Design Principles in Synthetic Biology

Transcription

1 Design Principles in Synthetic Biology Chris Myers 1, Nathan Barker 2, Hiroyuki Kuwahara 3, Curtis Madsen 1, Nam Nguyen 1, Michael Samoilov 4, and Adam Arkin 4 1 University of Utah 2 Southern Utah University 3 Microsoft Research, Trento, Italy 4 University of California, Berkeley Design Principles in Biological Systems April 24, 2008

2 Synthetic Biology Increasing number of labs are designing more ambitious and mission critical synthetic biology projects. These projects construct synthetic genetic circuits from DNA. These synthetic genetic circuits can potentially result in: A better understanding of how microorganisms function by examining differences in vivo compared to in silico (Sprinzak/Elowitz). More efficient pathways for the production of antimalarial drugs (Dae et al.). Bacteria that can metabolize toxic chemicals (Brazil et al.). Bacteria that can hunt and kill tumors (Anderson et al.).

3 Genetic Design Automation (GDA) Electronic Design Automation (EDA) tools have facilitated the design of ever more complex integrated circuits each year. Crucial to the success of synthetic biology is an improvement in methods and tools for Genetic Design Automation (GDA). Existing GDA tools require biologists to design at the molecular level. Roughly equivalent to designing electronic circuits at the layout level. Analysis of genetic circuits is also performed at this very low level. A GDA tool that supports higher levels of abstraction is essential.

4 Overview This talk describes our research to develop such a GDA tool. This tool has helped us examine design principles for synthetic biology. As a case study, will describe the design of a genetic Muller C-element.

5 Current State of GDA Tools MIT has created a registry of standard biological parts used to design synthetic genetic circuits ( Methods and tools are needed to assist in the design and analysis of synthetic genetic circuits using these parts. BioJADE provides a schematic capture interface to the MIT parts registry. Systems Biology Markup Language (SBML) has been proposed as a standard representation for the simulation of biological systems. Many simulation tools have been developed that accept models in the SBML format (BioPathwise, BioSPICE, CellDesigner, SimBiology, etc.).

6 Systems Biology Markup Language (SBML) SBML models biological systems at the molecular level. A typical SBML model is composed of a number of chemical species (i.e., proteins, genes, etc.) and reactions that transform these species. This is a very low level representation which is roughly equivalent to the layout level for electronic circuits. Designing and simulating genetic circuits at this level of detail is extremely tedious and time-consuming. Therefore, there is a need for higher-level abstractions for modeling, analysis, and design of genetic circuits.

7 BioSim Genetic Circuit Perform Experiments Insert into Host Plasmid Construct Plasmid Experimental Biological DNA Data Knowledge Sequence Learn Model GCM Synthesis Simulation Data Abstraction/ Simulation SBML Model

8 BioSim: Analysis Genetic Circuit Perform Experiments Insert into Host Plasmid Construct Plasmid Experimental Biological DNA Data Knowledge Sequence Learn Model GCM Synthesis Simulation Data Abstraction/ Simulation SBML Model

9 BioSim: Design Genetic Circuit Perform Experiments Insert into Host Plasmid Construct Plasmid Experimental Biological DNA Data Knowledge Sequence Learn Model GCM Synthesis Simulation Data Abstraction/ Simulation SBML Model

10 BioSim: Design Genetic Circuit Perform Experiments Insert into Host Plasmid Construct Plasmid Experimental Biological DNA Data Knowledge Sequence Learn Model GCM Synthesis Simulation Data Abstraction/ Simulation SBML Model

11 BioSim: Genetic Circuit Model Genetic Circuit Perform Experiments Insert into Host Plasmid Construct Plasmid Experimental Biological DNA Data Knowledge Sequence Learn Model GCM Synthesis Simulation Data Abstraction/ Simulation SBML Model

12 Phage λ Virus

13 Phage λ Decision Circuit

14 Phage λ Decision Circuit

15 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

16 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

17 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

18 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription RNAP RNAP RNAP RNAP cii Promoters Genes

19 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription RNAP RNAP RNAP cii Promoters Genes

20 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription RNAP RNAP RNAP cii Promoters Genes

21 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

22 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

23 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

24 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

25 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

26 Genetic Circuits CI Dimer DNA Dimerization ci CI Protein Repression Activation Pre O E O R Operator Sites CII Protein Transcription Pr RNAP RNAP RNAP cii Degradation Translation mrna Promoters Genes

27 Genetic Circuits CI Dimer DNA Dimerization ci CI Protein Repression Activation Pre O E O R Operator Sites CII Protein Transcription Pr RNAP RNAP RNAP cii Degradation Translation mrna Promoters Genes

28 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

29 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

30 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

31 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

32 Genetic Circuits CI Dimer DNA Dimerization CI Protein Repression Activation Pre O E O R ci Operator Sites CII Protein Degradation Translation mrna Transcription Pr RNAP RNAP RNAP cii Promoters Genes

33 Logical Representation CI CII

34 Graphical Representation CI Pre Pr CII

35 Genetic Circuit Model (GCM) Provides a higher level of abstraction than SBML. Includes only important species and their influences upon each other. A GCM is a tuple S,P,G,I,S d where: S is a finite set of species; P is a finite set of promoters; G : P 2 S maps promoters to sets of species; I S P {a,r} is a finite set of influences; S d S is a set of species that influence as dimers.

36 GCM Graphical Representation A bipartite graph with species and promoters as the two types of nodes. Species are connected to promoters using influences I, and promoters are connected to species using function G. To simplify presentation, graphs shown using only species as nodes, edges are inferred using I and G, and edges are labeled with the promoter that links the species.

37 Influences on the Same Promoter A B A B C P1 P1 C P1 c

38 Influences on the Same Promoter A B A B C P1 P1 C P1 c A B C

39 Influences on Different Promoters A C A B P1 C P2 B P1 c C P2 c

40 Influences on Different Promoters A C A B P1 C P2 B P1 c C P2 c A B C

41 GCM Parameters Parameter Sym Structure Value Units Initial species count n s species 0 molecule Dimerization equilibrium K d species.05 Degradation rate k d species molecule 1 sec Initial promoter count n g promoter 2 molecule Stoichiometry of production n p promoter 10 molecule Degree of cooperativity n c promoter 2 molecule RNAP binding equilibrium K o promoter.033 Open complex production rate k o promoter.05 Basal production rate k b promoter.0001 Activated production rate k a promoter.25 Repression binding equilibrium K r influence.5 Activation binding equilibrium K a influence molecule 1 sec 1 sec 1 sec 1 molecule nc 1 molecule (nc+1)

42 GCM versus SBML Representation CI Pre Pr CII

43 SBML Example

44 SBML Example

45 SBML Example

46 SBML Example

47 SBML Example

48 SBML Example

49 SBML Example

50 SBML Example

51 Synthesizing SBML from a GCM Representation Create degradation reactions Create open complex formation reactions Create dimerization reactions Create repression reactions Create activation reactions

52 Degradation Reactions

53 Open Complex Formation Reactions

54 Dimerization Reactions

55 Repression Reactions

56 Activation Reactions

57 Complete SBML Model

58 Classical Chemical Kinetics Uses ordinary differential equations (ODE) to represent the system to be analyzed, and it assumes: A system is well-stirred. Number of molecules in a cell is high. Concentrations can be viewed as continuous variables. Reactions occur continuously and deterministically. Genetic circuits involve small molecule counts. Gene expression can have substantial fluctuations. ODEs do not capture non-deterministic behavior.

59 Stochastic Chemical Kinetics To more accurately predict the temporal behavior of genetic circuits, stochastic chemical kinetics formalism can be used. Probabilistically predicts the dynamics of biochemical systems. Describes the time evolution of a system as a discrete-state jump Markov process governed by the chemical master equation (CME). Can simulate it using Gillespie s Stochastic Simulation Algorithm (SSA). It exactly tracks the quantities of each molecular species, and treats each reaction as a separate random event. Only practical for small systems with no major time-scale separations. Abstraction is essential for efficient analysis of any realistic system.

60 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

61 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

62 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

63 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

64 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

65 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

66 Automatic Abstraction Reaction Model Reaction-based Abstraction Abstracted Reaction Model State-based Abstraction SAC Model Markov Chain Analysis Stochastic Simulation Results Begins with a reaction-based model in SBML. Next, it automatically abstracts this model leveraging the quasi-steady state assumption, whenever possible. Finally, it encodes chemical species concentrations into Boolean (or n-ary) levels to produce a stochastic asynchronous circuit model. It can now be analyzed using Markov chain analysis.

67 Dimerization Reduction

68 Dimerization Reduction

69 Operator Site Reduction (PR)

70 Operator Site Reduction (PR)

71 Operator Site Reduction (PRE)

72 Operator Site Reduction (PRE)

73 Similar Reaction Combination

74 Modifier Constant Propagation

75 Final SBML Model 10 species and 10 reactions reduced to 2 species and 4 reactions

76 BioSim: Genetic Circuit Editor

77 BioSim: SBML Editor

78 BioSim: Simulator

79 BioSim: Parameter Editor

80 BioSim: Graph Editor

81 GCM Advantages Greatly increases the speed of model development and reduces the number of errors in the resulting models. Allows efficient exploration of the effects of parameter variation. Constrains SBML model such that it can be more easily abstracted resulting in substantial improvement in simulation time.

82 Genetic Muller C-Element A B C C A B C C 1 0 C 1 1 1

83 Toggle Switch C-Element (Genetic Circuit) A D X A B A B D E X F Y Z S R Q C P1 B P2 D P7 E d x E X e x P3 y F f F Z Y P8 f P4 z C Y Z P5 c y P6 z

84 Toggle Switch C-Element (GCM) A D X P1 B d E x X Y P2 D e x P3 y F E P7 f F Z P8 f P4 z C Y Z P5 c y P6 z

85 Toggle Switch C-Element (SBML)

86 Toggle Switch C-Element (Abstracted) Reduced from 34 species and 31 reactions to 9 species and 15 reactions.

87 Toggle Switch C-Element (Simulation) Simulation time improved from 312 seconds to 20 seconds.

88 Majority Gate C-Element (Genetic Circuit) A B X Y D E C Z A X Y D P1 x y P5 d B Z X D E C P2 z x P4 d P7 e P8 c D Y Z D P3 y z P6 d

89 Majority Gate C-Element (GCM) A X Y D P1 x y P5 d B Z X D E C P2 z x P4 d P7 e P8 c D Y Z D P3 y z P6 d

90 Majority Gate C-Element (Simulation)

91 Speed-Independent C-Element (Genetic Circuit) A B S1 S2 S4 S3 C A S4 X S1 S2 S3 P1 s4 x P4 s1 P5 s2 P6 s3 B S4 Y S1 S3 S2 Z P2 s4 y P7 s1 P8 s2 z S4 S3 Z C S2 S4 P3 s3 z P9 c P10 s4

92 Speed-Independent C-Element (GCM) A S4 X S1 S2 S3 P1 s4 x P4 s1 P5 s2 P6 s3 B S4 Y S1 S3 S2 Z P2 s4 y P7 s1 P8 s2 z S4 S3 Z C S2 S4 P3 s3 z P9 c P10 s4

93 Speed-Independent C-Element (Simulation)

94 Ordinary Differential Equation Analysis Use Law of Mass Action to derive an ODE model. Study behavior of our model at steady state. Analyze nullclines to characterize the gate.

95 ODE Analysis: Nullclines for Toggle C-Element Toggle, Inputs low 120 dy=0 dz= Y Z Stable

96 ODE Analysis: Nullclines for Toggle C-Element Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable Unstable Stable Z

97 ODE Analysis: Nullclines for Toggle C-Element 120 Stable Toggle, Inputs High dy=0 dz= Y Z

98 ODE Analysis: Nullclines for Toggle C-Element Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable Unstable Stable Z

99 Stochastic Simulation: State Change from Low to High Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable? 20 Unstable Stable Z

100 Stochastic Simulation: State Change from Low to High Low to High maj heat high maj light high tog heat high tog light high si heat high si light high Failure Rate Time (s)

101 Stochastic Simulation: State Change from High to Low Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable 40 20? Unstable Stable Z

102 Stochastic Simulation: State Change from High to Low High to Low maj heat low maj light low tog heat low tog light low si heat low si light low Failure Rate Time (s)

103 Effect of Gene Count Failure Rate Low to High maj heat high maj light high tog heat high tog light high si heat high si light high Number of Genes

104 Effect of Cooperativity Low to High maj heat high maj light high tog heat high tog light high si heat high si light high Failure Rate Cooperativity

105 Effect of Repression Strength Low to High maj heat high maj light high tog heat high tog light high si heat high si light high Failure Rate Repression

106 Effect of Decay Rates 1 Low to High Failure Rate maj heat high maj light high tog heat high tog light high si heat high si light high Decay Rate

107 Effect of Dual Rail Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable? 20 Unstable Stable Z

108 Effect of Dual Rail Low to High single tog heat high single tog light high dual tog heat high dual tog light high 0.02 Failure Rate Time (s)

109 Effect of Dual Rail Toggle, Inputs Mixed 120 dy=0 dz=0 100 Y Stable 40 20? Unstable Stable Z

110 Effect of Dual Rail High to Low single tog heat low single tog light low dual tog heat low dual tog light low Failure Rate Time (s)

111 Design Principles in Synthetic Biology Speed-independence does not necessarily imply better robustness. Higher gene counts improve production rates, higher equilibrium values, and more robust operation. Cooperativity of at least two is required to produce the necessary non-linearity for state-holding. Repressors should bind efficiently. Decay rates cannot be too high. Dual-rail outputs are essential.

112 Future Work: Modular Design More levels of hierarchy are needed in the GCM format. We plan to create structural constructs that allow us to connect GCM s for separate modules through species ports. Allow design at the logical and higher levels of abstraction.

113 Biologically Inspired Circuit Design Human inner ear performs the equivalent of one billion floating point operations per second and consumes only 14 µw while a game console with similar performance burns about 50 W (Sarpeshkar, 2006). We believe this difference is due to over designing components in order to achieve an extremely low probability of failure in every device. Future silicon and nano-devices will be much less reliable. For Moore s law to continue, future design methods should support the design of reliable systems using unreliable components. Biological systems constructed from very noisy and unreliable devices. GDA tools may be useful for future integrated circuit technologies. Biological systems tend to be more asynchronous and analog in nature, so future engineered circuits will likely need to be also.

114 Acknowledgments Nathan Barker Hiroyuki Kuwahara Nam Nguyen Curtis Madsen Michael Samoilov Adam Arkin This work is supported by the National Science Foundation under Grants No and CCF