Network Rigidity and Metabolic Engineering in ...

Network Rigidity and Metabolic Engineering in Metabolite Overproduction Gregory Stephanopoulos; Joseph J. Vallino Science, New Series, Vol. 252, No. 5013. (Jun. 21, 1991), pp. 1675-1681. Stable URL: http://links.jstor.org/sici?sici=0036-8075%2819910621%293%3A252%3A5013%3C1675%3ANRAMEI%3E2.0.CO%3B2-L Science is currently published by American Association for the Advancement of Science.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/aaas.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact [email protected].

http://www.jstor.org Mon Nov 12 16:50:34 2007

K. Walsh and D. E. Koshland, Jr., J. Biol. Chem. 259, 9646 (1984). D. G. Gadian, Nuclear Magnetic Resonance and Its Applications to Living Systems (Oxford Univ. Press, Oxford, 1982); E. J. Fernandez and D. S. Clark, Enzyme Microb. Technol. 9, 256 (1987). 73. J. V. Shanks and J. E. Bailey, Biotechnol. Bioeng. 32, 1138 (1988). 74. T. D. Brock, Ed., Thermophiles: General, Molecular and Applied Microbiology (Wiley-Interscience, New York, 1986). 75. H. Kacser and J. A. Burns, Symp. Soc. Exp. Biol. 27, 65 (1973);R. Heinrich, S. M. Rapoport, T. A. Rapoport, Prog. Biophys. Mol. Biol. 32, 1 (1977). 76. J. L. Galazu, and J. E. Bailey, Enzyme Microb. Technol. 12, 162 (1990);ibid. 13, 363 (1991). 77. C. Reder, J. Theor. Biol. 135, 175 (1988). 78. P. M. Schlosser and J. E. Bailey, Math. Biosci. 100, 87 (1990). 79. J. P. Delgado and J. Liao, Biotechnol. Prog. 7, 15 (1991).

80. M. M. Domach et al., ibid. 26, 203 (1984). 81. S. B. Lee and J. E. Bailey, Plasmid 11, 151 (1984);ibid., p. 166; Biotechnol. Bioeng. 26, 66 (1984);ibid., p. 1372; ibid., p. 1283. 82. J. E. Bailey, N. Da Silva, S. M. Peretti, J.-H. Seo, F. Srienc, A n n . N . Y . Acad. Sci. 469, 194 (1986). 83. N. Danhash, D. C. J. Gardner, S. G. Oliver, Biotechnology 9, 179 (1991). 84. K. Walsh, M. Schena, A. J. Flint, D. E. Koshland, Jr., Biochem. Soc. Symp. 54,183 11987). 85. M. ~dsinski,U . Rinas, J. E. Baile,y, unpublished data. 86. Supported by NSF and by the Catalysis and Biocatalysis Program of the Advanced Industrial Concepts Division of the U.S. Department of Energy. Advice and assistance in the preparation of this paper were provided by D. S. Clark, B. D. Ensley, T. Hashimoto, R. J. Kaufman, C. Khosla, R. A. Lazarus, A. J. Sinskey, K. N. Timmis, and colleagues at the California Institute of Technology.

Network Rigidity and Metabolic Engineering

in Metabolite Overproduction

In order to enhance the yield and productivity of metabolite production, researchers have focused almost exclusively on enzyme amplification or other modifications of the product pathway. However, overproduction of many metabolites requires significant redirection of flux distributions in the primary metabolism, which may not readily occur following product deregulation because metabolic pathways have evolved to exhibit control architectures that resist flux alterations at branch points. This problem can be addressed through the use of some general concepts of metabolic rigidity, which include a means for identifying and removing rigid branch points within an experimental framework.

LL ORGANISMS USE PRIMARY METABOLIC PATHWAYS TO

supply precursor metabolites and energy to anabolic pathways that synthesize cellular constituents that are necessary for growth and maintenance. In many industrial strains of microorganisms (as well as tissue and plant cultures), these anabolic pathways have been exploited for the overproduction of compounds (such as amino and nucleic acids, antibiotics, vitamins, enzymes, and proteins) that cannot be synthetically produced or for which it is not economical to do so. In general, a particular metabolite is overproduced by deregulating the pathway directly associated with the synthesis of that metabolite, or, more recently, by transforming a robust host organism (typically Escherichia coli) with the genes that encode for the synthesis of the desired product (1, 2). This approach, however, does not necessarily result in high product yields (defined as the moles of product formed per mole of substrate consumed) since carbon flux distributions at key branch points (nodes) in the primary metabolism [such as glycolysis, tricarboxylic The authors are in the Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139. *Present address: Scripps Institution of Oceanography, University of California at San Diego, La Jolla, CA 92093-0202. 21 JUNE 1991

acid (TCA) cycle, and pentose phosphate pathway] must often be radically redirected from the flux distributions that are normally associated with balanced growth. Such metabolic flux alterations are often directly opposed by mechanisms for controlling enzyme activity that have evolved to maintain flux distributions that are optimal for growth. We refer to this inherent resistance to flux alterations as metabolic or network rigidity and to the genetic modifications of specific nodes in the primary metabolism for the purpose of enhancing yield and productivity as metabolic engineering (3). Although genetic manipulations can now be readily performed, there are relatively few accounts of successhl metabolic flux alterations because of the complex, nonlinear nature of the metabolic control architectures. The nature and types of metabolic rigidity are reviewed in this article along with methods to identify and possibly circumvent such undesirable nodal controls. The overproduction of lysine by Corynebacterium glutamicum [and related strains (4)] is used as a vehicle to illustrate key points because of: (i) the lack of compartmentalization in bacteria; (ii) the need for significant flux alterations to optimize lysine biosynthesis; and (iii) the apparent marginal success of mutation-selection (5, 6 ) or genetic engineering ( 7 )techniques used to that end. The concepts, however, are of general value, and the methods are applicable to other metabolic products as well.

Basis of Metabolic Rigidity Although intracellular metabolite concentrations can fluctuate during growth, on average, the distributions of the major cellular groups (proteins, RNA, DNA, lipids, and so forth) remain relatively proportional to one another throughout balanced growth (8). In fact, metabolites and energy required to synthesize an E. coli cell have been calculated on the basis of its known composition ( 9 ) .In order to preserve this regularity in cellular composition, the primary metabolism has evolved coordination of pathway control, such that building-block metabolites, energy [such as adenosine triphosphate (ATP)], and biosynthetic reducing power [such as nicotinamide adenine dinucleotide phosphate (NADPH)] are synthesized in approximate stoichiometric ratios during balanced growth. AlARTICLES

1675

though cellular metabolism can support radically different flux distributions in response to different environmental stimuli (lo), such flux alterations are a consequence of the metabolic feedbacks that control the optimum synthesis of the major cellular groups required for the survival of the cell. Since primary metabolism is coordinated to synthesize a relatively uniform distribution of building block metabolites and energy, it is not well suited for the overproduction of any single metabolite. Consequently, if overproduction of a particular metabolite demands significant flux alterations in the primary metabolism compared to the flux distributions observed under balanced growth, then simple deregulation of product feedback inhibition frequently results in product yields that are significantly less than theoretical. An example of the effects that overproduction of a metabolite can elicit on metabolic flux distributions is shown in the flux diagrams in Fig. 1, which represent lysine overproduction by C. glutamicum ATCC 21253 under varying metabolic burdens. The flux distributions that support lysine production at typically observed yields (30 to 40% molar yield) (11) (Fig. 1B) do not deviate significantly from the flux distributions that support balanced growth (Fig. IA). However, the flux distributions required to support maximum lysine yield (Fig. 1C) represent a significant divergence from the flux distributions that occur during balanced growth or typical lysine production conditions. If primary metabolism could readily shift to match lysine synthesis demands, then significant yield improvements could be obtained by simply blocking pathways leading to by-products, such as the pyruvate dehydrogenase complex (PDC) that leads into the TCA cycle. However, such attempts have not resulted in significant yield improvements (5, 6 ) , which indicates that the metabolic network doks not respond readily to the demands of such perturbations. Since modification of the primary metabolism in order to redirect fluxes in support of maximum product yield is the objective of metabolic engineering, the sources of metabolic rigidity must be identified.

Principal Nodes It is often assumed that poor product yield resultsfrom an enzyme or enzymes in a reaction sequence that limit the throughput to the product. Although overall enzymatic activity along the product branch certainly governs product synthesis rate, product yield is ultimately a function of flux partitioning that occurs at intermediate branch points. The latter, also referred to as nodes, are points in a A

B 100

1

+=

8

+F6P

G6P

79

16

2

AcCoA

"1 5

100

RibuSP

*

1

Glucose 6

100

G6P -RibuSP

99

-

1

1 1

F6P

76

160

/

-Pyr AcCoA

a0

AcCoA

17

Co2

Lysines a

1

G6P -RibuSP

F6P

11

1

C

Glucose

Glucose

metabolic network where a reaction sequence bifurcates among two or more different pathways. Since the pathways that synthesize primary metabolites can typically support substantial fluxes, metabolic engineering efforts should be directed toward altering flux partitioning at selected metabolic branch points to enhance product yield. An important realization in the analysis of metabolic networks is that product yield typically depends on the flux partitioning that occurs at only a small subset of all possible nodes that comprise the metabolic network. Identification of these nodes, referred to as the principal nodes, is accomplished by searching for those nodes in the metabolic network where significant changes in flux partitioning occur as a function of product yield. Flux distributions required for this analysis can be determined from metabolite balance constraints with or without (as in Fig. 1 ) the use of experimental data (see below). The lysine example illustrates the point. Of -30 branch points in the lysine biosynthetic network, significant changes in flux partition occur at only three nodes, glucose-6-phosphate (G6P), phosphoenolpyruvate (PEP), and pyruvate (Pyr), as lysine yield increases from 35% (Fig. 1B) to 75% (Fig. 1C). Close inspection of the oxaloacetate (OaA) branch point reveals that it is a trivial principal node (12) since all of the OaA Synthesized from PEP condenses with Pyr to form lysine. The OaA synthesized through malate dehydrogenase is consumed by citrate synthase, so that net OaA production occurs only through the anaplerotic PEP carboxylase reaction (10). The fructose-6-phosphate (F6P) node can also be ruled out as a principal node. For a lysine yield of less than 60%, the F6P node is simply a condensation point; however, for yields greater than 60%, F6P becomes a principal node and G6P becomes a condensation point (Fig. 1C). This change is due to the pentose which must operate i n a cyclic manner to fullill phosphate NADPH requirements when lysine yield exceeds 60%. Since typically observed lysine yields are 30 to 40%, metabolic modifications need only target the G6P, PEP, and Pyr principal nodes. Hence metabolic engineering efforts need only focus on these three nodes and not on the entire network, which dramatically reduces problem complexity. The duality of the G6P and F6P nodes also illustrates the point that the location of the principal nodes depends on the product under investigation, the substrate utilized, and the byproducts that are typically associated with the product. The connectivity arrangement of the principal nodes in the network is also an important consideration. In general, network architectures fall into two basic classes: dependent and independent. If each principal node of a network, or subnetwork, contributes a stoichiometrically consumed component

J ( J65

Lysine

c02

,6kyJ

Lysine

CO,

Fig. 1. Flux distributions, normalized by glucose uptake in the metabolic network of C. glutamicum (illustrated in condensed form) during: (A) balanced growth (at a typically observed yield of 0.5 grams of biomass per gram glucose) without lysine overproduction; (6) lysine overproduction at 35% molar yield without growth; and (C) lysine overproduction at maximum molar yield of 75%. Fluxes at the indicated yields were determined solely from metabolic balance constraints (that is, experimental data were not required) (36, 37) for a network consistine of 34 reactions. including ass synthesis a lumped (9, 36, reaction 37, 39) to" account indicated forby biomthe

The . two fluxes between PEP double arrows ( j ) and Pyr account for reactions catalyzed by pyruvate kinase (orphosphotransferase PEP synthetase) (left) and (43) the g1ucose:PEP system

(right).A negative flux indicates that flow is in the opposite direction to that indicated by the arrow. Abbreviations: G6P, glucose-6-phosphate; Ribu5P, ribulose-5-phosphate; F6P, fructose-6-phosphate; PEP, phosphoenolpyruvate; Pyr, pyruvate; AcCoA, acetyl coenzyme A; and OaA, oxaloacetate. 1676

SCIENCE, VOL. 252

4: & A

S

B

S

Fig. 2. Examples of (A) dependent and ( 6 )independent two-node networks, where S is the substrate, B, and B, are by-products, and P is the desired product.

B,

P

8,

B2

P

of the final product, then the network or subnetwork is considered dependent (Fig. 2A), such as in the lysine network (NADPH, OaA, and Pyr are all lysine precursors). In such networks, the flux in all condensing branches must be stoichiometricdy balanced to prevent accumulation or excretion of intracellular components (for example, flux partitioning at nodes 1and 2 of Fig. 2A must be equal). Under this restriction, flux partitioning at all principal nodes that comprise a dependent network must be coordinated, and all principal nodes must be considered of equal importance, regardless of their "remoteness" from the product of interest. Conversely, if the metabolites produced from the principal nodes do not condense (Fig. 2B), then the network is considered independent, and product yield may be improved by altering the flux partitioning that occurs at a single principal node. In the simple two-node network (Fig. 2B), the yield of the desired product P can be enhanced if the flux partitioning at either node 1 or 2 is altered to favor the synthesis of P, while in the dependent network, flux partitioning at both nodes must be altered. Although network architecture affects metabolic response to enzyme modifications, in both dependent and independent networks the degree of metabolic rigidity ultimately resides with the ease by which flux partitioning at the principal nodes may be altered to obtain a flux distribution that is optimal for product formation. Since the type of nodal rigidity dictates the types of metabolic modifications required to render the rigid node flexible, factors that influence nodal flux partitioning must be examined.

Branch Point Classifications The metabolic control structures used by different organisms to regulate pathway fluxes vary markedly (13). It is, nevertheless, possible to classify these control structures into three general categories on the basis of their branch point rigidity. In the discussion that follows, branch kinetics refers to the lumped response of all of the enzymes that constitute the branch, and the branch split ratio refers to the carbon flux channeled through that branch normalized by the flux into the node. Nodes where flux partitioning into each branch readily changes to meet metabolic demands are referred to as flexible nodes. In a flexible node, enzymes participating in each branch show similar affinity for the node metabolite, and reaction velocities of each branch are of similar magnitude. The flux through each branch is controlled by feedback inhibition by the corresponding terminal metabolite (Fig. 3A). As a result, the branch split ratio can change from zero to one to meet the demands placed on the terminal metabolites. For example, if the feedback inhibition by metabolite P (Fig. 3A) is removed, or if P (but not by-product B) is a precursor to a desired product, the split-ratio of the P branch would approach unity as the product yield is maximized. Many terminal branch points in amino acid biosynthesis exhibit such flexibility (14). For example, the flux into the aspartate semialdehyde branch point can be completely redirected toward lysine synthesis by deregulating the concerted feedback inhibition of aspartokinase by lysine plus threonine (15). Flexible nodes are the most amenable to alterations of branch flux distributions and seldom require modification. Perhaps 21 JUNE 1991

because of this, many metabolic modifications intended to enhance product yield assume the principal nodes (hence the network) to be flexible, something that cannot be expected to be generally true. A node is said to be weakly rigid if the flux partitioning at the node is dominated by the kinetics of one of its branches. This can be the result of high enzymatic activity or high enzymatic &nity for the node metabolite and lack of feedback inhibition in the dominant branch (Fig. 3B). Even if the subordinate (product) branch is deregulated (or a high demand is placed on P), a significant fraction of the flux still enters the dominant branch and limits product yield. Product yield may still be enhanced, however, by attenuating the activity of the dominant branch. This may have to be accompanied by activity amplification of the subordinate branch or end product deregulation in order to prevent accumulation of the branch point intermediate (N). Nevertheless, weakly rigid nodes are typically amenable to mutation-selection techniques. Weakly rigid nodes are often f o ~ ind catabolic pathways leading to CO, formation. One example is the isocitrate branch point in E. coli between the glyoxylate shunt and the TCA cycle. When glucose is added to an E. coli culture grown on acetate, the flux supported by isocitrate lyase (ICL), the first enzyme of the glyoxylate shunt, drops to zero even though the glyoxylate shunt is completely active (16). Under these conditions, TCA cycle isocitrate dehydrogenase (ICDH) exhibits complete dominance over ICL since the Michaelis constant K, of ICDH for isocitrate is 1/75 that of ICL, and both enzymes exhibit similar activity (16-18). However, if the activity of ICDH is attenuated by 60% or more, a significant fraction of the isocitrate synthesized is diverted into the glyoxylate shunt (16-19). In E. coli, ICDH activity is attenuated by phosphorylation, so that growth on acetate can be sustained. A node is said to be strongly rigid if the split ratio of one or more of its branches is tightly controlled. This is commonly achieved by a combination of feedback control and enzyme trans-activation by a metabolite in an opposite branch (Fig. 3C). In this particular case, besides inhibiting their own synthesis, metabolites P and B also act as positive effectors of the opposite branch, a property exhibited by many dosteric enzymes (20). This architecture effectively stabilizes the flux partitioning at the node, provided the steady-state concentrations of P and B are sufficiently high to inhibit their own synthesis in the absence of their corresponding branch activators, B and P, respectively (21). For example, if the flux through the B branch is attenuated as means of increasing the split ratio of the P branch, the synthesis rate of P would also be attenuated because of the loss of activation by B and dominance of the remaining end product inhibition by P, leaving the split ratios of each branch relatively unaffected. Interestingly, simulations indicate that other nodal modifications, such as increasing the demand for P (by either atKnity or activity modulation), removing the feedback inhibition of P, or blocking the demand for B, have little effect on the split ratio of the P branch as well. Because of the hypothetical nature of this example, the details of these modifications are not discussed. However, several examples of rigid nodes have been previously documented. In Bacillus subtilis, histidine activates anthranilate synthetase, an enzyme required for tryptophan synthesis (22). This activation by histidine allows B. subtilis to grow in the presence of-5-methyltryptophan, an anti-metabolite of tryptophan. Furthermore, the addition of histidine increases tryptophan excretion by a factor of 4 in mutant strains that express anthranilate synthetase constitutively. I t is hypothesized that this trans-activation, which the investigators referred to as "metabolic interlock," has evolved to control the level of 5-phosphoribosyl-1-pyrophosphate, a metabolite that is common to the synthesis of both amino acids. This type of pathway interaction has also been referred to as compensatory control or activation (14, 23, 24). Remote effectors also influence phenylalanine biosynARTICLES

1677

B

A

1

1

N

N

;, 'P

1

C

s

s

S

'

B'

, 'P

1

1

1 \

I

--

N

B

1

Fia. 3. Control architectures that would render the illustrated branch ~ o i n t (A? flexible, (6) weakly rigid, or (C) strongly rigid to modifications i; flux partitioning (see text for details). Dashed lines indicate negative (11) or positive ( L)feedback from the corresponding metabolite.

thesis by modulating prephenate dehydratase in both coryneform bacteria and B. subtilis (25). In regard to primary metabolism, significant coupling interactions have been identified between anaplerotic pathways-PEP and Pyr carboxylases-and the catabolic pathways of glycolysis and the TCA cycle (24, 26). In general, flux partitioning at strongly rigid nodes cannot be easily modified since the kinetics of the enzymes associated with the branch point may require specific modifications, such as alterations in effector and substrateaffinities.

Network Responses The presence of a rigid node (either strong or weak) in a metabolic network can have profound implications with regard to the outcome of perturbations intended to enhance product yield, Consider again the simple independent network (Fig. 2B), where increasing the yield of P is the desired goal. It is further assumed that node 1 is flexible, node 2 is strongly rigid, and that B, does not inhibit its own synthesis. If the B, branch is blocked, the yield of P would be increased; however, if the B, branch is blocked, the yield of P would actually decrease because of the rigidity of node 2. This occurs because attenuation of the flux through the B2 branch causes a decrease in the flux to the product branch (due to node 2 rigidity). Since node 1is flexible, the flux to node 2 would be diverted to B,, thereby decreasing the yield of P. If the architecture is reversed, s; that node 1is rigid but node 2 is flexible, then blocking B2 synthesis would improve the yield of P. However, if B, synthesis is blocked, the yield of P would not be affected while the rate of P synthesis woGd be severely attenuated. Nodal rigidity in dependent networks presents even greater problems in metabolic engineering. For example, if either node 1or node 2 in the simple dependent network (Fig. 2A) is rigid, then attenuation of B, synthesis would not affect the yield of P, but would cause its synthesis rate to be diminished by the same extent as B, synthesis. Hence, modifications intended to improve flux distributions in dependent networks often result in overall metabolic collapse (or excretion of intermediate metabolites). In a dependent network, flux distribution response is independent of the location of the rigidity since alterations in flux partitioning at all principal nodes must b e coordinated if product yield is to be enhanced. Metabolic networks may not always exhibit the binary response to implied by the above classificationsbecause flux diverse control architectures and enzyme kinetics that are associated with branch points. Thus, attenuation of one branch of a stronglv - * rigid node miy only result in partial attenuation of the competing branch, so that some improvement in product yield may be attained in a network a rigid Also, attenuating nondesirable pathways may cause excretion of intermediate metabolites instead of metabolic flux collapse or flux redirection. Finally, 1678

as with the location of the principal nodes, the status of a branch point (that is, whether flexible or rigid) may depend on the product under investigation. For instance,-if a node consists of the two branches, A -+ B + C and A -+ D, where B and D (but not C) inhibit their own synthesis, then the node would appear to be flexible to products that consume B or D, but weakly rigid to products that require only C. In any event, the extent of rigidity of principal nodes in a network must first be assessed before any metabolic modifications can be suggested.

Assessment of Principal Node Rigidity Once the principal nodes for the product of interest have been identified in the metabolic network, their degree of rigidity must be assessed. Several techniques are available to analyze metabolic networks, all with their own advantages and disadvantages but none completely comprehensive. If detailed kinetic expressions are available for all of the enzymes included in the network representation of the organism's biochemistry, then a mathematical model of the metabolism can be constructed, such as those developed for glycolysis and respiration (27, 28) and for erythrocytes (29). With the aid of such models. simulations of the metabolic network, or subnetwork, are conducted to assess the rigidity of the principal nodes by comparing - the steady-state flux distributions for the nominal case to those generated from the same network following a perturbation intended to enhance product yield (that is, alteration of a principal node's split-ratio), such as eGyme'deletion or deregulation. s&gestions for metabolic alterations are then based on the type of principal nodes identified (Fig. 3). Although these techniques are governed by trial and error, they are quite informative since the models incorporate the full nonlinear nature of metabolic networks. The obvious disadvantage of simulation-based assessment techniques is that detailed kinetic information is limited and, when available, its applicability to in vivo situations not readily established. Other related approaches, such as metabolic control theory (MCT) (30) or biochemical systems theory (BST) (31), have received extensive attention (2, 32); however, their value in directing metabolic engineering efforts remains questionable (33). Both techniques rely bn approximating the actual metabolic system (assuming enzyme kinetics are available) around a particular operating point, so that the sensitivity of steady-state fluxes can be in enzyme velocities. The evaGa;ed with respect to A

Glucose

1

(1.37)

loo

G ~ P-

Q6P

Gluconate

7yy

F6p + trehalose

RibuSP trehalose

Fig. 4. Flux partitioning at the glucose-6-phosphate (G6P) branch point during lysine fermentations of C,glutamicum ATCC 21253 with (A) glucose (control) or (8)gluconate at the primary carbon source. Fermentations were conducted aerobically in a 10-liter fermentor in which the consumption of glucose (or gluconate), oxygen, and ammonium and the production of Ivsine. biomass. carbon dioxide. and bv-oroducts ftrehalose. acetate., pvru, vate, hanine, v k n e , and lactatej were trlquently konitored'(37). Ac-Ulation or consumption rates of extracellular metabolites were determined following the termination of exponential growth and the commencement of lysine overproduc~on, These rates were then used with the metabolite balance constraints to estimate the metabolic flux distributions. Absolute fluxes (in parentheses) are in units of millimoles per hour per gram of cells. - 4

SCIENCE, VOL. 252

results of this sensitivity analysis in turn provide the basis for relating metabolic response to nodal type since metabolic responses suggesting various metabolic modifications. A major drawback of are case-specific due to the variability of network structure (depenthese methods is that they are valid only in the local neighborhood dent versus independent), control architecture, and perturbation of the operating point evaluated. Yet, the objective in metabolic used. For this same reason, the types of perturbations used to assess engineering is to modify primary metabolism such that the resulting nodal rigidity are also case specific, but can be classified into four flux distributions (as in Fig. 1C) differ radically from those observed general categories: (i) attenuation of enzyme activity through the during growth (as in Fig. 1A). Furthermore, enzymes with large addition of a specific inhibitor; (ii) amplification or attenuation of flux control coefficients (30) are not always the ones to be modified, enzyme activity through genetic modifications; (iii) environmental especially if thev are involved in feedback control l o o ~ s .For perturbation, such as a change in carbon source; and (iv) deregulaexample, in a sequence of reactions with a simple feedback loop, the tion of a different metabolite to increase metabolic burden. These enzyme that consumes the feedback metabolite strongly controls the types of perturbations were applied to investigate the rigidity of the flux. However, it is usually the removal of the feedbHck loop that G6P, Pyr, and PEP principal nodes in the lysine biosynthetic should be considered, not activity amplification of the enzymes that network (Fig. 1) (37). consume the feedback metabolite. Nevertheless, MCT, BST, and In order to examine the possibility that lysine synthesis is limited related analysis techniques (34) are still useful tools in the analysis of by NADPH availability (that is, the rigidity of the G6P node), C. metabolic networks. glutamicum was cultivated on gluconate instead of glucose since the When enzyme kinetics are unavailable or incomplete, metabolic former is known to bypass the G6P node by directly entering the flux distributions can be obtained by applying stoichiometrically pentose phosphate pathway (43). Flux partitioning at the G6P node derived mass balance constraints to experimental data on the rates of (Fig. 4) calculated from the fermentation data reveals that 50% accumulation or depletion of extracellular metabolites. We (35-37), more NADP was reduced per mole of substrate consumed in the as well as others (19, 38, 39), have used this method to determine gluconate fermentation than in the glucose fermentation (a result of metabolic fluxes under different conditions and during different the higher flux supported by the second dehydrogenase in the phases of culture. There is obviously a limit to the extent that the oxidative branch of the pentose phosphate pathway). Lysine yield in structure of a biochemical network can be delineated by using both fermentations, however, was -30% molar, which indicates extracellular measurements only (40), and the use of radioactive (16, that lysine yield is not limited by flux partitioning at the G6P node. 41) or stable (42) isotope tracers can enhance the power of this In support of these findings, a second perturbation of the G6P node approach. However, such tracer measurements should be made in was conducted in which a strain of C. glutamicum was isolated that addition to the usual extracelluiar measurements taken in the course expressed G6P isomerase at only 10% of its native activity. Interof a fermentation if the metabolic network and its f l u distributions estingly, the mutation did not affect lysine yield or flux partitioning are to be further elucidated. Application of mass balances can also but did result in a 75% attenuation of all metabolic fluxes (37). ,, a yield flux distributions for a specified metabolic burden (such as in response consistent with a dependent network that harbors a rigid Fig. 1, A and B) as well as estimate the maximum theoretical yield node. These results, along with the efficiency by which f l u distrifor a product (such as in Fig. 1C) but cannot be used to predict f l u butions at the G6P node shifted in response to ah increased demand responses to metabolic perturbations. However, in conjunction with for NADPH during lysine synthesis in the control fermentation experimental perturbations, f l u distribution estimates can often be (37), established the G6P node as flexible and imply that either the used to identify the source or sources of metabolic rigidity, as Pyr or PEP nodes must be rigid. described below. The possibility that lysine yield is limited by the flux partitioning ,The degree by which f l u partitioning at a principal node limits at the Pyr principal node was examined by perturbing this node (37) product yield can be assessed by experimentally perturbing a branch by adding fluoropyruvate [a competitive inhibitor of PDC (6)] that is proximal to the principal node under investigation and during normal lysine fermentation. Flux distributions estimated comparing the resulting local f l u distributions to-those observed from the fermentation data (Fig. 5) demonstrate that pyruvate was under nominal conditions. If the flux partitioning at the node can be diverted from the TCA cycle, but the increase in pyruvate availability clearly redirected in a manner consistent with the perturbation, then did not enhance lysine yield. In a second perturbation experiment the node is either flexible or weakly rigid; otherwise, the node may (37), flux distributions were determined for a strain of C.glutamicum be strongly rigid. It is not possible to provide a comprehensive map that expressed only 1% of nominal PDC activity. This mutation did not effect instantaneous lysine yield or flux partitioning but did result in reducing metabolic flues by 75% in an analogous manner as the G6P isomerase mutation. These two perturbation experi(3.00) (3.64)

ments indicate that lysine yield is not limited by pyruvate availability (that is, Pyr node is flexible), which implies, in conjunction with the PEP PEP previous results, that the PEP node must be the source of the network rigidity. However, the control architecture underlying the cause of the PEP node rigidity cannot be determined from the perturbation f l u analysis technique; therefore, the kinetics of the enzymes associated with the PEP node were examined in detail. From inspection, the PEP node control architecture in C. gluLysine ACCOA Lyslne ACCOA tamicum exhibits characteristics of a strongly rigid node since PEP Fig. 5. Flux partitioning at the PEP and Pyr branch points during (A) the control lysine fermentation and ( 6 )an identical fermentation following the carboxylase (PPC) is inhibited strongly by aspartate, but this addition of 2 rnM fluoropyruvate (FP) at the termination of exponential inhibition is alleviated by activation by acetyl coenzyme A (AcCoA) growth and at the start of lysine overproduction (37). The addition of FP (44). To further examine the rigidity of the PEP branch point, a . resulted in the extracellular accumulation of pyruvate (Pyr,,) and dimin- kinetic model of the node was constructed utilizing enzymekinetics ished the flux into the TCA cycle; however, lysine yield remained unaffected at 30% molar. Absolute fluxes (in parentheses) are in units of millimoles per available from glutamic acid bacteria (37). The model consists of hour per gram of cells, and fluxes to biosynthesis are not shown. See caption kinetic expressions for enolase (ENO), -PPC, Pyr kinase (PK) (45), of Fig. 4 for measurements taken to construct flux diagrams. aspartate aminotransferase, PDC, aspartate kinase (AK), and citrate \

21 JUNE 1991

ARTICLES

1679

synthase (CS). The flux supported by the g1ucose:PEP phosphotransferase system (43) was assumed to be one-half the enolase flux, and all of the aspartyl phosphate synthesized was assumed to condense with pyruvate to produce lysine (kinetics of dihydrodipicolinate synthase were not included in the model). The rigidity of the PEP node is illustrated by the steady-state flux distributions obtained from the model following attenuation of PK activity (Fig. 6). For the nominal system (Fig 6, sblid lines), attenuation of PK activity results in a decrease of PPC flux and a general collapse of flux through the PEP node. If the effects of Asp and AcCoA are removed from the kinetics of PPC, the PEP node reverts to a weakly rigid node since attenuation of PK activity results in a diversion of flux into the PPC branch (Fig. 6, dashed lines). The strong rigidity of the PEP node is enhanced by (i) the high activity of PK and its strong affinity for PEP and (ii) the low activity of AK and its poor affinity for Asp, which maintains the high intracellular pool of Asp. O n the basis of these simulations, it appears that flux partitioning at the PEP node can be made favorable for lysine synthesis by transforming C. glutamicum with a deregulated PPC enzyme at approximately tenfold its native activity (amplification of the native PPC activity did not result in significant flux partitioning enhancement in simulations). The analysis of the PEP node brings up an important point regarding mitigation of strong rigidity. Although genetic engineering techniques can be used to modify a rigid node and render it flexible (such as changing substrate or inhibitor affinity), at this time such modifications a r e still quite difficult to achieve. Another procedure that may prove more advantageous is to use enzymes from organisms that have different control architectures for the same biochemical pathway. A good example is the regulation of the PPC flux in photoautotrophs. Since C, plants and some cyanobacteria use PPC in CO, fixation (46), the control architecture of the PEP node in many photoautotrophs differs from heterotrophs in that PPC is not inhibited (or notinhibited as strongly) by Asp nor does it require activation by AcCoA (47, 48). Consequently, flux partition at the PEP node in C. glutamicum may be improved by transforming this strain with PPC from a cyanobacter&, such as ~ n a c y s t ; nidulans (48), although other possibilities exist (49). It is important to realize that flux distributions that are optimal for product synthesis may not support growth. ~ o n s e ~ u e n tsuch l ~ , mddifica-

tions should be constructed so that they may be "turned on" after sufficient biomass has accumulated.

Conclusions Relatively few examples have been offered in strain improvement through alterations of primary metabolism. This lack of significant progress in the field is partly attributed to the complex interactions of metabolic networks but also to the lack of a general, practical approach to metabolic engineering. T o address this problem, we have proposed a working hypothesis regarding metabolic rigidity and have developed an approach to identify the source or sources of rigidity that should foster attempts to optimize metabolic flux distributions for metabolite overproduction. We have proposed that the control architecture of primary metabolism, which is optimized for the synthesis of cellular components, prevents radical alterations of metabolic flux distributions that would allow certain metabolites to be overproduced near maximum yield. Analysis of metabolic flux distributions that are optimal for product synthesis reveals that modifications in flux partitioning need occur only at the principal nodes of the network. Control architecture .or the kinetics of a principal node may limit the extent to which flux partitioning would occur following product deregulation. Flux mapping coupled with metabolic perturbations can be used to assess the flexibility of the principal nodes. Since the alteration or bypass of a strongly rigid node may require significant alterations of enzyme kinetics, we propose that nodal rigidity may be alleviated by tapping into the diversity of control architectures that are found in other organisms. REFERENCES AND NOTES

1. S. E. Lindow, N. J. Panopoulos, B. L. McFarland, Science 244, 1300 (1989). 2. D. B. KeU and H. V. Westerhoff, Trends Biotechnol. 4, 137 (1986). 3. J. E. Bailey, S. Birnbaum, J. L. Galazzo, C. Khosla, J. V. Shanks,Ann. N.Y. Acad. Sci. 589, 1 (1990); J. W. Cary, D. J. Petersen, G. N. Bennen, E. T. Papoutsakis, ibid., p. 69. 4. S. Kinoshita, in Biology of Industrial Microorganisms, A. L. Demain and N. A. Solomon, Eds. (Benjamin/C&ngs, Menlo Park, CA, 1985), pp. 115-142. 5. Although high lysine-producing strains have been isolated, the genetic modilications that lead to such strains remain uncertain due to the isolation techniques used: I. Shiio, A. Yokata, S.-i. Sugimoto,Agric. Biol. Chem. 51,2485 (1987); A. Yokata and I. Shiio, ibid. 52, 455 (1988); I. Shiio, Y. Toride, S.-i. Sugirnoto, ibid. 48, 3091 (1984). 6. 0. Tosaka, Y. Yoshiiara, S. Ikeda, K. Takinami, Agric. Biol. Chem. 49, 1305 (1985) \ - - - - I -

Attenuation of PK activity (%)

Fig. 6. Simulated steadystate fluxes (in millimoles per hour per gram of cells) supported by (A) enolase (ENO), (B) PEP carboxylase (PPC), and (C) pyruvate kinase (PK) following attenuation of PK activity in the PEP branch point model with either native PPC kinetics (solid lines) or deregulated PPC kinetics (dashed lines) (37). Concentrations of 2phosphoglycerate (2PG), ATP, and AMP were held constant at 4, 7, and 1 mM, respectively. These parameters were adjusted such that the steady-state fluxes matched those observed in a typical lysine fermentation.

7. Unfortunately, emphasis on transforming C. glutamicum (and related strains) has taken precedence over documenting the metabolic effects of the genetic alterations: n ~ ,J. Gen. ~iGobiol. 134, 3221 J. ~ r e k e r ,C. Treptow, L. ~ ~ ~ eH.c Sahm, (1988); E. Menkel, G. Thierbach, L. Eggeling, H. Sahm,Appl. Environ. Microbiol. 55, 684 (1989); M. O'Regan et al., Gene 77, 237 (1989). 8. D. Herbert, Symp. Soc. Gen. Microbiol. 11, 391 (1961). 9. J. L. Ingraham, 0. Madge, F. C. Neidhardt, Growth of the Bacterial Cell (Sinauer, Sunderland, MA, 1983), pp. 87-173; A. H. Stouthamer,Antonie van Leeuwenhoek 39, 545 (1973). 10. H. L. Kornberg, Essays Biochem. 2, 1 (1966). 11. S. Kinoshita and K. Nakayama, in Economic Microbiology, A. H. Rose, Ed. (Academic Press, New York, 1978), vol. 2, pp. 209-261. 12. The OaA branch point would be considered a principal node if glutamate (or some other TCA-derived metabolite) was produced as a by-product during lysine overproduction. For example, a significant fraction of the net OaA synthesized during balanced growth is consumed by the glutamate amino acid family (Fig. 1A). 13. B. D. Sanwal, Bacterial. Rev. 34, 20 (1970). 14. H. E. Umbarger, Annu. Rev. Biochem. 38,. 323 (1969). 15,. K. Nakayama, in Microbial Production $Amino Acids, K. Yamada, S. Kinoshita, T. Tsunoda, K. Aida, Eds. (Kodansha, Tokyo, 1972), pp. 369-397; 0. Tosaka, M. Ishihara, Y. Morinaga, K. Takinami, Agru. Biol. Chem. 43, 265 (1979). 16. K. Walsh and D. E. Koshland, Jr., J. Biol. Chem. 260, 8430 (1985). 17. D. C. LaPorte, K. Walsh, D. E. Koshland, Jr., ibid. 259, 14068 (1984). 18. D. E. Koshland, Jr., K. Walsh, D. C. LaPorte, Curr. Top. Cell. Regul. 27, 13 (1985). 19. W. H. Holms, Biochem. Soc. Symp. 54, 17 (1987). 20. B. I. Kurganov, Allosterk Entymes: Kinetic Behavior (Wiley, New York, 1982). 21. This is easily achieved if the enzymes that consume-Pand B exhibit poor affinity for these metabolites and are at or near saturation under nominal conditions. 22. Interestingly, histidine also represses anthranilate synthetase: J. F. Kane and R. A. SCLENCE, VOL. 252

Jensen, J. Biol. Chem. 245, 2384 (1970); R. A. Jensen, ibid. 244, 2816 (1969). 23. B. D. Sanwal,M. Kapoor, H. W. Duckworth, Curr. Top. Cell. Regul. 3, 1 (1971). 24. B. D. Sanwal and P. Maeba, J. Biol. Chem. 241, 4557 (1966); E. R. Stadunan, Enzymes 1, 397 (1970). 25. A. M. Faxel and R. A. Jensen, Arch. Biochem. Biophys. 200, 165 (1980). 26. A. L. Miller and D. E. Atkinson, ibid. 152, 531 (1972); C.-L. Liao and D. E. Atkinson, J. Bacterial. 106, 37 (1971);D. E. Atkinson, CellularEnergy Metabolism and Its Regulation (Academic Press, New York, 1977), pp. 193-200. 27. J. L. Galazzo and J. E. Bailey, Enzyme Microb. Technol. 12, 162 (1990). 28. B. Chance, D. Garfinkel, J. Higgins, B. Hess, J. Biol. Chem. 235, 2426 (1960); D. Garfinkel and B. Hess, ibid. 239,971 (1964);D. Garfinkel, Trends Biochem. Sci. 6, 69 (1981). 29. A. Joshi and B. 0.Palsson, J. Theor. Biol. 141,515 (1989); ibid. 142,41 (1990). 30. H. Kacser and J. A. Burns, Symp. Soc. Exp. Biol. 27,65 (1973); R. Heinrich and T. A. Rapoport, Eur. J. Biochem. 42, 89 (1974); R. Heinrich, S. M. Rapoport, T. A. Rapoport, Prog. Biophys. Mol. Biol. 32, 1 (1977). 31. M. A. Savageau, J. Theor. Biol. 25, 365 (1969); ibid. 26, 215 (1970). 32. Discussion Forum, Trends Biochem. Sci. 12, 4 (1987); ibid., p. 216; A. K. Groen and J. M. Tager, Biochem. J. 253, 619 (1988). 33. D. E. Atkinson et al., in Control ofMetabo1~Processes, A. Cornish-Bowden and M. L. CQrdenas,Eds. (Plenum, New York, 1989), pp. 3-27; ibid., pp. 413-427. Several other papers on MCT and related topics can also be found in this publication. 34. B. 0. Palsson and E. N. Lightfoot, J. Theor. Biol. 113, 261 (1985); R. A. Majewski and M. M. Domach, BioSystems 18, 15 (1985); J. C. Liao and E. N. Lightfoot, Biotechnol. Bioeng. 31, 869 (1988); R. A. Majewski and M. M. Domach, ibid. 36, 166 (1990). 35. J. J. Vallino and G. Stephanopoulos, Ann. N.Y. Acad. Sci. 506, 415 (1987). 36. , in Frontiers in Bioprocessing, S. K. Sikdar, M. Bier, P. Todd, Eds. (CRC Press, Boca Raton, FL, 1990), pp. 205-219. 37. J. J. Vallino, thesis, Massachusetts Institute of Technology, Cambridge (1991). 38. S. Aiba and M. Matsuoka, Biotechnol. Bioeng. 21, 1373 (1979);E. T. Papoutsakis, ibid. 26, 174 (1984);K. F. Reardon, T.-H. Scheper, J. E. BGley, Biotechnol. Prog.

3, 153 (1987); S. P. Tsai and Y. H. Lee, Biotechnol. Bioeng. 32, 713 (1988); S. C. Niranjan and K.-Y. San, ibid. 34, 496 (1989). 39. W. H. Holms, Curr. Top. Cell. Regul. 28, 69 (1986). 40. S. P. Tsai and Y. H. Lee, Biotechnol. Prog. 4, 82 (1988). 41. J. J. Blum and R. B. Stein, in Biological Regulation and Development, R. F. Goldberger and K. R. Yamamoto, Eds. (Plenum, New York, 1982), vol. 3A, pp. 99-125; R. B. Stein and J. J. Blum, J. Biol. Chem. 254, 10385 (1979). 42. L. Inbar and A. Lapidot, Eur. J. Biochem. 162, 621 (1987); J. A. den Hollander, K. Ugurbil, R. G. Shulman, Biochemistry 25,.212 (1986);K. Yamaguchi, S. Ishino, K. Araki, K. Shirahata,Agric. Biol. Chem. 50, 2453 (1986). 43. M. Mori and I. Shiio, Agric. Biol. Chem. 51, 129 (1987). 44. I Shiio and K. Ujigawa-Takeda, ibid. 43, 2479 (1979); M. Mori and I. Shiio, J. Biochem. 97, 1119 (1985); ibid. 98, 1621 (1985);Agric. Biol. Chem. 50, 2605 (1986). 45. The Monod allosteric model [J. Monod, J. Wyman, J.-P. Changeux,,]. Mol. Biol. 12,88 (1969)l was used to represent the kinetics of PPC and PK, where the kinetic parameters were obtained from nonlinear least-square fits to kinetic data published for PPC and PK [H. Ozaki and I. Shiio, J. Biochem. 66, 297 (1969)l. 46. M. D. Hatch and C. R. Slack, Annu. Rev. Plant Physiol. 21, 141 (1970); B. Colman, K. H. Cheng, R. K. Ingle, Plant Sci. Lett. 6,123 (1976); G. W. Owttrim and B. Colman, Biochem. Cell Biol. 66, 93 (1988). 47. M. F. Utter and H. M. Kolenbrander, Enzymes 6, 117 (1972). 48. T. Kodaki, F. Katagiri, M. Asano, K. Iqui, H . Katsuki, J. Biochem. 97, 533 (1985). 49. Another possibility is to use an effectorless pyruvate carboxylase, such as that found in Pseudomonasfluorescens: A. I. Higa, S. R. Milrad de Forchetti, J. J. Cazzulo, J. Gen. Microbial. 93, 69 (1976). 50. We acknowledge the contribution of A. J. Sinskey and M. Follettie of the MIT Department of Biology in our work on the analysis of C . glutamrcum metabolism. Financial assistance was provided by the National Science Foundation in part through a PYI grant (CBT-8514729)and in part through the MIT Biotechnology Process Engineering Center. Lysine fermentations were conducted in equipment donated by MBR Bioreactors.

Announcing:

As part of an ongoing effort to encourage the development of young scientists and to recognize their achievements in all fields of scientific research, the AAAS will highlight exceptional research by college and university students in a special poster session at the AAAS Annual Meeting, 6-1 1 February 1992, in Chicago. Undergraduate students and graduate students who wish to be considered for this distinction can apply by submitting brief abstracts of their research.

Accepted applicants will have the opportunity to present their research to AAAS members in a oneon-one poster session at the Annual Meeting, and their abstracts will be published in the Annual Meeting Program. In addition, a panel of distinguished scientists will evaluate the poster presentations. The students with the best presentations in their fields will receive cash awards and be recognized during the AAAS awards ceremony at the Annual Meeting.

For complete instructions on how to submit abstracts, watch for the "Call for Papers" in the 6 September 1991 issue of Science, or write: AAAS Meetings, Dept. SM, 1333 H Street, NW, Washington, DC 20005. (Deadline for abstracts is 1 November 1991.)

21 JUNE 1991

ARTICLES

1681

Microbiol Monogr DOI 10.1007/7171_2006_089/Published online: 24 February 2007 © Springer-Verlag Berlin Heidelberg 2007

The L-Lysine Story: From Metabolic Pathways to Industrial Production Christoph Wittmann (u) · Judith Becker Biochemical Engineering, Saarland University, Building A1.5, Im Stadtwald, 66041 Saarbrücken, Germany [email protected] 1

Introduction

2

Lysine Biosynthetic Pathways

3 3.1 3.2 3.3

Corynebacterium glutamicum as a Production Organism Lysine Biosynthesis Central Carbon Metabolism Maximal Lysine Production Capacity

4 4.1 4.2 4.3 4.4 4.5

Strain Engineering Classical Engineering Metabolic Engineering of Lysine Biosynthesis Metabolic Engineering of NADPH Metabolism Metabolic Engineering of Precursor Supply Global Strain Engineering through Systems Biotechnology Approaches

5 5.1 5.2

Industrial Production Processes Large-Scale Manufacturing Process Optimization

References Abstract l-lysine is an essential amino acid required for nutrition of animals and humans. It has to be present in food and feed, which, in many cases, is realized by supplementation of the feed-stuffs with pure lysine. The high importance of lysine in nutrition has stimulated intensive research on the lysine biosynthetic pathways and their regulation and the search for microorganisms capable of over-producing this amino acid. As an important milestone, the glutamate producing soil bacterium Corynebacterium glutamicum was isolated in 1956 and soon received interest to be used for production of another amino acid stemming from the TCA cycle: lysine. Within a few years the first lysine producing strains were obtained. The past 50 years following the discovery of C. glutamicum were characterized by a huge progress towards understanding the physiology of this organism and developing and optimizing industrial production strains. This has resulted in effective biotechnological processes currently used for producing about 750 000 tons of lysine per year. Today, systems-oriented approaches aiming at investigating the link between the different components of cellular physiology such as transcriptome, fluxome and metabolome, provide a novel powerful platform that will surely drive future research towards holistic understanding of lysine over-producing microorganisms as well as the creation of superior production strains.

C. Wittmann · J. Becker

1 Introduction The essential amino acid l-lysine is one of the most important amino acids applied as supplement in animal feed. Animal feed, which is typically based on corn, wheat or barley, is poor in lysine. The supplementation of such feed materials with a lysine rich source leads to optimized growth of e.g. pigs or chicken. The direct addition of lysine hereby has proven especially valuable. It does not cause an extra uptake and metabolization of other amino acids beyond their need so that superfluous formation of ammonia and environmental burden by increased nitrogen loads in the manure is avoided. The continuing development of an increased consumption of white meat in various countries of the western as well as the eastern world has led to an enormous market growth for lysine during the past decades (Fig. 1). Currently, the world-wide production of lysine is about 750 000 tons with a predicted market growth over the next years of about 8%. Since only the l-isomer of lysine is effective as feed supplement, all manufacturing processes utilize fermentative production (Leuchtenberger 1996). The cradle of biotechnological lysine production stands in Japan. Stimulated by the increased demand for amino acids, especially glutamate, a large screening program was initiated in Japan about 50 years ago which, in 1956, led to the discovery of the glutamate excreting microorganism Corynebacterium glutamicum (Kinoshita et al. 1957; Udaka 1960). Within a few years, the first lysine excreting mutants of C. glutamicum were available and applied for production in large scale (Kelle et al. 2005; Kinoshita et al. 1961). Since then a continuous optimization of producing strains and processes has led to efficient manufacturing of lysine from renewable resources covering the high demand for this compound required today (de Graaf et al. 2001; Eggeling and Sahm 1999; Hermann 2003; Pfefferle et al. 2003).

Fig. 1 Development of the world-wide annual biotechnological production of l-lysine

The l-Lysine Story: From Metabolic Pathways to Industrial Production

2 Lysine Biosynthetic Pathways In microorganisms, lysine can be synthesized by two completely different routes: either from 2-oxoglutarate and acetyl-CoA via the α-aminoadipate route or from aspartate via the diaminopimelate route. Two variants of the αaminoadipate route (for review see Velasco et al. 2002 and references therein) occur in higher fungi and in archaea like Thermoproteus neutrophilus on the one hand and in the bacterium Thermus thermophilus on the other hand. In five reactions that are catalyzed by homoisocitrate synthase (EC 4.1.3.21), homoaconitate hydratase/cis-homoaconitase (EC 4.2.1.36) and homoisocitrate dehydrogenase (EC 1.1.1.155), 2-oxoglutarate and acetyl-CoA are converted to α-aminoadipate. In one variant of the pathway, α-aminoadipate is converted to lysine by α-aminoadipate reductase (EC 1.2.1.32), saccharopine reductase (EC 1.5.1.10) and saccharopine dehydrogenase (EC 1.5.1.7 or 1.5.1.8). In another variant of the pathway, first described in Thermus thermophilus, the conversion of α-aminoadipate to lysine occurs via the acetylated intermediates N2 -acetyl-l-α-aminoadipate, N2 -acetyl-l-aminoadipyl-δphosphate, N2 -acetyl-l-α-aminoadipate semialdehyde and N2 -acetyl-l-lysine catalyzed by the gene products of lysX, lysZ, lysY and either lysJ or argD and either lysK or argE. In bacteria and plants, lysine may be synthesized from aspartate by one or several of four variants of the diaminopimelate route. These pathway variants diverge at the common intermediate tetrahydrodipicolinate (Born and Blanchard 1999; Schrumpf et al. 1991; McCoy et al. 2006). As shown in Fig. 2, one of these pathways involves succinylated intermediates, while the acetylase pathway comprises acetylated intermediates and the so-called dehydrogenase pathway directly forms dl-diaminopimelate from tetrahydrodipicolinate (Schrumpf et al. 1991; Wehrmann et al. 1998). The aminotransferase pathway, recently described to operate in Chlamydia, converts tetrahydrodipicolinate to ll-diaminopimelate, which can then be epimerized to dl-diaminopimelate (McCoy et al. 2006). Most bacteria only comprise one of these pathways (Bartlett and White 1985; White 1983). Whereas the succinylase pathway is present both in gram negative and gram positive bacteria, the acetylase variant seems to be exclusively used by some Bacillus species (Bartlett and White 1985; Born and Blanchard 1999; Weinberger and Gilvarg 1970). Only in a few organisms like different species of the genera Corynebacterium and in Bacillus macerans two lysine biosynthetic pathways operate together (Bartlett and White 1985; Malumbres and Martin 1996; Schrumpf et al. 1991). A common feature involved in the different pathways concerns the regulation of pathway flux by feedback inhibition of aspartate kinase. In C. glutamicum only one isoenzyme of aspartate kinase (or aspartokinase) exists, which is encoded by two genes, lysCα and lysCβ, representing the coding sequences for the two subunits of the enzyme (Kalinowski et al. 1990). Its activity is bio-

C. Wittmann · J. Becker

Fig. 2 The four pathways of d,l-diaminopimelate and lysine synthesis from aspartate in prokaryotes: succinylase pathway (A); acetylase pathway (B), dehydrogenase pathway (C), and aminotransferase pathway (D). Enzymes involved in the succinylase pathway are: (1) tetrahydrodipicolinate succinylase (DapD); (2) N-succinyl-aminoketopimelate aminotransferase (DapC); and (3) N-succinyl-diaminopimelate desuccinylase (DapE). Enzymes of the acetylase pathway are: (4) tetrahydrodipicolinate acetylase; (5) N-acetylaminoketopimelate aminotransferase; and (6) N-acetyl-diaminopimelate deacetylase. Via the aminotransferase pathway, l,l-diaminopimelate is formed by (8) tetrahydrodipicolinate aminotransferase. (9) Diaminopimelate epimerase (DapF) is common for these three pathways. The dehydrogenase pathway directly forms d,l-diaminopimelate via (7) diaminopimelate dehydrogenase (Ddh). Lysine is built from d,l-diaminopimelate by (10) diaminopimelate decarboxylase

The l-Lysine Story: From Metabolic Pathways to Industrial Production

chemically regulated through concerted feedback inhibition by lysine and threonine that bind at the regulatory β-subunits (Kalinowski et al. 1991; Malumbres and Martin 1996). A similar regulation has been described for lysine synthesis in B. megaterium which, however, has two different isoenzymes. One of the present isoenzymes is also inhibited by a concerted action of two amino acids, which are lysine and methionine, whereas the other isoenzyme is only subject to inhibition by threonine (Chatterjee and White 1982). Additionally to the modulation of enzyme activity by metabolites, some bacteria e.g. E. coli and B. megaterium also show a transcriptional regulation for enzymes involved in lysine biosynthesis (Chatterjee and White 1982; Cremer et al. 1988).

3 Corynebacterium glutamicum as a Production Organism Corynebacterium glutamicum, including its subspecies Brevibacterium flavum, Brevibacterium lactofermentum, Corynebacterium lilium, Corynebacterium efficiens and Brevibacterium divaricatum (Liebl 2005; Liebl et al. 1991) is the most important organism for industrial lysine production. The only other species used for lysine production are recombinant E. coli strains (Imaizumi et al. 2005, 2006). The capability of C. glutamicum to secrete amino acids was discovered in the 1950s (Kinoshita et al. 1957; Udaka 1960). It is a gram-positive, rod-like, non-motile, aerobic bacterium (Fig. 3). Fur-

Fig. 3 Raster electron micrograph of Corynebacterium glutamicum ATCC 13032 cultivated on minimal glucose medium. Samples were taken during exponential growth (Krömer, Heinzle and Wittmann 2006, unpublished results)

C. Wittmann · J. Becker

The l-Lysine Story: From Metabolic Pathways to Industrial Production
Fig. 4 Central metabolic pathways and lysine biosynthesis in Corynebacterium glutamicum. The numbers given for the reactions in the lysine biosynthesis pathway relate to Table 1, where further details, for example on the corresponding enzymes, the encoding genes, and the patents claimed by major industrial players in the field are given

ther characteristics comprise a cell wall with arabino-galactan and mycolic acids with 26 to 36 carbon atoms and a murein sacculus with peptido-glycan cross linked via meso-diaminopimelic acid (Goodfellow et al. 1976). The GCcontent of its genome is 53.8% (Kalinowski et al. 2003). Extensive biochemical studies in the last decades as reviewed in a recent handbook on Corynebacterium glutamicum (Eggeling and Bott 2005) and the recent unravelling of the genome sequence (Haberhauer et al. 2001; Ikeda and Nakagawa 2003; Tauch et al. 2002; Kalinowski et al. 2003) have contributed to a detailed knowledge of the reactions of lysine biosynthesis and central metabolism in this microorganism (Fig. 4). 3.1 Lysine Biosynthesis Lysine belongs to the aspartate amino acid family and in C. glutamicum is produced from pyruvate, oxaloacetate and two ammonia molecules involving the additional supply of four NADPH as reducing power (Michal 1999). Interestingly, the organism has a split pathway for the biosynthesis of lysine (Schrumpf et al. 1991; Sonntag et al. 1993). The two alternative branches give C. glutamicum an increased flexibility in response to changing environmental conditions, involving e.g. different ammonia levels (Sahm et al. 2000). dl-diaminopimelate as intermediate of the lysine pathway additionally is an essential building block for the synthesis of the murein sacculus (Wehrmann et al. 1998). Concerning regulation of lysine biosynthesis, aspartokinase (EC 2.7.2.4), catalyzing the formation of aspartyl phosphate from aspartate is the key enzyme. It is feedback regulated by concerted action of lysine and threonine (Kalinowski et al. 1991). Via the requirement for building blocks and cofactors lysine formation is closely linked to the central metabolism. The enzymes involved in lysine biosynthesis are summarized in Table 1. 3.2 Central Carbon Metabolism The most relevant substrates for industrial lysine production are starch and molasses. They are based on glucose, fructose, and sucrose as major carbon sources. These compounds are taken up via phosphoenolpyruvate-dependent phosphotransferase systems (Dominguez and Lindley 1996; Dominguez et al. 1998; Malin and Bourd 1991; Moon et al. 2005). The carbon source has an in-

aspartate kinase

aspartate semialdehyde dehydrogenase dihydrodipicolinate synthase dihydrodipicolinate reductase tetrahydrodipicolinate succinylase succinyl-amino-ketopimelate transaminase succinyl-diaminopimelate desuccinylase meso-diaminopimelate dehydrogenase diaminopimelate epimerase diaminopimelate decarboxylase lysine permease

lysC

asd

lysE

lysA

dapF

ddh

dapE

dapC

dapD

dapB

dapA

Enzyme

Gene

4.1.1.20

5.1.1.7

1.4.1.16

3.5.1.18

2.6.1.17

2.3.1.117

1.3.1.26

4.2.1.52

1.2.1.11

2.7.2.4

EC number lysine threonine lysine threonine

Inhibitory ligands

lysE

lysA

dapF

ddh

dapE

dapC

dapB-orf2dapA-orf4 dapB-orf2dapA-orf4 dapD

asd

lysC

Transcription unit

induced by lysine

Reaction number

4

3

2

8

(Vrljic et al. 1996)

11

10

(Hartmann et al. 2003) 9

(Cremer et al. 1988)

(Wehrmann et al. 1998) 7

(Hartmann et al. 2003) 6

(Wehrmann et al. 1998) 5

(Patek et al. 1997)

(Patek et al. 1997)

(Cremer et al. 1988)

(Kalinowski et al. 1991) 1

Reference

inhibited by lysine (Cremer et al. 1988)

Transcriptional regulation

•







 •

•

•

•

•

•

•

•

•



•

Patents

Table 1 Genes and enzymes involved in l-lysine biosynthesis in C. glutamicum. The information on patents and patent applications by the lysine producing companies Archer Daniels Midland (), Ajinomoto (), BASF(•), Degussa () and Kyowa Hakko Kogyo () claiming an improvement of lysine production through modification of the lysine biosynthetic pathway in C. glutamicum is taken from Kelle et al. (2005). The reaction numbers relate to the corresponding metabolic reactions of lysine biosynthesis in the metabolic network shown in Fig. 4

C. Wittmann · J. Becker

The L-Lysine Story: From Metabolic Pathways to Industrial Production

fluence on lysine production by C. glutamicum, which is related to the resulting different entry points into the central metabolism (Kiefer et al. 2002). The central catabolic network previously identified in C. glutamicum comprises the pathways of glycolysis, pentose phosphate pathway (PPP), tricarboxylic acid (TCA) cycle and glyoxylate cycle, while the Entner–Doudoroff pathway has not been detected (Eikmanns 2005; Kalinowski et al. 2003; Yokota and Lindley 2005). The oxidative part of the PPP comprises the NADPH dependent enzymes glucose 6-phosphate dehydrogenase (EC 1.1.1.49) (Ihnen and Demain 1969; Moritz et al. 2000) and 6-phophogluconate dehydrogenase (1.1.1.44) (Moritz et al. 2000; Ohnishi et al. 2005). Both enzymes were intensively studied regarding their kinetic properties (Moritz et al. 2000). C. glutamicum exhibits an extensive set of different enzymes for the interconversion of C4 metabolites of the TCA cycle and C3 metabolites of glycolysis (Fig. 4). For anaplerotic replenishment of the TCA cycle C. glutamicum possesses pyruvate carboxylase (EC 6.4.1.1) (Peters-Wendisch et al. 1998) and phosphoenolpyruvate carboxylase (EC 4.1.1.31). Malic enzyme (EC 1.1.1.40) (Gourdon et al. 2000), PEP carboxykinase (EC 4.1.1.32) (Jetten and Sinskey 1993; Riedel et al. 2001), oxaloacetate decarboxylase (EC 4.1.1.3) (Jetten and Sinskey 1995) and probably phosphoenolpyruvate synthetase (Jetten et al. 1994), of which the genes of oxaloacetate decarboxylase and phosphoenolpyruvate synthetase have not been found yet, catalyze decarboxylation reactions from the TCA cycle towards glycolysis. In selected cases, NADPH supply by malic enzyme supports growth and lysine production of C. glutamicum (Dominguez et al. 1998; Kim et al. 2006). However, deletion of malE does not affect growth of C. glutamicum on glucose or acetate, suggesting that the metabolic pathways involved in NADPH-production are highly flexible (Gourdon et al. 2000). It has been proposed that cyclic cooperation of the enzymes between pyruvate/phosphoenolpyruvate and malate/oxaloacetate is involved in the regeneration of excess ATP (Marx et al. 1996; Petersen et al. 2000; Riedel et al. 2001). In addition to the central catabolic routes, also anabolic pathways have been studied in detail providing detailed information on the cellular composition and the precursor demand for growth (Wittmann and de Graaf 2005). Overall about (16.4 mmol NADPH) g–1 is required for biomass synthesis. It is obvious that the anabolic NADPH requirement competes with lysine production. Considering a biomass yield of 0.5 (g dry biomass) (g glucose)–1 , which is achieved by C. glutamicum under aerobic conditions this results in 1.7 mol NADPH (mol glucose)–1 that have to be generated by the NADPH forming reactions in the PPP and the TCA cycle. It should be noted that the knowledge about the metabolic network is continuously being updated with novel findings such as the detection of pyruvate carboxylase as anaplerotic enzyme (Peters-Wendisch et al. 1998) or the consideration of new reactions for growth on specific substrates such as sucrose (Dominguez and Lindley 1996).

C. Wittmann · J. Becker

3.3 Maximal Lysine Production Capacity The maximal capacity, i.e. the theoretical maximum yield, of a C. glutamicum cell for lysine production is an important characteristic, since it provides an estimate of the remaining optimization potential of a running industrial process and gives advice for process or genetic engineers. Previous stoichiometric calculation considering only the major pathways involved in lysine production have yielded a molar lysine yield on glucose of 75% (Hollander 1994). A more detailed insight can be gained via stoichiometric network analysis considering the full set of available pathways in the central metabolism with information on reversibility or irreversibility of the different reactions and additional assumptions and restrictions posed on the metabolic network. This approach, named elementary flux mode analysis (Schilling et al. 1999; Schuster et al. 2002), has been recently applied to elucidate the potential of C. glutamicum for methionine production (Krömer et al. 2006). The theoretical maximum molar yield of C. glutamicum for lysine production obtained by such an analysis is 82% (Fig. 5). The in silico pathway analysis also provides information on the reactions contributing to this theoretical optimum, i.e. the corresponding theoretical metabolic flux distribution. Shown for one of the optimal elementary flux modes, i.e. optimal lysine production, PPP and malic enzyme supply the required NADPH (Fig. 5). Hereby the concerted action of pyruvate carboxylase, malate dehydrogenase (EC 1.1.1.37) and malic enzyme form a transhydrogenase-like cycle converting NADH into NADPH and thus providing an additional amount of this important cofactor. This metabolic cycle is different from a previously described cyclic pathway involving PCx, PCK, and MDH as shown to operate in vivo (Petersen et al. 2000; Riedel et al. 2001). One should further notice that, in addition to the soluble malate dehydrogenase C. glutamicum possesses another enzyme for interconversion of oxaloacetate and malate, the membrane-bound malate:quinone oxidoreductase (MQO, EC 1.1.99.16). Both enzymes probably catalyze a cyclic reaction in vivo leading to a net oxidation of NADH (Molenaar et al. 2000). The dehydrogenase branch of lysine biosynthesis is the sole lysine producing route. Admittedly, the given scenario implies zero flux through the TCA cycle and no biomass formation, which can hardly be realized. The achievable optimum can be expected to be somewhat lower than the value calculated here, but should still be significantly higher than the yield achieved in practice, which is in the range of about 55% (Ikeda 2003; Kawahara et al. 1990).

The l-Lysine Story: From Metabolic Pathways to Industrial Production

Fig. 5 Metabolic flux distribution of Corynebacterium glutamicum with maximal theoretical lysine carbon yield. All fluxes are given as relative molar fluxes to the glucose uptake in mol (mol)–1 × 100 (Krömer, Heinzle and Wittmann 2006, unpublished). One should note that additionally another elementary mode with optimal production exists which, however, only slightly differs in reaction steps involved in cofactor metabolism

C. Wittmann · J. Becker

4 Strain Engineering After the discovery of its ability to produce and excrete amino acids (Kinoshita et al. 1957), C. glutamicum was used to establish a biotechnological production process for several amino acids. Through the years various methods for strain engineering have been developed to create more efficient production strains. 4.1 Classical Engineering The first production strains were created within a few years after the discovery of C. glutamicum using an iterative procedure of random mutagenesis with UV light or chemical mutagens and subsequent strain selection (Nakayama et al. 1978). The key to success in these days was the use of toxic lysine analogues, such as S-(2-aminoethyl) cysteine, to screen for feedback resistant strains (Nakayama and Araki 1973). These strains later all revealed point mutations in the aspartokinase gene, through which the encoded enzyme was released from feedback inhibition by lysine and threonine (Kalinowski et al. 1991; Thierbach et al. 1990). This modification displays one of the most important characteristics of lysine production strains at all. Consequently, also strains were developed which exhibited a weakened or even blocked biosynthesis of threonine, i.e. auxotrophy for threonine (Nakayama and Araki 1973). Through further cycles of mutagenesis and selection strains with different auxotrophies for other amino acids, vitamins and resistance to other anti-metabolites were obtained (Kelle et al. 2005). The subsequent mutants from such a strain genealogy exhibited a stepwise improvement of production (Schrumpf et al. 1992; Wittmann and Heinzle 2002). Remarkable production properties such as a conversion yield up to 50% and a lysine·HCl titre above 100 g L–1 were achieved with such classically derived strains (Ikeda 2003; Leuchtenberger 1996). The additional nutrient requirement and the weak stress tolerance, due to the large number of undesired mutations accumulated during strain development (Ohnishi et al. 2002), display, however, severe disadvantages of conventional production strains and stimulated targeted approaches for strain optimization. After the gene targets improving lysine production had been identified (see below), it became possible to introduce mutant alleles of these genes isolated from the classically obtained producer strains into the wild type to generate stable and stress-tolerant lysine producer strains without additional nutrient requirements (Georgi et al. 2005; Hayashi et al. 2006a; Ohnishi et al. 2002, 2005).

The l-Lysine Story: From Metabolic Pathways to Industrial Production

4.2 Metabolic Engineering of Lysine Biosynthesis The possibility to perform targeted genetic modifications through developments of molecular biology and genetic engineering tools initiated a number of efforts towards rational optimization to C. glutamicum (Ohnishi et al. 2002, 2005; Sahm et al. 2000). Logically, many of these studies have focussed on the optimization of the flux through the lysine biosynthesis by directly modifying enzymes of the pathway (Fig. 5). The modification of three of the enzymes, i.e. aspartate kinase (LysC), dihydrodipicolinate synthase (DapA, EC 4.2.1.52) and the lysine exporter (LysE), was especially valuable with respect to improvement of lysine production (Fig. 6). Aspartate kinase is the key enzyme with regard to metabolic control of the lysine pathway as it is subject to a feedback inhibition by threonine and lysine (Kalinowski et al. 1991; Malumbres and Martin 1996). Different point mutations in the lysC gene, i.e. in the region coding for its regulatory β-subunit, have been shown to release the enzyme from feedback control and lead to enhanced lysine formation (Cremer et al. 1991; Follettie et al. 1993; Kalinowski et al. 1991; Sugimoto et al. 1997). Similarly, also overexpression of the aspartate kinase gene stimulated production (Jetten et al. 1995). Today, the release of aspartate kinase from feedback control is regarded as one of the most important features of industrial lysine producer strains. This is also underlined by the various patents claiming different amino acid exchanges in this enzyme (Table 1) (Kelle et al.

Fig. 6 Lysine biosynthetic pathway in Corynebacterium glutamicum. Metabolic regulation through feedback inhibition of asparate kinase by concerted action of lysine and threonine and important targets for optimization of lysine overproduction

C. Wittmann · J. Becker

2005). Plasmid encoded amplified expression of the dapA gene significantly increases lysine (Bonnassie et al. 1990; Cremer et al. 1991; Eggeling et al. 1998; Pisabarro et al. 1993). Amplification of dapA expression was further achieved through an extensive mutation of the promoter sequence (Vasicova et al. 1999), whereby a hot spot was discovered at the – 10 region (de Graaf et al. 2001). Also overproduction of diaminopimelate epimerase (DapF, EC 5.1.1.7) and succinyl-aminoketopimelate transaminase (DapC, EC 2.6.1.17), two enzymes of the succinylase branch, was beneficial for lysine formation (Kelle et al. 2005). A striking discovery with respect to lysine production was the discovery of the lysine exporter (LysE) and the subsequent overexpression of the lysE gene which resulted in an increased lysine secretion rate (Bellmann et al. 2001; Vrljic et al. 1996, 1999). The recently performed expression of lysE from C. glutamicum in a Methylophilus methylotrophus lysine producing strain was shown to also improve lysine production from methanol by this organism (Gunji and Yasueda 2006). Summarizing, the importance of engineering enzymes of the lysine pathway for efficient lysine production is underlined by the fact that today every single gene of the lysine biosynthetic pathway is covered with one or several patents by the major players in the field (Table 1). 4.3 Metabolic Engineering of NADPH Metabolism NADPH is consumed by the lysine biosynthetic pathway in four steps either directly or indirectly through the assimilation of ammonium and its efficient supply appears crucial for lysine overproduction. A detailed insight into the NADPH metabolism of C. glutamicum has been obtained by 13 C metabolic flux studies under various physiological conditions (Becker et al. 2005; Kiefer et al. 2004; Marx et al. 1996, 1997; Sonntag et al. 1995; Wittmann and de Graaf 2005; Wittmann and Heinzle 2001a, 2002; Wittmann et al. 2004a). These allow establishing a NADPH balance. Glucose 6-phosphate dehydrogenase, 6-phosphogluconate dehydrogenase and isocitrate dehydrogenase (EC 1.1.1.42), which uses NADP as cofactor in C. glutamicum (Chen and Yang 2000; Eikmanns et al. 1995), are hereby considered as NADPHgenerating reactions, whereas NADPH-consuming reactions comprise lysine production and growth with a stoichiometric demand of 16.4 mmol NADPH (g biomass)–1 (Wittmann and de Graaf 2005). Figure 7A illustrates that the NADPH metabolism of C. glutamicum is highly flexible adjusting to the given overall physiological needs of the cell. NADPH supply and consumption vary depending on the physiological growth state (Marx et al. 1997), the applied carbon source (Dominguez et al. 1998; Wendisch et al. 2000; Wittmann et al. 2004a), or the genetic background (Marx et al. 1999, 2003). In most cases this results in an apparent NADPH excess which seems to increase with diminishing NADPH oxidation. In selected cases even an apparent limitation for NADPH is observed. Examples are phases of maximal lysine production dur-

The l-Lysine Story: From Metabolic Pathways to Industrial Production

Fig. 7 NADPH metabolism of different C. glutamicum strains under various cultivation conditions as revealed by 13 C flux analysis: Overall flux through NADPH supply and demand (A), flux through the lysine pathway and the PPP (B). NADPH supply considers G6P dehydrogenase, 6PG dehydrogenase, and isocitrate dehydrogenase. Lysine production (4 NADPH/lysine) and anabolism are considered as NADPH-consuming reactions. Hereby, the NADPH demand for biomass formation, 16.4 mmol (g cell dry mass)–1 , was based on a detailed analysis of cellular composition (Wittmann and de Graaf 2005). Studies involved flux analysis of (i) non-producing C. glutamicum ATCC 13032 in batch culture on glucose (open diamond) (Sonntag et al. 1995), (ii) non-producing C. glutamicum ATCC 13032 and lysine producing C. glutamicum ATCC 13287, ATCC 21253, ATCC 21526, and ATCC 21543 in batch culture on glucose (open circle) (Wittmann and Heinzle 2002), (iii) lysine producing C. glutamicum ATCC 21253 during maximum production phase in batch culture on glucose (open square) (Wittmann and Heinzle 2001a), (iv) non-producing C. glutamicum LE4 and lysine producing C. glutamicum MH 20–22B in continuous culture on glucose (closed circles) (Marx et al. 1997), (v) lysine-producing C. glutamicum MH 20–22B in continuous culture on glucose (closed square) (Marx et al. 1996), lysine-producing C. glutamicum ATCC 21526 in batch culture on glucose, fructose (closed diamond) (Kiefer et al. 2004) and on sucrose (closed triangle) (Wittmann et al. 2004a), and lysine-producing C. glutamicum lysCfbr and C. glutamicum lysCfbr PEFTU fbp during batch culture on glucose (open triangle) (Becker et al. 2005)

C. Wittmann · J. Becker

ing batch culture or production on fructose allowing only a small PPP flux (Kiefer et al. 2004). The apparent imbalance between supply and demand for NADPH indicates that in C. glutamicum so far unassigned metabolic reactions are active in vivo, which either consume or supply NADPH. Possible candidates for NADPH consumption comprise superoxide-generating NADPH oxidase in the respiratory chain (Matsushita et al. 2001) or a metabolic cycle around the pyruvate node involving malic enzyme as hypothesized previously (de Graaf 2000; Petersen et al. 2000). Clear experimental evidence supporting the role of either one of these reactions has not been obtained to date. However, regarding the transhydrogenase-like cycle consisting of pyruvate carboxylase, malate dehydrogenase, and PEP carboxykinase, it has been shown that deletion of pckA improves lysine production and pckA overexpression reduces lysine production (Riedel et al. 2001) due to altered fluxes within this transhydrogenase-like cycle (Petersen et al. 2001). Concerning NADPH supply, the contribution of malic enzyme has been demonstrated for growth on fructose (Dominguez et al. 1998). Under these conditions cells carry an extremely low PPP flux due to an unfavorable entry point of the substrate into the central metabolism which probably activates malic enzyme (Kiefer et al. 2004). However, overexpression of the malic enzyme gene malE in a genetically defined lysine producing strain did neither improve lysine product yields on glucose nor on fructose or sucrose (Georgi et al. 2005). Since the unassigned NADPH consumption flux becomes minimal with increasing lysine yield (dashed line in Fig. 7A) or NADPH even shows an apparent deficiency, a NADPH limitation of lysine production appears likely. This and the simple fact that four NADPH are required for synthesis of one lysine has stimulated metabolic engineering of the NADPH supply in C. glutamicum (Marx et al. 1999, 2003). As a prerequisite for successful modification of the NADPH metabolism, the key pathways supplying NADPH for lysine production in C. glutamicum were identified by different approaches. Stoichiometric investigation of the lysine network in the early 1990s already predicted that an increased lysine yield is linked to an increased flux through the PPP and a decreased flux through the TCA cycle (Kiss and Stephanopoulos 1992). The importance of the PPP for efficient lysine production was later shown by metabolic flux analysis (Marx et al. 1996) and by genetic experiments (Georgi et al. 2005; Marx et al. 2003; Ohnishi et al. 2005). As a combined outcome of various flux studies with different strains and under different conditions a close correlation of lysine production with the flux through the PPP could be observed (Fig. 7B). The importance of the PPP for efficient lysine production becomes also obvious from the theoretical flux distribution corresponding to optimal production (Fig. 5). In contrast to the PPP, the contribution of isocitrate dehydrogenase decreases with increasing lysine production (Wittmann and Heinzle 2002). In summary, it turns out that (i) lysine production is likely limited by NADPH availability and that (ii) the

The l-Lysine Story: From Metabolic Pathways to Industrial Production

PPP is the major pathway for supply of NADPH. In light of this, different approaches have aimed at redirecting the flux through the PPP in order to increase lysine production by C. glutamicum. As example, deletion of the phosphoglucose isomerase gene pgi, forcing the cell to metabolize the substrate glucose completely via the PPP, resulted indeed in increased lysine production on glucose (Marx et al. 2003). This strategy, however, appears useful only for glucose-based processes, since sucrose-based processes require an active phosphoglucose isomerase for recycling carbon into the PPP and full NADPH supply (Becker et al. 2005; Wittmann et al. 2004a). Amplified expression of the fructose 1,6-bisphosphatase gene fbp in a genetically defined strain of C. glutamicum, only carrying a point mutation in the lysC gene, was shown to increase lysine yield on glucose, fructose, and sucrose up to about 40% (Becker et al. 2005). Hereby, flux analysis of the mutant revealed that the overexpression of fructose 1,6-bisphosphatase gene indeed resulted in a 10% enhanced PPP flux. On the other hand, it was shown that fbp overexpression in a genetically defined strain carrying point mutations in lysC, hom, pyc, and zwf improved lysine yields due to reduced intracellular concentrations of fructose 1,6-bisphosphate, an inhibitor of glucose 6-phosphate DH and 6-phosphogluconate DH, in the engineered strain only on sucrose (Georgi et al. 2005). These differences are very likely to be caused by the different background of the strains, as the strain used by Georgi et al. for amplified expression of fbp already had an engineered PPP (Georgi et al. 2005). Further successful examples comprise the introduction of point mutations into PPP genes, previously identified by comparative sequencing of the C. glutamicum wild type and a classically derived production strain. The substitution A243T in the zwf gene encoding for glucose 6-phosphate dehydrogenase (Georgi et al. 2005; Zelder et al. 2005) and the substitution T1083C in the gnd gene encoding for 6-phosphogluconate dehydrogenase (Ohnishi et al. 2005) both led to a significantly increased lysine titre. The modification of 6-phosphogluconate dehydrogenase caused an 8% increase in the PPP flux (Ohnishi et al. 2005) which probably resulted from positively changed kinetics of the enzyme. 4.4 Metabolic Engineering of Precursor Supply Oxaloacetate is a direct precursor of aspartate-derived amino acids, including lysine. In C. glutamicum, the anaplerotic enzymes phosphoenolpyruvate carboxylase (Eikmanns et al. 1989; O’Regan et al. 1989; Ozaki and Shiio 1969) and pyruvate carboxylase (Park et al. 1997; Peters-Wendisch et al. 1996, 1998) are involved in supplying oxaloacetate. The importance of these enzymes for lysine production becomes obvious from the correlation between the lumped anaplerotic net carboxylation flux and the flux into the lysine biosynthetic pathway under various conditions determined by 13 C metabolic flux analy-

C. Wittmann · J. Becker

sis (Fig. 8). Pyruvate carboxylase is today regarded as the major anaplerotic enzyme in C. glutamicum (Petersen et al. 2000) and overexpression of its gene has been shown to improve lysine production (Peters-Wendisch et al. 2001). This makes sense taking into account that, due to the presence of the phosphotransferase system for sugar uptake, high amounts of pyruvate are generally formed during growth of C. glutamicum. However, deletion of the pyruvate carboxylase gene pyc leads to the inability to grow on lactate as a sole carbon source (Peters-Wendisch et al. 1998), whereas overexpression of pyc in a lysine-producing strain strongly increased production (PetersWendisch et al. 2001). Knowing the importance of pyruvate carboxylase for lysine production, the point mutation C1372T, identified in a classically derived producer, was introduced into the pyc gene and resulted in a strong increase of lysine production (Ohnishi et al. 2002). It should be noted that, however, also overexpression of the phosphoenolpyruvate carboxylase gene ppc is beneficial for the formation of amino acids of the aspartate family (Sano et al. 1987). The metabolism around the pyruvate node in C. glutamicum, was further unravelled by genetic, biochemical, and 13 C metabolic flux analyses,

Fig. 8 Correlation between lysine yield and anaplerotic net flux in different C. glutamicum strains under various cultivation conditions as revealed by 13 C flux analysis. Studies involved flux analysis of (i) non-producing C. glutamicum ATCC 13032 in batch culture on glucose (open diamond) (Sonntag et al. 1995), (ii) non-producing C. glutamicum ATCC 13032 and lysine-producing C. glutamicum ATCC 13287, ATCC 21253, ATCC 21526, and ATCC 21543 in batch culture on glucose (open circle) (Wittmann and Heinzle 2002), (iii) lysine-producing C. glutamicum ATCC 21253 during maximum production phase in batch culture on glucose (open square) (Wittmann and Heinzle 2001a), (iv) non-producing C. glutamicum LE4 and lysine-producing C. glutamicum MH 20–22B in continuous culture on glucose (closed circles) (Marx et al. 1997), (v) lysine-producing C. glutamicum MH 20–22B in continuous culture on glucose (closed square) (Marx et al. 1996), lysine-producing C. glutamicum ATCC 21526 in batch culture on glucose, fructose (closed diamond) (Kiefer et al. 2004) and on sucrose (closed triangle) (Wittmann et al. 2004a), and lysine-producing C. glutamicum lysCfbr and C. glutamicum lysCfbr PEFTU fbp during batch culture on glucose (open triangle) (Becker et al. 2005)

The l-Lysine Story: From Metabolic Pathways to Industrial Production

showing the presence of decarboxylating enzymes converting C4 metabolites of the TCA cycle into C3 metabolites of glycolysis. These enzymes, phosphoenolpyruvate carboxykinase and malic enzyme, operate in addition to the carboxylation enzymes and establish a highly flexible metabolic reaction cycle around the pyruvate node, which is also present in various other microorganisms (Sauer and Eikmanns 2005). It might function in wasting excess ATP under certain conditions (Sauer et al. 1997; Wittmann and Heinzle 2001b) or equilibrating the intracellular pool sizes of metabolites around the pyruvate node (Sauer and Eikmanns 2005). Additionally, a contribution to the NADPH metabolism has been hypothesized (Cocaign-Bousquet and Lindley 1995; de Graaf 2000). In light of these findings, deletion of the genes coding for the decarboxylating enzymes displayed a promising strategy to enhance the anaplerotic net flux. Indeed deletion of the phosphoenolpyruvate carboxykinase gene pckA resulted in a significant improvement of lysine production (Riedel et al. 2001). Neither the deletion nor the overexpression of the malic enzyme gene, however, influenced the metabolism of C. glutamicum on sugars markedly (Gourdon et al. 2000; Netzer et al. 2004). Moreover, overexpression of the malic enzyme gene did not change lysine production by C. glutamicum (Georgi et al. 2005). This might be due to the fact that typically phosphoenolpyruvate carboxykinase catalyzes the major decarboxylation flux in C. glutamicum, whereas malic enzyme obviously only plays a minor role (Gourdon et al. 2000; Petersen et al. 2000). It can, however, not be excluded that the situation might be different under certain physiological conditions. 4.5 Global Strain Engineering through Systems Biotechnology Approaches The experience of the past clearly shows that detailed quantitative knowledge of metabolic physiology is required for rational design of superior production strains. Especially for the optimization of amino acid production by C. glutamicum, characterized by a close connection between central metabolism and product biosynthetic pathways, understanding of global metabolic regulation has turned out to be crucial. The powerful experimental and computational tools available today enable a detailed quantitative investigation of the metabolism of the industrial lysine producer C. glutamicum. A milestone in this research was the sequencing of the genome of C. glutamicum and the investigation of its genetic repertoire (Bathe et al. 1996; Haberhauer et al. 2001; Ikeda and Nakagawa 2003; Kalinowski et al. 2003; Tauch et al. 2002). Transcriptome analysis in C. glutamicum through DNA microarrays provided valuable insights into gene expression under various conditions (Wendisch 2003; Wendisch et al. 2006), such as growth on different carbon sources like glucose or acetate (Gerstmeir et al. 2003; Hayashi et al. 2002; Muffler et al. 2002) or production of lysine (Hayashi et al. 2006b; Krömer et al. 2004). Sim-

C. Wittmann · J. Becker

ilarly, also the analysis of the proteome, based on 2-D gel electrophoresis (Bendt et al. 2003; Hermann et al. 2001; Schaffer and Burkovski 2005; Schluesener et al. 2005) has proven valuable to understand important metabolic processes including, for example, nitrogen starvation (Schmid et al. 2000). The enormous contribution of the quantification of metabolic fluxes (fluxome) to our current understanding of the C. glutamicum metabolism has already been pointed out above. For this purpose comprehensive approaches combining 13 C tracer experiments, metabolite balancing, and isotopomer modelling have been developed and applied to C. glutamicum (de Graaf 2000; Wittmann and de Graaf 2005). Important studies to be mentioned are the comparative analysis of fluxes during growth, glutamate, and lysine production (Marx et al. 1997; Sonntag et al. 1995), in different mutants of a lysine-producing strain genealogy (Wittmann and Heinzle 2002), during co-utilization of acetate and glucose (Wendisch et al. 2000), and during lysine production on different industrially relevant carbon sources (Kiefer et al. 2004; Wittmann et al. 2004a). Further achievements comprise the miniaturization of flux analysis to the µL scale for screening purposes (Sauer 2004; Wittmann et al. 2004b), novel approaches towards the analysis of large-scale production processes (Drysch et al. 2003, 2004; El Massaoudi et al. 2003; Yang et al. 2003, 2006a,b) and current developments aiming at flux analysis under dynamic conditions (Nöh et al. 2006; Nöh and Wiechert 2006; Wiechert and Nöh 2005). These tools have contributed significantly to our current understanding of the C. glutamicum metabolism. To fully describe the physiological state of a biological system, however, not one but all of its functional components (genome, transcriptome, proteome, metabolome, and fluxome) have to be analyzed. First examples of such systems-oriented studies already reveal a great potential (Krömer et al. 2004; Lange et al. 2003; Silberbach et al. 2005) and have stimulated the development of systems biotechnology approaches for future characterization and engineering of C. glutamicum (Wendisch et al. 2006). Such approaches are especially promising for the targeted multidimensional alteration of complex regulatory networks towards better tolerance of production strains to high temperature or salt levels, or extreme pH values (Kelle et al. 2005).

5 Industrial Production Processes 5.1 Large-Scale Manufacturing Today large plants are in use for industrial lysine production and the feedgrade amino acid market is developing towards a few major suppliers (Ajinomoto, ADM, BASF, Cheil Jedang, Degussa, Global Biochem, Kyowa Hakko).

The l-Lysine Story: From Metabolic Pathways to Industrial Production

Fig. 9 Lysine production plant of the BASF AG located in Gunsan, South Korea with an annual capacity of about 100 000 tons. Copyright BASF AG—The chemical company (2003). Reproduced with permission

As an example Fig. 9 shows the lysine production facility of BASF AG in Gunsan, located at the west coast of South Korea. This plant has an annual production capacity of about 100 000 tons, which accounts for about 15% of the total world market. Industrial large-scale manufacturing of lysine can be separated into different steps involved in upstream processing, the fermentation process itself, and the downstream processing. Upstream processing comprises raw material testing, delivery and storage, the preparation of media from the raw materials, and the preparation of the inoculum for the production. The major industrial carbon sources for lysine production are cane molasses, beet molasses, sucrose and dextrose, whereby the latter is obtained from hydrolysis of starch (Ikeda 2003). Because of batch-to-batch variation of these complex nutrient sources, extensive media testing is carried out to ensure the suitability of a certain raw material with respect to product yield or titre. The carbon source is the major cost factor in industrial lysine production (Kelle et al. 2005). Related to this, sugar suppliers and lysine producers are subject to close alliances or companies cover even both, sugar supply and lysine production. Because of the strong impact of the sugar costs the conversion yield is of major importance for the economy of the production process. The preparation of the inoculum typically

C. Wittmann · J. Becker

involves successive cultivation of the production strain in increasing culture volumes, since C. glutamicum exhibits elevated lag phases when inoculated at a biomass concentration below 0.1 g L–1 (Kelle et al. 2005). Today, lysineproducing plants use large-scale fermenter vessels with 500 m3 volume or even more to benefit from the economy of scale. The purification and formulation of the product in the downstream processing is a further important cost factor (Hermann 2003). One applied route comprises cell separation by vacuum filtration, evaporation, and spray drying for product formulation as shown in the flow chart in Fig. 9. Additionally, alternative strategies are used, mainly depending on the lysine preparation finally obtained (Kelle et al. 2005). During the past, lysine was mainly purified from the broth by ion exchange with separation of the biomass, followed by addition of HCl, evaporation, and drying (Hermann 2003). The crystalline lysine·HCl formed is much less hygroscopic than the corresponding sulfate salt (Kelle et al. 2005) and displayed the major product form through the past. In such processes the biomass of C. glutamicum, classified as GRAS organism and thus suitable for animal feed, is utilized separately for feed purposes. Today different developments allow a more economical downstream processing and have led to a number of different lysine preparations, such as liquid lysine (50% purity), granulated lysine sulfate (40–50% purity), or liquid lysine sulfate (20–30% purity), which are today well established on the market (Kelle et al. 2005). The granulated product contains the entire fermentation broth without separation of biomass reducing costs and adding additional nutritional value to the product. 5.2 Process Optimization The optimization of the lysine production process can significantly contribute to reduction of the production costs. The improvement of the downstream processing leading to different lysine preparations as shown above is one illustrative example of the efficiency of such approaches. Other efforts aim at intensification of the process through a change of the operation mode. Repeated batch or fed-batch runs without preparation of the reactor or the inoculum display an interesting approach. After the production is finished a certain fraction of the broth is left in the reactor and is mixed with fresh medium, which significantly decreases the down-time and thus increases the volumetric productivity, i.e. the capacity of a given plant. Problems arise in cases where the production strain is genetically not stable, as in the case of lysine-producing mutants auxotrophic for other amino acids (Hermann 2003). Further improvement could be made by running the lysine production as a continuous process, but that has not been realized yet at a large scale (Ikeda 2003). Studies at a smaller scale, such as the optimization of lysine production in continuous culture under simultaneous phosphate and

The l-Lysine Story: From Metabolic Pathways to Industrial Production

carbon source limitation, however, show the potential of such approaches (Hirao et al. 1989). Instability of production strains is still a major problem in converting current batch or fed-batch processes into processes with extended runtime, but might be overcome in the near future through new insights into the underlying metabolic processes related to genetic instability or application of, for example, flow cytometry to quantify inoculum viability. Process optimization or process design can be additionally supported by process modelling. Process models do not only consider process stoichiometry via the mass streams into the process and operational parameters of the different unit operations, but also take environmental aspects into consideration by weighing the environmental burden of utilized and released components (Heinzle et al. 2006). In this regard, process modelling allows a detailed insight into the process, the estimation of process efficiency, and the identification of conditions for optimal yield, product titre, selectivity, or minimized environmental burden. Amongst various examples, this has been recently demonstrated for the case study of lysine production (Knoll and Büchs 2006). In this example, a process model of lysine production, related to the flow chart shown in Fig. 10, was established and implemented

Fig. 10 Process flow diagram of a lysine production plant. Picture taken from: Knoll A, Büchs J (2006) l-lysine—coupling of bioreaction and process model. In: Heinzle, Biwer, Cooney, (eds) Development of Sustainable Bioprocesses—Modelling and Assessment. Copyright: Wiley. Reproduced with permission

C. Wittmann · J. Becker

into a process modelling software. For the given process setup, coupled to a simple biological model of lysine production, the authors could nicely show that the operational conditions allowing minimum production costs are different from those providing maximum space-time yield. Such modelling approaches are not only interesting for existing processes, but can also support optimal development of new plants. Through variation of the process setup, for example the comparison of alternative down-stream processing routes, optimal process configurations can be identified in an early phase of development, when the degree of freedom is still high.

References Bartlett ATM, White PJ (1985) Species of Bacillus that make a vegetative peptidoglycan containing lysine lack diaminopimelate epimerase but have diaminopimelate dehydrogenase. J Gen Microb 131:2145–2152 Bathe B, Kalinowski J, Pühler A (1996) A physical and genetic map of the Corynebacterium glutamicum ATCC 13032 chromosome. Mol Gen Genet 252:255–265 Becker J, Klopprogge C, Zelder O, Heinzle E, Wittmann C (2005) Amplified expression of fructose 1,6-bisphosphatase in Corynebacterium glutamicum increases in vivo flux through the pentose phosphate pathway and lysine production on different carbon sources. Appl Environ Microbiol 71:8587–8596 Bellmann A, Vrljic M, Patek M, Sahm H, Krämer R, Eggeling L (2001) Expression control and specificity of the basic amino acid exporter LysE of Corynebacterium glutamicum. Microbiology 147:1765–1774 Bendt AK, Burkovski A, Schaffer S, Bott M, Farwick M, Hermann T (2003) Towards a phosphoproteome map of Corynebacterium glutamicum. Proteomics 3:1637–1646 Bonnassie S, Oreglia J, Sicard AM (1990) Nucleotide sequence of the dapA gene from Corynebacterium glutamicum. Nucleic Acids Res 18:6421 Born TL, Blanchard JS (1999) Structure/function studies on enzymes in the diaminopimelate pathway of bacterial cell wall biosynthesis. Curr Opin Chem Biol 3:607–613 Chatterjee SP, White PJ (1982) Activities and regulation of the enzymes of lysine biosynthesis in a lysine-excreting strain of Bacillus megaterium. J Gen Microb 128:1073–1081 Chen R, Yang H (2000) A highly specific monomeric isocitrate dehydrogenase from Corynebacterium glutamicum. Arch Biochem Biophys 383:238–245 Cocaign-Bousquet M, Lindley ND (1995) Pyruvate overflow and carbon flux within the central metabolic pathways of Corynebacterium glutamicum during growth on lactate. Enzyme Microb Technol 17:260–267 Cremer J, Eggeling L, Sahm H (1991) Control of the lysine biosynthesis sequence in Corynebacterium glutamicum as analyzed by overexpression of the individual corresponding genes. Appl Environ Microbiol 57(6):1746–1752 Cremer J, Treptow C, Eggeling L, Sahm H (1988) Regulation of enzymes of lysine biosynthesis in Corynebacterium glutamicum. J Gen Microbiol 134(Pt12):3221–3229 de Graaf AA (2000) Metabolic flux analysis of Corynebacterium glutamicum. In: Schügerl K, Bellgard KH (eds) Bioreaction engineering. Springer, Berlin Heidelberg New York, pp 506–555 de Graaf AA, Eggeling L, Sahm H (2001) Metabolic engineering for l-lysine production by Corynebacterium glutamicum. Adv Biochem Eng Biotechnol 73:9–29

The l-Lysine Story: From Metabolic Pathways to Industrial Production Dominguez H, Lindley ND (1996) Complete sucrose metabolism requires fructose phosphotransferase activity in Corynebacterium glutamicum to ensure phosphorylation of liberated fructose. Appl Environ Microbiol 62(10):3878–3880 Dominguez H, Rollin C, Guyonvarch A, Guerquin-Kern JL, Cocaign-Bousquet M, Lindley ND (1998) Carbon-flux distribution in the central metabolic pathways of Corynebacterium glutamicum during growth on fructose. Eur J Biochem 254:96– 102 Drysch A, El Massaoudi M, Mack C, Takors R, de Graaf AA, Sahm H (2003) Production process monitoring by serial mapping of microbial carbon flux distributions using a novel sensor reactor approach: 13 C-labeling-based metabolic flux analysis and llysine production. Metab Eng 5:96–107 Drysch A, El Massaoudi M, Wiechert W, de Graaf AA, Takors R (2004) Serial flux mapping of Corynebacterium glutamicum during fed-batch l-lysine production using the sensor reactor approach. Biotechnol Bioeng 85:497–505 Eggeling L, Bott M (2005) Handbook of Corynebacterium glutamicum. CRC Press, Boca Raton, Fl Eggeling L, Oberle S, Sahm H (1998) Improved l-lysine yield with Corynebacterium glutamicum: Use of dapA resulting in increased flux combined with growth limitation. Appl Microbiol Biotechnol 49:24–30 Eggeling L, Sahm H (1999) l-glutamate and l-lysine: Traditional products with impetuous developments. Appl Microbiol Biotechnol 52:146–153 Eikmanns BJ (2005) Central metabolism: Tricarboxylic acid cycle and anaplerotic reactions. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC Press, Boca Raton, Fl, pp 241–276 Eikmanns BJ, Follettie MT, Griot MU, Sinskey AJ (1989) The phosphoenolpyruvate carboxylase gene of Corynebacterium glutamicum: Molecular cloning, nucleotide sequence, and expression. Mol Gen Genet 218:330–339 Eikmanns BJ, Rittmann D, Sahm H (1995) Cloning, sequence analysis, expression, and inactivation of the Corynebacterium glutamicum icd gene encoding isocitrate dehydrogenase and biochemical characterization of the enzyme. J Bacteriol 177:774–782 El Massaoudi M, Spelthahn J, Drysch A, de Graaf A, Takors R (2003) Production process monitoring by serial mapping of microbial carbon flux distributions using a novel sensor reactor approach: I—sensor reactor system. Metab Eng 5:86–95 Follettie MT, Peoples OP, Agoropoulou C, Sinskey AJ (1993) Gene structure and expression of the Corynebacterium flavum N13 ask-asd operon. J Bacteriol 175:4096–4103 Georgi T, Rittmann D, Wendisch VF (2005) Lysine and glutamate production by Corynebacterium glutamicum on glucose, fructose and sucrose: Roles of malic enzyme and fructose-1,6-bisphosphatase. Metab Eng 7:291–301 Gerstmeir R, Wendisch VF, Schnicke S, Ruan H, Farwick M, Reinscheid D, Eikmanns BJ (2003) Acetate metabolism and its regulation in Corynebacterium glutamicum. J Biotechnol 104:99–122 Goodfellow M, Collins MD, Minnikin DE (1976) Thin-layer chromatographic analysis of mycolic acid and other long-chain components in whole-organism methanolysates of coryneform and related taxa. J Gen Microbiol 96:351–358 Gourdon P, Baucher MF, Lindley ND, Guyonvarch A (2000) Cloning of the malic enzyme gene from Corynebacterium glutamicum and role of the enzyme in lactate metabolism. Appl Environ Microbiol 66:2981–2987 Gunji Y, Yasueda H (2006) Enhancement of l-lysine production in methylotroph methylophilus methylotrophus by introducing a mutant LysE exporter. J Biotechnol 127:1–13

C. Wittmann · J. Becker Haberhauer G, Schröder H, Pompejus M, Zelder O, Kröger B (2001) Corynebacterium glutamicum genes encoding proteins involved in membrane synthesis and membrane transport. Patent WO 01/00805 Hartmann M, Tauch A, Eggeling L, Bathe B, Möckel B, Pühler A, Kalinowski J (2003) Identification and characterization of the last two unknown genes, dapC and dapF, in the succinylase branch of the l-lysine biosynthesis of Corynebacterium glutamicum. J Biotechnol 104:199–211 Hayashi M, Mizoguchi H, Ohnishi J, Mitsuhashi S, Yonetani Y, Hashimoto S, Ikeda M (2006a) A leuC mutation leading to increased l-lysine production and rel-independent global expression changes in Corynebacterium glutamicum. Appl Microbiol Biotechnol 72:783–789 Hayashi M, Mizoguchi H, Shiraishi N, Obayashi M, Nakagawa S, Imai J, Watanabe S, Ota T, Ikeda M (2002) Transcriptome analysis of acetate metabolism in Corynebacterium glutamicum using a newly developed metabolic array. Biosci Biotechnol Biochem 66:1337–1344 Hayashi M, Ohnishi J, Mitsuhashi S, Yonetani Y, Hashimoto S, Ikeda M (2006b) Transcriptome analysis reveals global expression changes in an industrial l-lysine producer of Corynebacterium glutamicum. Biosci Biotechnol Biochem 70:546–550 Heinzle E, Biwer A, Cooney CL (2006) Development of sustainable bioprocesses – modeling and assessment. Wiley, Chichester Hermann T (2003) Industrial production of amino acids by coryneform bacteria. J Biotechnol 104:155–172 Hermann T, Pfefferle W, Baumann C, Busker E, Schaffer S, Bott M, Sahm H, Dusch N, Kalinowski J, Pühler A, Bendt AK, Krämer R, Burkovski A (2001) Proteome analysis of Corynebacterium glutamicum. Electrophoresis 22:1712–1723 Hirao T, Nakano T, Azuma T, Sugimoto M, Nakanishi T (1989) l-lysine production in continuous culture of an l-lysine hyperproducing mutant of Corynebacterium glutamicum. Appl Microbiol Biotechnol 32:269–273 Hollander JA (1994) Potential metabolic limitations in lysine production by Corynebacterium glutamicum as revealed by metabolic network analysis. Appl Microbiol Biotechnol 42:508–515 Ihnen ED, Demain AL (1969) Glucose-6-phosphate dehydrogenase and its deficiency in mutants of Corynebacterium glutamicum. J Bacteriol 98:1151–1158 Ikeda M (2003) Amino acid production processes. Adv Biochem Eng Biotechnol 79: 1–35 Ikeda M, Nakagawa S (2003) The Corynebacterium glutamicum genome: Features and impacts on biotechnological processes. Appl Microbiol Biotechnol 62:99–109 Imaizumi A, Kojima H, Matsui K (2006) The effect of intracellular ppGpp levels on glutamate and lysine overproduction in Escherichia coli. J Biotechnol 125:328–337 Imaizumi A, Takikawa R, Koseki C, Usuda Y, Yasueda H, Kojima H, Matsui K, Sugimoto S (2005) Improved production of l-lysine by disruption of stationary phase-specific rmf gene in Escherichia coli. J Biotechnol 117:111–118 Jetten M, Sinskey AJ (1993) Characterization of phosphoenolpyruvate carboxykinase from Corynebacterium glutamicum. FEMS Microbiol Lett 111:183–188 Jetten MS, Follettie MT, Sinskey AJ (1995) Effect of different levels of aspartokinase on the lysine production by Corynebacterium lactofermentum. Appl Microbiol Biotechnol 43:76–82 Jetten MS, Pitoc GA, Follettie MT, Sinskey AJ (1994) Regulation of phospho(enol)pyruvate- and oxaloacetate-converting enzymes in Corynebacterium glutamicum. Appl Microbiol Biotechnol 41:47–52

The l-Lysine Story: From Metabolic Pathways to Industrial Production Jetten MS, Sinskey AJ (1995) Purification and properties of oxaloacetate decarboxylase from Corynebacterium glutamicum. Antonie Van Leeuwenhoek 67:221–227 Kalinowski J, Bachmann B, Thierbach G, Pühler A (1990) Aspartokinase genes lysC alpha and lysC beta overlap and are adjacent to the aspartate beta-semialdehyde dehydrogenase gene asd in Corynebacterium glutamicum. Mol Gen Genet 224:317–324 Kalinowski J, Bathe B, Bartels D, Bischoff N, Bott M, Burkovski A, Dusch N, Eggeling L, Eikmanns BJ, Gaigalat L, Goesmann A, Hartmann M, Huthmacher K, Krämer R, Linke B, McHardy AC, Meyer F, Möckel B, Pfefferle W, Pühler A, Rey DA, Rückert C, Rupp O, Sahm H, Wendisch VF, Wiegrabe I, Tauch A (2003) The complete Corynebacterium glutamicum ATCC 13032 genome sequence and its impact on the production of l-aspartate-derived amino acids and vitamins. J Biotechnol 104:5–25 Kalinowski J, Cremer J, Bachmann B, Eggeling L, Sahm H, Pühler A (1991) Genetic and biochemical analysis of the aspartokinase from Corynebacterium glutamicum. Mol Microbiol 5:1197–1204 Kawahara Y, Yoshihara Y, Ikeda S, Yoshii H, Hirose Y (1990) Stimulatory effect of glycine betaine on l-lysine fermentation. Appl Microbiol Biotechnol 34:87–90 Kelle R, Hermann T, Bathe B (2005) l-lysine production. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC Press, Boca Raton, Fl, pp 465– 488 Kiefer P, Heinzle E, Wittmann C (2002) Influence of glucose, fructose and sucrose as carbon sources on kinetics and stoichiometry of lysine production by Corynebacterium glutamicum. J Ind Microbiol Biotechnol 28:338–343 Kiefer P, Heinzle E, Zelder O, Wittmann C (2004) Comparative metabolic flux analysis of lysine-producing Corynebacterium glutamicum cultured on glucose or fructose. Appl Environ Microbiol 70:229–239 Kim HM, Heinzle E, Wittmann C (2006) Deregulation of aspartokinase by single nucleotide exchange leads to global flux rearrangement in the central metabolism of Corynebacterium glutamicum. J Microbiol Biotechnol 8:1174–1179 Kinoshita S, Nakayama K, Kitada S (1961) Method of producing l-lysine by fermentation. In: Office USP (ed) US Patent 2 979 439 Kinoshita S, Shigezo U, Shimono M (1957) Studies on the amino acid fermentation, Part I. Production of l-glutamic acid by various microorganisms. J Gen Appl Microbiol 3:193–205 Kiss RD, Stephanopoulos G (1992) Metabolic characterization of a l-lysine-producing strain by continuous culture. Biotechnol Bioeng 39:565–574 Knoll A, Büchs J (2006) l-lysine – coupling of bioreaction and process model. In: Heinzle E, Biwer A, Cooney CL (eds) Development of sustainable bioprocesses – modeling and assessment. Wiley, Chichester, pp 155–168 Krömer JO, Sorgenfrei O, Klopprogge K, Heinzle E, Wittmann C (2004) In-depth profiling of lysine-producing Corynebacterium glutamicum by combined analysis of the transcriptome, metabolome, and fluxome. J Bacteriol 186:1769–1784 Krömer JO, Wittmann C, Schröder H, Heinzle E (2006) Metabolic pathway analysis for rational design of l-methionine production by Escherichia coli and Corynebacterium glutamicum. Metab Eng 8:353–369 Lange C, Rittmann D, Wendisch VF, Bott M, Sahm H (2003) Global expression profiling and physiological characterization of Corynebacterium glutamicum grown in the presence of l-valine. Appl Environ Microbiol 69:2521–2532 Leuchtenberger W (1996) Amino acids—technical production and use. In: Rehm HJ, Reed G, Pühler A, Stadler P (eds) Biotechnology, Vol 6. VCH, Weinheim, Germany, pp 465–502

C. Wittmann · J. Becker Liebl W (2005) Corynebacterium taxonomy. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC, Boca Raton, Fl, pp 9–34 Liebl W, Ehrmann M, Ludwig W, Schleifer KH (1991) Transfer of Brevibacterium divaricatum DSM 20297t, Brevibacterium flavum DSM 20411, Brevibacterium lactofermentum DSM 20412 and DSM 1412, and Corynebacterium glutamicum and their distinction by rRNA gene restriction patterns. Int J Syst Bacteriol 41:255–260 Malin GM, Bourd GI (1991) Phosphotransferase-dependent glucose transport in Corynebacterium glutamicum. J Appl Bacteriol 71:517–523 Malumbres M, Martin JF (1996) Molecular control mechanisms of lysine and threonine biosynthesis in amino acid-producing Corynebacteria: Redirecting carbon flow. FEMS Microbiol Lett 143:103–114 Marx A, de Graaf A, Wiechert W, Eggeling L, Sahm H (1996) Determination of the fluxes in the central metabolism of Corynebacterium glutamicum by nuclear magnetic resonance spectroscopy combined with metabolite balancing. Biotechnol Bioeng 49(2):111–129 Marx A, Eikmanns BJ, Sahm H, de Graaf AA, Eggeling L (1999) Response of the central metabolism in Corynebacterium glutamicum to the use of an NADH-dependent glutamate dehydrogenase. Metab Eng 1:35–48 Marx A, Hans S, Möckel B, Bathe B, de Graaf AA (2003) Metabolic phenotype of phosphoglucose isomerase mutants of Corynebacterium glutamicum. J Biotechnol 104:185– 197 Marx A, Striegel K, de Graaf A, Sahm H, Eggeling L (1997) Response of the central metabolism of Corynebacterium glutamicum to different flux burdens. Biotechnol Bioeng 56(2):168–180 Matsushita K, Otofuji A, Iwahashi M, Toyama H, Adachi O (2001) NADH dehydrogenase of Corynebacterium glutamicum. Purification of an NADH dehydrogenase II homolog able to oxidize NADPH. FEMS Microbiol Lett 204:271–276 McCoy AJ, Adams NE, Hudson AO, Gilvarg C, Leustek T, Maurelli AT (2006) l,l-diaminopimelate aminotransferase, a trans-kingdom enzyme shared by Chlamydia and plants for synthesis of diaminopimelate/lysine. Proc Natl Acad Sci USA 103:17909–17914 Michal G (1999) Biochemical pathways. Wiley, Chichester Molenaar D, van der Rest ME, Drysch A, Yucel R (2000) Functions of the membraneassociated and cytoplasmic malate dehydrogenases in the citric acid cycle of Corynebacterium glutamicum. J Bacteriol 182:6884–6891 Moon MW, Kim HJ, Oh TK, Shin CS, Lee JS, Kim SJ, Lee JK (2005) Analyses of enzyme II gene mutants for sugar transport and heterologous expression of fructosekinase gene in Corynebacterium glutamicum ATCC 13032. FEMS Microbiol Lett 244:259–266 Moritz B, Striegel K, De Graaf AA, Sahm H (2000) Kinetic properties of the glucose6-phosphate and 6-phosphogluconate dehydrogenases from Corynebacterium glutamicum and their application for predicting pentose phosphate pathway flux in vivo. Eur J Biochem 267:3442–3452 Muffler A, Bettermann S, Haushalter M, Horlein A, Neveling U, Schramm M, Sorgenfrei O (2002) Genome-wide transcription profiling of Corynebacterium glutamicum after heat shock and during growth on acetate and glucose. J Biotechnol 98:255–268 Nakayama K, Araki K (1973) Process for producing l-lysine. US Patent 3 708 395 Nakayama K, Araki K, Kase H (1978) Microbial production of essential amino acid with Corynebacterium glutamicum mutants. Adv Exp Med Biol 105:649–661 Netzer R, Krause M, Rittmann D, Peters-Wendisch P, Eggeling L, Wendisch VF, Sahm H (2004) Roles of pyruvate kinase and malic enzyme in Corynebacterium glutamicum for growth on carbon sources requiring gluconeogenesis. Arch Microbiol 182:354–363

The l-Lysine Story: From Metabolic Pathways to Industrial Production Nöh K, Wahl A, Wiechert W (2006) Computational tools for isotopically instationary 13 C labeling experiments under metabolic steady state conditions. Metab Eng 8:554–577 Nöh K, Wiechert W (2006) Experimental design principles for isotopically instationary 13 C labeling experiments. Biotechnol Bioeng 94:234–251 O’Regan M, Thierbach G, Bachmann B, Villeval D, Lepage P, Viret JF, Lemoine Y (1989) Cloning and nucleotide sequence of the phosphoenolpyruvate carboxylase-coding gene of Corynebacterium glutamicum ATCC 13032. Gene 77:237–251 Ohnishi J, Katahira R, Mitsuhashi S, Kakita S, Ikeda M (2005) A novel gnd mutation leading to increased l-lysine production in Corynebacterium glutamicum. FEMS Microbiol Lett 242:265–274 Ohnishi J, Mitsuhashi S, Hayashi M, Ando S, Yokoi H, Ochiai K, Ikeda M (2002) A novel methodology employing Corynebacterium glutamicum genome information to generate a new l-lysine-producing mutant. Appl Microbiol Biotechnol 58:217–223 Ozaki H, Shiio I (1969) Regulation of the TCA and glyoxylate cycles in Brevibacterium flavum. II. Regulation of phosphoenolpyruvate carboxylase and pyruvate kinase. J Biochem (Tokyo) 66:297–311 Park SM, Shaw-Reid C, Sinskey AJ, Stephanopoulos G (1997) Elucidation of anaplerotic pathways in Corynebacterium glutamicum via 13 C-NMR spectroscopy and GC-MS. Appl Microbiol Biotechnol 47:430–440 Patek M, Bilic M, Krumbach K, Eikmanns B, Sahm H, Eggeling L (1997) Identification and transcriptional analysis of the dapB-orf2-dapA-orf4 operon of Corynebacterium glutamicum, encoding two enzymes involved in l-lysine synthesis. Biotechnol Letters 19(11):1113–1117 Peters-Wendisch PG, Kreutzer C, Kalinowski J, Patek M, Sahm H, Eikmanns BJ (1998) Pyruvate carboxylase from Corynebacterium glutamicum: Characterization, expression and inactivation of the pyc gene. Microbiology 144(4):915–927 Peters-Wendisch PG, Schiel B, Wendisch VF, Katsoulidis E, Möckel B, Sahm H, Eikmanns BJ (2001) Pyruvate carboxylase is a major bottleneck for glutamate and lysine production by Corynebacterium glutamicum. J Mol Microbiol Biotechnol 3:295– 300 Peters-Wendisch PG, Wendisch VF, de Graaf AA, Eikmanns BJ, Sahm H (1996) C3 -carboxylation as an anaplerotic reaction in phosphoenolpyruvate carboxylase-deficient Corynebacterium glutamicum. Arch Microbiol 165:387–396 Petersen S, de Graaf AA, Eggeling L, Mollney M, Wiechert W, Sahm H (2000) In vivo quantification of parallel and bidirectional fluxes in the anaplerosis of Corynebacterium glutamicum. J Biol Chem 275:35932–35941 Petersen S, Mack C, de Graaf AA, Riedel C, Eikmanns BJ, Sahm H (2001) Metabolic consequences of altered phosphoenolpyruvate carboxykinase activity in Corynebacterium glutamicum reveal anaplerotic regulation mechanisms in vivo. Metab Eng 3:344– 361 Pfefferle W, Möckel B, Bathe B, Marx A (2003) Biotechnological manufacture of lysine. Adv Biochem Eng Biotechnol 79:59–112 Pisabarro A, Malumbres M, Mateos LM, Oguiza JA, Martin JF (1993) A cluster of three genes (dapA, orf2, and dapB) of Brevibacterium lactofermentum encodes dihydrodipicolinate synthase, dihydrodipicolinate reductase, and a third polypeptide of unknown function. J Bacteriol 175:2743–2749 Riedel C, Rittmann D, Dangel P, Möckel B, Petersen S, Sahm H, Eikmanns BJ (2001) Characterization of the phosphoenolpyruvate carboxykinase gene from Corynebacterium glutamicum and significance of the enzyme for growth and amino acid production. J Mol Microbiol Biotechnol 3:573–583

C. Wittmann · J. Becker Sahm H, Eggeling L, de Graaf AA (2000) Pathway analysis and metabolic engineering in Corynebacterium glutamicum. Biol Chem 381:899–910 Sano K, Ito K, Miwa K, Nakamori S (1987) Amplification of the phosphoenolpyruvate carboxylase gene of Brevibacterium lactofermentum to improve amino acid production. Agric Biol Chem 51(2):597–599 Sauer U (2004) High-throughput phenomics: Experimental methods for mapping fluxomes. Curr Opin Biotechnol 15:58–63 Sauer U, Eikmanns BJ (2005) The PEP-pyruvate-oxaloacetate node as the switch point for carbon flux distribution in bacteria. FEMS Microbiol Rev 29:765–794 Sauer U, Hatzimanikatis V, Bailey JE, Hochuli M, Szyperski T, Wüthrich K (1997) Metabolic fluxes in riboflavin-producing Bacillus subtilis. Nat Biotechnol 15:448–452 Schaffer S, Burkovski A (2005) Proteomics. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC, Boca Raton, Fl, pp 99–118 Schilling CH, Schuster S, Palsson BO, Heinrich R (1999) Metabolic pathway analysis: Basic concepts and scientific applications in the post-genomic era. Biotechnol Prog 15:296– 303 Schluesener D, Fischer F, Kruip J, Rogner M, Poetsch A (2005) Mapping the membrane proteome of Corynebacterium glutamicum. Proteomics 5:1317–1330 Schmid R, Uhlemann EM, Nolden L, Wersch G, Hecker R, Hermann T, Marx A, Burkovski A (2000) Response to nitrogen starvation in Corynebacterium glutamicum. FEMS Microbiol Lett 187:83–88 Schrumpf B, Eggeling L, Sahm H (1992) Isolation and prominent characteristics of an l-lysine hyperproducing strain of Corynebacterium glutamicum. Appl Microbiol Biotechnol 37:566–571 Schrumpf B, Schwarzer A, Kalinowski J, Pühler A, Eggeling L, Sahm H (1991) A functionally split pathway for lysine synthesis in Corynebacterium glutamicium. J Bacteriol 173:4510–4516 Schuster S, Hilgetag C, Woods JH, Fell DA (2002) Reaction routes in biochemical reaction systems: Algebraic properties, validated calculation procedure and example from nucleotide metabolism. J Math Biol 45:153–181 Silberbach M, Schäfer M, Huser AT, Kalinowski J, Pühler A, Krämer R, Burkovski A (2005) Adaptation of Corynebacterium glutamicum to ammonium limitation: A global analysis using transcriptome and proteome techniques. Appl Environ Microbiol 71:2391– 2402 Sonntag K, Eggeling L, De Graaf AA, Sahm H (1993) Flux partitioning in the split pathway of lysine synthesis in Corynebacterium glutamicum. Quantification by 13 C- and 1 H-NMR spectroscopy. Eur J Biochem 213:1325–1331 Sonntag K, Schwinde J, de Graaf A, Marx A, Eikmanns B, Wiechert W, Sahm H (1995) 13 C NMR studies of the fluxes in the central metabolism of Corynebacterium glutamicum during growth and overproduction of amino acids in batch cultures. Appl Microbiol Biotechnol 44:489–495 Sugimoto M, Tanaka A, Suzuki T, Matsui H, Nakamori S, Takagi H (1997) Sequence analysis of functional regions of homoserine dehydrogenase genes from l-lysine and l-threonine-producing mutants of Brevibacterium lactofermentum. Biosci Biotechnol Biochem 61:1760–1762 Tauch A, Homann I, Mormann S, Ruberg S, Billault A, Bathe B, Brand S, BrockmannGretza O, Rückert C, Schischka N, Wrenger C, Hoheisel J, Möckel B, Huthmacher K, Pfefferle W, Pühler A, Kalinowski J (2002) Strategy to sequence the genome of Corynebacterium glutamicum ATCC 13032: Use of a cosmid and a bacterial artificial chromosome library. J Biotechnol 95:25–38

The l-Lysine Story: From Metabolic Pathways to Industrial Production Tauch A, Kaiser O, Hain T, Goesmann A, Weisshaar B, Albersmeier A, Bekel T, Bischoff N, Brune I, Chakraborty T, Kalinowski J, Meyer F, Rupp O, Schneiker S, Viehoever P, Pühler A (2005) Complete genome sequence and analysis of the multiresistant nosocomial pathogen Corynebacterium jeikeium K411, a lipid-requiring bacterium of the human skin flora. J Bacteriol 187:4671–4682 Thierbach G, Kalinowski J, Bachmann B, Pühler A (1990) Cloning of a DNA fragment from Corynebacterium glutamicum conferring aminoethyl cysteine resistance and feedback resistance to aspartokinase. Appl Microbiol Biotechnol 32:443–448 Udaka S (1960) Screening method for microorganisms accumulating metabolites and its use in the isolation of Micrococcus glutamicus. J Bacteriol 79:754–755 Vasicova P, Patek M, Nesvera J, Sahm H, Eikmanns B (1999) Analysis of the Corynebacterium glutamicum dapA promoter. J Bacteriol 181:6188–6191 Velasco AM, Leguina JI, Lazcano A (2002) Molecular Evolution of the Lysine Biosynthetic Pathways. J Mol Evol 55:445–459 Vrljic M, Garg J, Bellmann A, Wachi S, Freudl R, Malecki MJ, Sahm H, Kozina VJ, Eggeling L, Saier MH Jr, Eggeling L, Saier MH Jr (1999) The LysE superfamily: Topology of the lysine exporter LysE of Corynebacterium glutamicum, a paradyme for a novel superfamily of transmembrane solute translocators. J Mol Microbiol Biotechnol 1:327–336 Vrljic M, Sahm H, Eggeling L (1996) A new type of transporter with a new type of cellular function: l-lysine export from Corynebacterium glutamicum. Mol Microbiol 22:815– 826 Wehrmann A, Phillipp B, Sahm H, Eggeling L (1998) Different modes of diaminopimelate synthesis and their role in cell wall integrity: A study with Corynebacterium glutamicum. J Bacteriol 180:3159–3165 Weinberger S, Gilvarg C (1970) Bacterial distribution of the use of succinyl and acetyl blocking groups in diaminopimelic acid biosynthesis. J Bacteriol 101:323–324 Wendisch VF (2003) Genome-wide expression analysis in Corynebacterium glutamicum using DNA microarrays. J Biotechnol 104:273–285 Wendisch VF, Bott M, Kalinowski J, Oldiges M, Wiechert W (2006) Emerging Corynebacterium glutamicum systems biology. J Biotechnol 124:74–92 Wendisch VF, de Graaf AA, Sahm H, Eikmanns BJ (2000) Quantitative determination of metabolic fluxes during coutilization of two carbon sources: Comparative analyses with Corynebacterium glutamicum during growth on acetate and/or glucose. J Bacteriol 182:3088–3096 White PJ (1983) The essential role of diaminopimelate dehydrogenase in the biosynthesis of lysine by Bacillus spaericus. J Gen Microb 129:739–749 Wiechert W, Nöh K (2005) From stationary to instationary metabolic flux analysis. Adv Biochem Eng Biotechnol 92:145–172 Wittmann C, de Graaf A (2005) Metabolic flux analysis in Corynebacterium glutamicum. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC, Boca Raton, Fl, pp 277–304 Wittmann C, Heinzle E (2001a) Application of MALDI-TOF MS to lysine-producing Corynebacterium glutamicum: A novel approach for metabolic flux analysis. Eur J Biochem 268:2441–2455 Wittmann C, Heinzle E (2001b) MALDI-TOF MS for quantification of substrates and products in cultivations of Corynebacterium glutamicum. Biotechnol Bioeng 72:642–647 Wittmann C, Heinzle E (2002) Genealogy profiling through strain improvement by using metabolic network analysis: Metabolic flux genealogy of several generations of lysine-producing corynebacteria. Appl Environ Microbiol 68:5843– 5859

C. Wittmann · J. Becker Wittmann C, Kiefer P, Zelder O (2004a) Metabolic fluxes in Corynebacterium glutamicum during lysine production with sucrose as carbon source. Appl Environ Microbiol 70:7277–7287 Wittmann C, Kim HM, Heinzle E (2004b) Metabolic network analysis of lysine producing Corynebacterium glutamicum at a miniaturized scale. Biotechnol Bioeng 87:1–6 Yang TH, Wittmann C, Heinzle E (2003) Dynamic calibration and dissolved gas analysis using membrane inlet mass spectrometry for the quantification of cell respiration. Rapid Commun Mass Spectrom 17:2721–2731 Yang TH, Wittmann C, Heinzle E (2006) Respirometric 13 C flux analysis, part I: Design, construction and validation of a novel multiple reactor system using on-line membrane inlet mass spectrometry. Metab Eng 8:417–431 Yang TH, Wittmann C, Heinzle E (2006) Respirometric 13 C flux analysis, part II. In vivo flux estimation of lysine-producing Corynebacterium glutamicum. Metab Eng 8:432– 446 Yokota A, Lindley ND (2005) Central metabolism: Sugar uptake and conversion. In: Eggeling L, Bott M (eds) Handbook of Corynebacterium glutamicum. CRC, Boca Raton, Fl, pp 215–240 Zelder O, Pompejus M, Schröder H, Kröger B, Klopprogge C, Haberhauer G (2005) Genes coding for glucose-6-phosphate-dehydrogenase proteins. US Patent 20 050 014 235 A1

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

PERSPECTIVE

Production of isoprenoid pharmaceuticals by engineered microbes Michelle C Y Chang & Jay D Keasling Throughout human history, natural products have been the foundation for the discovery and development of therapeutics used to treat diseases ranging from cardiovascular disease to cancer. Their chemical diversity and complexity have provided structural scaffolds for small-molecule drugs and have consistently served as inspiration for medicinal design. However, the chemical complexity of natural products also presents one of the main roadblocks for production of these pharmaceuticals on an industrial scale. Chemical synthesis of natural products is often difficult and expensive, and isolation from their natural sources is also typically low yielding. Synthetic biology and metabolic engineering offer an alternative approach that is becoming more accessible as the tools for engineering microbes are further developed. By reconstructing heterologous metabolic pathways in genetically tractable host organisms, complex natural products can be produced from inexpensive sugar starting materials through large-scale fermentation processes. In this Perspective, we discuss ongoing research aimed toward the production of terpenoid natural products in genetically engineered Escherichia coli and Saccharomyces cerevisiae. Since the dawn of civilization, natural products have had an essential role in the treatment of human health and disease1. Many drugs still used in modern medicine, such as vinblastine (Madagascar periwinkle), digitalis (purple foxglove) and codeine (opium poppy), are derived from herbal remedies and are used for diverse purposes such as treatment of cancer, heart conditions and pain, respectively. More recently, the difficulties inherent in screening, isolating and identifying these low-abundance secondary metabolites have led to a change in focus from natural-product mining to newer technologies such as combinatorial chemistry and high-throughput screening. However, interest in natural-products research has resurged with the growing realization that the structural complexity and diversity that is key to the biological activity of natural products is not easily

Michelle C.Y. Chang and Jay D. Keasling are in the California Institute for Quantitative Biomedical Research, University of California Berkeley, Berkeley, California 94720-3224, USA. Jay D. Keasling is also in the Department of Chemical Engineering, Department of Bioengineering, University of California Berkeley, Berkeley, California 94720, USA and the Physical Bioscience Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA. e-mail: [email protected] Published online 15 November 2006; doi:10.1038/nchembio836

NATURE CHEMICAL BIOLOGY

ADVANCE ONLINE PUBLICATION

reproduced in de novo synthetic libraries2. In fact, 61% of the 877 small-molecule drugs introduced in the last 20 years can trace their ancestry to natural products, with 38% comprising natural products and their close derivatives3. Revitalization of bioprospecting as an important conduit for drug discovery leads to the question of how these complex molecules can be produced on an industrial scale. Chemical synthesis and modification of natural products can be extremely difficult and problematic because of the very chemical complexity that is so important for high affinity and selectivity toward biomolecular targets. Correspondingly, extraction of these molecules is often low-yielding and inefficient, and requires substantial expenditure of natural resources. For example, it would take approximately six 100-year-old Pacific yew trees to provide enough Taxol (paclitaxel) to treat one cancer patient4. Furthermore, as the search for new compounds takes us toward unmined biodiversity, consumption of limited and rare natural resources is becoming a larger issue. Indeed, for cases in which compounds have been isolated from exotic sources such as soft coral or unculturable microbes, there may not even exist enough raw material to provide the target for clinical trials. Moreover, the availability of compounds obtained through extraction of biological sources is difficult to quickly scale to changing demands, and it is dependent on unpredictable fluxes in environmental conditions or external circumstances. In one striking example this past year, seed shortages caused massive shortfalls of the antimalarial drug artemisinin, thereby limiting its distribution in third-world nations because no other sources were available5. Synthetic biology provides an alternative to traditional methods that are centered on extraction of bioactive compounds from their natural source or on chemical synthesis6. In this approach, the genes related to the biosynthetic pathway for a target natural product are transplanted from the natural host into a genetically tractable host system such as E. coli or S. cerevisiae6 (Fig. 1). This strategy offers advantages over traditional methods in scalability, consumption of natural resources, and use of more environmentally friendly ‘green’ chemistry. Another one of the primary goals in this area is to provide fine chemicals and pharmaceuticals at a significantly lower cost by scalable fermentation processes using microbes to carry out the biosynthesis of valuable small molecules from inexpensive sugarbased carbon sources. By combining process optimization and host genetic engineering aimed at generating high product titers, the hope is that therapeutics that are currently too expensive to be generally accessible can eventually be available to all those in need. Microbes have been engineered throughout the ages to carry out biosynthetic processes such as production of ethanol (wine and beer)

1

PERSPECTIVE of target compounds. We will then discuss efforts to implement downstream pathways for the biosynthesis and production of pharmaceutical precursors.

Transplantation of biosynthetic pathway

Carbon source

Fermentation Pharmaceutical or precursor

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

Microbial host Native host

Figure 1 Transplantation of a biosynthetic pathway from the natural host into a microbial host. The genes related to the biosynthetic pathway in the native host are imported heterologously into a microbial host. Growth on an inexpensive sugar carbon source allows the production of the small-molecule pharmaceutical or precursor by fermentation.

and lactic acid (cheese). Currently, many biochemicals including amino acids and vitamins are produced through microbial fermentation7,8. In addition, many natural products, such as the wide spectrum of antibiotics made by microbes including Streptomyces spp., Bacillus spp. and Aspergillus spp., are produced in their natural host strains, which have been mutated or engineered for increased production of the compound of interest7,8. However, these products are examples in which endogenous biosynthetic pathways have been optimized but do not involve using exogenous biosynthetic pathways from foreign species. Only more recently have researchers tackled the problem of importing pathways from other organisms into microbial hosts such as E. coli and S. cerevisiae for production of xenobiotic small molecules. One early achievement in the early 1980s was the production of indigo dye in E. coli by expressing a naphthalene dioxygenase enzyme9,10. Other successful examples include the production of polyhydroxylalkanoates11–13 and 1,3-propanediol14 in E. coli. The latter project was accomplished on an industrial scale in a collaborative effort between DuPont and Genencor International and highlights the significant resources and lead time required to successfully produce a target compound at the titers needed to make microbial fermentation economically feasible. With regard to pharmaceutical production, metabolic engineering has proven to be highly successful in producing polyketide natural products in hosts both with and without native polyketide biosynthesis machinery15–19. In E. coli, titers of 700 mg l–1 for production of the erythromycin core have been achieved in high-density fermentations20,21. S. cerevisiae has also shown potential as a platform host for production of simple polyketide products22,23. In this Perspective, we will discuss progress in genetic engineering of E. coli and S. cerevisiae for producing pharmaceuticals from the isoprenoid family of natural products. Isoprenoids are of particular interest for microbial production because they exist at low concentrations in their host organism and are structurally complex with multiple chiral centers, which makes them difficult subjects for both biological extraction and chemical synthesis. In addition, they are biosynthesized in a modular fashion, which allows us to build microbial production platforms that can be generalized to many targets. We will first focus on the metabolic engineering of E. coli and S. cerevisiae to channel flux through the isoprenoid pathway and provide sufficient precursors for the biosynthesis of high titers

2

Isoprenoids Isoprenoids comprise the largest class of natural products, encompassing 450,000 known compounds with an extremely diverse array of chemical structures24. They are traditionally valued as flavors and fragrances, as they are the primary constituents of the essential oils of plants and flowers (for example, b-damascenone (rose), limonene (citrus) and zingiberene (ginger)). However, isoprenoids are also now known for their many varied biological functions, such as their use as hormones (steroids, gibberellins and abscisic acid) and their roles in membrane fluidity maintenance (steroids), respiration (quinones), photosynthetic light harvesting (carotenoids), and protein targeting and regulation (prenylation and glycosylation). Although all organisms use isoprenoids for these basic cellular processes, it is in the plant kingdom that the structure and function of isoprenoids has most significantly evolved and diversified. In plants, isoprenoids have many essential roles in specialized processes such as defense, communication, pollination attraction and seed dispersion in addition to their involvement in primary growth and development25. It is among these secondary metabolites that many effective and promising pharmaceuticals, such as Taxol, vinblastine, artemisinin and prostratin, have been discovered and isolated. Biosynthesis of isoprenoids can be divided into four basic steps26: (i) production of the C5 monomers isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), (ii) polymerization to form the linear prenyl diphosphate precursors, (iii) cyclization and/or rearrangement to form the initial terpene carbon backbone and (iv) tailoring and modification of the terpene skeleton (Scheme 1). Biosynthesis of the IPP and DMAPP building blocks in the first phase takes place via two distinct pathways, the mevalonate pathway27 (Scheme 2a) and the more recently discovered 1-deoxylulose-5-phosphate (DXP) or methylerythritol phosphate (MEP) pathway28,29 (Scheme 2b). With some exceptions, the mevalonate pathway is primarily found in eukaryotes, whereas the DXP pathway is typically found in prokaryotes and the plastids of photosynthetic organisms30. DMAPP can be produced by isomerization from IPP or, in the case of E. coli, either by isomerization or directly by a branch-point reaction from the precursor to IPP31. In the next phase, DMAPP primes the sequential head-to-tail condensations of IPP molecules by prenyltransferases to form the linear prenyl diphosphate precursors geranyl diphosphate (GPP, C10, monoterpenoids), farnesyl diphosphate (FPP, C15, sesquiterpenoids) and geranylgeranyl diphosphate (GGPP, C20, diterpenoids), as well as larger units. Cyclizations and/or rearrangements of these linear precursors are catalyzed by terpene synthases to generate the many diverse carbon skeletons found in terpenoid natural products. These skeletons are then extensively decorated in multiple steps to form the final bioactive chemical structures. Increasing precursor flux for isoprenoid production The initial step toward building a platform in either E. coli or S. cerevisiae for high-level isoprenoid production requires engineering a strain with high potential for generating the universal IPP monomer. In this way, a pathway for IPP production from redirected cellular metabolites can be optimized while the downstream branch-point enzymes, such as cyclases or tailoring enzymes, are added in a modular fashion to generate different target molecules. The existence of two pathways for IPP biosynthesis allows for consideration of

ADVANCE ONLINE PUBLICATION

NATURE CHEMICAL BIOLOGY

PERSPECTIVE limitations for isoprenoid production in the direct carbon pathway to IPP. In the DXP IPP DMAPP pathway, glyceraldehyde-3-phosphate (G3P) 2 Prenyltransferase and pyruvate are both the immediate precursors to IPP biosynthesis; they are also central metabolites involved in several pathways (glycolysis, gluconeogensis, the tricarboxylic acid n OPP cycle and the pentose phosphate pathway) GPP FPP GGPP and thus subject to many layers of regulation. n=1 n=2 n=3 A combination of feeding and gene expression 3 Terpene synthase and deletion experiments has shown that isoprenoid production by the DXP pathway H is also controlled by the G3P/pyruvate ratio— not simply by substrate availibility38. On the basis of this discovery, dynamic control of H H H isoprenoid production using a glycolytic flux Taxadiene Limonene Amorphadiene sensor has been shown to increase carotenoid O 4 Tailoring enzymes AcO titers by 50 fold39. O NH O H Ph OH Though considerable improvements in isoO O Ph prenoid production were observed in the O O foregoing studies of the native E. coli DXP H OH H HO O O O AcO Ph pathway, it seems that flux through this O O pathway remains limited by intracellular Menthol Taxol Artemisinin regulatory control. Thus, we chose to bypass Scheme 1 Isoprenoid biosynthesis. (1) C5 monomers IPP and DMAPP are synthesized from primary the DXP pathway and instead import metabolites. (2) Head-to-tail condensations of IPP with a DMAPP starter unit generate the linear prenyl the heterologous mevalonate pathway from diphosphate precursors GPP, FPP and GGPP. (3) The terpene synthase catalyzes the cyclization or S. cerevisiae40 to supplement endogenous IPP rearrangement to the parent carbon skeleton with loss of pyrophosphate. (4) Tailoring enzymes biosynthesis41. The mevalonate pathway was elaborate the parental backbone to generate the final product. Me, methyl; Ac, Acetyl; Ph, Phenyl. divided into two synthetic operons, referred to as the top (MevT) and bottom (MBIS, for complementary approaches in both E. coli and S. cerevisiae platforms. MevB, idi and ispA) sections of the pathway. MevT is responsible for The first approach is to import a heterologous pathway to supplement the three-step conversion of acetyl coenzyme A (acetyl-CoA) to the native pathway; the alternative is to alter the metabolic flux and mevalonate, whereas MBIS takes mevalonate to FPP (Scheme 2a). regulation of the native pathway. Both approaches have their unique Initial work showed that expression of the mevalonate pathway in advantages and disadvantages. On the one hand, metabolites that are E. coli leads to severe growth inhibition, and this was attributed to foreign to the cell may be easier to produce at high yield because of the toxicity resulting from accumulation of the prenyl diphosphates (FPP, absence of feedback regulation and alternative dissipative pathways, IPP and/or DMAPP)40. Coexpression of a codon-optimized but they also have possibilities for toxicity that are difficult to predict terpene synthase42 from the artemisinin pathway, amorpha-4, or understand. In addition, the host may not contain the native 11-diene (amorphadiene) synthase (ADS)43,44, alleviates this toxicity machinery either for precursor biosynthesis or for managing key and can be used to evaluate IPP and DMAPP production as the heterologous enzymes (for example, post-translational modification downstream amorphadiene product. Production of amorphadiene or subcellular targeting) from a non-native pathway. On the other from the imported mevalonate pathway was found to be substantially hand, high flux through the desired pathway can be much more higher than production with a supplemental DXP pathway40, yielding difficult to achieve when endogenous metabolites intersect multiple titers of 280 mg l–1 after 24 h and 480 mg l–1 in a fed-batch pathways in the host network and are subject to strong regulatory bioreactor45. Closer examination of the mevalonate pathway suggests checkpoints. Toxicity resulting from either high concentrations or that availability of mevalonate may limit amorphadiene production46. depletions of native intracellular metabolites is also possible through However, when MevT expression is increased, small molecule– cytotoxicity or loss of resources needed for cell growth. dependent growth inhibition of E. coli is observed; this result has Both approaches have been used successfully to increase iso- been attributed to hydroxymethylglutaryl-CoA (HMG-CoA) toxiprenoid production in E. coli, which normally uses the DXP pathway city46. Expression of an additional copy of HMGR alleviates buildup (Scheme 2a). Although this pathway has only recently been fully of HMG-CoA and increases productivity of mevalonate by almost elucidated, most efforts in engineering E. coli for isoprenoid produc- two-fold by relieving growth inhibition46. Following up on these tion have focused on manipulation of this route using the carotenoid results, library-based engineering of the intergenic regions of the pigments (from GGPP). Initial forays into pathway optimization polycistronic message for the MevT operon was carried out to balance sought to define the main rate-limiting steps within the pathway expression of the individual genes in the MevT operon, resulting in a itself; these were found to be 1-deoxyxylulose-5-phosphate synthase seven-fold overall increase in mevalonate titers due to improvements (DXS)32–34, 1-deoxyxylulose-5-phosphate reductoisomerase (DXR, or in specific production and cell growth47. Because of this toxicity, IspC)32,34, IPP-DMAPP isomerase (IDI)32,35 and the respective pre- increasing gene stability by high-level chromosomal expression of the nyltransferase (FPP or GGPP synthase)36,37. Upon delineation of the mevalonate pathway may also greatly increase production, as it has main limitations within the DXP pathway, attention was turned with the DXP pathway48. These efforts show that carbon can toward finding higher-level factors for either regulation or flux be effectively channeled for production of exogenous isoprenoids in

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

1

Primary metabolites

NATURE CHEMICAL BIOLOGY

Isomerase

OPP

ADVANCE ONLINE PUBLICATION

OPP

3

PERSPECTIVE

a

O

O

O

AAS

CoA Acetyl-CoA

OH

Acetyl-CoA

2 CoA

OH

ATP OH

MK

2 NADPH

HOOC

HMGS

CoA Hydroxymethylglutaryl-CoA

Acetoacetyl-CoA

HOOC

O

ATP

OH

OH

PMK

HOOC

HMGR

PMD

HOOC

OP

Mevalonate

OPP Mevalonate diphosphate

Mevalonate-5-phosphate

IDI

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

OPP

b

O

O

NADPH

DXS

OH

O

OH

4-(Cyt-5'-diphospho)-ME

OH

OH

IspH

IspG

OPP cyt

2-Phospho-4-(cyt-5'-diphospho)-ME

IspH

PO

IspF

OPP cyt

OH

ME-4-phosphate OP

OP

ATP IspE

OH

OH

OH

1-Deoxylulose-5-phosphate

CTP OP IspD

OP

Glyceraldehyde-3-phosphate

OH

OH

DXR/IspC

OP

+ O Pyruvate

OH

–CO2

OH

OPP DMAPP

IPP

OPP OH

OH

OH

ME-2,4-cyclodiphosphate

HMB-4-diphosphate

Scheme 2 Biosynthesis of IPP and DMAPP. (a) Via the mevalonate pathway (AAS, acetoacetyl-CoA synthase; HMGS, HMG-CoA synthase; HMGR, HMG-CoA reductase; MK, mevalonate kinase; PMK, phosphomevalonate kinase; PMD, mevalonate diphosphate decarboxylase). (b) Via the DXP pathway (ME, 2-methylerythritol; cyt, cytidine; HMB, 1-hydroxy-2-methyl-2-butenyl; IspD, ME-4-phosphate cytidyltransferase; IspE, 4-(cyt-5¢-diphospho)-ME kinase; IspF, ME-2,4-cyclodiphosphate synthase; IspG, HMB-4-diphosphate synthase; IspH, HMB-4-diphosphate reductase).

E. coli even though the flux through native isoprenoid biosynthesis is relatively low. Much of the earlier work in isoprenoid production in yeast has focused on the biosynthesis of endogenous and engineered steroids49–52, as well as carotenoids53. Although limiting factors have been identified in these studies50,54, little of the effort has specifically addressed the generation of a high-level production host for isoprenoid production55. More recently, S. cerevisiae has become a valuable platform for production of heterologous terpenoids such as the precursors to artemisinin56 and Taxol57. Indeed, yeast has an important advantage over E. coli in that the functional expression of cytochrome P450 (P450) enzymes, which participate ubiquitously in the biosynthetic pathways of most plant-derived natural products, is much less problematic. Metabolic engineering in S. cerevisiae is centered on the native mevalonate pathway for IPP production; so far, transplantation of the DXP pathway into yeast has not yet been proven to measurably supplement isoprenoid production. Yeast has a higher native use of isoprenoids compared with E. coli; thus pathway optimization must take into account both increasing FPP productivity and reducing production of endogenous sterols. Expression of ADS alone on a high-copy plasmid yielded 4 mg l–1 amorphadiene56. Downregulation of squalene synthase in combination with overexpression of the rate-limiting HMGR further increased production by ten fold. Further genetic modifications, including expression of upc2-1 (transcriptional regulator of sterol biosynthesis), provided a strain capable of producing 150 mg l–1 amorphadiene56, which is a B500-fold increase in sesquiterpene production compared with previous reports55. Both E. coli and S. cerevisiae have demonstrated suitability as host platforms for production of the universal isoprenoid precursors. Thus,

4

choice of the host for a particular application will depend on the details of the downstream pathway for production of the final biosynthetic target. For plant natural products, S. cerevisiae is often considered to be the logical choice as it has the capacity for posttranslational modification of enzymes and has less natural codon bias. However, E. coli still remains more amenable for genetic modification, and more tools are available for its metabolic engineering and analysis. In addition, the faster growth rate of E. coli is a benefit with regard to economic concerns when considering industrial fermentation processes. For both host microbes, limitations remain for isoprenoid production; engineering the native metabolism of both hosts could increase terpenoid yield by allowing carbon to be used more efficiently and by minimizing loss to acetate and biomass production58–60. The directed changes previously described bring large increases in productivity and have been instrumental for optimizing isoprenoid titers; however, a key goal for synthetic biology is also to develop systematic methods for engineering the global metabolic network of the cell. The difficulty in implementing these methods is exacerbated by complex, nonlinear molecular and genetic interactions that have not yet been sufficiently defined. Many methods, both systematic and librarybased, have been used for genome-wide screens for genes affecting isoprenoid production in E. coli through the DXP pathway and have uncovered metabolic and regulatory gene candidates for deletion61,62 and overexpression63. At the same time, profiling techniques for monitoring global RNA, protein and metabolite levels are essential for determining the effects of the engineered pathway on the rest of the cell network. Metabolic flux experiments using labeled carbon sources64–67 are also useful for visualizing the relationships between major metabolic pathways and for identifying new opportunities to increase flux through the target pathway. The ultimate objective is to

ADVANCE ONLINE PUBLICATION

NATURE CHEMICAL BIOLOGY

PERSPECTIVE been isolated over the last ten years yet half of the genes remain unidentified. Taxol is a OR1 highly decorated diterpenoid that has been 18 OH OH found to be an extremely valuable chemother10 19 P450s 9 11 DMAPP GGPP TS 12 Acyltransferases 16 apeutic agent, especially for the treatment of 7 13 15 GGPP + HO 8 17 synthase 6 lung, ovarian and breast cancer. It was origin14 3 IPP 3 1 5 2 4 ally isolated from the bark of the Pacific yew H H H HO OR1 tree74 but is currently produced semisynthetiOR 2 20 Proposed intermediate cally from advanced intermediates isolated Taxadiene R1 = Ac or H; R2 = Bz or H from induced plant cell culture (Scheme 3). Oxidation at C9 Taxol biosynthesis is thought to take place in Oxetane formation 19 steps from GGPP. After the first committed RO O step to form the parental taxa-4(5),11(12)O AcO OH diene carbon skeleton, the olefin is modified O NH O Ph OH by several P450-mediated oxygenations and HO CoA-dependent acylations. Formation of the O Ph Chemical synthesis four-membered cyclic ether and subsequent H HO OH oxidation generates the 10-deacetylbaccatin III O BzO AcO H HO O intermediate; four additional steps are then BzO AcO 10-Deacetylbaccatin III (R = H) Taxol required to produce the final Taxol structure Baccatin III (R = Ac) (Scheme 3)75. The first gene identified was the Biosynthesis gene encoding taxadiene synthase, and this Scheme 3 Semisynthesis of Taxol using S. cerevisiae. GGPP is made from the endogeous pool of IPP was accomplished on the basis of sequence and DMAPP. Upon cyclization to taxadiene, which is catalyzed by taxadiene synthase (TS), a series of conservation76. Using EST libraries from hydroxylations and acylations occur. Oxetane ring formation and C9 oxidation result in the generation induced Taxus brevifolia cells, three acyltransof 10-deacetylbaccatin III and baccatin III, both of which are precursors for the semisynthesis of ferases were then identified77–79, followed by Taxol. Bz, Benzyl. isolation of the first P450 gene80. Four more P450 hydroxylases were subsequently identified81–84, as well as genes responsible for genuse cellular profiling to answer very specific questions about global erating the C13 side chain85,86. Initial work in production E. coli has metabolism and to quickly pinpoint the source of problems or been limited to the production of taxadiene, and production is limitations. Advances in understanding this network, as well as postulated to be constrained by the solubility and activity of the methods for implementing balanced gene expression68 on the terpene synthase87. Upon the expression of DXS, IDI, GGPP synthase basis of this knowledge, are needed to increase the pace of efforts and taxadiene synthase, the parent terpene backbone has been proin this area. duced at titers of 1.3 mg l–1. S. cerevisiae has also been tested as a production host for Taxol intermediates57. Toward this goal, 8 of the Implementing downstream pathways for target production 15 genes (GGPP synthase, taxadiene synthase, three P450s and three One of the significant challenges for metabolic engineering of acyltransferases) required for baccatin were tested for expression in microbes is isolation of the genes involved in these extended biosyn- yeast and were all found to be active88. Coexpression of the GGPP thetic pathways. This task is especially demanding in plants, whose synthase and taxadiene synthase resulted in taxadiene production on genomes often well exceed the size of the human genome (for the order of 1 mg l–1, however further elaboration of the terpene example, the Assyrian fritillary genome is 110,000 megabases in backbone was limited by the first P450-dependent step and resulted in size69). In contrast to the genes encoding microbial natural products low-product titers of 25 mg l–1. Efforts toward the microbial producsuch as polyketides, the biosynthetic genes of plants are not arranged tion of Taxol illustrate the current difficulty in transplanting large in operons, but are instead scattered across the genome. Plants also biosynthetic pathways from a heterologous host into microbes such as typically contain many more pathways for production of secondary E. coli and S. cerevisiae. Although the rate of identification of metabolites than other types of organisms; these pathways are often biosynthetic genes is increasing, a substantial bottleneck remains for branched and produce structurally similar yet diverse products using functional expression of these genes at levels that are capable of closely related genes. Thus, the same families of enzymes are ubiqui- producing high titers of the target molecules. tously found in biosynthetic pathways, and consequently the genes for Our group has focused on the semisynthetic production of the a specific pathway are difficult to pinpoint. As more plant genome and highly effective antimalarial drug artemisinin using both S. cerevisiae expressed sequence tag (EST) sequencing projects are being under- and E. coli platforms (Scheme 4). Malaria is a growing global health taken, the use of comparative genomics in combination with tradi- threat that affects approximately 40% of the world’s population and is tional methods in plant physiology and chemical analysis has proven responsible for 1 to 3 million deaths annually89. Though treatments essential for more rapid gene identification. In a complementary area, such as the quinine class of drugs have been successful in managing advances in protein design and directed evolution of terpene malaria in the past, occurrence of multi-drug-resistant strains of the synthases70,71 and related P450s72,73 may help researchers to substitute malaria parasite, Plasmodium falciparum, have substantially increased the mortality rate over the last 20 years. Artemisinin-based combinathe activities of the native enzymes with engineered enzymes. Prominent work in the elucidation of the Taxol biosynthetic path- tion therapies (ACTs), recommended by the World Health Organizaway underscores the difficulty in isolating the full set of genes involved tion (WHO), are highly effective against even drug-resistant strains in the biosynthesis of a complex natural product; ten genes have but are currently inaccessible to malaria victims in developing nations Acetyl-CoA

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

Native mevalonate pathway

NATURE CHEMICAL BIOLOGY

ADVANCE ONLINE PUBLICATION

5

PERSPECTIVE Scheme 4 Semisynthesis of artemisinin using either E. coli or S. cerevisiae. IPP and DMAPP are synthesized using an engineered mevalonate pathway from the primary metabolite, acetyl-CoA. Cyclization to amorphadiene catalyzed by ADS provides the substrate for the next step. Oxidation is carried out in three steps via the corresponding alcohol and aldehyde intermediates. Isolation of the product, artemisinic acid, allows the chemical synthesis of artemisinin in two steps. IspA, Fpp synthase.

Engineered mevalonate pathway 2

Acetyl-CoA DMAPP

OPP +

OPP IPP

IspA

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

H

because of high cost and short supply. At present, raw artemisinin is obtained by extraction from the natural source, Artemisia annua (sweet wormwood), and represents a large part of the drug cost. The cost of ACTs could be greatly reduced if this plantsourced artemisinin were replaced with a less expensive microbially sourced artemisinin. The full biosynthetic pathway for artemisinin has yet to be illuminated, and it is not known whether installation of the key endoperoxide bridge is enzyme dependent90 or photochemical91. The first committed step in artemisinin biosynthesis is cyclization of FPP by ADS to yield amorphadiene43,44. Functionalization of the terpene carbon skeleton to yield artemisinic acid provides a biosynthetic intermediate that is then used as a precursor for the semisynthesis of artemisinin and related derivatives (Scheme 4). Upon harvesting artemisinic acid from the culture medium, chemical conversion of artemisinic acid to artemisinin can be achieved in two steps with an overall yield of 30–40%92. Oxidation of the terminal isopropenyl group of amorphadiene to the allylic carboxylic acid has been proposed to be catalyzed by P450 mono-oxygenases via the corresponding alcohol and aldehyde intermediates93. However, P450s in particular are known to have a wide range of sequence homology (1–99%) and are very abundant in plant genomes (for example, there are 103 P450s in rice), which makes the search for the specific gene quite challenging94. The search was narrowed by the observation that many plants in the Asteraceae family, comprising A. annua, chicory, lettuce and sunflower biosynthesize sesquiterpene lactones such as artemisinin (Fig. 2a). Using EST libraries for sunflower and lettuce, three main families of P450s have been targeted: CYP71, CYP76

a

b

Basal angiosperm Monocots Lower eudicots

Angiosperms

H O O H

O O

H

O

O

Eudicots

H O Lettucenin A Lettuce

O Artemisinin A. annua

Core eudicots Rosids I

O

OH

O

Eurosids II

O HO O

O O

Costunolide Chicory

HO

O Niveusin A Sunflower

Asterids

I

II Euasterids

ADS H FPP

3-step oxidation H

H O Reduction HO H

hν, O2

Chemical synthesis

O Artemisinic acid

O O H O

H

O Artemisinin

Microbial biosynthesis

and CYP82. The target list was narrowed again by the additional observation that of the known terpene hydroxylases (except those involved in Taxol biosynthesis), four out of five belong to the CYP71 family (Fig. 2b). The one outlier is found in cotton, which is only distantly related to the genus Artemisia, whereas mint and tobacco, which do contain CYP71 terpene hydroxylases, are closely related to this genus. Using CYP71-specific primers, a P450 was isolated from an A. annua cDNA library and was found to oxidize amorphadiene to artemisinic acid in three steps when coexpressed in S. cerevisiae with ADS and a cytochrome P450 reductase redox partner56. In vitro studies with purified yeast microsomes indicated that the isolated P450 gene is capable of carrying out each individual step and that endogenous alcohol and/or aldehyde dehydrogenases are not solely responsible for the oxidation of the alcohol and aldehyde. Parallel studies using EST libraries generated from trichromes (specialized organelles for synthesis and segregation of artemisinin) led to the identification of the same P450 for amorphadiene oxidation95. When the amorphadiene Gymnosperms oxidase (AMO) was expressed in the optiAmborella Laurales mized S. cerevisiae strain, up to 100 mg l–1 Liliopsida Ranunculales of artemisinic acid was produced in vivo56. Caryophyllales Although expression of plant P450s is known Santalales Berberidopsidaceae to be especially difficult in E. coli, work with Myrothamnaceae Gunneraceae 8-cadinene hydroxylase from cotton has shown Saxifragales Crossomatales that E. coli has a capacity similar to that of Gerianales Myrtales S. cerevisiae for producing oxidized sesquiterCelestrales penoids (4100 mg l–1; unpublished data). Malpighiales Oxalidales Zygophyllaceae Cucurbitales Fagales Fabales Rosales Brassicales Malvales Sapindales Cornales Ericales Garryales Gentianales Lamiales Solanales Aquifoliales Apiales Asterales Dipsacales

Cotton

Mint Tobacco A. annua

Figure 2 The search for a P450 for the oxidation of amorphadiene. (a) Sesquiterpene lactones isolated from different members of the Asteraceae family. (b) Phylogenetic tree indicating the relationship of A. annua with other plants from which terpene hydroxylases have been isolated.

6

PPO

Amorphadiene

Summary These developments indicate great promise for the production of isoprenoid pharmaceuticals in either E. coli or S. cerevisiae host platforms. Though increasing flux to isoprenoid precursors remains an ongoing area of research as production levels are directly tied to final cost, current titers of these precursors produced by fermentation would be sufficient for economic production of most drug targets. Current work has focused on a limited number of molecules, but this approach can be extended to other members of the isoprenoid

ADVANCE ONLINE PUBLICATION

NATURE CHEMICAL BIOLOGY

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

PERSPECTIVE family upon elucidation of the necessary biosynthetic pathways. Because of the effort required to identify the relevant genes, it is preferable to choose targets that are already clinically tested and approved, such as Taxol and artemisinin. This approach would also be valuable for compounds produced by rare species or unculturable microbes, such as eleutherobin from soft coral, as microbial production may be the only source of sufficient material to initiate clinical trials. At this time, functional expression of key enzymes, especially P450s, remains the primary challenge for microbial production of isoprenoids and other plant natural products. Tailoring enzymes such as acetylases, glycosylases and dehydrogenases, which are also commonly found in isoprenoid biosynthesis, tend to be more easily expressed in heterologous hosts. It is also an advantage to have a semisynthetic method in place for the compound of interest, because production of a less bioactive or toxic intermediate helps to avoid toxicity to the host and allows higher production titers. Another benefit of a semisynthetic approach is that these precursors can be used more easily for generating more potent derivatives and for carrying out diversity-oriented synthesis from a core structure or early intermediate. We hope that the methods developed and knowledge gained during the course of these efforts will allow us to take on more complex targets in the future. ACKNOWLEDGMENTS We would like to thank D. Pitera, E. Paradise and D.-K. Ro for helpful discussions. M.C.Y.C acknowledges a postdoctoral fellowship from the Jane Coffin Childs Memorial Fund. Isoprenoid research in the Keasling laboratory has been funded by the University of California Discovery Grant Program, the US National Science Foundation, Maxygen, Diversa and the Bill & Melinda Gates Foundation. COMPETING INTERESTS STATEMENT The authors declare competing financial interests (see the Nature Chemical Biology website for details). Published online at http://www.nature.com/naturechemicalbiology Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Newman, D.J., Cragg, G.M. & Snader, K.M. The influence of natural products upon drug discovery (Antiquity to late 1999). Nat. Prod. Rep. 17, 215–234 (2000). 2. Koehn, F.E. & Carter, G.T. The evolving role of natural products in drug discovery. Nat. Rev. Drug Discov. 4, 206–220 (2005). 3. Newman, D.J., Cragg, G.M. & Snader, K.M. Natural products as sources of new drugs over the period 1981–2002. J. Nat. Prod. 66, 1022–1037 (2003). 4. Horwitz, S.B. How to make taxol from scratch. Nature 367, 593–594 (1994). 5. McNeil, D.G.J. ‘‘Plant shortage leaves campaigns against malaria at risk.’’ New York Times 14 November 2004. 6. Khosla, C. & Keasling, J.D. Metabolic engineering for drug discovery and development. Nat. Rev. Drug Discov. 2, 1019–1025 (2003). 7. Glazer, A.N. & Nikaido, H. Microbial Biotechnology: Fundamentals of Applied Microbiology (WH. Freeman and Co., New York, 1995). 8. Stephanopoulos, G.N., Aristidou, A.A. & Nielsen, J. Metabolic Engineering: Principles and Methodologies (Academic, New York, 1998). 9. Ensley, B.D. et al. Expression of naphthalene oxidation genes in Escherichia coli results in the biosynthesis of indigo. Science 222, 167–169 (1983). 10. Mermod, N., Haryama, S. & Timmis, K.N. New route to bacterial production of indigo. Bio/Technology 4, 321–324 (1986). 11. Slater, S.C., Voige, W.H. & Dennis, D.E. Cloning and expression in Escherichia coli of the Alcaligenes eutrophus H16 poly-b-hydroxybutyrate biosynthetic pathway. J. Bacteriol. 170, 4431–4436 (1988). 12. Schubert, P., Steinbu¨chel, A. & Schlegel, H.G. Cloning of the Alcaligenes eutrophus genes for synthesis of poly-b-hydroxybutyric acid (PHB) and synthesis of PHB in Escherichia coli. J. Bacteriol. 170, 5837–5847 (1988). 13. Peoples, O.P. & Sinskey, A.J. Poly-b-hydroxybutyrate (PHB) biosynthesis in Alcaligenes eutrophus H16. Identification and characterization of the PHB polymerase gene (phbC). J. Biol. Chem. 264, 15298–15303 (1989). 14. Nakamura, C.E. & Whited, G.M. Metabolic engineering for the microbial production of 1,3-propanediol. Curr. Opin. Biotechnol. 14, 454–459 (2003). 15. Hopwood, D.A. et al. Production of ’hybrid’ antibiotics by genetic engineering. Nature 314, 642–644 (1985). 16. Stanzak, R., Matsushima, P., Baltz, R.H. & Rao, R.N. Cloning and expression in Streptomyces lividans of clustered erythromycin biosynthesis genes from Streptomyces erythreus. Bio/Technology 4, 229–232 (1986).

NATURE CHEMICAL BIOLOGY

ADVANCE ONLINE PUBLICATION

17. Tang, L. et al. Cloning and heterologous expression of the epothilone gene cluster. Science 287, 640–642 (2000). 18. Pfeifer, B.A. & Khosla, C. Biosynthesis of polyketides in heterologous hosts. Microbiol. Mol. Biol. Rev. 65, 106–118 (2001). 19. Menzella, H.G. et al. Combinatorial polyketide biosynthesis by de novo design and rearrangement of modular polyketide synthase genes. Nat. Biotechnol. 23, 1171–1176 (2005). 20. Pfeifer, B.A., Admiraal, S.J., Gramajo, H., Cane, D.E. & Khosla, C. Biosynthesis of complex polyketides in a metabolically engineered strain of E. coli. Science 291, 1790–1792 (2001). 21. Lau, J., Tran, C., Licari, P. & Galazzo, J. Development of a high cell-density fed-batch bioprocess for the heterologous production of 6-deoxyerythronolide B in Escherichia coli. J. Biotechnol. 110, 95–103 (2004). 22. Mutka, S.C., Bondi, S.M., Carney, J.R., Da Silva, N.A. & Kealey, J.T. Metabolic pathway engineering for complex polyketide biosynthesis in S. cerevisiae. FEMS Yeast Res. 6, 40–47 (2006). 23. Kealey, J.T., Liu, L., Santi, D.V., Betlach, M.C. & Barr, P.J. Production of a polyketide natural product in nonpolyketide-producing prokaryotic and eukaryotic hosts. Proc. Natl. Acad. Sci. USA 95, 505–509 (1998). 24. Conolly, J.D. & Hill, R.A. Dictionary of Terpenoids (Chapman & Hall, London, 1991). 25. Harborne, J.B. in Ecological Chemistry and Biochemistry of Plant Terpenoids (eds. Harborne, J.B. & Tomas-Barberan, F.F.) 399–426 (Clarendon Press, Oxford, 1991). 26. McCaskill, D. & Croteau, R. Prospects for the bioengineering of isoprenoid biosynthesis. Adv. Biochem. Eng. Biotechnol. 55, 107–146 (1997). 27. Spurgeon, S.L. & Porter, J.W. (eds.) Biosynthesis of Isoprenoid Compounds (Wiley, New York, 1981). 28. Rohmer, M., Knani, M., Simonin, P., Sutter, B. & Sahm, H. Isoprenoid biosynthesis in bacteria: a novel pathway for the early steps leading to isopentenyl diphosphate. Biochem. J. 295, 517–524 (1993). 29. Rohdich, F., Hecht, S., Bacher, A. & Eisenreich, W. Deoxyxylulose phosphate pathway of isoprenoid biosynthesis. Discovery and function of ispDEFGH genes and their cognate enzymes. Pure Appl. Chem. 75, 393–405 (2003). 30. Lange, B.M., Rujan, T., Martin, W. & Croteau, R. Isoprenoid biosynthesis: the evolution of two ancient and distinct pathways across genomes. Proc. Natl. Acad. Sci. USA 97, 13172–13177 (2000). 31. Rohdich, F. et al. Studies on the nonmevalonate terpene biosynthetic pathway: metabolic role of IspH (LytB) protein. Proc. Natl. Acad. Sci. USA 99, 1158–1163 (2002). 32. Albrecht, M., Misawa, N. & Sandmann, G. Metabolic engineering of the terpenoid biosynthetic pathway of Escherichia coli for production of the carotenoids b-carotene and zeaxanthin. Biotechnol. Lett. 21, 791–795 (1999). 33. Matthews, P.D. & Wurtzel, E.T. Metabolic engineering of carotenoid accumulation in Escherichia coli by modulation of the isoprenoid precursor pool with expression of deoxyxylulose phosphate synthase. Appl. Environ. Microbiol. 53, 396–400 (2000). 34. Kim, S.-W. & Keasling, J.D. Metabolic engineering of the nonmevalonate isopentenyl diphosphate synthesis pathway in Escherichia coli enhances lycopene production. Biotechnol. Bioeng. 72, 408–415 (2001). 35. Kajiwara, S., Fraser, P.D., Kondo, K. & Misawa, N. Expression of an exogenous isopentenyl diphosphate isomerase gene enhances isoprenoid biosynthesis in Escherichia coli. Biochem. J. 324, 421–426 (1997). 36. Neudert, U., Martinez-Ferez, I.M., Fraser, P.D. & Sandmann, G. Expression of an active phytoene synthase from Erwinia uredovora and biochemical properties of the enzyme. Biochim. Biophys. Acta 1392, 51–58 (1998). 37. Wang, C.W., Oh, M.K. & Liao, J.C. Engineered isoprenoid pathway enhances astaxantin production in Escherichia coli. Biotechnol. Bioeng. 62, 235–241 (1999). 38. Farmer, W.R. & Liao, J.C. Precursor balancing for metabolic engineering of lycopene production in Escherichia coli. Biotechnol. Prog. 17, 57–61 (2001). 39. Farmer, W.R. & Liao, J.C. Improving lycopene production in Escherichia coli by engineering metabolic control. Nat. Biotechnol. 18, 533–537 (2000). 40. Martin, V.J., Pitera, D.J., Withers, S.T., Newman, J.D. & Keasling, J.D. Engineering a mevalonate pathway in Escherichia coli for production of terpenoids. Nat. Biotechnol. 21, 796–802 (2003). 41. Kakinuma, K. et al. New approach to multiply deuterated isoprenoids using triply engineered Escherichia coli and its potential as a tool for mechanistic enzymology. J. Am. Chem. Soc. 123, 1238–1239 (2001). 42. Martin, V.J., Yoshikuni, Y. & Keasling, J.D. The in vivo synthesis of plant sesquiterpenes by Escherichia coli. Biotechnol. Bioeng. 75, 497–503 (2001). 43. Mercke, P., Bengtsson, M., Bouwmeester, H.J., Posthumus, M.A. & Brodelius, P.E. Molecular cloning, expression, and characterization of amorpha-4,11-diene synthase, a key enzyme of artemisinin biosynthesis in Artemisia annua L. Arch. Biochem. Biophys. 381, 173–180 (2000). 44. Chang, Y.J., Song, S.H., Park, S.H. & Kim, S.U. Amorpha-4,11-diene synthase of Artemisia annua: cDNA isolation and bacterial expression of a terpene synthase involved in artemisinin biosynthesis. Arch. Biochem. Biophys. 383, 178–184 (2000). 45. Newman, J.D. et al. High-level production of amorpha-4,11-diene in a two-phase partitioning bioreactor of metabolically engineered Escherichia coli. Biotechnol. Bioeng. 95, 684–691 (2006). 46. Pitera, D.J., Paddon, C., Newman, J.D. & Keasling, J.D. Rebuilding a balanced heterologous mevalonate pathway for isoprenoid production in Escherichia coli. Metab. Eng. (submitted). 47. Pfleger, B.F., Pitera, D.J., Smolke, C.D. & Keasling, J.D. Combinatorial engineering of intergenic regions in operons tunes expression of multiple genes. Nat. Biotechnol. 24, 1027–1032 (2006).

7

© 2006 Nature Publishing Group http://www.nature.com/naturechemicalbiology

PERSPECTIVE 48. Yuan, L.Z., Rouviere, P.E., Larossa, R.A. & Suh, W. Chromosomal promoter replacement of the isoprenoid pathway for enhancing carotenoid production in E. coli. Metab. Eng. 8, 79–90 (2006). 49. Donald, K., Hampton, R. & Fritz, I. Effects of overproduction of the catalytic domain of 3-hydroxy-3-methylglutaryl coenzyme A reductase on squalene synthesis in Saccharomyces cerevisiae. Appl. Environ. Microbiol. 63, 3341–3344 (1997). 50. Polakowski, T., Stahl, U. & Lang, C. Overexpression of a cytosolic hydroxymethylglutaryl-CoA reductase leads to squalene accumulation in yeast. Appl. Microbiol. Biotechnol. 49, 66–71 (1998). 51. Duport, C., Spagnoli, R., Degryse, E. & Pompon, D. Self-sufficient biosynthesis of pregnenolone and progesterone in engineered yeast. Nat. Biotechnol. 16, 186–189 (1998). 52. Szczebara, F.M. et al. Total biosynthesis of hydrocortisone from a simple carbon source in yeast. Nat. Biotechnol. 21, 143–149 (2003). 53. Shimada, H. et al. Increased carotenoid production by the food yeast Candida utilis through metabolic engineering of the isoprenoid pathway. Appl. Environ. Microbiol. 64, 2676–2680 (1998). 54. Szkopinska, A., Swiezewska, E. & Karst, F. The regulation of activity of main mevalonic acid pathway enzymes: farnesyl diphosphate synthase, 3-hydroxy-3-methylglutarylCoA reductase, and squalene synthase in yeast S. cerevisiae. Biochem. Biophys. Res. Commun. 267, 473–477 (2000). 55. Jackson, B.E., Hart-Wells, E.A. & Matsuda, S.P.T. Metabolic engineering to produce sesquiterpenes in yeast. Org. Lett. 5, 1629–1632 (2003). 56. Ro, D.K. et al. Production of the antimalarial drug precursor artemisinic acid in engineered yeast. Nature 440, 940–943 (2006). 57. DeJong, J.M. et al. Genetic engineering of taxol biosynthetic genes in Saccharomyces cerevisiae. Biotechnol. Bioeng. 93, 212–224 (2006). 58. Causey, T.B., Zhou, S., Shanmugam, K.T. & Ingram, L.O. Engineering the metabolism of Escherichia coli W3110 for the conversion of sugar to redox-neutral and oxidized products: homoacetate production. Proc. Natl. Acad. Sci. USA 100, 825–832 (2003). 59. Causey, T.B., Shanmugam, K.T., Yomano, L.P. & Ingram, L.O. Engineering Escherichia coli for efficient conversion of glucose to pyruvate. Proc. Natl. Acad. Sci. USA 101, 2235–2240 (2004). 60. Aristidou, A.A., San, K.-Y. & Bennett, G.N. Modification of central metabolic pathway in Escherichia coli to reduce acetate accumulation by heterologous expression of the Bacillus subtilis acetolactate synthase gene. Biotechnol. Bioeng. 44, 944–951 (2004). 61. Alper, H., Jin, Y.S., Moxley, J.F. & Stephanopoulos, G. Identifying gene targets for the metabolic engineering of lycopene biosynthesis in Escherichia coli. Metab. Eng. 7, 155–164 (2005). 62. Alper, H., Miyaoku, K. & Stephanopoulos, G. Construction of lycopene-overproducing E. coli strains by combining systematic and combinatorial gene knockout targets. Nat. Biotechnol. 23, 612–616 (2005). 63. Hemmi, H., Ohnuma, S., Nagaoka, K. & Nishino, T. Identification of genes affecting lycopene formation in Escherichia coli transformed with carotenoid biosynthetic genes: candidates for early genes in isoprenoid biosynthesis. J. Biochem. 123, 1088–1096 (1998). 64. Varma, A. & Palsson, B.O. Metabolic flux balancing: basic concepts, scientific and practical use. Bio/Technology 12, 994–998 (1994). 65. Stephanopoulos, G. Metabolic fluxes and metabolic engineering. Metab. Eng. 1, 1–11 (1999). 66. Fischer, E. & Sauer, U. Metabolic flux profiling of Escherichia coli mutants in central carbon metabolism using GC-MS. Eur. J. Biochem. 270, 880–891 (2003). 67. Sauer, U. High-throughput phenomics: experimental methods for mapping fluxomes. Curr. Opin. Biotechnol. 15, 58–63 (2004). 68. Bayer, T.S. & Smolke, C.D. Programmable ligand-controlled riboregulators of eukaryotic gene expression. Nat. Biotechnol. 23, 337–343 (2005). 69. Bennett, M.D. & Smith, J.B. Nuclear DNA amounts in angiosperms. Philos. Trans. R. Soc. Lond. B Biol. Sci. 274, 227–274 (1976). 70. Yoshikuni, Y., Martin, V.J., Ferrin, T.E. & Keasling, J.D. Engineering cotton (+)-dcadinene synthase to an altered function: germacrene D-4-ol synthase. Chem. Biol. 13, 91–98 (2006). 71. Yoshikuni, Y., Ferrin, T.E. & Keasling, J.D. Designed divergent evolution of enzyme function. Nature 440, 1078–1082 (2006). 72. Cirino, P.C. & Arnold, F.A. Protein engineering of oxygenases for biocatalysis. Curr. Opin. Chem. Biol. 6, 130–135 (2002).

8

73. Sowden, R.J., Yasmin, S., Rees, N.H., Bell, S.G. & Wong, L.L. Biotransformation of the sesquiterpene (+)-valencene by cytochrome P450cam and P450BM-3. Org. Biomol. Chem. 3, 57–64 (2005). 74. Wani, M.C., Taylor, H.L., Wall, M.E., Coggon, P. & McPhail, A.T. Plant antitumor agents. VI. The isolation and structure of taxol, a novel antileukemic and antitumor agent from Taxus brevifolia. J. Am. Chem. Soc. 93, 2325–2327 (1971). 75. Jennewein, S., Wildung, M.R., Chau, M., Walker, K. & Croteau, R. Random sequencing of an induced Taxus cell cDNA library for identification of clones involved in taxol biosynthesis. Proc. Natl. Acad. Sci. USA 101, 9149–9154 (2004). 76. Wildung, M.R. & Croteau, R. A cDNA clone for taxadiene synthase, the diterpene cyclase that catalyzes the committed step of taxol biosynthesis. J. Biol. Chem. 271, 9201–9204 (1996). 77. Walker, K. & Croteau, R. Molecular cloning of a 10-deacetylbaccatin III-10-O-acetyl transferase cDNA from Taxus and functional expression in Escherichia coli. Proc. Natl. Acad. Sci. USA 97, 583–587 (2000). 78. Walker, K. & Croteau, R. Taxol biosynthesis: molecular cloning of a benzoyl-CoA:taxane 2a-O-benzoyltransferase cDNA from Taxus and functional expression in Escherichia coli. Proc. Natl. Acad. Sci. USA 97, 13591–13596 (2000). 79. Walker, K., Schoendorf, A. & Croteau, R. Molecular cloning of a taxa-4(20),11(12)dien-5a-ol-O-acetyl transferase cDNA from Taxus and functional expression in Escherichia coli. Arch. Biochem. Biophys. 374, 371–380 (2000). 80. Schoendorf, A., Rithner, C.D., Williams, R.M. & Croteau, R.B. Molecular cloning of a cytochrome P450 taxane 10b-hydroxylase cDNA from Taxus and functional expression in yeast. Proc. Natl. Acad. Sci. USA 98, 1501–1506 (2001). 81. Jennewein, S., Rithner, C.D., Williams, R.M. & Croteau, R. Taxol biosynthesis: taxane 13a-hydroxylase is a cytochrome P450-dependent monooxygenase. Proc. Natl. Acad. Sci. USA 98, 13595–13600 (2001). 82. Jennewein, S., Long, R.M., Williams, R.M. & Croteau, R. Cytochrome P450 taxadiene 5a-hydroxylase, a mechanistically unusual monooxygenase catalyzing the first oxygenation step of taxol biosynthesis. Chem. Biol. 11, 379–387 (2004). 83. Chau, M., Jennewein, S., Walker, K. & Croteau, R. Taxol biosynthesis: molecular cloning and characterization of a cytochrome P450 taxoid 7b-hydroxylase. Chem. Biol. 11, 663–672 (2004). 84. Chau, M. & Croteau, R. Molecular cloning and characterization of a cytochrome P450 taxoid 2a-hydroxylase involved in taxol biosynthesis. Arch. Biochem. Biophys. 427, 48–57 (2004). 85. Walker, K., Long, R. & Croteau, R. The final acylation step in taxol biosynthesis: cloning of the taxoid C13-side chain N-benzoyltransferase. Proc. Natl. Acad. Sci. USA 99, 9166–9171 (2002). 86. Walker, K.D., Klettke, K., Akiyama, T. & Croteau, R. Cloning, heterologous expression, and characterization of a phenylalanine aminomutase involved in taxol biosynthesis. J. Biol. Chem. 279, 53947–53954 (2004). 87. Huang, Q., Roessner, C.A., Croteau, R. & Scott, A.I. Engineering Escherichia coli for the synthesis of taxadiene, a key intermediate in the biosynthesis of taxol. Bioorg. Med. Chem. 9, 2237–2242 (2001). 88. Hefner, J., Ketchum, R.E.B. & Croteau, R. Cloning and functional expression of a cDNA encoding geranylgeranyl diphosphate synthase from Taxus canadensis and assessment of the role of this prenyltransferase in cells induced for Taxol production. Arch. Biochem. Biophys. 360, 62–74 (1998). 89. Korenromp, E., Miller, J., Nahlen, B., Wardlaw, T. & Young, M. World Malaria Report 2005 (World Health Organization, Roll Back Malaria, Geneva, 2005). 90. Dhingra, V. & Narasu, M.L. Purification and characterization of an enzyme involved in biochemical transformation of arteannuin B to artemisinin from Artemisia annua. Biochem. Biophys. Res. Commun. 281, 558–561 (2001). 91. Wallaart, T.E. et al. Isolation and identification of dihydroartemisinic acid from Artemisia annua and its possible role in the biosynthesis of artemisinin. J. Nat. Prod. 62, 430–433 (1999). 92. Roth, R.J. & Acton, N. A simple conversion of artemisinic acid into artemisinin. J. Nat. Prod. 52, 1183–1185 (1989). 93. Bertea, C.M. et al. Identification of intermediates and enzymes involved in the early steps of artemisinin biosynthesis in Artemisia annua. Planta Med. 71, 40–47 (2005). 94. Schuler, M.A. & Werck-Reichhart, D. Functional genomics of P450s. Annu. Rev. Plant Biol. 54, 629–667 (2003). 95. Teoh, K.H., Polichuk, D.R., Reed, D.W., Nowak, G. & Covello, P.S. Artemisia annua L. (Asteraceae) trichome-specific cDNAs reveal CYP71AV1, a cytochrome P450 with a key role in the biosynthesis of the antimalarial sesquiterpene lactone artemisinin. FEBS Lett. 580, 1411–1416 (2006).

ADVANCE ONLINE PUBLICATION

NATURE CHEMICAL BIOLOGY

Leading Edge

Essay Engineering Static and Dynamic Control of Synthetic Pathways William J. Holtz1,4 and Jay D. Keasling2,3,4,5,*

Department of Electrical Engineering and Computer Science, University of California at Berkeley Departments of Chemical Engineering and Bioengineering, University of California at Berkeley 3 Physical Bioscience Division, Lawrence Berkeley National Laboratory Berkeley, CA 94720, USA 4 Synthetic Biology Engineering Research Center 5 Joint Bio Energy Institute Emeryville, CA 94608, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2009.12.029 1 2

Maximizing the production of a desired small molecule is one of the primary goals in metabolic engineering. Recent advances in the nascent field of synthetic biology have increased the predictability of small-molecule production in engineered cells growing under constant conditions. The next frontier is to create synthetic pathways that adapt to changing environments. Introduction Metabolic engineering is the genetic manipulation of intracellular, enzymecatalyzed, chemical reactions for the production of a desired molecule. For relatively simple products that require few gene manipulations to effect their production, natural or slightly modified control systems (such as promoters or ribosome-binding sites) have been widely used to control the expression of heterologous pathways or deregulate native genes. Most of these non-native controllers are static; that is, the level of gene expression is set without sensing changes in pathway output or cellular environment. Generally, these systems are optimized only for a particular environment and any deviations away from that set of conditions are likely to result in suboptimal productivity. Metabolic engineering is being called upon to tackle increasingly demanding challenges, like the production of complicated pharmaceuticals and nutraceuticals that require very long metabolic pathways or the convergence of multiple heterologous pathways. For example, the pathway from acetyl-CoA to β-carotene has been implemented with 12 enzymatic steps, and the production of the anti-malaria drug precursor artemisinic acid from acetyl-CoA uses 10 enzymatic steps (Yoon et al., 2009; Ro et al., 2006). As heterologous pathways become larger and more complicated, optimizing

them becomes increasingly difficult such that piecewise optimizations, where a subset of the pathway is optimized at a time, are utilized (Martin et al., 2003). Because segments of pathways are not independent, performing pathway optimization in sections and then combining the sections together can result in the final state of the cell being different from its state during the earlier optimizations. The ability to sense the output of each pathway segment and control of the levels of expression of pathway enzymes (or activities of pathway enzymes) would allow each pathway segment to optimize itself relative to the other pathway segments and the cell’s native metabolism. Another difficult, and timely, challenge in metabolic engineering is the production of molecules with a low profit margin (such as commodity chemicals and fuels), where the need to maximize yield and productivity are essential for economic viability (Zhang et al., 2008; Nakamura and Whited, 2003). The ability to sense critical metabolic intermediates and control levels of pathway gene expression could eliminate pathway bottlenecks or the accumulation of potentially toxic intermediates. Most of these molecules are produced by cells grown in large bioreactors where heterogeneities in the nutrient concentrations, oxygen levels, or pH are often the norm and markedly different from the conditions under which the strains were originally

constructed and optimized (that is, in the uniform environment of the shaker flask or small fermenter) (Amanullah et al., 2004; Schmidt, 2005). In applications such as these, the ability to sense the environment in which a particular cell finds itself and respond could vastly improve production by the entire culture. Additional improvements in yield or productivity could be achieved if the pathway has the capacity to monitor and respond to the cell’s growth phase and density. In contrast to the static control engineered into most heterologous pathways, native metabolic pathways generally utilize dynamic regulatory networks to compensate for changing conditions by altering fluxes. That is, the turnover rate of molecules by an enzyme is altered, often by allosteric inhibition or by a negative feedback loop regulating expression of the enzyme. As the environment outside the cell changes (such as a change in nutrient availability or pH), the cell modulates its metabolic pathways dynamically to adjust fluxes so that required metabolic intermediates are delivered at the appropriate levels and times to optimize growth. Coupling sensory inputs with control devices allows for pathways to be dynamically controlled so that resources can be more efficiently utilized and productivity gains realized. By sensing cellular conditions and adjusting fluxes in a feedback loop,

Cell 140, January 8, 2010 ©2010 Elsevier Inc.  19

the engineered cell can remain near the optimal productivity that corresponds to each set of conditions encountered. The lack of dynamic controls in metabolic engineering is due in part to the complexity of designing and implementing pathways that both sense the environment and control the formation of product. Dynamic controls have the added requirements of a sensing component, a method of modulating flux, and a connection between those two elements. An additional complexity of dynamic control is the requirement for more parameters to determine transient behavior than would be required for static behavior. Thus, the benefits of dynamic control may not be perceived to outweigh the additional time and money required for implementation and testing. The predisposition against working on system dynamics is a problem that extends beyond synthetic biology. The majority of molecular biology and biomedical research also focuses on steady-state conditions, and this is likely due to the increased costs and difficultly of analyzing dynamic properties. Growing interest in implementing dynamic controls will result in the development of new experimental and computational procedures. These new techniques could lead to a synergy between synthetic biology and molecular biology that rapidly advances research on dynamic systems. Recent work on the dynamics of native metabolic pathways demonstrates how molecular biology can provide insights into how to measure and optimize their responses (Chin et al., 2008). Although there are barriers to the adoption of dynamic controls, advances in synthetic and systems biology will continue to lower these barriers and make dynamic controls more attractive. In particular, progress in modeling, computer-aided design, quantitative gene expression, and DNA synthesis and assembly will play important roles in enabling dynamic controls. And the development of advanced biological devices (for example, logic gates, bi-stable switches, counters, and ring oscillators) has resulted in the availability of a large number of well-characterized biological components that can be used to construct dynamic controllers. Many of these devices make use of standardized

parts or interfaces that simplify functional composition and allow for rapid reuse of devices in new contexts. Hopefully these community-driven standards will result in a lowering of the costs involved in building complex systems. Not only will these well-characterized and standardized components aid the development of dynamic controls, they will also further static control systems, which are likely to be used extensively for controlling metabolic pathways long into the future. Static Controls Static controllers of flux are genetically encoded components that play a role in determining the level of flux through a pathway but do not sense cellular conditions and modulate the pathway flux based on the sensed information. There are many parameters that can be used to change the static control of a particular flux. Examples of such parameters are strength and type of constitutive promoter, ribosome binding site strength, or copy number of the vector. There is a vast literature on promoters and transcription initiation. However, much of the application of promoters in metabolic engineering involves the use of natural or slightly modified, inducible promoters or constitutive promoters of different strengths. Precise quantitative data on promoter activities from relevant culture conditions are required for predicting gene expression levels to alter flux. Recently, there has been an effort to standardize a technique for measuring relative promoter strengths (Kelly et al., 2009). In this method, the fluorescence level from green fluorescent protein from a standard reference plasmid is compared to a plasmid with only the promoter changed. The ratio of the two expression levels is expected to remain constant if the measurements are repeated with a different host strain or growth condition. Therefore, if the fluorescence from the standard plasmid is measured under a unique set of conditions, then the transcription rate can be predicted for any promoter that has been previously measured in another context along with the reference promoter. One major challenge in specifying the level of gene expression is designing the appropriate levels of translation. In native systems, the folding of mRNA

20  Cell 140, January 8, 2010 ©2010 Elsevier Inc.

near the ribosome-binding site (RBS) can strongly contribute to the gene expression level. Because the portion of the mRNA containing the RBS can interact with the gene-coding region, it is not trivial to predict how an RBS-gene pair will express based on their individual behavior in other contexts. The energetics of translation initiation in prokaryotes has been extensively studied. However, many of the existing models have an unacceptable failure rate or are difficult to use when designing systems. A thermodynamic model of translation initiation has been developed by Salis et al. (2009) to predict the relative level of translation initiation for a given mRNA. Using the nucleotide sequence of a gene and a relative translation initiation rate, the algorithm generates a corresponding RBS sequence that when incorporated into the construct will provide the desired level of gene expression. Use of such a tool reduces the unintentional introduction of inhibitory RNA secondary structures and allows for levels of gene expression to be easily and predictably adjusted. Another method of achieving predictable translation levels is to use bicistronic operons, which have long been employed to ensure high levels of expression from arbitrary genes (Schoner et al., 1986). The existence of mRNA secondary structure between the second RBS and downstream gene does not cause translational inhibition, and therefore the translation initiation of the second gene is independent of its coding sequence (Kimura et al., 2005). The first gene can be a truncated protein that only exists to affect the translation of the second gene. Use of synthetic bicistronic systems has been limited to achieving high protein levels. However, there is the potential to create libraries of bicistronic static controllers of flux that cover a broad range of translation initiation rates. It has not yet been demonstrated that bicistronic constructs can give predictable and precise expression of the second gene; however they are likely to outperform pairings of genes with independently characterized RBSs. Increasing enzyme expression levels is an easy route to achieving high fluxes with static controls; however, other methods may make more efficient use of

Figure 1. Dynamic Control of a Synthetic Pathway A dynamic controller of flux can achieve higher productivity than constitutive expression of the proteins in the pathway. The dynamic system only expresses the proteins involved in the production pathway once a threshold cell density is reached. (A) All dynamic controllers of flux contain a sensor, an output, and an interface between them. (B) At low cell densities the diffusible molecule acyl-homoserine lactone produced by LuxI is too dilute to activate its receptor LuxR. (C) Once a critical cell density is reached LuxR is activated and turns on expression of the pathway genes from the Lux promoter (Plux). (D) Plots showing how higher productivity can be obtained by delaying expression of the pathway until a high cell density is reached.

cellular resources. By localizing tagged pathway enzymes to a scaffold using protein-protein interaction domains, Dueber et al. (2009) have been able to increase the flux 77-fold through a biosynthetic pathway for mevalonate in the bacterium Escherichia coli. They use an engineered scaffold, containing mouse and rat protein-protein interaction domains, to increase the local concentration of the enzymes and prevent the accumulation of pathway intermediates that may be harmful to the cells. Ideally, one would use a model to specify the gene expression levels needed to achieve the desired flux in any one particular reaction. The existence of genome-scale metabolic models has lead to the development of new computational algorithms for improving production strains. Microbes appear to maximize their formation of biomass, which often does not lead to high yields

of the desired product. By altering or eliminating expression levels of native genes it is possible to improve a strain’s yield or productivity. However, the number of possible combinatorial changes is too large to test in vivo and therefore in silico methods are used to predict the best set of modifications. These predictions are generally made using flux balance analysis, which makes an assumption of steady-state conditions in order to reduce computational complexity (Llaneras and Picó, 2008). Because flux balance analysis does not incorporate any regulatory information, these methods can only suggest changes to expression levels of catalytic enzymes but not to regulatory genes. Another drawback of flux balance analysis is the absence of any kinetic information. Despite these short-comings, flux balance analysis has proven to be a useful tool for choosing where to imple-

ment static controllers of flux. Several algorithms have been published for determining targets for altered expression levels. To simplify calculations some methods only determine genes to be knocked out. The OptKnock framework uses linear programming techniques to efficiently find combinations of knockout candidates (Burgard et al., 2003). OptReg, an extension of the OptKnock framework, can select genes that should be expressed at higher levels, lower levels, or not at all (Pharkya and Maranas, 2006). An alternative iterative search methodology was used by Alper et al. (2005a) for optimizing lycopene biosynthesis in E. coli. They first identified advantageous single knockouts and then additional knockouts were introduced in the single knockout strains, which were then tested for lycopene production. Strains with single, double, and triple knockouts

Cell 140, January 8, 2010 ©2010 Elsevier Inc.  21

were created and tested in vivo (Alper et al., 2005a). For most strains they find good agreement between in silico and in vivo results. To explore the limits of flux balance analysis, Alper et al. performed a global transposon search, which found new knockout targets, and used a convenient screening method based on colony color to rapidly locate the advantageous knockouts. Although the transposon knockouts were able to improve production, they did not exceed the production of the best strain from flux balance analysis, nor did the combination of both knockout pools result in a new maximum level of lycopene production (Alper et al., 2005b). This illustrates that developing in silico methods for determining which regulatory genes are good targets for knockouts or modification of expression level is an open area of research. Dynamic Control Dynamic control is a common feature of native metabolic pathways, and allosteric regulation is an important mechanism to maintain flux or limit accumulation of a metabolic intermediate through the pathway in the face of changing conditions. In general, the first enzyme or a key branchpoint of a pathway is downregulated by the pathway’s product. For example, the first enzyme in E. coli’s serine biosynthesis pathway, D-3-phosphoglycerate dehydrogenase (SerA), is allosterically inhibited by serine (Grant et al., 1996). There are few published examples of engineered dynamic control of fluxes in heterologous pathways. However, these systems will become increasingly common as engineered networks become more predictable (Figure 1). Farmer and Liao (2000) were the first to demonstrate that engineering dynamic control of fluxes could improve yield and productivity of a heterologous pathway. Excess glucose flux and the diversion of carbon to acetate formation reduce the productivity of lycopene-generating E. coli cultures. An acetyl phosphate-activated transcription factor and promoter were used as a sensor of excess glucose flux. When acetyl phosphate accumulates inside the cell and is detected by the engineered system, transcription of two genes regulated by the engineered system (pps and idi) are upregulated, which diverts flux from acetate produc-

tion to lycopene. The strain with dynamic control of pps and idi produces titers of lycopene that are 18-fold higher (and comparable improvement to productivity) than those from a strain with constitutive control of the genes. Growing cultures to an intermediate or high density before inducing product formation is a common practice because it can greatly improve productivity. However using inducer adds cost and potential regulatory hurdles. An alternative is to engineer microbes to sense their own density and activate gene expression at the appropriate time (Kobayashi et al., 2004). By linking a genetic toggle switch to a synthetic quorum sensing system based on the lux system from Vibrio fischeri, cells activate gene expression when they reach a threshold density (Kobayashi et al., 2004). At low cell densities the protein of interest is not detectable, at intermediate densities there is a bimodal distribution with the majority of cells not expressing the protein, and at high densities all the cells express the protein. Although this system has been used in the context of protein production, it could easily be applied to metabolic engineering to upregulate an enzyme or pathway. More recently, Gadkar et al. (2005) have shown that it is possible to use metabolic flux analysis to predict improvements in product formation resulting from dynamic pathway control. Using a metabolic model of a glycerol-producing strain of E. coli, the optimal induction time for the glycerol kinase gene, glpK, was determined so that growth is maximized in one phase and production maximized in the other phase. Introducing this level of dynamic control of enzyme expression increased biomass, leading to higher glycerol productivity. The same modeling framework has also been applied to ethanol production by an E. coli strain deficient in the lactate biosynthetic enzyme lactate dehydrogenase and expressing a heterologous pyruvate decarboxylase and alcohol dehydrogenase (Gadkar et al., 2005). Knocking out the gene encoding acetate kinase, ackA, is known to drastically decrease the growth rate but increase the flux of carbon to ethanol production. The authors compared an ackA knockout to the repression of ackA expres-

22  Cell 140, January 8, 2010 ©2010 Elsevier Inc.

sion at an optimal time during the process. The general results are similar to those revealed for glycerol production; a higher initial growth rate allows the culture with a modulated ackA to have an increased productivity in experiments of limited duration. This work provides a mathematical framework for dynamically controlling fluxes and demonstrates the advantages of doing so. Anesiadis et al. (2008) assessed in silico the ability of the dynamic controller devised by Kobayashi et al. (2004) to modulate fluxes in strains engineered to produce ethanol or succinate. In all cases, Anesiadis et al. had the dynamic controller repress expression of one or more genes when cell density reached a threshold. The affinity between LuxR (a quorum sensing transcriptional activator) and its ligand acyl-homoserine lactone is used to tune the threshold cell density. The modulated genes were in competing pathways or consumed an upstream metabolite. With an in silico genomescale model of E. coli, the authors demonstrate that a self-regulating biphasic culture could be obtained and that such a culture could result in increased productivity. Opportunities for Dynamic Controls The literature on dynamic control that we have presented provides examples of added layers of regulation that could be integrated with existing heterologous pathways to turn on or off genes at certain stages of the culture. This is a powerful paradigm and for the near term will likely constitute the majority of research on dynamic controls. However there are other ways of applying dynamic controls to metabolic engineering that have not yet been demonstrated. Mimicking allosteric regulation is difficult to implement for an arbitrary product, in part because protein engineering is not yet sophisticated enough to perform such a task. However it would be easier to implement a feedback loop that modulates the expression level of an upstream enzyme. An engineered aptazyme (a fusion of an aptamer and ribozyme to form an allosteric catalytic RNA) could be used to sense product and accordingly inhibit translation of an enzyme in its metabolic pathway. Alternatively, the dynamic control could forgo

sensing the product and simply attempt to maintain an enzyme at a constant concentration by creating a fusion between the enzyme and a transcriptional inhibitor so that the enzyme autoregulates its own transcription. Such a feedback loop could reduce the noise present in gene expression to give a more constant level of enzyme (Becskei and Serrano, 2000). One opportunity is to engineer dynamic controls that will respond to cell heterogeneities and “dead zones” present in large-scale industrial cultures. Through scale up, it can be difficult to maintain a homogeneous culture in part due to gradients of carbon source, O2, pH, or CO2 in large fermentation vessels (Amanullah et al., 2004). Strains with only static control of flux through the heterologous pathway cannot be optimized for production in all of these different environments. By using dynamic controls it may be possible to partially rescue the productivity of cells while they are in the undermixed regions of the bioreactor. This is a challenging proposition because time constants of the dynamic controls will need to be less than the circulation time of cells in the bioreactor. One final application for dynamic controls is to enable piecewise optimization of a large pathway. If a strain is designed and optimized to produce a common precursor, then later work to add more genes downstream has the potential to interact with the upstream portion of the pathway and perturb it away from the optimal flux. A system that could sense and respond to the change in flux would allow downstream

engineering to proceed with less concern about how additions to the strain might negate previous work. Conclusions Synthetic biology has the potential to reshape how heterologous pathways are designed and controlled. Static controls for synthetic pathways are becoming more predictable, and in combination with advances in DNA synthesis, new levels of complexity in pathway design are now feasible. This additional complexity can enable new, longer pathways or can be used to add dynamic controls that can increase productivity and make strains more robust to changing conditions.

Farmer, W.R., and Liao, J.C. (2000). Nat. Biotechnol. 18, 533–537. Gadkar, K.G., Doyle III, F.J., Edwards, J. S., and Mahadevan, R. (2005). Biotechnology and Bioengineering 89, 243-251. Grant, G.A., Schuller, D.J., and Banaszak, L.J. (1996). Protein Sci. 5, 34–41. Kelly, J., Rubin, A., Davis, J., Ajo-Franklin, C., Cumbers, J., Czar, M., de Mora, K., Glieberman, A., Monie, D., and Endy, D. (2009). J. Biol. Eng. 3, 4. Kimura, S., Umemura, T., and Iyanagi, T. (2005). J. Biochem. 137, 523–533. Kobayashi, H., Kærn, M., Araki, M., Chung, K., Gardner, T.S., Cantor, C.R., and Collins, J.J. (2004). Proc. Natl. Acad. Sci. USA 101, 8414–8419. Llaneras, F., and Picó, J. (2008). J. Biosci. Bioeng. 105, 1–11.

References

Martin, V.J.J., Pitera, D.J., Withers, S.T., Newman, J.D., and Keasling, J.D. (2003). Nat. Biotechnol. 21, 796–802.

Alper, H., Jin, Y., Moxley, J., and Stephanopoulos, G. (2005a). Metab. Eng. 7, 155–164.

Nakamura, C.E., and Whited, G.M. (2003). Curr. Opin. Biotechnol. 14, 454–459.

Alper, H., Miyaoku, K., and Stephanopoulos, G. (2005b). Nat. Biotechnol. 23, 612–616.

Pharkya, P., and Maranas, C.D. (2006). Metab. Eng. 8, 1–13.

Amanullah, A., Buckland, B.C., and Nienow, A.W. (2004). Mixing in the fermentation and cell culture industries. Handbook of Industrial Mixing: Science and Practice (New York: John Wiley & Sons), pp. 1071–1170.

Ro, D., Paradise, E.M., Ouellet, M., Fisher, K.J., Newman, K.L., Ndungu, J.M., Ho, K.A., Eachus, R.A., Ham, T.S., Kirby, J., et al. (2006). Nature 440, 940–943.

Anesiadis, N., Cluett, W.R., and Mahadevan, R. (2008). Metab. Eng. 10, 255–266. Becskei, A., and Serrano, L. (2000). Nature 405, 590–593. Burgard, A.P., Pharkya, P., and Maranas, C.D. (2003). Biotechnol. Bioeng. 84, 647–657.

Salis, H.M., Mirsky, E.A., and Voigt, C.A. (2009). Nat. Biotechnol. 27, 946–950. Schmidt, F.R. (2005). Appl. Microbiol. Biotechnol. 68, 425–435. Schoner, B.E., Belagaje, R.M., and Schoner, R.G. (1986). Proc. Natl. Acad. Sci. USA 83, 8506–8510.

Chin, C., Chubukov, V., Jolly, E.R., DeRisi, J., and Li, H. (2008). PLoS Biol. 6, e146.

Yoon, S., Lee, S., Das, A., Ryu, H., Jang, H., Kim, J., Oh, D., Keasling, J.D., and Kim, S. (2009). J. Biotechnol. 140, 218–226.

Dueber, J.E., Wu, G.C., Malmirchegini, G.R., Moon, T.S., Petzold, C.J., Ullal, A.V., Prather, K.J., and Keasling, J.D. (2009). Nat. Biotech. 27, 753–759.

Zhang, K., Sawaya, M.R., Eisenberg, D.S., and Liao, J.C. (2008). Proc. Natl. Acad. Sci. USA 105, 20653–20658.

Cell 140, January 8, 2010 ©2010 Elsevier Inc.  23