MINIREVIEW Deciphering enzymes Genetic selection as a probe of structure and mechanism Kenneth J. Woycechowsky and Donald Hilvert Laboratorium fu ¨ r Organische Chemie, Swiss Federal Institute of Technology, ETH-Ho ¨ nggerberg, Zu ¨ rich, Switzerland The efficient engineering of enzymes with novel activities remains an ongoing challenge. Towards this end, genetic selection techniques provide a method for finding rare solutions to catalytic problems that requires only a limited foreknowledge of structure–function relationships. We have used genetic selections to extensively probe the structure and mechanism of chorismate mutases. The insights gained from these investigations will aid future enzyme design efforts. Keywords: chorismate mutase; functional selection; protein engineering; protein folding. Introduction The incredible catalytic power of enzymes is well-documen- ted [1,2], but its source remains elusive. Enzymes catalyze a vast array of reactions with high specificity, under mild conditions [3]. These properties make enzymes potentially useful for organic synthesis [4,5]. Still, our current under- standing of protein structure–function relationships remains insufficient for the de novo design of enzymes with tailored catalytic activities [6]. Evolution provides, through multiple rounds of random mutagenesis and selection, a means to circumvent this problem. This process has produced the vast array of proteins found in nature. In the laboratory, directed evolution offers a promising strategy for the thorough study of protein structure–function relationships and for producing novel proteins with properties favorable for diverse applications, including catalysis [7,8]. Principles of genetic selection Natural evolution selects for the survival and reproduction of organisms. By introducing DNA libraries encoding potential enzymes into microorganisms such as bacteria, this process can be harnessed in the laboratory to concen- trate the selection process on an individual catalytic activity. A great advantage of genetic selection systems is the ability to perform parallel processing of huge libraries (rather than the serial analysis required by high-throughput screening). During selection, only sequences that encode functional enzymes are observed, which enables the efficient detection of rare solutions to a catalytic problem (with frequencies as low as one in 10 10 ). Furthermore, while the only know- ledge of protein structure or function required for this approach is the DNA sequence encoding the starting protein, structural and functional information can guide the choice of residues to be mutated or the content of the amino acid set to be sampled at these positions [9]. Such choices may focus the search of sequence space on areas with a higher frequency of success and thus increase the probability of their detection. In principle, any enzyme activity can be selected for in vivo, provided that catalysis of the desired reaction can be linked to cell growth. One general strategy for in vivo enzyme selection is the introduction of a metabolic requirement for the desired activity. A genetic selection system for chorismate mutase (CM) activity provides an example of this strategy (Fig. 1) [10]. CMs catalyze the Claisen rearrangement of chorismate to prephenate, which is the first committed step in the biosynthesis of phenylalanine and tyrosine [11]. In this system, a strain of Escherichia coli was engineered in which the genes encoding the bifunctional CM–prephanate dehy- dratase and CM–prephenate dehydrogenase protein com- plexes were replaced by genes encoding monofunctional versions of the dehydratase and the dehydrogenase. The growth of this strain on minimal media lacking phenyl- alanine and tyrosine requires an added source of CM activity. This source can be provided by transformation with a plasmid carrying a gene encoding the enzyme. This selection system has been used to reveal structural and mechanistic requirements for enzyme catalysis of this reaction. Selection for restructured enzymes Catalysis requires the fulfilment of exacting structural criteria; only properly folded proteins are active. Protein folding is dictated by amino acid sequence [12]. In an ensemble of proteins composed from the standard set of 20 amino acids with completely random sequences, however, the chance of encountering a significantly structured Correspondence to D. Hilvert, Laboratorium fu ¨ r Organische Chemie, Swiss Federal Institute of Technology, ETH-Ho ¨ nggerberg, CH-8093, Zu ¨ rich, Switzerland. Fax: + 41 1632 1486, Tel.: + 41 1632 3176, E-mail: donald.hilvert@org.chem.ethz.ch Abbreviations:BsCM,Bacillus subtilis CM; CM, chorismate mutase; EcCM, Escherichia coli CM; MjCM, Methanococcus jannaschii CM; MLE II, muconate lactonizing enzyme II; OSBS, ortho- succinylbenzoate; TIM, triose phosphate isomerase. (Received 5 January 2004, accepted 5 March 2004) Eur. J. Biochem. 271, 1630–1637 (2004) Ó FEBS 2004 doi:10.1111/j.1432-1033.2004.04073.x molecule is minute. For successful enzyme engineering, it would be extremely useful to bias protein libraries in favor of foldable sequences. Genetic selection experiments using CMs have helped to illuminate factors that influence protein structure and stability. One view of enzymes holds that the large size of these molecules is required for the precise positioning of a few active site residues around a substrate in three dimensions and enables their tremendous stabilization of transition states. Most amino acids in an enzyme thus serve both as spacers between, and as a scaffold for, the critical active site residues. This reasoning may account for the widely varying tolerances of protein structures to substitution at different sequence positions [13]. The uneven distribution of structural information in amino acid sequences presents both great opportunities and great challenges for enzyme engineering. Genetic selections of Methanococcus jannaschii CM (MjCM) [14] were used to assess the tolerance of a protein fold to secondary structures of varying sequence [15]. MjCM belongs to the AroQ class of CMs whose members adopt a homodimeric, a-helical bundle fold (Fig. 2) [16]. Each monomer consists of three a-helices and two turns. Three libraries were constructed and subjected to in vivo selection for CM activity (Fig. 3): first, the N-terminal helix alone was randomized, secondly, the two C-terminal helices were randomized simultaneously, and finally, positives from the first two libraries were randomly combined to give proteins whose sequences had been varied over all three Fig. 1. Selection system for chorismate mutase activity in Escherichia coli. An E. coli strain (KA12) was engineered in which the genes encoding the bifunctional enzymes chorismate mutase–prephenate dehydratase and chorismate mutase–prephenate dehydratase were deleted. Monofunctional versions of the dehydratase and the dehydrogenase are provided by plasmid pKIMP-UAUC. Random gene libraries are introduced into this strain and the ability of a cell harboring an individual library member to form a colony on minimal agar media lacking added phenylalanine and tyrosine reports on the chorismate mutase activity of the encoded protein [10]. Fig. 2. Structure of AroQ chorismate mutases. E. coli chorismate mutase, the prototypical AroQ chorismate mutase, is shown in a ribbon diagram representation [16]. AroQ chorismate mutases form homodimers of intimately entwined a-helices. The three helices of one subunit are indicated. A transition state analog inhibitor is bound in each active site and is represented in ball-and-stick form. Fig. 3. Design of the three binary-patterned libraries of Methanococcus jannaschii chorismate mutase. The residues within the secondary structural elements of M. jannaschii chorismate mutase were changed to a random distribution of only eight different amino acids: four polar and four hydrophobic [15]. An individual sequence position was ran- domised using the four amino acid set of similar polarity to the wild type residue. The libraries were constructed in two stages. First, helix 1 (Library 1) and helices 2 and 3 (Library 2) were randomized and introduced into the chorismate mutase selection system. Second, the successful clones from the initial libraries were crossed (Library 3) and subjected to selection. In Library 3, approximately 80% of the protein sequence was randomized. Binary patterned segments are depicted in red and blue; the segments of wild type sequence are colorless. Ó FEBS 2004 Deciphering enzymes (Eur. J. Biochem. 271) 1631 helices. The turn sequences and six active site residues were held constant in all cases. Libraries were designed using a restricted set of eight amino acids to incorporate a random distribution of four polar residues (Asp, Glu, Asn and Lys) and four nonpolar residues (Phe, Ile, Leu and Met) at positions of corresponding polarity in the a-helical regions of the wild type protein [15]. By using binary patterning of amino acids with high a-helical propensities, the libraries were designed to favor correct secondary structure forma- tion [17]. In addition, folded structures may be more common in proteins built from a small set of appropriately chosen amino acid building blocks [18]. Proper protein folding requires not only the formation of secondary structures, but also their packing together to form appropriate tertiary and quaternary interactions, particularly a hydrophobic core. For all three libraries, the complementation rate was about 0.01% [15]. This low frequency illustrates the challenge of packing elements of secondary structure to form proper tertiary and quaternary interactions. The importance of precise templates for proper protein folding is underlined by the low (one in 10 4 ) complementation rate of library 3. In this library, the sequences of helix 1 (itself) and helices 2 and 3 (together) were each functional in a context where C- and N-terminal halves of the protein, respectively, had the wild type sequence. The successful packing of these preselected segments against each other required an equally extensive search of sequence space as did the selection for proper folding of the initially randomized helices with the wild type template. A sequential strategy of randomizing helix 1 first and then randomizing helices 2 and 3 of an active variant from the first library (or vice versa), might prove more efficient than the convergent library approach outlined in Fig. 3. In this study of MjCM, about 80% of the protein sequence was subjected to randomization. Functional enzymes were found with less than 50% sequence identity to the wild type. While active catalysts were rare in these libraries, their presence demonstrates the ability of this protein fold to tolerate extensive substitutions. Harnessing this structural plasticity should be advantageous for enzyme redesign. Examination of the selected sequences revealed that some positions are more important than others and thus showed stronger preferences for one particular residue. For exam- ple, Ile14, Asn84 and Lys85 are all highly conserved in the active variants. This lack of permissiveness is perhaps unsurprising given that these residues probably contact the substrate and transition state during catalysis; active site sequences tend to be highly conserved. Additionally, Asp15 and Asp18 were also relatively nonpermissive. While they probably do not directly contact the substrate or transition state, these residues are thought to help orient catalytically essential residues that were held invariant in these libraries. Important second sphere interactions can be easily over- looked, but were readily apparent in these selections. The high enrichment for Phe at position 77 shows the import- ance of certain interactions in the hydrophobic core [19]. Phe77 may represent a Ôhot spotÕ for binding energy during protein folding [20], analogous to those found for receptor– ligand interactions [21]. If a smaller set of amino acids is structurally and functionally viable, then complete sampling becomes feasible for libraries in which a larger number of amino acids are varied simultaneously. Primordial protein catalysts may have had to manage with significantly fewer than the 20 amino acids commonly found in modern-day enzymes [22]. The active MjCM variants identified in this study lend credence to this evolutionary hypothesis. Furthermore, proteins built from a smaller set of building blocks should simplify the computational study and rational design of enzymes [23]. In addition to the packing of secondary structural elements, protein folding also requires the polypeptide backbone to turn back on itself. The requirements for the formation of an interhelical turn were examined by selecting active sequences from libraries of E. coli CM (EcCM) variants [24]. The solvent-exposed turn between helices 2 and 3 is composed of three amino acids: Ala65, His66 and His67 (Fig. 4). When these three residues were simulta- neously changed to a random distribution of the 20 standard amino acids, almost two-thirds of the resulting tripeptide sequences were functional. When Lys64, the solvent-exposed C-terminal residue of helix 2, was included in the randomization, the fraction of functional sequences dropped to 50%, but all four residues showed similar, high tolerances to substitution. Despite this high permissiveness, and in contrast to a previous study on the sequence requirements for a turn in cytochrome b 562 [25], a close examination of the sequence data showed a subtle, but strong, bias for hydrophilic amino acids in these positions. This bias may have gone undetected in cytochrome b 562 because that study, which found a similar low stringency for an interhelical turn sequence, relied on an assay for structure that was probably less sensitive than functional selection. The thermodynamic benefit resulting from minimizing the water accessible surface area of hydrophobic residues placed at these solvent-exposed positions may lead to aggregation or to local conformational disruptions. Fig. 4. The turn between helices 2 and 3 in E. coli chorismate mutase. Random mutagenesis of Lys64, Ala65, His66 and His67 followed by selection for chorismate mutase activity showed that these solvent exposed positions are highly permissive. In contrast, a similar experi- ment including Leu68, which is buried, instead of Lys64 produced much fewer complementing sequences. Apparently, tertiary contacts necessitate a hydrophobic amino acid at position 68, preferably one with a branched aliphatic side chain [24]. 1632 K. J. Woycechowsky and D. Hilvert (Eur. J. Biochem. 271) Ó FEBS 2004 A markedly different result was obtained when Leu68, a buried loop residue (Fig. 4), was randomized in tandem with the three turn residues [24]. This library had a complementation rate of about 6%, a 10-fold drop from the library in which the three turn residues were randomized alone. This drop was largely attributable to the absolute requirement for a hydrophobic residue at position 68. Furthermore, these successful clones exhibited a marked preference for branched, aliphatic residues at this position. This bias further highlights the functional importance of proper tertiary packing in the hydrophobic core of proteins. The secondary structural context may be relatively unim- portant, but the tertiary structural context and the pattern- ing of polar and nonpolar residues can greatly restrict allowable turn sequences. Engineering a new turn sequence into a protein structure presents a greater challenge than simply changing the sequence of a pre-existing turn. AroQ CMs have composite active sites, consisting of residues from helix 1 of one monomer and residues from helices 2 and 3 of the other monomer. It has been proposed that such domain-swapped dimers might have evolved from active, monomeric precur- sors [26]. By inserting a 180° turn into the middle of helix 1 of the thermostable MjCM, it was possible to form an active site with residues from a single polypeptide chain, and thus to perform domain swapping in reverse [27]. Like the (proposed) natural domain-swapping evolution- ary process, domain unswapping relied on selection for catalytic activity. In this case, two amino acids, Lys20 and Leu21, were duplicated and a random sequence of six residues was introduced between them. Introduction of this library into the CM selection system followed by screening of the positives using size-exclusion chromatography allowed the identification of a monomeric variant of MjCM that retained nearly 30% of the wild type activity (Fig. 5) [27]. Statistical analysis indicated that < 0.05% of the sequences produced well-behaved monomers, a surprisingly small fraction given the broad sequence tolerance of interhelical turn sequences noted above. The tertiary structural context may place imposing constraints on this turn sequence. Genetic selection has proved useful in generating other changes in CM quaternary structure. In a similar strategy to that described above, a randomized sequence of four to seven residues was inserted between Ala23 and Leu24 in the N-terminal helix of the mesostable EcCM (Fig. 5) [28]. Selection of these libraries showed that functional turn sequences were again rare, giving complementation rates of < 0.5% in all cases. While EcCM variants with four or seven amino acid insertions gave unstable monomers that were prone to precipitation, a five amino acid insertion surprisingly gave a stable, well-behaved hexamer [28]. The sequence of the insertion was nonpolar, suggesting that oligomerization through hydrophobic interactions may be an easy way to increase enzyme stability. This hexameric variant, however, suffered a 200-fold decrease in catalytic efficiency. In contrast, the unstable monomeric variants had near wild type activity. Over the limited area of sequence space covered by these libraries, there may be a trade-off between protein stability and catalytic activity. The AroQ CMs can retain function despite large changes in sequence and structure. The studies described above have helped both to estimate the tolerance of protein structural elements towards substitution and to identify structural constraints, such as packing interactions and polar/non- polar patterning, on functional sequences. Genetic selection of CM libraries has been an invaluable tool in the engineering of drastically restructured variants. Selection for altered active sites Genetic selection can also be extremely useful for studying structure–function relationships in enzymes. The simulta- neous in vivo analysis of variants randomized at one or several positions allows for a more thorough analysis of important functional residues than the traditional one-at-a- time approach of site-directed mutagenesis, protein purifi- cation and in vitro kinetic analysis. The development of such structure–function relationships in enzyme active sites is particularly useful for examining the important and often overlooked roles played by the multiple, subtle interactions between active-site residues. The rearrangement of chorismate to prephenate is arguably one of the simplest enzyme-catalyzed reactions. Like its uncatalyzed counterpart, this pericyclic reaction utilizes a concerted, but asynchronous, transition state, with C-O bond breakage preceding C-C bond formation [29–31]. Yet, the catalytic mechanism of CMs remains controversial. Specifically, a topic of current debate is whether transition- state stabilization by electrostatic interactions [32–34] or the preferential binding of reactive ground-state conformers [35–37] is of greater importance. Selection experiments provide persuasive evidence for the importance of electro- static interactions in catalysis by Bacillus subtilis CM (BsCM). BsCM is a member of the AroH class of CMs. This class adopts a trimeric, pseudo-a/b barrel fold [38] (Fig. 6). AroH and AroQ CMs share some common active site features. For example, in the crystal structures with an oxabicyclic transition state analog (TSA), both enzymes show multiple cationic groups (Arg and Lys) interacting with the carb- oxylates and the ether oxygen [16,38]. Additionally, both Fig. 5. Topological rearrangement of dimeric AroQ chorismate mutase into a monomer. Insertionofaflexibleloopintohelix1,whichspans the dimer, allows the N-terminal portion of the helix to bend back on itself and thus form a complete active site within a monomeric four- helix bundle. The insertion site is indicated by a horizontal red line. Ó FEBS 2004 Deciphering enzymes (Eur. J. Biochem. 271) 1633 possess a Glu residue that hydrogen bonds to the hydroxyl group of the TSA. Despite their different folds, both enzymes are likely to utilize similar catalytic mechanisms. The transition state for the chorismate mutase reaction is highly polarized [39]. In the structure of BsCM complexed with TSA [38], Arg90 seems poised to stabilize developing negative charge during the C-O bond cleavage (Fig. 7). An R90A variant exhibits a more than 10 6 -fold decrease in activity [40]. To further assess the role of this residue in catalysis, libraries were constructed in which both Arg90 alone and Arg90 and Cys88 together were randomized [10]. Selection revealed that, when the rest of the protein sequence is held constant, no other residue at position 90 is able to successfully replace Arg in vivo. In contrast, simultaneous substitution of positions 88 and 90 produces some alternat- ive, selectable solutions. In particular, a Lys was able to replace Arg90 if a residue smaller than Cys was present at position 88, even with the conservative change of Cys to Ser. Remarkably, it is also possible for a Lys at position 88 to substitute for Arg90, provided a Gly, Ser, Leu or Met is present at position 90. The selection of variants with rearranged active sites shows that, while rare, alternative active site structures capable of efficient catalysis within a given enzyme fold are experimentally accessible. Crystal structures of the R90K/C88S and R90S/C88K variants reveal the small but significant structural rearrangements within the active site caused by these mutations that probably allow the introduced ammonium group to interact with the developing negative charge on the ether oxygen in the transition state [41]. Apparently, subtle packing inter- actions are crucial for proper active site structure, and (similar to the requirements for proper protein folding discussed above) the local structural context imposes strict criteria for efficient function. During C-O bond breaking, a positive charge develops within the cyclohexadiene ring of chorismate. Although not as obvious as the interaction of Arg90 with the oxyanion in the transition state, the BsCM structure suggests that Glu78 could be important for carbocation stabilization in the transition state [38]. Glu78 is certainly important for catalysis; the E78A variant of BsCM is 10 4 -fold less active than wild type [40]. To examine its role in catalysis, Glu78 was randomized alone and together with Cys75 [42]. Unlike the strict requirement for Arg90, several other residues were able to directly replace Glu78, although the selection produced a bias for residues capable of hydrogen bonding. Interestingly, Asp was unable to substitute for Glu78, providing a further indication of the subtle interactions that dictate active site structure and function. When positions 75 and 78 were varied in tandem, however, an Asp at position 75 was able to substitute for Glu78, provided Ala, Ser, Met or Val was present at position 78. As functional solutions lacking an anion were found, the interaction of Glu78 with the transition state carbocation is not clear-cut. The crucial role of Glu78 may be to orient the substrate through a hydrogen bond with the hydroxyl group of chorismate [42a,42b]. Enzyme catalysis is a dynamic process. Yet, the import- ance of highly mobile, crystallographically unresolved residues is often overlooked. At the C-terminus of BsCM, residues 111–115 adopt a 3 10 helix and the following 11 residues have poorly defined structure (Fig. 6). This C-terminal tail lies close to the entrance of the substrate binding pocket and therefore may be important for catalysis. In the absence of structural information, however, it is difficult to postulate functional roles for individual residues. To help provide a functional definition for these residues, libraries of BsCM variants were constructed using a random protein truncation mutagenesis strategy and these libraries were subjected to selection [43]. Individually, none of the original 17 C-terminal residues are essential for complementation. Moreover, a truncated variant lacking the last 11 residues is still active in vivo, despite a 250-fold Fig. 6. Structure of Bacillus subtilis chorismate mutase. The mono- functional chorismate mutase from B. subtilis is a homotrimer and adopts a pseudo-a/b barrel fold [38]. A transition state analog, shown in a ball-and-stick representation, is bound in each of the active sites, which are located at the trimer interfaces. The location of the cystal- lographically unresolved residues at the C-termini are indicated by dashed lines. Fig. 7. Important interactions in the B. subtilis chorismate mutase active site. Electrostatic interactions are used to bind the transition state analogintheactivesiteofB. subtilis chorismate mutase. The guan- idinium group of Arg90 is poised to stabilize the developing oxyanion. Glu78 is positioned to hydrogen bond with the substrate hydroxyl group, and may also stabilize the developing carbocation in the cyclohexadiene ring. 1634 K. J. Woycechowsky and D. Hilvert (Eur. J. Biochem. 271) Ó FEBS 2004 decrease in catalytic efficiency relative to the wild type. The 3 10 helix (residues 111–115) is permissive but shows a modest preference for the wild type residues. In particular, Ala112 and Leu115 are the most highly conserved residues. These residues pack against the hydrophobic interior of BsCM and so are probably more important for structural stability than catalytic activity. The selected enzyme variants all showed little change in k cat , but significant increases in K m , which precludes their direct participation in catalysis. Instead, these residues probably contribute to catalytic efficiency via uniform binding of the substrate and trans- ition state. The versatility of functional selection So far, we have focused on genetic selection of CMs. Other selection systems have also proved useful for investigating enzyme structure and mechanism, and have been recently reviewed elsewhere [8]. Expanding the lessons learned with CMs, a few recent examples of selections with eight- stranded b/a-barrel [or triose phosphate isomerase (TIM) barrel] enzymes have examined the structural requirements for this fold and the active site differences that separate members of an enzyme superfamily. The TIM barrel is the most frequently encountered enzyme fold [44], and its natural catalytic versatility demon- strates its enormous potential for enzyme engineering. The robustness of triose phosphate isomerase (the prototypical TIM barrel enzyme) to substitutions was examined by combinatorial mutagenesis and selection for activity using a TIM-deficient strain of E. coli [45]. In this experiment, 182 residues outside of the TIM active site were mutated to one of seven amino acids (using binary polar/nonpolar patterning similar to that described above for MjCM) and introduced into the selection system. Analysis of complementing sequences shows that, while most individual sequence positions were tolerant to substitution by at least one member of the restricted amino acid set, only about one in 10 10 sequences randomized over the full length of the protein should be able to complement in vivo.Structuralelements such as the a/b interface, loops connecting secondary structures and a-helix caps were found to be permissive. In contrast, b-strand stop signals (particularly Gly), the central core of the barrel and a buried salt bridge were highly conserved. These results provide a more detailed view of how TIM barrel enzymes decouple catalytic activity and struc- tural stability [46] and should facilitate the de novo design of novel TIM barrel proteins [47]. Selections with another TIM barrel enzyme have been used to evaluate the plasticity of enzyme active sites. Variants of muconate lactonizing enzyme II (MLE II) with ortho-succinylbenzoate (OSBS) activity have been identified using random mutagenesis and genetic selection [48]. This selection system utilizes a mutant strain of E. coli that requires an added source of OSBS activity for anaerobic growth. OSBS and MLE II catalyze different overall reactions, but both catalytic mechanisms begin with the formation of an enolate intermediate (Fig. 8). Wild type MLE II, however, lacks detectable OSBS activity despite 24% sequence identity with E. coli OSBS and a similar TIM barrel fold. Three MLE II variants, each containing an E223G mutation were identified from the selection experi- ment. Indeed, this single mutation alone is sufficient to allow complementation of the mutant strain, despite a 10 3 -fold lower catalytic efficiency compared with E. coli OSBS. Interestingly, this variant retains residual activity for the MLE II reaction and may therefore resemble a catalytically promiscuous intermediate of a natural divergent evolution- ary process. Both MLE II and OSBS are members of the TIM barrel- containing enolase superfamily, and therefore both enzymes catalyze a common chemical step during catalysis of their respective reactions (Fig. 8) [49]. The gain of function seen for the MLE II variants, which can be considered as an extreme case of changing substrate specificity, still represents the first successful interconversion of catalytic activities within the well-characterized enolase superfamily. This result extends prior work that used random mutagenesis and selection to change substrate specificity without chan- ging the overall reaction [50]. A rationally designed variant of L -Ala- D / L -Glu epimerase (a third member of the enolase superfamily, Fig. 8), containing a mutation (D297G) analogous to that of the E223G MLE II, also exhibited measurable OSBS activity, albeit 100-fold lower than that of the selected MLE II variant [48]. The generality of this single mutation in conferring OSBS activity on enolases shows the potential utility of selection experiments in aiding rational design of enzymes. Apparently, the active sites of enolases may require only minor restructuring to accommodate the substrates of other superfamily members. As an alternative to metabolic requirements, in vivo selection systems can also be designed that couple enzyme Fig. 8. Reactions catalyzed by enolase superfamily members. Enolase superfamily members catalyze different overall reactions using a common mechanistic step: formation of an enolate intermediate. Some examples of the different reactions catalyzed by this superfamily are shown; (A) muconate lactonizing enzyme II (B) ortho-succinylbenzo- ate synthase and (C) L -Ala- D / L -Glu epimerase. The enolate inter- mediate for each reaction is enclosed in brackets. Ó FEBS 2004 Deciphering enzymes (Eur. J. Biochem. 271) 1635 function to antibiotic resistance. In one interesting system, a rationally designed DNA polymerase was used for the intro- duction in vivo of mutations in the TEM-1 b-lactamase gene at a desired frequency [51]. The cells containing this engineered DNA polymerase and mutated TEM-1 b- lactamase were grown in the presence of the antibiotic aztreonam. This system combines library generation and selection for enzyme activity (in this case, antibiotic detoxification) into one step. In this selection, three different mutations were identified that led to a 150-fold increase in aztreonam resistance. Two of these mutations matched those found in clinical isolates. Such rapid laboratory evolution may be useful to better anticipate the natural evolution of bacterial antibiotic resistance. Hydrolysis of aztreonam requires a change in the substrate specificity of TEM-1 b-lactamase. The chance of finding the three mutations that effected this change was estimated at one in 10 10 . Genetic selection was used to beat these odds and find a functional active site with altered structure. The power of selection is undeniable. Depending on the application, however, screening can also be useful for identifying enzymes with novel activities. High-throughput technology is advancing the size of libraries that can be thoroughly screened, providing an ever more appealing alternative to selection [52,53]. Conclusions and outlook Enzyme function is the product of multiple, subtle inter- actions within the protein structure. Functional selection of randomized libraries provides a general, sensitive and efficient probe of these interactions. The use of selection techniques with CMs has allowed us to explore the limits of structure and function for these enzymes. Although we have learned much from the studies described above, questions remain. CM, TIM and OSBS all catalyze reactions with fairly high background rates [2]. Would the complementation rates found in the studies with these enzymes be lower if the uncatalyzed reactions were energetically more demanding [54]? Are many different protein folds capable of catalyzing a given chemical reaction? Conversely, what is the potential for catalytic diversity within a given protein fold? The studies described above have shown that it is feasible to change the arrangement of functional groups within an active site. Can we harness divergent evolution to endow an existing enzyme scaffold with a completely new activity, changing both substrate specificity and chemical mechanism? Binary patterning [17] and restricted amino acid sets [18,22] can produce proteins capable of folding. What is the smallest set of amino acids that can still produce a catalytically competent protein? How can the knowledge gained from these studies be applied to the computational design of proteins with novel activities [55]? These questions are important for enzyme engineering. Genetic selection will probably play a key role in finding the answers. Acknowledgements This paper is dedicated to Professor Duilio Arigoni on the occasion of his 75th birthday. We thank Katherina Vamvaca for helpful comments on the manuscript. Financial support for the work on chorismate mutases was provided by the ETH-Zu ¨ rich and the Schweizer Nationalfonds. References 1. Knowles, J.R. & Albery, W.J. (1977) Perfection in enzyme catalysis – energetics of triose phosphate isomerase. Acc. Chem. Res. 10, 105–111. 2. Wolfenden, R. & Snider, M.J. (2001) The depth of chemical time and the power of enzymes as catalysts. Acc. Chem. Res. 34, 938–945. 3. Walsh, C. (2001) Enabling the chemistry of life. Nature 409, 226–231. 4. Koeller, K.M. & Wong, C H. (2001) Enzymes for chemical synthesis. Nature 409, 232–240. 5. Schmid, A., Dordick, J.S., Hauer, B., Kiener, A., Wubbolts, M. & Wittholt, B. (2001) Industrial biocatalysts for today and tomorrow. Nature 409, 258–268. 6. DeGrado, W.F., Summa, C.M., Pavone, V., Nastri, F. & Lom- bardi, A. (1999) De novo design and structural characterization of proteins and metalloproteins. Annu. Rev. Biochem. 68, 779–819. 7. Arnold, F.H. (1998) Design by directed evolution. Acc. Chem. Res. 31, 125–131. 8. Taylor, S.V., Kast, P. & Hilvert, D. (2001) Investigating and engineering enzymes by genetic selection. Angew. Chem. Int. Ed. 40, 3311–3335. 9. Kast, P. & Hilvert, D. (1997) 3D structural information as a guide to protein engineering using genetic selection. Curr. Opin. Sruct. Biol. 7, 470–479. 10. Kast, P., Asif-Ullah, M., Jiang, N. & Hilvert, D. (1996) Exploring theactivesiteofchorismatemutasebycombinatorialmutagen- esis and selection: the importance of electrostatic catalysis. Proc. NatlAcad.Sci.USA93, 5043–5048. 11. Haslam, E. (1993) Shikimic Acid: Metabolism and Metabolites. Wiley, New York. 12. Anfinsen, C.B. (1973) Principles that govern the folding of pro- tein chains. Science 181, 223–230. 13.Bowie,J.U.,Reidhaar-Olson,J.F.,Lim,W.A.&Sauer,R.T. (1990) Deciphering the message in protein sequences: tolerance to amino acid substitutions. Science 247, 1306–1310. 14. Macbeath, G., Kast, P. & Hilvert, D. (1998) A small, thermo- stable, and monofunctional chorismate mutase from the archeon Methanococcus janaschii. Biochemistry 37, 10062–10073. 15. Taylor, S.V., Walter, K.U., Kast, P. & Hilvert, D. (2001) Searching sequence space for protein catalysts. Proc.NatlAcad. Sci. USA 98, 10596–10601. 16. Lee, A.Y., Karplus, P.A., Ganem, B. & Clardy, J. (1995) Atomic structure of the buried catalytic pocket of Escherichia coli chor- ismate mutase. J. Am. Chem. Soc. 117, 3627–3628. 17. Kamtekar, S., Schiffer, J.M., Xiong, H., Babik, J.M. & Hecht, M.H. (1993) Protein design by binary patterning of polar and nonpolar amino acids. Science 262, 1680–1685. 18. Davidson, A.R. & Sauer, R.T. (1994) Folded proteins occur frequently in libraries of random amino acid sequences. Proc. NatlAcad.Sci.USA91, 2146–2150. 19. Lim, W.A. & Sauer, R.T. (1989) Alternative packing arrange- ments in the hydrophobic core of k repressor. Nature 339, 31–36. 20. Fersht, A.R., Matouschek, A. & Serrano, L. (1992) The folding of an enzyme. I. Theory of protein engineering anaylsis of sta- bility and pathway of protein folding. J. Mol. Biol. 224, 771–782. 21. Clackson, T. & Wells, J.A. (1995) A hot spot of binding energy in a hormone–receptor interface. Science 267, 383–386. 22. Riddle, D.S., Santiago, J.V., Bray-Hall, S.T., Doshi, N., Grant- charova, V.P., Yi, Q. & Baker, D. (1997) Functional rapidly folding proteins from simplified amino acid sequences. Nat. Struct. Biol. 4, 805–809. 1636 K. J. Woycechowsky and D. Hilvert (Eur. J. Biochem. 271) Ó FEBS 2004 23. Chan, H.S. (1999) Folding alphabets. Nat. Struct. Biol. 6,994– 996. 24. Macbeath, G., Kast, P. & Hilvert, D. (1998) Exploring sequence constraints on an interhelical turn using in vivo selection for catalytic activity. Protein Sci. 7, 325–335. 25. Brunet, A.P., Huang, E.S., Huffine, M.E., Loeb, J.E., Weltman, R.J. & Hecht, M.H. (1993) The role of turns in the structure of an a-helical protein. Nature 364, 355–358. 26. Bennett, M.J., Sclunegger, M.P. & Eisenberg, D. (1995) 3D domain swapping: a mechanism for oligomer assembly. Protein Sci. 4, 2455–2468. 27. Macbeath, G., Kast, P. & Hilvert, D. (1998) Redesigning enzyme topology by directed evolution. Science 279, 1958–1961. 28. Macbeath, G., Kast, P. & Hilvert, D. (1998) Probing enzyme quaternary structure by mutagenesis and selection. Protein Sci. 7, 1757–1767. 29. Addadi, L., Jaffe, E.K. & Knowles, J.R. (1983) Secondary tritium isotope effects as probes of the enzymic and nonenzymic conversion of chorismate to prephenate. Biochemistry 22, 4494– 4501. 30. Copley, S.D. & Knowles, J.R. (1985) The uncatalyzed Claisen rearrangement of chorismate to prephenate prefers a transition state of chairlike geometry. J. Am. Chem. Soc. 107, 5306–5308. 31. Guilford, W.J., Copley, S.D. & Knowles, J.R. (1987) On the mechanism of the chorismate mutase reaction. J. Am. Chem. Soc. 109, 5013–5019. 32. Lyne, P.D., Mulholland, A.J. & Richards, W.G. (1995) Insights into chorismate mutase catalysis from a combined QM/MM simulation of the enzyme reaction. J. Am. Chem. Soc. 117, 11345– 11350. 33. Kienho ¨ fer, A., Kast, P. & Hilvert, D. (2003) Selective stabiliza- tion of the chorismate mutase transition state by a postiviely charged hydrogen bond donor. J. Am. Chem. Soc. 125, 3206– 3207. 34. Strajbl, M., Shurki, A., Kato, M. & Warshel, A. (2003) Apparent NAC effect in chorismate mutase reflects electrostatic transition state stabilization. J. Am. Chem. Soc. 125, 10228–10237. 35. Khanjin, N.A., Snyder, J.P. & Menger, F.M. (1999) Mechanism of chorismate mutase: contribution of conformational restriction to catalysis in the Claisen rearrangement. J. Am. Chem. Soc. 121, 11831–11846. 36. Guo, H., Cui, Q., Lipscomb, W.N. & Karplus, P.A. (2003) Understanding the role of active-site residues in chorismate mutase catalysis from molecular dynamics simulations. Angew. Chem. Int. Ed. 42, 1508–1511. 37. Hur, S. & Bruice, T.C. (2003) Just a near attack conformer for catalysis (chorismate to prephenate rearrangements in water, antibody, enzymes, and their mutants). J. Am. Chem. Soc. 125, 10540–10542. 38. Chook, Y.M., Ke, H. & Lipscomb, W.N. (1993) Crystal struc- tures of the monofunctional chorismate mutase from Bacillus subtilis and its complex with a transition state analog. Proc. Natl Acad. Sci. USA 90, 8600–8603. 39. Gustin, D.J., Mattei, P., Kast, P., Wiest, O., Lee, L., Cleland, W.W. & Hilvert, D. (1999) Heavy atom isotope effects reveal a highly polarized transition state for chorismate mutase. J. Am. Chem. Soc. 121, 1756–1757. 40. Cload, S.T., Liu, D.R., Pastor, R.M. & Schultz, P.G. (1996) Mutagenesis study of active site residues in chorismate mutase from Bacillus subtilis. J. Am. Chem. Soc. 118, 1787–1788. 41. Kast, P., Grisostomi, C., Chen, I.A., Li, S., Krengel, U., Xue, Y. & Hilvert, D. (2000) A strategically positioned cation is crucial for efficient catalysis by chorismate mutase. J. Biol. Chem. 275, 36832–36838. 42. Kast, P., Hartgerink, M., Asif-Ullah, M. & Hilvert, D. (1996) Electrostatic catalysis of the Claisen rearrangement: probing the role of Glu78 in Bacillus subtilis chrorismate mutase by genetic selection. J. Am. Chem. Soc. 118, 3069–3070. 42a. Worthington, S.E., Roitberg, A.E. & Krauss, M. (2001) An MD/QM study of the chorismate mutase-catalyzed Claisen rearrangement reaction. J. Phys. Chem. B 105, 7087–7095. 42b. Ranaghan, K.E., Riddler, L., Szefczyk, B., Sokalski, W.A., Hermann, J.C. & Mulholland, A.J. (2004) Transition state sta- bilization and substrate strain in enzyme catalysis: ab initio QM/MM modelling of the chorismate mutase reaction. Org. Biomol. Chem. 2, 968–980. 43. Gamper, M., Hilvert, D. & Kast, P. (2000) Probing the role of the C-terminus of Bacillus subtilis chorismate mutase by a novel random protein termination strategy. Biochemistry 39, 14087– 14094. 44. Reardon, D. & Farber, G.K. (1995) The structure and evolution of a/b barrel proteins. FASEB J. 9, 497–503. 45. Silverman, J.A., Balakrishnan, R. & Harbury, P.B. (2001) Reverse engineering the (b/a) 8 barrel fold. Proc.NatlAcad.Sci. USA 98, 3092–3097. 46. Ho ¨ cker, B., Ju ¨ rgens,C.,Wilmanns,M.&Sterner,R.(2001) Stability, catalytic versatility and evolution of the (b/a) 8 -barrel fold. Curr. Opin. Biotechnol. 12, 376–381. 47. Offredi,F.,Dubail,F.,Kischel,P.,Sarinski,K.,Stern,A.S.,Van de Weerdt, C., Hoch, J.C., Prosperi, C., Franc¸ ois, J.M., Mayo, S.L. & Martial, J.A. (2003) De novo backbone and sequence design of an idealized a/b-barrel protein: evidence of stable tertiary structure. J. Mol. Biol. 325, 163–174. 48. Schmidt, D.M.Z., Mundorff, E.C., Dojka, M., Bermudez, E., Ness, J.E., Govindarajan, S., Babbitt, P.C., Minshull, J. & Gerlt, J.A. (2003) Evolutionary potential of (b/a) 8 -barrels: functional promiscuity produced by single substitutions in the enolase superfamily. Biochemistry 42, 8387–8393. 49. Babbitt, P.C. & Gerlt, J.A. (1997) Understanding enzyme superfamilies. J. Biol. Chem. 272, 30591–30594. 50. Ju ¨ rgens,C.,Strom,A.,Wegener,D.,Hettwer,S.,Wilmanns,M. & Sterner, R. (2000) Directed evolution of a (ba) 8 -barrel enzyme to catalyze related reactions in two different metabolic pathways. Proc.NatlAcad.Sci.USA97, 9925–9930. 51. Camps, M., Naukkarinen, J., Johnson, B.P. & Loeb, L.A. (2003) Targeted gene evolution in Escherichia coli usingahighly error-prone DNA polymerase I. Proc. Natl Acad. Sci. USA 100, 9727–9732. 52. Olsen, M., Iverson, B. & Georgiou, G. (2000) High-throughput screening of enzyme libraries. Curr. Opin. Biotechnol. 11, 331– 337. 53. Arnold, F.H. (2001) Combinatorial and computational chal- lenges for biocatalyst design. Nature 409, 253–257. 54. Taylor, E.A., Palmer, D.R.J. & Gerlt, J.A. (2001) The lesser Ôburden borneÕ by o-succinylbenzoate synthase: an ÔeasyÕ reaction involving a carboxylate carbon acid. J. Am. Chem. Soc. 123, 5824–5825. 55. Looger,L.L.,Dwyer,M.A.,Smith,J.J.&Hellinga,H.W.(2003) Computational design of receptor and sensor proteins with novel functions. Nature 423, 185–190. Ó FEBS 2004 Deciphering enzymes (Eur. J. Biochem. 271) 1637 . chorismate mutase–prephenate dehydratase were deleted. Monofunctional versions of the dehydratase and the dehydrogenase are provided by plasmid pKIMP-UAUC. Random. MINIREVIEW Deciphering enzymes Genetic selection as a probe of structure and mechanism Kenneth J. Woycechowsky and Donald Hilvert Laboratorium fu ¨ r Organische