1. Flux Balance Analysis Evangelos Simeonidis v.simeonidis@manchester.ac.uk Metabolic Engineering

2. Simulating living cells Living cells contain >300 gene products E.coli: 4000 gene products yeast: 6000 gene products ~1000 are enzymes Thousands of processes occur simultaneously   immense complexity How can Metabolic Engineering cope with this complexity?

3. Stoichiometry Stoichiometry, together with Mass Balances help us overcome much of this complexity

4. Linear Pathway This system has 3 degrees of freedom = number of rates What will happen to the rates after time 0? After a long time? The system is said to be in “steady state” SXPY v1v1 v2v2 v3v3

5. Flux Analysis The rates at steady state are commonly called “fluxes” Degrees of freedom = number of rates – flux constraints Because this is a linear pathway, there is ONE flux through the system Value of flux determined by kinetics; but the equality of the 3 fluxes is determined ONLY by stoichiometry How do we increase the yield (production) of P? SXPY v1v1 v2v2 v3v3

6. Flux Analysis Write the mass balances that are possible for this system. What changed from the previous example? Degrees of freedom ? How do we change the yield in the first example? How do we change it in this example? What is the difference from a Metabolic Engineering Perspective? SX P Y v1v1 v2v2 v3v3

7. Flux Analysis and Optimisation Knowing enough about the system is often difficult; assume for example that in the above “cell” we can only measure consumption of S and production of P What is the relationship between the two fluxes in ss? Is there a way to evaluate the internal fluxes? An assumption is needed; often this assumption is that the organism is operating OPTIMALLY e.g. assume the above organism optimal with respect to energetic efficiency SXPY ATP ADP + Pi

8. Topics covered Constraint-based approach to metabolic modelling Mathematics behind FBA: Optimisation Mathematical examples

9. Introduction A major goal of systems biology is to relate genome sequence to cell physiology This requires the identification of the components and their interactions in the system + mathematical modelling Small molecule metabolism is the best described molecular network in the cell and there are various computational tools to model its behaviour

10. Metabolic network reconstructions Network reconstruction  delineation of the chemical and physical interactions between the components

11. Metabolic network reconstructions Automated metabolic reconstructions for > 500 organisms based on genome sequence data (e.g. KEGG database) Automated reconstructions are usually not suitable for modelling Manual assembly gives higher quality networks and is based on genomic + biochemical + physiological data

12. Incorporate information on: reaction reversibility cofactor usage transport reactions cellular compartments (e.g. mitochondrion) biomass composition Only available for well studied microbes (e.g. yeast, E. coli and ~10 other bacteria) High quality manual reconstructions

13. Example: Escherichia coli metabolic reconstruction * the best characterized network 931 reactions 625 different metabolites But: 67 are dead end! High quality manual reconstructions * Reed et al. (2003) Genome Biol 4: R54

14. How to analyse such a complex network? ExPASy – Metabolic Pathways

15. Problem with kinetic modelling A lot of data is required to parameterize large-scale models, experimentally intractable at present. The largest kinetic metabolic model available:  Human red blood cell (35 enzymes)

16. Two modelling approaches Kinetic rate equation for each reaction + parameters Simulation of system behaviour Mechanistic (kinetic) Constraint-based (stoichiometric) Consider all possible behaviours of the system (large solution space) Imposing constraints (physicochemical laws, biological constraints) Smaller allowable solution space

17. Types of constraints Physico-chemical constraints  mass, charge and energy conservation, laws of thermodynamics Biological constraints:  external environment, regulatory constraints

18. Stoichiometric modelling of metabolism Metabolic networkThermodynamic constraints (irreversibility) Mass balance constraint in steady state Maximum enzyme capacity etc.

19. Problem: Addition of constraints reduces the allowable solution space, but usually not to a single point (underdetermined system). How to find a particular solution? We can look for a solution which optimises a particular network function (e.g. production of ATP or biomass)  FBA

20. Topics covered Constraint-based approach to metabolic modelling Mathematics behind FBA: Optimisation Mathematical examples

21. Optimisation and Mathematical Programming optimisation problem or mathematical programming problem: a formulation in which a function is minimised by systematically choosing the values of variables from within an allowed set Given a function f: A  R (e.g. min x 2 +1) Find an element x 0 in A such that f(x 0 ) ≤ f(x) for all x in A The domain A of f is called the search space, while the elements of A are called feasible solutions A is specified by a set of constraints (equalities or inequalities) function f is called an objective function A feasible solution that minimizes the objective function is called an optimal solution

22. Subfields Linear programming studies the case in which the objective function f is linear and the constraints linear equalities and inequalities Integer linear programming studies linear programs in which some or all variables take on integer values Nonlinear programming studies the general case in which the objective function or the constraints or both are nonlinear

23. Techniques for solving mathematical programming problems There exist robust, fast numerical techniques for optimising mathematical programming problems –Gradient descent (steepest descent) –Nelder-Mead method –Simplex method –Ellipsoid method –Newton's method –Quasi-Newton methods –Interior point methods –Conjugate gradient method

24. Alternatives for optimisation Mathematical programming and its techniques for solving of optimisation problems are powerful tools, but are not the only solutions available. Other approaches (that usually apply numerical analysis approximations) are: –Hill climbing –Simulated annealing –Quantum annealing –Tabu search –Beam search –Genetic algorithms –Ant colony optimization –Evolution strategy –Stochastic tunneling –Particle swarm optimization –Differential evolution

25. Linear Programming (LP) LP model  objective function + linear constraints extensively used optimisation technique allocation of limited resources to competing activities in the optimal way examples of application: graphs, network flows, plant management, economics, business management most prominent method for solving: simplex method Prominent solver: CPLEX minimise c 1. x 1 + c 2. x 2 + … +c n. x n subject to: equality constraints a i1. x 1 + a i2. x 2 + … +a in. x n = a i0 inequality constraints b i1. x 1 + b i2. x 2 + … +b in. x n  b i0 or in matrix form: min c T. x subject to: A. x = a B. x  b

26. Mixed Integer Programming If variables are required to be integer, then the problem is an integer programming (IP) or mixed integer programming (MIP) problem In contrast to linear programming, which can be solved efficiently in the worst case, integer programming problems are in the worst case undecidable, and in many practical situations NP-hard MIP problems are solved using advanced algorithms such as branch and bound or branch and cut LP and MILP solvers are in widespread use for optimization of various problems in industry, such as optimization of flow in transportation networks –CPLEX –MINTO –AIMMS –SYMPHONY –Xpress-MP –GNU Linear Programming Kit –Qoca –Cassowary constraint solver

27. Optimisation in FBA Linear Programming may be used to study the stoichiometric constraints on metabolic networks optimisation is used to predict metabolic flux distributions at steady state based on the assumption of maximised growth performance along evolution only stoichiometric data and cellular composition required valuable for identifying flux distribution boundaries for the metabolic function of cellular systems

28. Application FBA involves carrying out a steady state analysis, using the stoichiometric matrix (S) for the system in question The system is assumed to be optimised with respect to objectives such as maximisation of biomass production or minimisation of nutrient utilisation At steady state: The required flux distribution is the null space of S. Since the number of fluxes typically exceeds the number of metabolites, the system is under-determined and may be solved by selecting an optimisation criterion, following which, the system translates into an LP problem

29. Mathematical Model max Σ c j. v j Σ S ij. v j = 0  i L j ≤ v j ≤ U j  j reversible 0 ≤ v j ≤ U j  j irreversible S ij : stoichiometric matrix v j : reaction fluxes (mmol / gDW hr) c j : weight j j i: metabolites j: reactions s.t. L j, U j : bounds

30. Small example for S matrix construction A  B A + B  C B + C  2 A

31. Exercise 1 A  B A + E  2 C B + C  D + E 2 E + C  2 A + B

32. Exercise 2

33. Topics covered Constraint-based approach to metabolic modelling Mathematics behind FBA: optimisation Mathematical examples

34. Model Network Simplified model of E. coli metabolic network Varma and Palsson, J.Theor.Biol. (1993) 165, 477 Reactions: 35 –17 reversible –18 irreversible Metabolites: 30

35. Glc Mal G6PR5P E4P F6P T3P 3PG PEP Pyr AcCoA LacEthAc CitICit aKG SuccCoA SuccFum Glx OA NADPH ATPH exp QH 2 FADHNADH

36. ATP18.67 NADH11.57 NADPH11.00 3PG2.00 PEP2.00 Pyr2.00 OA1.50 G6P0.91 F6P0.91 R5P1.08 E4P1.33 T3P1.74 AcCoA2.00 aKG1.00 SuccCoA1.00 Single metabolite production maximisation Maximum stoichiometric yields (mmol/mmol Glc)

37. Glc Mal G6PR5P E4P F6P T3P 3PG PEP Pyr AcCoA LacEthAc CitICit aKG SuccCoA SuccFum Glx OA NADPH ATPH exp QH 2 FADHNADH

38. Glc Mal G6P F6P T3P 3PG PEP Pyr AcCoA CitICit aKG SuccCoA SuccFum OA NADPH ATPH exp QH 2 FADHNADH Flux distribution map for maximal ATP production

39. Glc Mal G6PR5P E4P F6P T3P 3PG PEP Pyr AcCoA LacEthAc CitICit aKG SuccCoA SuccFum Glx OA NADPH ATPH exp QH 2 FADHNADH

40. Glc Mal G6P F6P T3P 3PG PEP Pyr OA NADPH ATPH exp QH 2 NADH Flux distribution map for maximal PEP production

41. ATP41.2570 NADH-3.5470 NADPH18.2250 G6P0.2050 F6P0.0709 R5P0.8977 E4P0.3610 T3P0.1290 3PG1.4960 PEP0.5191 Pyr2.8328 AcCoA3.7478 OA1.7867 aKgG1.0789 Biomass production maximisation Metabolic Demands for 1g of biomass yield (mmol) Maximum biomass yield: 0.589 g DW / g Glc

42. Glc Mal G6PR5P E4P F6P T3P 3PG PEP Pyr AcCoA LacEthAc CitICit aKG SuccCoA SuccFum Glx OA NADPH ATPH exp QH 2 FADHNADH

43. Glc Mal G6PR5P E4P F6P T3P 3PG PEP Pyr AcCoA CitICit aKG SuccFum Glx OA ATPH exp QH 2 FADHNADH Optimal flux distribution for aerobic growth on glucose

44. FBA Summary Simple, no kinetic information needed Can be applied to large networks In accordance with experimental results Can be used for defining wider limits of metabolic behaviour

45. Further reading Bernhard O. Palsson: Systems Biology: Properties of reconstructed networks, Cambridge University Press, 2006 Metabolic network reconstruction: Nature Reviews Genetics (2006) 7:130-141. Metabolic modelling approaches: Journal of Biotechnology (2002) 94: 37-63. Biotechnology and Bioengineering (2003) 84: 763-772 Flux Balance Analysis: Current Opinion in Biotechnology (2003) 14: 491-496. Nature Reviews Microbiology (2004) 2: 886-897 Mathematical programming: Paul Williams: Model Building in Mathematical Programming