- Email: [email protected]
Directed evolution: selecting today's biocatalysts
Biomolecular Engineering 22 (2005) 1–9 www.elsevier.com/locate/geneanabioeng
Review
Directed evolution: selecting today’s biocatalysts Linda G. Otten 1, Wim J. Quax * University of Groningen, University Centre for Pharmacy, Pharmaceutical Biology, Antonius Deusinglaan 1, 9713 AV Groningen, The Netherlands Received 1 October 2004; received in revised form 21 November 2004; accepted 22 November 2004
Abstract Directed evolution has become a full-grown tool in molecular biology nowadays. The methods that are involved in creating a mutant library are extensive and can be divided into several categories according to their basic ideas. Furthermore, both screening and selection can be used to target the enzyme towards the desired direction. Nowadays, this technique is broadly used in two major applications: (industrial) biocatalysis and research. In the first field enzymes are engineered in order to produce suitable biocatalysts with high catalytic activity and stability in an industrial environment. In the latter area methods are established to quickly engineer new enzymes for every possible catalytic step, thereby creating a universal biotechnological toolbox. Furthermore, directed evolution can be used to try to understand the natural evolutionary processes. This review deals with new mutagenesis and recombination strategies published recently. A full overview of new methods for creating more specialised mutant libraries is given. The importance of selection in directed evolution strategies is being exemplified by some current successes including the b-lactam acylases. # 2005 Elsevier B.V. All rights reserved. Keywords: Directed evolution; Selection; Mutagenesis; Recombination; New methods
1. Introduction Industrial and household catalysis becomes more and more dependent onenzymes.Thisis not surprisingsince enzymes are able to catalyse all kinds of chemical reactions. They can perform conversions in minutes or even seconds, which would take hundreds of years without their interference [1]. Furthermore, they catalyse reactions, which are difficult to perform by chemical methods, like the enantio- or regioseAbbreviations: HTS, high throughput screening; epPCR, error-prone polymerase chain reaction; NRR, non-homologous random recombination; SeSaM, sequence saturation mutagenesis; ADO, assembly of designed oligonucleotides; DHR, degenerate homoduplex gene family recombination; adipyl-7-ADCA, adipyl-7-aminodesacetoxycephalosporanic acid; glutaryl-7-ACA, glutaryl-7-aminocephalosporanic acid; MMP, matrix metalloprotease; AATase, aspartate aminotransferase; TATase, tyrosine aminotransferase; LBD, ligand binding domain; DBD, DNA binding domain; IVC, in vitro compartmentalisation; FACS, fluorescence activated cell sorting; MABS, methylation activity-based selection; FSS, functional salvage screen; GFP, green fluorescent protein * Corresponding author. Tel.: +31 50 3632558; fax: +31 50 3633000. E-mail address: [email protected] (W.J. Quax). 1 Present address: University of Cambridge, Department of Biochemistry, 80 Tennis Court Road, Cambridge CB2 1GA, UK. 1389-0344/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.bioeng.2005.02.002
lective hydrolysis or addition of chiral groups. Since all of these features are generally displayed at room temperature under mostly aqueous conditions the research towards the use of biocatalysts is mainly driven by the desire of using sustainable technologies for the production of chemicals (green routes) and complex active ingredients in a pharmaceutical and agrobiological context [2]. This ‘‘white biotechnology’’ remains a big challenge, since new biocatalytic processes have to compete economically with the well-established chemical processes that have been optimised for years. Although many complicated chemical reactions can be efficiently performed by biocatalysts, industrial conditions are usually different from those in nature with respect to substrate concentrations, sheering forces, temperature and organic solvents. Therefore, most enzymes found in soil and water may display the desired activity, but are generally not suited for industrial use [3]. Furthermore, for numerous industrial chemical processes an adequate enzyme cannot be readily found in nature, implying that there is a need for novel biocatalysts. Enzymes with the desired activity under industrial conditions can be obtained by optimising process conditions and by protein engineering. The latter deals with changing the genetic code for an enzyme in such a manner that a better biocatalyst is obtained. In this
2
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
review, we will focus on the strategy of evolution in the laboratory as a method for protein engineering.
2. Directed evolution Directed evolution has developed quickly to become a method of choice for protein engineers in order to create enzymes having desired properties for all kind of processes. Over the last decade this technique has become a daily part of the molecular toolbox of every biochemist. This is emphasised by the increasing number of publications about the subject. In the 1990s most publications were about new methods for mutagenesis and recombination. They described the evolution of a model enzyme to establish the new methods. In the past years, however, more research deals with enzymes that can be used in industry or research. In the first half of 2004, for example, at least 14 reviews about directed evolution were published, the majority of these dealing with the application of engineered enzymes. With so many excellent reviews already written, we will only describe articles published in 2003 and early 2004. In nature, evolution and creation of new functionalities is achieved by mutagenesis, recombination and survival of the
fittest. Directed evolution mimics this and is a process of iterative cycles of producing mutants and finding the mutant with the desired properties. The constantly continuing process of mutagenesis is the easiest one to copy in the lab. Mutations can be introduced at specific places using (site-) directed mutagenesis or throughout the gene by random mutagenesis. Several mutagenesis techniques have been developed in order to avoid codon bias. These were recently described by Neylon and Lutz and Patrick in two outstanding reviews [4,5]. More radical changes in evolution originate from recombination. Pim Stemmer was the first to recognise this and evolution was mimicked using a technique he called DNA-shuffling [6,7]. This method is based on the mixing and subsequent joining of different related small DNA fragments in order to form a complete new gene. In the process of shuffling, the recombination frequency is dependent on the degree of homology. A high level of recombination is important to get all possible combinations of mutations. Since recombination can be biased, several methods to overcome problems arising by the use of shuffling in the early years where tackled by novel strategies, all having their own advantages and disadvantages. These new recombination methods can be divided into six categories according to their basic idea (Table 1).
Table 1 Different recombination techniques with their main advantages and disadvantages Group
Technique
Members
Shuffling
Recombination of small fragments based on homology in the sequence between mutations that stem from all kinds of mutagenesis strategies or different family members Aims for high recombination, but difficult to separate close mutations
Shuffling [6,7] Family shuffling [45] RE cut shuffling [46] ssDNA shuffling [47] Mn2+ DNase cut [48] Endonuclease V cut [49] RPR [50] RETT [51] SCRATCHY [52]
Full length parent shuffling
Recombination of small fragments from different origin using one or more full length parent strands Higher recombination frequency, but more elaborate
StEP [53] RACHITT [54]
Single cross-over
Recombination of non-homologous genes by ligating front and back of two different genes, selection of new genes on size Recombination possible between low or non-homologous genes, but only one recombination point
(THIO)ITCHY [55,56] SHIPREC [57]
Recombination of structural, functional or less homologous parts of different family members More active enzymes in the resulting library, but only a few recombination points, which are hard to find
Exon shuffling [58] DOGS [59]
In vivo recombination
Recombination using the gap repair system of yeast or the recE/recT system of E. coli High yield, since no ligation necessary, but specialised vectors and multiple steps necessary
CLERY [63] ET-recombination [64]
Synthetic shuffling
Recombination of (un)known mutations in synthetic oligonucleotides Recombination of close mutations possible, but expensive and good selection necessary
Single step shuffling [65] DHR [16] Synthetic shuffling [66] ADO [15]
Domain swapping
SCRATCHY [52]
SISDC [60] YLBS [61] SCOPE [62]
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
Depending on the goal of the experiment one should decide which process is the best to create the optimal library. Although the production of an extensive and unbiased library is very important, the success of a directed evolution experiment highly depends on the method that is used to find the best mutant enzyme [8]. Since most directed evolution experiments generate a huge mutant library, a high throughput screening strategy or, more preferably, a selection has to be developed for different enzyme functions. These methods should be sufficiently sensitive and specific to identify positive mutants [9]. The advantage of screening is that the difference between substrate and product of an enzymatic reaction can be determined directly or indirectly in almost every case. However, relatively simple visual screens such as the formation of coloured or fluorescent products or halos around a colony on a plate are not always available. This implies the use of high throughput screening (HTS) in microtiter plates. HTS demands miniaturisation and automation of enzymatic assays. In the past decade a lot of research has been focused on finding better, cheaper, quicker and more accurate HTS assays [10]. This has made HTS feasible for many laboratories all over the world now, resulting in a lot of smart enzymatic screening methods. The disadvantage of screening lays in the fact that all individual mutants have to be tested for the desired enzymatic reaction, even those that might not be active or accurately folded. In a directed evolution approach using recombination these nonactive enzymes are typically 50–80% of the total library. After randomisation of a whole gene, which can result in unpredictable interesting new mutations outside the active site, the frequency of inactive mutants may be even higher. Therefore, the most preferred way of finding the best mutant is by selection. Selection mimics the natural survival of the fittest strategy and is the most efficient method to find the best mutant, since only mutants of interest will appear. Unfortunately, this approach is not possible for all enzymatic activities. Selection is based on the fact that mutants with the desired enzyme function provide an advantage to the host cell over bacteria bearing wild type enzymes. However, a lot of enzymatic activities are not essential to the bacterium. This can be overcome by coupling the activity of interest to an essential feature of the bacterium. This is very challenging and has already resulted in very elegant solutions from several labs. In this review, we will focus on new mutagenesis and selection techniques and the directed evolution methods using selection.
3
one problem concessions have to be made to other aspects of library construction. Bichet et al. tried to solve the problem of the losses of mutants through ligation [11]. They designed a ‘new’ errorprone PCR (epPCR) strategy, they call the Bringer Strategy, that does not involve a ligation step to bring the mutant gene back into the vector. The strategy is based on amplification of ring closed plasmids (e.g. QuikChange, Stratagene). The whole plasmid containing the gene of interest is amplified by PCR, but instead of using primers containing mutations, wild type primers are used and epPCR is performed on the whole plasmid. After discarding the wild type with DpnI, the resulting mutant plasmid is transformed. Although it is a real advantage not to be dependent on ligation efficiency of the PCR product into the vector, a lot of controls are necessary to be sure that only mutant plasmids are transformed and no other regions in the plasmids are damaged by mistakes in the PCR. In order to be able to recombine low homologous stretches of DNA, Bittker et al. developed the nonhomologous random recombination (NRR) strategy [12]. Recombination is accomplished by adding random blunt end DNA fragments together and select for binding of the newly prepared DNA stretch. Recently, they improved this technique into protein-NRR [13]. They start with fragments of one or more genes and control the size of the newly recombined DNA by adding a certain ratio of hairpin loops. DNA of the right size is obtained from an agarose gel and cloned into an expression vector. The expressed mutant proteins are selected for the desired enzyme activity (Fig. 1).
3. New library production methods Every technique to create mutant libraries has its own advantages and drawbacks. The objectives to develop a new method vary from creating more random (=unbiased) libraries to generating libraries from low homologous genes and producing a library more rapid. Usually, in the solving of
Fig. 1. Schematic representation of protein-NRR. Blunt-ended DNA fragments are ligated, size-selected and cloned into an expression vector. The size of the first pool of ligated fragments is controlled by the amount of blunt hairpins added to the ligation.
4
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
In this way they found active chorismate mutases with several insertions and deletions at certain defined points in the enzyme, but not at other points. Protein-NRR is therefore a useful tool for finding important functional parts of an enzyme in a broad and unbiased manner. It is, however, essential to have a good selection method, since many inactive proteins will be formed. The bias towards the original codon in random mutagenesis strategies was tackled by Wong et al. [14]. They designed a new mutagenesis strategy called sequence saturation mutagenesis (SeSaM). In this method, mutagenesis is performed by incorporating dNTPs nucleotides at random positions in the gene by PCR. Cutting at these positions reveals random fragments to which (different) universal bases are added. The whole gene is restored using PCR and the universal base is replaced by a normal one. Using dATPs and the universal base deoxyinosine, they found a library of well-distributed point mutations and some insertions or deletions. To get a totally unbiased library, this 2–3 days strategy should be repeated for all four dNTPs and several different universal bases, which opposes the claimed speed of the strategy. Using assembly of designed oligonucleotides (ADO), Zha et al. tried to get a high recombination frequency between different parents by producing oligonucleotides comprising all possible recombinations of alleles [15]. These designed oligo’s on alternating strands were assembled using homologous overlapping parts of the different parents. This resulted in chimaeras of lipA and lipB with a higher enantioselectivity towards the desired substrate. The difference between this method and the previously published degenerate homoduplex gene family recombination (DHR) [16] is the fact that the oligo’s are more random. Therefore, they may introduce new mutations that are not present in one of the parents.
4. Evolution of biocatalysts Biocatalysis becomes more important in industries. In the pharmaceutical industry, for example, it is important nowadays to produce enantiopure drugs to reduce the side effects and metabolic burden on the body. Enantioselectivity is, however, difficult to select for and libraries are usually screened as reviewed by Reetz [17,18]. Other industrially important enzymes may be selected by sophisticated methods. In our lab, a glutaryl acylase was engineered towards an adipyl acylase by randomising the gene using epPCR followed by a selection on a mimic substrate. The small set of selected mutants was subsequently screened on the real substrates (Fig. 2) [19]. This directed evolution step guided us to the important residues, which we subsequently randomised into all 19 others to find the best amino acid at each position [20,21]. Determining the catalytic parameters from each purified enzyme on adipyl-7-aminode-
Fig. 2. The mimic substrates adipyl-serine and adipyl-leucine, were used as selection substrate. Selected mutants were tested for their activity on the desired substrate adipyl-7-ADCA, which is quite similar.
sacetoxycephalosporanic acid (adipyl-7-ADCA) revealed a single mutant having a 15 times improved catalytic efficiency resulting from an improved Km and a kcat similar to the kcat of wild type enzyme on glutaryl-7-aminocephalosporanic acid (glutaryl-7-ACA). This enzyme is almost good enough to be used in an industrial process to produce 7-ADCA from adipyl-7-ADCA. The production of a library and developing of a selection procedure that renders only wanted mutants can be difficult as is obvious from the research done by the group of Buchholz [22]. They used directed evolution techniques on retroviruses. Retroviruses can be used in gene therapy to bring beneficial DNA into cells, but also to bring suicide genes into tumour cells in order to kill them. One of the strategies to target retroviruses towards the tumour cell is by inserting a protein sequence into the retroviral envelope protein. Activation, and thereby spreading of the retrovirus, will only take place if this sequence is cleaved by a tumour specific protease. It is, however, not known which protein sequence will be recognised by proteases from the tumour cells. Therefore, a directed evolution experiment was designed in order to change the linker peptide into a better target for a matrix metalloprotease (MMP), which is very common in tumour cells. The first generation of this library was already described in 1998 [23], but resulted only in protein sequences recognised by omnipresent serine proteases. In the present research a new strategy was used to create a library without target peptides for serine proteases. Furthermore, another envelop protein was used to be able
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
5
Fig. 3. Different methods to use a selection system to study protein–protein interactions. In both systems a reporter activity is restored upon binding of two proteins. This activity can be used to give the host a growth advantage or to assay the amount of reporter produced. (A) In the yeast two-hybrid method the gene of interest is fused to a DNA binding domain and the mutant library is fused to an activation domain, which activates the reporter gene. In the yeast three-hybrid system, a ligand promotes the binding of the target and binding protein. (B) In the split-protein system, the interaction between two proteins brings the two halves of the reporter protein together.
to select retroviruses in tumour cells from human origin. Two consecutive cycles of diversification and in vivo selection resulted in a more efficient cutting of the target site by the MMP produced by the cancer cell-line. This selected MMP-activatable retrovirus spread more effectively around the tumour. Since the method is reasonably rapid, it can be used to make an individual highly effective therapy possible. Van der Veen et al. developed a selection method to use as a measure of amylosucrase activity [24]. The procedure is based on a zero background expression cloning strategy in order to select against plasmids containing no amylosucrose gene. Transformants were scraped from the first plates containing sucrose and selection was performed on agar plates containing sucrose as the only carbon source, thereby selecting for active enzymes. The selected enzymes are subsequently screened in an automated screening that was designed to discriminate between different actions of the enzyme. A pilot experiment was performed using random mutagenesis and a round of shuffling to lose the high percentage of inactive clones. Two libraries were selected using the new strategy and selected mutants were screened for amylosucrase activity. From the 1250 transformants plated, only 60 were active, stressing the importance of the (pre)selection. Two improved mutants were found having mutations, which were not considered changing from the crystal structure.
5. Understanding natural evolution and biological functions Although directed evolution mimics its natural variant, the actual evolution process is still not fully understood. A better understanding of natural evolution will allow mimicking the process better. To do so, different labs are trying to evolve one family member into another or different functionalities from a so-called ancient parent. In Kirsch’s lab it is tried to evolve aspartate aminotransferase (AATase) into tyrosine aminotransferase (TATase) to show that they are evolutionary related. Both rational design and directed evolution were performed to reach this goal. The best mutant from the directed evolution approach gave a complementation of tyrosine auxotrophy, but the rationally designed HEX mutant did not, although it had some TATase activity in vitro [25]. Therefore, HEX was shuffled (with itself) and selected for complementation [26]. Two mutants showed to be active in vivo and these appeared to have a lower Km. Furthermore, they displayed less product inhibition. The best new mutation in this shuffled HEX library, A293D, was introduced in the best mutant from the directed evolution library, which resulted in a TATase activity similar to the wild type AATase activity [27]. The three new mutants were crystallised and TATase activity of the mutants could very well be explained by the crystal structure, although
6
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
this residue was not considered changing in the first rational library. Christ and Winter try to find proteins that are related to Rop, an antisense RNA binding protein [28]. To do so, they randomised five important residues that were found in a previous Ala-scanning experiment. Selection was done using a very elegant liquid growing system in which the original Rop gene of the plasmid was deleted. This triggers the bacterium to make more plasmid and thereby grow slower. Upon restoring the (mutant) gene into the bacterium, it grows faster and outgrows others in the culture medium. Three rounds of selection were performed. A reporter gene on the plasmid can be used as a second screen. Eight active variants were found after three rounds of selection and appeared to have a low kd as determined by surface plasmon resonance. The sequence of the selected mutants helped to define a search tool to find distantly related proteins. The writers, however, fairly admit that these are related in function/structure and not necessarily in an evolutionary sense. Buskirk et al. try to unravel the complex biological function of transcription activation [29]. Their goal was to make active transcriptional elements from random RNA sequences with comparable potency as the natural ones. The random RNA sequences were synthesised as a variable region of DNA coding for a larger RNA structure with a stable secondary structure. Another part of this RNA consists of two MS2 hairpins, which bind to one of the partners of the yeast three-hybrid system (Fig. 3A). Active RNA sequences can trigger histidine production, which makes them selectable in a histidine–auxotrophic host on selective plates containing no histidine. The activation was quantified by using a second yeast three-hybrid system with b-galactosidase as reporter enzyme. Two rounds of evolution and selection were performed in which surprisingly high amounts of survivors were found (0.2%). Only one RNA sequence appeared to be really active when looking into the b-galactosidase activity. This RNA activator was only two times lower than a normal protein activator. Peimbert and Segovia used a two step approach to change a transpeptidase into a b-lactamase to show their evolutionary relationship [30]. They started with a rational site directed mutagenesis library in which important residues (according to the crystal structure) were randomised. This library was subjected to random mutagenesis by epPCR and recloned into an expression plasmid. Transformants were enriched in liquid media containing cefotaxime, plasmids were retransformed and plated onto LB-plates containing different concentrations of cefotaxime. This resulted in a mutant with a 105-fold increase in cefotaxime deacylation. Performing several epPCR rounds and a shuffling step with the wild type gene or saturation mutagenesis of important residues found in the first library did not result in better mutants. This suggests that using this selection system no better mutant can be found and another, maybe
more stringent, method should be used for subsequent rounds.
6. Evolution of a molecular toolbox Every directed evolution experiment is an individual search for the best enzyme for the desired reaction. In order to be able to attack each question as quick and direct as possible it is most convenient to have some general tools at hand. One of the possibilities is to maintain a huge variety of similar enzymes in the freezer and screen them for the desired activity when necessary. Another complementary strategy is to have some general directed evolution methods ‘on the shelf’ with readily adjustable screening or selection methods. Azizi et al. described a very elegant new selection strategy in yeast based on the yeast two-hybrid system in order to find new ligands or new receptors for existing ligands (Fig. 3A) [31]. In this system the nuclear receptor’s ligand binding domain (LBD) is fused to the Gal4 DNA binding domain (Gal4 DBD). The fusion protein binds to the Gal4 response element controlling the expression of the His3 gene. Only in the presence of a good ligand the fusion protein will enable the His3 gene to be expressed, which makes yeast cells able to grow on media without histidine. The benefit of this system is that the library of ligands is not co-expressed with the strain, but added to the growth medium. This strategy can be very well combined with random mutagenesis of the LBD resulting in a total directed evolution strategy, even with a simultaneous selection on different ligands. Another selection strategy to investigate protein–protein interaction is the use of split-protein sensors (Fig. 3B). Creating split-proteins used to be a process of trial and error, but Tafelmeyer et al. developed a method to make a new split-protein using directed evolution [32]. They fused the N and C terminus of a Trp gene together and generated a library with splits in every possible position. This library was cloned into an expression vector containing two polypeptides that associate into an antiparallel coiled coil in order to bring the two halves together. The library was then selected for active enzyme, which resulted in several candidates. One of the selected split-Trp pairs was successfully applied to monitor protein–protein interactions both at the membrane as well as in the cytosol of yeast. This powerful selection complements the repertoire of the currently used split-protein sensors and provides a new tool for high-throughput interaction screening. Gimble et al. used the bacterial two-hybrid selection system (Fig. 3A) [33] in E. coli to change the substrate specificity of the homing endonuclease PI-SceI protein [34]. Homing endonucleases have been used in genomic engineering and variants recognising other DNA stretches will be useful. From the crystal structure, they determined ˚ of the binding basepairs +18 which amino acids within 7 A
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
and +19 had direct interactions or possible hydrogen bonding with these residues. The five selected residues were randomised in a PCR using NNK codons at the target positions. The bacterial two-hybrid system was used to select for mutants that recognise a new target DNA. Some mutants recognising new target DNA with little conservation of the wild type protein residues were found that surprisingly still recognise the wild type sequence, pointing to the plasticity of the recognition site of these enzymes. Griffiths and Tawfik improved their in vitro compartmentalisation (IVC) selection system that uses oil-in-water emulsions [35]. They made a link between phenotype and genotype of the target enzyme by attaching its DNA on a streptavidin coated bead. The enzyme is produced in the compartment and attached to the same bead as the gene, thereby making a genotype/phenotype linkage. The compartments are broken and new ones are formed in which substrate bound to biotin is added. The substrate can be converted by the bound enzyme and will be caught by the bead as well. Beads can then be selected using antibodies against the product with affinity tags or fluorescently labelled and sorted by fluorescence activated cell sorting (FACS). This completely in vitro process selects for all enzymatic features simultaneously (substrate recognition, product formation, rate acceleration and turnover) and single enzyme molecules can be detected. The strength of the method was shown by improving the catalytic activity of a phosphotriesterase. The most enhanced mutant showed a 63fold improved kcat, compared to the already very efficient wild type enzyme [36]. Zhu et al. changed the substrate specificity from restriction enzyme BsoBI [37]. They started with rational design, by changing the amino acid in close contact with the substrate as seen in the crystal structure. One of the mutants with changed substrate specificity was then further mutagenised using epPCR, in order to increase enzymatic activity. Selection was performed in vivo by looking at the dark blue colour of colonies from an indicator strain on a plate. This resulted in more active variants with changed substrate specificity. Surprisingly, most mutations were found outside the protein–DNA interface region. Rimseliene et al. also described a new selection system to produce restriction enzymes with a new specificity, which they call MABS (methylation activity-based selection) [38]. The procedure can only be used on bifunctional type II endonucleases and includes three steps. The bifunctional restriction enzyme is transformed into a monofunctional DNA-methyltransferase by cleavage centre disruption. Then mutants are selected which show altered DNA modification specificity based on their ability to protect predetermined DNA targets. In the final step the cleavage centre’s wild type structure is reconstituted. New enzymes made by this method can be used as new molecular tools in cloning and restriction analysis. Another publication in which a methyltransferase was changed by directed evolution is the paper from Cohen et al.
7
[39]. In order to change substrate specificity of HaeIII methyltransferase, they used a two-step mutagenesis strategy followed by selection using in vitro compartmentalisation. The initial mutagenesis of DNA contacting residues was followed by randomisation of the loop that lies behind these residues. The selection system was based on the addition of the desired DNA stretch to the gene resulting in a covalent linkage between the gene and the methylated DNA, in case of a good mutant. This resulted in the remarkable finding of a mutant with a catalytic activity towards the new substrate that was even better than wild type activity towards the original substrate. Recombination is one of the cornerstones of directed evolution. Voziyanov et al. are using the Flp recombinase to recombine DNA fragments. The drawback of this method is the fact that the enzyme only recognises a certain stretch of DNA, which has to be inserted into the gene(s) of interest to get recombination. To overcome this problem they used mutagenesis and shuffling to rapidly evolve different mutant recombinases that recognised other stretches of DNA [40]. Combining two different mutants in one experiment can lead to a recombination system that utilises bi-specific hybrid target sites. A hybrid site is assembled from two half-sites, each harbouring a distinct binding specificity. Directed evolution was used to find mutations that improve the association of the two halves, thereby expanding the use of these enzymes for recombination [41]. In directed evolution experiments it is usually attempted to search the largest sequence space possible. Starting from a known protein limits this sequence space. The group of Kim HS has described a new method to expand the sequence space, which they call functional salvage screen (FSS). First they introduced deletions into green fluorescent protein (GFP) to inactivate them. Starting from these defective GFPs they added random sequences at the mutated sites and screened for proteins that regained fluorescence. This resulted in a mutant with low fluorescence intensity and stability [42]. Using directed evolution they got an improved variant after three rounds of evolution with four mutations and a ‘totally’ different stretch of amino acids than the wild type enzyme at that position. The fluorescence and stability from this mutant is, however, similar to the original enzyme [43]. Fa et al. tried to expand the substrate repertoire of a DNA polymerase by directed evolution [44]. Their goal was to allow enzymatic synthesis of unnatural polymers in vivo or in vitro. Therefore, they developed an activity-based selection method, which isolates polymerase mutants with the desired property from libraries of the enzyme displayed on phage. The method is based on the co-display (on a single phage particle) of a member of a mutant enzyme library and an acidic peptide that is used to attach a DNA primertemplate substrate. A mutant polymerase that is able to incorporate methylated and biotinylated bases into the growing strain will be selected by streptavidin-coated beads. Two libraries were constructed. In each one, six residues
8
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
Fig. 4. General directed evolution scheme.
chosen by looking at the crystal structure were randomised. After four rounds of selection one mutant enzyme incorporated methylated dATP at a similar rate as wild type incorporates non-methylated dATP.
7. Conclusions Directed evolution becomes more and more an integrated research field. Both rational and random strategies are alternately used to create the optimal library for finding the desired enzyme. New in vitro and in vivo selection strategies are developed to select for all kinds of new activities. All these components are used in very diverse formats to reach the same goal: creating new enzymes with desired properties (Fig. 4). These properties can be dictated by industrial or scientific needs and directed evolution is even used as a tool to better understand the principles of natural evolution. This will encourage the development of even more directed evolution strategies to find the optimal library size and the best selection or screening method for each new enzyme.
Acknowledgements This review was written on the basis of discussions with many colleagues at the Department of Pharmaceutical Biology of the University of Groningen, with financial support from STW, a part of the Dutch Organisation for Science (NWO). W.Q. further acknowledges support from the EU commission in project QLTR-2001-00519.
References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
Dalby PA. Curr Opin Struct Biol 2003;13:500–5. Schoemaker HE, Mink D, Wubbolts MG. Science 2003;299:1694–7. Marrs B, Delagrave S, Murphy D. Curr Opin Microbiol 1999;2:241–5. Neylon C. Nucl Acids Res 2004;32:1448–59. Lutz S, Patrick WM. Curr Opin Biotechnol 2004;15:291–7. Stemmer WP. Nature 1994;370:389–91. Stemmer WP. Proc Natl Acad Sci USA 1994;91:10747–51. You L, Arnold FH. Protein Eng 1996;9:77–83. Zhao H, Arnold FH. Curr Opin Struct Biol 1997;7:480–5. Delagrave S, Murphy DJ. Assay Drug Develop Technol 2003;1:187– 98. [11] Bichet A, Bureik M, Lenz N, Bernhardt R. Appl Biochem Biotechnol 2004;117:115–22.
[12] Bittker JA, Le BV, Liu DR. Nat Biotechnol 2002;20:1024–9. [13] Bittker JA, Le BV, Liu JM, Liu DR. Proc Natl Acad Sci USA 2004;101:7011–6. [14] Wong TS, Tee KL, Hauer B, Schwaneberg U. Nucl Acids Res 2004;32:e26. [15] Zha D, Eipper A, Reetz MT. Chembiochem 2003;4:34–9. [16] Coco WM, Encell LP, Levinson WE, Crist MJ, Loomis AK, Licato LL, et al. Nat Biotechnol 2002;20:1246–50. [17] Reetz MT. Tetrahedron 2002;58:6595–602. [18] Reetz MT. Proc Natl Acad Sci USA 2004;101:5716–22. [19] Otten LG, Sio CF, Vrielink J, Cool RH, Quax WJ. J Biol Chem 2002;277:42121–7. [20] Sio CF, Otten LG, Cool RH, Quax WJ. Biochem Biophys Res Comm 2003;312:755–60. [21] Otten LG, Sio CF, van der Sloot AM, Cool RH, Quax WJ. Chem Bio Chem 2004;5:820–5. [22] Schneider RM, Medvedovska Y, Hartl I, Voelker B, Chadwick MP, Russell SJ, et al. Gene Ther 2003;10:1370–80. [23] Buchholz CJ, Peng KW, Morling FJ, Zhang J, Cosset FL, Russell SJ. Nat Biotechnol 1998;16:951–4. [24] van der Veen BA, Potocki-Veronese G, Albenne C, Joucla G, Monsan P, Remaud-Simeon M. Febs Lett 2004;560:91–7. [25] Rothman SC, Kirsch JF. J Mol Biol 2003;327:593–608. [26] Rothman SC, Voorhies M, Kirsch JF. Protein Sci 2004;13:763–72. [27] Chow MA, McElroy KE, Corbett KD, Berger JM, Kirsch JF. Biochemistry 2004;43:12780–7. [28] Christ D, Winter G. Proc Natl Acad Sci USA 2003;100:13202–6. [29] Buskirk AR, Kehayova PD, Landrigan A, Liu DR. Chem Biol 2003;10:533–40. [30] Peimbert M, Segovia L. Protein Eng 2003;16:27–35. [31] Azizi B, Chang EI, Doyle DF. Biochem Biophys Res Comm 2003;306:774–80. [32] Tafelmeyer P, Johnsson N, Johnsson K. Chem Biol 2004;11:681–9. [33] Joung JK, Ramm EI, Pabo CO. Proc Natl Acad Sci USA 2000;97:7382–7. [34] Gimble FS, Moure CM, Posey KL. J Mol Biol 2003;334:993–1008. [35] Tawfik DS, Griffiths AD. Nat Biotechnol 1998;16:652–6. [36] Griffiths AD, Tawfik DS. EMBO J 2003;22:24–35. [37] Zhu ZY, Zhou J, Friedman AM, Xu SY. J Mol Biol 2003;330:359– 72. [38] Rimseliene R, Maneliene Z, Lubys A, Janulaitis A. J Mol Biol 2003;327:383–91. [39] Cohen HM, Tawfik DS, Griffiths AD. Prot Eng Des Sel 2004;17:3–11. [40] Voziyanov Y, Konieczka JH, Stewart AF, Jayaram M. J Mol Biol 2003;326:65–76. [41] Konieczka JH, Paek A, Jayaram M, Voziyanov Y. J Mol Biol 2004;339:365–78. [42] Kim GJ, Cheon YH, Park MS, Park HS, Kim HS. Protein Eng 2001;14:647–54. [43] Nam SH, Oh KH, Kim GJ, Kim HS. Protein Eng 2003;16:1099–105. [44] Fa M, Radeghieri A, Henry AA, Romesberg FE. J Am Chem Soc 2004;126:1748–54. [45] Crameri A, Raillard SA, Bermudez E, Stemmer WP. Nature 1998;391:288–91. [46] Kikuchi M, Ohnishi K, Harayama S. Gene 1999;236:159–67.
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9 [47] [48] [49] [50] [51] [52] [53] [54] [55] [56]
Kikuchi M, Ohnishi K, Harayama S. Gene 2000;243:133–7. Lorimer IA, Pastan I. Nucl Acids Res 1995;23:3067–8. Miyazaki K. Nucl Acids Res 2002;30:e139. Shao Z, Zhao H, Giver L, Arnold FH. Nucl Acids Res 1998;26:681–3. Lee SH, Ryu EJ, Kang MJ, Wang ES, Piao Z, Choi YJ, et al. J Mol Catal B-Enzym 2003;26:119–29. Lutz S, Ostermeier M, Moore GL, Maranas CD, Benkovic SJ. Proc Natl Acad Sci USA 2001;98:11248–53. Zhao H, Giver L, Shao Z, Affholter JA, Arnold FH. Nat Biotechnol 1998;16:258–61. Coco WM, Levinson WE, Crist MJ, Hektor HJ, Darzins A, Pienkos PT, et al. Nat Biotechnol 2001;19:354–9. Ostermeier M, Shim JH, Benkovic SJ. Nat Biotechnol 1999;17: 1205–9. Lutz S, Ostermeier M, Benkovic SJ. Nucl Acids Res 2001;29:16.
9
[57] Sieber V, Martinez CA, Arnold FH. Nat Biotechnol 2001;19:456– 60. [58] Kolkman JA, Stemmer WP. Nat Biotechnol 2001;19:423–8. [59] Gibbs MD, Nevalainen KMH, Bergquist PL. Gene 2001;271:13–20. [60] Hiraga K, Arnold FH. J Mol Biol 2003;330:287–96. [61] Kitamura K, Kinoshita Y, Narasaki S, Nemoto N, Husimi Y, Nishigaki K. Protein Eng 2002;15:843–53. [62] O’Maille PE, Bakhtina M, Tsai MD. J Mol Biol 2002;321:677–91. [63] Abecassis V, Pompon D, Truan G. Nucl Acids Res 2000;28:88. [64] Strausberg SL, Alexander PA, Gallagher DT, Gilliland GL, Barnett BL, Bryan PN. Biotechnology (NY) 1995;13:669–73. [65] Stemmer WP, Crameri A, Ha KD, Brennan TM, Heyneker HL. Gene 1995;164:49–53. [66] Ness JE, Kim S, Gottman A, Pak R, Krebber A, Borchert TV, et al. Nat Biotechnol 2002;20:1251–5.
Review
Directed evolution: selecting today’s biocatalysts Linda G. Otten 1, Wim J. Quax * University of Groningen, University Centre for Pharmacy, Pharmaceutical Biology, Antonius Deusinglaan 1, 9713 AV Groningen, The Netherlands Received 1 October 2004; received in revised form 21 November 2004; accepted 22 November 2004
Abstract Directed evolution has become a full-grown tool in molecular biology nowadays. The methods that are involved in creating a mutant library are extensive and can be divided into several categories according to their basic ideas. Furthermore, both screening and selection can be used to target the enzyme towards the desired direction. Nowadays, this technique is broadly used in two major applications: (industrial) biocatalysis and research. In the first field enzymes are engineered in order to produce suitable biocatalysts with high catalytic activity and stability in an industrial environment. In the latter area methods are established to quickly engineer new enzymes for every possible catalytic step, thereby creating a universal biotechnological toolbox. Furthermore, directed evolution can be used to try to understand the natural evolutionary processes. This review deals with new mutagenesis and recombination strategies published recently. A full overview of new methods for creating more specialised mutant libraries is given. The importance of selection in directed evolution strategies is being exemplified by some current successes including the b-lactam acylases. # 2005 Elsevier B.V. All rights reserved. Keywords: Directed evolution; Selection; Mutagenesis; Recombination; New methods
1. Introduction Industrial and household catalysis becomes more and more dependent onenzymes.Thisis not surprisingsince enzymes are able to catalyse all kinds of chemical reactions. They can perform conversions in minutes or even seconds, which would take hundreds of years without their interference [1]. Furthermore, they catalyse reactions, which are difficult to perform by chemical methods, like the enantio- or regioseAbbreviations: HTS, high throughput screening; epPCR, error-prone polymerase chain reaction; NRR, non-homologous random recombination; SeSaM, sequence saturation mutagenesis; ADO, assembly of designed oligonucleotides; DHR, degenerate homoduplex gene family recombination; adipyl-7-ADCA, adipyl-7-aminodesacetoxycephalosporanic acid; glutaryl-7-ACA, glutaryl-7-aminocephalosporanic acid; MMP, matrix metalloprotease; AATase, aspartate aminotransferase; TATase, tyrosine aminotransferase; LBD, ligand binding domain; DBD, DNA binding domain; IVC, in vitro compartmentalisation; FACS, fluorescence activated cell sorting; MABS, methylation activity-based selection; FSS, functional salvage screen; GFP, green fluorescent protein * Corresponding author. Tel.: +31 50 3632558; fax: +31 50 3633000. E-mail address: [email protected] (W.J. Quax). 1 Present address: University of Cambridge, Department of Biochemistry, 80 Tennis Court Road, Cambridge CB2 1GA, UK. 1389-0344/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.bioeng.2005.02.002
lective hydrolysis or addition of chiral groups. Since all of these features are generally displayed at room temperature under mostly aqueous conditions the research towards the use of biocatalysts is mainly driven by the desire of using sustainable technologies for the production of chemicals (green routes) and complex active ingredients in a pharmaceutical and agrobiological context [2]. This ‘‘white biotechnology’’ remains a big challenge, since new biocatalytic processes have to compete economically with the well-established chemical processes that have been optimised for years. Although many complicated chemical reactions can be efficiently performed by biocatalysts, industrial conditions are usually different from those in nature with respect to substrate concentrations, sheering forces, temperature and organic solvents. Therefore, most enzymes found in soil and water may display the desired activity, but are generally not suited for industrial use [3]. Furthermore, for numerous industrial chemical processes an adequate enzyme cannot be readily found in nature, implying that there is a need for novel biocatalysts. Enzymes with the desired activity under industrial conditions can be obtained by optimising process conditions and by protein engineering. The latter deals with changing the genetic code for an enzyme in such a manner that a better biocatalyst is obtained. In this
2
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
review, we will focus on the strategy of evolution in the laboratory as a method for protein engineering.
2. Directed evolution Directed evolution has developed quickly to become a method of choice for protein engineers in order to create enzymes having desired properties for all kind of processes. Over the last decade this technique has become a daily part of the molecular toolbox of every biochemist. This is emphasised by the increasing number of publications about the subject. In the 1990s most publications were about new methods for mutagenesis and recombination. They described the evolution of a model enzyme to establish the new methods. In the past years, however, more research deals with enzymes that can be used in industry or research. In the first half of 2004, for example, at least 14 reviews about directed evolution were published, the majority of these dealing with the application of engineered enzymes. With so many excellent reviews already written, we will only describe articles published in 2003 and early 2004. In nature, evolution and creation of new functionalities is achieved by mutagenesis, recombination and survival of the
fittest. Directed evolution mimics this and is a process of iterative cycles of producing mutants and finding the mutant with the desired properties. The constantly continuing process of mutagenesis is the easiest one to copy in the lab. Mutations can be introduced at specific places using (site-) directed mutagenesis or throughout the gene by random mutagenesis. Several mutagenesis techniques have been developed in order to avoid codon bias. These were recently described by Neylon and Lutz and Patrick in two outstanding reviews [4,5]. More radical changes in evolution originate from recombination. Pim Stemmer was the first to recognise this and evolution was mimicked using a technique he called DNA-shuffling [6,7]. This method is based on the mixing and subsequent joining of different related small DNA fragments in order to form a complete new gene. In the process of shuffling, the recombination frequency is dependent on the degree of homology. A high level of recombination is important to get all possible combinations of mutations. Since recombination can be biased, several methods to overcome problems arising by the use of shuffling in the early years where tackled by novel strategies, all having their own advantages and disadvantages. These new recombination methods can be divided into six categories according to their basic idea (Table 1).
Table 1 Different recombination techniques with their main advantages and disadvantages Group
Technique
Members
Shuffling
Recombination of small fragments based on homology in the sequence between mutations that stem from all kinds of mutagenesis strategies or different family members Aims for high recombination, but difficult to separate close mutations
Shuffling [6,7] Family shuffling [45] RE cut shuffling [46] ssDNA shuffling [47] Mn2+ DNase cut [48] Endonuclease V cut [49] RPR [50] RETT [51] SCRATCHY [52]
Full length parent shuffling
Recombination of small fragments from different origin using one or more full length parent strands Higher recombination frequency, but more elaborate
StEP [53] RACHITT [54]
Single cross-over
Recombination of non-homologous genes by ligating front and back of two different genes, selection of new genes on size Recombination possible between low or non-homologous genes, but only one recombination point
(THIO)ITCHY [55,56] SHIPREC [57]
Recombination of structural, functional or less homologous parts of different family members More active enzymes in the resulting library, but only a few recombination points, which are hard to find
Exon shuffling [58] DOGS [59]
In vivo recombination
Recombination using the gap repair system of yeast or the recE/recT system of E. coli High yield, since no ligation necessary, but specialised vectors and multiple steps necessary
CLERY [63] ET-recombination [64]
Synthetic shuffling
Recombination of (un)known mutations in synthetic oligonucleotides Recombination of close mutations possible, but expensive and good selection necessary
Single step shuffling [65] DHR [16] Synthetic shuffling [66] ADO [15]
Domain swapping
SCRATCHY [52]
SISDC [60] YLBS [61] SCOPE [62]
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
Depending on the goal of the experiment one should decide which process is the best to create the optimal library. Although the production of an extensive and unbiased library is very important, the success of a directed evolution experiment highly depends on the method that is used to find the best mutant enzyme [8]. Since most directed evolution experiments generate a huge mutant library, a high throughput screening strategy or, more preferably, a selection has to be developed for different enzyme functions. These methods should be sufficiently sensitive and specific to identify positive mutants [9]. The advantage of screening is that the difference between substrate and product of an enzymatic reaction can be determined directly or indirectly in almost every case. However, relatively simple visual screens such as the formation of coloured or fluorescent products or halos around a colony on a plate are not always available. This implies the use of high throughput screening (HTS) in microtiter plates. HTS demands miniaturisation and automation of enzymatic assays. In the past decade a lot of research has been focused on finding better, cheaper, quicker and more accurate HTS assays [10]. This has made HTS feasible for many laboratories all over the world now, resulting in a lot of smart enzymatic screening methods. The disadvantage of screening lays in the fact that all individual mutants have to be tested for the desired enzymatic reaction, even those that might not be active or accurately folded. In a directed evolution approach using recombination these nonactive enzymes are typically 50–80% of the total library. After randomisation of a whole gene, which can result in unpredictable interesting new mutations outside the active site, the frequency of inactive mutants may be even higher. Therefore, the most preferred way of finding the best mutant is by selection. Selection mimics the natural survival of the fittest strategy and is the most efficient method to find the best mutant, since only mutants of interest will appear. Unfortunately, this approach is not possible for all enzymatic activities. Selection is based on the fact that mutants with the desired enzyme function provide an advantage to the host cell over bacteria bearing wild type enzymes. However, a lot of enzymatic activities are not essential to the bacterium. This can be overcome by coupling the activity of interest to an essential feature of the bacterium. This is very challenging and has already resulted in very elegant solutions from several labs. In this review, we will focus on new mutagenesis and selection techniques and the directed evolution methods using selection.
3
one problem concessions have to be made to other aspects of library construction. Bichet et al. tried to solve the problem of the losses of mutants through ligation [11]. They designed a ‘new’ errorprone PCR (epPCR) strategy, they call the Bringer Strategy, that does not involve a ligation step to bring the mutant gene back into the vector. The strategy is based on amplification of ring closed plasmids (e.g. QuikChange, Stratagene). The whole plasmid containing the gene of interest is amplified by PCR, but instead of using primers containing mutations, wild type primers are used and epPCR is performed on the whole plasmid. After discarding the wild type with DpnI, the resulting mutant plasmid is transformed. Although it is a real advantage not to be dependent on ligation efficiency of the PCR product into the vector, a lot of controls are necessary to be sure that only mutant plasmids are transformed and no other regions in the plasmids are damaged by mistakes in the PCR. In order to be able to recombine low homologous stretches of DNA, Bittker et al. developed the nonhomologous random recombination (NRR) strategy [12]. Recombination is accomplished by adding random blunt end DNA fragments together and select for binding of the newly prepared DNA stretch. Recently, they improved this technique into protein-NRR [13]. They start with fragments of one or more genes and control the size of the newly recombined DNA by adding a certain ratio of hairpin loops. DNA of the right size is obtained from an agarose gel and cloned into an expression vector. The expressed mutant proteins are selected for the desired enzyme activity (Fig. 1).
3. New library production methods Every technique to create mutant libraries has its own advantages and drawbacks. The objectives to develop a new method vary from creating more random (=unbiased) libraries to generating libraries from low homologous genes and producing a library more rapid. Usually, in the solving of
Fig. 1. Schematic representation of protein-NRR. Blunt-ended DNA fragments are ligated, size-selected and cloned into an expression vector. The size of the first pool of ligated fragments is controlled by the amount of blunt hairpins added to the ligation.
4
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
In this way they found active chorismate mutases with several insertions and deletions at certain defined points in the enzyme, but not at other points. Protein-NRR is therefore a useful tool for finding important functional parts of an enzyme in a broad and unbiased manner. It is, however, essential to have a good selection method, since many inactive proteins will be formed. The bias towards the original codon in random mutagenesis strategies was tackled by Wong et al. [14]. They designed a new mutagenesis strategy called sequence saturation mutagenesis (SeSaM). In this method, mutagenesis is performed by incorporating dNTPs nucleotides at random positions in the gene by PCR. Cutting at these positions reveals random fragments to which (different) universal bases are added. The whole gene is restored using PCR and the universal base is replaced by a normal one. Using dATPs and the universal base deoxyinosine, they found a library of well-distributed point mutations and some insertions or deletions. To get a totally unbiased library, this 2–3 days strategy should be repeated for all four dNTPs and several different universal bases, which opposes the claimed speed of the strategy. Using assembly of designed oligonucleotides (ADO), Zha et al. tried to get a high recombination frequency between different parents by producing oligonucleotides comprising all possible recombinations of alleles [15]. These designed oligo’s on alternating strands were assembled using homologous overlapping parts of the different parents. This resulted in chimaeras of lipA and lipB with a higher enantioselectivity towards the desired substrate. The difference between this method and the previously published degenerate homoduplex gene family recombination (DHR) [16] is the fact that the oligo’s are more random. Therefore, they may introduce new mutations that are not present in one of the parents.
4. Evolution of biocatalysts Biocatalysis becomes more important in industries. In the pharmaceutical industry, for example, it is important nowadays to produce enantiopure drugs to reduce the side effects and metabolic burden on the body. Enantioselectivity is, however, difficult to select for and libraries are usually screened as reviewed by Reetz [17,18]. Other industrially important enzymes may be selected by sophisticated methods. In our lab, a glutaryl acylase was engineered towards an adipyl acylase by randomising the gene using epPCR followed by a selection on a mimic substrate. The small set of selected mutants was subsequently screened on the real substrates (Fig. 2) [19]. This directed evolution step guided us to the important residues, which we subsequently randomised into all 19 others to find the best amino acid at each position [20,21]. Determining the catalytic parameters from each purified enzyme on adipyl-7-aminode-
Fig. 2. The mimic substrates adipyl-serine and adipyl-leucine, were used as selection substrate. Selected mutants were tested for their activity on the desired substrate adipyl-7-ADCA, which is quite similar.
sacetoxycephalosporanic acid (adipyl-7-ADCA) revealed a single mutant having a 15 times improved catalytic efficiency resulting from an improved Km and a kcat similar to the kcat of wild type enzyme on glutaryl-7-aminocephalosporanic acid (glutaryl-7-ACA). This enzyme is almost good enough to be used in an industrial process to produce 7-ADCA from adipyl-7-ADCA. The production of a library and developing of a selection procedure that renders only wanted mutants can be difficult as is obvious from the research done by the group of Buchholz [22]. They used directed evolution techniques on retroviruses. Retroviruses can be used in gene therapy to bring beneficial DNA into cells, but also to bring suicide genes into tumour cells in order to kill them. One of the strategies to target retroviruses towards the tumour cell is by inserting a protein sequence into the retroviral envelope protein. Activation, and thereby spreading of the retrovirus, will only take place if this sequence is cleaved by a tumour specific protease. It is, however, not known which protein sequence will be recognised by proteases from the tumour cells. Therefore, a directed evolution experiment was designed in order to change the linker peptide into a better target for a matrix metalloprotease (MMP), which is very common in tumour cells. The first generation of this library was already described in 1998 [23], but resulted only in protein sequences recognised by omnipresent serine proteases. In the present research a new strategy was used to create a library without target peptides for serine proteases. Furthermore, another envelop protein was used to be able
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
5
Fig. 3. Different methods to use a selection system to study protein–protein interactions. In both systems a reporter activity is restored upon binding of two proteins. This activity can be used to give the host a growth advantage or to assay the amount of reporter produced. (A) In the yeast two-hybrid method the gene of interest is fused to a DNA binding domain and the mutant library is fused to an activation domain, which activates the reporter gene. In the yeast three-hybrid system, a ligand promotes the binding of the target and binding protein. (B) In the split-protein system, the interaction between two proteins brings the two halves of the reporter protein together.
to select retroviruses in tumour cells from human origin. Two consecutive cycles of diversification and in vivo selection resulted in a more efficient cutting of the target site by the MMP produced by the cancer cell-line. This selected MMP-activatable retrovirus spread more effectively around the tumour. Since the method is reasonably rapid, it can be used to make an individual highly effective therapy possible. Van der Veen et al. developed a selection method to use as a measure of amylosucrase activity [24]. The procedure is based on a zero background expression cloning strategy in order to select against plasmids containing no amylosucrose gene. Transformants were scraped from the first plates containing sucrose and selection was performed on agar plates containing sucrose as the only carbon source, thereby selecting for active enzymes. The selected enzymes are subsequently screened in an automated screening that was designed to discriminate between different actions of the enzyme. A pilot experiment was performed using random mutagenesis and a round of shuffling to lose the high percentage of inactive clones. Two libraries were selected using the new strategy and selected mutants were screened for amylosucrase activity. From the 1250 transformants plated, only 60 were active, stressing the importance of the (pre)selection. Two improved mutants were found having mutations, which were not considered changing from the crystal structure.
5. Understanding natural evolution and biological functions Although directed evolution mimics its natural variant, the actual evolution process is still not fully understood. A better understanding of natural evolution will allow mimicking the process better. To do so, different labs are trying to evolve one family member into another or different functionalities from a so-called ancient parent. In Kirsch’s lab it is tried to evolve aspartate aminotransferase (AATase) into tyrosine aminotransferase (TATase) to show that they are evolutionary related. Both rational design and directed evolution were performed to reach this goal. The best mutant from the directed evolution approach gave a complementation of tyrosine auxotrophy, but the rationally designed HEX mutant did not, although it had some TATase activity in vitro [25]. Therefore, HEX was shuffled (with itself) and selected for complementation [26]. Two mutants showed to be active in vivo and these appeared to have a lower Km. Furthermore, they displayed less product inhibition. The best new mutation in this shuffled HEX library, A293D, was introduced in the best mutant from the directed evolution library, which resulted in a TATase activity similar to the wild type AATase activity [27]. The three new mutants were crystallised and TATase activity of the mutants could very well be explained by the crystal structure, although
6
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
this residue was not considered changing in the first rational library. Christ and Winter try to find proteins that are related to Rop, an antisense RNA binding protein [28]. To do so, they randomised five important residues that were found in a previous Ala-scanning experiment. Selection was done using a very elegant liquid growing system in which the original Rop gene of the plasmid was deleted. This triggers the bacterium to make more plasmid and thereby grow slower. Upon restoring the (mutant) gene into the bacterium, it grows faster and outgrows others in the culture medium. Three rounds of selection were performed. A reporter gene on the plasmid can be used as a second screen. Eight active variants were found after three rounds of selection and appeared to have a low kd as determined by surface plasmon resonance. The sequence of the selected mutants helped to define a search tool to find distantly related proteins. The writers, however, fairly admit that these are related in function/structure and not necessarily in an evolutionary sense. Buskirk et al. try to unravel the complex biological function of transcription activation [29]. Their goal was to make active transcriptional elements from random RNA sequences with comparable potency as the natural ones. The random RNA sequences were synthesised as a variable region of DNA coding for a larger RNA structure with a stable secondary structure. Another part of this RNA consists of two MS2 hairpins, which bind to one of the partners of the yeast three-hybrid system (Fig. 3A). Active RNA sequences can trigger histidine production, which makes them selectable in a histidine–auxotrophic host on selective plates containing no histidine. The activation was quantified by using a second yeast three-hybrid system with b-galactosidase as reporter enzyme. Two rounds of evolution and selection were performed in which surprisingly high amounts of survivors were found (0.2%). Only one RNA sequence appeared to be really active when looking into the b-galactosidase activity. This RNA activator was only two times lower than a normal protein activator. Peimbert and Segovia used a two step approach to change a transpeptidase into a b-lactamase to show their evolutionary relationship [30]. They started with a rational site directed mutagenesis library in which important residues (according to the crystal structure) were randomised. This library was subjected to random mutagenesis by epPCR and recloned into an expression plasmid. Transformants were enriched in liquid media containing cefotaxime, plasmids were retransformed and plated onto LB-plates containing different concentrations of cefotaxime. This resulted in a mutant with a 105-fold increase in cefotaxime deacylation. Performing several epPCR rounds and a shuffling step with the wild type gene or saturation mutagenesis of important residues found in the first library did not result in better mutants. This suggests that using this selection system no better mutant can be found and another, maybe
more stringent, method should be used for subsequent rounds.
6. Evolution of a molecular toolbox Every directed evolution experiment is an individual search for the best enzyme for the desired reaction. In order to be able to attack each question as quick and direct as possible it is most convenient to have some general tools at hand. One of the possibilities is to maintain a huge variety of similar enzymes in the freezer and screen them for the desired activity when necessary. Another complementary strategy is to have some general directed evolution methods ‘on the shelf’ with readily adjustable screening or selection methods. Azizi et al. described a very elegant new selection strategy in yeast based on the yeast two-hybrid system in order to find new ligands or new receptors for existing ligands (Fig. 3A) [31]. In this system the nuclear receptor’s ligand binding domain (LBD) is fused to the Gal4 DNA binding domain (Gal4 DBD). The fusion protein binds to the Gal4 response element controlling the expression of the His3 gene. Only in the presence of a good ligand the fusion protein will enable the His3 gene to be expressed, which makes yeast cells able to grow on media without histidine. The benefit of this system is that the library of ligands is not co-expressed with the strain, but added to the growth medium. This strategy can be very well combined with random mutagenesis of the LBD resulting in a total directed evolution strategy, even with a simultaneous selection on different ligands. Another selection strategy to investigate protein–protein interaction is the use of split-protein sensors (Fig. 3B). Creating split-proteins used to be a process of trial and error, but Tafelmeyer et al. developed a method to make a new split-protein using directed evolution [32]. They fused the N and C terminus of a Trp gene together and generated a library with splits in every possible position. This library was cloned into an expression vector containing two polypeptides that associate into an antiparallel coiled coil in order to bring the two halves together. The library was then selected for active enzyme, which resulted in several candidates. One of the selected split-Trp pairs was successfully applied to monitor protein–protein interactions both at the membrane as well as in the cytosol of yeast. This powerful selection complements the repertoire of the currently used split-protein sensors and provides a new tool for high-throughput interaction screening. Gimble et al. used the bacterial two-hybrid selection system (Fig. 3A) [33] in E. coli to change the substrate specificity of the homing endonuclease PI-SceI protein [34]. Homing endonucleases have been used in genomic engineering and variants recognising other DNA stretches will be useful. From the crystal structure, they determined ˚ of the binding basepairs +18 which amino acids within 7 A
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
and +19 had direct interactions or possible hydrogen bonding with these residues. The five selected residues were randomised in a PCR using NNK codons at the target positions. The bacterial two-hybrid system was used to select for mutants that recognise a new target DNA. Some mutants recognising new target DNA with little conservation of the wild type protein residues were found that surprisingly still recognise the wild type sequence, pointing to the plasticity of the recognition site of these enzymes. Griffiths and Tawfik improved their in vitro compartmentalisation (IVC) selection system that uses oil-in-water emulsions [35]. They made a link between phenotype and genotype of the target enzyme by attaching its DNA on a streptavidin coated bead. The enzyme is produced in the compartment and attached to the same bead as the gene, thereby making a genotype/phenotype linkage. The compartments are broken and new ones are formed in which substrate bound to biotin is added. The substrate can be converted by the bound enzyme and will be caught by the bead as well. Beads can then be selected using antibodies against the product with affinity tags or fluorescently labelled and sorted by fluorescence activated cell sorting (FACS). This completely in vitro process selects for all enzymatic features simultaneously (substrate recognition, product formation, rate acceleration and turnover) and single enzyme molecules can be detected. The strength of the method was shown by improving the catalytic activity of a phosphotriesterase. The most enhanced mutant showed a 63fold improved kcat, compared to the already very efficient wild type enzyme [36]. Zhu et al. changed the substrate specificity from restriction enzyme BsoBI [37]. They started with rational design, by changing the amino acid in close contact with the substrate as seen in the crystal structure. One of the mutants with changed substrate specificity was then further mutagenised using epPCR, in order to increase enzymatic activity. Selection was performed in vivo by looking at the dark blue colour of colonies from an indicator strain on a plate. This resulted in more active variants with changed substrate specificity. Surprisingly, most mutations were found outside the protein–DNA interface region. Rimseliene et al. also described a new selection system to produce restriction enzymes with a new specificity, which they call MABS (methylation activity-based selection) [38]. The procedure can only be used on bifunctional type II endonucleases and includes three steps. The bifunctional restriction enzyme is transformed into a monofunctional DNA-methyltransferase by cleavage centre disruption. Then mutants are selected which show altered DNA modification specificity based on their ability to protect predetermined DNA targets. In the final step the cleavage centre’s wild type structure is reconstituted. New enzymes made by this method can be used as new molecular tools in cloning and restriction analysis. Another publication in which a methyltransferase was changed by directed evolution is the paper from Cohen et al.
7
[39]. In order to change substrate specificity of HaeIII methyltransferase, they used a two-step mutagenesis strategy followed by selection using in vitro compartmentalisation. The initial mutagenesis of DNA contacting residues was followed by randomisation of the loop that lies behind these residues. The selection system was based on the addition of the desired DNA stretch to the gene resulting in a covalent linkage between the gene and the methylated DNA, in case of a good mutant. This resulted in the remarkable finding of a mutant with a catalytic activity towards the new substrate that was even better than wild type activity towards the original substrate. Recombination is one of the cornerstones of directed evolution. Voziyanov et al. are using the Flp recombinase to recombine DNA fragments. The drawback of this method is the fact that the enzyme only recognises a certain stretch of DNA, which has to be inserted into the gene(s) of interest to get recombination. To overcome this problem they used mutagenesis and shuffling to rapidly evolve different mutant recombinases that recognised other stretches of DNA [40]. Combining two different mutants in one experiment can lead to a recombination system that utilises bi-specific hybrid target sites. A hybrid site is assembled from two half-sites, each harbouring a distinct binding specificity. Directed evolution was used to find mutations that improve the association of the two halves, thereby expanding the use of these enzymes for recombination [41]. In directed evolution experiments it is usually attempted to search the largest sequence space possible. Starting from a known protein limits this sequence space. The group of Kim HS has described a new method to expand the sequence space, which they call functional salvage screen (FSS). First they introduced deletions into green fluorescent protein (GFP) to inactivate them. Starting from these defective GFPs they added random sequences at the mutated sites and screened for proteins that regained fluorescence. This resulted in a mutant with low fluorescence intensity and stability [42]. Using directed evolution they got an improved variant after three rounds of evolution with four mutations and a ‘totally’ different stretch of amino acids than the wild type enzyme at that position. The fluorescence and stability from this mutant is, however, similar to the original enzyme [43]. Fa et al. tried to expand the substrate repertoire of a DNA polymerase by directed evolution [44]. Their goal was to allow enzymatic synthesis of unnatural polymers in vivo or in vitro. Therefore, they developed an activity-based selection method, which isolates polymerase mutants with the desired property from libraries of the enzyme displayed on phage. The method is based on the co-display (on a single phage particle) of a member of a mutant enzyme library and an acidic peptide that is used to attach a DNA primertemplate substrate. A mutant polymerase that is able to incorporate methylated and biotinylated bases into the growing strain will be selected by streptavidin-coated beads. Two libraries were constructed. In each one, six residues
8
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9
Fig. 4. General directed evolution scheme.
chosen by looking at the crystal structure were randomised. After four rounds of selection one mutant enzyme incorporated methylated dATP at a similar rate as wild type incorporates non-methylated dATP.
7. Conclusions Directed evolution becomes more and more an integrated research field. Both rational and random strategies are alternately used to create the optimal library for finding the desired enzyme. New in vitro and in vivo selection strategies are developed to select for all kinds of new activities. All these components are used in very diverse formats to reach the same goal: creating new enzymes with desired properties (Fig. 4). These properties can be dictated by industrial or scientific needs and directed evolution is even used as a tool to better understand the principles of natural evolution. This will encourage the development of even more directed evolution strategies to find the optimal library size and the best selection or screening method for each new enzyme.
Acknowledgements This review was written on the basis of discussions with many colleagues at the Department of Pharmaceutical Biology of the University of Groningen, with financial support from STW, a part of the Dutch Organisation for Science (NWO). W.Q. further acknowledges support from the EU commission in project QLTR-2001-00519.
References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
Dalby PA. Curr Opin Struct Biol 2003;13:500–5. Schoemaker HE, Mink D, Wubbolts MG. Science 2003;299:1694–7. Marrs B, Delagrave S, Murphy D. Curr Opin Microbiol 1999;2:241–5. Neylon C. Nucl Acids Res 2004;32:1448–59. Lutz S, Patrick WM. Curr Opin Biotechnol 2004;15:291–7. Stemmer WP. Nature 1994;370:389–91. Stemmer WP. Proc Natl Acad Sci USA 1994;91:10747–51. You L, Arnold FH. Protein Eng 1996;9:77–83. Zhao H, Arnold FH. Curr Opin Struct Biol 1997;7:480–5. Delagrave S, Murphy DJ. Assay Drug Develop Technol 2003;1:187– 98. [11] Bichet A, Bureik M, Lenz N, Bernhardt R. Appl Biochem Biotechnol 2004;117:115–22.
[12] Bittker JA, Le BV, Liu DR. Nat Biotechnol 2002;20:1024–9. [13] Bittker JA, Le BV, Liu JM, Liu DR. Proc Natl Acad Sci USA 2004;101:7011–6. [14] Wong TS, Tee KL, Hauer B, Schwaneberg U. Nucl Acids Res 2004;32:e26. [15] Zha D, Eipper A, Reetz MT. Chembiochem 2003;4:34–9. [16] Coco WM, Encell LP, Levinson WE, Crist MJ, Loomis AK, Licato LL, et al. Nat Biotechnol 2002;20:1246–50. [17] Reetz MT. Tetrahedron 2002;58:6595–602. [18] Reetz MT. Proc Natl Acad Sci USA 2004;101:5716–22. [19] Otten LG, Sio CF, Vrielink J, Cool RH, Quax WJ. J Biol Chem 2002;277:42121–7. [20] Sio CF, Otten LG, Cool RH, Quax WJ. Biochem Biophys Res Comm 2003;312:755–60. [21] Otten LG, Sio CF, van der Sloot AM, Cool RH, Quax WJ. Chem Bio Chem 2004;5:820–5. [22] Schneider RM, Medvedovska Y, Hartl I, Voelker B, Chadwick MP, Russell SJ, et al. Gene Ther 2003;10:1370–80. [23] Buchholz CJ, Peng KW, Morling FJ, Zhang J, Cosset FL, Russell SJ. Nat Biotechnol 1998;16:951–4. [24] van der Veen BA, Potocki-Veronese G, Albenne C, Joucla G, Monsan P, Remaud-Simeon M. Febs Lett 2004;560:91–7. [25] Rothman SC, Kirsch JF. J Mol Biol 2003;327:593–608. [26] Rothman SC, Voorhies M, Kirsch JF. Protein Sci 2004;13:763–72. [27] Chow MA, McElroy KE, Corbett KD, Berger JM, Kirsch JF. Biochemistry 2004;43:12780–7. [28] Christ D, Winter G. Proc Natl Acad Sci USA 2003;100:13202–6. [29] Buskirk AR, Kehayova PD, Landrigan A, Liu DR. Chem Biol 2003;10:533–40. [30] Peimbert M, Segovia L. Protein Eng 2003;16:27–35. [31] Azizi B, Chang EI, Doyle DF. Biochem Biophys Res Comm 2003;306:774–80. [32] Tafelmeyer P, Johnsson N, Johnsson K. Chem Biol 2004;11:681–9. [33] Joung JK, Ramm EI, Pabo CO. Proc Natl Acad Sci USA 2000;97:7382–7. [34] Gimble FS, Moure CM, Posey KL. J Mol Biol 2003;334:993–1008. [35] Tawfik DS, Griffiths AD. Nat Biotechnol 1998;16:652–6. [36] Griffiths AD, Tawfik DS. EMBO J 2003;22:24–35. [37] Zhu ZY, Zhou J, Friedman AM, Xu SY. J Mol Biol 2003;330:359– 72. [38] Rimseliene R, Maneliene Z, Lubys A, Janulaitis A. J Mol Biol 2003;327:383–91. [39] Cohen HM, Tawfik DS, Griffiths AD. Prot Eng Des Sel 2004;17:3–11. [40] Voziyanov Y, Konieczka JH, Stewart AF, Jayaram M. J Mol Biol 2003;326:65–76. [41] Konieczka JH, Paek A, Jayaram M, Voziyanov Y. J Mol Biol 2004;339:365–78. [42] Kim GJ, Cheon YH, Park MS, Park HS, Kim HS. Protein Eng 2001;14:647–54. [43] Nam SH, Oh KH, Kim GJ, Kim HS. Protein Eng 2003;16:1099–105. [44] Fa M, Radeghieri A, Henry AA, Romesberg FE. J Am Chem Soc 2004;126:1748–54. [45] Crameri A, Raillard SA, Bermudez E, Stemmer WP. Nature 1998;391:288–91. [46] Kikuchi M, Ohnishi K, Harayama S. Gene 1999;236:159–67.
L.G. Otten, W.J. Quax / Biomolecular Engineering 22 (2005) 1–9 [47] [48] [49] [50] [51] [52] [53] [54] [55] [56]
Kikuchi M, Ohnishi K, Harayama S. Gene 2000;243:133–7. Lorimer IA, Pastan I. Nucl Acids Res 1995;23:3067–8. Miyazaki K. Nucl Acids Res 2002;30:e139. Shao Z, Zhao H, Giver L, Arnold FH. Nucl Acids Res 1998;26:681–3. Lee SH, Ryu EJ, Kang MJ, Wang ES, Piao Z, Choi YJ, et al. J Mol Catal B-Enzym 2003;26:119–29. Lutz S, Ostermeier M, Moore GL, Maranas CD, Benkovic SJ. Proc Natl Acad Sci USA 2001;98:11248–53. Zhao H, Giver L, Shao Z, Affholter JA, Arnold FH. Nat Biotechnol 1998;16:258–61. Coco WM, Levinson WE, Crist MJ, Hektor HJ, Darzins A, Pienkos PT, et al. Nat Biotechnol 2001;19:354–9. Ostermeier M, Shim JH, Benkovic SJ. Nat Biotechnol 1999;17: 1205–9. Lutz S, Ostermeier M, Benkovic SJ. Nucl Acids Res 2001;29:16.
9
[57] Sieber V, Martinez CA, Arnold FH. Nat Biotechnol 2001;19:456– 60. [58] Kolkman JA, Stemmer WP. Nat Biotechnol 2001;19:423–8. [59] Gibbs MD, Nevalainen KMH, Bergquist PL. Gene 2001;271:13–20. [60] Hiraga K, Arnold FH. J Mol Biol 2003;330:287–96. [61] Kitamura K, Kinoshita Y, Narasaki S, Nemoto N, Husimi Y, Nishigaki K. Protein Eng 2002;15:843–53. [62] O’Maille PE, Bakhtina M, Tsai MD. J Mol Biol 2002;321:677–91. [63] Abecassis V, Pompon D, Truan G. Nucl Acids Res 2000;28:88. [64] Strausberg SL, Alexander PA, Gallagher DT, Gilliland GL, Barnett BL, Bryan PN. Biotechnology (NY) 1995;13:669–73. [65] Stemmer WP, Crameri A, Ha KD, Brennan TM, Heyneker HL. Gene 1995;164:49–53. [66] Ness JE, Kim S, Gottman A, Pak R, Krebber A, Borchert TV, et al. Nat Biotechnol 2002;20:1251–5.