Optimization of enzyme parameters for fermentative production of biorenewable fuels and chemicals

Microbial biocatalysts such as Escherichia coli and Saccharomyces cerevisiae have been extensively subjected to Metabolic Engineering for the fermentative production of biorenewable fuels and chemicals. This often entails the introduction of new enzymes, deletion of unwanted enzymes and efforts to fine-tune enzyme abundance in order to attain the desired strain performance. Enzyme performance can be quantitatively described in terms of the Michaelis-Menten type parameters Km, turnover number kcat and Ki, which roughly describe the affinity of an enzyme for its substrate, the speed of a reaction and the enzyme sensitivity to inhibition by regulatory molecules. Here we describe examples of where knowledge of these parameters have been used to select, evolve or engineer enzymes for the desired performance and enabled increased production of biorenewable fuels and chemicals. Examples include production of ethanol, isobutanol, 1-butanol and tyrosine and furfural tolerance. The Michaelis-Menten parameters can also be used to judge the cofactor dependence of enzymes and quantify their preference for NADH or NADPH. Similarly, enzymes can be selected, evolved or engineered for the preferred cofactor preference. Examples of exporter engineering and selection are also discussed in the context of production of malate, valine and limonene.


Introduction
In the time since Escherichia coli was first engineered to produce ethanol as its major fermentation product [1] and the coining of the term "metabolic engineering" in that same year [2,3], a variety of microbes have been engineered for the production of a wide range of products. These products include, but are not limited to, fuels [4], chemicals [5] and neutraceuticals [6]. Here we focus on the use of microbial biocatalysts to produce biorenewable fuels and chemicals.
Metabolic Engineering is defined as "the directed improvement of production, formation or cellular properties through the modification of specific biochemical reactions or the introduction of new ones with the use of recombinant DNA technology" [7]. Straightforward expression of a new pathway is often sufficient for production of the desired compound. However, an economically viable process requires that the target compound be formed at high titer (concentration), yield and rate, where the target values for these parameters can obviously vary according to the value of the product. Deletion of competing pathways and increasing expression of the target pathway are standard tools for increasing titer, yield and rate [8]. A variety of tools exist for increasing gene and enzyme abundance including the use of inducible promoters [9][10][11][12], engineering or evolution of the promoter and ribosome binding region [13], mutation of transcriptional regulators [14], transcript stabilization [15], optimization of translation initiation [16], codon optimization [17,18] and others [8,19,20].
However, pathway function is determined by more than just the expression level of the constituent enzymes. The affinity of an enzyme for substrate(s) and/or cofactor(s), catalytic efficiency, cofactor requirements and allosteric regulation, as well as substrate uptake and product export, are all important drivers of flux through the desired pathway. Here we describe key examples where knowledge and manipulation of these parameters have enabled increased process performance in terms of the production of biorenewable fuels and chemicals. Note that it is often difficult to determine a priori which enzyme is limiting biocatalyst performance. There are several recent examples of methods for identifying problematic, or "bottleneck" enzymes [21][22][23][24][25]; this topic is not addressed in this review.

Overview of Michaelis-Menten Parameters
The enzymatic conversion of substrate S to product P by enzyme E can be represented by the following simplified two-step reaction schematic (rxn 1) Engineering for the fermentative production of biorenewable fuels and chemicals. This often entails the introduction of new enzymes, deletion of unwanted enzymes and efforts to fine-tune enzyme abundance in order to attain the desired strain performance. Enzyme performance can be quantitatively described in terms of the Michaelis-Menten type parameters Km, turnover number kcat and Ki, which roughly describe the affinity of an enzyme for its substrate, the speed of a reaction and the enzyme sensitivity to inhibition by regulatory molecules. Here we describe examples of where knowledge of these parameters have been used to select, evolve or engineer enzymes for the desired performance and enabled increased production of biorenewable fuels and chemicals. Examples include production of ethanol, isobutanol, 1-butanol and tyrosine and furfural tolerance. The Michaelis-Menten parameters can also be used to judge the cofactor dependence of enzymes and quantify their preference for NADH or NADPH. Similarly, enzymes can be selected, evolved or engineered for the preferred cofactor preference. Examples of exporter engineering and selection are also discussed in the context of production of malate, valine and limonene. In this manner, v reflects the overall velocity (rate) of a given reaction as a function of substrate concentration cs, concentration of active enzyme cET, Michaelis constant Km, and turnover number kcat. Note that cE and cE-S represent the concentration of enzyme in the unbound and substrate-bound states, respectively. This formulation was first described in 1913 and has recently been translated into English and revisited with some interesting insights [26].
vmax and Km are the two most-commonly quantified values for a particular enzyme-substrate pair, as they can be determined by measuring reaction rate v over a range of substrate concentrations. When the substrate concentration becomes saturating, the reaction velocity will approach vmax. Km is the substrate concentration at which the reaction velocity is one half of vmax. Thus, Km reflects the affinity of an enzyme for its substrate, with a lower value indicating a stronger affinity. kcat, also known as the turnover number, represents the speed at which a particular enzyme can convert substrate to product; higher values represent a faster-acting enzyme. The theoretical upper limit of kcat is generally considered to be in the range of 10 6 -10 7 s -1 [27]. The ratio of kcat/Km is often referred to as the 'specificity constant' and used to compare the activity of a particular enzyme with multiple substrates; the theoretical upper limit of kcat/Km is estimated as 10 8 -10 9 M -1 s -1 [27]. This ratio is also said to reflect an enzyme's catalytic efficiency, though there are concerns about the validity of this term [28]. A recent compilation and analysis of data for more than 1,800 enzymes reported that median values for kcat, Km and kcat/Km are 13.7 s -1 , 130 µM and 125,000 M -1 s -1 , respectively [27].

Impact of Michaelis-Menten parameters on biocatalyst performance
Km values are especially important at metabolic nodes, where multiple enzymes compete for one substrate. When engineering E. coli for homoethanol production, Ohta et al [1] introduced pyruvate decarboxylase (PDC, Km pyruvate = 0.4 mM) into an existing pyruvate node, where other enzymes (pyruvate formate lyase, Km pyruvate = 2.0 mM; lactate dehydrogenase, Km pyruvate = 7.2 mM) were already competing for pyruvate. However, PDC had the lowest Km pyruvate and was able to effectively out-compete the other enzymes, enabling production of ethanol at 95% of the theoretical yield without deletion of the competing enzymes [1,29].
Metabolic cofactors, such as ATP and NAD(P)H can be considered among the most highly-connected metabolic nodes. In these cases, enzymes with a high affinity (low Km) for these valuable metabolites can be problematic for a well-performing strain if these enzymes are not involved in product formation. For example, E. coli's YqhD is an NADPH-dependent promiscuous aldehyde reductase that normally functions to reduce the toxic aldehydes that are produced by lipid peroxidation [30]. It has a Km NADPH of 0.8 µM [29,31]. However, in the presence of exogenous aldehydes, such as the furfural that can be relatively abundant in hydrolyzed biomass, YqhDmediated furfural reduction results in depletion of the NADPH pool [31,32]. This depletion is so extreme that there is insufficient NADPH for sulfite reductase (Km NADPH = 80µM) to produce the hydrogen sulfide required for production of cysteine [31,32]. This depletion of cysteine results in a lack of growth and therefore a lack of product formation. Elimination of this NADPH depletion via silencing or removal of yqhD results in increased furfural tolerance, both in terms of biocatalyst growth and product formation [31,32].
A high Km value can be problematic when it results in incomplete substrate utilization. A demonstration of this problem is the levoglucosan kinase (LGK) enzyme. Levoglucosan is an anhydrosugar produced during biomass pyrolysis that can be utilized with the same ATP and redox demand as glucose [33]. However, LGK has a relatively high Km levoglucosan of 75 mM [34]. The problem incurred by this Km value is reflected by the fact that a substantial amount of levoglucosan is left unutilized, resulting in a loss in product formation [33]. This problem could potentially be alleviated by modifying the enzyme to have a lower Km; examples of this type of modification are described below.
Improving Km, kcat and kcat/Km to improve strain performance As highlighted above, the use of enzymes with appropriate Michaelis-Menten parameters can enhance the performance of a microbial biocatalyst. The question becomes how to obtain enzymes with the appropriate parameters. In some cases, there exist characterized isozymes for a given enzymatic reaction. However, in many cases it becomes necessary to generate variants of an enzyme in order to obtain the desired function. These variants can either be generated by evolution [39][40][41][42] or through rational design [43,44].
Chen et al [21] recently provided an excellent example of the how improving the Michaelis-Menten parameters of one enzyme can improve process performance. Having identified transaldolase (TAL), a component of the non-oxidative branch of the pentose phosphate pathway, as the enzyme limiting the utilization of pentose sugars by ethanol-producing Pichia stipitis, Chen et al set out to generate improved variants of this enzyme through directed evolution and screening. The most promising variant (Gln263Arg) had a two-fold decrease in Km F6P and 3-fold increase in kcat F6P , resulting in a 5-fold increase in the kcat/Km ratio (Table 1). When the fermentative performance of the strain expressing this improved enzyme was compared to the strain with the original TAL enzyme, an increase in both the xylose consumption rate and ethanol production rate were observed (Table 1).
As part of an engineered pathway for isobutanol production, the Lactococcus lactis alcohol dehydrogenase (AdhA) was demonstrated as effective for converting isobutyraldehyde to isobutanol, though the Km value was higher than other existing enzyme alternatives [45]. Screening of nearly 4,000 random variants identified amino acid changes that were useful in lowering the Km. Three of these changes were engineered into a final mutant termed RE1 [35]. RE1 showed a 10-fold decrease in Km, 4-fold increase in kcat and thus 40-fold increase in kcat/Km and enabled a nearly 2-fold increase in isobutanol titer (Table 1).

Cofactor requirements
The above example of YqhD-mediated drainage of NADPH highlights the importance of this valuable cofactor. Relative to the glycolysis-associated NADH, NADPH can be relatively scarce. Therefore pathway designs in which NADPH is required for production of the target compound can suffer from a lack of NADPH availability. One method for dealing with this problem is to use transhydrogenase enzymes to intercovert NADH and NADPH [32,35,[46][47][48][49]. Another method is to exchange NADPH-dependent enzyme activity for NADH-dependent enzyme activity, either by Enzyme optimization for biorenewables production selecting an appropriate isozyme or by modifying the NADPHdependent enzyme. This exchange of NADPH/NADH dependency was recently reviewed [50] and a few key examples are described here.
The reduction of furfural to the less-inhibitory furfuryl alcohol is performed by the NADPH-dependent aldehyde reductase YqhD in wild-type E. coli [31]. Deletion or silencing of yqhD increases tolerance of approximately 1.0 g/L of furfural by sparing NADPH for biosynthesis [31]. However, this results in a lack of detoxification of furfural to furfuryl alcohol. Wang et al [37] addressed this problem by increasing expression of the NADH-dependent furfural reductase FucO, enabling a 50% increase in furfural tolerance. Note that the Km furfural of FucO is 0.4 mM, enabling it to outcompete YqhD's m furfural of 9 mM [31,37] (Table 1), further highlighting the importance of using enzymes with appropriate Km values. Watanabe et al [36] used an enzyme modification approach to switch the P. stipitis xylose reductase (PsXR) enzyme from a preference for NADPH to a preference for NADH (Table 1). This cofactor switching was motivated by the goal to maintain redox balance with the NAD+-dependent xylitol dehydrogenase, the enzyme which is immediately downstream of PsXR in the conversion of xylose to ethanol. The original enzyme had a 10-fold higher Km NADH relative to Km NADPH , reflecting a 10-fold lower affinity for NADH, though the kcat NADH was about 30% lower than kcat NADPH . By contrast, the evolved enzyme had a 10-fold lower Km NADH relative to Km NADPH and a 25-fold lower kcat NADPH relative to kcat NADH . This combination of changes in Km and kcat means that the evolved enzyme has a 3-fold higher (kcat/Km) NADH relative to (kcat/Km) NADPH , relative to the original enzyme's 20-fold higher (kcat/Km) NADPH relative to (kcat/Km) NADH . Simply put, the original enzyme's preference for NADPH was evolved to a preference for NADH, where this preference is reflected in the Km, kcat and kcat/Km values. Use of this evolved PsXR enzyme in S. cerevisiae resulted in increased ethanol production from xylose and decreased formation of the side product xylitol (

Addressing enzyme inhibition (Ki)
The Michaelis-Menten parameters described above all relate to an active enzyme, its affinity for the substrate and its speed in forming product. However, many enzymes have at least some degree of posttranslational allosteric regulation which serves to fine-tune enzyme activity in response to the abundance of key metabolites. This activity control occurs in the form of both activation and inhibition; here we focus on examples of enzyme inhibition.
As with the Michaelis-Menten model of enzyme activity, there also exist quantitative models for enzyme inhibition. These describe both competitive and non-competitive inhibition. In standard cases of competitive inhibition the inhibitor (I) competes with the substrate for binding to the active site, resulting in the additional reaction (rxn 2) to the simplified schematic described above. This reversible binding is described with the inhibition parameter Ki, which reflects the affinity of the enzyme for the inhibitor according to Ki = cEcI/cE-I Note that ci is the concentration of the inhibitor.
By contrast, in standard cases of non-competitive inhibition, the inhibitor binds to a site distinct from the active site and this binding induces a conformational change in the enzyme that decreases enzymatic activity. Thus, in addition to Rxn 2, it is possible for the inhibitor to bind to the E-S complex; this E-S-I complex can revert to E-I by dissociation of the substrate or possibly proceed to product formation, though at a much slower rate than the E-S complex in the absence of bound inhibitor. This non-competitive inhibition is modeled Competitive and non-competitive inhibition can be distinguished by the use of Lineweaver-Burk plots, which are not discussed here. The relevance of these equations to the current work is the fact that enzyme sensitivity to inhibition can be quantified by the parameter Ki, where a higher value indicates decreased sensitivity to inhibition.
This regulatory control of enzyme activity presumably serves to balance metabolic flux distribution and can be problematic when one desires to produce a single metabolic product at high concentration and yield, as this can conflict with the microbial need to balance production of biomass constituents. Thus, enzyme inhibition is a problem that often needs to be addressed in the fermentative production of biorenewable fuels and chemicals.
As with the other enzyme properties described above, the problem of enzyme inhibition can often be addressed by selecting from existing characterized isozymes. For example, Shen et al [51] observed relatively low metabolic flux through their engineered 1-butanol and 1-propanol pathways that was presumably due to inhibition of homoserine dehydrogenase (ThrA) by threonine, where threonine is an intermediate of the engineered pathway downstream of ThrA.
Replacement of the native E. coli ThrA with a feedback-resistant mutant (ThrA fbr ) resulted in a more than 3-fold increase in the final titers of 1-butanol and 1-propanol (Table 2) [51]. Similarly, the use of feedback-resistant mutants of 3-deoxy-D-arabino-heptulosonate-7phosphate (DHAP) synthase (AroG) and chorismate mutase/ prephenate dehydrogenase (TyrA) each increased tyrosine production more than 10-fold when expressed individually (Table 2) and enabled even further increases in production when expressed simultaneously (data not shown) [52]. Note that AroG performs the first dedicated step of the tyrosine biosynthesis pathway and is inhibited by Lphenylalanine. TyrA performs the next-to-last step in tyrosine biosynthesis and is inhibited by tyrosine.
Biomass formation by ethanologenic E. coli KO11 was limited in defined growth media due to NADH-mediated inhibition of citrate synthase, resulting in limitation of the biomass precursor alphaketoglutarate and limitation of overall growth and therefore product formation [53]. Replacement of the native E. coli citrate synthase with an NADH-resistant isozyme from Bacillus subtilis resulted in a 50% increase in growth and ethanol production in the desired growth condition [53].
An alternative approach to replacing an inhibition-sensitive enzyme with an inhibition-resistant isozyme is to modify the original enzyme so that the inhibition sensitivity is reduced or eliminated. This approach was taken by Kim et al [54,55] in regards to pyruvate dehydrogenase (PDH). The PDH complex is normally subject to inhibition by NADH; presumably this serves to balance generation of NADH in glycolysis and the subsequent regeneration of NAD + through fermentative pathways. The lack of PDH activity during fermentative growth, when NADH is abundant, has resulted in reliance on recombinant expression of the Zymomonas mobilis PET pathway for ethanol production by E. coli [29]. However, mutations within the dihydrolipoamide dehydrogenase (LPD) subunit of PDH reduced this feedback sensitivity approximately 10-fold, resulting in a 10-fold improvement of ethanol production without dependence on the Z. mobilis PET pathway (Table 2).
Appropriate transporters for substrate uptake and product export Finally, effective pathway flux requires the presence of appropriate uptake systems for the desired substrate and effective means of excreting or sequestering the product compound.
Transporters that are discovered when searching for importers can also be useful as exporters. The Schizosaccharomyces pombe malate transporter Mae1 (SpMae1p) was first demonstrated as useful for malate uptake by S. cerevisiae [59], but was also able to support a 10fold increase in the malate titer achieved by a malate-producing S. cerevisiae [60].
Product export becomes increasingly important when the target compound is inhibitory to the microbial biocatalyst. Here we discuss two examples of the selection of appropriate exporters in order to improve the microbial production of an inhibitory compound.
Despite the fact that it is naturally produced by E. coli and is necessary for protein translation, the branched-chain amino acid valine has long been known to be toxic to E. coli [ Alcanivorax borkumensis, enabled an approximately 50% increase in limonene titer when expressed in an E. coli strain engineered for limonene production.
These three examples highlight the use of native transporters, recombinant transporters and engineered/evolved transporters to increase production of biorenewable fuels and chemicals.

Summary and Outlook
Here we have highlighted recent examples of how improvement of enzyme parameters, as reflected in the Michaelis-Menten-type parameters Km, kcat and Ki, can improve the fermentative performance of a microbial biocatalyst. Each of the examples that we have described represent improved biocatalyst performance in the context of production of biorenewable fuels and chemicals. While the Michaelis-Menten is a simplified model of enzyme kinetics [26,[65][66][67][68], these parameters provide a useful quantification of enzyme properties that can be enormously valuable to other researchers when selecting enzymes during pathway design. Databases such as BRENDA [69] are a useful repository of this type of information. However, it is critical that researchers continue to quantify and report these parameters for engineered or evolved enzymes so that others can make informed choices and use these enzymes when appropriate.
There are some enzymes that are tantalizing targets for improvement in order to increase production of biorenewable fuels and chemicals, yet these enzymes remain remarkably intractable to such improvement. The most well-known example is photosynthesis pathway enzyme Rubisco, which has a low catalytic efficient and poor substrate specificity [27,70]. A recent cross-species analysis of the evolutionary landscape for Rubisco has provided interesting insight into why it has proven so difficult to improve its function [27]. Thus, despite the fact that we have described many successful examples of improving strain performance by improving enzyme parameters, it should be noted that enzyme improvement is not always feasible. Note that others have managed to obtain (slightly?) improved Rubisco mutants [70,71].