- Email: [email protected]
Combinatorial engineering of microbes for optimizing cellular phenotype
Available online at www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Christine Nicole S Santos and Gregory Stephanopoulos Although random mutagenesis and screening and evolutionary engineering have long been the gold standards for strain improvement in industry, the development of more sophisticated recombinant DNA tools has led to the introduction of alternate methods for engineering strain diversity. Here, we summarize several combinatorial cell optimization methods developed in recent years, many of which are more amenable to phenotypic transfer and more efficient in probing greater dimensions of the available phenotypic space. They include tools that enable the finetuning of pathway expression (synthetic promoter libraries, tunable intergenic regions (TIGRs)), methods for generating randomized knockout and overexpression libraries, and more global techniques (artificial transcription factor engineering, global transcription machinery engineering, ribosome engineering, and genome shuffling) for eliciting complex, multigenic cellular properties. Addresses Department of Chemical Engineering, Massachusetts Institute of Technology, Room 56-469, Cambridge, MA 02139, United States Corresponding author: Stephanopoulos, Gregory ([email protected])
Current Opinion in Chemical Biology 2008, 12:168–176 This review comes from a themed issue on Biocatalysis and Biotransformation Edited by Stephen G. Withers and Lindsay Eltis Available online 29th February 2008 1367-5931/$ – see front matter # 2008 Elsevier Ltd. All rights reserved. DOI 10.1016/j.cbpa.2008.01.017
been successful in many applications, it was established early on that the interconnectivity and sheer complexity of biological networks often preclude the recognition of simple genotype–phenotype relationships to guide these modifications. Indeed, a single genetic perturbation often has a variety of unpredictable secondary responses within the cell. In a similar vein, the performance of biosynthetic pathways frequently depends on distal genes through kinetic and regulatory interactions whose origins are poorly understood [3,4]. Finally, as an added complication, engineering a complex phenotype may call for the simultaneous modulation of several of these potentially unknown factors [5]. Such challenges led to the development of a new concept called ‘inverse metabolic engineering’ (IME) for cell optimization. This methodology involves three main steps: (1) the construction or identification (by selection) of strains possessing a desired cellular phenotype, (2) the evaluation and determination of genetic and/or environmental factors that confer the phenotype, and (3) the transfer of this phenotype to another strain through direct modifications of the identified genetic and/or environmental factors [6]. Various ‘-omics’ approaches established in the past several years have greatly facilitated the analysis of identified strains and have been the subject of other excellent reviews [7,8]. Here, we focus on recently developed techniques for the generation of strains possessing a phenotype of interest, which, in many cases, remains a significant bottleneck of the IME approach. Owing to the difficulty of predicting these complex genotype–phenotype relationships, many of these methods are combinatorial in nature, that is, they are based on generating genetic (and hence, phenotypic) diversity in a population followed by screening and selection for improved phenotypes.
Introduction In the past 15 years, metabolic engineering has emerged as the discipline that utilizes modern genetic tools for the construction of organisms capable of fuel and chemical production. It was formalized partly from the advent of more sophisticated recombinant DNA techniques that allowed for the targeted genetic manipulation of microbes, either for the modification of existing biochemical reactions or the introduction of completely heterologous pathways. As such, the earliest examples in the field focus on engineering cellular phenotype using rational modifications (typically gene deletions/overexpressions and pathway deregulation) based on existing stoichiometric, kinetic, and regulatory knowledge of a system [1,2]. Although this ‘rational design’ approach has Current Opinion in Chemical Biology 2008, 12:168–176
Fine-tuning expression levels of pathway components It is now broadly accepted that most metabolic pathways are not limited by a single rate-limiting step and that optimized pathways require the balanced expression of several enzymes [9,10]. Without such coordination, metabolic imbalance can lead to the accumulation of gene products or intermediate metabolites with potentially cytotoxic effects or, in some cases, may result in the depletion of a metabolite needed for normal cell growth. Furthermore, the overexpression of genes/proteins often results in an undue metabolic burden on the cell [11]. Thus, many recent metabolic engineering efforts have focused on the development of tools for fine-tuning www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 169
expression to facilitate the search for an optimum balance between pathway expression and cell viability. Synthetic promoter libraries
One such approach involves the creation of libraries of constitutive promoters for modulating gene expression in a variety of prokaryotic (Escherichia coli, Lactococcus lactis, and Lactobacillus plantarum) and eukaryotic (Saccharomyces cerevisiae) organisms. The novelty of these promoters lies in their ability to cover a wide range of expression levels with incremental changes in activity, thus providing the full continuum of gene expression [12,13,14,15,16–19]. Such a tool eliminates reliance on inducible systems that are not only difficult to control but are also cost-prohibitive when considered for large-scale industrial processes. There are currently two established techniques for generating these promoter libraries. The first method, developed by Jensen and Hammer [14,15], is performed by modulating the spacer sequences in between the 10 and 35 regions of the promoter while keeping the consensus regions intact. When applied toward the construction of a promoter library in L. lactis, promoters that possessed sequence variations in the spacer region exhibited a 400-fold range in activity and were shown to alter the expression of pfk and the las operon upon chromosomal delivery [19]. Since then, this technique has been used to create constitutive promoter libraries in E. coli and L. plantarum [13,18], as well as a library of synthetic stationary-phase/stress promoters featuring a range of expression levels and induction times upon exit out of exponential growth [16]. A second method for constructing promoter libraries has been developed in our lab and utilizes error-prone PCR to introduce point mutations into well-characterized native promoters. Proof-of-concept studies have been conducted with the PLtetO-1 promoter for the generation of an E. coli promoter library and the TEF promoter for a S. cerevisiae counterpart [12,17]. The E. coli library features 22 promoters spanning a wide range of activity, which were extensively characterized using three separate orthogonal metrics—GFP fluorescence, mRNA transcript levels, and chloramphenicol minimum inhibitory concentrations. Additionally, each promoter was analyzed on a single-cell level to ensure homogeneity of expression among clonal populations. The utility of such a tool was demonstrated by placing the dxs gene under the control of several different promoters and examining its effects on lycopene production through the nonmevalonate pathway. Interestingly, the highest lycopene titer was seen for an intermediate level of expression, again highlighting the importance of balancing pathway overexpression and cell metabolism [12]. Applications of these promoter libraries have thus far been limited to the manipulation of a single gene or www.sciencedirect.com
operon at a time [12,17,19]. A clear disadvantage of such an approach is that multiple rounds of cell modifications are required to find the optimum expression levels of all genes in a given pathway. Furthermore, it is likely that changing the expression level of one gene will result in different optimum levels for the other genes, particularly when such a change results in a substantial increase in flux (indicating a shift in the location of the pathway bottleneck). Despite the narrow focus of these initial studies, synthetic promoters hold great potential in enabling the simultaneous transcriptional variation of multiple genes in a pathway. Expression libraries can be created simply by placing each gene under the control of several promoters and then combinatorially reassembling the pathway from a mixed pool of promoter-gene constructs (Figure 1a). Recombinant DNA techniques, such as chain reaction cloning, can facilitate the assembly of these libraries [20], and high-throughput screens can then be implemented to select the best performing mutants. Engineering post-transcriptional control
Although the previous examples have focused on exerting control at the transcriptional level, it is also possible to tune the expression of multiple genes in synthetic operons by employing post-transcriptional mechanisms. This was demonstrated in a study featuring a library of tunable intergenic regions (TIGRs) consisting of several control elements, including mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences (Figure 1b). By incorporating these TIGRs in between the coding regions of a transcript, the authors were able to differentially alter the processes of transcription termination, mRNA stability, and translation initiation [21]. In particular, the inclusion of RNAse cleavage sites allowed them to decouple the stabilities of each coding region, thus facilitating the independent variation of expression levels. This TIGR approach was utilized in optimizing flux through a heterologous mevalonate pathway in E. coli through manipulation of the operon containing atoB, HMGS, and tHMGR. A sevenfold increase in mevalonate production was observed via the counterintuitive mechanism of HMGS and tHMGR activity reduction. The ability to recover these unexpected yet phenotypically beneficial changes points to the power of these combinatorial search strategies.
Global perturbations for phenotypic improvement Although fine-tuning the relative levels of gene expression in a specific pathway may lead to improved phenotypes, the majority of applications require engineering techniques that can alter the entire cellular milieu in a less targeted and more global fashion. This concept of strain randomization and selection/screening is not a foreign one and is a methodology that has been employed extensively for the generation of industrial strains. For example, ‘classical strain improvement’ relies on mutaCurrent Opinion in Chemical Biology 2008, 12:168–176
170 Biocatalysis and Biotransformation
Figure 1
Metabolic engineering tools for fine-tuning pathway expression. (a) Promoter libraries can be used to optimize entire pathways by placing individual genes under the control of a series of promoters and then combinatorially reassembling the pathway. (b) Tunable intergenic region (TIGR) libraries consisting of mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences can be used to differentially control posttranscriptional processing of genes in a synthetic operon (adapted from reference [21]).
genic agents such as nitrosoguanidine or UV light to introduce genetic diversity (in the form of point mutations) in a population of strains [22]. Other techniques, such as ‘evolutionary engineering,’ apply a selection pressure and allow the organism to naturally evolve in a chemostat, thus relying on the cell’s inherent capacity to introduce adaptive mutations [23]. Although these methods have been successful in engineering several phenotypes, they suffer from being laborious and timeconsuming, often requiring multiple rounds of screening and mutagenesis and, for the case of evolutionary engineering, extensive cultivation periods [24,25]. Furthermore, because silent mutations tend to accumulate, it becomes difficult to identify which mutations are necessary for conferring a particular phenotype. A full ‘-omics’ analysis of the strain may be needed to understand the basis for these phenotypic differences [7,8], and full genome sequencing/reconstructions may be necessary to initiate phenotypic transfer into a new strain [26,27]. Because of the inherent disadvantages with classical strain improvement and evolutionary engineering, several groups have focused their attentions on designing novel methods for generating phenotypic diversity in a population of strains. Here, we review several alternatives to these more established classical methods that are traceCurrent Opinion in Chemical Biology 2008, 12:168–176
able (thus making them amenable to phenotypic transfer) and are more effective in sampling a larger portion of the phenotypic space available to a cell. Random knockout and overexpression libraries
Methods for predicting knockout and overexpression targets rely heavily on a priori knowledge of the biochemical network and often take the form of stoichiometric models and algorithms, such as flux balance analysis [28] and minimization of metabolic adjustment [29]. However, a variety of recombinant DNA techniques are also available for generating these genetic changes in a randomized fashion, thus introducing the possibility of uncovering regulatory, kinetic, or unknown/poorly understood targets not encompassed in the model. For example, transposon mutagenesis, which involves the transposase-mediated integration of an antibiotic marker cassette into the genome, is an effective tool for creating random knockout libraries (Figure 2a). Although previously prone to insertions into preferred ‘hot spot’ regions, this system has now been optimized for effective and unbiased integration into the entire genome [30,31]. Other knockout approaches have also been developed, including one based on antisense RNA inhibition and promoter interference in Candida albicans. For this system, a library of antisense complementary DNA (cDNA) www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 171
fragments is placed under the control of the GAL1 promoter which, when induced, leads to the selective inhibition of gene expression [32]. Overexpression studies are often conducted simply by transforming the parental strain with a plasmid-encoded genomic library (Figure 2b) [33–35]. The advantage of these knockout/overexpression approaches is that the phenotype can be easily linked back to a single genetic perturbation. The presence of an integration cassette within the transposon libraries allows one to identify the region of insertion through a modified ‘Thermal Asymmetric Interlaced PCR’ procedure [3], and the genomic pieces selected for overexpression can be identified by plasmid sequencing [34]. In addition, many groups have developed microarray-based techniques to increase the throughput of target identification, a particularly important parameter when several unique genes are found to confer phenotypic improvements. The incorporation of a T7 promoter within a knockout cassette allows for the creation and analysis of cDNA libraries at the point of insertion and has been demonstrated to be successful in identifying gene clusters that are negatively selected during E. coli growth competition experiments [30]. More recently, a method for a multiscale analysis of library enrichment (SCALEs) has been developed, which facilitates the identification of single gene or whole operon overexpressions that contribute to a phenotype of interest [33]. Although these knockout and overexpression approaches are often studied in the context of growth phenotypes, they have also been shown to be remarkably successful for the case of lycopene production in E. coli. A random knockout search in the background of a previously constructed high-lycopene-producing parental strain led to the identification of three targets – hnr, yjfP, and DpyjiD – that resulted in further increases in lycopene content. It is important to note that each of these genes encode for proteins with regulatory or unknown functions and thus would not have been predicted by a simple stoichiometric model [3]. In a later study, the introduction of a plasmidencoded genomic library in the background of a mutant containing systematically identified and transposonbased knockouts yielded additional overexpression targets (rpoS, yjiD, ycgW) conferring a lycopene-producing phenotype [34]. This clearly demonstrates the benefits of combining these two complementary approaches for phenotypic improvement. Targeting transcription/translation for introducing global perturbations
Sequential approaches for cellular optimization are not always capable of finding the global optimum of a phenotype owing to the often-convoluted form of the metabolic landscape [3]. As a result, accessing these complex phenotypes from a single round of strain diversification www.sciencedirect.com
requires techniques for introducing genetic changes with known pleiotropic effects. Most of the approaches that have been developed target proteins that have widespread regulatory functions in important cellular processes such as transcription and translation. Thus, the alteration of these proteins holds enormous potential in generating widely diverse libraries for strain improvement studies. Artificial transcription factor libraries based on zinc-finger motifs
Zinc fingers are highly specific DNA-binding domains that recognize three base pair sequences and are found in a variety of transcriptional regulatory proteins. A single transcription factor may contain several of these motifs, which can be assembled in a highly modular fashion. This modularity, coupled with the ability to attach additional effector (activator/repressor) domains, makes zinc-finger proteins powerful tools for introducing transcriptional diversity (Figure 2c). The utility of this approach was first demonstrated in S. cerevisiae through the creation of a randomized plasmid library of over 105 three-finger and four-finger proteins. Transformation of these plasmids and subsequent screening led to the isolation of zincfinger proteins that were capable of conferring a variety of tolerance phenotypes in S. cerevisiae, including resistance to heat treatment, osmotic pressure, and the antifungal drug ketoconazole [36]. In this study, 11 different zincfinger proteins were found to grant varying degrees of ketoconazole resistance. Significantly, coexpression of two of these factors often led to enhanced resistance with the best performing double expression strain exhibiting a 1000-fold enhancement in the resistance phenotype. The most surprising result, however, was the discovery that swapping the effector domain of a zinc-finger protein selected for in one phenotype (from activator to repressor or vice versa) could confer the opposite phenotype. Although the broad applicability of such a result needs to be examined, it introduces the possibility of utilizing this technique for situations in which the opposite phenotype is more amenable to screening. In a later study conducted in E. coli, 40 different zinc fingers were used to construct libraries of 6.4 104 and 2.6 106 three-fingered and four-fingered proteins, respectively. Despite the absence of an activator or repressor domain in these proteins, the authors were successful in identifying a zinc-finger protein conferring a thermotolerance phenotype [37]. After measuring the transcript levels of genes containing potential binding sites, the thermotolerance trait was shown to be a result of the downregulation (but not complete inhibition) of the ubiquinone biosynthesis gene ubiX. Thus, these zincfinger proteins are also capable of directing subtle changes in gene expression that would otherwise remain inaccessible through the previously discussed knockout and overexpression approaches. Current Opinion in Chemical Biology 2008, 12:168–176
172 Biocatalysis and Biotransformation
Figure 2
Current Opinion in Chemical Biology 2008, 12:168–176
www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 173
Global transcription machinery engineering (gTME)
Global transcription machinery engineering (gTME) is a powerful new tool that enables a complete reprogramming of the cellular transcriptome. It does so through the targeted mutagenesis (via error-prone PCR) of select components of the transcriptional machinery, particularly those that are involved in DNA sequence recognition and thus dictate the promoter preferences of RNA polymerase (Figure 2d). In prokaryotic systems, sigma factors bind to the promoter regions of genes with varying degrees of affinity and help to preferentially recruit the RNA polymerase holoenzyme to initiate transcription. Thus, slight variations in these proteins have the potential to greatly affect the subset of genes that are bound by RNA polymerase and expressed. The introduction of plasmidencoded mutants of the principal sigma factor in E. coli (s70) resulted in strains optimized for a variety of phenotypes, including ethanol tolerance, lycopene overproduction, and simultaneous tolerance to sodium dodecyl sulfate and ethanol [38]. These phenotypes, some of which were previously unattainable through traditional methods of strain improvement, were generated in a highly efficient manner and required only a few rounds of screening to promote substantial phenotypic improvements. This methodology has also been shown to be effective in S. cerevisiae through the mutation of the TATA-binding protein (encoded by SPT15) and one of the TATA-binding protein-associated factors (TAF25). With this approach, Alper et al. were successful in isolating a strain that acquired a high degree of tolerance to high glucose/high ethanol conditions and exhibited a 70% improvement in the volumetric productivity of ethanol. We also note that the phenotypes of cells harboring the mutant sigma (or transcription) factors were indistinguishable from that of the control in the absence of stress [39]. This work has recently been extended to another prokaryotic organism, L. plantarum, and this study reveals several important findings about the gTME approach. Through a direct comparison of gTME with the more established method of nitrosoguanidine mutagenesis, the authors were able to demonstrate that the former is significantly more effective in generating phenotypic diversity as measured by the complex phenotype of growth/colony size. This increased diversity was in turn found to be linked to a greater probability of finding a
desirable mutant during selection/screening of a library for malic acid tolerance. Finally, phenotypic variation was shown to be well correlated with the frequency of error generation during the mutagenic PCR reaction, with higher mutation frequencies leading to a wider range of phenotypes [40]. This last point is particularly important, since it offers an additional knob for controlling this system and adapting it to various applications. Ribosome engineering
Modulating ribosomal proteins and rRNA is an alternative way for altering bacterial gene expression at the level of translation. This technique, termed ‘ribosome engineering,’ begins with the introduction of mutations that confer resistance to a variety of antibiotics that attack the ribosome, such as streptomycin, gentamicin, kanamycin, and chloramphenicol. Selection on different concentrations of antibiotics facilitates the isolation of strains with a wide variety of mutations in known ribosomal components (Figure 2e). Although the mechanism leading to phenotypic variation is not well understood, the changes are hypothesized to be a result of the activation of a ‘stringent response’ that leads to elevated levels of protein synthesis during the late growth phase and is beneficial for a select number of specialized phenotypes. Such a technique resulted in antibiotic-producing properties in several Streptomyces organisms, enzyme (a-amylase and protease) overproduction in Bacillus subtilis, and improved tolerance to aromatic compounds, such as 4-hydroxybenzoate, toluene, and m-xylene in Pseudomonas putida [41,42]. It has yet to be determined whether a more randomized mutagenesis approach will be capable of yielding strains optimized in other cellular phenotypes. However, because mutations are introduced directly into the chromosomal copies of these translational proteins, the majority of these changes will probably lead to an inhibition of growth and may ultimately limit the broader utilization of this technique. Genome shuffling
Whole-genome shuffling allows for the generation of combinatorial libraries of complex progeny from a few previously selected parental strains exhibiting subtle improvements in a desired property. The utilization of a method called recursive protoplast fusion ensures a thorough shuffling of the genome, which then allows for the creation of strains that accumulate mutations with additive or even synergistic effects with respect to the
(Figure 2 Legend) Phenotypic diversification through whole-cell engineering techniques. (a) Random knockout libraries are created through the arbitrary insertion of a transposon cassette into the host genome, resulting in the disruption of the genetic element at the site of integration. (b) Random overexpression libraries are generated by transforming the host strain with a plasmid-encoded genomic library. (c) Artificial transcription factor (zincfinger protein) libraries are constructed by combinatorially assembling three or four zinc-finger motifs, followed by attachment of an activator (A) or repressor (R) domain. Expression of these proteins in a strain leads to changes at the level of transcription (adapted from reference [36]). (d) Global transcription machinery engineering (gTME) begins with the mutagenesis of a component of the transcription machinery (often in charge of DNA recognition and binding) by error-prone PCR. The introduction of this mutant constituent results in a complete alteration of the global transcriptome. (e) In ribosome engineering, strains are first selected on media containing antibiotics that target ribosomal proteins and rRNA. The surviving strains, which contain mutations in the translational machinery of a cell, exhibit changes in protein translation and hence, cellular phenotype. (f) Genome shuffling employs recursive protoplast fusion in order to create a library of cells containing shuffled genomic DNA from two or (often) more parental strains. www.sciencedirect.com
Current Opinion in Chemical Biology 2008, 12:168–176
174 Biocatalysis and Biotransformation
phenotype of interest (Figure 2f). Compared with classical strain improvement strategies, this technique has been shown to be more effective in producing strains with enhanced phenotypes. Impressively, when applied toward tylosin production in Streptomyces fradiae, two rounds of genome shuffling were found to be successful in achieving the same titers generated from 20 rounds of classical strain improvement. As stated aptly, ‘this is the difference between 24 000 assays and approximately 1 year of effort, and roughly 1 000 000 assays and 20 years of effort’ [43]. Since then, this tool has been used to engineer a myriad of other complex phenotypes, including improved acid tolerance in L. plantarum, pesticide degradation by Sphingobium chlorophenolicum, hydroxycitric acid production in Streptomyces sp. U121, and epithilone production by Sorgangium cellulosome [44– 47]. Thus, genome shuffling is clearly an important metabolic engineering tool for strain improvement, particularly when strains with intermediate properties have already been isolated.
Concluding remarks As demonstrated by all these approaches, many techniques are now available for the generation of optimized strains that can be studied within an IME framework. Each of these methods taps into its own subset of the phenotypic space available to a cell and works through mostly orthogonal mechanisms. It is therefore likely that combining these search strategies may yield synergistic effects. Early successes in engineering high levels of lycopene production through a combination of systematic/random knockout and gTME approaches point to the distinct possibility of such synergism [38]. As an extension to this thought, these strategies, when implemented in combination, may also be successful in selecting strains with multiple desirable characteristics. In such a scenario, the traceability of these introduced genetic perturbations would greatly facilitate the compilation of each individual mutation into a single ‘super-engineered’ strain. Although we now have access to a diverse collection of tools for the generation of libraries, these combinatorial approaches have thus far been limited to the engineering of only a few select phenotypes. As demonstrated by the body of literature in this area, the major thrusts have largely been on identifying genetic changes that alter the growth characteristics of a strain [30,33,35– 37,39,47], a phenotype that is easily accessed through serial subculturing or by cultivating libraries in a chemostat environment. Such strategies effectively enrich a mixed population with only strains exhibiting desirable growth properties, thus making strain selection a simple process. Lycopene production in E. coli is another major trait that has greatly benefited from a variety of these techniques. Although metabolite overproduction has long been known to be a difficult phenotype to screen, the red pigmentation of this carotenoid molecule makes it an Current Opinion in Chemical Biology 2008, 12:168–176
exception to the rule. Desirable clones that overproduce lycopene assume a red coloration and are therefore easily selected by visual inspection [3,48]. Despite these reported successes, the broader applicability of these combinatorial techniques for other systems of interest clearly depends on the availability of high-throughput screens for probing large (>105 variants) libraries. Although several studies have been successful in generating screens tailored for individual phenotypes such as tyrosine, mevalonate, and poly-3-hydroxybutyrate production [49–51], more modular and flexible platforms will be needed to access a wider array of cellular properties. It is only after such technological advances that the true potential of these combinatorial approaches can be realized.
Acknowledgements This work was supported by a National Science Foundation Graduate Fellowship (CNSS), Grant CBET-0730238, and the Singapore-MIT Alliance (SMA2).
References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Bailey JE: Toward a science of metabolic engineering. Science 1991, 252:1668-1675.
2.
Stephanopoulos G: Metabolic fluxes and metabolic engineering. Metab Eng 1999, 1:1-11.
3.
Alper H, Miyaoku K, Stephanopoulos G: Construction of lycopene-overproducing E. coli strains by combining systematic and combinatorial gene knockout targets. Nat Biotechnol 2005, 23:612-616. This study demonstrates the ability for random knockout approaches to uncover regulatory or unknown protein targets not encompassed in standard stoichiometric models. 4.
Stephanopoulos G, Alper H, Moxley J: Exploiting biological complexity for strain improvement through systems biology. Nat Biotechnol 2004, 22:1261-1267.
5.
Bailey JE: Lessons from metabolic engineering for functional genomics and drug discovery. Nat Biotechnol 1999, 17:616-618.
6.
Bailey JE, Sburlati A, Hatzimanikatis V, Lee K, Renner WA, Tsai PS: Inverse metabolic engineering: a strategy for directed genetic engineering of useful phenotypes. Biotechnol Bioeng 2002, 79:568-579.
7.
Bro C, Nielsen J: Impact of ‘ome’ analyses on inverse metabolic engineering. Metab Eng 2004, 6:204-211.
8.
Gill RT: Enabling inverse metabolic engineering through genomics. Curr Opin Biotechnol 2003, 14:484-490.
9.
Koffas MA, Jung GY, Stephanopoulos G: Engineering metabolism and product formation in Corynebacterium glutamicum by coordinated gene overexpression. Metab Eng 2003, 5:32-41.
10. Pitera DJ, Paddon CJ, Newman JD, Keasling JD: Balancing a heterologous mevalonate pathway for improved isoprenoid production in Escherichia coli. Metab Eng 2007, 9:193-207. 11. Glick BR: Metabolic load and heterologous gene expression. Biotechnol Adv 1995, 13:247-261. 12. Alper H, Fischer C, Nevoigt E, Stephanopoulos G: Tuning genetic control through promoter engineering. Proc Natl Acad Sci U S A 2005, 102:12678-12683. This study describes the multifaceted characterization and subsequent application of an E. coli promoter library generated by error-prone PCR mutagenesis of a PLl derivative. www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 175
13. De Mey M, Maertens J, Lequeux GJ, Soetaert WK, Vandamme EJ: Construction and model-based analysis of a promoter library for E. coli: an indispensable tool for metabolic engineering. BMC Biotechnol 2007, 7:34. 14. Jensen PR, Hammer K: Artificial promoters for metabolic optimization. Biotechnol Bioeng 1998, 58:191-195. 15. Jensen PR, Hammer K: The sequence of spacers between the consensus sequences modulates the strength of prokaryotic promoters. Appl Environ Microbiol 1998, 64:82-87. This study describes the creation of a synthetic promoter library for L. lactis based on randomization of the spacer region in between the 35 and 10 consensus sequences. 16. Miksch G, Bettenworth F, Friehs K, Flaschel E, Saalbach A, Twellmann T, Nattkemper TW: Libraries of synthetic stationaryphase and stress promoters as a tool for fine-tuning of expression of recombinant proteins in Escherichia coli. J Biotechnol 2005, 120:25-37. 17. Nevoigt E, Kohnke J, Fischer CR, Alper H, Stahl U, Stephanopoulos G: Engineering of promoter replacement cassettes for fine-tuning of gene expression in Saccharomyces cerevisiae. Appl Environ Microbiol 2006, 72:5266-5273.
mutants using subgenic-resolution arrays. Nat Biotechnol 2001, 19:1060-1065. 31. Alexeyev MF, Shokolenko IN: Mini-Tn10 transposon derivatives for insertion mutagenesis and gene delivery into the chromosome of gram-negative bacteria. Gene 1995, 160:59-62. 32. De Backer MD, Nelissen B, Logghe M, Viaene J, Loonen I, Vandoninck S, de Hoogt R, Dewaele S, Simons FA, Verhasselt P et al.: An antisense-based functional genomics approach for identification of genes critical for growth of Candida albicans. Nat Biotechnol 2001, 19:235-241. 33. Lynch MD, Warnecke T, Gill RT: SCALEs: multiscale analysis of library enrichment. Nat Methods 2007, 4:87-93. 34. Jin YS, Stephanopoulos G: Multi-dimensional gene target search for improving lycopene biosynthesis in Escherichia coli. Metab Eng 2007, 9:337-347. 35. Gill RT, Wildt S, Yang YT, Ziesman S, Stephanopoulos G: Genome-wide screening for trait conferring genes using DNA microarrays.. Proc Natl Acad Sci U S A 2002, 99:7033-7038.
19. Solem C, Jensen PR: Modulation of gene expression made easy. Appl Environ Microbiol 2002, 68:2397-2403.
36. Park KS, Lee DK, Lee H, Lee Y, Jang YS, Kim YH, Yang HY, Lee SI, Seol W, Kim JS: Phenotypic alteration of eukaryotic cells using randomized libraries of artificial transcription factors. Nat Biotechnol 2003, 21:1208-1214. This paper describes the use of zinc-finger motifs attached to effector (activator/repressor) domains to create libraries of artificial transcription factors. When expressed in S. cerevisiae, the authors were able to isolate strains exhibiting a number of tolerance phenotypes.
20. Pachuk CJ, Samuel M, Zurawski JA, Snyder L, Phillips P, Satishchandran C: Chain reaction cloning: a one-step method for directional ligation of multiple DNA fragments. Gene 2000, 243:19-25.
37. Park KS, Jang YS, Lee H, Kim JS: Phenotypic alteration and target gene identification using combinatorial libraries of zinc-finger proteins in prokaryotic cells. J Bacteriol 2005, 187:5496-5499.
21. Pfleger BF, Pitera DJ, Smolke CD, Keasling JD: Combinatorial engineering of intergenic regions in operons tunes expression of multiple genes. Nat Biotechnol 2006, 24:1027-1032. This paper employs libraries of post-transcriptional control elements, such as mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences, to differentially control the expression of genes in a synthetic operon. It is an excellent example of the potential of combinatorial searches for pathway optimization.
38. Alper H, Stephanopoulos G: Global transcription machinery engineering: a new approach for improving cellular phenotype. Metab Eng 2007, 9:258-267. This study outlines a new method for modulating the entire prokaryotic transcriptome through mutagenesis of DNA-recognition elements of the transcription machinery. This method for diversification was shown to be effective for several distinct classes of phenotypes.
18. Rud I, Jensen PR, Naterstad K, Axelsson L: A synthetic promoter library for constitutive gene expression in Lactobacillus plantarum. Microbiology 2006, 152:1011-1019.
22. Adrio JL, Demain AL: Genetic improvement of processes yielding microbial products. FEMS Microbiol Rev 2006, 30:187-214. 23. Sauer U: Evolutionary engineering of industrially important microbial phenotypes. Adv Biochem Eng Biotechnol 2001, 73:129-169. 24. Wisselink HW, Toirkens MJ, del Rosario Franco Berriel M, Winkler AA, van Dijken JP, Pronk JT, van Maris AJ: Engineering of Saccharomyces cerevisiae for efficient anaerobic alcoholic fermentation of L-arabinose. Appl Environ Microbiol 2007, 73:4881-4891. 25. Yomano LP, York SW, Ingram LO: Isolation and characterization of ethanol-tolerant mutants of Escherichia coli KO11 for fuel ethanol production. J Ind Microbiol Biotechnol 1998, 20:132-138. 26. Herring CD, Raghunathan A, Honisch C, Patel T, Applebee MK, Joyce AR, Albert TJ, Blattner FR, van den Boom D, Cantor CR et al.: Comparative genome sequencing of Escherichia coli allows observation of bacterial evolution on a laboratory timescale. Nat Genet 2006, 38:1406-1412. 27. Ikeda M, Ohnishi J, Hayashi M, Mitsuhashi S: A genome-based approach to create a minimally mutated Corynebacterium glutamicum strain for efficient L-lysine production. J Ind Microbiol Biotechnol 2006, 33:610-615. 28. Alper H, Jin YS, Moxley JF, Stephanopoulos G: Identifying gene targets for the metabolic engineering of lycopene biosynthesis in Escherichia coli. Metab Eng 2005, 7:155-164. 29. Segre D, Vitkup D, Church GM: Analysis of optimality in natural and perturbed metabolic networks. Proc Natl Acad Sci U S A 2002, 99:15112-15117. 30. Badarinarayana V, Estep PW 3rd, Shendure J, Edwards J, Tavazoie S, Lam F, Church GM: Selection analyses of insertional www.sciencedirect.com
39. Alper H, Moxley J, Nevoigt E, Fink GR, Stephanopoulos G: Engineering yeast transcription machinery for improved ethanol tolerance and production. Science 2006, 314:1565-1568. This paper extends the gTME approach to eukaryotic cells and describes the isolation of a S. cerevisiae strain exhibiting improved growth in high glucose/high ethanol conditions and an increased volumetric productivity of ethanol. 40. Klein-Marcuschamer D, Stephanopoulos G: Assessing the potential of mutational strategies to elicit new phenotypes in industrial strains. Proc Natl Acad Sci USA 2008, doi:10.1073/ pnas.0712177105. This study provides a direct comparison of gTME with the more classical approach of nitrosoguanidine mutagenesis and is able to show that gTME is more effective in generating phenotypic diversity in L. plantarum. 41. Ochi K: From microbial differentiation to ribosome engineering. Biosci Biotechnol Biochem 2007, 71:1373-1386. This is a comprehensive review of much of the work that has been done in engineering ribosomal components, particularly for eliciting antibioticproducing phenotypes in Streptomyces. 42. Ochi K, Okamoto S, Tozawa Y, Inaoka T, Hosaka T, Xu J, Kurosawa K: Ribosome engineering and secondary metabolite production. Adv Appl Microbiol 2004, 56:155-184. 43. Zhang YX, Perry K, Vinci VA, Powell K, Stemmer WP, del Cardayre SB: Genome shuffling leads to rapid phenotypic improvement in bacteria. Nature 2002, 415:644-646. This paper introduces the concept of genome shuffling (recursive protoplast fusion) for generating strains featuring significant enhancements in phenotype from a select pool of phenotypically intermediate parental strains. 44. Dai M, Copley SD: Genome shuffling improves degradation of the anthropogenic pesticide pentachlorophenol by Sphingobium chlorophenolicum ATCC 39723. Appl Environ Microbiol 2004, 70:2391-2397. Current Opinion in Chemical Biology 2008, 12:168–176
176 Biocatalysis and Biotransformation
45. Gong GL, Sun X, Liu XL, Hu W, Cao WR, Liu H, Liu WF, Li YZ: Mutation and a high-throughput screening method for improving the production of Epothilones of Sorangium. J Ind Microbiol Biotechnol 2007, 34:615-623. 46. Hida H, Yamada T, Yamada Y: Genome shuffling of Streptomyces sp. U121 for improved production of hydroxycitric acid. Appl Microbiol Biotechnol 2007, 73:1387-1393. 47. Patnaik R, Louie S, Gavrilovic V, Perry K, Stemmer WP, Ryan CM, del Cardayre S: Genome shuffling of Lactobacillus for improved acid tolerance. Nat Biotechnol 2002, 20:707-712. 48. Jin YS, Alper H, Yang YT, Stephanopoulos G: Improvement of xylose uptake and ethanol production in recombinant
Current Opinion in Chemical Biology 2008, 12:168–176
Saccharomyces cerevisiae through an inverse metabolic engineering approach. Appl Environ Microbiol 2005, 71:8249-8256. 49. Pfleger BF, Pitera DJ, Newman JD, Martin VJ, Keasling JD: Microbial sensors for small molecules: development of a mevalonate biosensor. Metab Eng 2007, 9:30-38. 50. Santos CNS, Stephanopoulos G: Melanin-based highthroughput screen for L-tyrosine production in Escherichia coli. Appl Environ Microbiol 2008, 74:1190-1197. 51. Tyo KE, Zhou H, Stephanopoulos GN: High-throughput screen for poly-3-hydroxybutyrate in Escherichia coli and Synechocystis sp. strain PCC6803. Appl Environ Microbiol 2006, 72:3412-3417.
www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Christine Nicole S Santos and Gregory Stephanopoulos Although random mutagenesis and screening and evolutionary engineering have long been the gold standards for strain improvement in industry, the development of more sophisticated recombinant DNA tools has led to the introduction of alternate methods for engineering strain diversity. Here, we summarize several combinatorial cell optimization methods developed in recent years, many of which are more amenable to phenotypic transfer and more efficient in probing greater dimensions of the available phenotypic space. They include tools that enable the finetuning of pathway expression (synthetic promoter libraries, tunable intergenic regions (TIGRs)), methods for generating randomized knockout and overexpression libraries, and more global techniques (artificial transcription factor engineering, global transcription machinery engineering, ribosome engineering, and genome shuffling) for eliciting complex, multigenic cellular properties. Addresses Department of Chemical Engineering, Massachusetts Institute of Technology, Room 56-469, Cambridge, MA 02139, United States Corresponding author: Stephanopoulos, Gregory ([email protected])
Current Opinion in Chemical Biology 2008, 12:168–176 This review comes from a themed issue on Biocatalysis and Biotransformation Edited by Stephen G. Withers and Lindsay Eltis Available online 29th February 2008 1367-5931/$ – see front matter # 2008 Elsevier Ltd. All rights reserved. DOI 10.1016/j.cbpa.2008.01.017
been successful in many applications, it was established early on that the interconnectivity and sheer complexity of biological networks often preclude the recognition of simple genotype–phenotype relationships to guide these modifications. Indeed, a single genetic perturbation often has a variety of unpredictable secondary responses within the cell. In a similar vein, the performance of biosynthetic pathways frequently depends on distal genes through kinetic and regulatory interactions whose origins are poorly understood [3,4]. Finally, as an added complication, engineering a complex phenotype may call for the simultaneous modulation of several of these potentially unknown factors [5]. Such challenges led to the development of a new concept called ‘inverse metabolic engineering’ (IME) for cell optimization. This methodology involves three main steps: (1) the construction or identification (by selection) of strains possessing a desired cellular phenotype, (2) the evaluation and determination of genetic and/or environmental factors that confer the phenotype, and (3) the transfer of this phenotype to another strain through direct modifications of the identified genetic and/or environmental factors [6]. Various ‘-omics’ approaches established in the past several years have greatly facilitated the analysis of identified strains and have been the subject of other excellent reviews [7,8]. Here, we focus on recently developed techniques for the generation of strains possessing a phenotype of interest, which, in many cases, remains a significant bottleneck of the IME approach. Owing to the difficulty of predicting these complex genotype–phenotype relationships, many of these methods are combinatorial in nature, that is, they are based on generating genetic (and hence, phenotypic) diversity in a population followed by screening and selection for improved phenotypes.
Introduction In the past 15 years, metabolic engineering has emerged as the discipline that utilizes modern genetic tools for the construction of organisms capable of fuel and chemical production. It was formalized partly from the advent of more sophisticated recombinant DNA techniques that allowed for the targeted genetic manipulation of microbes, either for the modification of existing biochemical reactions or the introduction of completely heterologous pathways. As such, the earliest examples in the field focus on engineering cellular phenotype using rational modifications (typically gene deletions/overexpressions and pathway deregulation) based on existing stoichiometric, kinetic, and regulatory knowledge of a system [1,2]. Although this ‘rational design’ approach has Current Opinion in Chemical Biology 2008, 12:168–176
Fine-tuning expression levels of pathway components It is now broadly accepted that most metabolic pathways are not limited by a single rate-limiting step and that optimized pathways require the balanced expression of several enzymes [9,10]. Without such coordination, metabolic imbalance can lead to the accumulation of gene products or intermediate metabolites with potentially cytotoxic effects or, in some cases, may result in the depletion of a metabolite needed for normal cell growth. Furthermore, the overexpression of genes/proteins often results in an undue metabolic burden on the cell [11]. Thus, many recent metabolic engineering efforts have focused on the development of tools for fine-tuning www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 169
expression to facilitate the search for an optimum balance between pathway expression and cell viability. Synthetic promoter libraries
One such approach involves the creation of libraries of constitutive promoters for modulating gene expression in a variety of prokaryotic (Escherichia coli, Lactococcus lactis, and Lactobacillus plantarum) and eukaryotic (Saccharomyces cerevisiae) organisms. The novelty of these promoters lies in their ability to cover a wide range of expression levels with incremental changes in activity, thus providing the full continuum of gene expression [12,13,14,15,16–19]. Such a tool eliminates reliance on inducible systems that are not only difficult to control but are also cost-prohibitive when considered for large-scale industrial processes. There are currently two established techniques for generating these promoter libraries. The first method, developed by Jensen and Hammer [14,15], is performed by modulating the spacer sequences in between the 10 and 35 regions of the promoter while keeping the consensus regions intact. When applied toward the construction of a promoter library in L. lactis, promoters that possessed sequence variations in the spacer region exhibited a 400-fold range in activity and were shown to alter the expression of pfk and the las operon upon chromosomal delivery [19]. Since then, this technique has been used to create constitutive promoter libraries in E. coli and L. plantarum [13,18], as well as a library of synthetic stationary-phase/stress promoters featuring a range of expression levels and induction times upon exit out of exponential growth [16]. A second method for constructing promoter libraries has been developed in our lab and utilizes error-prone PCR to introduce point mutations into well-characterized native promoters. Proof-of-concept studies have been conducted with the PLtetO-1 promoter for the generation of an E. coli promoter library and the TEF promoter for a S. cerevisiae counterpart [12,17]. The E. coli library features 22 promoters spanning a wide range of activity, which were extensively characterized using three separate orthogonal metrics—GFP fluorescence, mRNA transcript levels, and chloramphenicol minimum inhibitory concentrations. Additionally, each promoter was analyzed on a single-cell level to ensure homogeneity of expression among clonal populations. The utility of such a tool was demonstrated by placing the dxs gene under the control of several different promoters and examining its effects on lycopene production through the nonmevalonate pathway. Interestingly, the highest lycopene titer was seen for an intermediate level of expression, again highlighting the importance of balancing pathway overexpression and cell metabolism [12]. Applications of these promoter libraries have thus far been limited to the manipulation of a single gene or www.sciencedirect.com
operon at a time [12,17,19]. A clear disadvantage of such an approach is that multiple rounds of cell modifications are required to find the optimum expression levels of all genes in a given pathway. Furthermore, it is likely that changing the expression level of one gene will result in different optimum levels for the other genes, particularly when such a change results in a substantial increase in flux (indicating a shift in the location of the pathway bottleneck). Despite the narrow focus of these initial studies, synthetic promoters hold great potential in enabling the simultaneous transcriptional variation of multiple genes in a pathway. Expression libraries can be created simply by placing each gene under the control of several promoters and then combinatorially reassembling the pathway from a mixed pool of promoter-gene constructs (Figure 1a). Recombinant DNA techniques, such as chain reaction cloning, can facilitate the assembly of these libraries [20], and high-throughput screens can then be implemented to select the best performing mutants. Engineering post-transcriptional control
Although the previous examples have focused on exerting control at the transcriptional level, it is also possible to tune the expression of multiple genes in synthetic operons by employing post-transcriptional mechanisms. This was demonstrated in a study featuring a library of tunable intergenic regions (TIGRs) consisting of several control elements, including mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences (Figure 1b). By incorporating these TIGRs in between the coding regions of a transcript, the authors were able to differentially alter the processes of transcription termination, mRNA stability, and translation initiation [21]. In particular, the inclusion of RNAse cleavage sites allowed them to decouple the stabilities of each coding region, thus facilitating the independent variation of expression levels. This TIGR approach was utilized in optimizing flux through a heterologous mevalonate pathway in E. coli through manipulation of the operon containing atoB, HMGS, and tHMGR. A sevenfold increase in mevalonate production was observed via the counterintuitive mechanism of HMGS and tHMGR activity reduction. The ability to recover these unexpected yet phenotypically beneficial changes points to the power of these combinatorial search strategies.
Global perturbations for phenotypic improvement Although fine-tuning the relative levels of gene expression in a specific pathway may lead to improved phenotypes, the majority of applications require engineering techniques that can alter the entire cellular milieu in a less targeted and more global fashion. This concept of strain randomization and selection/screening is not a foreign one and is a methodology that has been employed extensively for the generation of industrial strains. For example, ‘classical strain improvement’ relies on mutaCurrent Opinion in Chemical Biology 2008, 12:168–176
170 Biocatalysis and Biotransformation
Figure 1
Metabolic engineering tools for fine-tuning pathway expression. (a) Promoter libraries can be used to optimize entire pathways by placing individual genes under the control of a series of promoters and then combinatorially reassembling the pathway. (b) Tunable intergenic region (TIGR) libraries consisting of mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences can be used to differentially control posttranscriptional processing of genes in a synthetic operon (adapted from reference [21]).
genic agents such as nitrosoguanidine or UV light to introduce genetic diversity (in the form of point mutations) in a population of strains [22]. Other techniques, such as ‘evolutionary engineering,’ apply a selection pressure and allow the organism to naturally evolve in a chemostat, thus relying on the cell’s inherent capacity to introduce adaptive mutations [23]. Although these methods have been successful in engineering several phenotypes, they suffer from being laborious and timeconsuming, often requiring multiple rounds of screening and mutagenesis and, for the case of evolutionary engineering, extensive cultivation periods [24,25]. Furthermore, because silent mutations tend to accumulate, it becomes difficult to identify which mutations are necessary for conferring a particular phenotype. A full ‘-omics’ analysis of the strain may be needed to understand the basis for these phenotypic differences [7,8], and full genome sequencing/reconstructions may be necessary to initiate phenotypic transfer into a new strain [26,27]. Because of the inherent disadvantages with classical strain improvement and evolutionary engineering, several groups have focused their attentions on designing novel methods for generating phenotypic diversity in a population of strains. Here, we review several alternatives to these more established classical methods that are traceCurrent Opinion in Chemical Biology 2008, 12:168–176
able (thus making them amenable to phenotypic transfer) and are more effective in sampling a larger portion of the phenotypic space available to a cell. Random knockout and overexpression libraries
Methods for predicting knockout and overexpression targets rely heavily on a priori knowledge of the biochemical network and often take the form of stoichiometric models and algorithms, such as flux balance analysis [28] and minimization of metabolic adjustment [29]. However, a variety of recombinant DNA techniques are also available for generating these genetic changes in a randomized fashion, thus introducing the possibility of uncovering regulatory, kinetic, or unknown/poorly understood targets not encompassed in the model. For example, transposon mutagenesis, which involves the transposase-mediated integration of an antibiotic marker cassette into the genome, is an effective tool for creating random knockout libraries (Figure 2a). Although previously prone to insertions into preferred ‘hot spot’ regions, this system has now been optimized for effective and unbiased integration into the entire genome [30,31]. Other knockout approaches have also been developed, including one based on antisense RNA inhibition and promoter interference in Candida albicans. For this system, a library of antisense complementary DNA (cDNA) www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 171
fragments is placed under the control of the GAL1 promoter which, when induced, leads to the selective inhibition of gene expression [32]. Overexpression studies are often conducted simply by transforming the parental strain with a plasmid-encoded genomic library (Figure 2b) [33–35]. The advantage of these knockout/overexpression approaches is that the phenotype can be easily linked back to a single genetic perturbation. The presence of an integration cassette within the transposon libraries allows one to identify the region of insertion through a modified ‘Thermal Asymmetric Interlaced PCR’ procedure [3], and the genomic pieces selected for overexpression can be identified by plasmid sequencing [34]. In addition, many groups have developed microarray-based techniques to increase the throughput of target identification, a particularly important parameter when several unique genes are found to confer phenotypic improvements. The incorporation of a T7 promoter within a knockout cassette allows for the creation and analysis of cDNA libraries at the point of insertion and has been demonstrated to be successful in identifying gene clusters that are negatively selected during E. coli growth competition experiments [30]. More recently, a method for a multiscale analysis of library enrichment (SCALEs) has been developed, which facilitates the identification of single gene or whole operon overexpressions that contribute to a phenotype of interest [33]. Although these knockout and overexpression approaches are often studied in the context of growth phenotypes, they have also been shown to be remarkably successful for the case of lycopene production in E. coli. A random knockout search in the background of a previously constructed high-lycopene-producing parental strain led to the identification of three targets – hnr, yjfP, and DpyjiD – that resulted in further increases in lycopene content. It is important to note that each of these genes encode for proteins with regulatory or unknown functions and thus would not have been predicted by a simple stoichiometric model [3]. In a later study, the introduction of a plasmidencoded genomic library in the background of a mutant containing systematically identified and transposonbased knockouts yielded additional overexpression targets (rpoS, yjiD, ycgW) conferring a lycopene-producing phenotype [34]. This clearly demonstrates the benefits of combining these two complementary approaches for phenotypic improvement. Targeting transcription/translation for introducing global perturbations
Sequential approaches for cellular optimization are not always capable of finding the global optimum of a phenotype owing to the often-convoluted form of the metabolic landscape [3]. As a result, accessing these complex phenotypes from a single round of strain diversification www.sciencedirect.com
requires techniques for introducing genetic changes with known pleiotropic effects. Most of the approaches that have been developed target proteins that have widespread regulatory functions in important cellular processes such as transcription and translation. Thus, the alteration of these proteins holds enormous potential in generating widely diverse libraries for strain improvement studies. Artificial transcription factor libraries based on zinc-finger motifs
Zinc fingers are highly specific DNA-binding domains that recognize three base pair sequences and are found in a variety of transcriptional regulatory proteins. A single transcription factor may contain several of these motifs, which can be assembled in a highly modular fashion. This modularity, coupled with the ability to attach additional effector (activator/repressor) domains, makes zinc-finger proteins powerful tools for introducing transcriptional diversity (Figure 2c). The utility of this approach was first demonstrated in S. cerevisiae through the creation of a randomized plasmid library of over 105 three-finger and four-finger proteins. Transformation of these plasmids and subsequent screening led to the isolation of zincfinger proteins that were capable of conferring a variety of tolerance phenotypes in S. cerevisiae, including resistance to heat treatment, osmotic pressure, and the antifungal drug ketoconazole [36]. In this study, 11 different zincfinger proteins were found to grant varying degrees of ketoconazole resistance. Significantly, coexpression of two of these factors often led to enhanced resistance with the best performing double expression strain exhibiting a 1000-fold enhancement in the resistance phenotype. The most surprising result, however, was the discovery that swapping the effector domain of a zinc-finger protein selected for in one phenotype (from activator to repressor or vice versa) could confer the opposite phenotype. Although the broad applicability of such a result needs to be examined, it introduces the possibility of utilizing this technique for situations in which the opposite phenotype is more amenable to screening. In a later study conducted in E. coli, 40 different zinc fingers were used to construct libraries of 6.4 104 and 2.6 106 three-fingered and four-fingered proteins, respectively. Despite the absence of an activator or repressor domain in these proteins, the authors were successful in identifying a zinc-finger protein conferring a thermotolerance phenotype [37]. After measuring the transcript levels of genes containing potential binding sites, the thermotolerance trait was shown to be a result of the downregulation (but not complete inhibition) of the ubiquinone biosynthesis gene ubiX. Thus, these zincfinger proteins are also capable of directing subtle changes in gene expression that would otherwise remain inaccessible through the previously discussed knockout and overexpression approaches. Current Opinion in Chemical Biology 2008, 12:168–176
172 Biocatalysis and Biotransformation
Figure 2
Current Opinion in Chemical Biology 2008, 12:168–176
www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 173
Global transcription machinery engineering (gTME)
Global transcription machinery engineering (gTME) is a powerful new tool that enables a complete reprogramming of the cellular transcriptome. It does so through the targeted mutagenesis (via error-prone PCR) of select components of the transcriptional machinery, particularly those that are involved in DNA sequence recognition and thus dictate the promoter preferences of RNA polymerase (Figure 2d). In prokaryotic systems, sigma factors bind to the promoter regions of genes with varying degrees of affinity and help to preferentially recruit the RNA polymerase holoenzyme to initiate transcription. Thus, slight variations in these proteins have the potential to greatly affect the subset of genes that are bound by RNA polymerase and expressed. The introduction of plasmidencoded mutants of the principal sigma factor in E. coli (s70) resulted in strains optimized for a variety of phenotypes, including ethanol tolerance, lycopene overproduction, and simultaneous tolerance to sodium dodecyl sulfate and ethanol [38]. These phenotypes, some of which were previously unattainable through traditional methods of strain improvement, were generated in a highly efficient manner and required only a few rounds of screening to promote substantial phenotypic improvements. This methodology has also been shown to be effective in S. cerevisiae through the mutation of the TATA-binding protein (encoded by SPT15) and one of the TATA-binding protein-associated factors (TAF25). With this approach, Alper et al. were successful in isolating a strain that acquired a high degree of tolerance to high glucose/high ethanol conditions and exhibited a 70% improvement in the volumetric productivity of ethanol. We also note that the phenotypes of cells harboring the mutant sigma (or transcription) factors were indistinguishable from that of the control in the absence of stress [39]. This work has recently been extended to another prokaryotic organism, L. plantarum, and this study reveals several important findings about the gTME approach. Through a direct comparison of gTME with the more established method of nitrosoguanidine mutagenesis, the authors were able to demonstrate that the former is significantly more effective in generating phenotypic diversity as measured by the complex phenotype of growth/colony size. This increased diversity was in turn found to be linked to a greater probability of finding a
desirable mutant during selection/screening of a library for malic acid tolerance. Finally, phenotypic variation was shown to be well correlated with the frequency of error generation during the mutagenic PCR reaction, with higher mutation frequencies leading to a wider range of phenotypes [40]. This last point is particularly important, since it offers an additional knob for controlling this system and adapting it to various applications. Ribosome engineering
Modulating ribosomal proteins and rRNA is an alternative way for altering bacterial gene expression at the level of translation. This technique, termed ‘ribosome engineering,’ begins with the introduction of mutations that confer resistance to a variety of antibiotics that attack the ribosome, such as streptomycin, gentamicin, kanamycin, and chloramphenicol. Selection on different concentrations of antibiotics facilitates the isolation of strains with a wide variety of mutations in known ribosomal components (Figure 2e). Although the mechanism leading to phenotypic variation is not well understood, the changes are hypothesized to be a result of the activation of a ‘stringent response’ that leads to elevated levels of protein synthesis during the late growth phase and is beneficial for a select number of specialized phenotypes. Such a technique resulted in antibiotic-producing properties in several Streptomyces organisms, enzyme (a-amylase and protease) overproduction in Bacillus subtilis, and improved tolerance to aromatic compounds, such as 4-hydroxybenzoate, toluene, and m-xylene in Pseudomonas putida [41,42]. It has yet to be determined whether a more randomized mutagenesis approach will be capable of yielding strains optimized in other cellular phenotypes. However, because mutations are introduced directly into the chromosomal copies of these translational proteins, the majority of these changes will probably lead to an inhibition of growth and may ultimately limit the broader utilization of this technique. Genome shuffling
Whole-genome shuffling allows for the generation of combinatorial libraries of complex progeny from a few previously selected parental strains exhibiting subtle improvements in a desired property. The utilization of a method called recursive protoplast fusion ensures a thorough shuffling of the genome, which then allows for the creation of strains that accumulate mutations with additive or even synergistic effects with respect to the
(Figure 2 Legend) Phenotypic diversification through whole-cell engineering techniques. (a) Random knockout libraries are created through the arbitrary insertion of a transposon cassette into the host genome, resulting in the disruption of the genetic element at the site of integration. (b) Random overexpression libraries are generated by transforming the host strain with a plasmid-encoded genomic library. (c) Artificial transcription factor (zincfinger protein) libraries are constructed by combinatorially assembling three or four zinc-finger motifs, followed by attachment of an activator (A) or repressor (R) domain. Expression of these proteins in a strain leads to changes at the level of transcription (adapted from reference [36]). (d) Global transcription machinery engineering (gTME) begins with the mutagenesis of a component of the transcription machinery (often in charge of DNA recognition and binding) by error-prone PCR. The introduction of this mutant constituent results in a complete alteration of the global transcriptome. (e) In ribosome engineering, strains are first selected on media containing antibiotics that target ribosomal proteins and rRNA. The surviving strains, which contain mutations in the translational machinery of a cell, exhibit changes in protein translation and hence, cellular phenotype. (f) Genome shuffling employs recursive protoplast fusion in order to create a library of cells containing shuffled genomic DNA from two or (often) more parental strains. www.sciencedirect.com
Current Opinion in Chemical Biology 2008, 12:168–176
174 Biocatalysis and Biotransformation
phenotype of interest (Figure 2f). Compared with classical strain improvement strategies, this technique has been shown to be more effective in producing strains with enhanced phenotypes. Impressively, when applied toward tylosin production in Streptomyces fradiae, two rounds of genome shuffling were found to be successful in achieving the same titers generated from 20 rounds of classical strain improvement. As stated aptly, ‘this is the difference between 24 000 assays and approximately 1 year of effort, and roughly 1 000 000 assays and 20 years of effort’ [43]. Since then, this tool has been used to engineer a myriad of other complex phenotypes, including improved acid tolerance in L. plantarum, pesticide degradation by Sphingobium chlorophenolicum, hydroxycitric acid production in Streptomyces sp. U121, and epithilone production by Sorgangium cellulosome [44– 47]. Thus, genome shuffling is clearly an important metabolic engineering tool for strain improvement, particularly when strains with intermediate properties have already been isolated.
Concluding remarks As demonstrated by all these approaches, many techniques are now available for the generation of optimized strains that can be studied within an IME framework. Each of these methods taps into its own subset of the phenotypic space available to a cell and works through mostly orthogonal mechanisms. It is therefore likely that combining these search strategies may yield synergistic effects. Early successes in engineering high levels of lycopene production through a combination of systematic/random knockout and gTME approaches point to the distinct possibility of such synergism [38]. As an extension to this thought, these strategies, when implemented in combination, may also be successful in selecting strains with multiple desirable characteristics. In such a scenario, the traceability of these introduced genetic perturbations would greatly facilitate the compilation of each individual mutation into a single ‘super-engineered’ strain. Although we now have access to a diverse collection of tools for the generation of libraries, these combinatorial approaches have thus far been limited to the engineering of only a few select phenotypes. As demonstrated by the body of literature in this area, the major thrusts have largely been on identifying genetic changes that alter the growth characteristics of a strain [30,33,35– 37,39,47], a phenotype that is easily accessed through serial subculturing or by cultivating libraries in a chemostat environment. Such strategies effectively enrich a mixed population with only strains exhibiting desirable growth properties, thus making strain selection a simple process. Lycopene production in E. coli is another major trait that has greatly benefited from a variety of these techniques. Although metabolite overproduction has long been known to be a difficult phenotype to screen, the red pigmentation of this carotenoid molecule makes it an Current Opinion in Chemical Biology 2008, 12:168–176
exception to the rule. Desirable clones that overproduce lycopene assume a red coloration and are therefore easily selected by visual inspection [3,48]. Despite these reported successes, the broader applicability of these combinatorial techniques for other systems of interest clearly depends on the availability of high-throughput screens for probing large (>105 variants) libraries. Although several studies have been successful in generating screens tailored for individual phenotypes such as tyrosine, mevalonate, and poly-3-hydroxybutyrate production [49–51], more modular and flexible platforms will be needed to access a wider array of cellular properties. It is only after such technological advances that the true potential of these combinatorial approaches can be realized.
Acknowledgements This work was supported by a National Science Foundation Graduate Fellowship (CNSS), Grant CBET-0730238, and the Singapore-MIT Alliance (SMA2).
References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Bailey JE: Toward a science of metabolic engineering. Science 1991, 252:1668-1675.
2.
Stephanopoulos G: Metabolic fluxes and metabolic engineering. Metab Eng 1999, 1:1-11.
3.
Alper H, Miyaoku K, Stephanopoulos G: Construction of lycopene-overproducing E. coli strains by combining systematic and combinatorial gene knockout targets. Nat Biotechnol 2005, 23:612-616. This study demonstrates the ability for random knockout approaches to uncover regulatory or unknown protein targets not encompassed in standard stoichiometric models. 4.
Stephanopoulos G, Alper H, Moxley J: Exploiting biological complexity for strain improvement through systems biology. Nat Biotechnol 2004, 22:1261-1267.
5.
Bailey JE: Lessons from metabolic engineering for functional genomics and drug discovery. Nat Biotechnol 1999, 17:616-618.
6.
Bailey JE, Sburlati A, Hatzimanikatis V, Lee K, Renner WA, Tsai PS: Inverse metabolic engineering: a strategy for directed genetic engineering of useful phenotypes. Biotechnol Bioeng 2002, 79:568-579.
7.
Bro C, Nielsen J: Impact of ‘ome’ analyses on inverse metabolic engineering. Metab Eng 2004, 6:204-211.
8.
Gill RT: Enabling inverse metabolic engineering through genomics. Curr Opin Biotechnol 2003, 14:484-490.
9.
Koffas MA, Jung GY, Stephanopoulos G: Engineering metabolism and product formation in Corynebacterium glutamicum by coordinated gene overexpression. Metab Eng 2003, 5:32-41.
10. Pitera DJ, Paddon CJ, Newman JD, Keasling JD: Balancing a heterologous mevalonate pathway for improved isoprenoid production in Escherichia coli. Metab Eng 2007, 9:193-207. 11. Glick BR: Metabolic load and heterologous gene expression. Biotechnol Adv 1995, 13:247-261. 12. Alper H, Fischer C, Nevoigt E, Stephanopoulos G: Tuning genetic control through promoter engineering. Proc Natl Acad Sci U S A 2005, 102:12678-12683. This study describes the multifaceted characterization and subsequent application of an E. coli promoter library generated by error-prone PCR mutagenesis of a PLl derivative. www.sciencedirect.com
Combinatorial engineering of microbes for optimizing cellular phenotype Santos and Stephanopoulos 175
13. De Mey M, Maertens J, Lequeux GJ, Soetaert WK, Vandamme EJ: Construction and model-based analysis of a promoter library for E. coli: an indispensable tool for metabolic engineering. BMC Biotechnol 2007, 7:34. 14. Jensen PR, Hammer K: Artificial promoters for metabolic optimization. Biotechnol Bioeng 1998, 58:191-195. 15. Jensen PR, Hammer K: The sequence of spacers between the consensus sequences modulates the strength of prokaryotic promoters. Appl Environ Microbiol 1998, 64:82-87. This study describes the creation of a synthetic promoter library for L. lactis based on randomization of the spacer region in between the 35 and 10 consensus sequences. 16. Miksch G, Bettenworth F, Friehs K, Flaschel E, Saalbach A, Twellmann T, Nattkemper TW: Libraries of synthetic stationaryphase and stress promoters as a tool for fine-tuning of expression of recombinant proteins in Escherichia coli. J Biotechnol 2005, 120:25-37. 17. Nevoigt E, Kohnke J, Fischer CR, Alper H, Stahl U, Stephanopoulos G: Engineering of promoter replacement cassettes for fine-tuning of gene expression in Saccharomyces cerevisiae. Appl Environ Microbiol 2006, 72:5266-5273.
mutants using subgenic-resolution arrays. Nat Biotechnol 2001, 19:1060-1065. 31. Alexeyev MF, Shokolenko IN: Mini-Tn10 transposon derivatives for insertion mutagenesis and gene delivery into the chromosome of gram-negative bacteria. Gene 1995, 160:59-62. 32. De Backer MD, Nelissen B, Logghe M, Viaene J, Loonen I, Vandoninck S, de Hoogt R, Dewaele S, Simons FA, Verhasselt P et al.: An antisense-based functional genomics approach for identification of genes critical for growth of Candida albicans. Nat Biotechnol 2001, 19:235-241. 33. Lynch MD, Warnecke T, Gill RT: SCALEs: multiscale analysis of library enrichment. Nat Methods 2007, 4:87-93. 34. Jin YS, Stephanopoulos G: Multi-dimensional gene target search for improving lycopene biosynthesis in Escherichia coli. Metab Eng 2007, 9:337-347. 35. Gill RT, Wildt S, Yang YT, Ziesman S, Stephanopoulos G: Genome-wide screening for trait conferring genes using DNA microarrays.. Proc Natl Acad Sci U S A 2002, 99:7033-7038.
19. Solem C, Jensen PR: Modulation of gene expression made easy. Appl Environ Microbiol 2002, 68:2397-2403.
36. Park KS, Lee DK, Lee H, Lee Y, Jang YS, Kim YH, Yang HY, Lee SI, Seol W, Kim JS: Phenotypic alteration of eukaryotic cells using randomized libraries of artificial transcription factors. Nat Biotechnol 2003, 21:1208-1214. This paper describes the use of zinc-finger motifs attached to effector (activator/repressor) domains to create libraries of artificial transcription factors. When expressed in S. cerevisiae, the authors were able to isolate strains exhibiting a number of tolerance phenotypes.
20. Pachuk CJ, Samuel M, Zurawski JA, Snyder L, Phillips P, Satishchandran C: Chain reaction cloning: a one-step method for directional ligation of multiple DNA fragments. Gene 2000, 243:19-25.
37. Park KS, Jang YS, Lee H, Kim JS: Phenotypic alteration and target gene identification using combinatorial libraries of zinc-finger proteins in prokaryotic cells. J Bacteriol 2005, 187:5496-5499.
21. Pfleger BF, Pitera DJ, Smolke CD, Keasling JD: Combinatorial engineering of intergenic regions in operons tunes expression of multiple genes. Nat Biotechnol 2006, 24:1027-1032. This paper employs libraries of post-transcriptional control elements, such as mRNA secondary structures, RNase cleavage sites, and RBS sequestering sequences, to differentially control the expression of genes in a synthetic operon. It is an excellent example of the potential of combinatorial searches for pathway optimization.
38. Alper H, Stephanopoulos G: Global transcription machinery engineering: a new approach for improving cellular phenotype. Metab Eng 2007, 9:258-267. This study outlines a new method for modulating the entire prokaryotic transcriptome through mutagenesis of DNA-recognition elements of the transcription machinery. This method for diversification was shown to be effective for several distinct classes of phenotypes.
18. Rud I, Jensen PR, Naterstad K, Axelsson L: A synthetic promoter library for constitutive gene expression in Lactobacillus plantarum. Microbiology 2006, 152:1011-1019.
22. Adrio JL, Demain AL: Genetic improvement of processes yielding microbial products. FEMS Microbiol Rev 2006, 30:187-214. 23. Sauer U: Evolutionary engineering of industrially important microbial phenotypes. Adv Biochem Eng Biotechnol 2001, 73:129-169. 24. Wisselink HW, Toirkens MJ, del Rosario Franco Berriel M, Winkler AA, van Dijken JP, Pronk JT, van Maris AJ: Engineering of Saccharomyces cerevisiae for efficient anaerobic alcoholic fermentation of L-arabinose. Appl Environ Microbiol 2007, 73:4881-4891. 25. Yomano LP, York SW, Ingram LO: Isolation and characterization of ethanol-tolerant mutants of Escherichia coli KO11 for fuel ethanol production. J Ind Microbiol Biotechnol 1998, 20:132-138. 26. Herring CD, Raghunathan A, Honisch C, Patel T, Applebee MK, Joyce AR, Albert TJ, Blattner FR, van den Boom D, Cantor CR et al.: Comparative genome sequencing of Escherichia coli allows observation of bacterial evolution on a laboratory timescale. Nat Genet 2006, 38:1406-1412. 27. Ikeda M, Ohnishi J, Hayashi M, Mitsuhashi S: A genome-based approach to create a minimally mutated Corynebacterium glutamicum strain for efficient L-lysine production. J Ind Microbiol Biotechnol 2006, 33:610-615. 28. Alper H, Jin YS, Moxley JF, Stephanopoulos G: Identifying gene targets for the metabolic engineering of lycopene biosynthesis in Escherichia coli. Metab Eng 2005, 7:155-164. 29. Segre D, Vitkup D, Church GM: Analysis of optimality in natural and perturbed metabolic networks. Proc Natl Acad Sci U S A 2002, 99:15112-15117. 30. Badarinarayana V, Estep PW 3rd, Shendure J, Edwards J, Tavazoie S, Lam F, Church GM: Selection analyses of insertional www.sciencedirect.com
39. Alper H, Moxley J, Nevoigt E, Fink GR, Stephanopoulos G: Engineering yeast transcription machinery for improved ethanol tolerance and production. Science 2006, 314:1565-1568. This paper extends the gTME approach to eukaryotic cells and describes the isolation of a S. cerevisiae strain exhibiting improved growth in high glucose/high ethanol conditions and an increased volumetric productivity of ethanol. 40. Klein-Marcuschamer D, Stephanopoulos G: Assessing the potential of mutational strategies to elicit new phenotypes in industrial strains. Proc Natl Acad Sci USA 2008, doi:10.1073/ pnas.0712177105. This study provides a direct comparison of gTME with the more classical approach of nitrosoguanidine mutagenesis and is able to show that gTME is more effective in generating phenotypic diversity in L. plantarum. 41. Ochi K: From microbial differentiation to ribosome engineering. Biosci Biotechnol Biochem 2007, 71:1373-1386. This is a comprehensive review of much of the work that has been done in engineering ribosomal components, particularly for eliciting antibioticproducing phenotypes in Streptomyces. 42. Ochi K, Okamoto S, Tozawa Y, Inaoka T, Hosaka T, Xu J, Kurosawa K: Ribosome engineering and secondary metabolite production. Adv Appl Microbiol 2004, 56:155-184. 43. Zhang YX, Perry K, Vinci VA, Powell K, Stemmer WP, del Cardayre SB: Genome shuffling leads to rapid phenotypic improvement in bacteria. Nature 2002, 415:644-646. This paper introduces the concept of genome shuffling (recursive protoplast fusion) for generating strains featuring significant enhancements in phenotype from a select pool of phenotypically intermediate parental strains. 44. Dai M, Copley SD: Genome shuffling improves degradation of the anthropogenic pesticide pentachlorophenol by Sphingobium chlorophenolicum ATCC 39723. Appl Environ Microbiol 2004, 70:2391-2397. Current Opinion in Chemical Biology 2008, 12:168–176
176 Biocatalysis and Biotransformation
45. Gong GL, Sun X, Liu XL, Hu W, Cao WR, Liu H, Liu WF, Li YZ: Mutation and a high-throughput screening method for improving the production of Epothilones of Sorangium. J Ind Microbiol Biotechnol 2007, 34:615-623. 46. Hida H, Yamada T, Yamada Y: Genome shuffling of Streptomyces sp. U121 for improved production of hydroxycitric acid. Appl Microbiol Biotechnol 2007, 73:1387-1393. 47. Patnaik R, Louie S, Gavrilovic V, Perry K, Stemmer WP, Ryan CM, del Cardayre S: Genome shuffling of Lactobacillus for improved acid tolerance. Nat Biotechnol 2002, 20:707-712. 48. Jin YS, Alper H, Yang YT, Stephanopoulos G: Improvement of xylose uptake and ethanol production in recombinant
Current Opinion in Chemical Biology 2008, 12:168–176
Saccharomyces cerevisiae through an inverse metabolic engineering approach. Appl Environ Microbiol 2005, 71:8249-8256. 49. Pfleger BF, Pitera DJ, Newman JD, Martin VJ, Keasling JD: Microbial sensors for small molecules: development of a mevalonate biosensor. Metab Eng 2007, 9:30-38. 50. Santos CNS, Stephanopoulos G: Melanin-based highthroughput screen for L-tyrosine production in Escherichia coli. Appl Environ Microbiol 2008, 74:1190-1197. 51. Tyo KE, Zhou H, Stephanopoulos GN: High-throughput screen for poly-3-hydroxybutyrate in Escherichia coli and Synechocystis sp. strain PCC6803. Appl Environ Microbiol 2006, 72:3412-3417.
www.sciencedirect.com