PEDS Advance Access originally published online on April 27, 2007
Protein Engineering Design and Selection 2007 20(5):243-256; doi:10.1093/protein/gzm017
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Multivariate-activity mining for molecular quasi-species in a glutathione transferase mutant library
Department of Biochemistry and Organic Chemistry, Uppsala University, BMC, Box 576, SE-75123 Uppsala, Sweden
3 To whom correspondence should be addressed. E-mail: Bengt.Mannervik{at}biorg.uu.se
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
A library of recombinant glutathione transferases (GSTs) generated by shuffling of DNA encoding human GST M1-1 and GST M2-2 was screened with eight alternative substrates, and the activities were subjected to multivariate analysis. Assays were made in lysates of bacteria in which the GST variants had been expressed. The primary data showed clustering of the activities in eight-dimensional substrate-activity space. For an incisive analysis, the rows of the data matrix, corresponding to the different enzyme variants, were individually scaled to unit length, thus accounting for different expression levels of the enzymes. The columns representing the activities with alternative substrates were subsequently individually normalized to unit variance and a zero mean. By this standardization, the data were adjusted to comparable orders of magnitude. Three molecular quasi-species were recognized by multivariate K-means and principal component analyses. Two of them encompassed the parental GST M1-1 and GST M2-2. A third one diverged functionally by displaying enhanced activities with some substrates and suppressed activities with signature substrates for GST M1-1 and GST M2-2. A fourth cluster contained mutants with impaired functions and was not regarded as a quasi-species. Sequence analysis of representatives of the mutant clusters demonstrated that the majority of the variants in the diverging novel quasi-species were structurally similar to the M1-like GSTs, but distinguished themselves from GST M1-1 by a Ser to Thr substitution in the active site. The data show that multivariate analysis of functional profiles can identify small structural changes influencing the evolution of enzymes with novel substrate-activity profiles.
Keywords: directed evolution/DNA shuffling/glutathione transferase/library/multivariate analysis
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
Redesign of protein structures is a powerful approach to evolving novel functions in already existing enzymes (Anantharaman et al., 2003
; Orengo and Thornton, 2005). However, the rational creation of enzymes with novel activities on the basis of structural data remains a major challenge, especially since even highly similar sequences may display totally different functions (White, 2006). As a result, the empirical approach of directed evolution has over a relatively short period of time emerged as a powerful alternative to structure-based redesign for altering and enhancing properties of enzymes and other proteins (Svendsen, 2004).
There are numerous strategies to create mutant libraries with adequate sequence variations (Valetti and Gilardi, 2004; Bloom et al., 2005; Meyer et al., 2006) and diverse procedures for examining such libraries for recombinant proteins with the targeted properties (Farinas et al., 2001; Aharoni et al., 2005). However, methods for analyzing the functional consequences have not received comparable attention. Structural studies of genomes and proteomes are essentially limited to a small number of dimensions. The primary structure of a polynucleotide or a polypeptide is a one-dimensional entity, whereas folded structures can be traced in three-dimensions, and their temporal changes can be described by adding time as a fourth dimension. In contrast, functional properties can be expressed in numerous dimensions defined by the parameters investigated.
In addition to the eventual goal of evolving enzymes with tailor-made activities, we wanted to explore the distribution of members of a mutant library in functional substrate-activity space. This information could provide a basis for choosing the optimal variants from a suitable ‘quasi-species’ of a mutant library (Eigen et al., 1988) to be parents for a subsequent generation of mutants. The optimal parentage for a new generation is not limited to the mutants that have evolved the furthest in a desired direction, since their genetic background may be too narrow (Ness et al., 1999). Therefore, the identification of an entire group of near-optimal mutants is an important task.
The gene superfamily of glutathione transferases (GSTs) illustrates how similar structures can display widely different properties. GSTs were originally discovered as a group of detoxication enzymes catalyzing the conjugation of a wide variety of electrophilic substances with the sulfhydryl group of the tripeptide glutathione (GSH) (Josephy and Mannervik, 2006). The three-dimensional structure is highly conserved among the soluble GSTs, and consists of two identical, or similar, protein subunits (Armstrong, 1997). Each subunit has an N-terminal domain with the same (
/ß)-fold as the thioredoxin molecule. The second domain is a bundle of helical segments. This canonical two-domain structure of soluble GSTs is also found in other proteins with unrelated assignments, such as lens crystallins of cephalopod eyes and an eukaryotic translation elongation factor (Koonin et al., 1994). These alternative functions could probably not have been predicted by currently available bioinformatics tools; neither from the nucleotide sequence of the corresponding DNA nor from the crystal structure of expressed proteins.
In recent investigations, we explored the concept of ‘molecular quasi-species’ as the optimal group of recombinant mutants for recursive mutagenesis toward novel properties (Larsson et al., 2004; Emrén et al., 2006). In one model system, a GST M1/M2 library obtained by shuffling of DNA encoding human GST M1-1 and GST M2-2 (Hansson et al., 1999a) was characterized with a panel of alternative substrates (Emrén et al., 2006). Multivariate analysis of the activities of variant GSTs clearly demonstrated that the enzymes segregate into functionally discrete distributions, rather than displaying variants with properties ranging in a continuum between those of the parental GSTs. However, the analysis did not approach the bias that could arise when mutant proteins are expressed at different levels in the samples analyzed. Further, the structural basis for the clustering of the variant GSTs was lacking as a validation of the functional analysis. These issues are addressed in this paper.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
GST M1/M2 library
A library of mutant enzymes was produced by shuffling of cDNAs encoding human GST M1-1 and GST M2-2 (Hansson et al., 1999a). DNA from the library was used to transform electrocompetent Escherichia coli XL1-Blue cells (Stratagene). Transformants were allowed to recover in 2TY medium [1.6% (w/v) Tryptone, 1% (w/v) yeast extract and 0.5% (w/v) NaCl] at 37°C for 1 h before spreading them on LB-ampicillin plates [1% (w/v) Tryptone, 0.5% yeast extract, 1% (w/v) NaCl, 1.5% Bacto-agar and 100 µg/ml of ampicillin]. The M1/M2 library was estimated to contain 5 x 106 independent clones.
Preparation of bacterial lysates
The parental GST M1-1 and GST M2-2, as well as 384 mutants randomly picked from the library, were expressed in E.coli. The clones were individually grown overnight at 37°C with agitation in 2 ml of LB medium [1% (w/v) Tryptone, 0.5% (w/v) yeast extract and 1% (w/v) NaCl] supplemented with 100 µg/ml of ampicillin. Samples of the cultures were diluted 100-fold into 10 ml 2TY medium supplemented as above. After 2 hours at 37°C the production of GST was induced by addition of isopropyl-ß-D-thiogalactopyranoside to a final concentration of 0.2 mM. The bacteria were then grown for an additional 16 h before being harvested by centrifugation at 1500g for 10 min at 4°C. The supernatant was removed and the bacteria were resuspended in a 50-ml Falcon tube with 250 µl 0.1 M sodium phosphate buffer, pH 6.5, supplemented with 0.2 mg/ml of lysozyme in order to lyse the cells. The suspension was left on ice for 1 h prior to completing the lysis by freezing and thawing at –80°C for 10 min and at 37°C for 5 min, respectively, performed three times. After the final thawing, the samples were centrifuged for 30 min at 15 000 g. The supernatants were collected and transferred to 96-well plates and stored at –80°C before use.
Assays of catalytic activities
The GST activities of the lysates were tested with eight different substrates. The measurements were made at 30°C on a SPECTRAmaxPLUS 384 microplate spectrophotometer (Molecular Devices, Sunnyvale, CA, USA). The exception was monochlorobimane (MCB), which was assayed fluorometrically in microplates on a Fluoroskan Ascent (Labsystems, Helsinki, Finland) at room temperature (Eklund et al., 2002). All measurements were made in duplicate within 12 h after thawing of the lysates. The volume of lysate added to the assay system was adjusted to give linear progress curves of the reaction velocity for each substrate. Reaction traces of every clone with all substrates were inspected individually. The activity measurements were performed in 0.1 M sodium phosphate buffer at pH 6.5, except for p-nitrophenylacetate (pNPA) (Keen and Jakoby, 1978), with which the measurements were performed in sodium phosphate buffer at pH 7.0. The activity measured with pNPA in the lysate was proven to be GST-catalyzed by addition of S-(2-iodobenzyl) glutathione, a known GST inhibitor, as well as by measuring the activity in lysates lacking expressed GST. The assays for 1-chloro-2,4-dinitrobenzene (CDNB) and epoxy-3-(4-nitrophenoxy)-propane (EPNP) were performed as described by Shokeer et al. (2005). The measurements with 3-(4-nitrophenyl)-glycidol (NPG) were performed according to Comstock et al. (1994). Trans-4-phenyl-3-buten-2-one (tPBO) was measured as described by Habig et al. (1974). The assay with phenethyl isothiocyanate (PEITC) was according to Kolm et al. (1995). The conditions for the novel substrate 1-nitro-1-cyclohexene (NCH) were 0.5 mM GSH and 0.05 mM NCH in 0.1 M sodium phosphate buffer, pH 6.5, at 30°C. The product formation was monitored at 280 nm, and the net extinction coefficient was determined to be 4.2 mM–1 cm–1.
Purification and specific activity determinations of selected GST variants
Purification of variants 72, 74, 295, 342 and 383 was made from overnight cultures of E.coli XL1-Blue using expression vectors carrying the respective clones. The bacteria were grown in 2TY-medium supplemented with 100 µg/ml of ampicillin, and induction and purification of the GST variants were performed as described by Johansson et al. (1999). Determination of specific activities of the purified GSTs was performed as described for the lysates. The substrate-activity profiles were expanded with three additional substrates, trans-stilbene oxide (tSBO) (Ivarsson et al., 2003), 2-cyano-1,3-dimethyl-1-nitrosoguanidine (cyanoDMNG) (Jensen and Stelman, 1987) and 4-chloro-7-nitro-1,2,3-benzoxadiazole (NBD-Cl) (Ricci et al., 1994).
The amino-acid sequences of GSTs in bacterial 34 clones, including the parental enzymes, were deduced from DNA sequences determined by using DYEnamic ET Dye Terminator Cycle Sequencing Kit for MegaBACE 1000 DNA Analysis Systems (GE Healthcare).
The molecular structures of the eight substrates used in screening of the library and the additional three substrates used in purified protein measurements were modeled and represented in energy-minimized conformations in H2O. The calculations were made in MMFF94x (Halgren, 1999) force field using Molecular Operating Environment (MOE) software (Chemical Computing Group Inc., Canada). All multivariate analyses were performed using mean values of duplicate measurements. Throughout the whole analysis, the measured reaction rates were used without subtraction of the non-enzymatic contribution, which was relatively small for measurements with high-enzyme activities. Subtraction of the uncatalyzed reaction rates was undesirable, since it gave rise to a large number of negative values. This occurred because of the experimental scatter and a large proportion of bacterial lysates with activities not significantly different from zero. However, analyses were also made on data corrected for the non-enzymatic contribution, and the overall results were similar to those presented, even though the assignments of some individual clones were changed.
With the exception of the initial scatter plots of untransformed data, all rows in the data matrix, corresponding to different GST variants, were scaled to unit length. Furthermore, the eight columns, corresponding to the alternative substrates, were individually normalized to a mean value of zero and unit variance. In order to evaluate any bias arising from the scaling of rows with uniformly low activities, a comparative analysis was made between the complete data set and a subset restricted to the rows of the original untransformed matrix in which at least one substrate gave activity in the upper 30% percentile. The restricted data set contained 62% of the original enzyme variants and was analyzed in the scaled and normalized form. Multivariate analysis showed that the cluster assignments were essentially unaffected by elimination of the clones with no or low activity. In particular, the 34 sequenced clones (see Results) were found in the corresponding clusters as in the multivariate analysis of the complete data set. The software used was SIMFIT (W.G. Bardsley, http://www.simfit.man.ac.uk, University of Manchester, UK) and SIMCA (Umetrics, Umeå, Sweden).
Experimental error in the screening of lysatesand assays of purified proteins
GST activities of the lysates were measured in duplicate in order to check the precision of the measurements. For the entire data set, the medians of the relative standard deviations of the activities were <14% for each substrate. The substrates of special significance for the cluster assignments showed lower relative standard deviations: pNPA (1.4%), EPNP (8.9%) and NCH (2.3%). If inactive clones were disregarded, the relative standard deviations were <8%. The day-to-day variability of the measurements in lysates was determined with GST M1-1 and GST M2-2 by assays with the same substrate on eight different days. The standard error for these measurements was <24% for all substrates. The relative standard deviation of assays with purified protein variants was <9%. EPNP and pNPA are substrates showing low relative standard deviation in purified protein fractions, 4.4% and 3.8%, respectively, and the value for NCH was 8.3%.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
Recombinant GST M1/M2 mutants catalyze both substitution and addition reactions
The GST M1-1 and GST M2-2 catalyze the nucleophilic attack of GSH on electrophilic centers in a variety of organic molecules, but differ markedly in their substrate selectivity profiles. It is particularly noteworthy that the activities are not restricted to one single class of chemical reactions, but comprise both addition and substitution reactions. The probing of substrate-activity space therefore also includes investigation of alternative chemical mechanisms.
Fig. 1 shows substrate structures and the electrophilic sites undergoing nucleophilic attack by the sulfur of GSH. CDNB, pNPA and MCB react with GSH via substitution reactions, whereas NPG, tPBO, EPNP, NCH and PEITC are conjugated through addition reactions. Both GST M1-1 and GST M2-2 have high activity with CDNB, but the specific activity of GST M2-2 is three times higher than the specific activity of GST M1-1 (Ivarsson et al., 2003). The GSH-dependent release of p-nitrophenol from pNPA (Keen and Jakoby, 1978) represents a transacylation reaction. The third substitution reaction involves MCB, a fluorogenic substrate of both GST M1-1 and GST M2-2 (Eklund et al., 2002). The NPG and EPNP are epoxides, and the opening of the oxirane ring occurs via nucleophilic addition to either of the oxirane carbons. tPBO is an enone to which GSH adds via the ß-carbon. NCH is a novel GST substrate in which GSH is added to the double bond of the cyclohexene ring. The last substrate of the addition reactions is PEITC, an isothiocyanate that forms a dithiocarbamate with GSH. These eight substrates were all used for screening of the library. For the characterization of purified enzymes (see below), the activities were also measured with NBD-Cl (aromatic substitution), cyanoDMNG (transnitrosylation) and tSBO (epoxide addition).
|
Activities of GSTs from the GST M1/M2 library demonstrate divergent functional properties among the mutants
The two parental GSTs and 384 clones, randomly chosen from the mutant library, were assayed in bacterial lysates with the eight electrophilic substrates. The rosters of eight-dimensional substrate-activity data for the 386 GST variants were first examined by scatter plots. Projections in two or three dimensions indicated multiple correlations between some of the substrates (Fig. 2). For example, CDNB versus NPG or EPNP versus PEITC suggested subgroupings radiating in different directions. In contrast, pNPA versus EPNP did not show such well-defined divergent patterns. In three dimensions, CDNB, NPG and EPNP strongly indicated three diverging subdivisions of the activity values (Fig. 2D).
|
Pair-wise plots of rosters of multiple functional values can be used to distinguish clones of dissimilar GSTs and identify those with similar properties
A sensitive test for functional similarities of two enzyme variants is the plot of the functional parameter values of the two variants against one another. The parental enzymes GST M1-1 and GST M2-2 clearly have divergent specific activities with the alternative substrates (Fig. 3A). GST M2-2 and variant 74 show strong similarities with seven substrates (Fig. 3B), but differ with respect to the eighth (MCB). If two clones contain identical enzyme variants, the coordinates will fall along a straight line, and the slope will be an estimate of the relative amounts of active enzyme in the two samples assayed. For example, variants 72 and 119, which represent identical proteins, as evidenced by their amino-acid sequences (see below), are expressed at different levels: the slope of the regression line (R2 = 0.941, P = 0.0001, testing the probability that the rows are not correlated; not depicted) shows that the lysate containing variant 72 holds 2.1 ± 0.2 times more active enzyme than does the lysate containing variant 119 (Fig. 3C). When the corresponding vectors of data are scaled to unit length (see the following paragraph), the respective parameter values of two functionally identical variants should fall along a straight line through the origin with a slope equal to 1.0. Variants 71 and 72 are such examples of catalytically identical GSTs, as judged by analysis of the eight parameters values (Fig. 3D).
|
The functional properties of variant GSTs display clustering in multidimensional substrate-activity space
The primary analysis gave clear indications for subgroups of the data, but further refinement was made in order to avoid bias that can be introduced by variations in the relative amounts of enzyme in the lysates analyzed, as well as by differences in the magnitudes of the activities measured with the alternative substrates. Therefore, each row in the data matrix was scaled to a vector of unit length. By this scaling, all clones that produce functionally identical GSTs are treated as if they were expressing the same amount of active enzyme. Further, the eight columns of activities in the data matrix were individually normalized to a mean value of zero and unit variance. This normalization allows all substrates to influence the analysis of the total data set to the same extent. All subsequent analyses of cell lysates were made with the transformed data.
The transformed data could be divided into four divergent clusters based on the Euclidian distances among the GST variants in eight-dimensional functional space. Fig. 4 shows the results of a K-means analysis under the assumption of four nodes. CDNB versus pNPA (Fig. 4A), NPG versus EPNP (Fig. 4B) and CDNB versus NCH (Fig. 4C) are three of the 28 possible two-dimensional projections of the eight-dimensional substrate-activity data. The analysis shows two clusters encompassing the parental GST M1-1 (58 variants colored blue) and GST M2-2 (31 variants colored red). In addition, two diverging clusters could be identified. One of them (253 variants colored brown) is formed by GST mutants with low absolute values of catalytic activities overall. The scaling of activity vectors to unit length makes the activities appear relatively high, but if the data matrix is transformed only by normalizing the columns to unit variance and a zero mean, the brown cluster is centered at the origin, as expected. Disregarding this cluster of ‘null’ mutants, the centroid of the second novel cluster (containing 44 variants colored yellow) is the highest on the pNPA, EPNP and NCH axes (Fig. 4), and intermediate between the centroids of the M1-like and M2-like clusters on the CDNB and NPG axes. High relative activities with pNPA, EPNP and NCH in comparison with other activities appear to characterize members of both the yellow and the brown clusters. However, the NCH activity is primarily linked to the brown cluster (Fig. 4C), and the pNPA and the EPNP activities are the hallmarks of the yellow cluster. As in the K-means cluster analysis, a similar division of the data into a minimum of four groups was indicated by hierarchical analysis based on dendrograms (not shown), using a number of alternative distance measures and group linkages. Indications for further subgrouping of the data needed corroboration by a larger data set. Furthermore, the structural identification of some identical enzyme variants, which had to be joined, set a limit to the maximum number of clusters.
|
Principal component analysis
A principal component (PC) analysis showed the dominating functional relationships among the M1/M2 library mutants in a PC1–PC2 score plot (Fig. 5A). The data points are colored on the basis of the previous clustering in the K-means analysis (Fig. 4). The two methods of analysis give consistent groupings of the variants. The parental enzymes GST M1-1 and GST M2-2 are located in different quadrants and segregate with variants in the two dominating swarms of the data points. The two-dimensional projection, of the eight-dimensional PC scores, in the PC1–PC2 plane (Fig. 5A) suggests that, in addition to the M1-like and M2-like clusters diverging in PC2, a third cluster (colored yellow) is emerging in between the parental enzymes. In addition, a fourth cluster (brown) encompasses mutants with low activities. The first three eigenvalues account for 45, 27, and 9%, respectively, of the variability in the PC analysis. Examination of PC3 (not shown) demonstrates that the members of the yellow novel cluster diverge in a direction distinct from those of the parental distributions. In addition, the yellow and brown clusters clearly segregate in this dimension. The loading plot (Fig. 5B) suggests that EPNP and pNPA are the substrates contributing particularly prominently to forming the yellow cluster, as indicated by loadings 1 and 3 (not shown), whereas the other substrates primarily are responsible for the separation of the M1- and M2-like clusters. In PC1, the loading of NCH is similar to those of EPNP and pNPA (Fig. 5B). In PC2-PC3, however, the loading of NCH is distinctly separated into a different quadrant from the loadings of EPNP and pNPA. NCH is thus linked to the distribution of the brown cluster, whereas EPNP and pNPA are associated with the yellow cluster.
|
Canonical variate analysis
PCs are obtained by a distance-preserving rotation of the original axes of the substrate-activity space. In order to give emphasis to the groupings of the data, a canonical variate analysis was performed on the four clusters obtained by K-means analysis. The transformation underlying canonical variates best represents the Mahalanobis distance, which highlights differences between groups (Krzanowski, 2000). The canonical variate plot (Fig. 6) indeed underscores the distinction between the clusters. The separation is clear between the M1-like cluster (blue) and the M2-like cluster (red), but somewhat less distinct between the ‘null’ individuals (brown) and the novel distribution (yellow). The data were also analyzed under the assumption of three or five clusters (not shown). However, the hierarchical dendrograms analysis (mentioned above) strongly suggested a minimum of four clusters, and for lack of additional experimental data, the parsimonious choice of four was made. In an analysis in which the members of the brown cluster were eliminated, the remaining data still separated into the same M1-like, M2-like and yellow clusters (not shown).
|
DNA sequences of 34 clones demonstrate the structural diversity in the GST M1/M2 library
DNA sequences were determined for 34 variants in the GST M1/M2 library. The GST variants were chosen such that representatives of all four clusters were included (cf. Fig. 5). Comparison with the GST M1-1 and GST M2-2 sequences identified 23 different nucleotide segments in the variants that had been exchanged between the two parental sequences (data not shown). Owing to the degeneracy of the genetic code, the resulting amino-acid sequences contained 14 alternative segments (Fig. 7). The exact size of each segment exchanged remains undefined, because of the 89% sequence identity of GST M1-1 and GST M2-2 at the DNA level; the sequences separating the exchanged segments are identical and therefore could derive from either GST M1-1 or GST M2-2.
|
Fig. 7 displays a schematic alignment of the exchanged segments of parental amino-acid sequences. Arrows indicate the location of the four regions of primary structure that form the binding site for the electrophilic substrate. Squares mark two segments containing active-site residues that differ between GST M1-1 and GST M2-2 and circles mark the other segments. Deviations from both parents were noted in the form of spurious amino-acid substitutions indicated by vertical bars. The average number of mutations was about two per DNA sequence, of which one was silent and the other was altering the encoded amino acid. A noteworthy modification is the Q110R point mutation, which appears in 14 clones all containing a segment originating from the parental GST M2-2. Residue 110 is located in the same segment as the catalytically important Tyr116, but is still distant from the active site and does not appear to influence the catalytic properties. The structural diversity of the library is quite high. Out of the 34 variants, 30 had unique DNA sequences; only clones 72 and 119 as well as clones 284 and 383 had exactly the same nucleotide sequence. Four pairs of identical protein variants appeared when the codons were translated: clones 23 and 295, clones 6 and 276, clones 161 and 320, as well as clones 141 and 245. The other sequences translated into different protein variants.
The brown cluster, containing low-activity GSTs, was represented by five variants: 185, 95, 376, 297 and 140. Their sequences did not reveal any strikingly similar traits. Clone 140 has seven point mutations, including the start codon, which may compromise its expression. The M1-like blue cluster contributed nine variants to the sequence analysis: 72, 119, 23, 295, 71, 291, 12, 48 and 275. With the exception of 48 and 275, they were dominated by segments derived from GST M1-1. All enzymes other than variant 48 contained Ser in position 210. Variant 48 had Thr in position 210, but contained the centrally located active-site segment distinguishing GST M1-1 and GST M2-2. The yellow cluster was represented by ten variants: 161, 320, 141, 245, 383, 284, 343, 240, 7 and 28. Excepting variant 28, their primary structures were dominated by segments from GST M1-1, but in contrast to members of the M1-like cluster they contained Thr in position 210. The deviating structure, 28, was similar to the sequence of GST M2-2, but in distinction from the latter contained Ser in position 210. Finally, ten variants from the M2-like red cluster were sequenced: 286, 342, 168, 34, 74, 6, 276, 8, 5 and 31. They are composed primarily of segments from GST M2-2 and have either Thr or Ser in position 210. Instead, they all without exception contain the central active-site segment distinguishing the parental enzymes, and in this respect mimic GST M2-2 (black square in Fig. 7). As mentioned above, clones 72 and 119 are identical at the protein level, but they are also identical with the parental GST M1-1. In contrast, there are no clones analyzed that are structurally identical with the parental GST M2-2.
Purification and characterization of selected GST variants
After screening of the 384 GST mutants from the M1/M2 library, five variants (72, 74, 295, 342 and 383), together with the parental enzymes GST M1-1 and GST M2-2, were selected for purification. These five variants were chosen to represent the three clusters recognized among active library individuals (Fig. 5) in order to compare their substrate selectivity profiles. Variants 72 and 295 belong to the M1-like cluster (blue) and variant 74 belongs to the M2-like cluster (red), while variant 383 is a member of the third distribution of active mutants (yellow). Variant 342 clustered with the M2-like GSTs by the K-means and dendrograms analyses, but clearly appeared close to variant 383 of the yellow cluster in dimensions three and four of the PC analysis. This ambiguity apparently arises because variant 342 is in the region between the red and yellow clusters in the original eight-dimensional substrate-activity space. In the PC coordinates, the major variability (81%), also including correlations between activities with different substrates, is accounted for in the first two dimensions. The higher dimensions three (9%) and four (8%) can then display more weakly contributing, but still distinguishing, properties. In the present analysis, they signify a distribution of activities diverging in a new direction. The functional properties of variant 342 were undoubtedly most similar to those of variant 383 (see below).
The purified GST variants were assayed with eleven substrates (Table I). Specific activities were determined with NBD-Cl, cyanoDMNG and tSBO, in addition to the eight substrates used in the screening. NBD-Cl and cyanoDMNG are characteristic GST M2-2 substrates, showing 5 and 200 times, respectively, higher activities with GST M2-2 than with GST M1-1. On the other hand, tSBO is a GST M1-1 distinctive substrate, giving 600 times higher activity with GST M1-1 than with GST M2-2. GST variants belonging to the M1-like cluster (variants 72 and 295) display higher specific activity with tSBO than with the other two substrates. On the other hand, variant 74 typifying the M2-like cluster has low-specific activity with tSBO, but high-specific activity with cyanoDMNG and NBD-Cl (Table I). The remaining variants, 342 and 383, both clearly deviate in their substrate-selectivity profile from the parent-like individuals, having pronounced specific activities with NBD-Cl but not with cyanoDMNG or tSBO (Table I). Variant 383 belongs to the yellow cluster of mutants with novel activity profiles recognized in multivariate analysis (Figs 4 and 5), and this mutant definitely displays distinguishing properties also in the purified form when characterized by activities with parent-specific substrates. Variant 342, in spite of being assigned to the M2-like cluster by the K-means analysis (Fig. 4), displays essentially the same specific activity profile as variant 383 (Table I), at least as a first approximation. This finding is in agreement with the PC analysis, which demonstrates, in the higher principal components PC3 and PC4 (not shown), a clear association between variants 342 and 383 as well as with other mutants of the yellow cluster.
|
Substrate-selectivity profiles for purified members of different clusters
The fractions of specific activity for the 11 substrates in relation to the sum of all the specific activities are given as illustrations of the profiles in ‘exploded doughnut’ representation (Fig. 8). All of the seven purified GST variants share high activities with CDNB. The M1-like variants 72 and 295 display a second high activity with PEITC, whereas the second highest activity for the M2-like variants is found with cyanoDMNG. Variants 342 and 383 are both markedly dominated by the CDNB activity, indicating that no alternative substrate tested gives an activity of the same magnitude as the CDNB activity. In particular, these mutants have suppressed activities with PEITC and cyanoDMNG, as compared to the respective parental GSTs. On the other hand, the members of the yellow cluster are also characterized by the emerging activities with pNPA and EPNP (cf. Fig. 4), even though the specific activities with these substrates are relatively low in absolute numbers. The substrate-activity profiles of GST M1-1 and variants 72 and 295 are very similar, as also concluded from measurements in crude lysates. GST M1-1 and variant 72 have the same amino-acid sequence and are expected to have identical properties within the experimental variance. They differ from variant 295 only by the amino-acid residue in position 16. GST M2-2 and variant 74 are also functionally highly similar, in agreement with their major structural similarities (Fig. 7), but variant 74 shows decreased CDNB activity, and increased MCB and cyanoDMNG activities in comparison with GST M2-2.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
Expressivity as a variable in protein engineering
In the search for recombinant proteins with enhanced properties, mutants with increased expressivity may confound the identification of proteins with intrinsically altered functions. For example, mutations of a ß-lactamase that provide bacteria with enhanced resistance to an antibiotic may result in both increased production of the enzyme protein and enhanced catalytic efficiency of the enzyme (Ness et al., 2000). Direct monitoring of the expression of the enzyme protein could resolve this issue, but suitable molecular identifiers, such as intrinsic chromophores or engineered labels, may not be available. Besides, the added task of measuring the absolute levels of the proteins could limit the search for mutants with targeted properties in large mutant libraries.
In an earlier study of the GST M1/M2, library cluster assignments were made by inspection of distributions in a PC analysis, and only GST variants with relatively high activities were annotated as members of the three clusters of active enzymes (Emrén et al., 2006). The input data were measurements of enzyme activities in crude lysates of bacteria in which the enzymes had been expressed. An enzyme produced in small amounts could therefore inappropriately have been regarded as essentially inactive, in spite of the fact that its intrinsic properties may be similar to enzymes in one of the other groupings. In a similar manner, clones may be misassigned when their expression levels are markedly higher than those of the majority in the cluster to which they properly belong. The present investigation demonstrates how the bias caused by differences in protein concentration can be circumvented and the analysis improved.
Enzymes represented as vectors in functional space
Eleven alternative substrates were used to characterize GST variants, eight in the initial screening and three additional substrates in the assay of purified enzymes. Each individual in the mutant population can thus be represented by a point, or vector, in multidimensional substrate-activity space. In enzymes with broad substrate specificities, such as GSTs (Norrgård et al., 2006) and cytochrome P450s (Otey et al., 2004; Taly et al., 2007), mutations often differentially affect the catalytic efficiencies with alternative substrates. This feature provides a tool for distinguishing emerging mutants with enhanced (or decreased) activity with a chosen substrate from those that are simply characterized by altered expression of the protein. In the case of a pure effect on expression, all measurable catalytic activities should change proportionally in a unique direction in multidimensional substrate-activity space, given a linear relationship between activity and enzyme concentration. In contrast, if the intrinsic properties of the enzyme have changed by mutations, the substrate-activity vector will deviate from that of the parental enzyme.
In the data set of variant GST activities in bacterial lysates, the length of the vector in the multidimensional substrate-activity space will depend on the amount of enzyme sampled and assayed. Silent mutations in the coding sequence could markedly alter the expression levels of GSTs (Mannervik, 2005). It is also possible that different yields of enzyme are obtained in the different preparations of bacterial lysates. Consequently, separate clones of the same enzyme variant could be located at unequal distances from the origin, even though the vectors have the same direction in multidimensional space. Fig. 9 sketches this situation with three clones (rendered in red) having the same GST protein sequence and two additional clones with diverging vectors (black). Experimental data obviously are subject to experimental error such that the vectors of the same protein sequence do not overlap exactly, but fall within a ‘cone-shaped’ boundary. In the actual GST M1/M2, library variants 72 and 119 are sequence identical to the parental GST M1-1. As expected, their vectors coincide along a line in substrate-activity space as evidenced by the linear relationship between their respective parameter values in eight-dimensional space (cf. Fig. 3C). In principle, mutations that give a proportional change of all activities tested could be an alternative explanation of coinciding vectors of different lengths. However, in the present case, quantification of the GST protein by immunoblots with cross-reactive anti-GST M2-2 antibodies (Hao et al., 1994) verified the different expression levels in the lysates of variants 72, 119 and GST M1-1 (data not shown).
|
Contributions of the alternative substrates to the functional clustering of the GST variants
A notable result of the PC analysis was that pNPA, EPNP and NCH together, but separate from all other substrates, distinguish the novel yellow and brown distributions from the parental ones. The effect of NCH was most closely related to the brown cluster of variants with low or zero activity (Fig. 5B), and the association of NCH with the yellow and the brown clusters was not evident unless the rows of the data matrix were scaled to unit length. Apparently, the low-enzyme activities with NCH in comparison to the non-enzymatic reaction make the scaling of rows to exaggerate the importance of NCH for the identification of GST variants with novel activities.
The chemical reactions of pNPA and EPNP are of different types: pNPA is transformed via a transacetylation reaction to form S-acetylglutathione and 4-nitrophenol, whereas EPNP is subject to a GSH-addition reaction involving opening of the oxirane ring, and NCH undergoes a Michael addition. The structures of pNPA and EPNP share a 4-nitrophenoxy moiety, and the reaction with GSH occurs at approximately the same site in the second moiety of the molecules (Fig. 1). In view of the similar contributions of pNPA and EPNP in the PC analysis, we have considered the possibility that EPNP would rather react in a substitution reaction liberating 4-nitrophenol and S-2,3-epoxypropylglutathione, thus mimicking the pNPA reaction. However, HPLC analysis of the reaction products did not give any evidence for such an alternative reaction with EPNP (data not shown). In spite of the apparently similar effects of these two substrates on PC1 and PC2 (Fig. 5), the higher dimensions PC3 and PC4 and the corresponding loading plots (not shown) demonstrate that the effects of EPNP and pNPA are diverging, suggesting a further subdivision of the yellow distribution.
The other substrates used in the screening are also correlated and form groups in the loading plots (cf. Fig. 5B). MCB and CDNB are positively contributing to formation of the M2-like cluster, and both substrates undergo substitution reactions (Fig. 1). NPG and PEITC form a third group in the loading plot (Fig. 5B), and tPBO is most closely associated with these latter substrates, which all three positively contribute to forming the M1-like cluster. These latter substrates all undergo alternative addition reactions and have similarities in size and structure (Fig. 1). Nevertheless, both of the two major types of reaction mechanisms, substitution and addition, are catalyzed by members of all clusters, and these separate reaction types are therefore not fundamental in dividing the GST variants into different distributions.
The influence of data transformation
It is common practice in multivariate analysis to standardize data in order to handle variables of different dimensions. In the present data set, groupings signifying at least three diverging distributions can be recognized even in the absence of any transformation (Fig. 2). The yellow cluster identified in the K-means and PC analyses is represented by points in the elongated distribution with low PEITC and high EPNP activities in Fig. 2C, and in the distribution with low CDNB and NPG activities and high EPNP activity in Fig. 2D.
Normalizing the activities to unit variance for each substrate was previously shown to provide a clear separation of different clusters (Emrén et al., 2006), but the issue of different enzyme concentrations was not addressed. In the present study, the data were first transformed by the previous normalization, and subjected to K-means and PC analyses (not shown). In this manner, the cluster assignments were accomplished by the algorithm rather than by the investigator. However, scaling of the activities of all enzyme variants to vectors of unit length prior to normalizing the columns to unit variance allowed us to identify variants that apparently were misassigned, because of their low concentration in the bacterial lysates analyzed. The 291 ‘null’ mutants of the analysis of the normalized data were redistributed after the additional scaling (performed before the normalization of columns), such that 15 apparent ‘null’ mutants were assigned to the new yellow cluster, 17 to the M1-like blue cluster and seven to the M2-like red cluster. The remaining 252 were found in the new brown ‘null’ cluster, which in addition received one mutant from the 21 members of the first M2-like distribution. No further reassignments of mutants in the first M2-like cluster were made. The 15 members of the first M1-like distribution were all present in the new blue M1-like cluster. The most striking redistribution of variants was noted in the first ‘novel’ cluster consisting of 59 GSTs with altered substrate-activity profiles. The new yellow cluster retained 29 of these variants, but 26 were reassigned as M1-like and four as M2-like. Thus, half of the first ‘novel’ cluster members had intrinsic properties that were obviously more similar to those of GST M1-1 or GST M2-2. The assignments based on the double transformation were in better agreement with the sequences determined (Fig. 7).
Any transformation of the data may cause bias in the analysis, and different data treatments could serve different purposes. It could be argued that the scaling of rows to vectors of unit length may give unjustified weight to weakly active enzymes. However, an enzyme with low activity in all substrate dimensions will not show any of the distinctive patterns that characterize different clusters, but owing to experimental error will display a random scatter in all dimensions of activity space. The present study shows that the majority of the original ‘low-activity’ variants are still ‘null’ mutants in the brown cluster, even if their activities with different substrates appear exaggerated (cf. Fig. 4). The soundness of the double transformation in the present study is supported by the finding that five sequenced GSTs of the original ‘novel’ cluster (compared with cluster assignment of data only normalized with respect to the eight substrates) were reassigned to the M1-like (variants 119, 48 and 275) and M2-like clusters (variants 286 and 168), in improved accord with their primary structures.
It should be emphasized that the current multivariate analysis does not attempt a definitive cluster assignment of every enzyme variant. Obviously, not only the different transformations, but also the composition and size of the data set analyzed will influence the outcome. Furthermore, the vectors of experimental substrate-activity data, as well as the corresponding quasi-species, are stochastic variables subject to variance. The main result of the analysis was that a minimum of three divergent distributions, or quasi-species, can be identified among the active members in the GST M1/M2 library. The double transformation facilitated a more accurate assignment of differentially expressed enzymes. In comparison with the more subjective cluster assignments (Emrén et al., 2006), it is noteworthy that the results of the two approaches are in general agreement for GST variants expressed at high levels. However, a number of mutants present in low concentrations in the bacterial lysates had to be reassigned. It should also be noted that the higher dimensions of the PC analysis were most reliable in the assignment of some variants with novel properties, as exemplified with variant 342 (cf. Emrén et al., 2006).
Linking primary structure to functional changes
The clustering of the GST M1/M2 variants was based entirely on their functional properties without any prior information about their amino-acid sequences. In order to explore to what extent the clustering was a reflection of similarities in structure, 34 enzyme variants were chosen for sequence analysis. Pair-wise comparison of the activity data (exemplified in Fig. 3) made it possible to evaluate the effects of segment exchanges or spurious point mutations in variants that have similar primary structures. In this manner, it can be concluded that in the GST M1-1 sequence the mutations of Ala16 into Ser, Ala67 into Thr or Asn173 into Ser are without effect on the catalytic properties. The spurious mutations in, e.g., clones 284, 343 and 383 also appear to be without functional consequence. In general, the comparison of functional data and amino-acid sequences do not indicate that the spurious mutations have had any noteworthy effect on the cluster assignments of catalytically active GST mutants.
The GST mutants assigned to the brown cluster have low, or impaired, activity with the substrates tested. Since inactive proteins can derive from a variety of modifications of any structure, this cluster should be expected to be highly heterogeneous and lack unifying structural characteristics. The other three clusters, as a first approximation based on the 34 sequences determined, can be characterized by two features, one is the parental GST scaffold, and the second is the nature of active-site residue 210 (C-terminal square in Fig. 7). Thus, the members of the M1-like blue cluster have high structural similarity to GST M1-1, and Ser in position 210 (with one exception). The variants in the M2-like red cluster are highly similar to the parental GST M2-2, and all contain the centrally located segment distinguishing the active sites of GST M1-1 and GST M2-2 (central square in Fig. 7). The members of the red cluster have either Thr or Ser in position 210. The majority of enzymes in the novel yellow cluster appear to be characterized by a GST M1-1-like structure with Thr instead of Ser in position 210. Even though some key features in the structure are recognized as identifiers of the three clusters of active mutants, it should be borne in mind that the library could possibly be further subdivided into additional clusters with other functional and structural characteristics. It may also contain outliers that cannot readily be classified with other mutants.
A full agreement between cluster assignments and structural traits cannot be expected. Even though in our study, the protein sequence analysis is largely in accord with cluster assignments based on the functional characterization. On the other hand, the parental amino-acid sequences are 85% identical, and only modest variations could be predicted for a single round of recombinations. Therefore, it is reasonable to find the members of the yellow cluster with the largest number of peptide segments from either one or the other of the parental structures GST M1-1 or GST M2-2. Nevertheless, the majority of the members of the yellow distribution are structurally most similar to GST M1-1, some of them differing only in one or a few amino acids. The locations of the mutations characterizing variant 383 are shown in a three-dimensional model of the parental GST M1-1 subunit (Fig. 10).
|
The emergence of molecular quasi-species
The evolution of novel proteins with new functions in biological systems is largely based on remodeling of pre-existing structural elements (Orengo and Thornton, 2005; Aravind et al., 2006; Choi and Kim, 2006). At the genomic level, this is accomplished by fusions and recombinations of nucleotide sequences accompanied by stochastic point mutations. Similar processes are expedient also in protein engineering. It has been recognized that populations of mutants can form clusters of similar sequences, called molecular quasi-species (Eigen et al., 1988) that are superior evolving entities in comparison with single individuals. This paradigm can also be applied to catalytic properties of evolving proteins (Larsson et al., 2004; Emrén et al., 2006).
The functional properties of the GST mutants in the M1-like and M2-like clusters are largely similar to those of the parental enzymes, which are well characterized (Hansson et al., 1999b). In some cases, the M1-like variants are indistinguishable in their catalytic properties within the experimental variance. This is in excellent agreement with the fact that some of them, e.g., 23, 295, 71 and 291, are merely altered by single point mutations superficially located and distant from the active site of GST M1-1. This concordance validates the independent cluster assignments based on multivariate analyses of functional data. The blue and red quasi-species are thus basically clusters of GSTs varying modestly in both structure and function from their parents GST M1-1 and GST M2-2, respectively. The third quasi-species is functionally characterized by enhanced activities with pNPA and EPNP. This distinction is evident in the K-means cluster analysis (Fig. 4), and even more obvious in the higher dimensions of the PC analysis. In addition, the activities with signature substrates of the parental GSTs are suppressed, e.g., the tSBO and cyanoDMNG activities (Table I). It would appear that the formative attributes in the present case involve a combination of enhanced and suppressed properties.
Residue 210 is a pivotal determinant in the functional clustering of the GST M1/M2 variants
A particularly noteworthy finding was that most of the variants in the yellow cluster had M1-like structures, in which the active-site residue Ser210 is replaced by Thr. This minor structural difference of one methylene group has previously been shown to cause dramatic alterations of the catalytic properties in the Mu class GSTs (Norrgård et al., 2006). Further variations from the parental amino acids in position 210 were not found among the sequenced mutants from the library.
Residue 210 is particularly important with respect to the activity with epoxide substrates (Ivarsson and Mannervik, 2005). GST M1-1, as well as GST M2-2 with a Thr210Ser mutation, has high activity with epoxides, but low activity with the same substrates when Ser is changed into Thr (Ivarsson and Mannervik, 2005). The activities of the sequenced mutants with NPG and tSBO are in general agreement with these previous findings (Table I). The epoxide EPNP is an exception, in affording good activity also with Thr in position 210. The activity with the organic isothiocyanate PEITC is clearly also dependent on Ser in the active site, as evidenced by comparison of GST M1-1 and variant 383 (Table I).
In the naturally occurring Mu class GSTs, the residue in this position has been identified as hypervariable and undergoing positive selection in evolution (Ivarsson et al., 2003; Norrgård et al., 2006). The present results, from a completely different approach, underscore the functional importance of residue 210 and suggest that, in the Mu class GSTs, mutation of this residue may provide access to altered activity profiles by evolution.
In summary, the present investigation provides an approach to handling the problem of varying expressivity among mutants in the mining of an enzyme library for novel substrate-activity profiles. The multivariate analysis of the experimental data established that the crossing of GST M1-1 and GST M2-2 leads to a minimum of three distributions of functional recombinants, which can be regarded as molecular quasi-species. Assays with multiple alternative substrates provide substrate-activity signatures characterizing members of the different distributions. The sequence analyses of representatives of the distributions corroborate the conclusion drawn from functional studies.
Color versions of Figures 1, 4, 5, 6, 8, 9 and 10 are available online at http://proeng.oxfordjournals.org.
|
Footnotes |
|---|
1 Authors contributed equally to the work.
|
Acknowledgements |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
We are grateful to Dr W.G. Bardsley, Manchester, UK for generously giving guidance in the data analysis and providing subroutines in Simfit for our needs. We also thank Birgit Olin for advice on protein purification. The Swedish Research Council, the Swedish Cancer Society, and the Carl Trygger Foundation supported this work.
|
References |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsReferences |
|---|
Aharoni A., Griffiths A.D., Tawfik D.S. Curr. Opin. Chem. Biol. (2005) 9:210–216.[CrossRef][Web of Science][Medline]
Anantharaman V., Aravind L., Koonin E.V. Curr. Opin. Chem. Biol. (2003) 7:12–20.[CrossRef][Web of Science][Medline]
Aravind L., Iyer L.M., Koonin E.V. Curr. Opin. Struct. Biol. (2006) 16:409–419.[CrossRef][Web of Science][Medline]
Armstrong R.N. Chem. Res. Toxicol. (1997) 10:2–18.[CrossRef][Web of Science][Medline]
Bloom J.D., Meyer M.M., Meinhold P., Otey C.R., MacMillan D., Arnold F.H. Curr. Opin. Struct. Biol. (2005) 15:447–452.[CrossRef][Web of Science][Medline]
Choi I.G., Kim S.H. Proc. Natl Acad. Sci. USA (2006) 103:14056–14061.
Comstock K.E., Widersten M., Hao X.-Y., Henner W.D., Mannervik B. Arch. Biochem. Biophys. (1994) 311:487–495.[CrossRef][Web of Science][Medline]
Eigen M., McCaskill J., Schuster P. J. Phys. Chem. (1988) 92:6881–6891.[CrossRef][Web of Science]
Eklund B.I., Edalat M., Stenberg G., Mannervik B. Anal. Biochem. (2002) 309:102–108.[CrossRef][Web of Science][Medline]
Emrén L.O., Kurtovic S., Runarsdottir A., Larsson A-K., Mannervik B. Proc. Natl Acad. Sci. USA (2006) 103:10866–10870.
Farinas E.T., Bulter T., Arnold F.H. Curr. Opin. Biotechnol. (2001) 12:545–551.[CrossRef][Web of Science][Medline]
Habig W.H., Pabst M.J., Jakoby W.B. J. Biol. Chem. (1974) 249:7130–7139.
Halgren T.A. J. Comput. Chem. (1999) 20:730–748.[CrossRef][Web of Science]
Hansson L.O., Bolton-Grob R., Massoud T., Mannervik B. J. Mol. Biol. (1999a) 287:265–276.[CrossRef][Web of Science][Medline]
Hansson L.O., Bolton-Grob R., Widersten M., Mannervik B. Protein Sci. (1999b) 8:2742–2750.[Web of Science][Medline]
Hao X.-Y., Castro V.M., Bergh J., Sundström B., Mannervik B. Biochim. Biophys. Acta (1994) 1225:223–230.[Medline]
Ivarsson Y., Mannervik B. Protein Eng. Des. Sel. (2005) 18:607–616.
Ivarsson Y., Mackey A.J., Edalat M., Pearson W.R., Mannervik B. J. Biol. Chem. (2003) 278:8733–8738.
Jensen D.E., Stelman G.J. Carcinogenesis (1987) 8:1791–1800.
Johansson A-S., Bolton-Grob R., Mannervik B. Protein Expr. Purif. (1999) 17:105–112.[CrossRef][Web of Science][Medline]
Josephy P.D., Mannervik B. Molecular Toxicology (2006) New York: Oxford University Press, Inc.
Keen J.H., Jakoby W.B. J. Biol. Chem. (1978) 253:5654–5657.
Kolm R.H., Danielson U.H., Zhang Y.S., Talalay P., Mannervik B. Biochem. J. (1995) 311:453–459.[Web of Science][Medline]
Koonin E.V., Mushegian A.R., Tatusov R.L., Altschul S.F., Bryant S.H., Bork P., Valencia A. Protein Sci (1994) 3:2045–2054.[Web of Science][Medline]
Krzanowski W.J. Principles of Multivariate Analysis - A Users's Perspective (2000) New York: Oxford University Press, Inc.
Larsson A-K., Emrén L.O., Bardsley W.G., Mannervik B. Protein Eng. Des. Sel. (2004) 17:49–55.
Mannervik B. Methods Enzymol. (2005) 401:254–265.[Web of Science][Medline]
Meyer M.M., Hochrein L., Arnold F.H. Protein Eng. Des. Sel. (2006) 19:563–570.
Ness J.E., Welch M., Giver L., Bueno M., Cherry J.R., Borchert T.V., Stemmer W.P., Minshull J. Nat. Biotechnol. (1999) 17:893–896.[CrossRef][Web of Science][Medline]
Ness J.E., Del Cardayre S.B., Minshull J., Stemmer W.P. Adv. Protein Chem. (2000) 55:261–292.[Web of Science][Medline]
Norrgård M.A., Ivarsson Y., Tars K., Mannervik B. Proc. Natl. Acad. Sci. USA (2006) 103:4876–4881.
Orengo C.A., Thornton J.M. Annu. Rev. Biochem. (2005) 74:867–900.[CrossRef][Web of Science][Medline]
Otey C.R., Silberg J.J., Voigt C.A., Endelman J.B., Bandara G., Arnold F.H. Chem. Biol. (2004) 11:309–318.[CrossRef][Web of Science][Medline]
Ricci G., Caccuri A.M., Lo Bello M., Pastore A., Piemonte F., Federici G. Anal. Biochem. (1994) 218:463–465.[CrossRef][Web of Science][Medline]
Shokeer A., Larsson A-K., Mannervik B. Biochem. J. (2005) 388:387–392.[CrossRef][Web of Science][Medline]
Svendsen A. Enzyme Functionality: Design, Engineering, and Screening (2004) New York: Marcel Dekker Inc.
Taly V., Urban P., Truan G., Pompon D. Biochim. Biophys. Acta (2007) 1770:446–457.[Medline]
Valetti F., Gilardi G. Nat. Prod. Rep. (2004) 21:490–511.[CrossRef][Web of Science][Medline]
White R.H. J. Bacteriol. (2006) 188:3431–3432.
Received March 12, 2007; revised March 12, 2007; accepted March 14, 2007.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||