYST44 and CHO-HYST42, CHO-HYST7 exhibited the highest SEAP expression ([tTA]). The OFF-to-ON and ON-to-OFF SEAP expression profiles of CHO-HYST7 were graded between 0 and 2,000 ng/ml and completely reversible without showing any signs suggestive of hysteresis. The SEAP dose-response profiles of CHO-HYST7 matched those of CHO-E-SEAP, a stable EM-inducible cell line devoid of positive feedback control, at a high standard (see figure 3A in ref. 22).

Fig. 6.
SEAP expression profiling of CHO-HYST42 (A) and CHO-HYST7 (B). (A) CHO-HYST42 was cultivated in the presence (+) and absence (-) of EM for 3 days to set SEAP expression to an ON (+EM) or OFF (-EM) state. The cells then were reseeded under different [EM] ...

Encouraged by the high degree of correlation between the model and the SEAP expression profiles of CHO-HYST44, we also tested our model for CHO-HYST7. The SEAP expression capacity of CHO-HYST7 is 25 times higher compared with that of CHO-HYST44. Furthermore, SEAP expression of CHO-HYST7 is only one order of magnitude below maximum expression levels, and the expression of this glycoprotein is repressed to the detection level under identical conditions in CHO-HYST44. The model Eq. 7 was adapted to account for the different expression profiles of CHO-HYST44 and CHO-HYST7 as follows: The maximally allowed [tTA] [characterized by the term (1 - [tTA]/2.5)], which is the measured output in the model and is proportional to the SEAP expression in vivo was raised 25-fold to result in 1 - [tTA]/62.5. [tTA]0 values were set to the maximum level of 62.5 for pretreatment in EM-containing medium and to 6.25 for precultivation in EM-free medium (reflecting a 10-fold reduction). A parameter plot for the final [tTA] in dependence of [tTA]0 and the current [EM] did not reveal a hysteretic effect. Independent of their cultivation history, the [tTA]0↑ and [tTA]0↓ populations were predicted to show identical SEAP expression profiles. Furthermore, we found that the concentration window for the ON-to-OFF switch broadened significantly, reaching a range of 0 to ≈3,000 ng/ml. Between 0 and 2,000 ng/ml, the output of the network was almost linearly dependent on the antibiotic input (Fig. 6B).


Transcriptional coupling of a tetracycline-dependent transactivator (tTA), which induces its own transcription along with the transgene in a positive-feedback-loop manner, with an EM-dependent transrepressor (E-KRAB) able to repress positive feedback expression, enables hysteretic transgene control in mammalian cells. For hysteretic operation of the synthetic gene network, E-KRAB and tTA compete for Phybrid interaction. At high [EM] (high levels of Phybrid binding-incompetent E-KRAB), Phybrid-induction is biased toward tTA, resulting in ON-to-OFF SEAP expression switches (or maintenance of high-level SEAP expression). At low [EM] (high levels of Phybrid binding-competent E-KRAB), E-KRAB takes precedence over tTA for Phybrid binding, which results in ON-to-OFF SEAP expression switches (or maintenance of low-level SEAP expression). Only at intermediate [EM], when either tTA or E-KRAB may prevail for Phybrid modulation and the gene network manages either high-level ([tTA]) or low-level ([tTA]) SEAP expression, hysteresis could occur. The resulting transgene expression state depends on [tTA]0, which is itself a function of the cell's cultivation history. The amount of E-KRAB required to prevent induction of the tTA-dependent positive feedback loop increases with increasing [tTA]0.

Our observation that well-balanced relative transcription factor concentrations are required to generate hysteretic target gene expression is congruent with previous findings showing conversion of hysteretic signal integration of the E. coli lactose utilization network into a graded response profile by providing mock lactose operators to titrate the lac repressor away from the lactose promoter (5). As exemplified by CHO-HYST44, CHO-HYST42, and CHO-HYST7, our synthetic gene network gradually changes from hysteretic to graded expression profiles the more the tTA/E-KRAB levels are biased toward tTA. Binding competition between tetracycline-dependent transactivators and transrepressors for the same operator has been suggested to account for conversion of a graded rheostat to a bistable all-or-nothing switch (30, 31). Akin to such conversion, the bistable hysteretic network behavior (CHO-HYST44) moves toward a graded rheostat as tTA outcompetes E-KRAB for binding to the hybrid operator module (CHO-HYST7). CHO-HYST42 represents a transition state characterized by a typical expression profile resulting from chimeric hysteretic-rheostat expression qualities. Besides binding competition between transcription factors, their binding kinetics were suggested to impact graded vs. all-or-nothing response profiles (32).

Transcriptional noise is known to contribute to expression heterogeneity within clonal eukaryotic populations (33). Noise intensities are typically low for fully repressed or induced expression states and become significant at intermediate expression levels (33). Because the bistable hysteretic network operates at either ON or OFF state where noise levels are considered low, noise is expected to have little impact on hysteretic network behavior. Also, all hysteretic network components are encoded on the chromosome that is known for its noise-filtering capacity (33).

The design and construction of systems that exhibit complex dynamic behavior remain the major goal of the synthetic biology community (34). An educated choice of network components as well as their assembly in precise crosstalk configurations may enable synthetic networks to increase our understanding of natural processes as well as foster therapeutic advances (14).

Positive and double-negative feedback regulation circuits, able to convert a graded into a binary response, are common control themes in nature, which evolved to manage cell fate decisions including chemotaxis, myogenesis, and cell cycle across species (19, 35, 36). Therefore, synthetic networks incorporating virus-, bacteria-, or mammalian cell-derived building blocks show a high degree of interoperability and network design principles evolved in E. coli also function in a mammalian cell (15). Such generic-ness among gene control circuitries increases hope for successful therapeutic interventions in future gene therapy and tissue engineering initiatives.


We thank Wilfried Weber (Institute for Chemical and Bio-Engineering, Eidgenössische Technische Hochschule Zurich, Zurich) for providing cell line CHO-WW198 and critical comments on the manuscript and Susanne Moelbert for help with the mathematical model. This work was supported by Swiss National Science Foundation Grant 631-065946, the Swiss State Secretariat for Education and Research within EC Framework 6, and Cistronics Cell Technology GmbH (Zurich).


Author contributions: B.P.K. and M.F. designed research; B.P.K. performed research; B.P.K. and M.F. analyzed data; and B.P.K. and M.F. wrote the paper.

This paper was submitted directly (Track II) to the PNAS office.

Abbreviations: EM, erythromycin; SEAP, human placental secreted alkaline phosphatase.


1. Ball, P. (2004) Critical Mass: How One Thing Leads to Another (William Heinemann, London).
2. Ludwig, D., Jones, D. D. & Holling, C. S. (1978) J. Anim. Ecol. 47, 315-332.
3. Bren, A. & Eisenbach, M. (2001) J. Mol. Biol. 312, 699-709. [PubMed]
4. Barrett, J. (2001) Int. J. Biochem. Cell Biol. 33, 105-117. [PubMed]
5. Ozbudak, E. M., Thattai, M., Lim, H. N., Shraiman, B. I. & Van Oudenaarden, A. (2004) Nature 427, 737-740. [PubMed]
6. Sha, W., Moore, J., Chen, K., Lassaletta, A. D., Yi, C. S., Tyson, J. J. & Sible, J. C. (2003) Proc. Natl. Acad. Sci. USA 100, 975-980. [PMC free article] [PubMed]
7. Tyson, J. J. & Novak, B. (2001) J. Theor. Biol. 210, 249-263. [PubMed]
8. Hartwell, L. H., Hopfield, J. J., Leibler, S. & Murray, A. W. (1999) Nature 402, C47-C52. [PubMed]
9. Barabasi, A. L. & Oltvai, Z. N. (2004) Nat. Rev. Genet. 5, 101-113. [PubMed]
10. Milo, R., Shen-Orr, S., Itzkovitz, S., Kashtan, N., Chklovskii, D. & Alon, U. (2002) Science 298, 824-827. [PubMed]
11. Hasty, J., McMillen, D. & Collins, J. J. (2002) Nature 420, 224-230. [PubMed]
12. Gardner, T. S., Cantor, C. R. & Collins, J. J. (2000) Nature 403, 339-342. [PubMed]
13. Elowitz, M. B. & Leibler, S. (2000) Nature 403, 335-338. [PubMed]
14. Atkinson, M. R., Savageau, M. A., Myers, J. T. & Ninfa, A. J. (2003) Cell 113, 597-607. [PubMed]
15. Kramer, B. P., Viretta, A. U., Daoud-El-Baba, M., Aubel, D., Weber, W. & Fussenegger, M. (2004) Nat. Biotechnol. 22, 867-870. [PubMed]
16. Angeli, D., Ferrell, J. E., Jr., & Sontag, E. D. (2004) Proc. Natl. Acad. Sci. USA 101, 1822-1827. [PMC free article] [PubMed]
17. Xiong, W. & Ferrell, J. E., Jr. (2003) Nature 426, 460-465. [PubMed]
18. Isaacs, F. J., Hasty, J., Cantor, C. R. & Collins, J. J. (2003) Proc. Natl. Acad. Sci. USA 100, 7714-7719. [PMC free article] [PubMed]
19. Becskei, A., Seraphin, B. & Serrano, L. (2001) EMBO J. 20, 2528-2535. [PMC free article] [PubMed]
20. Ninfa, A. J. & Mayo, A. E. (May 11, 2004) Sci. STKE, .10.1126/stke.2322004pe20 [PubMed] [Cross Ref]
21. Gossen, M. & Bujard, H. (1992) Proc. Natl. Acad. Sci. USA 89, 5547-5551. [PMC free article] [PubMed]
22. Weber, W., Fux, C., Daoud-El Baba, M., Keller, B., Weber, C. C., Kramer, B. P., Heinzen, C., Aubel, D., Bailey, J. E. & Fussenegger, M. (2002) Nat. Biotechnol. 20, 901-907. [PubMed]
23. Fussenegger, M., Moser, S., Mazur, X. & Bailey, J. E. (1997) Biotechnol. Prog. 13, 733-740. [PubMed]
24. Kramer, B. P., Fischer, C. & Fussenegger, M. (2004) Biotechnol. Bioeng. 87, 478-484. [PubMed]
25. Wurm, F. M. (2004) Nat. Biotechnol. 22, 1393-1398. [PubMed]
26. Schlatter, S., Rimann, M., Kelm, J. & Fussenegger, M. (2002) Gene 282, 19-31. [PubMed]
27. Berger, J., Hauber, J., Hauber, R., Geiger, R. & Cullen, B. R. (1988) Gene 66, 1-10. [PubMed]
28. Strogatz, S. H. (1994) Nonlinear Dynamics and Chaos (Perseus, Reading, MA).
29. Pratt, J. M., Petty, J., Riba-Garcia, I., Robertson, D. H., Gaskell, S. J., Oliver, S. G. & Beynon, R. J. (2002) Mol. Cell Proteomics 1, 579-591. [PubMed]
30. Rossi, F. M., Kringstein, A. M., Spicher, A., Guicherit, O. M. & Blau, H. M. (2000) Mol. Cell 6, 723-728. [PubMed]
31. Louis, M. & Becskei, A. (July 30, 2002) Sci. STKE, .10.1126/stke.2002.143.pe33 [PubMed] [Cross Ref]
32. Pirone, J. R. & Elston, T. C. (2004) J. Theor. Biol. 226, 111-121. [PubMed]
33. Blake, W. J., Kaern, M., Cantor, C. R. & Collins, J. J. (2003) Nature 422, 633-637. [PubMed]
34. Brent, R. (2004) Nat. Biotechnol. 22, 1211-1214. [PubMed]
35. Pomerening, J. R., Sontag, E. D. & Ferrell, J. E., Jr. (2003) Nat. Cell Biol. 5, 346-351. [PubMed]
36. Ferrell, J. E., Jr. (2002) Curr. Opin. Cell Biol. 14, 140-148. [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...