This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Copyright Information Books from ASM Press MicrobeWorld Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Umeno, D. Articles by Arnold, F. H. Search for Related Content PubMed PubMed Citation Articles by Umeno, D. Articles by Arnold, F. H.

Diversifying Carotenoid Biosynthetic Pathways by Directed Evolution

Daisuke Umeno ,^, Alexander V. Tobias, and Frances H. Arnold^*

Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California

SUMMARY

INTRODUCTION

Overview of Carotenoid Biosynthetic Pathways

Carotenoids as a Model System for Laboratory Pathway Evolution

Evolvable Pathways and Enzymes

APPROACHES TO ENGINEERING NOVEL BIOSYNTHETIC PATHWAYS

Engineering Pathways by Gene Assembly

Engineering Pathways by Directed Enzyme Evolution

PROBING THE EVOLVABILITY OF CAROTENOID BIOSYNTHETIC ENZYMES AND PATHWAYS

Directed Evolution of Key Carotenoid Biosynthetic Enzymes

Isoprenyl diphosphate synthases.

Carotenoid synthases.

Carotenoid desaturases.

Carotenoid cyclases.

Enzymes catalyzing further modifications.

Creation of Pathways to Carotenoids with New Carbon Scaffolds

A C35 carotenoid pathway.

Carotenoids with longer backbones.

DISCUSSION AND FUTURE DIRECTIONS

Revised View of Carotenoid Biosynthetic Pathways

Future Challenge: Specific Pathways

Evolving enzyme specificity.

Designed channeling.

Specificity by flux balance.

Specificity by diversion of unwanted intermediates.

Prospects and Challenges for Diversifying Other Pathways by Directed Evolution

CONCLUSIONS

ACKNOWLEDGMENTS

REFERENCES

Top Next References

INTRODUCTION

Top Previous Next References

 
Nature is a masterful and prolific chemist. The ever-expanding catalog of natural products from plants and microbes currently stands at more than 100,000 different compounds (58). This collection has profoundly influenced human well-being. Natural products have been used to treat pain, infections, and disease for thousands of years; their equally ancient and diverse uses as colorants, spices, fragrances, aphrodisiacs, cosmetics, and toxins have been fundamental to human culture and development. This trend shows no signs of subsiding. Of the 877 new small-molecule pharmacological entities introduced worldwide between 1981 and 2002, 61% can be traced to natural products (142). A similar proportion (59%) of the 137 small molecules in phase II or III clinical trials for cancer treatment (as of June 2003) are natural products or closely related compounds (172). Natural products and their derivatives therefore remain an essential component of the pharmaceutical industry (172). Given the wealth of functional small molecules in nature, it is perhaps not surprising that so much effort today goes into prospecting for new ones or modifying existing ones. Academic and industrial laboratories, often in massive screening programs, continue to isolate and characterize promising new compounds from biological extracts. In addition, molecular scaffolds inspired by natural products are increasingly being used as the basis for synthesizing combinatorial chemical libraries which can be screened for protein-binding or other biochemical activities (30).

Although impressive in number, the known products of natural biosynthetic pathways account for but a tiny fraction of the structures that could be produced. This essentially infinite space of possible functional molecules represents an irresistible frontier for those seeking new, bioactive compounds. For example, only 1/10 of the 877 new small-molecule pharmacological entities mentioned above are bona fide natural products—most are derivatives of natural products, synthetic compounds with natural product-derived moieties, or natural-product mimics (142). The chemical space surrounding natural products is particularly rich in biologically functional molecules, and structurally related compounds can serve human purposes even better than natural products themselves. Researchers are now beginning to explore some of these nearly natural products by mixing, matching, and mutating biosynthetic genes from diverse sources in organisms such as Escherichia coli and screening for production of novel metabolites. In this review we illustrate how pathways can be "evolved" rapidly in the laboratory to generate new natural products. Used in this way, directed laboratory evolution is a powerful tool for discovering new pathways and for in vivo combinatorial chemistry of natural products, i.e., "combinatorial biosynthesis."

In addition to providing access to novel metabolites, laboratory evolution studies can offer insights into the natural evolution of metabolic pathways and their constituent enzymes. Comparative studies of homologous enzymes can readily identify similarities such as conserved residues and folds. However, identifying the exact genetic changes responsible for differences in the specificity, stability, or other properties of related enzymes is much more difficult. When comparing sequences that have diverged over millions of years, it is almost impossible to distinguish the handful of adaptive mutations from the plethora of differences that reflect evolutionary drift. In contrast, evolution in the laboratory allows us to capture new metabolic pathways in the act of emerging and permits the identification and analysis of the molecular events that gave rise to the new product distributions.

With carotenoid biosynthesis as our model, we describe how pathway engineers have used laboratory evolution experiments to diversify biosynthetic pathways and what they have learned from those efforts. This work has demonstrated the remarkable ability of carotenoid biosynthetic enzymes to evolve and acquire new specificities and has illuminated pathway features that facilitate the creation of new natural products based on "unnatural" molecular scaffolds. Although evolutionary pathway engineering is in its infancy, we show that it is capable of generating whole new families of natural product analogs. Some of these compounds may await discovery in nature, while others will probably never be found by natural evolution.

View larger version (65K): [in this window] [in a new window]   FIG. 1. Natural carotenoid biosynthetic pathways can be organized in a tree-like hierarchy. The biosynthesis of all natural carotenoids begins with the enzymatic assembly of a C30 or C40 backbone. These backbones are desaturated, cyclized, oxidized, and otherwise modified by downstream enzymes in various species-specific combinations. Shown are several common types of enzymatic transformations that occur in natural carotenoid pathways. Common carotenoids formed early in a pathway, such as lycopene, are modified in different organisms depending on the enzymes present. The multiple enzymatic routes originating from intermediates common to many end products result in extensive pathway branching (and subbranching).

Some organisms synthesize modified C₄₀ carotenoids with additional or fewer isoprenyl units. For example, Corynebacterium glutamicum synthesizes "homocarotenoids" such as flavuxanthin (C₄₅) and decaprenoxanthin (C₅₀) via enzymatic prenylation of lycopene (Fig. 1) (109, 112). In contrast, several fungi produce C₃₅ "apocarotenoids" such as neurosporaxanthin via oxidative cleavage of monocyclic C₄₀ carotenoids (Fig. 1) (12, 14, 164, 215). Despite their non-C₄₀ structures, these natural homo- and apocarotenoids are, from a biosynthetic perspective, part of the C₄₀ family, since their biosynthesis proceeds via the C₄₀ carotenoid backbone phytoene.

Isoprenoids and isoprenyl diphosphate building blocks are found in all organisms. E. coli has three isoprenyl diphosphate synthases: a farnesyl diphosphate (C₁₅PP) synthase (67, 68), an octaprenyl diphosphate (C₄₀PP) synthase (154), and a Z-type (see below) undecaprenyl diphosphate (C₅₅PP) synthase (94). A C₃₀ carotenoid pathway can be built in E. coli branching directly from the organism's endogenous C₁₅PP supply (226). However, since E. coli lacks a C₂₀PP synthase function, a geranylgeranyl diphosphate (C₂₀PP) synthase gene must be heterologously expressed along with the other carotenoid biosynthetic genes in order to make C₄₀ carotenoids in this organism.

Most of the enzymes involved in carotenoid biosynthesis are membrane associated, and the hydrophobic intermediates and products of carotenoid pathways partition to cytoplasmic or organelle membranes, depending on the type of organism (29). It is not known whether the carotenoid biosynthetic enzymes associate into complexes. Over the years, evidence for the existence of multienzyme complexes that function as assembly lines for carotenoid synthesis has been presented (summarized in reference 32). Some of the most convincing evidence has come from experiments performed in the laboratory of Cerdá-Olmedo with the fungus Phycomyces blakesleeanus (6, 39, 57, 138). On the other hand, the fact that carotenoid enzymes from different organisms are readily combined to generate functional pathways in heterologous hosts is evidence that the formation of a specific enzyme complex is not a prerequisite for carotenoid biosynthesis. Throughout this review, we refer to carotenoid-synthesizing enzyme complexes when they provide a possible alternative explanation for particular observations from our laboratory and others'.

Carotenoids are diverse, and their biosynthetic pathways are representative of other secondary metabolic pathways.Hundreds of carotenoids are known in nature. They are made using a few basic transformations, the earliest of which are highly conserved. The reactions involved in these early steps, such as backbone formation by carotenoid synthases and cyclization by carotenoid cyclases, are mechanistically similar to other reactions in isoprenoid biosynthesis. Modification steps later in the carotenoid pathways cover a wide range of biotransformations, many of which appear in other natural products. Thus, experience with carotenoid pathway evolution can be applied to diversifying other biosynthetic pathways.

Carotenoid biosynthetic enzymes are highly portable.The carotenoid pathway emerges directly from the central isoprenoid pathway that exists in all organisms. Thus, in principle, any organism can supply the precursors required for an engineered carotenoid pathway. Carotenoid biosynthetic genes can be expressed in a wide range of organisms to extend or redirect an existing pathway (69, 70, 105). For instance, a bacterial phytoene desaturase (CrtI) introduced into cyanobacteria (227) and tobacco plants (136) elevated their resistance to the herbicide norflurazon (known to block plant-type phytoene desaturases). Plastid-targeted introduction of an algal ß,ß-carotene ketolase resulted in transgenic tobacco plants that accumulated astaxanthin and changed the color of the plant's nectory tissue from yellow to red (128). The highly publicized introduction of the ß,ß-carotene pathway into rice resulted in substantial ß,ß-carotene levels in the normally carotenoid-free endosperm and therefore greater nutritional value for this food staple (234). Phytoene-producing human cells (HeLa, NIH 3T3 etc.) were created by the addition of a bacterial phytoene synthase and showed increased tolerance to oxidative shock as well as a lower rate of H-ras-induced carcinogenesis (143, 144, 179). Almost all carotenoid biosynthetic genes cloned to date, including those from plants, can be functionally expressed in E. coli (119, 175, 177, 180), as can animal oxidative cleavage enzymes (98, 123, 158, 220, 231). The "portability" of these enzymes greatly facilitates the assembly and evolution of novel biosynthetic pathways.

Evolution of carotenoid pathways can be tracked visually.Carotenoids are natural pigments, and their characteristic colors are ideal for product-based high-throughput screening. Scientists have made good use of this feature throughout the history of carotenoid research. Experiments involving the generation and analysis of mutant strains exhibiting altered colors have elucidated biosynthetic routes in various organisms (12, 15, 18, 40, 66, 73, 113, 120, 130, 155, 162, 173, 178, 196, 201, 216). Many carotenoid biosynthetic genes were cloned based on their ability to confer or alter color development in E. coli (52, 54, 133, 134, 160, 197). Carotenoid titers are an indicator of precursor levels and can be used to evaluate and tune upstream (isoprenoid) pathways (189, 223, 224). The colors generated in E. coli colonies provide a facile screen for new pathway products in a laboratory evolution experiment.

Carotenoids are valuable in their own right.The polyene chromophores of carotenoids, which absorb light in the 400 to 550-nm range, provide the basis for their characteristic yellow-to-red colors and their ability to quench singlet oxygen (5, 74, 79, 95, 103, 137, 218, 230). Carotenoids act as antioxidants in vivo (170, 183, 210), and many beneficial effects of carotenoids on human health have been reported, ranging from cancer prevention and tumor suppression to upregulation of immune function, reduction of the risk of coronary heart disease or age-related degeneration, and cataract prevention (19, 21, 42, 47, 83, 108). As natural pigments, carotenoids are used as colorants in foods, cosmetics, and flowers (117, 128, 209). Given these diverse and important properties of known carotenoids, we can assume that novel compounds of this family will also possess interesting biological or chemical functions.

In general, evolvable pathways appear to contain enzymes that are "locally specific" (32). These enzymes are not unspecific; rather, they recognize a particular structural motif common to a variety of possible substrates. Thus, if a change upstream of a locally specific enzyme generates a novel compound, there is a good chance that the enzyme will metabolize this compound further (as long as it retains the required motif) and generate a new derivative. More such enzymes further downstream will accept the new derivative(s) and produce yet more new metabolites. A carotenoid desaturase, for example, might need at most a few mutations in order to catalyze double-bond formation on a new carotenoid-like backbone. Locally specific modifying enzymes further downstream would, with high probability, accept these desaturated substrates, generating still more new compounds.

Pathway structure also contributes to evolvability: highly branched pathways can propagate discoveries along many routes. Pathway branches occur where one substrate is potentially converted to multiple products by the action of one or more enzymes. The existence of nested branches geometrically increases the number of new products that can be synthesized from a new metabolite produced upstream, provided that the downstream enzymes accept the new substrates. Evolvable and consequently diverse, carotenoid pathways are "bushy," with multiple products possible from each basic transformation. Product possibilities are dictated by the nature of the chemical reaction and substrate and ultimately by the specificity of the enzyme that controls that branch point. For example, desaturation of a carotenoid backbone can produce a number of different products—the one(s) produced by any given organism reflect the specificity of its particular desaturase in its environment. That specificity can change upon mutation and can open new pathway branches.

Pathways constructed entirely of locally specific enzymes would not serve a host organism well in terms of regulation and production of tried-and-true metabolites. Thus, we expect evolvable pathways to contain more specific enzymes as well. To maximize the ease of exploring new pathways while preserving key metabolic processes, we would expect these specific enzymes to be located at the earliest steps of a pathway. Here, they can serve as molecular gatekeepers, allowing only certain primary metabolites to enter but not impeding the discovery of new structures within the pathway. Manipulating these gatekeeper enzymes to admit different substrates should be an effective way to generate whole new families of natural products.

Thus, features contributing to the evolvability of a biosynthetic pathway include the use of locally specific enzymes and biotransformations that contribute to pathway branching. More-specific gatekeeper enzymes limit what flows through the pathway, providing necessary insurance against chemical chaos. In the end, however, pathways are evolvable because their component enzymes are themselves evolvable, changing properties such as substrate and product specificity readily on mutation. Evolvable enzymes should also exhibit multiple mutational routes to a given altered phenotype and may be comparatively robust to mutation. The former property increases the chances that new phenotypes will emerge, while the latter property allows "scanning" of similar sequences without loss of function. Perhaps surprisingly, it is still an open question whether promiscuous or locally specific enzymes are more evolvable than enzymes that are highly specific. As we detail in this review, as little as a single amino acid substitution can change the specificity of highly specific carotenoid biosynthetic enzymes.

An evolvable pathway, then, would consist of an evolvable gatekeeper enzyme (or set of enzymes) coupled with locally specific downstream enzymes. Such an arrangement maximally exploits mutations in the gatekeeper enzyme, converting a single newly discovered molecule into a potentially large number of new metabolites. Several lines of evidence support the view that the enzymes involved in carotenoid biosynthesis are highly evolvable and are embedded in pathways with such evolvable structures. This review represents the first attempt at gathering together this evidence, which collectively paints a picture of carotenoid biosynthetic pathways as dynamic systems capable of exploring a diversity of product structures much greater than is seen in nature. We now examine the various ways in which this evolvability has been exploited in the laboratory to explore and extend the product diversity of these pathways.

APPROACHES TO ENGINEERING NOVEL BIOSYNTHETIC PATHWAYS

Top Previous Next References

 
New biosynthetic pathways can be created by recruiting foreign catalytic machinery that adds a new branch; such enzyme recruitment happened frequently during the evolution of natural-product pathways (20, 27, 116). A pathway can also diversify when the function of an existing enzyme is altered by mutation or other genetic change. The wide functional diversity that exists within the evolutionarily related carotenoid pathways is evidence that enzyme evolution has played a central role in creating chemical diversity, primarily through modification of substrate and product specificity. In the laboratory, researchers can now employ the same strategies of enzyme recruitment and modification to create new biosynthetic pathways. In this review, we refer to coexpressing genes from different sources in a recombinant organism as gene assembly. This approach can be distinguished from directed evolution of the component enzymes (or their regulation), where libraries of mutant alleles are expressed along with other pathway genes in host cells, which are then screened or selected for desired properties. These two approaches generate novel compounds by mimicking the major modes of natural pathway evolution: gene transfer and divergence. Thus, the two approaches complement one another, and we consider both to fall under the umbrella of laboratory pathway evolution. In this section, we discuss efforts to create new carotenoid biosynthetic pathways by using gene assembly, directed enzyme evolution, and combined approaches.

View larger version (42K): [in this window] [in a new window]   FIG. 2. Selected examples of gene assembly leading to novel carotenoids in E. coli. Combining enzymes from different source organisms can lead to rare or novel carotenoids. In some cases, such as with CrtC and CrtD (4, 5), new pathway branches can be uncovered by replacing an enzyme with its counterpart from a different organism (192). Carotenoids in boxes had not been previously identified in biological material or chemically synthesized. Enzyme abbreviations: Al-1, phytoene desaturase from Neurospora crassa (five-step); CrtC, neurosporene 1,2-hydratase; CrtD, 1-OH-neurosporene 3,4-desaturase; CrtIEu, phytoene desaturase from Erwinia uredovora (four-step); CrtIRc, phytoene desaturase from Rhodobacter capsulatus (three-step); CrtW, ß,ß-carotene 4,(4')-ketolase; CrtX, zeaxanthin glycosylase; CrtY, lycopene ß-cyclase; CrtZ, ß,ß-carotene 3,(3')-hydroxylase. Other enzyme subscripts: Aa, "Agrobacterium aurantiacum" (current designation, Paracoccus sp. strain MBIC1143); Eu, Erwinia uredovora (current approved name, Pantoea ananatis); Rg, Rubrivivax gelatinosus; Rs, Rhodobacter sphaeroides. Natural pathways leading to the synthesis of demethylspheroidene and ß,ß-carotene are shaded. The dotted arrow represents trace enzymatic conversion, and the crossed-out arrow denotes undetectable enzymatic conversion. References to original publications describing the biosynthesis of various compounds are given in brackets beside compound names.

View larger version (23K): [in this window] [in a new window]   FIG. 3. "Matrix" pathway resulting from the coexpression of carotenoid biosynthetic enzymes with broad specificity. Coexpression in E. coli of CrtW [ß,ß-carotene 4,(4')-ketolase] and CrtZ [ß,ß-carotene 3,(3')-hydroxylase] from "Agrobacterium aurantiacum" (current designation, Paracoccus sp. strain MBIC1143) along with the genes for ß,ß-carotene synthesis from E. uredovora (current approved name, Pantoea ananatis) resulted in the biosynthesis of at least nine different ß,ß-carotene derivatives (135). This figure was made using data from reference 135.

Some enzyme specificities not available in nature become readily available by evolution in the laboratory, opening pathways to new metabolites. Decoupled from the biological constraints imposed by the need to fulfill a particular function throughout the course of evolution, a pathway can in principle explore all chemically accessible products, not just those that are biologically relevant (10). Thus, directed evolution gives us the ability to explore the space of possible chemical products much more rapidly and also more thoroughly than does natural evolution. By evolving biosynthetic enzymes, we can anticipate the discovery of large numbers of new natural products on the benchtop in convenient laboratory hosts.

Although each directed evolution project is unique, all involve two main steps: making a library of mutants and searching that library for desired properties. The mutant library is typically generated in an error-prone PCR amplification (38, 124) tuned to generate a certain average point mutation rate or by DNA shuffling, a method that makes new point mutations and recombines existing mutations (193) (many methods are available for making libraries of mutant genes for directed evolution; useful protocols are given in reference 9). The library of mutated DNA molecules is then subcloned into an appropriate vector for expression with other pathway genes in a convenient host organism such as E. coli. Hundreds or thousands of clones are screened in search of (typically) rare ones expressing favorably altered enzymes. A powerful aspect of this process is that it can be iterated—improved variants can be subjected to further rounds of mutagenesis and screening, often leading to further improvements. A sensitive, reliable screen is key to a successful directed evolution experiment: the screen allows the researcher to identify the typically rare mutants with new and interesting properties.

Different strategies exist for evolving metabolic pathways in the laboratory. In an early study, multiple genes in an arsenate degradation pathway were subjected to mutagenesis and recombination all at once (49). This "fully blind" approach makes the most sense when little is known about which genes or noncoding regions (such as ribosomal binding sites) should be mutated to obtain a desired phenotype. However, a significant disadvantage of this approach is that the combinatorial complexity (the size of the library) increases concomitantly with the length of DNA targeted for mutation. Consequently, more clones must be screened to find the rare ones with improved properties. Therefore, when possible, it is attractive to target individual genes for mutation and then screen those mutants in the context of the whole pathway; this can be thought of as a "partially blind" approach. Carotenoid biosynthetic pathways are sufficiently well characterized that investigators generally know which enzyme to target for mutagenesis in order to achieve a new product distribution; however, the specific mutation(s) needed is unknown. Accordingly, researchers have thus far chosen to extend carotenoid pathway branches in a stepwise manner by evolving one gene at a time and expressing the library of variant alleles together with other wild-type or previously laboratory-evolved pathway genes. Schmidt-Dannert et al. demonstrated the effectiveness of this approach for diversifying the range of carotenoid structures accessible from a given set of enzymes (181).

PROBING THE EVOLVABILITY OF CAROTENOID BIOSYNTHETIC ENZYMES AND PATHWAYS

Top Previous Next References

 
In the Introduction, we reasoned that highly evolvable pathways would be assembled from locally specific downstream enzymes and (a small number of) more specific "gatekeeper" enzymes. Furthermore, the enzymes themselves should readily alter specificity on mutation. Modifying a gatekeeper enzyme, accompanied by fine-tuning of downstream enzymes where needed, could allow whole new families of natural products to emerge. In this section, we assess the evolvability of carotenoid biosynthetic pathways and enzymes in view of the (laboratory) data now available.

2,500

View larger version (31K): [in this window] [in a new window]   FIG. 4. Chain elongation reactions catalyzed by natural all-E IDSs. Each IDS adds a defined number of IPP molecules to an allylic diphosphate substrate before the final product is released. Wild-type IDSs have quite stringent product specificity; catalytic ranges of known all-E IDSs are shown with arrows on the right. Natural products derived from the various isoprenyl diphosphates are shown on the left. This figure was made using data from reference 225.

This strategy of substituting residues just upstream of the FARM with smaller or larger amino acids has also been applied to a plant short-chain (C₂₀PP) IDS (104), heterodimeric medium-chain (C₃₀PP and C₃₅PP) IDSs (78, 81, 238), as well as long-chain (C₄₀PP-C₅₀PP) synthases (75, 153) to alter the product chain length. In several cases, substitution of residues lining the substrate binding pocket but distant from the FARM (or even located on the opposite subunit in heterodimeric enzymes) has also modified the product chain length (75, 77, 80, 92, 96, 148, 152, 238, 239). Many of these substitutions enhance the effects of mutations upstream of the FARM, leading to even greater changes in the product chain length.

Several C₁₅PP synthases have some degree of substrate tolerance (56, 107) and can be used for stereospecific synthesis of various insect hormones and related analogs (102, 106, 126). BstFPS was found to accept a number of nonnatural substrates (140), and its substrate range was further broadened by substitutions at Y81, previously described as a chain length-determining residue (151) (see above). The Y81A variant of BstFPS, for example, accepts various isoprenyl diphosphate analogs with -oxygen atoms in their prenyl chains as substrates, leading to the enzymatic synthesis of butterfly hair pencil pheromone and analogs (125). In theory, all of these isoprenoid diphosphate analogs could serve as building blocks for the enzymatic synthesis of carotenoid analogs.

IDSs from diverse species have exhibited evolvability in the laboratory. Although the wild-type IDSs are quite specific with respect to product size, minor genetic changes were found to profoundly alter the number of elongation steps and therefore the size of the products formed by these enzymes. The residues that were found to change the specificity of these IDSs upon mutation form a key part of the reaction pocket that accommodates the elongating isoprenoid chain, and it appears that product specificity is largely dependent on the size of this pocket (61, 203). Inspection of the data from the reports describing the modification of product specificity reveals that, in general, the mutant IDSs have a broadened rather than a shifted product range. In most cases, mutants synthesize multiple new products as well as the old one. In contrast, wild-type IDSs, although related by evolution, are very specific with respect to the length of their products. We now know that a single amino acid substitution can broaden the product range of an IDS. However, we do not know the degree of genetic change required to completely shift IDS product specificity such that it synthesizes only isoprenyl diphosphates of a new length.

It has long been recognized that the enzyme-catalyzed reaction leading to formation of the carotenoid backbone is very similar to the reaction catalyzed by squalene synthase (SqS), the first committed enzyme in cholesterol biosynthesis (32). Carotenoid synthases and SqS employ the same mechanism, and SqS produces the C₃₀ carotenoid backbone if deprived of NADPH (88, 202, 237). Carotenoid synthases and SqS also share several highly conserved domains, and it is likely that the two have a common evolutionary origin (176).

The biosynthesis of carotenoid backbones (and squalene) has proven to be a complex process (22, 88, 89) (Fig. 5a). The reaction proceeds in two distinct steps. The first consists of abstraction of a diphosphate group from a prenyl donor followed by 1-1' condensation of the donor and acceptor and loss of a proton to form a stable cyclopropyl intermediate. In the second step, the cyclopropyl intermediate is rearranged, the second diphosphate group is lost, and the resultant carbocation is quenched after the loss of another proton (Fig. 5a). Biochemical (156) and structural (159) studies have shown that these two subreactions occur in physically distinct sites in the enzyme (Fig. 5b).

View larger version (36K): [in this window] [in a new window]   FIG. 5. Product length determination by CrtM and mutants thereof. (a) Reaction mechanism of squalene synthase (left pathway) and CrtM (right pathway). Both enzymes catalyze the head-to-head condensation of two molecules of C15PP to yield linear C30 hydrocarbons. The enzymes employ similar mechanisms involving a common stable intermediate, presqualene diphosphate (PSPP). The mechanisms differ only in the final rearrangement step, which is followed in the case of squalene synthase by reduction of the central double bond. (b) Crystal structure of human squalene synthase (PDB ID 1EZF). Green residues are involved in the first half-reaction (formation of PSPP), while blue residues form the pocket of the second half-reaction in which PSPP is rearranged. Red residues correspond by sequence alignment to F26 and W38 of CrtM, which were found to control the product size of that enzyme (213, 214). Shown in yellow are the residues that align with mutations in CrtM that improve both its C30 and C40 carotenoid synthase activities. (c) Mutation analysis of F26 and W38 of CrtM. Various site-directed variants of CrtM (indicated by their one-letter amino acid code) were constructed and tested for their synthase activity in both C30 and C40 carotenoid pathways. Variants are arranged from left to right in increasing order with respect to the van der Waals' volume of the substituted amino acid. The C30 carotenoid synthase activity of each variant was estimated from the level of yellow pigmentation when coexpressed with CrtN. Similarly, the C40 carotenoid synthase function of each variant was evaluated in terms of the level of red pigmentation when coexpressed with CrtE and CrtI. Panels a and b were made using data from reference 212.

Carotenoid synthases, positioned early in the biosynthetic pathway and possessing the ability to discriminate between isoprenyl diphosphate substrates with different numbers of isoprene units, appear to play the role of pathway gatekeeper, particularly in organisms in which multiple potential substrates are available. Indeed, as we illustrate, downstream carotenoid biosynthetic enzymes are less specific than the synthases and appear to recognize only a particular motif of the substrate. In the Introduction, we reasoned that gatekeeper enzymes should be attractive targets for pathway diversification. If carotenoid synthases could be engineered to accept different prenyl substrates and synthesize carotenoid backbones from them, the locally specific downstream enzymes might accept these new backbones and generate the corresponding metabolites. This would give rise to whole new pathway families. Compared to the IDSs, however, little is known about the molecular basis for the substrate and product specificity of carotenoid synthases.

Following the lead of those who studied the IDSs (96, 122, 125, 141, 147, 148, 150-152, 203, 238, 239), we probed the evolvability of size specificity in the carotenoid synthases by looking for mutants that complement a non-native pathway. Using error-prone PCR to perform random mutagenesis, expressing the mutant genes with the remaining genes necessary to produce lycopene in E. coli, and screening to find red colonies, we identified single amino acid substitutions in the S. aureus C₃₀ carotenoid synthase CrtM (F26L or F26S) that confer the ability to make the C₄₀ backbone just as efficiently as CrtB (214). Repeating this experiment using a PCR method designed to guarantee that the libraries would be free from mutation at F26 allowed us to uncover two additional amino acid substitutions, W38C and E180G, that also confer C₄₀ function on CrtM (213). Mapped onto the crystal structure of human squalene synthase (SqS [Fig. 5b]), F26 and W38 appear in helices B and C, respectively, and the side chains of both residues point into the pocket that accommodates the second half-reaction (rearrangement of the cyclopropyl intermediate). Replacement of these amino acids with smaller ones significantly increased the C₄₀ synthase activity of CrtM (Fig. 5c). Conversely, the C₃₀ synthase performance was the highest for wild-type CrtM and decreased with decreasing size of the amino acid residues at these positions. This analysis led to our proposal that wild-type CrtM is able to perform the rate-limiting first half-reaction of phytoene synthesis—condensation of two molecules of C₂₀PP to form the cyclopropyl intermediate, prephytoene diphosphate—but the second half-reaction is prevented from going to completion by steric inhibition of the normally fast rearrangement step. Unable to convert into phytoene in wild-type CrtM, prephytoene diphosphate either remains stuck in the enzyme, departs, or undergoes other types of rearrangement to yield noncarotenoid products. (A similar phenomenon is well known for SqS; in the absence of NADPH, which is required to convert presqualene diphosphate to squalene, SqS produces a complex mixture of nonsqualene compounds including rillingol, 10-hydroxybotryococcene, and 12-hydroxysqualene [22, 89].) However, in CrtM mutants where F26 or W38 is replaced with a smaller or more flexible amino acid, the prephytoene diphosphate formed from two molecules of C₂₀PP is efficiently rearranged to form phytoene. Thus, the F26 and W38 mutations apparently modify the product specificity, not the substrate specificity of CrtM: they do not act by allowing the enzyme to bind and accept GGPP as a substrate but, rather, allow the intermediate prephytoene diphosphate to be converted to phytoene.

Gain of C₄₀ function by mutagenesis of CrtM usually came at a cost to the original C₃₀ synthase activity. For example, some enzymes doubly mutated at F26 and W38 showed only negligible C₃₀ synthase activity. However, other modes of obtaining C₄₀ synthase activity were possible. The E180G substitution increased performance in both the C₃₀ and C₄₀ contexts (212). Mapped onto the SqS structure, E180G is positioned outside the reaction pocket but close to the site of the first half-reaction. We think this mutation accelerates the rate-limiting first half-reaction. Several such activating mutations were found by random mutagenesis of CrtM_F26S in a successful search for improved or restored C₃₀ function (D. Umeno, unpublished results). All of these activating mutations were rather far from the reaction center (Fig. 5b, indicated in yellow) and enhanced both C₃₀ and C₄₀ synthase activity. These mutations could be increasing expression level, stability (half-life in vivo), or specific activity.

Like IDSs, carotenoid synthases appear to be specific, but at least in the case of CrtM, that specificity is readily altered by mutation. Furthermore, there are multiple mutational routes to the same altered phenotype. In contrast, in similar experiments with CrtB, we were unable to find any mutations that conferred C₃₀ carotenoid synthase function. We are tempted to speculate that CrtM, by virtue of the fact that it does not need to select C₁₅PP from C₂₀PP in its natural host (C₂₀PP is not made in C₃₀ carotenoid-producing organisms), is inherently more evolvable toward accepting C₂₀PP than is CrtB toward accepting C₁₅PP, because CrtB has always had to select C₂₀PP over C₁₅PP (the latter is present in all organisms). CrtM can in fact accept C₂₀PP in a hybrid reaction: when supplied with both C₁₅PP and C₂₀PP, wild-type CrtM can condense one molecule of each to synthesize a C₃₅ carotenoid backbone (see below). CrtM has thus proven to be an evolvable enzyme while CrtB has not, at least with respect to the specific tasks of accepting a larger (CrtM) or smaller (CrtB) substrate. As was the case with the IDSs, a single amino acid substitution was sufficient to increase the size of the products synthesized by CrtM. Furthermore, as we describe below, the mutants of CrtM that function in a C₄₀ carotenoid biosynthetic pathway are also capable of synthesizing even larger carotenoid backbones when supplied with the appropriate precursors.

In general, bacterial C₄₀ desaturases are functional on C₃₀ carotenoids and vice versa. It is proposed that carotenoid desaturases recognize only a portion of the substrate molecule common to both C₃₀ and C₄₀ carotenoid backbones (163, 174). Subsequent carotenoid-modifying enzymes are also often locally specific (see below). Because each of the desaturation intermediates represents a branch point for further diversification by downstream enzymes, altering and controlling the desaturation step number is key for creating extensive molecular diversity (70). Nature has done this: many different carotenoid desaturases are known in C₄₀ pathways, each with its specific step number. Carotenoid desaturases which primarily catalyze two, three, four, and five desaturation steps are known; one-step and six-step desaturated carotenoids have been reported only as minor products in natural carotenogenic organisms. The ability of carotenoid desaturases to accept different substrates was demonstrated when C₃₀ and C₄₀ desaturases were tested in each other's pathway (163, 214). C₄₀ desaturases from Erwinia (four-step enzymes), Rhodobacter (three-step), and Anabaena (two-step) are all active on C₃₀ substrates. Similarly, the C₃₀ desaturase CrtN showed measurable activity in a C₄₀ pathway. The localized specificity of these desaturases has been exploited in engineered pathways (211) (see below).

The desaturase step number can be inferred from the color (or color change) of the carotenoids produced in vivo. This provides an excellent basis for screening mutant desaturases in the laboratory for the ability to accept new substrates or for changes in product specificity. Using this principle, Schmidt-Dannert et al., working in this laboratory, readily isolated desaturase variants with altered step number in a C₄₀ pathway (181). Starting from a library made by DNA shuffling of two closely related four-step phytoene desaturases (crtI from P. agglomerans [E. herbicola] and crtI from Pantoea ananatis [E. uredovora]), they isolated one four- to six-step variant, CrtI₁₄ (Fig. 6), as well as 20 variants that catalyze fewer (less than four) steps. Wang and Liao conducted similar experiments with the three-step desaturase from Rhodobacter sphaeroides (222). Two rounds of random mutagenesis by error-prone PCR and color screening resulted in variants that accumulated the four-step product, lycopene. At least five different mutations increased the step number of this desaturase. Furthermore, some combinations of mutations resulted in yet higher production of lycopene (222).

View larger version (31K): [in this window] [in a new window]   FIG. 6. Extension of laboratory-evolved pathways to 3,4,3',4'-tetradehydrolycopene and torulene by coexpression of downstream carotenoid modifying enzymes. Boxed structures had not been previously identified in biological material. Native enzyme reactions are shown with black arrows. Gray arrows depict reactions not seen in natural organisms, and laboratory-evolved enzymes are written in gray lettering. Double arrows indicate enzymatic modification of both ends of a carotenoid substrate. Enzyme abbreviations: CrtA, spheroidene monooxygenase; CrtI, phytoene desaturase from E. uredovora (current approved name, Pantoea ananatis) (four-step), CrtI14, laboratory-evolved mutant of CrtI (four- to six-step) (180); CrtO, ß,ß-carotene 4,(4')-ketolase; CrtU, ß,ß-carotene desaturase; CrtX, zeaxanthin glycosylase; CrtY, lycopene ß-cyclase; CrtY2, laboratory-evolved mutant of CrtY (180); CrtZ, ß,ß-carotene 3,(3')-hydroxylase. Biosynthesis of oxygenated derivatives of -carotene and neurosporene was attributed to early termination by CrtA of the desaturation sequence of CrtI, which is normally a four-step desaturase (118). Although CrtA is primarily a ketolase, it is thought to catalyze the introduction of both hydroxy groups of phillipsiaxanthin (C. Schmidt-Dannert, personal communication). This figure was made using data from reference 118.

View larger version (43K): [in this window] [in a new window]   FIG. 7. The step number of the C30 desaturase CrtN is readily modified by mutation. A library of mutant crtN alleles was coexpressed with the C30 carotenoid synthase crtM in E. coli. Wild-type CrtN is a three- to four-step desaturase when expressed in E. coli, causing colonies to appear orange (inset). Colonies harboring CrtN mutants with increased step number were deep orange and red (highlighted by arrows), reflecting the longer carotenoid chromophores formed by these mutant desaturases. Colonies harboring mutants with decreased step number appear yellow/lemon, while colonies with nonfunctional or single-step desaturases are pale since these products are colorless. Shown below are the possible C30 desaturation products and the desaturases that synthesize them. CrtN1-3 are discovered mutants of CrtN with altered desaturation step number. Each colored box depicts the approximate color of the carotenoid in white light. Molecules without colored boxes are colorless. Abbreviations: lacP, lac promoter; wt, wild type.

Carotenoid desaturases tend to exhibit more well-defined product specificity in their natural hosts than when (over)expressed in a heterologous host. For instance, CrtN appears to be a three-step enzyme and produces almost exclusively 4,4'-diaponeurosporene in S. aureus (226). In a heterologous host, however, its desaturation step number is less distinct. Several groups have reported that CrtN produces the four-step 4,4'-diapolycopene as a major product in an E. coli system (118, 163, 211, 214). Desaturase step number can also be altered by manipulation of enzymes further downstream in the carotenoid pathway (69). It is not surprising that altering the environment of a desaturase can alter its product specificity. If downstream enzymes that normally remove certain desaturation products are not present, the desaturase may have an opportunity to catalyze further desaturation steps. In addition, if these enzymes associate in complexes, expression of a desaturase with other carotenoid biosynthetic genes from different organisms could yield complexes with altered or suboptimal substrate transfer properties, which could affect the desaturation step number. Further research should shed light on the source of desaturase specificity and on the ways in which specificity can be altered.

View larger version (24K): [in this window] [in a new window]   FIG. 8. Cyclization of carotenoids. (a) Mechanism of carotenoid cyclization. Lycopene with its two -end groups is the preferred substrate for all known carotenoid cyclases. All cyclases produce the same carbocationic intermediate on protonation of C-2 of the substrate. Product ring type (ß, , or ) is determined by which proton (a, b, or c in the figure) is eliminated in the subsequent rearrangement step. This panel was made using data from reference 32. (b) Substrates accepted by lycopene cyclases. The shaded portion(s) of each structure is cyclized. The first three compounds, with their -end groups, have long been known as natural substrates for lycopene cyclases. The remaining seven structures, including neurosporene with its 7,8-dihydro--end (left side of molecule) and -carotene with two 7,8-dihydro--ends, have more recently been shown to be cyclized.

All known ß-cyclases, except one recently discovered in a marine bacterium (207), create rings at both ends of lycopene. In contrast, known -cyclases except those from Lactuca sativa (lettuce) and the flower Adonis aestivalis generate monocyclic carotenoids (52). In an attempt to manipulate the ring number and understand how it is determined, Cunningham and Gantt constructed a series of single-crossover chimeras of one-step (Arabidopsis thaliana) and two-step (Lactuca sativa) -cyclases (52). Analysis revealed a region of six amino acids involved in ring number determination. Further mutagenesis experiments identified single-amino-acid substitutions that alter ring number specificity: in the lettuce two-step cyclase Dy4, substitution of H457 with leucine converted the enzyme into a monocyclase; the corresponding L448H mutation in Y2, the Arabidopsis monocyclase, resulted in an enzyme that forms two rings. Thus, ring number specificity can be modulated with a single amino acid substitution. The authors of this study suggested that the residues they identified may play a role in dimer formation (52). Lycopene may be oriented in the plasma membrane such that one of its ends is more accessible to the cyclase than the other. Only when cyclase dimers are formed does binding of the more accessible end bring the less accessible end close enough to the other subunit for it to be cyclized as well. An alternative but similar explanation for the above observations is that the mutations alter the cyclase oligomerization state within a carotenoid-synthesizing complex. For example, when one cyclase subunit is present in each complex, carotenoid products are cyclized on only one end whereas complexes with two cyclase subunits would cyclize carotenoids at both ends. Association of cyclase monomers in a complex would depend on their interactions with each other and the other constituents of the complex, and a single amino acid substitution could be sufficient to disrupt (or promote) this association. It is also possible that the single amino acid substitutions described above alter the enzyme's intrinsic preference for the carotenoid substrate that has already been cyclized on one end. The H457L mutation may reduce the preference of Dy4 for the monocyclic substrate, while the L448H mutation increases the ability of Y2 to cyclize it.

CrtI₁₄ is a four- to six-step desaturase, and E. coli cells expressing this variant along with CrtE (a C₂₀PP synthase) and CrtB (a C₄₀ carotenoid synthase) can synthesize 3,4,3',4'-tetradehydrolycopene (Fig. 6) (181). When CrtY, the ß-cyclase from Erwinia, was coexpressed with these enzymes, the cells synthesized exclusively ß,ß-carotene—the same product made by cells expressing wild-type CrtE, CrtB, CrtI, and CrtY. Directed evolution, however, could create a pathway leading to a new cyclized product: after performing DNA shuffling to make a library of CrtY variants and coexpressing these with CrtI₁₄ and CrtE, Schmidt-Dannert et al. identified clones that accumulated the monocyclic carotenoid torulene (Fig. 6) (181). Pigment analysis from cells harboring the CrtY mutants including CrtY₂ (Fig. 6) revealed torulene together with lycopene, 3,4,3',4'-tetradehydrolycopene, ß,-carotene, and ß,ß-carotene. Torulene-producing variants were also discovered when the CrtY library was created by error-prone PCR (Umeno, unpublished). These mutants made up 2 to 5% of the library. This high frequency of torulene-producing mutants suggests that the torulene pathway emerged by down-regulation of cyclase activity. Mutants of CrtY with decreased catalytic activity would compete less efficiently with CrtI₁₄ for the ends of the carotenoid substrate, with the result that only one end is cyclized while the other is desaturated, leading to torulene. Alternatively, the CrtY mutants found in the torulene-producing clones may be compromised in their ability to form dimers, either as part of larger carotenoid enzyme complexes or not.

Although the underlying mechanisms for the changes in cyclase step number remain obscure, it is clear that carotenoid cyclases are evolvable: their phenotype can change dramatically with a small number of mutations, and there appear to be multiple pathways to a particular phenotype. As we show in the next section, combining evolved carotenoid desaturases and cyclases with locally specific enzymes further downstream allows an even larger natural product space to be sampled. To date, carotenoid cyclases have not been evolved in the laboratory for altered product specificity, for example to convert a ß-cyclase to an -cyclase or vice versa. These cyclases have very similar chemical mechanisms (Fig. 8a), and we predict that cyclase product specificity will be easily modified by mutation.

Lee et al. exploited the catalytic promiscuity of a number of carotenoid-modifying enzymes in an E. coli system to extend laboratory-evolved pathways to 3,4,3',4'-tetradehydrolycopene and torulene, generating biosynthetic routes to several carotenoids not identified in nature (118) (Fig. 6). CrtA, an oxygenase from Rhodobacter capsulatus that normally acts on the methoxy carotenoid spheroidene, was found to insert one keto group into -carotene, neurosporene, and lycopene and to insert two keto groups into tetradehydrolycopene. Monocyclic torulene, synthesized via the pathway constructed using the laboratory-evolved desaturase and cyclase (181), served as a substrate for CrtO, a ß,ß-carotene ketolase from Synechocystis sp. strain PCC 6803; CrtU, a ß,ß-carotene ring desaturase from Brevibacterium linens; and CrtZ, a ß,ß-carotene hydroxylase from E. herbicola. In addition, 3-hydroxytorulene, the product of the action of CrtZ on torulene, was further metabolized by CrtX, a zeaxanthin glycosylase from E. herbicola, leading to the synthesis of torulene-3-ß-D-glucoside (Fig. 6). In this work, the pathway to torulene generated by directed evolution of two upstream carotenoid biosynthetic enzymes was extended in several different directions by adding genes for further transformations. Owing to their intrinsic localized specificity, coexpression of these downstream modifying enzymes with the foreign torulene pathway resulted in a series of novel torulene derivatives.

There are no reports yet of directed evolution of carotenoid oxygenases or post-cyclase carotenoid-modifying enzymes. Work by Sun et al. (197), however, provides an interesting example of how modifying the sequence of a carotenoid hydroxylase can alter the enzyme's product distribution. The ß,ß-carotene hydroxylase from A. thaliana catalyzes two hydroxylation steps, converting ß,ß-carotene to zeaxanthin (>90%) in A. thaliana and when expressed in E. coli (Fig. 9). Noticing that this enzyme had an N-terminal extension of more than 130 amino acids compared with other known ß,ß-carotene hydroxylases, Sun et al. expressed in E. coli a truncated version of the Arabidopsis hydroxylase lacking the N-terminal 129 amino acids. The truncated enzyme primarily catalyzed only one hydroxylation step, and ß-cryptoxanthin (Fig. 9) accumulated as the main product (>75%) (197). The molecular basis for this altered hydroxylation step number is not known; the authors speculated that truncation yielded an enzyme deficient in the ability to form dimers.

View larger version (23K): [in this window] [in a new window]   FIG. 9. Alteration of product specificity of the ß,ß-carotene hydroxylase from Arabidopsis thaliana. The wild-type enzyme catalyzes two hydroxylation steps, converting ß,ß-carotene to zeaxanthin when expressed in either A. thaliana or E. coli. When the N-terminal 129 amino acids (aa) were removed, the truncated enzyme expressed in E. coli catalyzed only one hydroxylation step, leading primarily to synthesis of ß-cryptoxanthin (197).