Complementary changes

Do different interactions stabilise protein structures having similar folds? Figure 14 shows the distribution of the percentage of complementary changes versus . For some type and similarities, the proportion of complementary changes is as high as of the total number of possible interacting pairs. Some pairs of 3D structures with an extraordinarily high number of complementary changes are labelled in the Figure. Many similar 3D structures have a proportion of complementary changes similar to that expected by chance (Table 4), suggesting a fundamental difference in how they are stabilised. The Figure suggests that interactions between residues at equivalent positions in similar 3D structures can differ substantially in character, and has many implications for methods which attempt to use protein 3D structural information to find sequences compatible with a fold. In particular, methods not taking long-range interactions into account [Johnson et al., 1993][Overington et al., 1992][Overington et al., 1990][Bowie et al., 1991] will encounter difficulties in differentiating many genuine structural similarities from noise, since they will be unable to detect such complementary changes.

Though comparatively rare, many interesting varieties of complementary changes occur within protein structure pairs. Most involve an interaction changing from a predominantly hydrophobic pair to a charge pair or a pair of polar residues. Five examples are shown. In all of the examples, the regions shown are extracted from a larger structural alignment and superimposition and the residues shown to be in contact fall within core or structurally equivalent regions and all have relative accessibilities of less than .

The first example Figure 15a shows how residues interacting between a strand and an helix vary in one - - supersecondary structure in the barrel family of protein structures. In xylose isomerase (7XIA) two hydrophobic residues (Phe and Leu) are in contact; in trimethylamine dehydrogenase (1TMD) the hydrophobic residues are replaced by a charge pair (Glu and Lys); and in Rubisco (8RUB_L) the two residues are cysteines (though not oxidised to form a disulphide). The second example Figure 15b shows how residues on one sheet within the Rossmann fold family of structures differ. In glycogen phosphorylase (1GPB) two largely electrostatic interactions (Glu with Gln and Asp with Arg) are replaced by two hydrophobic interactions in 6-phosphogluconate dehydrogenase (PGD; Ile with Leu and Ile with Phe). The third example Figure 15c shows how a disulphide bond around the `pin' in Ig folds (2FBJ_Lconstant domain in the Figure) is replaced in the antibacterial protein macromycin (2MCM) by a hydrophobic (Val - Ala) interaction. The fourth example Figure 15d shows how a helix-strand interaction differs between two Rossmann fold domains. In malate dehydrogenase (4MDH) two Leucine residues are in contact; in glyceraldehyde-3-phosphate dehydrogenase a histidine forms a hydrogen bond with a glutamic acid. The fifth example Figure 15e shows how an electrostatic interaction (Asp - His) within innkeeper worm haemoglobin (1ITH_A) is replaced by a hydrophobic interaction (Phe - Ile) in the bacterial toxin colicin A (1COL_A).

Much recent work has concentrated on attempting to predict 3D contacts in proteins of unknown structures by analysis of complementary changes in multiple protein sequence alignments [Göbel et al., 1994][Neher, 1994][Shindylav et al., 1994][Taylor \& Hatrick, 1994]. The details of these previous studies differ, though the general conclusion is that it is not possible to predict such contacts with confidence. Though perhaps not directly comparable, the results of this study shed some light on why these predictive methods are unsuccessful. Although subtle changes to side chains can have disastrous effects on specific protein function (e.g., Lim &Sauer, 1989), pairs of genuinely similar 3D protein structures (even with similar, but not identical, functions) can have very different patterns of long range side-chain to side-chain interactions. This observation would suggest that the detection of long-range interactions by slight compensations in side-chain volume seen in sequence alignments may be difficult. A search for the types of complementary changes described in this study might prove more fruitful, since the change of a pair of hydrophobics to a charge pair or pair of polar residues is more likely to correlate to spatial separation [Neher, 1994]. However, Figure 14 shows that these are very rare events in proteins having similar sequences (i.e., type similarities or those sequences that are alignable without resort to 3D structure comparison), so detection of such sites from multiple sequence alignments is likely to prove difficult.