Metabolic Engineering Flux Balance Analysis Evangelos Simeonidis v

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

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?

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

Linear Pathway S v 1 X v 2 Y v 3 P • 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”

Flux Analysis S v 1 X v 2 Y v 3 P • 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?

S v 1 Flux Analysis v 2 X Y v 3 P • 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?

Flux Analysis and Optimisation ATP S X ADP + Pi Y P • 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

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

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

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

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

High quality manual reconstructions • 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 Example: Escherichia coli metabolic reconstruction* • the best characterized network • 931 reactions • 625 different metabolites But: 67 are dead end! * Reed et al. (2003) Genome Biol 4: R 54

How to analyse such a complex network? Ex. PASy – Metabolic Pathways

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)

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

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

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

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

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

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

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

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

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

Linear Programming (LP) • LP model objective function + linear constraints minimise c 1. x 1+ c 2. x 2+ … +cn. xn • extensively used optimisation technique subject to: • allocation of limited resources to competing activities in the optimal way • examples of application: graphs, network flows, plant management, economics, business management equality constraints ai 1. x 1+ ai 2. x 2+ … +ain. xn = ai 0 inequality constraints bi 1. x 1+ bi 2. x 2+ … +bin. xn bi 0 or in matrix form: • most prominent method for solving: simplex method • Prominent solver: CPLEX min c. T. x subject to: A. x = a B. x b

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

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

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

Mathematical Model max s. t. Σ cj. vj j Σ Sij. vj = 0 j L j ≤ v j ≤ Uj 0 ≤ v j ≤ Uj i jreversible jirreversible i: metabolites j: reactions Sij: stoichiometric matrix vj: reaction fluxes (mmol / g. DW hr) cj: weight Lj, Uj: bounds

Small example for S matrix construction A B A+B C B+C 2 A

Exercise 1 A B A+E 2 C B+C D+E 2 E+C 2 A+B

Exercise 2

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

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

Glc G 6 P ATP NADH E 4 P Fum Succ T 3 P Succ. Co. A Mal 3 PG Glx PEP OA a. KG Cit Pyr Lac QH 2 R 5 P NADPH F 6 P Hexp Ac. Co. A Eth Ac ICit FADH

Single metabolite production maximisation Maximum stoichiometric yields (mmol/mmol Glc) ATP NADH NADPH 3 PG PEP Pyr OA G 6 P 18. 67 11. 57 11. 00 2. 00 1. 50 0. 91 F 6 P R 5 P E 4 P T 3 P Ac. Co. A a. KG Succ. Co. A 0. 91 1. 08 1. 33 1. 74 2. 00 1. 00

Glc G 6 P ATP NADH E 4 P Fum Succ T 3 P Succ. Co. A Mal 3 PG Glx PEP OA a. KG Cit Pyr Lac QH 2 R 5 P NADPH F 6 P Hexp Ac. Co. A Eth Ac ICit FADH

Flux distribution map for maximal ATP production Glc ATP Hexp QH 2 G 6 P NADPH NADH F 6 P Fum Succ T 3 P Mal Succ. Co. A OA a. KG 3 PG PEP Cit Pyr Ac. Co. A ICit FADH

Glc G 6 P ATP NADH E 4 P Fum Succ T 3 P Succ. Co. A Mal 3 PG Glx PEP OA a. KG Cit Pyr Lac QH 2 R 5 P NADPH F 6 P Hexp Ac. Co. A Eth Ac ICit FADH

Flux distribution map for maximal PEP production Glc ATP Hexp G 6 P NADPH F 6 P T 3 P Mal 3 PG PEP Pyr OA NADH QH 2

Biomass production maximisation Metabolic Demands for 1 g of biomass yield (mmol) ATP 41. 2570 T 3 P 0. 1290 NADH -3. 5470 3 PG 1. 4960 NADPH 18. 2250 PEP 0. 5191 G 6 P 0. 2050 Pyr 2. 8328 F 6 P 0. 0709 Ac. Co. A 3. 7478 R 5 P 0. 8977 OA 1. 7867 E 4 P 0. 3610 a. Kg. G 1. 0789 Maximum biomass yield: 0. 589 g DW / g Glc

Glc G 6 P ATP NADH E 4 P Fum Succ T 3 P Succ. Co. A Mal 3 PG Glx PEP OA a. KG Cit Pyr Lac QH 2 R 5 P NADPH F 6 P Hexp Ac. Co. A Eth Ac ICit FADH

Optimal flux distribution for aerobic growth on glucose Glc G 6 P ATP Hexp R 5 P NADH F 6 P E 4 P Fum Succ T 3 P Mal 3 PG Glx PEP OA a. KG Cit Pyr QH 2 Ac. Co. A ICit FADH

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

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