Enhanced dead-end elimination in the search for the global minimum energy conformation of a collection of protein side chains

Although the conformational states of protein side chains canbe described using a library of rotamers, the determinationof the global minimum energy conformation (GMEC) of a largecollection of side chains, given fixed backbone coordinates,represents a challenging combinatorial problem with importantapplications in the field of homology modelling. Recently, wehave developed a theoretical framework, called the dead-endelimination method, which allows us to identify efficientlyrotamers that cannot be members of the GMEC. Such dead-endingrotamers can be iteratively removed from the system under studythereby tracking down the size of the combinatorial problem.Here we present new developments to the dead-end eliminationmethod that allow us to handle larger proteins and more extensiverotamer libraries. These developments encompass (i) a procedureto determine weight factors in the generalized dead-end eliminationtheorem thereby enhancing the elimination of dead-ending rotamersand (ii) a novel strategy, mainly based on logical argumentsderived from the logic pairs theorem, to use dead-ending rotamerpairs in the efficient elimination of single rotamers. Thesedevelopments are illustrated for proteins of various sizes andthe flow of the current method is discussed in detail. The effectivenessof dead-end elimination is increased by two orders of magnitudeas compared with previous work. In addition, it now becomesfeasible to use extremely detailed libraries. We also providean appendix in which the validity of the generalized dead-endcriterion is shown. Finally, perspectives for further applicationswhich may now become within reach are discussed.     

Influence of secondary structure on the hydration of serine, threonine and tyrosine residues in proteins

Previous analysis of experimental data on the solvation of highresolution protein structures has shown that preferred interactionsites for water molecules exist around most amino acid sidechains. We have extended this analysis to look in more detailat the distributions around serine, threonine and tyrosine.We find that for serine and threonine side chains the preferredinteraction sites of solvent molecules with the hydroxyl groupdepends on secondary structure and the _X1 torsion angle of theside chain. For tyrosine side chains the hydroxyl group is toofar from the main chain to reflect secondary structure influences.Specific patterns of hydration are observed in which water molecules‘bridge’ between the hydroxyl side-chain atom andanother main chain or side-chain atom.     

The fuzzy-end elimination theorem: correctly implementing the side chain placement algorithm based on the dead-end elimination theorem

Recently it has been shown that the dead-end elimination theoremis a powerful tool in the search for the global minimum energyconformation (GMEC) of a large collection of protein side chainsgiven known backbone coordinates and a library of allowed sidechain conformational states, also known as rotamers. A sidechain placement algorithm based on this theorem iterativelyapplies this theorem to single as well as to pairs of rotamersleading to the identification of rotamers, single or pairs,that are incompatible with the GMEC and that can thus be qualifiedas ‘dead-ending’. Here we formulate a theorem whichproves that contrary to intuition dead-end rotamer pairs cannot simply be discarded from consideration in the iterativeprocess leading to the further elimination of dead-end rotamers.We refer to this theorem as the fuzzy-end elimination theorem.We also describe how the obtained dead-end rotamer pairs cancontribute to the search for the GMEC in the protein side chainplacement problem. Hence the present work forms a theoreticalbasis for the correct implementation of a side chain placementalgorithm based on the dead-end elimination theorem. In addition,possible future perspectives are presented.     

Hydration of amino acid side chains: dependence on secondary structure

Energy calculations have been used to study the hydration sitesaround the polar groups of serine, threonine and tyrosine sidechains. These hydration sites depend not only on the hybridizationof the polar group but also on the local secondary structure,the X₁ side chain torsion angle and the position of the hydroxylhydrogen atom. For tyrosine side chains, two solvent sites arefound approximately in the plane of the ring. Even for serineand threonine side chains only two minimum energy sites arefound in general of which one is in an expected position withinhydrogen bonding of the hydroxyl hydrogen atom (unless thisis blocked from interaction with solvent molecules by, for example,O_i–4 or O_i–3. The position of the second of thesesites depends not only on the position of the hydroxyl oxygenbut also on neighbouring main chain atoms to which it can alsohydrogen bond. There is good agreement with the solvent distributionsobtained from crystallographic data.     

Finding a new vaccine in the ricin protein fold

Previous attempts to produce a vaccine for ricin toxin have been hampered by safety concerns arising from residual toxicity and the undesirable aggregation or precipitation caused by exposure of hydrophobic surfaces on the ricin A-chain (RTA) in the absence of its natural B-chain partner. We undertook a structure-based solution to this problem by reversing evolutionary selection on the 'ribosome inactivating protein' fold of RTA to arrive at a non-functional, compacted single-domain scaffold (sequence RTA1-198) for presentation of a specific protective epitope (RTA loop 95-110). An optimized protein based upon our modeling design (RTA1-33/44-198) showed greater resistance to thermal denaturation, less precipitation under physiological conditions and a reduction in toxic activity of at least three orders of magnitude compared with RTA. Most importantly, RTA1-198 or RTA1-33/44-198 protected 100% of vaccinated animals against supra-lethal challenge with aerosolized ricin. We conclude that comparative protein analysis and engineering yielded a superior vaccine by exploiting a component of the toxin that is inherently more stable than is the parent RTA molecule.     

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     

Refinement of protein cores and protein-peptide interfaces using a potential scaling approach

Refinement of side chain conformations in protein model structures and at the interface of predicted protein-protein or protein-peptide complexes is an important step during protein structural modelling and docking. A common approach for side chain prediction is to assume a rigid protein main chain for both docking partners and search for an optimal set of side chain rotamers to optimize the steric fit. However, depending on the target-template similarity in the case of comparative protein modelling and on the accuracy of an initially docked complex, the main chain template structure is only an approximation of a realistic target main chain. An inaccurate rigid main chain conformation can in turn interfere with the prediction of side chain conformations. In the present study, a potential scaling approach (PS-MD) during a molecular dynamics (MD) simulation that also allows the inclusion of explicit solvent has been used to predict side chain conformations on semi-flexible protein main chains. The PS-MD method converges much faster to realistic protein-peptide interface structures or protein core structures than standard MD simulations. Depending on the accuracy of the protein main chain, it also gives significantly better results compared with the standard rotamer search method.     

Redesigning a sweet protein: increased stability and renaturability

Monellin is one of two natural proteins from African berrieswith potent sweet taste. Monellin is the smaller of the two,and consists of two peptides. The protein loses sweetness whenheated above 50°C under acidic pH. Based on the crystalstructure of monellin we have fused the two chains into a singlechain using several different linkers copied and ‘transplanted’from the same molecule. One of the newly designed proteins isas potently sweet as the natural one, is more stable upon temperatureor pH changes, and renatures easily even after heating to 100°Cat low pH.     

Solution conformation of a synthetic bis-headed inhibitor of trypsin and carboxypeptidase A: new structural alignment between the squash inhibitors and the potato carboxypeptidase inhibitor

The trypsin carboxypeptidase peptide inhibitor (TCPI) whichinhibits both trypsin and carboxypeptidase A has been chemicallyengineered by modification of the Ecballium elaterium trypsininhibitor II (EETI-II). The solution conformation of TCPI, studiedby two-dimensional nuclear magnetic resonance, was shown tobe very close to those of squash inhibitors. Only limited deviationsof the trypsin binding loop compared to its location in theEETI-II/trypsin complex were detected. It was also shown thatthe position of the C-terminal tail did not significantly changefrom the position observed in the complex between carboxypeptidaseA and the potato carboxypeptidase inhibitor (PCI). The conformationof TCPI was carefully compared with the PCI one and a new structuralalignment between the two microproteins is proposed. This alignmentpoints out the very good conservation in the two inhibitorsof a subdomain comprising segments 7–15, 19–22 and25–28. Most importantly, the 2–19 disulfide bridgeof TCPI was not structurally conserved in PCI and appeared tobe rather unimportant for the early folding process of thesemolecules. This result agrees with the recent observation thatthe 2–19 bridge is the last to be formed in the foldingof the squash inhibitor EETI-II and suggests that this is alsothe case during the folding of the potato carboxypeptidase inhibitor.     

Distance distributions in proteins: a six-parameter representation

We present a statistical analysis of protein structures basedon interatomic C_a distances. The overall distance distributionsreflect in detail the contents of sequence-specific substructuresmaintained by local interactions (such as -helixes) and longerrange interactions (such as disulfide bridges and ß-sheets).We also show that a volume scaling of the distances makes distancedistributions for protein chains of different length superimposable.Distance distributions were also calculated specifically foramino acids separated by a given number of residues. Specificfeatures in these distributions are visible for sequence separationsof up to 20 amino acid residues. A simple representation, whichpreserves most of the information in the distance distributions,was obtained using six parameters only. The parameters giverise to canonical distance intervals and when predicting coarse-graineddistance constraints by methods such as data-driven artificialneural networks, these should preferably be selected from theseintervals. We discuss the use of the six parameters for determiningor reconstructing 3-D protein structures.     

Confirmation of the predicted source of a slow folding reaction: proline 8 of bovine pancreatic trypsin inhibitor

A major question in protein structural analysis concerns theapplicability of results from model systems to other proteins.Theoretical approaches seem the best manner of transferringinformation from one system to another, but their accuracy inthe model systems must first be tested with results from experiment.Since bovine pancreatic trypsin inhibitor (BPTI) is a modelsystem for the evaluation of energy minimization and moleculardynamics routines, we can use folding and stability measurementsto examine the reliability of these methods. All two-disulfidemutants of BPTI investigated thus far have two very slow foldingreactions which have characteristics of proline isomerization.These reactions may occur because the non-native cis form oftwo of the four prolines in BPTI significantly destabilizesthe folded state of the protein. Previous energy minimizationstudies of wild-type BPTI suggested that the cis form of Pro8was the most destabilizing of the four prolines [Levitt,M. (1981)J. Mol. Biol., 145, 251–263]. In this paper, we show thatmutation of Pro8 - Gln in the two-disulfide bond mutant Val30–Ala51results in a loss of the slowest folding reaction, consistentwith Levitt's prediction.     

DOMPLOT: a program to generate schematic diagrams of the structural domain organization within proteins, annotated by ligand contacts

A program is described for automatically generating schematiclinear representations of protein chains in terms of their structuraldomains. The program requires the co-ordinates of the chain,the domain assignment, PROSITE information and a file listingall intermolecular interactions in the protein structure. Theoutput is a PostScript file in which each protein is representedby a set of linked boxes, each box corresponding to all or partof a structural domain. PROSITE motifs and residues involvedin ligand interactions are highlighted. The diagrams allow immediatevisualization of the domain arrangement within a protein chain,and by providing information on sequence motifs, and metal ion,ligand and DNA binding at the domain level, the program facilitatesdetection of remote evolutionary relationships between proteins.     

GANNPhos: a new phosphorylation site predictor based on a genetic algorithm integrated neural network

With the advance of modern molecular biology it has become increasingly clear that few cellular processes are unaffected by protein phosphorylation. Therefore, computational identification of phosphorylation sites is very helpful to accelerate the functional understanding of huge available protein sequences obtained from genomic and proteomic studies. Using a genetic algorithm integrated neural network (GANN), a new bioinformatics method named GANNPhos has been developed to predict phosphorylation sites in proteins. Aided by a genetic algorithm to optimize the weight values within the network, GANNPhos has demonstrated a high accuracy of 81.1, 76.7 and 73.3% in predicting phosphorylated S, T and Y sites, respectively. When benchmarked against Back-Propagation neural network and Support Vector Machine algorithms, GANNPhos gives better performance, suggesting the GANN program can be used for other prediction tasks in the field of protein bioinformatics.     

A novel method for predicting transmembrane segments in proteins based on a statistical analysis of the SwissProt database: the PRED-TMR algorithm

We present a novel method that predicts transmembrane domainsin proteins using solely information contained in the sequenceitself. The PRED-TMR algorithm described, refines a standardhydrophobicity analysis with a detection of potential termini(`edges', starts and ends) of transmembrane regions. This allowsone both to discard highly hydrophobic regions not delimitedby clear start and end configurations and to confirm putativetransmembrane segments not distinguishable by their hydrophobiccomposition. The accuracy obtained on a test set of 101 non-homologoustransmembrane proteins with reliable topologies compares wellwith that of other popular existing methods. Only a slight decreasein prediction accuracy was observed when the algorithm was appliedto all transmembrane proteins of the SwissProt database (release35). A WWW server running the PRED-TMR algorithm is availableat http://o2.db.uoa.gr/PRED-TMR/     

Predicting the point at which transmembrane helices protrude from the bilayer: a model of the antenna complexes from photosynthetic bacteria

We describe a method for predicting the point at which a transmembranehelix leaves the bilayer and enters the more polar region ofthe aqueous exterior. This is achieved by comparing the relativedirections of the hydrophobic and internal faces of the transmembranehelices which should be opposite for the regions within thebilayer but equivalent for the regions on the outside. Thisinformation provides a strong constraint in the process of modellingmembrane proteins. We go on to use the approach to model themonomers of the bacterial light-harvesting antenna complexes.This information is then combined with some preliminary crystallographkdata and biochemical results to produce a 3-D model of a tetramer.     

Engineering the aggregation properties of dodecameric glutamine synthetase: a single amino acid substitution controls 'salting out'

Escherichia coli glutamine synthetase (GS) is a dodecamer ofidentical subunits which are arranged as two face-to-face hexamericrings. In the presence of 10% ammonium sulfate, wild type GSexhibits a pH-dependent ‘salting out’ with a pK_aof 4.51. Electron micrographs indicate that the pH-dependentaggregation corresponds to a highly specific self-assembly ofGS tubules, which result from stacking of individual dodecamers.This stacking of dodecamers is similar to the metal ion-inducedGS tubule formation previously described. Site-directed mutagenesisexperimentsindicate that the N-terminal helix of each subunit is involvedin the salting out reaction, as it is in the metal-induced stacking.A single substitution of alanine for His4 completely abolishesthe (NH₄₂SO₄-induced aggregation. However, the H4C mutant proteindoes nearly completely precipitate under the same salting outconditions. Mutations at other residues within the helix haveno effect on the stacking reaction. Differential catalyticactivityof unadenylylated GS versus adenylylated GS has been used todetermine whether wild type dodecamers ‘complement’the H4A mutant in the stacking reaction. The complementationexperiments indicate that His4 residues on bothsides of thedodecamer-dodecamer interfaces are not absolutely required forsalting out, although the wild type dodecamers clearly stackpreferentially with other wild type dodecamers. Approximately20% of the protein precipitated fromthe mixtures containingthe wild type GS and the H4A mutant is the mutant. The implicationsof these results for protein engineering are discussed.     

Identifying the mechanism of protein loop closure: a molecular dynamics simulation of the Bacillus stearothermophilus LDH loopin solution

The ‘loop’ involving residues 98–110 in Bacillusstearothermophilus lactate dehydrogenase (BSLDH) is of greatinterest as substrate-induced ‘loop’ closure isthought to berate-limiting and essential in catalyzing the reactionand in determining substrate specificity. Consequently, we haveexplored the mechanism underlying ‘loop’ openingin BSLDH through a molecular dynamics simulation at high temperature(1000 K) in the presence of explicit solvent, starting fromthe X-ray structure of BSLDH complexed with the co-enzyme NAD+and oxamate at 2.5 Å. During the simulation, a significantconformational change occurred, as evidenced by sharp dihedralangle transitions, hydrogen bond breaking and formation andlarge root mean square deviations from the starting structure;these changes define the criteria for ‘loop’ opening.The mechanical elements responsible for ‘loop’ opening,i.e. ‘loop’ hinges and flap, are defined througha combination of order parameters, dihedral angle changes andtheir correlations and the dynamical cross-correlation map ofatomic displacements for the ‘loop’ residues. Theresults indicate that the ‘loop’ consists of twoflexible hinge regions on either side of a relatively rigidthree-residue segment that undergoes a significant spatial displacementduring ‘loop’opening. ‘Loop’ openingis made possible through an array of correlated dihedral anglechanges and intra-& ‘loop’ rearrangements ofhydrogen-bond interactions. The presentfindings are comparedto previous work related to ‘loop’ opening and site-directedmutagenesis experiments.     

Functional diversity of the phosphoglucomutase superfamily: structural implications

Three-dimensional structural models of three members of thephosphoglucomutase (PGM) superfamily, parafusin, phosphoglucomutase-relatedprotein and sarcoplasmic reticulum phosphoglucomutase, wereconstructed by homology modeling based on the known crystalstructure of rabbit muscle phosphoglucomutase. Parafusin, phosphoglucomutase-relatedprotein and sarcoplasmic reticulum phosphoglucomutase each have50% or more identity with rabbit muscle phosphoglucomutase atthe amino acid level and all are reported to exhibit no or minorphosphoglucomutase activity. There are four major insertionsand two deletions in the parafusin sequence relative to PGM,all of which are located in surface-exposed loops connectingsecondary structural elements. The remaining amino acid substitutionsare distributed throughout the sequence and are not predictedto alter the polypeptide fold. Parafusin contains a putativeprotein kinase C site located on a surface loop in domain IIthat is not present in the homologs. Although the general domainstructure and the active site of rabbit muscle phosphoglucomutaseare preserved in the model of phosphoglucomutase-related protein,a major structural difference is likely to occur in domain 1due to the absence of 55 amino acid residues in PGM-RP. Thisdeletion predicts the loss of three -helices and one ß-strandfrom an anti-parallel ß-sheet in this domain as comparedwith the rabbit muscle phosphoglucomutase.