Constructing de novo biosynthetic pathways for chemical synthesis inside living cells

Abstract

Living organisms have evolved a vast array of catalytic functions that make them ideally suited for the production of medicinally and industrially relevant small-molecule targets. Indeed, native metabolic pathways in microbial hosts have long been exploited and optimized for the scalable production of both fine and commodity chemicals. Our increasing capacity for DNA sequencing and synthesis has revealed the molecular basis for the biosynthesis of a variety of complex and useful metabolites and enables the de novo construction of novel metabolic pathways for the production of new and exotic molecular targets in genetically tractable microbes. However, the development of commercially viable processes for these engineered pathways is currently limited by our ability to quickly identify or engineer enzymes with the correct reaction and substrate selectivity as well as the speed by which metabolic bottlenecks can be determined and corrected. Efforts in understanding the relationship between sequence, structure, and function in the basic biochemical sciences can advance these goals for synthetic biology applications while also serving as an experimental platform to elucidate the in vivo specificity and function of enzymes and to reconstitute complex biochemical traits for study in a living model organism. Furthermore, the continuing discovery of natural mechanisms for the regulation of metabolic pathways has revealed new principles for the design of high-flux pathways with minimized metabolic burden and has inspired the development of new tools and approaches to engineer synthetic pathways in microbial hosts for chemical production.

Living systems have discovered diverse solutions to fundamental problems in chemical catalysis that have the potential to transform society if they could be tapped for synthetic chemistry. For example, the ability of autotrophs to fix and activate carbon dioxide from the atmosphere for use as a universal C₁ building block in biosynthesis has been a longstanding objective for human chemists and could find great utility in the industrial-scale production of commodity chemicals (1–3). With regard to production of complex bioactive compounds, the evolution of enzymes to regio- and stereoselectively utilize molecular oxygen to modify and functionalize complex hydrocarbon skeletons leads to extremely efficient and modular syntheses of entire families of drug-like structures with lower step counts (4–8). The synthetic capacity of organisms has long been adapted for the industrial production of commodity and fine chemicals that can be made in their native hosts at high yield and low cost (9–12); however, the full combinatorial potential of cellular metabolism for designing new synthetic routes to novel targets has yet to be fully realized (13–16). With advances in DNA sequencing (17–19) and the resulting explosion in sequence information, we have collected a vast array of possible genetic components from which to assemble and construct pathways for de novo reaction sequences. In addition, our growing proficiency in large-scale DNA synthesis (20) and sequence manipulation (21) is beginning to provide the necessary tools to modify the chemical programming of cells at a genome level that would allow use of living cells for synthetic applications in medicine, alternative energy, and materials science.

Despite the enormous promise that synthetic biology offers for building new chemical function at an organism level, the development of technical tools to achieve these goals has outpaced our understanding of how chemistry works inside the cell and thus the fundamental design principles for the construction of new pathways. In contrast to traditional synthetic approaches, organismal chemistry must take place in the presence of the thousands of other chemical processes that occur simultaneously within the cell to maintain life. Naturally occurring pathways take advantage of the evolutionary optimization of connections between enzyme partners mediated by protein-protein interactions (22, 23), subcellular localization (24, 25), or complex homeostatic and regulatory networks to channel intermediates to product. In contrast, engineered pathways are built from individual components that have been extracted based on their native functions and reconstituted out of context within a new pathway or host and may produce metabolites and end products that are foreign to the cell. Despite these challenges, several chimeric pathways have been successfully constructed in tractable genetic hosts, such as Escherichia coli (26–29) and Saccharomyces cerevisiae (30), which are robust enough for the design of scalable industrial processes for commercial production.

These examples highlight the potential impact of synthetic pathway construction on the production of small molecule targets; however, each example has required significant resources to develop from the gene discovery stage and/or identification of key pathway bottlenecks to a genetically optimized strain that produces commercially viable product titers (Figure 1). One of the most significant roadblocks in this process is the initial pathway design and selection or engineering of enzymes capable of supporting sufficient flux through the desired synthetic route. The bottlenecks in a synthetic pathway can arise from many sources, including problems in protein folding, co-factor availability or assembly, the absence of protein partners and pathways that might be required for catalysis, the presence of unexpected promiscuous activities, and crosstalk of non-native metabolic intermediates with other pathways in the cell. A second obstacle is the elucidation of the factors that govern intrapathway flux and kinetic behavior that maintains carbon within the pathway rather than allowing it to be lost or dissipated to other pathways within the cell through metabolic crosstalk. This review will address biochemical developments related to these questions and goals and their application to synthetic pathway design.

Figure 1.

Pipeline for construction of a de novo metabolic pathway. Enzymes from environmental hosts are identified, assembled, and transplanted into a heterologous host for target molecule production. Bottlenecks that decrease product titer are identified by a combination of in vivo and in vitro characterization to insight into their source. Incorporation of additional metabolic and regulatory elements are used to alleviate these bottlenecks, which reveals new factors that limit production yields for further optimization.

IDENTIFICATION OF USEFUL CHEMICAL TRANSFORMATIONS

Many interesting and useful chemical phenotypes are apparent at the organism level; among the more industrially relevant transformations are multistep processes like the degradation of cellulose and lignin, fermentation of sugars to produce sustainable fuels, detoxification of environmental pollutants, and production of complex bioactive compounds (Figure 2). In addition, various specialized structural motifs can be found in natural products, such as strained ring structures (31–36) and unusual functional groups (37–39) or bond couplings (40, 41) that could be used as useful handles for chemical synthesis (Figure 3). However, these attributes need to be distilled into a minimal set of genes, including chaperones, activators, and upstream and downstream partners, required for robust enzymatic activity in a heterologous host in order to transplant the reaction of interest into a new host. Despite the wealth of genome information, it remains very difficult to pinpoint the genetic basis for these traits without biochemical insight to narrow the scope of possible candidates. Thus, the most successful gene identification approaches rely on work- or information-intensive routes such as protein purification and sequencing, comparative genomic studies, functional gene clustering, or analysis of gene expression patterns to successfully identify candidates of interest. Consequently, we need to further develop methods to exploit the wealth of sequence information to make predictions about the reactivity and specificity of new enzymes in order to expand our ability to more rapidly design new or enhanced pathways in silico. In this regard, recent advances in functional gene annotation have sought to improve our understanding of the evolutionary relationships between exotic enzymes and their well-studied counterparts and to leverage that understanding for assignment of activities to uncharacterized enzymes and for engineering new enzymatic activities. In addition to sequence-based analysis, physical and computational methods for substrate and/or transition state docking have aided in the identification of specific substrates by restricting the possibilities to a limited set of metabolites. These initial studies show promise for future applications if they can be generalized for the non-specialist user or used for high-throughput and accurate genome annotation so that the production of new small-molecule targets can be approached without requiring the commitment of expensive targeted gene discovery efforts.

Figure 2.

Chemical phenotypes of interest for de novo metabolic pathway construction.

Figure 3.

Specialized structural motifs and unusual functional groups in natural products. Structural motifs and functional groups of interested are highlighted in red. (A) Unique functional groups. (B) Unusual bond couplings. (C) Strained ring structures.

The input and curation of reliable and specific gene annotations within the sequence database is essential to expanding the scope of synthetic pathway design, but inaccuracy and imprecision in functional assignments remains a major roadblock given the stringent requirements for enzyme behavior for synthetic biology purposes. For example, bottlenecks in the production of the target molecule (29, 42, 43) can result from the assignment of a minor catalytic activity to the enzyme of interest (44, 45) or non-specific and promiscuous activities that may cause carbon to exit the synthetic pathway. In addition to having reaction selectivity, synthetic pathway components also must be able to be expressed in a heterologous host with high productivity to prevent bottlenecks resulting from low functional expression. Beyond application to chemical industry, metabolic engineering studies also allow us to step beyond in vitro validation of gene annotations to begin testing enzyme function in vivo when placed within the context of a synthetic pathway.

Targeted gene identification

The discovery of new activities or genes responsible for the biosynthesis of a desired target has been most rapid in bacteria because their genomes are rich in operons and gene clusters that place functionally related genes into close physical proximity with one another (46). Indeed, the number of genes known to encode the production and tailoring of common microbial natural products such as polyketides, non-ribosomal peptides, and ribosomally synthesized peptides has dramatically increased since the development of high throughput sequencing techniques for genome and metagenome sequencing (47, 48). If the natural product of interest is structurally characterized and conditions are known for its production in the native host, the identification of any gene involved its biosynthesis allows the location of unknown classes of enzymes responsible for installation of the particular structural motif of interest by limiting the possibilities to those found within the gene cluster (49, 50). These approaches have identified ununusal and synthetically useful transformations such as aliphatic radical halogenation (49), cyclopropanation (51), enediyne formation (52), new sugar tailoring (53), nitro group installation (54), phosphonate biosynthesis (55), and others. Although these enzymes are typically highly specific for a particular biosynthetic intermediate, mechanistic studies aid in approaching the design of activities of interest from a related class of enzymes (56) or engineering the active site to accept new substrates (57–59). In some cases, mining of genome sequences can pinpoint new enzyme orthologs that naturally catalyze reactions with the desired specificity. For example, Höhne and coworkers predicted mutations that would reverse the enantioselectivity of the known (S)-specific transaminases and biochemically validated 17 previously undiscovered naturally occurring (R)-selective transaminases that were identified from protein databases by the presence of these mutations (60).

The proliferation of completed microbial genome sequences also allows the use of comparative genomics for gene discovery and has uncovered new patterns of analysis. The recent identification of a fatty aldehyde decarbonylase from the cyanobacteria, Synechococcus sp. PCC7002, relied on substractive genome analysis between alkane producing and non-producing cyanobacteria to limit the number of likely candidates (61). In fact, only a single hypothetical protein was left after analysis and its activity was validated in vitro and in vivo for the production of alkanes within the biodiesel range. Interestingly, this decarbonylase represents a different type of active site than those of the decarbonylases found in algae (62) or plants (63) and was clustered with a partner acyl-coenzyme A reductase. In addition to these unusual classes of transformations, ubiquitous enzymes, such as aldolases or dehydrogenases, can also be very useful for the construction of pathways for new targets and genomic context can help to identify their specific substrates. Even without a clear pathway defined for a particular gene of interest, functional correlation between genes that are consistently proximal to each other can be made if the pattern is found across many genomes (64–66). While this method has been used most widely in bacteria, recent evidence suggests that gene order conservation in eukaryotes also has implications for biological function (67–69). Genomic context mining has also been expanded to correlate evolutionary loss or gain of genes with their biological functions through the construction of phylogenetic profiles (70, 71), which correlate the presence or absence of a gene with the presence or absence of other genes across genomes, allowing a functional correlation to be drawn between genes that are lost and gained together.

Although microbial genomes have served an important function in the discovery of new biosynthetic genes, plants remain a large source for much of the diversity of bioactive small molecules at this time (72). For example, taxol, made by the Pacific Yew, is utilized as an important therapy for the treatment of human cancers; however, it is highly tailored and requires and estimated nineteen committed steps after formation of geranylgeranyl diphosphate (73). Despite its importance and the potential for lowering its cost through synthetic biology approaches utilized for other molecules of this family (74, 75), the identification of the necessary genes for engineering its production is a significant challenge and only nine genes have been found to date (74). Similar to taxol and other members of the isoprenoid class, entire families of medicinally relevant alkaloids remain cryptic with respect to their biosynthetic machinery (76, 77). Although the initiation of plant genome sequencing projects has accelerated, most studies rely on expressed sequence tag¹ (EST) libraries or RNA sequencing for biosynthetic gene identification (78–82). Part of the difficulty lies in the fact that tailoring enzymes are key for natural product biosynthesis but are widely represented across plant genomes. For instance, cytochrome P450s play an essential role in structural tailoring, but even a relatively biosynthetically silent organism, such as Arabidopsis thaliana, contains 272 of these enzymes within its genome. Functional genomic approaches such as phylogenetic and metabolite profiling and analysis of spatiotemporal patterns of gene expression and small molecule production hold great promise for addressing this problem (83–88) Thus, these methods are necessary for connecting known pathways of precursor production to important semisynthetic intermediates that may expand our synthetic capabilities such as in the production of artemisinin, an antimalarial isoprenoid (30).

Integrating sequence- and structure-based prediction of enzyme function

Given the almost overwhelming amount of sequence information now available, parallel work in functional gene annotation of biochemically uncharacterized proteins based on bioinformatics and computational approaches (Figure 4AB) can also contribute to providing new components for pathway design. The most common approaches operate at the sequence level and assume functional inheritance through sequence similarity (89) (Figure 4A). An obvious problem with this approach is that homologs arise through divergent evolution and may have developed different functions based on selective pressure. Because genes evolve at different rates depending on selection pressure, the species in which they are found (90), and the family to which they belong (91), it is difficult to establish a threshold for percent sequence identity above which functional annotation can be accurately transferred. Some studies have estimated that a minimum of 40% pairwise identity is required to transfer the first three digits of the Enzyme Commission (EC) number (representing the type of reaction, type of bond acted upon, and type of group acted upon) and that a minimum of 60% identity is required to transfer all four digits (which contain additional details about the substrate) (92, 93).

Figure 4.

Methods for functional gene annotation. (A) Multiple sequence alignments are often sufficient to classify an uncharacterized protein into a superfamily or family, limiting the scope of possible reactions. (B) In silico docking utilizes structural information about the protein of interest for docking potential substrates and ranking them according to favorability of binding. The results limit the size of libraries that must be screened in vitro to determine enzyme function. (C) Functional genomic approaches including protein-protein interaction screens, genetic interaction mapping, and microarray analysis facilitate gene annotation based on biological rather than biochemical function.

The relative inaccuracy of functional assignment based on pairwise identity alone has motivated the development of methods that incorporate position-specific information (94), manually curated family information (95), and structural data (96) to increase the precision of functional predictions. While the availability of structural information for a protein of unknown function increases the chances of successful function prediction, overall protein fold comparison is limited by the high structural similarity between divergent superfamily members and by the existence of ‘superfolds’ (97), which may span several superfamilies that are dissimilar on both the sequence and functional levels. However, if a range of activities are known within a superfamily, hidden Markov models constructed from structure-based alignments can help to pinpoint the precise reaction class and substrate class of an uncharacterized protein for mechanistically characterized superfamilies (98). To evaluate structures’ functional potential at a higher level of detail, methods have been developed to assess the catalytic potential of residues near the active site by scoring co-location of potentially catalytic groups (99), while others map sequences onto known 3D structures (100, 101) with manually annotated catalytic residues (102) for comparison or calculate protein electrostatic surfaces to predict interaction motifs for ligands or other proteins (103). Utilization of 3D structural models can also accelerate the directed evolution of new protein function by increasing the information content involved in library construction, thereby maximizing the number of positive hits from smaller library sizes (104–107).

Physical and computational methods can also be utilized to pinpoint the identity of a substrate for structurally characterized enzyme families by virtual screening of small molecule libraries, which can be computationally docked into the active site and scored based on the energetic favorability of their binding (Figure 4B). Adapted from a widely used method in medicinal chemistry for prediction of inhibitors for particular molecular targets (reviewed in (108)), the use of computational docking in identifying enzyme substrates has been validated in retrospective studies on several different superfamilies (109–111). In silico screening of a collection of ground-state metabolites has also predicted substrates for members of the enolase superfamily in studies that were validated in parallel by in vitro screening of the same set of possible substrates (110, 112). For enzymes that undergo large conformational changes upon ligand binding, these ground-state metabolite docking methods can adapted for use with homology models generated using ligand-bound structures of homologs as templates even in the absence of an experimentally determined crystal structure (110).

Docking methods can also take advantage of the fact that many enzyme active sites show preferential binding to the transition state of the reactions that they catalyze rather than the ground-state substrate. In exploring the substrate specificity of cryptic members of the amidohydrolase superfamily, Hermann and coworkers reduced the scale of their docking study from the complete set of metabolites found in the KEGG database to only those that would provide reactive groups for the range of reactions catalyzed by this superfamily. The members of this reduced library of metabolites were then converted to their respective high-energy intermediate forms for docking studies, which performed as well or better than analogous studies using ground-state metabolites (109). The high-energy intermediate approach was subsequently used to predict the function of an amidohydrolase superfamily member with a novel S-inosylhomocysteine deaminase activity using a docking screen of over 4,000 potential substrates (113). Many other studies have used docking of high-energy intermediates to assist in functional annotation of amidohydrolase superfamily members (114–116) and the method holds promise for other superfamilies and the possible annotation of new enzymes and metabolic pathways.

High-throughput experimental approaches for gene discovery and functional annotation

Although these methods provide insight into the reactivity of characterized families of enzymes, there are also many open reading frames for which homology cannot be used to infer function. The availability of experimental information derived from high-throughput studies presents an opportunity to overcome these obstacles by facilitating annotation based on biological rather than biochemical function (Figure 4C). High-throughput protein-protein interaction screens (117–120) have provided vast amounts of data on the physical association of proteins, allowing enzymes to be grouped in terms of their interaction partners. Accordingly, several databases have been developed to catalog this information (121–123). Microarray and RNA sequencing datasets have likewise allowed the construction of co-expression profiles (124–127), which examine gene expression across various conditions or species, allowing inference of functional linkage between genes whose expression is co-regulated.

In several model organisms, it is also possible to use high-throughput approaches to examine genetic interaction by comparing the fitness or phenotypes of single mutants in the presence of an additional perturbation such as a deletion, overexpression, treatment with chemical inhibitors, or gene knockdown by RNA interference (128). The resulting information can then be used to construct a genetic interaction profile for the gene of interest, providing information about biological function. These methods are best developed for Saccharomyces cerevisiae, for which many quantitative single-gene genetic interaction maps (129–133) as well as a genome-scale genetic interaction map (134) have been constructed. While most high-throughput genetic interaction mapping has been done in model organisms, the method has recently been expanded to mammalian cell culture (135, 136) and has great potential for use in other types of systems.

The increasing availability of knockout collections provides a useful tool for evaluating the output of high-throughput functional genomic screens. Deletion strains for genes identified through high-throughput screening can be evaluated for the phenotype of interest. In one recent example, microarray analysis (137) combined with targeted knockout collection screening identified a key cellodextrin transporter in Neurospora crassa, which was then characterized and used to engineer cellodextrin transport in S. cerevisiae for improved ethanol production (138). One can also take advantage of knockout collections to run highly parallel screens to monitor physiological traits. The construction of knockout libraries in a number of yeast and bacterial species containing molecular ‘barcode’ identifiers has facilitated pooled screening in which fitness can be assessed by the use of oligonucleotide microarrays complementary to the barcodes (139). The benefit of these approaches is that uncharacterized enzyme families or new linkages across pathways can be revealed in this process.

ENGINEERING NEW OR ALTERED ENZYME FUNCTION

One of the major challenges for synthetic biology is not only to find enzymes that catalyze the reaction of interest with high specificity, but also to identify enzymes with expression characteristics that are amenable for synthetic pathway construction because in vivo productivity is a function of both enzyme concentration and rate of turnover. As the functional expression of an enzyme relies on complex pathways for protein synthesis, folding, and maturation, it can often be difficult to overcome the bottleneck of protein solubility. Thus, advances in protein engineering and design can help to alter the characteristics of a specific enzyme by utilizing a well-behaved scaffold to engineer new activity or change substrate specificity in ways that allow the particular enzyme component to meet the metabolic demands of the pathway of interest. If the reaction of interest can be made essential for cell survival, then a genetic selection can also be used in order to search through greater library diversity but most enzymes typically require more intensive screening methods to find sufficiently robust hits for synthetic biology applications. In these cases, protein engineering efforts are typically more successful when the mechanism or source of substrate specificity is well understood, which allows the design of focused libraries that reduce the number of members to be screened or aid in training design algorithms (104–106, 140, 141). Growing insight into physiological mechanisms of evolution can also help to inform the design of laboratory-based evolution while allowing us to explore basic mechanisms of evolution and the contribution that neutral drift and promiscuous function may play in generation of new activities.

In vitro evolution of new and altered enzyme characteristics

While many of the methods described above are useful for isolating the genes responsible for a particular biological activity, in some cases an enzyme is desired to catalyze a chemical transformation not known to occur in nature. In vitro laboratory evolution, using both combinatorial and rational approaches, has been successful at producing new enzyme activities that are useful for synthetic biology applications (59, 142–144). In some cases, the evolution of new activities allows an alternative route to a biosynthetic intermediate (145). In other cases, the improvement in an enzyme that serves as a bottleneck in the overall pathway can lead to large amplifications in yield (146, 147). Furthermore, there are times when simple changes, such as the conversion of an NADPH-dependent enzyme to an NADH-dependent one, can have large implications in overall cellular processes like that observed for C₅ sugar assimilation in the production of biofuels (148–151). As the number of protein sequences increases and their annotation improves, bioinformatics approaches can be used for protein design to focus on residues of interest (152–154) or even for prediction of precise amino acid changes that would alter specificity.

The mechanistic diversity of certain enzyme superfamilies can be utilized for the design of new reactivities from a single scaffold. Mechanistically diverse superfamiles are often considered to be particularly ‘evolvable’ because they share mechanistic features and a common fold while catalyzing a wide variety of different reactions (reviewed in (155)). The enolase superfamily was the first to be systematically explored in this regard and members were found to share a common Mg²⁺-stabilized enolate intermediate formed upon abstraction of the a-proton of a carboxylic acid (156). From this intermediate, a wide range of reaction paths ensue depending on the details of the active site (157) (Figure 3A). This understanding has been used to rationally design enolase superfamily members to catalyze new reactions and can also be used to identify markers of function on a sequence level (158). In other mechanistically diverse superfamilies, such as the Asp/Glu racemase superfamily and the enoyl-CoA hydratase superfamily, there are also several examples in which an understanding of mechanism has facilitated engineering of large changes in activity based on a small number of amino acid changes, which may indicate the generality of this approach (159).

Enzyme promiscuity and neutral drift

While many enzymes have evolved as proficient catalysts with high efficiency and specificity, many others have been found that adventitiously catalyze secondary reactions (160–164). The efficiency of catalysis for these reactions is usually quite low, but can represent a rate acceleration of many orders of magnitude above the uncatalyzed reaction. Mutations conferring this catalytic promiscuity are believed to accumulate as a result of neutral drift (165, 166), in which enzymes evolve under a selective pressure to maintain their original activities (Figure 5A). Enzyme promiscuity has been proposed to play an important role in the evolution of new enzyme functions, requiring that when a selective pressure arises that makes a secondary activity beneficial, it can be improved with just a few mutations to enhance the selective advantages (Figure 5B). This plasticity is believed to account for the apparent rapid evolution of enzymes that proficiently degrade anthropogenic toxins that only appeared in the 20^th century, including tetrachlorohydroquinone dehalogenase (167), which is proposed to have evolved from maleylacetoacetate isomerase (Figure 5C), and phosphotriesterase (164, 168), which is evolutionarily related to the phophotriesterase-like lactonases (Figure 5D). Recent experiments have shown that beyond providing a platform for evolution of single new activities, promiscuous reactions can also result in the appearance of serendipitous metabolic pathways (169). Because of their high evolutionary potential, promiscuous enzymes are an excellent starting point for enzyme redesign. Indeed, some enzyme folds are thought to be more permissive and flexible towards evolving new functions (155, 170–172)

Figure 5.

The neutral drift mechanism of enzyme evolution. (A) A sequence encoding an enzyme with a specific substrate accumulates mutations under selective pressure to maintain the original activity, resulting in the emergence of promiscuous activities and the maintenance of the initial activity. (B) When a selective pressure favoring the promiscuous activity arises, the accumulation of a small number of mutations enhances the promiscuous activity. (C) Phosphotriesterases are proposed to have evolved recently from lactonases. (D) Tetrachlorohydroquinone dehalogenase is proposed to have evolved from maleylacetoacetate isomerase (GSH, glutathione; GSSG, glutathione disulfide).

Terpene synthases and oxidosqualene cyclases have been extensively explored with respect to their catalytic promiscuity because of the relatively small number of enzymes known compared to the quantity of high value products that they produce. Furthermore, their product distribution can also quantitatively report on the alternative reaction pathways taken by the carbocation intermediates as a result of rearrangements, quenching with water, or deprotonation at the incorrect position (173). The primary determinants of reaction outcome are thought to be the shape of the active site and the placement of reactive functional groups; accordingly, mutation to introduce or remove a hydroxyl group can lead to large changes in product distribution. For example, mutation of Ile to Thr in the ent-kaurene synthases converts them to pimaradiene synthases (174), while an Ala to Ser mutation in abietadiene synthase produces a pimaradiene cyclase (175). Similary, removal of a hydroxyl group through a Thr to Ile mutation in syn-pimara-7,15-diene synthase produces an aphidicolene-specific synthase (176). The latter case is notable because rather than short-circuiting the natural reaction to produce a simpler diterpene, the mutation increases reaction complexity and confers a specificity not observed in nature. Steric bulk at the active site also plays a role in determining the reaction path. In cycloartenol synthase, a decrease in bulk at the active site through a Tyr to Thr mutation produced a lanosterol synthase (177). Because all terpene synthases share a conserved fold, it has also been possible to reciprocally interconvert the activities of pairs of enzymes using analogous mutations (178, 179). Fold conservation also facilitated the identification of key plasticity residues at the active site of γ-humulene synthase, allowing the ‘designed evolution’ of seven distinct terpene synthases from a small saturation mutagenesis library (154). Taken together, this work demonstrates the potential for application of our understanding of the origins of catalytic promiscuity for the synthesis of new and potentially bioactive molecules.

OPTIMIZING FLUX THROUGH SYNTHETIC METABOLIC PATHWAYS

The biosynthesis of complex molecules through synthetic metabolic pathways involves the expression or overexpression of several enzymatic components, often from a variety of different organisms. Because these genes are expressed outside their natural contexts and reconnected in unique ways, native regulatory mechanisms are often missing or compromised, which can decrease target molecule output. Without regulatory mechanisms to control expression of the pathway and flux, the host organism may exhibit slow growth owing to the metabolic burden or toxicity of protein overexpression (180), the depletion of host-derived precursors to such an extent that its needs for growth are not met (181–184), or the accumulation of toxic intermediates (185). Metabolic flux may also be compromised by bottlenecks caused by enzyme activity levels that are not commensurate with the activity of more efficient enzymes in the pathway or by the expression of the subunits of multienzyme complexes at suboptimal stoichiometric ratios. Synthetic metabolic pathways must therefore include engineered regulatory mechanisms to balance protein expression and to direct intermediates down the target pathway to maximize target molecule production (Figure 6).

Figure 6.

Engineering pathway balance. (A) Expression of pathways genes appropriate levels can be achieved by adding RNA regulatory elements. (B) Control of ribosome binding site accessibility or (C) ribosome binding site optimization can be used to tune protein expression at the translational level. (D) Variation of promoter strength or inducer concentration can be used to tune protein expression at the transcriptional level.

Identifying and overcoming pathway bottlenecks

Owing to variation in specific activity, reaction equilibrium, and solubility between the enzymes in synthetic pathways, bottlenecks often arise that limit flux and product titers. A major challenge following the construction of pathway interactions is to find these limiting steps and to develop approaches to manage them. In the course of these efforts, new bottlenecks often appear each time previous ones are alleviated. The origin of these bottlenecks does not necessarily result from only poor solubility or expression but can also have a root in the kinetic behavior of a pathway. For example, various bottlenecks have been identified during the development of a high-yielding engineered pathway for the production of an antimalarial drug precursor and were found to arise from kinetic mismatching for the clearance of a toxic intermediate (185–187), loss of diffusible intermediates from the cell, or limitations in rate of an irreversible step to drive the pathway flux in the forward direction (188). Thus, the elucidation of the biochemical basis for the performance of a synthetic pathway and its dependence on individual steps often aids in the development of strategies to increase titers of the target small molecule. In general, improvement of product yields can be achieved by titrating the expression of a particular protein (29, 188) or by in vitro evolution of improved kinetic parameters of the enzyme in question (154, 189).

Working toward understanding the underlying physiological source of a particular bottleneck can also lead to changes in pathway design itself and provide insight into the function of native pathways. During the course of engineering E. coli for the production of n-butanol, a second-generation biofuel, several groups had identified a similar bottleneck in the enoyl-CoA reduction step that is required to produce butyryl-CoA, a key intermediate, from crotonyl-CoA (29, 42, 171). Cellular studies demonstrated that butyryl-CoA is not dissipated through native cellular pathways, suggesting that if an effectively irreversible reduction of crotonyl-CoA could be achieved, it would commit carbon to the synthetic n-butanol pathway (29). Replacement of butyryl-CoA dehydrogenase (Bcd), the native clostridial system for crotonyl-CoA reduction, with a member of the more unusual and mechanistically distinct trans-enoyl-CoA reductase (Ter) family led to an order of magnitude increase in n-butanol titers to 4.7 g L⁻¹ (29). Further analysis of different strains led to the conclusion that Ter is capable of acting as a kinetic trap because of its chemical mechanism of direct hydride transfer from NADH. In contrast, the native Bcd enzyme utilizes a flavin cofactor, which decreases the kinetic barrier to the reverse reaction via a more energetically accessible intermediate, allowing butyryl-CoA to revert to crotonyl-CoA, which can be lost to competing cellular pathways (29). The use of in vivo product titers to assay enzyme activity in this case allows us to build strains that produce commercially viable levels of a target compound as well as to assess the function of an enzyme inside the cell and to begin identifying the basis for the parallel evolution of mechanististically distinct enzymes that catalyze the same reaction. In addition to experimental approaches for synthetic pathway construction, the continual development of more robust methods for modeling pathways and identifying bottlenecks through computational approaches should greatly facilitate this process (190–192). Advances in rapid multiplexed in vitro evolution at the genome level also hold much promise as a combinatorial method to examine relationships between enzymes in a native or engineered pathway and to identify bottlenecks or points of regulatory control (21).

Engineering pathway balance

Adjusting promoter strength has long been a widely used method for controlling and optimizing protein expression levels (Figure 6D). In addition, promoter titrations are often the fastest approach to identifying key bottlenecks. A wide variety of promoters are available that provide a means for regulation of expression levels with a small molecule inducer. Some promoters, like that controlled by arabinose (araC-P_BAD) (193) and propionate (prpR-P_prpB) (194), offer both tunable expression in the presence of the inducer and tight control of expression in the absence of inducer and have been developed further for use in synthetic biology. Given that yields depend on individual cellular behavior, these expression systems sometimes require genomic modifications in order to enforce cell-to-cell tunability rather then the formation of mixed sub-populations (195). The use of promoters as the primary means of controlling gene expression remains problematic, however, in that genes that must be expressed at different levels to achieve pathway balance must each be expressed under a different promoter requiring a different inducer.

More recent work has focused on uncoupling transcriptional control of gene expression from translational control, allowing several genes to be encoded in a single operon but to be expressed at different levels (Figure 6A). An early approach utilized the incorporation of RNase E recognition sites between coding regions that would be endonucleolytically cleaved following transcription, generating two independent secondary transcripts (196). The stability of these transcripts can modulated by adding engineered secondary structural elements, allowing control of protein expression through mRNA stabilization or destabilization. This idea was subsequently expanded through the incorporation of a library of tunable intergenic regions (TIGRs) encoding two variable hairpins flanking variable RNase E sites (186). Using this library, relative expression of the two coding regions could be varied over a 100-fold range. Application of TIGRs to the heterologous mevalonate pathway in E. coli attenuated the effects of the accumulation of the toxic intermediate 3-hydroxymethylglutaryl-CoA, which limits target molecule production, leading to a seven-fold enhancement in mevalonate output. Interestingly, the mechanisms by which the improvement was achieved were counterintuitive and involved differential mRNA processing, transcription termination, and sequestration of ribosome binding sites (Figure 6B). Ribozymes and RNA aptamers have also been applied to modulate gene expression. RNA aptamers have been engineered that bind to transcriptional repressors, activating gene expression. Combination ribozyme-aptamer (‘aptazyme’) RNAs have been designed in which ligand binding to the aptamer mediates ribozyme self-cleavage, altering mRNA stability (197) or ribosome binding site accessibility (198). These systems show great potential for application in the construction of feedback loops to couple gene expression to changes in the concentration of a small molecule of interest.

In many cases, the metabolic demands of protein expression interfere with cellular requirements for growth, leading to lower yields of the target molecule. Linking gene transcription to the metabolic state of the cell could alleviate this problem by only allowing the transcription and expression of protein when cellular resources become available. One approach utilized a reengineered Ntr regulon to couple glucose availability to transcription by sensing acetyl phosphate (199). Placement of a gene whose expression is normally toxic to E. coli under the promoter of the Ntr regulon relieved growth inhibition by repressing protein expression until stationary phase, when glucose became abundant. Incorporation of this system into a synthetic pathway for the production of lycopene increased yields ~20-fold compared to use of standard inducible promoters, demonstrating the utility of dynamic control of protein expression in increasing yields.

As a complement to transcriptional regulation, gene expression can also be tuned at the translational level (Figure 6BC). In bacteria, ribosome binding sites (RBSs) are used to control translation initiation, which in most cases is the rate-limiting step in translation. The creation of libraries of RBSs with different rates of translation initiation has been studied for the optimization of synthetic pathways (200). Because the library size required for optimization increases combinatorially with the number of genes in a pathway, a computational RBS optimization algorithm was developed to predict the translation initiation rate for a particular sequence, decreasing the time and resources required for RBS optimization (201). This model correctly predicted that the rate of translation initiation varies depending on RBS context and was experimentally verified for over 100 genes, demonstrating that RBS sequence can vary gene expression over a range of 100,000-fold.

Maximizing pathway flux through engineered spatial organization

Rather than acting in isolation, natural enzymes often participate in multienzyme complexes, are localized to specific cellular compartments, or are found as fusion proteins in which a single polypeptide catalyzes two or more activities (Figure 7). Based on whole-genome affinity purification-mass spectrometry studies in Saccharomyces cerevisiae, it was estimated that there are at least 491 protein complexes in yeast (117, 118). A later study using a protein-fragment complementation assay suggested that at least 1124 yeast proteins participate in 2270 multiprotein complexes (119), demonstrating the ubiquity of relatively stable protein-protein interactions in cellular pathways. Such spatial organization often confers kinetic benefits by increasing local concentration of intermediates, by avoiding loss of reactive intermediates, and by preventing intermediates from entering other pathways. Further, such metabolic channels can help to avoid toxicity of poisonous metabolites by keeping local concentrations high while keeping overall cellular concentrations low. In addition to function of the pathway itself, approaches to increasing the efficiency of metabolic pathways without increasing intracellular enzyme concentration are important as the metabolic burden that protein synthesis places on the cell can be substantial and detrimental to product titers.

Figure 7.

Spatial organization in natural enzyme systems. (A) Carbamoyl phosphate synthetase utilizes a physical channel that protects the labile intermediates ammonia, carboxy phosphate, and carbamate from the cellular environment. (B) In the fungal shikimate pathway, a pentafunctional polypeptide is utilized to catalyze the five reactions required to convert 3-deoxy-D-arabino-heptulosonate 7-phosphate to 5-enolpyruvylshikimate 3-phosphate. (C) The breakdown of cellulose requires multiple enzymes of the glycosyl hydrolase superfamily, including reducing and non-reducing end exoglucanases as well as endoglucanases. The spatial organization of cellulosomes allows the modular docking and exchange of different enzymes in a single scaffold for synergistic and tunable degradation of the sugar polymer (CBM, celluolose binding module). (D) Formation of the purinosome complex between the enzymes that catalyze de novo purine biosynthesis is believed to protect the short-lived intermediates, phosphoribosylamine (PRA) and 4-carboxyaminoimidazole ribonucleotide (CAIR), in the mammalian pathway between 5-phosphoribosyl-α-pyrophosphate (PRPP) and inosine monophosphate (IMP). In most bacteria, yeasts, and plants, the precursor to CAIR, N⁵-carboxy-4-aminoimidazole ribonucleotide (N⁵-CAIR, t_½ = 15 s), also exists as a short-lived intermediate but is channeled by a multifunctional enzyme in mammals.

Many different mechanisms have evolved for achieving metabolic channeling through spatial organization. Some enzymes, such as tryptophan synthase (202) and carbamoyl phosphate synthetase (203), use physical channels to deliver reactive intermediates from one active site to another within multisubunit complexes (Figure 7A). Other pathways utilize fusion proteins that carry out several reactions, such as synthesis of 5-enolpyruvylshikimate 3-phosphate in the shikimate pathway (204) (Figure 7B). Many examples exist in which spatial organization is achieved through the formation of multienzyme, multiple activity complexes, including the tricarboxylic acid cycle (205), the Calvin cycle (206), glycolysis (207), fatty acid oxidation (208), and protein degradation (209), that allow the cell to rapidly control both flux and selectivity of multistep reaction pathways. For example, the assembly and disassembly of the enzymes of purine biosynthesis, which involves intermediates unstable in the cellular milieu, is regulated by cellular conditions and can be rapidly controlled to turn biosynthesis on or off in low and high purine levels, respectively (22) (Figure 7D). Cellulosomes similarly comprise multisubunit complexes, but complex formation is triggered by scaffold proteins that specify the stoichiometry of each enzyme involved in cellulose breakdown depending on the carbon source available to the organism (23) (Figure 7C). Enzymes can also be compartmentalized by protein organelles, such as the carboxysome structure found in cyanobacteria that traps CO₂ within in order to manage the poor selectivity and kinetic behavior of ribulose-1,6-bisphosphate carboxylase/oxygenase (RuBisCO) in carbon fixation (25).

While multienzyme complexes often mediate metabolic channeling, they also play roles in the regulation of metabolic pathways by using stable and transient protein-protein interactions or enzyme localization to control the output and product distribution of promiscuous enzymes and pathways. In the biosynthesis of dhurrin, for example, the formation of transient, low-affinity complexes both prevents the diffusion of reactive and toxic intermediates and enforces the need for expression of a specific glucosyltransferase despite the promiscuity and ubiquity of these enzymes (210, 211). For enzymes that participate in multiple pathways, such as glyceraldehyde-3-phosphate dehydrogenase of glycolysis and the Calvin cycle, differential regulation mechanisms can be used to control its behavior when it is complexed with different enzyme partners (206). Although the precise organization of complexes involved in plant phenylpropanoid biosynthesis remains to be established, the distribution of the diverse set of possible products, including lignin, sinapate esters, stilbenes, and flavonoids, is coordinated by the localization of several different enzymes in the pathway (212).

Engineered spatial organization has long been considered as a means to enhance productivity of enzymes by mimicking metabolic channeling. Early studies focused on enforcing the proximity of enzymes by physically immobilizing them on polymer beads (213). While modest enhancements in product formation were achieved, enzyme immobilization is problematic in terms of scaling up for industrial applications. More desirable would be to mimic nature by designing self-replicating, genetically encoded systems. The use of fusion proteins for this application was explored in the incipient days of recombinant DNA technology (214, 215). While some fusion proteins were successful, in other cases, fusions led to decreased activity due to interference with the formation of other multiprotein complexes that were required for activity, highlighting the limited scope of this strategy.

Recent efforts to engineer spatial organization have focused on using scaffold proteins to template enzyme colocalization. In an example based on natural cellulosomes, two cellulases were fused to scaffold binding domains known as dockerins, allowing their precise assembly onto the scaffold protein scaffoldin (216). The resulting ‘designer cellulosomes’ showed synergistic behavior that conferred a two-fold enhancement in their ability to degrade cellulose. A second approach utilized scaffolds with interaction domains derived from the metazoan signaling machinery. By tagging the enzymes of the mevalonate pathway with the cognate peptide ligands for the interaction domains, the stoichiometry of the mevalonate-producting enzymes was optimized, increasing product titers by 77-fold (187). The same scaffold strategy was applied to enhance production of glucaric acid from an engineered pathway by 5-fold, demonstrating the system’s modularity (187, 217). By colocalizing pathway proteins, the scaffold decreases the metabolic burden of overexpression, suggesting that it could be useful for difficult to express or poorly soluble proteins. In addition to product yield enhancement, this approach could also be used to control the product distribution of promiscuous enzymes, especially when considering that substrate flexibility can often arise as a result of protein engineering efforts.

CONCLUSIONS

The remarkable diversity of enzyme-catalyzed transformations observed in nature makes biological systems ideal for addressing a wide variety of problems in chemical synthesis. The availability of thousands of genome sequences and improving technologies for the low-cost assembly of large synthetic DNAs means that we are limited by our ability to design rather than to construct de novo synthetic pathways that are sufficiently robust to displace existing processes for small molecule production. Beyond applications to low-cost chemical production, synthetic biology also offers an alternative and complementary platform to more traditional reductionist approaches to study and elucidate how enzymes and other biochemical components work inside living cells to produce organism-level phenotypes. In this regard, synthetic pathway construction can also serve as a powerful tool for enzymology by providing an interesting intermediate level of study between in vitro studies where we can carefully measure physicochemical properties of enzymes and genetic studies where we can assess and validate physiological function. In comparison to these approaches, product titers from synthetic pathways can be used both as a measure of enzyme activity when filtered through the context of thousands of other highly regulated and potentially opposing chemical reactions within a living cell as well as a genetic phenotype with high dynamic range to score the fitness of individual components or a pathway as a whole. Furthermore, the perturbation to host metabolism caused by introduction of an exogenous metabolic pathway has the potential to uncover the organizational and regulatory principles that control the complex metabolic network of the cell. Thus, the synergy between deconstruction and de novo construction can advance both our understanding of the complex behavior of metabolic pathways and networks and our ability to engineer microbes capable of producing commercially viable titers of an expanding repertoire of target molecules.

ABBREVIATIONS

EST

expressed sequence tag

EC

Enzyme Commission

OMPDC

orotidine 5′-monophosphate decarboxylase

NAL

N-acetylneuraminate lyase

TIGR

tunable intergenic region

RBS

ribosome binding site