Progress in predicting protein function from structure: unique features of O-glycosidases.

The Structural Genomics Initiative promises to deliver between 10,000 and 20,000 new protein structures within the next ten years. One challenge will be to predict the functions of these proteins from their structures. Since the newly solved structures will be enriched in proteins with little sequence identity to those whose structures are known, new methods for predicting function will be required. Here we describe the unique structural characteristics of O-glycosidases, enzymes that hydrolyze O-glycosidic bonds between carbohydrates. O-glycosidase function has evolved independently many times and enzymes that carry out this function are represented by a large number of different folds. We show that O-glycosidases none-the-less have characteristic structural features that cross sequence and fold families. The electrostatic surfaces of this class of enzymes are particularly distinctive. We also demonstrate that accurate prediction of O-glycosidase function from structure is possible.

Examination of the folding of E. coli CspA through tryptophan substitutions.

Escherichia coli cold shock protein, CspA, folds very rapidly (time constant, tau = 4 msec) by an apparent two-state mechanism. However, recent time-resolved infrared (IR) temperature-jump experiments indicate that the folding trajectory of CspA may be more complicated. The sole tryptophan of wild-type CspA (Trp11), which is used to monitor the folding process by fluorescence spectroscopy, is located in an unusual aromatic cluster on the surface of CspA within the nucleic acid binding site. To gain a more global picture of the folding kinetics of CspA and to determine if there are any previously undetected intermediates, we have introduced a second tryptophan at three different surface locations in the protein. The three mutations did not significantly alter the tertiary structure of CspA, although two of the substitutions were found to be slightly stabilizing. Two-state folding, as detected by stopped-flow fluorescence spectroscopy, is preserved in all three mutants. These results indicate that the fast folding of CspA is driven by a concerted mechanism.

Contributions of residue pairing to beta-sheet formation: conservation and covariation of amino acid residue pairs on antiparallel beta-strands.

In an effort to better understand beta-sheet assembly, we have investigated the evolutionary behavior of neighboring residues on adjacent antiparallel beta-strands. Residue pairs were classified according to solvent exposure as well as by whether their backbone NH and C==O groups are hydrogen bonded. The conservation and covariation of 19,241 pairs in 219 sequence alignments was analyzed. Buried pairs were found to be the most conserved, while stronger covariation was detected in the solvent-exposed pairs. However, residues on neighboring strands showed a degree of conservation and covariation similar to that of well-separated residues on the same strand, suggesting that evolutionary pressure to maintain complementarity between pairs on neighboring strands is weak. Moreover, in spite of the preference of certain amino acid pairs to occupy neighboring positions on adjacent strands, such favored pairs are neither more strongly mutually conserved nor covary more strongly than pairs of the same type in non-interacting positions. Although the beta-sheet pairs did not show outstanding evolutionary coupling, in many protein families significant conservation and covariation patterns were detected for some of the residue pairs. Overall, the weak evolutionary conservation and covariation of the beta-sheet pairs indicates that sheet structure is unlikely to be dictated by specific side-chain interactions.

Role of a solvent-exposed aromatic cluster in the folding of Escherichia coli CspA.

Escherichia coli CspA is a member of the cold shock protein family. All cold shock proteins studied to date fold rapidly by an apparent two-state mechanism. CspA contains an unusual cluster of aromatic amino acids on its surface that is necessary for nucleic acid binding and also provides stability to CspA (Hillier et al., 1998). To elucidate the role this aromatic cluster plays in the determining the folding rate and pathway of CspA, we have studied the folding kinetics of mutants containing either leucine or serine substituted for Phe 18, Phe20, and/or Phe31. The leucine substitutions are found to accelerate folding and the serine substitutions to decelerate folding. Because these residues exert effects on the free energy of the folding transition state, they may be necessary for nucleating folding. They are not responsible, however, for the very compact, native-like transition state ensemble seen in the cold shock proteins, as the refolding rates of the mutants all show a similar, weak dependence of unfolding rate on denaturant concentration. Using mutant cycle analysis, we show that there is energetic coupling among the three residues between the unfolded and transition states, suggesting that the cooperative nature of these interactions helps to determine the unfolding rate. Overall, our results suggest that separate evolutionary pressures can act simultaneously on the same group of residues to maintain function, stability, and folding rate.

Predicting protein function from structure: unique structural features of proteases.

We have noted consistent structural similarities among unrelated proteases. In comparison with other proteins of similar size, proteases have smaller than average surface areas, smaller radii of gyration, and higher C(alpha) densities. These findings imply that proteases are, as a group, more tightly packed than other proteins. There are also notable differences in secondary structure content between these two groups of proteins: proteases have fewer helices and more loops. We speculate that both high packing density and low alpha-helical content coevolved in proteases to avoid autolysis. By using the structural parameters that seem to show some separation between proteases and nonproteases, a neural network has been trained to predict protease function with over 86% accuracy. Moreover, it is possible to identify proteases whose folds were not represented during training. Similar structural analyses may be useful for identifying other classes of proteins and may be of great utility for categorizing the flood of structures soon to flow from structural genomics initiatives.

Protein folding and unfolding on a complex energy landscape.

Recent theories of protein folding suggest that individual proteins within a large ensemble may follow different routes in conformation space from the unfolded state toward the native state and vice versa. Herein, we introduce a new type of kinetics experiment that shows how different unfolding pathways can be selected by varying the initial reaction conditions. The relaxation kinetics of the major cold shock protein of Escherichia coli (CspA) in response to a laser-induced temperature jump are exponential for small temperature jumps, indicative of folding through a two-state mechanism. However, for larger jumps, the kinetics become strongly nonexponential, implying the existence of multiple unfolding pathways. We provide evidence that both unfolding across an energy barrier and diffusive downhill unfolding can occur simultaneously in the same ensemble and provide the experimental requirements for these to be observed.

Context-dependence of amino acid residue pairing in antiparallel beta-sheets.

In an effort to understand the driving forces behind antiparallel beta-sheet assembly, we have investigated the mutational tolerance of four pairs of residues in CspA, the major cold shock protein of E. coli. Two buried pairs and two exposed pairs of neighboring amino acids were separately randomized and the corresponding effects on protein stability were assessed using a protein expression screen. The thermal denaturation of a subset of the recovered proteins was measured by circular dichroism spectroscopy in order to determine the range of stabilities sampled by the expressed mutants. As anticipated, buried sites are substantially less tolerant of substitutions than exposed sites with more than half of the exposed residue combinations giving rise to stably folded proteins. The two exposed residue pairs, however, display different degrees of tolerance to substitution and accept different residue pair combinations. Except for the prohibition of proline from interior strand positions, no obvious correlations of mutant stability with any single parameter such as beta-sheet propensity or hydrophobicity can be detected. Mutant combinations recovered in both orientations (e.g. XY and YX) at a given exposed pair site often show markedly different stabilities, indicating that the local environment plays a substantial role in modulating the pairing preferences of residues in beta-sheets.

Coupling protein stability and protein function in Escherichia coli CspA.

BACKGROUND: CspA is a small protein that binds single-stranded RNA and DNA. The binding site of CspA consists of a cluster of aromatic amino acids, which form an unusually large nonpolar patch on the surface of the protein. Because nonpolar residues are generally found in the interiors of proteins, this cluster may have evolved to bind nucleic acids at the expense of protein stability. RESULTS: Three neighboring phenylalanines have been mutated singly and in combination to leucine and to serine. All mutations adversely affect DNA binding. Surprisingly, all mutations, and especially those to serine, are destabilizing. CONCLUSIONS: The aromatic cluster in CspA is required not only for protein function but also for protein stability. This result is pertinent to the design of beta-sheet proteins and single-stranded nucleic acid binding proteins, whose binding mode is proposed to be of aromatic-aromatic intercalation.

Tolerance of a protein helix to multiple alanine and valine substitutions.

BACKGROUND: Protein stability is influenced by the intrinsic secondary structure propensities of the amino acids and by tertiary interactions, but which of these factors dominates is not known in most cases. We have used combinatorial mutagenesis to examine the effects of substituting a good helix-forming residue (alanine) and a poor helix-forming residue (valine) at many positions in an alpha helix of a native protein. This has allowed us to average over many molecular environments and assess to what extent the results reflect intrinsic helical propensities or are masked by tertiary effects. RESULTS: Alanine or valine residues were combinatorially substituted at 12 positions in alpha-helix lambda repressor. Functional proteins were selected and sequenced to determine the degree to which each residue type was tolerated. On average, valine substitutions were accommodated slightly less well than alanine substitutions. On a positional basis, however, valine was tolerated as well as alanine at the majority of sites. In fact, alanine was preferred over valine statistically significantly only at four sites. Studies of mutant protein and peptide stabilities suggest that tertiary interactions mask the intrinsic secondary structure propensity differences at most of the remaining residue positions in this alpha helix. CONCLUSIONS: At the majority of positions in alpha-helix lambda repressor, tertiary interactions with other parts of the protein can be viewed as an environmental "buffer" that help to diminish the helix destabilizing effects of valine mutations and allow these mutations to be tolerated at frequencies similar to alanine mutations.

Stability and folding properties of a model beta-sheet protein, Escherichia coli CspA.

Although beta-sheets represent a sizable fraction of the secondary structure found in proteins, the forces guiding the formation of beta-sheets are still not well understood. Here we examine the folding of a small, all beta-sheet protein, the E. coli major cold shock protein CspA, using both equilibrium and kinetic methods. The equilibrium denaturation of CspA is reversible and displays a single transition between folded and unfolded states. The kinetic traces of the unfolding and refolding of CspA studied by stopped-flow fluorescence spectroscopy are monoexponential and thus also consistent with a two-state model. In the absence of denaturant, CspA refolds very fast with a time constant of 5 ms. The unfolding of CspA is also rapid, and at urea concentrations above the denaturation midpoint, the rate of unfolding is largely independent of urea concentration. This suggests that the transition state ensemble more closely resembles the native state in terms of solvent accessibility than the denatured state. Based on the model of a compact transition state and on an unusual structural feature of CspA, a solvent-exposed cluster of aromatic side chains, we propose a novel folding mechanism for CspA. We have also investigated the possible complications that may arise from attaching polyhistidine affinity tags to the carboxy and amino termini of CspA.

The origins of protein secondary structure. Effects of packing density and hydrogen bonding studied by a fast conformational search.

Globular proteins fold to create compact structures rich in alpha-helices and beta-sheets. While studies of cubic lattice models of simplified polypeptide chains have concluded that secondary structure is a necessary consequence of chain compactness, different conclusions have been reached from studies of off-lattice models of simplified chains. In an attempt to resolve this controversy, we study an all-atom off-lattice model of a protein subject to a variety of simplified energy functions. A Monte Carlo simulated annealing algorithm is used to search conformational space quickly. The algorithm uses pivot-type moves in which a residue is selected at random and the values of its main-chain dihedral angles are changed. The energy function used to accept or reject moves is taken to be either a term proportional to the volume occupied by a structure (to mimic the hydrophobic effect), a term proportional to the energy of main-chain hydrogen bonding, or a combination of these two terms. Secondary structure content is evaluated using several different definitions. For all the definitions used, compactness alone produces a 10% increase in secondary structure content. However, this is a small fraction of the secondary structure observed in native protein structures. Structures produced by minimizing the hydrogen bond energy have extensive secondary structure but are not densely packed. Structures having both the high density of native structures and extensive secondary structure are produced by minimizing combinations of the volume and hydrogen bond energy terms. Our results emphasize the close relationship between secondary structure and the geometry of main-chain hydrogen bonding. The results are consistent with a description of protein folding in which the hydrophobic effect favors dense packing while hydrogen bonding determines the specific local geometry which generates secondary structure. To make an analogy with lattice studies of packing density and secondary structure, it seems that hydrophobicity provides the packing density while hydrogen bonding provides the lattice.

Additivity of mutant effects assessed by binomial mutagenesis.

Eleven amino acid positions in the helix-turn-helix of lambda repressor have been mutagenized by using a combinatorial method in which alanine is substituted at each position with a probability of 0.5. Approximately 25% of the 2048 proteins in the resulting binomial library are active, including some variants with as many as seven alanine substitutions. The frequency of alanine substitutions in the set of active variants is a measure of the importance of the wild-type residue at each mutagenized position, and comparison of the frequencies of pairwise mutations with those expected based upon the single-position frequencies allows the additivity of mutant effects to be tested. For the positions examined here, we find that the effects of multiple substitutions are largely additive and are able to predict the activity class of the binomial mutants with 90% accuracy by using a model that simply sums penalty scores derived from the alanine substitution frequencies. We also find, however, that several residue pairs, including some that are distant in the three-dimensional structure, do display nonadditive effects that appear to be statistically significant.

Modeling side-chain conformation for homologous proteins using an energy-based rotamer search.

We have developed a computational method for accurately predicting the conformation of side-chain atoms when building a protein structure from a known homologous structure. A library of rotamers is used to model the side-chains, allowing an average of five to six different conformations per residue. Local sites of adjacent side-chains are defined throughout the protein, and all combinations of side-chain rotamers are evaluated within each site using a molecular mechanics force field enhanced by the inclusion of a solvation term. At each site, the lowest energy combination of side-chains is identified and added onto the fixed protein backbone. A series of test cases using the refined X-ray structure of alpha-lytic protease has shown that: (1) the force field can correctly predict up to 90% of side-chain rotamers; (2) the assumption of side-chain rotamer geometry is usually a very good approximation; and (3) the complete combinatorial conformation search is able overcome local minima and identify the lowest energy rotamer set for the protein in the absence of a starting bias to the correct structure. Tests with several pairs of homologous proteins have shown that the algorithm is quite successful at predicting side-chain conformation even when the protein backbone used to generate side-chain positions deviates from the correct conformation. The root-mean-square (r.m.s.) deviation of predicted side-chain atoms rises from 1.31 A (average r.m.s.d. 0.73 A) in a test case with the correct backbone to only 2.68 A (1.95 A average r.m.s.d.) in a test case with < 35% homology. The high accuracy of this method suggests that it may be a useful automated tool for modeling protein structure.

Arresting tissue invasion of a parasite by protease inhibitors chosen with the aid of computer modeling.

Computer modeling of the three-dimensional structure of an enzyme, based upon its primary sequence alone, is a potentially powerful tool to elucidate the function of enzymes as well as design specific inhibitors. The cercarial (larval) protease from the blood fluke Schistosoma mansoni is a serine protease hypothesized to assist the schistosome parasite in invading the human circulatory system via the skin. A three-dimensional model of the protease was built, taking advantage of the similarity of the sequence of the cercarial enzyme to the trypsin-like class of serine proteases. A large hydrophobic S-1 binding pocket, suspected from previous kinetic studies, was located in the model and confirmed by new kinetic studies with both synthetic peptide substrates and inhibitors. Unexpected structural characteristics of the enzyme were also predicted by the model, including a large S-4 binding pocket, again confirmed by assays with synthetic peptides. The model was then used to design a peptide inhibitor with 4-fold increased solubility, and a series of synthetic inhibitors were tested against live cercariae invading human skin to confirm that predictions of the model were also applicable in a biologic assay.

Protein folding. Effect of packing density on chain conformation.

Recent lattice polymer simulations by Chan & Dill suggest that compactness may be a significant driving force in the formation of secondary structure. We have addressed the robustness of this conclusion for non-lattice polymers using a rotational isomeric model of proteins. Boundary conditions are used to enforce compactness and excluded volume effects are explicitly incorporated. As in the cubic lattice studies, compactness is seen to influence secondary structure content. This effect is modest for densities comparable to native proteins but dramatic for chains that are approximately 30% more dense than native proteins. alpha-Helical structure is common but beta-sheet structure is rare. It appears that lattices impart to compact chains an organizational bias that favors beta-sheet structure. The strengths and weakness of various simplified representations of polypeptide chains are also discussed.

Hydrogen bonds involving sulfur atoms in proteins.

Intrachain hydrogen bonds are a hallmark of globular proteins. Traditionally, these involve oxygen and nitrogen atoms. The electronic structure of sulfur is compatible with hydrogen bond formation as well. We surveyed a set of 85 high-resolution protein structures in order to evaluate the prevalence and geometry of sulfur-containing hydrogen bonds. This information should be of interest to experimentalists and theoreticians interested in protein structure and protein engineering.

Protein model structure evaluation using the solvation free energy of folding.

A systematic study of solvation free energy of folding for proteins with known crystallographic structures is presented. There is a linear relationship between the solvation free energy of folding and the protein size. This relationship, which can be rationalized by a simple model of chain folding, allows prediction of the solvation free energy of folding for proteins for which no high resolution structures are available. All misfolded structures analyzed show solvation free energies higher than predicted; however, some of the misfolded structures have values close enough to the predicted values so that one must be very careful when using such a criterion to check the correctness of a protein model.

Novel method for the rapid evaluation of packing in protein structures.

There has been considerable effort to predict the structure of proteins from their amino acid sequences. A major problem in all prediction efforts has been that, short of a direct comparison with crystallographic co-ordinates, it is often difficult to evaluate the merit of a model, or "proposed" protein structure. Here, we present a method for evaluating proposed protein structures that does not require a structural model of complete atomic detail. Our method evaluates residue-residue packing density using a simplified model of the polypeptide chain where amino acids are represented as one, two (histidine, tyrosine and phenylalanine), or three (tryptophan) spheres. This method also gives a measure of the appropriateness of residue-residue contacts, thus giving a measure of the amino acid distribution throughout the protein. Amino acid packing and amino acid distribution, as evaluated by this technique, are consistent with the accuracy of model-built structures. We have been able to select the best structures from a set of combinatorially generated models using this method, and we anticipate that it will be useful as a general tool for model-building.