SPECIAL SECTION
NETWORKS IN BIOLOGY
22. 23. 24. 25. 26. 27. 28.
of the Primate Cerebral Cortex, D. Falk, K. R. Gibson, Eds. (Cambridge Univ. Press, Cambridge, 2001), pp. 57–78. D. Attwell, S. B. Laughlin, J. Cereb. Blood Flow Metab. 21, 1133 (2001). P. Lennie, Curr. Biol. 13, 493 (2003). V. Braitenberg, A. Schu¨tz, Cortex: Statistics and Geometry of Neuronal Connectivity (Springer, Berlin, ed. 2, 1998). J. A. White, J. T. Rubinstein, A. R. Kay, Trends Neurosci. 23, 131 (2000). A. Faisal, J. A. White, S. B. Laughlin, in preparation. S. Schreiber, C. K. Machens, V. A. Herz, S. B. Laughlin, Neural Comput. 14, 1323 (2002). K. M. Franks, T. J. Sejnowski, Bioessays 12, 1130 (2002).
29. M. C. W. van Rossum, B. O’Brien, R. G. Smith, J. Neurophysiol. 89, 406 (2003). 30. J. von Neumann, The Computer and the Brain (Yale Univ. Press, New Haven, CT, 1958). 31. R. Sarpeshkar, Neural Comput. 10, 1601 (1998). 32. T. von der Twer, D. I. MacLeod, Network 12, 395 (2001). 33. H. B. Barlow, Perception 1, 371 (1972). 34. E. P. Simoncelli, B. A. Olshausen, Annu. Rev. Neurosci. 24, 1193 (2001). 35. R. Baddeley et al., Proc. R. Soc. London Ser. B 264, 1775 (1997). 36. V. Balasubramanian, M. J. Berry, Network 13, 531 (2002). 37. D. J. Field, Neural Comput. 6, 559 (1994).
38. W. B. Levy, R. A. Baxter, Neural Comput. 8, 531 (1996). 39. V. N. Murthy, T. J. Sejnowski, C. F. Stevens, Neuron 18, 599 (1997). 40. W. B. Levy, R. A. Baxter, J. Neurosci. 22, 4746 (2002). 41. Z. F. Mainen, T. J. Sejnowski, Science 268, 1503 (1995). 42. L. E. Dobrunz, C. F. Stevens, Neuron 18, 995 (1997). 43. T. J. Sejnowski, Neuron 24, 773 (1999). 44. C. von der Malsburg, Neuron 24, 95 (1999). 45. P. Fries, J. H. Reynolds, A. E. Rorie, R. Desimone, Science 291, 1560 (2001). 46. B. Pesaran, J. S. Pezaris, M. Sahani, P. P. Mitra, R. A. Andersen, Nature Neurosci. 5, 805 (2002).
REVIEW
A Bacterial Cell-Cycle Regulatory Network Operating in Time and Space Harley H. McAdams* and Lucy Shapiro Transcriptional regulatory circuits provide only a fraction of the signaling pathways and regulatory mechanisms that control the bacterial cell cycle. The CtrA regulatory network, important in control of the Caulobacter cell cycle, illustrates the critical role of nontranscriptional pathways and temporally and spatially localized regulatory proteins. The system architecture of Caulobacter cell-cycle control involves topdown control of modular functions by a small number of master regulatory proteins with cross-module signaling coordinating the overall process. Modeling the cell cycle probably requires a top-down modeling approach and a hybrid control system modeling paradigm to treat its combined discrete and continuous characteristics. In the past few years, microarrays and other high-throughput experimental methods have supported cellwide or “global” assays of gene expression activity in cells. We can realistically expect these methods to yield virtually complete descriptions of transcriptional wiring diagrams of some yeast and bacterial species, perhaps within the next decade. However, the transcriptional regulatory circuitry provides only a fraction of the signaling pathways and regulatory mechanisms that control the cell. Rates of gene expression are modulated through posttranscriptional mechanisms that affect mRNA half-lives and translation initiation and progression, as well as DNA structural or chemical state modifications that affect transcription initiation rates. Phosphotransfer cascades provide fast point-to-point signaling and conditional signaling mechanisms to integrate internal and external status signals, activate regulatory molecules, and coordinate the progress of diverse asynchronous pathways. As if this were not complex enough, we are now finding that the interior of bacterial cells is highly spatially structured, with the cellular position Department of Developmental Biology, Stanford University School of Medicine, B300 Beckman Center, Stanford, CA 94305, USA. *To whom correspondence should be addressed. Email:
[email protected]
1874
of many regulatory proteins as tightly controlled at each time in the cell cycle as are their concentrations.
Architecture of Bacterial Cell-Cycle Control System A few top-level master regulatory proteins organize and coordinate multiple processive modular functions that do much of the work to execute the cell cycle (1, 2). Herein, “modules” refers to groups of proteins that work together to execute a function (3). Bacterial genes for module proteins are frequently organized in one or several operons. Proteins in a module generally interact strongly with each other, either by forming multiprotein machines that accomplish complex functions (e.g., gene transcription, mRNA translation, and chromosome replication) or by creating signaling or reaction pathways. Production of the proteins constituting these modules is tightly controlled so that they are expressed just at the time needed and in the order required for assembly (1, 4, 5) and destroyed when they are not needed. Stoichiometry and, frequently, the place of expression in the cell also are often controlled (5). Once launched, modules act autonomously (e.g., modular metabolic pathways) or processively (e.g., the replisome, RNA polymerase, and the construction of structures such as flagella and pili) to complete their job. Because of inherent stochastic differences in the time required to execute any reaction cascade
(6), intermodule regulatory linkages are required for synchronization when maintenance of relative timing is important. Top-level master regulators work together to switch modular functions on and off in orderly succession (1) as needed to progress through the cell cycle. Many of the genes controlled by master regulator proteins are themselves regulatory molecules, so that coordinated activation of a complex mix of activities can be achieved. For example, in the yeast Saccharomyces cerevisiae, 12 cell-cycle regulatory proteins control a subset of regulators forming an interconnected loop of activators that contribute to cell-cycle progression (7). Other master regulator proteins control complex cell responses such as the Escherichia coli general stress response (8) and the activation of the Bacillus subtilis sporulation pathway (9). The advantages of a regulatory circuit architecture involving hierarchical top-down control of modular functions by master regulators may relate to simplification of control signal communication pathways, as well as to facilitation of both short-term (i.e., immediate within the resources of the current cell design) and long-term [i.e., evolutionary evolvability (10)] adaptation to changes in environmental conditions. Disadvantages include the creation of lethal failure points when regulation of large numbers of critical functions depends on a limited number of master regulatory proteins. Failures can also occur because of stochastic risk of failure in any given pathway (6). This risk is alleviated in cells by redundant control of critical regulatory proteins, as we see in the case of CtrA in Caulobacter. Caulobacter crescentus is a productive model system for the study of bacterial cell-cycle regulation and the mechanisms of asymmetric cell division. Caulobacter always divides asymmetrically, producing daughter cells with differing polar structures, different cell fates, and asymmetric
26 SEPTEMBER 2003 VOL 301 SCIENCE www.sciencemag.org
NETWORKS IN BIOLOGY
Synchronization of Cell Functions and Control of CtrA Timing During the course of the cell cycle, functions such as flagella construction, chromosome replication, and cell division must be both started and completed in proper order. The time required for the modules executing these complex multistage actions to complete their work will vary depending on (i) environmental considerations, such as variations in the cell’s nutritional state; (ii) the inherently stochastic temporal pattern of protein production from activated genes (6); and (iii) natural stochastic differences in the time to complete a chemical reaction cascade. Reaction cascades can be quite complex, and multiple cascades may have to be completed before a given reaction can proceed. The time to complete any reaction cascade is never precisely predictable. However, by adding regulatory complexity, the timing can be
SPECIAL SECTION
regulation of the initiation of chromosome replication (Fig. 1A). The role of master regulatory proteins in top-down control of functional modules is well illustrated by Caulobacter’s CtrA regulatory network (Fig. 1B), which is parallel in many ways to the Spo0A master regulator of sporulation in B. subtilis (11). Examples of CtrAregulated modular functions include flagella biogenesis, the chemotaxis control subsystem, the divisiome (FtsZ), the pili structure, and the replisome. Both CtrA and Spo0A are response regulator members of two-component signal-transduction pathways that control large numbers of genes involved in multiple modular functions. CtrA controls, directly or indirectly, 26% of the Caulobacter cell-cycle–regulated genes and directly controls 95 genes in 55 different operons (1, 2). There are presumably additional as-yet-unidentified master regulators of comparable scope that in coordination with CtrA control the remaining 74% of the cell-cycle–regulated genes. The “just-in-time” principle for expression of genes supporting modular functions can be seen in the early derepression of metabolic and ribosomal genes as swarmer cells differentiate into stalked cells, in the timing of ftsZ (the earliest expressed gene in cell division) expression, and in the timing of flagella biogenesis and the associated chemotaxis control subsystem (Fig. 1B). Another example is the timing of production of CtrA. In a narrow window of the cell cycle, Caulobacter produces ⬃22,000 CtrA molecules (12). Why so many CtrA molecules are needed so quickly is unclear. Perhaps the purpose is to minimize stochastic variations in start times of important cellcycle functions that must complete before cell division. In any case, the Caulobacter cell is designed to produce the CtrA molecules at just the right time relative to the progress of chromosome replication, and then, a little later, to remove them with equal alacrity by activating an as-yet-unidentified cell-cycle–regulated recognition factor that delivers CtrA to the ClpP/ClpX proteolysis system (Fig. 1).
Fig. 1. (A) Schematic of the Caulobacter cell cycle. The motile swarmer cell has a single polar flagellum, polar chemotaxis receptors, and multiple polar pili. Shaded cells contain CtrA. Chromosome replication is repressed in swarmer cells by binding of CtrA to the origin of replication. After about a third of its cell cycle, the swarmer discards the flagellum, chemoreceptors, and pili as it differentiates into a stalked cell. At the swarmer-to-stalked cell transition, CtrA is cleared from the cell, allowing the initiation of DNA replication (shown as the -like structure in the cell). Upon completion of separation of the predivisional cell into two compartments before cell division, CtrA is cleared just from the nascent stalked cell compartment, allowing replisome formation while maintaining repression of replication initiation in the swarmer compartment (12). (B) Genetic network controlled by the CtrA response regulator. The transcriptional network directly controlled by CtrA (green lines) is integrated with nontranscriptional regulatory pathways (red lines). A negative feedback loop through the CcrM DNA methylation pathway ties the timing of CtrA synthesis to a specific point in the progress of chromosome replication. Activation by one or more phosphorelays involving the membranelocalized CckA histidine kinase (15), and probably the DivL tyrosine kinase (32) and the DivK response regulator (33), are required and thought to integrate other currently unknown signals into the control of timing of CtrA activation. Red indicates regulatory genes. Timing in the cell cycle of transcriptional regulation (as assayed by gene expression microarrays and lacZ transcriptional fusions) is indicated by the bars on the right. Numbers over each bar indicate the numbers of genes regulated in that time period. Redundant control of the timing of chromosome replication involves both repression of the origin of replication at five CtrA binding sites (34) and transcriptional repression of components of the replication machinery by an unknown factor (35). The metabolic and ribosomal proteins controlled by CtrA are derepressed in the swarmer cell at about the time of the swarmer-to-stalked differentiation, perhaps to support accelerated growth in the stalked cell.
www.sciencemag.org SCIENCE VOL 301 26 SEPTEMBER 2003
1875
SPECIAL SECTION
NETWORKS IN BIOLOGY made more predictable, as in the case of the regulatory network controlling circadian rhythms (13). Even when different functions are started at exactly the proper times in the cell cycle, maintenance of synchronization requires regulatory signaling between modules at intermediate points (checkpoints) to coordinate the overall process (14). This cross-module signaling involving mechanisms such as two-component systems or phosphorelays, either blocks cell-cycle progress until necessary precursor functions are completed or initiates coordinated responses to emergencies such as nutrient shortages or DNA damage. Improper timing of some events in the cell cycle will have fatal consequences; thus, highly reliable synchronization mechanisms are required.
An obvious example is the need to complete separation of the two chromosomes before completing cell division. Other critical timing events in Caulobacter include: (i) somewhat surprisingly, the timing of CtrA expression versus the state of progression of chromosome replication noted above, and (ii) the timing of initiation of CtrA degradation in the stalked compartment of the late predivisional cell. In the swarmer cell, CtrA binds to and silences the chromosomal origin of replication; later, CtrA has to be cleared from the stalked cell for initiation of replication. Because chromosome replication is differentially regulated in the swarmer and stalked progeny, Caulobacter faces a tricky problem in the predivisional cell: It has to quickly and reliably get rid of CtrA in the
Fig. 2. Spatial and temporal dynamics of selected regulatory and structural proteins. (A) Scaled cell cross sections starting with the newly divided swarmer cell (0 min) and proceeding to cell division (240 min). Notable features include the diverse dynamics of histidine kinases and response regulators that are, for example, variously synthesized and degraded (CtrA), localized to the membrane and delocalized to the cytoplasm (DivK), localized to the pole and delocalized in the membrane (DivJ), and transiently present and localized at a pole (PleC); the controlled positioning of the chromosome origin and terminus of replication as well as the replisome; the continuously changing complement of regulatory and structural proteins at the cell poles; and the bifurcating cytoplasmic chemistry at the time of cell compartmentalization (⬃220 min). (B) Regulatory connections between localized features in (A). CckA is delocalized in the membrane of the swarmer cell and early stalked cell and accumulates at the predivisional cell pole, concomitant with transfer of its phosphate moiety to newly synthesized CtrA (140 min). The membrane-bound PleC histidine kinase is required for signaling polar pili assembly, biogenesis of holdfast material at the tip of the stalk, and chemotaxis functions. PleC is localized at the flagellated pole (⬃140 min) and delocalized at the swarmer-to-stalked cell transition time. Positioning of the PodJ localization factor at the pole opposite the stalk (90 min) is necessary for subsequent positioning of PleC and the components of the pili secretion apparatus. As the swarmer cell changes into a stalked cell, the replisome assembles on the polar chromosome origin of replication and PleC is released from the pole and replaced by the DivJ histidine kinase. In the late predivisional cell (⬃220 min), CtrA proteolysis commences in the stalked compartment, the CckA histidine kinase is released from the cell poles, and DivK is released from the pole of the swarmer compartment.
1876
nascent stalked cell while reliably keeping the concentration high in the nascent swarmer cell. Failure to maintain a high concentration of CtrA in the swarmer compartment risks premature activation of chromosome replication and disruption of the swarmer cell’s distinctive behavior and morphology. Caulobacter has a robust mechanism for avoiding this problem, involving the compartmentalization of the predivisional cell. Partitioning the cell apparently enables activation of a CtrA degradation signal, perhaps involving a phosphorylation pathway, which accelerates the CtrA degradation rate (12, 15). CtrA proteolysis is only initiated in the stalked half of the predivisional cell after cytoplasmic compartmentalization has occurred, about 15 to 18 min before cell separation. This ensures that DNA replication begins only in the progeny stalked cell (12, 16) Another novel synchronization mechanism is a resettable switch that ties the timing of CtrA expression to the progress of chromosome replication. The hemimethylated state of newly replicated DNA containing the ctrA gene is used as a signal for ctrA promoter activation (15). Before the initiation of replication, the chromosome is fully methylated at sites recognized by the CcrM methyltransferase. As the replication fork passes through the ctrA gene, the promoter region goes from the fully methylated to the hemimethylated state. CtrA is transcribed from two promoters: the P1 promoter, which can only be activated when it is hemimethylated, and the P2 promoter, which is autoactivated when the CtrA produced from P1 accumulates to threshold levels and is activated by phosphorylation. This system ensures that two gene copies are present before beginning CtrA expression, thereby increasing the rate of CtrA production and reducing the stochastic variations in the rate of its production. The increasing CtrA level represses P1 and activates many genes, including those that express the CcrM methyltransferase (Fig. 1B). CcrM methylates a site at the P1 promoter. Finally, both CtrA and CcrM are degraded to reset the switch for the next cell cycle. When the ctrA gene is moved to a site on the chromosome near the termination of DNA replication, the cells lose size control, probably as a result of defective coordination of timing (17).
Subcellular Localization of Proteins and Regions of the Genome In the Caulobacter cell, and presumably in any cell undergoing asymmetric division, asymmetries in the intracellular environment, particularly differentially localized regulatory molecules, are essential for the implementation of spatially differentiated cell morphogenesis. The design principles underlying these spatially distributed circuits are poorly understood. Also, the role of the physical geometry of the cellular reaction chamber (the cell interior and the membrane) in which the circuitry operates has not been well characterized in bacterial cells. During B. subtilis sporulation, differing chemistry on the two sides of the septum separating the compartments (9) as cell
26 SEPTEMBER 2003 VOL 301 SCIENCE www.sciencemag.org
NETWORKS IN BIOLOGY of structures in the bacterial cell. Integration of these imaging technologies with global gene expression studies of wild-type and mutant cells will enable new advances in genomics by integrating the spatial dimension of regulation.
Implications for Modeling the Cell Success with subsystem models has inspired ambitions to develop “whole-cell” models to put the individual parts together to create so-called “in silico” models (27, 28). However, major, perhaps insurmountable, difficulties must be overcome before whole-cell models based on extensions of the current “low-level” modeling and simulation methodologies, which emphasize kinetics of coupled reaction systems, will be feasible. Problems include lack of quantitative data on molecular concentrations and kinetic parameters, as well as only piecemeal characterization of the cell’s regulatory circuitry. Although great progress is being made in identifying global regulatory pathways by using high-throughput assays and sophisticated algorithms, it is still a difficult and labor-intensive biochemical and genetic job to verify details of any pathway. This suggests that it will be a long time before the data to support bottom-up models of the whole cell will be feasible. On the other hand, the hierarchical, top-down regulatory architecture and the modular structure of the functional elements that implement the cell cycle suggest that it may be useful to consider a top-down modeling paradigm. This approach would focus on the network of global regulators and the orderly initiation and coordinated execution of modular functions. Details of the wiring diagrams of global regulatory networks controlling the cell cycle, responses to environmental signals, and stress responses will have to be identified. With these results, along with characterization of the top-level regulation of modular functions and intermodule signaling pathways, realistic modeling of overall cell behavior and responses to environmental events should be possible. With all the switches and checkpoint mechanisms controlling cell-cycle progress, the resulting behavior of the cell is very machinelike, with the cell cycle progressing in a ratchetlike manner as conditional gates controlling forward progression from one state to another are successively opened. This discontinuous progression of the cell through the cell cycle is not well suited to modeling with differential equation-based cellcycle models with their implicit assumptions of continuity. Alternative modeling paradigms may prove necessary, probably involving modeling the cell cycle as a hybrid control system, i.e., a system with both discrete and continuous dynamics (29). Hybrid control models reflect the combination of discrete and continuous control mechanisms found in the cell (29). Biological applications of hybrid system models approximate transcriptional regulation with piecewiselinear approximations of sigmoidal responses (29–31) and map modular biological functions into corresponding software objects (29). This
approach, although promising, is less developed than the differential equation-based modeling of systems of coupled chemical reactions, and, at least for the Caulobacter cell, the three-dimensional nature of regulation with polar-localized regulatory molecules and bifurcating cytoplasmic chemistry adds complications. At a minimum, this leads to complex boundary conditions for the models. If reaction-diffusion models have to be incorporated (for example, to properly treat the function of localized components of phospho-cascades), the mathematics will be even more complicated. These will be exciting challenges for the new breed of “systems biologists” to address in the next decade.
SPECIAL SECTION
division nears completion is critical to differential development of the mother cell and prespore. There is no evidence for such asymmetric chemistry at the division plane in Caulobacter. Rather, in Caulobacter differential polar chemistry seems to be central to establishing the distinctive identities of the daughter cells. There is a complex interaction between the regulatory mechanisms operative at transcription initiation sites, the dynamically changing portfolio of regulatory proteins, the subset of those proteins localized to the cell poles (18, 19), and the instantaneous physical deployment of the replicating chromosome (19, 20) within the cell. As a result of these spatially distributed interactions, the dynamic deployment of regulatory proteins throughout the cell has to be considered when analyzing how individual components of the cell-cycle control circuitry work. Figure 2 provides a schematic view of the Caulobacter cell’s geometry and spatial deployment of proteins over the course of a cell cycle. Multiple two-component signal-transduction proteins, localization proteins, replication proteins, membrane-associated secretion and export proteins, and pili and flagellar structural proteins are dynamically created and destroyed at the poles as the cell cycle progresses and the polar organelles mature. The Caulobacter property of having only one flagellum made at a precise time in the cell cycle and strictly located at the pole is one result of precise temporal and spatial regulatory controls. Although the regulatory mechanisms for construction of a flagellum are well characterized (5), the mechanism that establishes its polar location is unknown (21). Some proteins that control or affect polar positioning of structures have been identified, but mechanisms of action are not known. In the case of polar pili biogenesis, a localization factor, PodJ (which is present in both long, PodJL, and short, PodJS, forms), has been identified that functions to place both the PleC histidine kinase required for pilus assembly and the pilus secretory protein at the cell pole (22, 23). PleC, in turn, signals the accumulation of the pilin subunit that assembles into the polar pilus; if active PleC is mislocalized, pili are not assembled (24). Mutations in polar-localized regulatory proteins can cause aberrant cell morphologies, defective or missing polar structures, failures in progression of the cell cycle, or cell death. Thus, it now appears that the transcriptional circuitry that implements the temporal control of cell-cycle progression is an integrated function of the spatial deployment of structural and regulatory proteins, so that the changing transcriptome both affects, and is affected by, polar developmental progress. Solving the puzzle of the connection between spatial positioning of regulatory proteins, cellcycle progression, and asymmetry in cell division and polar organelle development is a current challenge. Recent advances in protein labeling for electron microscope imaging (25) and in biological imaging using soft x-ray microscopy (26) offer prospects for unprecedented visibility into the dynamic three-dimensional positioning
References and Notes
1. M. T. Laub, H. H. McAdams, T. Feldblyum, C. M. Fraser, L. Shapiro, Science 290, 2144 (2000). 2. M. T. Laub, S. L. Chen, L. Shapiro, H. H. McAdams, Proc. Natl. Acad. Sci. U.S.A. 99, 4632 (2002). 3. L. H. Hartwell, J. J. Hopfield, S. Leibler, A. W. Murray, Nature 402, C47 (1999). 4. U. Jenal, FEMS Microbiol. Rev. 24, 177 (2000). 5. P. Aldridge, K. T. Hughes, Curr. Opin. Microbiol. 5, 160 (2002). 6. H. H. McAdams, A. Arkin, Trends Genet. 15, 65 (1999). 7. I. Simon et al., Cell 106, 697 (2001). 8. R. Hengge-Aronis, Curr. Opin. Microbiol. 2, 148 (1999). 9. P. Levin, R. Losick, in Prokaryotic Development, Y. Brun, L. Shimkets, Eds. (American Society for Microbiology, Washington, DC, 1999), pp. 167–189. 10. T. F. Hansen, Biosystems 69, 83 (2003). 11. M. Fujita, R. Losick, Genes Dev. 17, 1166 (2003). 12. W. E. Judd, K. Ryan, W. E. Moerner, L. Shapiro, H. H. McAdams, Proc. Natl. Acad. Sci. U.S.A. 100, 8235 (2003). 13. N. Barkai, S. Leibler, Nature 403, 267 (2000). 14. U. Jenal, C. Stephens, Curr. Opin. Microbiol. 5, 558 (2002). 15. C. Jacobs, N. Ausmees, S. J. Cordwell, L. Shapiro, M. T. Laub, Mol. Microbiol. 47, 1279 (2003). 16. K. R. Ryan, L. Shapiro, Annu. Rev. Biochem. (2003). 17. A. Reisenauer, L. Shapiro, EMBO J. 21, 4969 (2002). 18. S. R. Lybarger, J. R. Maddock, J. Bacteriol. 183, 3261 (2001). 19. L. Shapiro, R. Losick, Cell 100, 89 (2000). 20. R. B. Jensen, L. Shapiro, Trends Cell Biol. 10, 483 (2000). 21. L. Shapiro, H. H. McAdams, R. Losick, Science 298, 1942 (2002). 22. P. Viollier, N. Sternheim, L. Shapiro, Proc. Natl. Acad. Sci. U.S.A. 99, 13831 (2002). 23. A. J. Hinz, D. E. Larson, C. S. Smith, Y. V. Brun, Mol. Microbiol. 47, 929 (2003). 24. P. Viollier, N. Sternheim, L. Shapiro, EMBO J. 21, 4420 (2002). 25. S. R. Adams et al., J. Am. Chem. Soc. 124, 6063 (2002). 26. W. Meyer-Ilse et al., J. Microsc. 201, 395 (2001). 27. J. J. Tyson, K. Chen, B. Novak, Nature Rev. Mol. Cell Biol. 2, 908 (2001). 28. M. Tomita, Trends Biotechnol. 19, 205 (2001). 29. R. Alur et al., Comput. Sci. Eng. 4 (no. 1), 20 (2002). 30. R. Ghosh, C. J. Tomlin, in Hybrid Systems: Computation and Control, Lecture Notes in Computer Science 2034, M. D. Di Benedetto, A. Sangiovanni-Vincentelli, Eds. (Springer-Verlag, Berlin, Heidelberg, 2001), pp. 232–246. 31. R. Ghosh, K. Amonlirdviman, C. J. Tomlin, paper presented at the 41st IEEE Conference on Decision and Control, Las Vegas, NV, 2002. 32. J. Wu, N. Ohta, J. L. Zhao, A. Newton, Proc. Natl. Acad. Sci. U.S.A. 96, 13068 (1999). 33. G. B. Hecht, T. Lane, N. Ohta, J. M. Sommer, A. Newton, EMBO J. 14, 3915 (1995). 34. K. C. Quon, B. Yang, I. J. Domian, L. Shapiro, G. T. Marczynski, Proc. Natl. Acad. Sci. U.S.A. 95, 120 (1998). 35. K. C. Keiler, L. Shapiro, J. Bacteriol. 183, 4860 (2001). 36. H.H.M. is supported by Defense Advanced Research Projects Agency grant MDA972-00-1-0032, Department of Energy grant DE-FG03-01ER63219, and Office of Naval Research grant N00014-02-1-0538. L.S. is supported by National Institutes of Health grants GM32506/5120 M2 and GM51426.
www.sciencemag.org SCIENCE VOL 301 26 SEPTEMBER 2003
1877
