Alteration of Product Specificity of Rhodobacter sphaeroides Phytoene Desaturase by Directed Evolution*

Phytoene desaturases occurring in nature convert phytoene to either neurosporene or lycopene in most eubacteria. Approximately 10% of known phytoene desaturases, as in Rhodobacter , produce neurosporene, whereas the rest produce lycopene. These two types of enzymes, although similar in function, have relatively low similarity (below 60%) in terms of nucleotide or amino acid sequence. The mechanism controlling the product specificity of these enzymes is unclear. Here we used directed evolution to change the product of Rhodobacter sphaeroides phytoene desaturase ( crtI gene product), a neurosporene-producing enzyme, to lycopene. Two generations of random mutagenesis were performed, from which three positive mutants were isolated and sequenced. We then used site-directed mutagenesis to determine the effect of each amino acid change. Gathering information from random mutagenesis, we further recombined the beneficial mutations by site-directed mutagenesis and increased the percent of lycopene production to 90%.

Directed evolution has been used in many aspects of biological systems, such as protein engineering (1), vaccine development (2), biochemical production (3), and metabolic engineering (4). In general, it involves fast mutation of the target gene(s), followed by an efficient screening or selection system for a desirable trait. Products with improved functions or novel compounds were generated using this approach (5). Directed evolution can also be used to generate desirable mutants from which sequence alternation can be correlated with function. In this study, we demonstrate an example of directed evolution for the alteration of product specificity. Analysis of point mutations sheds light on the mechanism determining the product specificity.
Phytoene desaturase in eubacteria (crtI gene product) (6) is an important enzyme in the carotenoid pathway. It catalyzes the desaturation of phytoene to either neurosporene or lycopene (see Fig. 1). Lycopene is a red pigment and an antioxidant that has been shown to have preventive effects against certain cancers (7). The yellow carotenoid neurosporene is structurally different from lycopene by one double bond with one-step earlier termination in the desaturation reaction (see Fig. 1). These compounds are metabolites downstream from geranylgeranyl diphosphate (GGPP) 1 in the isoprenoid pathway via phytoene synthase encoded by crtB. About 10% of analyzed phytoene desaturases produce neurosporene, such as the enzymes from Rhodobacter, whereas the rest produce lycopene, such as in Agrobacterium aurantiacum (8), Erwinia sp. (9), and other photosynthetic bacteria (10). At the DNA level, the crtI gene encoding phytoene desaturase of the Rhodobacter species is significantly different from the ones from other organisms. Between the crtI genes of Rhodobacter and others, the amino acid sequence similarity is below 60% although there are some conserved regions, especially in the C-terminal domain (11). Because it is highly distinct between these two classes of phytoene desaturases, it would be relatively difficult to perform DNA family shuffling (12) or other in vitro recombination methods.
Because these two types of phytoene desaturases share relatively low nucleotide identity or amino acid sequence similarity (below 60%), it cannot be deduced from a sequence comparison which are the residues that determine product specificity. Recently, Schmidt-Dannert et al. (5) mutated Erwinia phytoene desaturase to produce a more desaturated compound, 3,4,3Ј,4Ј-tetradehydrolycopene (see Fig. 1). To determine the amino acid residue(s) important for controlling product specificity in phytoene desaturases, we used error-prone PCR mutagenesis to evolve Rhodobacter sphaeroides phytoene desaturase to produce lycopene. The mutated amino acids were individually tested by site-directed mutagenesis for their importance in determining product specificity.

EXPERIMENTAL PROCEDURES
Strains, Plasmids, and Materials-The crtI gene encoding phytoene desaturase was cloned from R. sphaeroides ATCC17025, according to the published crtI sequence of another R. sphaeroides NCIB 8253 (13). The 1.7-kb crtI fragment was amplified from the genomic DNA using PCR with a pair of designed degenerate primers (5Ј-TGGGGTACCTG-CATGTCATTCCTC-3Ј and 5Ј-TGCTCTAGATCATTCCGCGGCAAG-3Ј). The PCR product was ligated into the pCR2.1® vector (Invitrogen Inc., Carlsbad, CA) by the TA-cloning technique and cotransformed into competent TOP10 host (Invitrogen, Inc.) with plasmid pCW18, which overexpresses the necessary enzymes for producing the substrate, phytoene, in Escherichia coli (Fig. 1). The plasmid pCW18 is spectinomycinresistant and harbors the dxs gene encoding 1-deoxy-D-xylulose 5-phosphate synthase from E. coli, gps encoding GGPP synthase from Archaeoglobus fulgidus (14), and crtB encoding phytoene synthase from Erwinia uredovora on the vector pCL1920 (15). This plasmid was constructed by destroying the crtI gene on pCW9 (containing dxs, gps, and crtBI from E. uredovora) (16) with BamHI and then filled in the overhang and self-ligated. Among the yellow transformants due to the activity of the R. sphaeroides crtI gene, one was picked, and the plasmid harboring crtI was isolated, sequenced, and designated as pCW19. The sequence of our cloned gene is 87% identical to the published one for R. sphaeroides NCIB 8253 (13) at the nucleotide level and around 93% similar at the amino acid level. The plasmid pCW19 was found to contain the crtI-lacZ␣ fusion with 10 amino acids (32 nucleotides) of crtI truncated at the C terminus (see Fig. 3A). However, the yellow pigment was confirmed to be neurosporene by high performance liquid chromatography (HPLC) measurement.
Error-prone PCR Mutagenesis-Random mutagenesis of the R. sphaeroides crtI gene on the plasmid pCW19 was performed using methods * This work was supported in part by the National Science Foundation (BES 9814097). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ To whom correspondence should be addressed: Dept. of Chemical Engineering, 5531 Boelter Hall, University of California, Los Angeles, CA 90095. Tel.: 310-825-1656; Fax: 310-206-4107; E-mail: liaoj@ucla.edu. 1 The abbreviations used are: GGPP, geranylgeranyl diphosphate; described previously (17). A pair of primers (5Ј-AGCGGATAACAATT-TCACACAGGA-3Ј and 5Ј-CGCCAGGGTTTTCCCAGTCACGAC-3Ј) flanking the crtI gene on the pCR2.1 vector was designed to amplify the 1.7-kb crtI fragment by PCR under mutagenic conditions. The PCR products were purified using a QIAquick gel extraction kit (Qiagen, Valencia, CA) followed by digestion with the restriction enzymes HindIII and EcoRV. The fragments were cloned into the corresponding sites of vector pCR2.1 and transformed into competent TOP10/pCW18 cells. The colonies were grown on Luria-Bertani (LB) plates containing ampicillin (200 g/ml) and spectinomycin (50 g/ml) at 30°C in the dark for 24 h (or until color developed). The colonies were first screened visually on the plates, and the ones with (slight) color change were cultured in liquid LB medium supplemented with the same antibiotics. The fermentation products were analyzed further by HPLC. The wildtype plasmid pCW19 was used as the template in the first generation of random mutagenesis. The mutant plasmid pCW19 m1 generated from the first round of directed evolution was used as the template in the second generation of random mutagenesis.
Site-directed Mutagenesis-A QuikChange (Stratagene, La Jolla, CA) site-directed mutagenesis kit was used. Various mutagenic primers were designed individually according to the desired mutations. The cycling parameters for the mutagenesis reactions were chosen based on the protocol suggested by the manufacturer. After PCR amplification by PfuTurbo DNA polymerase, the reaction mixture was digested with 10 units of the restriction enzyme DpnI for 1 h. Transformation of 10 l of DpnI-treated reaction mixture into the competent cells yielded hundreds of colonies with the desired mutations.
Analysis of Carotenoid Production-After the cells were centrifuged, carotenoids were extracted from the pellets by 30:70 methanol-MTBE (methyl tert-butyl ether). The extracts were filtered through Sep-Pak cartridges (Waters, Milford, MA), eluted with the same solvent and characterized using an Alliance HPLC system (Waters, Milford, MA) equipped with a Waters 996 photodiode array detector. A 20-l aliquot of extract was injected into the Carotenoid TM column (2.0 ϫ 150 mm, 3 m; YMC, Inc., Wilmington, NC), and eluted with methanol-MTBE (30:70, isocratic) at a flow rate of 0.2 ml/min.

Directed Evolution of Phytoene Desaturase by Random PCR
Mutagenesis-Similar to the two-plasmid system in our previous work (4), we cloned the target crtI gene from R. sphaeroides ATCC17025 on one plasmid and the rest of the pathway genes (dxs, gps, and crtB) onto another compatible plasmid (pCW18). Random mutations were introduced to the crtI gene by PCR under mutagenic conditions. The 1.7-kb PCR fragment contains the upstream promoter region, the open reading frame of the crtI gene, and the downstream region. After mutation, restriction digestion, purification, and re-ligation, the plasmid DNA was transformed into the host strain TOP10 containing pCW18. The colonies were grown on plates to screen for different colors due to the accumulation of different carotenoids. Approximately 7,500 colonies were screened on the plates visually for the first generation of directed evolution. The E. coli cells expressing the wild-type R. sphaeroides phytoene desaturase appeared deep yellow. One colony showed a color change from yellow to slightly golden. This candidate mutant was grown in LB medium with the appropriate antibiotics, and the products were identified by HPLC. Results showed that this mutant strain produced a mixture of carotenoids, including neurosporene (45%), lycopene (32%), and some amounts of -carotene (Fig. 2A). The plasmid harboring the mutated crtI gene was isolated and designated as pCW19 m1.
We proceeded to perform the next mutagenesis by errorprone PCR using pCW19 m1 as a template. About 10,000 colonies were screened, and their colors were compared with the control harboring pCW19 m1. Twenty colonies showed a color change, and two of them appeared completely pink. These two mutants were grown in LB medium with antibiotics, and the carotenoids produced were subjected to HPLC analysis. Both mutants produced a mixture of carotenoids with the majority being lycopene (77 and 78%) and small amounts of neurosporene (20 and 16%) and -carotene (3 and 6%) ( Fig. 2A). The two plasmids harboring the mutated crtI genes from the second mutant generation were isolated and designated as pCW19 m2 and pCW19 m3, respectively.
Sequence Analysis-The three mutants m1, m2, and m3 were sequenced. Three point mutations were found on m1. Two of them led to the amino acid changes, F220S and E508G, whereas the third one caused the early truncation of the protein by changing the amino acid arginine into a stop codon (see Fig. 3A and Table I). Three pairs of mutagenic primers for site-directed mutagenesis were designed to investigate the effect of each point mutation. We found the change to stop codon caused the most significant change to product distribution (Fig.  2B), followed by E508G, and the mutation F220S had no effect on product distribution. These results are summarized in Table  I and shown schematically in Fig. 3B.
Four additional point mutations were found in mutant m2 compared with the parental template (m1), which lead to four amino acid changes, H12Q, V68D, F166I, and M402T. Again, site-directed mutagenesis was performed to determine the effect of each point mutation. We found that V68D and F166I were the two beneficial mutations that help the conversion of product from neurosporene to lycopene (see Fig. 2C and Fig.  3B). The mutation H12Q has little effect, whereas M402T has a negative effect on percent lycopene production.
Mutant m3 has three additional point mutations compared with the m1 template, including one amino acid change, L148H, and two silent mutations that cause no amino acid changes. We verified the effect of L148H by introducing this mutation into m1 by site-directed mutagenesis and, as expected, found that it increased the production of lycopene dramatically to about 80%. Moreover, we also introduced this mutation alone into the wild-type template pCW19, and the effect is also prominent (52% lycopene) (Fig. 2D).
Protein Evolution by Site-directed Mutagenesis-Based on the information obtained from two generations of directed evolution by random mutagenesis, we know that several amino acid changes were beneficial for achieving higher production of lycopene. The two final mutants generated from random mutagenesis, m2 and m3, showed nearly no difference in color compared with the positive control harboring pCW9 that overexpressed the lycopene-producing CrtI from E. uredovora (16), although they still synthesize small amounts of neurosporene. Thus, this visual limitation hinders the further use of color screening to isolate the next-generation mutants by evolution.
However, we can resort to site-directed mutagenesis based on information generated from directed evolution. Our results showed that L148H had the most significant positive effect among all the mutations, followed by F166I, R514Stop, V68D, and E508G. On the other hand, M402T had a negative effect for lycopene production whereas F220S and H12Q have negligible effects. To optimize lycopene production, we then combined all the positive mutations and eliminated negative mutations by site-directed mutagenesis. First, L148H was introduced into the mutant m2, because it does not have this mutation, and its lycopene production is already high (80%). The result showed that L148H further increased the lycopene production up to 90% (see Fig. 2D and Fig. 3B).
We suspected that we could further increase the production by removing the negative mutation M402T. To do so, we started from the mutant m1 and introduced mutations V68D and F116I and found the lycopene production of this strain was just slightly higher than that of m2 or m3. Finally, we incorporated L148H into the above strain, and its lycopene production level was around 90% (Fig. 3B). Note that this level of lycopene production was achieved with or without the negative mutation M402T. It appears that the negative effect was compensated by multiple positive mutations. DISCUSSION In this work, we evolved the R. sphaeroides phytoene desaturase to change its product from neurosporene to lycopene. We used site-directed mutagenesis to investigate the relative importance of each mutation to product specificity and combined the positive mutations. By combining error-prone PCR and site-directed mutagenesis, we changed the neurosporene-producing enzyme to lycopene-producing. We conclude that there are two important regions of this enzyme in determining the product distribution from neurosporene to lycopene. One is in the C-terminal region; the other is close to the N terminus (Val 68 , Phe 166 , and Leu 148 ) that may contain the substratebinding motifs.
Although many carotenoid biosynthetic genes have been cloned from various organisms, no crystal structures of these enzymes, including phytoene desaturase, are available so far. Previously, the hydrophobic C-terminal domain of the phytoene desaturase was identified to be conserved among carotenoid desaturases and required for the interaction in the dehydrogenation reaction (11). Schmidt-Dannert et al. (5) also showed that one point mutation in this domain changes product distribution. Sequence analysis of our mutant m1 showed two positive mutations near this C-terminal hydrophobic domain. The major one introduced a stop codon; the other caused the amino acid change E508G. The crtI genes of Rhodobacter sp. are longer in the C terminus compared with those of the organisms that produce lycopene. Thus, the C-terminal domain appears to be important in stopping the desaturation process at the level of neurosporene. Without this domain, the desaturation process continues to lycopene.
From the number of the colonies in the screening process, we noticed that it is relatively difficult to obtain a positive mutant in the first generation of random mutagenesis. It may be because of mutating a larger fragment (1.7-kb) by error-prone PCR, the use of the color screening system, or the rigidity of the enzyme. Once a mutant was obtained, several positive mutants FIG. 2. HPLC measurements of the carotenoids produced from the various crtI mutants by random and site-directed mutagenesis. All contain the mixture of two major carotenoids. The retention time of neurosporene is 3.70 min and that of lycopene is 4.85 min, and there are trace amounts of -carotene at the retention time of 3.1 min. AU represents the absorbance unit at 450 nm on the HPLC chromatograms. A, the carotenoids production of the mutants generated by random mutagenesis. B, site-directed mutagenesis to determine which amino acid changes are important in the mutant m1. C, analysis of the four new mutations introduced into the mutant m2 in second mutant generation. D, the significance of the residue change L148H from m3 in the evolutions of WT, m1, and m2. Combining the most effective mutations from m2 and L148H from m3 can achieve a shift of product from neurosporene up to 90% lycopene (Table I).
appeared in the second generation of directed evolution. We were able to achieve nearly 100% of product conversion by two generations of directed evolution without affecting the overall enzyme activity.
The mutants generated from the second mutagenesis process give us more information on the (primary) structure-function relationship of this enzyme. The mutant m2 has two positive mutations, V68D and F166I, and m3 has only one beneficial mutation, L148H, in addition to their parent template, m1. The R. sphaeroides CrtI protein sequence was compared against the Conserved Domain data base on the NCBI website (www. ncbi.nlm.nih.gov/structure/cdd/cdd.shtml). Interestingly, we found that five other phytoene desaturases that produce lycopene, including the ones from E. uredovora, Erwinia herbicola, A. aurantiacum, Flavobacterium sp. ATCC21588, and Erythrobacter longus, all have aspartate around position 68 of this protein from the multiple sequence alignment. The other phytoene desaturases that do not produce lycopene, such as Rhodobacter sp., have Val, Leu, or Met. We also noticed that the second beneficial mutation on m2, F166I, falls into one of the flavin adenine dinucleotide-binding sequence motifs (Fig.  3A), which belong to the class of glutathione reductase (20). However, the other two beneficial mutations, V68D and L148H, are in the regions that are not identified previously to be associated with any properties. We suspect that these two amino acid residues may be in the domains that are involved in the substrate binding. Further investigation of these mutations would elucidate the mechanisms that determine product specificity. FIG. 3. A, sequences of the R. sphaeroides ATCC17025 wild-type crtI and mutants that cause color change from yellow to red. The substituted amino acid residues are in boldface under the wild-type sequence with the ones mutated in the first generation of directed evolution underlined. There are two regions for the distributions of the positive mutations. One is near the C terminus; the other, which causes the higher production of lycopene by introducing more point mutations, is close to the N terminus (V68D, L148H, and F166I). Possible transmembrane domains are dash-overlined (13). The putative flavin adenine dinucleotide-binding sequence motifs are underlined, and their conserved residues are in boldface with dots underneath (18 -20). Asterisks represent the mutation to stop codon and the C-terminal hydrophobic amino acids. B, the schematic representation of the relative mutational effects caused by several amino acid changes. The heights of the upward arrows represent the extents of the positive mutations, the asterisks label the neutral ones, and a downward arrow represents the negative mutation. The lycopene production data were derived from the HPLC measurements in Fig. 2. WT, wild-type.