Engineering aggregation-resistant proteins by directed evolution

A method was recently described for selecting aggregation-resistant antibody domains. A repertoire of domains displayed on filamentous bacteriophage were heated/cooled and selected for binding to the affinity ligand protein A specific for the folded domains. Here we describe a generalization of the method based on the selection for retained phage infectivity and for the binding of an appended sequence tag, and applicable to any protein displayable in multivalent form on phage.     

Use of a quantitative structure-property relationship to design larger model proteins that fold rapidly

A quantitative structure–property relationship (QSPR)was used to design model protein sequences that fold repeatedlyand relatively rapidly to stable target structures. The specificmodel was a 125-residue heteropolymer chain subject to MonteCarlo dynamics on a simple cubic lattice. The QSPR was derivedfrom an analysis of a database of 200 sequences by a statisticalmethod that uses a genetic algorithm to select the sequenceattributes that are most important for folding and a neuralnetwork to determine the corresponding functional dependenceof folding ability on the chosen attributes. The QSPR dependson the number of anti-parallel sheet contacts, the energy gapbetween the native state and quasi-continuous part of the spectrumand the total energy of the contacts between surface residues.Two Monte Carlo procedures were used in series to optimize boththe target structures and the sequences. We generated 20 fullyoptimized sequences and 60 partially optimized control sequencesand tested each for its ability to fold in dynamic MC simulations.Although sequences in which either the number of anti-parallelsheet contacts or the energy of the surface residues is non-optimalare capable of folding almost as well as fully optimized ones,sequences in which only the energy gap is optimized fold markedlymore slowly. Implications of the results for the design of proteinsare discussed.     

Conservation of residue interactions in a family of Ca-binding proteins

In the TNC family of Ca-binding proteins (calmodulin, parvalbumin,intestinal calcium binding protein and troponin C) {small tilde}70 well-conserved amino acid sequences and six crystal structuresare known. We find a clear correlation between residue contactsin the structures and residue conservation in the sequences:residues with strong sidechain–sidechain contacts in thethree-dimensional structure tend to be the more conserved inthe sequence. This is one way to quantify the intuitive notionof the importance of sidechain interactions for maintainingprotein three-dimensional structure in evolution and may usefullybe taken into account in planning point mutations in proteinengineering.     

Comparison of solvent-inaccessible cores of homologous proteins: definitions useful for protein modelling

The three-dimensional structures of 41 homologous proteins (belongingto eight families) were compared by pairwise superposition.A subset of "core" residues was defined as those whose sidechains have <7% of their surface exposed to solvent. Thissubset has significantly higher sequence identity and lowerroot mean square (RMS) carbon separation than for all topologicallyequivalent residues in the structure, when members of a proteinfamily are superposed. For such superpositions the relationshipbetween RMS distance and percentage sequence identity of thissubset of residues is similar to that for all equivalent residues,although some variation is observed between families of proteinswhich are predominantly ß sheet and those which aremainly helix. The definition of a structurally more conservedcore may be ueful in model building proteins from an homologousfamily. The RMS differences of coordinates of structures ofproteins with identical sequences are found to be related tothe resolutions of the structures.     

High solubility of random-sequence proteins consisting of five kinds of primitive amino acids

Searching for functional proteins among random-sequence librariesis a major challenge of protein engineering; the difficultiesinclude the poor solubility of many random-sequence proteins.A library in which most of the polypeptides are soluble andstable would therefore be of great benefit. Although modernproteins consist of 20 amino acids, it has been suggested thatearly proteins evolved from a reduced alphabet. Here, we haveconstructed a library of random-sequence proteins consistingof only five amino acids, Ala, Gly, Val, Asp and Glu, whichare believed to have been the most abundant in the prebioticenvironment. Expression and characterization of arbitrarilychosen proteins in the library indicated that five-alphabetrandom-sequence proteins have higher solubility than do 20-alphabetrandom-sequence proteins with a similar level of hydrophobicity.The results support the reduced-alphabet hypothesis of the primordialgenetic code and should also be helpful in constructing optimizedprotein libraries for evolutionary protein engineering.     

Amino acid alphabet size in protein evolution experiments: better to search a small library thoroughly or a large library sparsely?

We compare the results obtained from searching a smaller librarythoroughly versus searching a more diverse, larger library sparsely.We study protein evolution with reduced amino acid alphabets,by simulating directed evolution experiments at three differentalphabet sizes: 20, 5 and 2. We employ a physical model forevolution, the generalized NK model, that has proved successfulin modeling protein evolution, antibody evolution and T-cellselection. We find that antibodies with higher affinity arefound by searching a library with a larger alphabet sparselythan by searching a smaller library thoroughly, even with well-designedreduced libraries. We also find ranked amino acid usage frequenciesin agreement with observations of the CDR-H3 variable regionof human antibodies.     

Detection and reduction of evolutionary noise in correlated mutation analysis

Direct or indirect inter-residue interactions in proteins are often reflected by mutations at one site that compensate for mutations at another site. Various bioinformatic methods have been developed for detecting such correlated mutations in order to obtain information about intra- and inter-protein interactions. Here, we show by carrying out a correlated mutation analysis for non-interacting proteins that the signal due to inter-residue interactions is of similar magnitude to the 'noise' that arises from other evolutionary processes related to common ancestry. A new method for detecting correlated mutations is presented that reduces this evolutionary noise by taking into account evolutionary distances in the protein family. It is shown that this method yields better signal-to-noise ratios and, therefore, can much better resolve, for example, correlated mutations that reflect true inter-residue interactions.     

Engineering stability into Escherichia coli secreted Fabs leads to increased functional expression

The recombinant expression of immunoglobulin domains, Fabs and scFvs in particular, in Escherichia coli can vary significantly from antibody to antibody. We hypothesized that poor Fab expression is often linked to poor intrinsic stability. To investigate this further, we applied a novel approach for stabilizing a poorly expressing anti-tetanus toxoid human Fab with a predisposition for being misfolded and non-functional. Forty-five residues within the Fab were chosen for saturation mutagenesis based on residue frequency analysis and positional entropy calculations. Using automated screening, we determined the approximate midpoint temperature of thermal denaturation (TM) for over 4000 library members with a maximum theoretical diversity of 855 unique mutations. This dataset led to the identification of 11 residue positions, primarily in the Fv region, which when mutated enhanced Fab stability. By combining these mutations, the TM of the Fab was increased to 92 degrees C. Increases in Fab stability correlated with higher expressed Fab yields and higher levels of properly folded and functional protein. The mutations were selected based on their ability to increase the apparent stability of the Fab and therefore the exact mechanism behind the enhanced expression in E.coli remains undefined. The wild-type and two optimized Fabs were converted to an IgG1 format and expressed in mammalian cells. The optimized IgG1 molecules demonstrated identical gains in thermostability compared to the Fabs; however, the expression levels were unaffected suggesting that the eukaryotic secretion system is capable of correcting potential folding issues prevalent in E.coli. Overall, the results have significant implications for the bacterial expression of functional antibody domains as well as for the production of stable, high affinity therapeutic antibodies in mammalian cells.     

The {alpha}/{beta} hydrolase fold

We have identified a new protein fold—the /ßhydrolase fold—that is common to several hydrolytic enzymesof widely differing phylogenetic origin and catalytic function.The core of each enzyme is similar: an /ß sheet, notbarrel, of eight ß-sheets connected by -helices. Theseenzymes have diverged from a common ancestor so as to preservethe arrangement of the catalytic residues, not the binding site.They all have a catalytic triad, the elements of which are borneon loops which are the best-conserved structural features inthe fold. Only the histidine in the nucleophile-histidine-acidcatalytic triad is completely conserved, with the nucleophileand acid loops accommodating more than one type of amino acid.The unique topological and sequence arrangement of the triadresidues produces a catalytic triad which is, in a sense, amirror-image of the serine protease catalytic triad. There arenow four groups of enzymes which contain catalytic triads andwhich are related by convergent evolution towards a stable,useful active site: the eukaryotic serine proteases, the cysteineproteases, subtilisins and the /ß hydrolase fold enzymes.     

Searching for common sequence patterns among distantly related proteins

We have developed a program Gap Allowing Pattern Explorer (GAPE)to extract amino acid sequence motifs conserved among distantlyrelated proteins. The GAPE program is designed to allow gapsin the sequences. First, this program generates all possibleamino acid patterns comprising up to five amino acids. Sequencescontaining the amino acid residues in the same order as a generatedpattern are selected as subsequences, where the differencesin the distances between two consecutive amino acids are ignored.Next, the motifs are extracted from the subsequences under conditionsin which all four distances between the five amino acids arefixed. At this stage, motifs with gaps in their subsequenceare also found by relaxing one of the four fixed distances.The statistical significance for a motif obtained is calculatedbased on the amino acid composition of the sequences under consideration.When the GAPE program was applied to 59 pyridoxal-phosphaterelatedsequences and 64 ATP (AMP-forming)-related sequences, motifsextracted with a low expectation of occurrence contained someof the amino acid residues chemically proved to be involvedin the ligand recognition.     

The prediction and characterization of metal binding sites in proteins

The rational engineering of novel functions into proteins canonly be attempted when the underlying structural scaffold onwhich the new function is displayed and the structure of thetarget protein are both well understood. To introduce functionsmediated by metals it is therefore necessary to identify theprincipal liganding residues for the chosen metal, the requiredarchitecture of the metal-ligand complex and sites within thetarget protein that could accommodate such sites. Here we presenta method that applies structural information from the proteindata bank to the ab initio design and characterization of novelmetal binding sites. The prediction method has been tested on28 metalloprotein structures from the Brookhaven Protein DataBank. It successfully identified >90% of the metal bindingsites. In addition, we have used the method to design and characterizezinc binding sites in two antibody structures. Metal bindingstudies on one of these putative metalloantibodies showed metalbinding, confirming the predictive power of the method.     

Directed evolution converts subtilisin E into a functional equivalent of thermitase

We used directed evolution to convert Bacillus subtilis subtilisinE into an enzyme functionally equivalent to its thermophilichomolog thermitase from Thermoactinomyces vulgaris. Five generationsof random mutagenesis, recombination and screening created subtilisinE 5-3H5, whose half-life at 83°C (3.5 min) and temperatureoptimum for activity (T_opt, 76°C) are identical with thoseof thermitase. The T_opt of the evolved enzyme is 17°C higherand its half-life at 65°C is >200 times that of wild-typesubtilisin E. In addition, 5-3H5 is more active towards thehydrolysis of succinyl-Ala-Ala-Pro-Phe-p-nitroanilide than wild-typeat all temperatures from 10 to 90°C. Thermitase differsfrom subtilisin E at 157 amino acid positions. However, onlyeight amino acid substitutions were sufficient to convert subtilisinE into an enzyme equally thermostable. The eight substitutions,which include known stabilizing mutations (N218S, N76D) andalso several not previously reported, are distributed over thesurface of the enzyme. Only two (N218S, N181D) are found inthermitase. Directed evolution provides a powerful tool to unveilmechanisms of thermal adaptation and is an effective and efficientapproach to increasing thermostability without compromisingenzyme activity.     

Potential of genetic algorithms in protein folding and protein engineering simulations

Genetic algorithms are very efficient search mechanisms whichmutate, recombine and select amongst tentative solutions toa problem until a near optimal one is achieved. We introducethem as a new tool to study proteins. The identification andmotivation for different fitness functions is discussed. Theevolution of the zinc finger sequence motif from a random startis modelled. User specified changes of the repressor structurewere simulated and critical sites and exchanges for mutagenesisidentified. Vast conformational spaces are efficiently searchedas illustrated by the ab initio folding of a model protein ofa four ß strand bundle. The genetic algorithm simulationwhich mimicked important folding constraints as overall hydrophobicpackaging and a propensity of the betaphilic residues for transpositions achieved a unique fold. Cooperativity in the ßstrand regions and a length of 3–5 for the interconnectingloops was critical. Specific interaction sites were considerablyless effective in driving the fold.     

Gene synthesis and functional expression of a protein exhibiting monodomain IgG Fc binding

A gene encoding a bacterial IgG Fc binding domain was designedand synthesized. The synthetic DNA fragment was cloned 3' toan inducible trpE promoter such that expression of the genein Escherichia coli produced abundant Fc binding protein fusedto the first seven amino acids of the trpE protein. The recombinantprotein contained a single Fc binding domain and demonstratedefficient binding to'human IgG in Western blot analysis. Thisprotein degraded rapidly following cell lysis in the absenceof protease inhibitors, but could be effectively protected bythe addition of protease inhibitor. After purification of theprotein by IgG affinity chromatography, IgG Fc binding abilitywas retained for at least 24 h at either 23 or 37°C andon heating for 15 min at temperatures up to 65°C. No immunoprecipitationwas observed in interactions between the monodomain Fc bindingprotein and IgG molecules. Unlike staphylococcal protein A,no detectable binding of the monodomain IgG Fc binding proteinwas observed to either IgM or IgA. Truncated proteins, expressedfrom a series of 3' deletions of the synthetic gene, were usedto estimate the minimum portion of a monodomain Fc binding proteinthat retained Fc binding ability.     

Protein structural domain identification

A simple method for the definition of protein structural domainsis described that requires only -carbon coordinate data. Thebasic method, which encodes no specific aspects of protein structure,captures the essence of most domains but does not give highenough priority to the integrity of ß-sheet structure.This aspect was encouraged both by a bias toward attaining intactß-sheets and also as an acceptance condition on the finalresult. The method has only one variable parameter, reflectingthe granularity level of the domains, and an attempt was madeto set this level automatically for each protein based on thebest agreement attained between the domains predicted on thenative structure and a set of smoothed coordinates. While notperfect, this feature allowed some tightly packed domains tobe separated that would have remained undivided had the bestfixed granularity level been used. The quality of the resultswas high and, when compared with a large collection of accepteddomain definitions, only a few could be said to be clearly incorrect.The simplicity of the method allowed its easy extension to thesimultaneous definition of domains across related structuresin a way that does not involve loss of detail through averagingthe structures. This was found to be a useful approach to reconcilingdifferences among structural family members. The method is fast,taking less than 1 s per 100 residues for medium-sized proteins.     

Identification of important functional environs in protein tertiary structures from the analysis of residue variation in 3-D: application to cytochromes c and carboxypeptidases A and B

A simple methodology is described to apply to aligned proteinsequence sets for which at least one representative 3-D C structureis known. The evolutionary variation observed at each residueposition in the sequence alignment is qualified by taking intoaccount the residue variation that has occurred at other positionslocated within 7 A (according to the probable chain fold). Thisexpresses the evolutionary behaviour of any residue positionin the more appropriate context of its immediate surroundingsand distinguishes between invariant residues on the basis ofthe variation of their environment. The highest mechanisticsignificance is attached to conserved residues in conservedsurroundings, but the quantitative nature of the analysis meansthat all residue vicinities can be ranked and merged accordingto the degree of conservation that they exhibit and the residuepositions that comprise them. Therefore, with the aid of thechain fold, contour maps can be constructed that show gradedfoci of evolutionary conservation in the underlying superstructureof the protein type, and the irregular shapes and extents oflarge conserved areas. To test the methodology, it was appliedto cytochromes c and the carboxypeptidases A and B.     

The elusive role of the N-terminal extension of {beta}A3- and {beta}Al-crystallin

ß-Crystallins are structural lens proteins with aconserved two-domain structure and variable N- and C-terminalextensions. These extensions are assumed to be involved in quaternaryinteractions within the ß-crystallin oligomers orwith other lens proteins. Therefore, the production of ßA3-and ßAl-crystallin from the single ßA3/A1mRNA by dual translation initiation is of interest. These crystallinsare identical, except that ßAl has a much shorterN-terminal extension than ßA3. This rare mechanismhas been conserved for over 250 million years during the evolutionof the ßA3/A1 gene, suggesting that the generationof different N-terminal extensions confers a selective advantage.We therefore compared the stability and association behaviourof recombinant ßA3- and ßAl-crystallin.Both proteins are equally stable in urea- and pH-induced denaturationexperiments. Gel filtration and analytical ultracentrifugationestablished that ßA3 and ßA1 both form homodimers.In the water-soluble proteins of bovine lens, ßA3and ßA1 are present in the same molecular weight fractions,indicating that they oligomerize equally with other ß-crystallins.¹H-NMR spectroscopy showed that residues Met1 to Asn22 of theN-terminal extension of ßA3 have great flexibilityand are solvent exposed, excluding them from protein interactionsin the homodimer. These results indicate that the differentN-terminal extensions of ßA3 and ßA1 donot affect their homo- or heteromeric interactions.     

A mixed helix--beta-sheet model of the transmembrane region of the nicotinic acetylcholine receptor

We have modelled the transmembrane region of the 7 nicotinicacetylcholine receptor as a mixed -helical/ß-sheetstructure. The model was mainly based on the crystal structureof a pore-forming toxin, heat-labile enterotoxin. This is apentameric protein having a central pore or channel composedof five -helices, one from each of the 5 B subunits that formthis pentamer. The remainder of this structure is ß-sheet,loops and a short -helix, not included in the model. The modeluses this channel as a template to build the transmembrane region,from M1 to the middle of M3. The remainder of M3 and M4 werebuilt de novo as -helices. Great consideration was given tolabelling data available for the transmembrane region. In generalterms, the shape of the model agrees very well with that obtainedindependently by electron microscopic analysis and the secondarystructure predicted by the model is in accord with that estimatedindependently by Fourier transform infrared spectroscopy. TheM2 helical region of the model is only slightly kinked, contraryto what is inferred from electron microscopic analysis, buthas the same overall shape and form. On the membrane face ofthe model, the presence of deep pockets may provide the structuralbasis for the distinction between annular and non-annular lipidbinding sites. Also, the transmembrane region is clearly asymmetricin the direction perpendicular to the membrane, and this mayhave strong influence on the surrounding lipid composition ofeach leaflet of the cytoplasmic membrane.     

Natural history as a predictor of protein evolvability

Natural selection generally produces specific and efficient enzymes. In contrast, directed evolution experiments usually produce enzyme variants with broadened substrate specificity or enhanced catalytic promiscuity. Some proteins may be more evolvable than others, but few workers consider this problem when choosing starting points for laboratory evolution. Here, we review the variables associated with enzyme evolvability, namely promiscuity and mutational robustness. We present a qualitative model of adaptive evolution and recommend that protein engineers exploit their knowledge of natural history to identify evolvable wild-type proteins. Three examples of 'generalist' proteins that evolved in the laboratory into 'specialists' are described to illustrate the practical utility of this point.     

Structure-guided SCHEMA recombination of distantly related beta-lactamases

We constructed a library of beta-lactamases by recombining three naturally occurring homologs (TEM-1, PSE-4, SED-1) that share 34-42% sequence identity. Most chimeras created by recombining such distantly related proteins are unfolded due to unfavorable side-chain interactions that destabilize the folded structure. To enhance the fraction of properly folded chimeras, we designed the library using SCHEMA, a structure-guided approach to choosing the least disruptive crossover locations. Recombination at seven selected crossover positions generated 6561 chimeric sequences that differ from their closest parent at an average of 66 positions. Of 553 unique characterized chimeras, 111 (20%) retained beta-lactamase activity; the library contains hundreds more novel beta-lactamases. The functional chimeras share as little as 70% sequence identity with any known sequence and are characterized by low SCHEMA disruption (E) compared to the average nonfunctional chimera. Furthermore, many nonfunctional chimeras with low E are readily rescued by low error-rate random mutagenesis or by the introduction of a known stabilizing mutation (TEM-1 M182T). These results show that structure-guided recombination effectively generates a family of diverse, folded proteins even when the parents exhibit only 34% sequence identity. Furthermore, the fraction of sequences that encode folded and functional proteins can be enhanced by utilizing previously stabilized parental sequences.