Water in protein structure prediction

Abstract

Proteins have evolved to use water to help guide folding. A physically motivated, nonpairwise-additive model of water-mediated interactions added to a protein structure prediction Hamiltonian yields marked improvement in the quality of structure prediction for larger proteins. Free energy profile analysis suggests that long-range water-mediated potentials guide folding and smooth the underlying folding funnel. Analyzing simulation trajectories gives direct evidence that water-mediated interactions facilitate native-like packing of supersecondary structural elements. Long-range pairing of hydrophilic groups is an integral part of protein architecture. Specific water-mediated interactions are a universal feature of biomolecular recognition landscapes in both folding and binding.

Fig. 1.

The detailed interactions in the bioinformatic prediction energy functions are compared. We partition 210 interresidue pairs into 165 pairs having at least one hydrophobic partner (hydrophobic group) and 45 pairs having only charged and polar residues in the pair (polar group). More positive values for the matrix elements indicate more favorable interactions. (A) The 165 interaction matrix elements for first-well interactions among residue pairs having at least one hydrophobic partner are shown. The hydrophobic group first-well interactions show very similar profiles among the AM/C and AM/W potentials. (B) The 45 interactions for first-well interactions among all charged and polar residue pairs are shown. The first-well polar group contact interactions are rather similar in each potential, except for the most charged pairs, which are more destabilized in the AM/W-0 and AM/W-1 potentials. (C) The 45 interactions for second-well protein-mediated (filled symbols) and second-well water-mediated (dotted symbols) interactions among all charged and polar residue pairs are shown. Again the second-well protein-mediated interactions in AM/W are similar to those in AM/C, but the AM/W water-mediated interactions stand out as being different (see the text for discussion).

Fig. 2.

Structure prediction performance and the comparison of AM/C, AM/W-0, and AM/W-1 potentials. The maximum Q scores versus chain length attained during five annealing runs for each of 14 proteins using three different potentials are shown. PDB codes for the training proteins are in violet, and the test proteins are in green.

Fig. 3.

Structure predictions for ferritin, PDB code 2FHA. (A) The best (of five for each potential) Q-score annealing trajectories are shown for three different potentials. (B) The average thermodynamic energy vs. Q.(C) Superposition of the AM/W-0 best Q-score structure (blue) and the native structure (red) is indicated. Spheres indicate charged residue Cα atoms. (D) The distance plot for the AM/W-0 best Q-score structure (blue, upper triangle) and the native structure (red, lower triangle). (E) Distance plot for the AM/C best Q-score structure (blue, upper triangle) and the native structure (red, lower triangle) are compared. In the AM/W-0 structure (D), only a small number of contacts are missing and a small registry shift near residue 70 occurs. In the AM/C structure (E) the C-terminal half misses on a major interhelical interface.

Fig. 4.

Structure predictions for 1HdeA, PDB code 1bg8 (CASP3). (A) A superposition of the best Q-score structure from the AM/W-1 potential (blue) and the native structure (red) is shown. Spheres indicate charged residue Cα atoms. (B) The superposition of the best Q-score structure from the AM/C potential (blue) and the native structure (red) is shown. (C) Free energy vs. Q as computed with a histogramming technique. (D) Annealing trajectories of individual fragment Q scores, large N-terminal fragment containing residues 1–61, and small C-terminal domain containing residues 62–76 are shown as a function of the instantaneous temperature through the run. (E) Annealing trajectories of interfragment Q scores are indicated. (F) Annealing trajectories of interfragment Q scores partitioned into the first-well and second-well contributions are shown.

Fig. 5.

Structure predictions for CASP5 target protein T129a (PDB code 1IZM, structural information not yet officially released at the time of writing). (A) The distance plot for AM/W-1 best Q-score structure (blue, upper triangle) and the native structure (red, lower triangle) is shown. (B) The distance plot for AM/W-1 structure with the best sum of individual domain Q scores (blue, upper triangle) and the native structure (red, lower triangle) is shown. (C) Annealing trajectories of individual domain Q scores, N-terminal domain containing residues 1–75, and C-terminal domain containing residues 76–170 are indicated. (D) Annealing trajectories of interdomain Q scores are shown. (E) Annealing trajectories of interdomain first-well Q scores are plotted. (F) Annealing trajectories of interdomain second-well Q scores are plotted.