Stabilization of TRAIL, an all-beta-sheet multimeric protein, using computational redesign

Protein thermal stability is important for therapeutic proteins, both influencing the pharmacokinetic and pharmacodynamic properties and for stability during production and shelf-life of the final product. In this paper we show the redesign of a therapeutically interesting trimeric all-beta-sheet protein, the cytokine TRAIL (tumor necrosis factor-related apoptosis-inducing ligand), yielding variants with improved thermal stability. A combination of tumor necrosis factor (TNF) ligand family alignment information and the computational design algorithm PERLA was used to propose several mutants with improved thermal stability. The design was focused on non-conserved residues only, thus reducing the use of computational resources. Several of the proposed mutants showed a significant increase in thermal stability as experimentally monitored by far-UV circular dichroism thermal denaturation. Stabilization of the biologically active trimer was achieved by monomer subunit or monomer-monomer interface modifications. A double mutant showed an increase in apparent T(m) of 8 degrees C in comparison with wild-type TRAIL and remained biologically active after incubation at 73 degrees C for 1 h. To our knowledge, this is the first study that improves the stability of a large multimeric beta-sheet protein structure by computational redesign. A similar approach can be used to alter the characteristics of other multimeric proteins, including other TNF ligand family members.     

GRAFTER: a computational aid for the design of novel proteins

The GRAFTER suite of programs provides geometric search andevaluation functions that simplify and automate the processof identifying the best scaffolds for a particular structuralmotif. Three applications of the GRAFTER suite are presented.Potential grafts between repressor and 434 repressor were identifiedthat should change the DNA binding specificity of these repressors.These results are compared with site-directed mutagenesis experimentsthat have been shown to alter repressor-DNA binding specificity.Next, 26 loops from antibody structures were grouped into familiesof similar structure. Grafts of antibody loops onto a pre-existingscaffold are an essential component of antibody humanization.Finally, interleukin (lL)-4 was searched as a scaffold thatmight accept the graft of a five residue epitope from humangrowth hormone (hGH). The existence of a crystal structure ofthe hGH-hGH receptor complex, extensive mutagenesis studiesof the hGH residues that contribute to the energetics of ligand-receptorinteractions and the gross structural homology between hGH andIL-4 make this an appealing computational target. The approachpresented here could aid the development of novel enzymes andbinding proteins     

A new method for predicting binding affinity in computer-aided drug design

A new semi–empirical method for calculating free energiesof binding from molecular dynamics (MD) simulations is presented.It is based on standard thermodynamic cycles and on a linearapproximation of polar and non–polar free energy contributionsfrom the corresponding MD averages. The method is tested ona set of endothiapepsin inhibitors and found to give accurateresults both for absolute as well as relative free energies.     

Comparison of experimental and computational functional group mapping of an RNase A structure: implications for computer-aided drug design

One relatively new computational approach to the drug discoveryprocess involves calculating functional group maps of a targetstructure. Experimental functional group mapping techniqueshave also recently emerged. In this paper, the structure ofRNase A with two bound formates (i.e. carboxylate functionalities)is used as a model system to test the computational methodology.Functional group maps of the RNase A structure were calculatedusing the Multiple Copy Simultaneous Search (MCSS) method andcompared with experimentally determined formate and water positions.The calculations indicate that the protonation state of active-sitehistidines determines the ability of the enzyme to bind formate.The results also suggest an ordered binding mechanism for thetwo formates. An improved strategy for using the MCSS methodto design new candidate ligands is discussed.     

Highly effective protease inhibitors from variants of human pancreatic secretory trypsin inhibitor (hPSTI): an assessment of 3-D structure-based protein design

The results of a protein design project are used to comparedifferent predictive strategies with respect to proteinproteininteractions. We have been able to generate variants of humanpancreatic secretory trypsin inhibitor (hPSTI) optimized withrespect to the affinity and specificity for human leukocyteelastase relative to trypsin and chymotrypsin, and in particularchymotrypsin. The extremely strong and specific human leukocyteelastase inhibitors were thus developed in three rounds of mutagenesisand two rounds of 3-D modelling; only 24 variants in total weresynthesized, although variations at seven different amino acidpositions were involved (i.e. from 20⁷ possible variants). Anexcellent elastase inhibitor could be designed with the minimumof two amino acid exchanges. The value of structural modellingand actual structure determination is discussed in the lightof the experimental results of the designed protein variantsand the results of tertiary structure determinations of thefree variant and the inhibitorprotease complex. Particular referenceis given to the strategy to be followed in protein design projectsin general and to the development of protease inhibitors inparticular.     

Protein design to understand peptide ligand recognition by tetratricopeptide repeat proteins

Protein design aims to understand the fundamentals of protein structure by creating novel proteins with pre-specified folds. An equally important goal is to understand protein function by creating novel proteins with pre-specified activities. Here we describe the design and characterization of a tetratricopeptide (TPR) protein, which binds to the C-terminal peptide of the eukaryotic chaperone Hsp90. The design emphasizes the importance of both direct, short-range protein-peptide interactions and of long-range electrostatic optimization. We demonstrate that the designed protein binds specifically to the desired peptide and discriminates between it and the similar C-terminal peptide of Hsp70.     

Protein design for non-aqueous solvents

Improving protein stability in unnatural and suboptimal environmentsis a promising application of protein engineering technology.Carefully designed amino acid alterations may lead to dramaticpositive effects on the stability of proteins under highly perturbingconditions, such as in non-aqueous solvents. Applications ofbiocatalysts and proteins with specific binding capabilitiesin the chemical industry have been severely limited by constraintsplaced on the solvent environment. With the advent of convenientmethods for altering the amino acid composition and even synthesizingentirely new protein molecules, it is worthwhile to considerengineering proteins for stability in non-aqueous solvents.In order to identify the features that a protein would needfor stability in organic media, we have been studying the structureand properties of the hydrophobic protein crambin. Crambin isunique in that it is soluble and stable in very high concentrationsof polar organic solvents. Crambin and its water-soluble homologsoffer a powerful demonstration of protein engineering for non-aqueoussolvents. This paper describes the structural features thatcontribute to crambin's special properties. Based on these observationsand consideration of how non-aqueous solvents affect the interactionsimportant in protein folding, a set of rules for designing non-aqueoussolvent-stable proteins is proposed.     

Glycine 85 of the trp-repressor of E.coli is important in forming the hydrophobic tryptophan binding pocket: experimental and computational approaches

Experimental and computational analyses were performed on thecorepressor (L-tryptophan) binding site of the trp-repressorof Escherichia coli to investigate the ligandprotein interactions.Gly 85, one of the residues forming the hydrophobic pocket ofthe binding site, was systematically replaced with Ala, Val,Leu and Trp by cassette mutagenesis. Biochemical characterizationshowed that all these mutations caused significant decreasesin tryptophan binding activity. Free energy perturbation calculationswere performed for the mutants and were consistent with theexperimental results. The lack of a side chain at position 85was concluded to be essential for binding the corepressor; thestructure of the binding pocket was suggested to be tight inthe vicinity of Gly85.     

Generation and analysis of proline mutants in protein G

The pyrrolidine ring of the amino acid proline reduces the conformational freedom of the protein backbone in its unfolded form and thus enhances protein stability. The strategy of inserting proline into regions of the protein where it does not perturb the structure has been utilized to stabilize many different proteins including enzymes. However, most of these efforts have been based on trial and error, rather than rational design. Here, we try to understand proline's effect on protein stability by introducing proline mutations into various regions of the B1 domain of Streptococcal protein G. We also applied the Optimization of Rotamers By Iterative Techniques computational protein design program, using two different solvation models, to determine the extent to which it could predict the stabilizing and destabilizing effects of prolines. Use of a surface area dependent solvation model resulted in a modest correlation between the experimental free energy of folding and computed energies; on the other hand, use of a Gaussian solvent exclusion model led to significant positive correlation. Including a backbone conformational entropy term to the computational energies increases the statistical significance of the correlation between the experimental stabilities and both solvation models.     

A novel method for the modelling of peptide ligands to their receptors

A knowledge-based approach to the modelling of enzyme- peptideinhibitor comlexes is described. Given the structure of an enzyme,and knowledge of its bindings site, the method seeks to predictthe binding geometry of a peptide ligand. This novel methodinvolves using examples of sidechain packing derived from proteinsof known three dimensional structure to define possible packingarrangements of a peptide inhibitor group to its bindings site.Asuite of progams, GEMINI, was written and used to predict thepacking of pairs of amino acid groups from three inhibitorscomplexed to their enzymes for which the X-ray strutures wereavailable. These included the Phe group of the inhibitor H142bound to endothiapepsin, the Leu group of CLT complexed to thermolysinand the C-terminus of Gly-L-Tyr bound to carboxypeptidase A.A detailed comparison of the modelled and observed inhibitorcoordinates was made. This approach may be extended to modellingother types of protein interactions.     

Computer-aided gene design

A computer program, which runs on MS-DOS personal computers,is described that assists in the design of synthetic genes codingfor proteins. The goal of the program is the design of a genewhich (0 contains as many unique restriction sites as possibleand (ii) uses a specific codon usage. The gene designed accordingto the criteria above is (i) suitable for ‘modular mutagenesis’experiments and (ii) optimized for expression. The program 'reverse-translates'protein sequences into degenerated DNA sequences, generatesa map of potential restriction sites and locates sequence positionswhere unique restriction sites can be accommodated. The nucleicacid sequence is then ‘refined’ according to a specificcodon usage to remove any degeneration. Unique restriction sites,if potentially present, can be ‘forced’ into thedegenerated nucleic acid sequence by using 'priority codes'assigned to different restriction sequences.     

Artificial protein vaccines with predetermined tertiary structure: application to anti-HTV-1 vaccine design

A successful approach to the development of a safe and effectivesynthetic vaccine requires that different B and T cell epitopesof the infectious agent be included in the vaccine construction.In this paper we suggest a new approach to vaccine design inthe form of an artificial protein with a predetermined tertiarystructure (PTS vaccines). Based on B and T cell epitope properties,we substantiate the possible use for vaccine construction ofone well-known protein spatial motifthe four-a-helix bundle.Antigenk determinants of cellular immunity (amphipathic a-helkes)and humoral immunity (flexible hydrophilk loop regions) areused as blocks for vaccine design. General principles of PTSvaccine construction have been applied to anti-HTV-1 vaccinedesign.     

Site-directed protein recombination as a shortest-path problem

Protein function can be tuned using laboratory evolution, in which one rapidly searches through a library of proteins for the properties of interest. In site-directed recombination, n crossovers are chosen in an alignment of p parents to define a set of p(n + 1) peptide fragments. These fragments are then assembled combinatorially to create a library of p(n+1) proteins. We have developed a computational algorithm to enrich these libraries in folded proteins while maintaining an appropriate level of diversity for evolution. For a given set of parents, our algorithm selects crossovers that minimize the average energy of the library, subject to constraints on the length of each fragment. This problem is equivalent to finding the shortest path between nodes in a network, for which the global minimum can be found efficiently. Our algorithm has a running time of O(N(3)p(2) + N(2)n) for a protein of length N. Adjusting the constraints on fragment length generates a set of optimized libraries with varying degrees of diversity. By comparing these optima for different sets of parents, we rapidly determine which parents yield the lowest energy libraries.     

Context dependence of phenotype prediction and diversity in combinatorial mutagenesis

Two different combinatorial mutagenesis experiments on the light-harvestingII (LH2) protein of Rhodobacter capsulatus indicate that heuristicrules relating sequence directly to phenotype are dependenton which sets or groups of residues are mutated simultaneously.Previously reported combinatorial mutagenesis of this chromogenicprotein (based on both phylogenetic and structural models) showedthat substituting amino acids with large molar volumes at Gly^ß31caused the mutated protein to have a spectrum characteristicof light-harvesting I (LH1). The six residues that underwentcombinatorial mutagenesis were modeled to lie on one side ofa transmembrane -helix that binds bacteriochlorophyll. In asecond experiment described here, we have not used structuralmodels or phylogeny in choosing mutagenesis sites. Instead,a set of six contiguous residues was selected for combinatorialmutagenesis. In this latter experiment, the residue substitutedat Gly^ß31 was not a determining factor in whetherLH2 or LH1 spectra were obtained; therefore, we conclude thatthe heuristic rules for phenotype prediction are context dependent.While phenotype prediction is context dependent, the abilityto identify elements of primary structure causing phenotypediversity appears not to be. This strengthens the argument forperforming combinatorial mutagenesis with an arbitrary groupingof residues if structural models are unavailable.     

Automated design of degenerate codon libraries

Degenerate codon libraries are frequently used in protein engineering and evolution studies but are often limited to targeting a small number of positions to adequately limit the search space. To mitigate this, codon degeneracy can be limited using heuristics or previous knowledge of the targeted positions. To automate design of libraries given a set of amino acid sequences, an algorithm (LibDesign) was developed that generates a set of possible degenerate codon libraries, their resulting size, and their score relative to a user-defined scoring function. A gene library of a specified size can then be constructed that is representative of the given amino acid distribution or that includes specific sequences or combinations thereof. LibDesign provides a new tool for automated design of high-quality protein libraries that more effectively harness existing sequence-structure information derived from multiple sequence alignment or computational protein design data.     

A new approach to the design of a sequence with the highest affinity for a molecular surface

We describe an algorithm to design the primary structures forpeptides which must have the strongest binding to a given molecularsurface. This problem cannot be solved by a direct combinatorialsorting, because of an enormous number of possible primary andspatial structures. The approach to solve this problem is todescribe a state of each residue by two variables: (i) aminoacid type and (ii) 3-D coordinate, and to minimize binding energyover all these variables simultaneously. For short chains whichhave no long-range interactions within themselves, this minimizationcan be done easily and efficiently by dynamic programming. Wealso discuss the problem of how to estimate specificity of bindingand how to deduce a sequence with maximal specificity for agiven surface. We show that this sequence can be deduced bythe same algorithm after some modification of energetic parameters.     

Construction of stabilized proteins by combinatorial consensus mutagenesis

We constructed stabilized variants of beta-lactamase (BLA) from Enterobacter cloacae by combinatorial recruitment of consensus mutations. By aligning the sequences of 38 BLA homologs, we identified 29 positions where the E.cloacae gene differs from the consensus sequence of lactamases and constructed combinatorial libraries using mixtures of mutagenic oligonucleotides encompassing all 29 positions. Screening of 90 random isolates from these libraries identified 15 variants with significantly increased thermostability. The stability of these isolates suggest that all tested mutations make additive contributions to protein stability. A statistical analysis of sequence and stability data identified 11 mutations that made stabilizing contributions and eight mutations that destabilized the protein. A second-generation library recombining these 11 stabilizing mutations led to the identification of BLA variants that showed further stabilization. The most stable variant had a mid-point of thermal denaturation (Tm) that was 9.1 degrees C higher than the starting molecule and contained eight consensus mutations. Incubation of three stabilized BLA variants with several proteases showed that all tested isolates have significantly increased resistance to proteolysis. Our data demonstrate that combinatorial consensus mutagenesis (CCM) allows the rapid generation of protein variants with improved thermal and proteolytic stability.     

Stabilization of the autoproteolysis of TNF-alpha converting enzyme (TACE) results in a novel crystal form suitable for structure-based drug design studies

The crystallization of TNF-alpha converting enzyme (TACE) has been useful in understanding the structure-activity relationships of new chemical entities. However, the propensity of TACE to undergo autoproteolysis has made enzyme handling difficult and impeded the identification of inhibitor soakable crystal forms. The autoproteolysis of TACE was found to be specific (Y352-V353) and occurred within a flexible loop that is in close proximity to the P-side of the active site. The rate of autoproteolysis was found to be proportional to the concentration of TACE, suggesting a bimolecular reaction mechanism. A limited specificity study of the S(1)' subsite was conducted using surrogate peptides and suggested substitutions that would stabilize the proteolysis of the loop at positions Y352-V353. Two mutant proteases (V353G and V353S) were generated and proved to be highly resistant to autoproteolysis. The kinetics of the more resistant mutant (V353G) and wild-type TACE were compared and demonstrated virtually identical IC(50) values for a panel of competitive inhibitors. However, the k(cat)/K(m) of the mutant for a larger substrate (P6 - P(6)') was approximately 5-fold lower than that for the wild-type enzyme. Comparison of the complexed wild-type and mutant structures indicated a subtle shift in a peripheral P-side loop (comprising the mutation site) that may be involved in substrate binding/turnover and might explain the mild kinetic difference. The characterization of this stabilized form of TACE has yielded an enzyme with similar native kinetic properties and identified a novel crystal form that is suitable for inhibitor soaking and structure determination.