Stereochemical analysis on protein structures - Lessons for design, engineering and prediction

Abstract

The results of the present analysis demonstrate that local van der Waals contacts are overwhelmingly dominant in determining the backbone conformations of amino acid residues in globular protein structures. The number of Ramachandran-disallowed non-Gly residues obtained from a carefully screened database of 110 unrelated high-resolution protein crystal structures (18,708 residues) is only 66 (0.3%). There are no examples of the occurrence of consecutive disallowed residues; a feature which may be considered in model building studies. The amino acids which most frequently adopt these disallowed conformations are Asn, Asp, His, Ser, and Thr. The preponderance of small polar residues with hydrogen bonding functions in the side chain suggests that the local compensating interactions may offset the strain energy penalty for adopting unfavorable (?), (?) values.

2. Disallowed Conformations

Disallowed residues also form clusters in (?), (?) space, suggesting that specific (?), (?) distortions are preferred under compulsions imposed by local constraints. The clusters in the lower right-handed quadrant of the Ramachandran map are largely constituted by residues adopting ?-turn conformations or those occupying the i+1 position of type-II ?-turns. While no clear correlation can be drawn at present, it is possible that strained backbone conformations may have a special significance at enzyme active sites. Indeed, the present dataset provides several examples. The increasing number of independent protein structure determinations permits comparisons between the structures of closely related proteins and sometimes even the comparison of the same protein determined independently by different groups of investigators. Significantly, 33 disallowed residues are in fact conserved in 22 proteins where multiple structure determinations are available. The analysis carried out recently on a dataset which consists of structures solved at higher resolution (<1.5 Å) confirms these conclusions (see Appendix A of this thesis).

While conformationally disallowed residues constitute only a very small fraction of the total number of residues in available protein structures, there are several instances where backbone stereochemical distortions are clearly an integral feature of the folded conformation. These sites may provide attractive targets for future mutagenesis studies aimed at relating local stereochemistry with energetics of folded structures. The present analysis suggests that motifs, largely determined by local interactions, which act to terminate secondary structure elements may be recognizable in protein sequences. The analysis also establishes that an ?-helix can effectively be terminated either by left-handed helical (?L) or by extended (E) conformations, with the two modes of termination exhibiting unique local structural features. The former leads to the classical Schellman motif (Schellman, 1980; Milner-White, 1988) with Gly being overwhelmingly preferred at the terminating position (Aurora et al., 1994; Viguera and Serrano, 1995). For the E-terminated structures, there is a remarkable preference of Pro residues to follow the helix-terminating residue. The analysis reveals that Pro residues flanked by polar amino acids have a very strong tendency to terminate helices. The local sequence patterns that determine whether a helix will terminate near a Pro residue or continue with incorporation of Pro into the body of the helix have been identified. The two types of helix termination (?L, E) signals also differ dramatically in their solvent accessibility. Gly and Pro residues at helix termini appeared to be strongly conserved in homologous sequences.

The importance of short-range structural features like ?-turns (Venkatachalam, 1968) in determining folding and stability has been emphasized in a recent study on an immunoglobulin variable domain, which correlates database analysis with the results of site-directed mutagenesis (Ohage et al., 1997). Although polypeptide helices have been known for nearly half a century, the precise sequences and structural features which act as helix start and stop signals have still not been completely defined. A detailed understanding of stereochemical punctuation marks encoded in protein sequences will facilitate more definitive structure prediction and permit rational design of synthetic mimics for protein structures.

The present analysis has revealed several important conformational and compositional features of ?-hairpins which may be of value in peptide design and protein engineering. The design of hairpins with the loop segment ranging from 2 to 4 residues appears to be a distinct possibility in view of the strong preference for specific conformational features in short loops. Indeed, the earlier realization by Sibanda and Thornton (1985) that type I' and type II' ?-turns are frequently found in hairpins has led to the successful design of synthetic peptides that adopt hairpin conformation, with D-Pro residues acting as a strong conformational determinant (Awasthi et al., 1995), permitting crystallographic characterization (Karle et al., 1996b). The present analysis suggests that 2-residue loops with longer strand-length hairpins nucleated by type I / ?H turns may also be attractive targets for future design, as they are fairly widespread.

Furthermore, the rational design of three and four-residue loops appears possible in view of the strong preferences for the type I-?L motif in the former and the ?R-?R-?R-?L motif in the latter. Also, the motifs ?-type I'-P, ?-type II'-P, ?-type I-?L-P, and ?-?R-?R-?R-?L-P have comparable propensity for the nucleation of ?-hairpins, and ?-type I-?L-P and ?-type I'-P have the same sense of twist. The strong preference for specific amino acid residues in the loop segments, together with rigid conformational requirements, augurs well for the design of consensus loops.

4. ?-Hairpins in Proteins

The additional possibility of introducing covalent constraints by disulfide bridging across ?-hairpins provides a means of locking specific conformations, a feature that has indeed been realized in short synthetic peptides (Karle et al., 1988). The absence of strong preferences of amino acids in strands and of strong pair correlations across strands suggests that a high degree of sequence variability can be built into designed hairpin structures. ?-Turn interconversion appears to be a significant feature of protein crystal structure, as demonstrated by the analysis of protein crystal structures. Such conformational interconversion leaves the adjacent residues largely unperturbed, and the concerted mechanism may indeed provide a convenient pathway for rationalizing ?-turn dynamics. Of the 55 ?-turns, 7 examples are found to occur in the loop region of ?-hairpins, indicating the formation of ordered secondary structures on either side of the ?-turn does not preclude local conformational dynamics. ?-Turns, undergoing conformational interconversions, are mostly located at the surface of the proteins. Gly has an overwhelming preference to occur at the i+2 position of the flippable ?-turns, with the majority of the examples having Pro at the i+1 position. In several cases (16 examples), ?-turns in which the sequence of residues is totally conserved were observed to undergo conformational interconversions. The computational results suggest that concerted flips of central peptide units involving correlated single bond rotation can occur with a low activation energy barrier (nearly 3 kcal/mol).

The results of the analysis presented here establish that the peptide backbone, which adopts a left-handed helical conformation, plays an important role in structure and hence in the function of the proteins. The classification suggests that most of the residues adopting ?L conformations in protein structures occur in the hydrogen-bonded motifs, like ?-turns, Schellman motifs, etc. These motifs are frequently found in the loop regions of ?-hairpins and in the helix-terminating positions (see Chapters 3 and 4). Examination of the backbone dihedral angles (?, ?) for the Gly and non-Gly residues adopting ?L conformation shows that the Gly residues have a greater tendency to occur near ? = 84° and ? = 180°, and the non-Gly residues near ? = 58° and ? = 36°. This apparent tendency is reflected in the preferences of amino acids in the ?-turns.

The results of the present analysis also establish that the peptide backbones, which show a marked structural similarity to the V3 loop tip sequence, can fall into two distinct families which are observed in many well-determined protein crystal structures. A strong preference for the amino acid at residue 3 has been established, with Gly overwhelmingly favored in family 1 and Asn/Asp in family 2. These observations are significant for the design of synthetic analog peptides, which may be used as viral antagonists or synthetic antigens.

The use of D-residues at position 3 to anchor ?L conformations and helix-stabilizing residues at position 5 or 6 would leave the critical Pro-2, Arg-4, and Ala-5 residues intact. Indeed, a more recent V3 loop peptide design study has replaced Ala-5 by Aib, which is a strong helix-promoting residue, in order to conformationally constrain the tip sequence (Ghiara et al., 1997). The mutagenesis of selected surface-exposed sequences from Table 6.6 and the incorporation of the critical central tetrapeptide (PGRA) segment into a surface loop of a soluble protein by insertion mutagenesis, as demonstrated for a cell adhesion sequence (Yamada et al., 1994), are attractive possibilities.

The known three-dimensional crystal structure of membrane proteins belongs to two distinct structural classes, helical bundles and ?-barrels. The preliminary comparison of transmembrane and globular protein helices reveals that (i) most residues have similar N-cap, C-cap, and C-tail preference and only a few amino acids show differences, for example, Glu and Trp at the N-cap; Lys, Ser, and Tyr at the C-cap; Cys, Glu, and Thr at the C-tail; Cys at the middle. The incorporation of the helix positional preference into the transmembrane helix prediction algorithms must aid in the accurate prediction of start and stop positions of helices. Hydrophobic sequences long enough to form six or seven turns are assumed to adopt helix in the membrane environment. The recent structural studies on the nicotinic acetylcholine receptor, however, question such assumptions. The comparison of buried residues in the globular protein and the transmembrane segments shows that they have similar characteristics in the residue composition. The use of secondary structure information derived from the buried residues, which is a much larger dataset compared to transmembrane residues, in assessing the conformation of membrane-spanning helices is an attractive possibility.