The intrinsic conformational propensities of the 20 naturally occurring amino acids and reflection of these propensities in proteins

AA dipeptide: Effect of environment on conformational properties.

Because Ramachandran's original work was based on hard sphere space-filling models without solvent, we simulated the Ala-Ala dipeptide without solvent and with all partial charges set to zero. The distributions of (Φ,Ψ) angles in the uncharged solvent-free simulations are similar to the classically accepted Ramachandran preferences (Fig. 1A), with the dominant populations occurring in the α- and β-quadrants. The results are similar when water is included, but the system remains uncharged [supporting information (SI) Table S1] and Fig. 1B. However, there is a shift toward helical/turn conformers (the α_R and α_L quadrants) when standard, more realistic charges are used and water is included (Fig. 1C). The populations are tabulated in Table S1 by quadrant and by the specific conformational region [see labeled regions in Fig. 1 and specific (Φ,Ψ) definitions in Methods] for these different dipeptide scenarios. These results show that it is important to consider the environment: The picture provided by uncharged hard sphere in vacuo Ramachandran maps may not be accurate for solvated systems, particularly polar solvents (16). It is also noteworthy that there is little change in the distributions upon doubling the sampling from 50 to 100 ns (Table S1).

GGAGG: Sampling behavior.

Before analyzing the conformational properties of all 20 aa in depth, sampling and convergence must be addressed. For an end-capped pentapeptide, we assume that each (Φ,Ψ) pair for the three internal residues can exist in four possible conformations with respect to the quadrants in the Ramachandran map, yielding 64 (4³) possible conformational states. We exclude the first and last residues and the capping groups, because they appear to freely sample conformational space and are not neighbors of the central Ala residue. The fractional population of each of the 64 substates was compared among four independent GGAGG simulations (Fig. S1). Consideration of the conformational properties of three residues instead of just the central residue provides a much more sensitive and robust metric of convergence.

Over short time periods, the conformers sampled in two separate simulations were unlikely to be the same (Fig. S1A). However, as sampling approached the ergodic limit, the 64 conformations exhibited equivalent populations in two separate 100-ns trajectories (Fig. S1B). One way to measure the sampling convergence is to plot the correlation coefficient (R) of the conformer populations between two simulations (Fig. S1C). These two trajectories require ≈100 ns for the substate populations to be correlated to >95%, and at 50 ns, they are 90% correlated. Two additional trajectories starting from very different structures converged at the same rate. Based on these findings, all simulations were performed for 100 ns.

GGAGG: Analysis of conformational behavior.

Fig. 1 shows a net surface and the accompanying contour plot of the Ramachandran map of Ala for the capped GGAGG peptide (Fig. 1 D and E). All points sampled by Ala during the simulations are displayed (10⁵ points per 100-ns simulation). There are peaks in three different quadrants. The relative peak heights and volumes define Ala's sampling preferences. The same results are obtained in four independent 100-ns simulations. Ala spends ≈60% of its time in the α_R-quadrant, 32% in the β-quadrant, and 8% in the α_L-quadrant (Table 1).

The conformational space sampled by the peptide can be divided more accurately to reflect the underlying local structure adopted (Table 1), as discussed above. Also, given the large number of points and sampling of many different conformations in this small Gly-rich peptide, it can be helpful to convert the population scale into a free energy surface (Fig. 1F). The central Ala residue prefers the helical regions of (Φ,Ψ) space, but the bulk of the conformers are shifted from α_R to the elbow region, or near the α_R region, reflective of turns and kinks. Nevertheless, the peptide was very dynamic and covered 51% of the (Φ,Ψ) space during the simulation (using 5° × 5° bins). Detailed breakdowns of the conformations sampled and their populations over the 10⁵ structures from each simulation are presented in Table 1. Furthermore, these data are provided for the four independent GGAGG simulations and the results are comparable (Table 1).

The GGAGG peptide has been probed in depth by NMR experiments via proton and heavy atom chemical shifts (13, 14). Chemical shifts were calculated over the simulations by using ShiftX (17) for H_N, H_α, H_β, N_H, C_α, C_β, and C′ for GGAGG are very close to the corresponding experimentally derived values (correlation coefficient R = 0.999 over 28 points, 7 atoms × 4 independent simulations). The calculated and experimental values are provided in Table S2.

Another quantity from NMR that provides information regarding the conformations sampled by MD is the vicinal spin–spin coupling constant ³J_NHCα, which depends on the torsion angle φ (18). The Karplus equation is typically used to estimate φ from ³J_HNCα:

where θ = | φ−60° | (19). The coefficients differ depending on the parameterization and folded protein reference set used. The difference between the calculated and experimental values is 0–2.2 Hz (20), and a single coupling constant can map to four different φ values. Nevertheless, we have calculated the coupling constant for Ala in the four independent GGAGG simulations by using the Ludvigsen et al. (21) parameterization: A = 6.7, B = −1.3, C = 1.5, leading to 〈³J_HNCα〉 = 7.3 ± 1.8 Hz. The experimental values for Ala in GGXGG peptides range from 5.8 to 6.1 Hz (13–15). Although our calculated value is high, it agrees with experiment within our uncertainty.

The conformations sampled by Ala in the GGAGG peptide at 298 K by MD are not random and do not fully cover conformational space, even though this small Gly peptide is a model for “random coil” structures. Instead Ala has preferred conformations, namely local α_R, α_L, β, and P_II (quantified in Table 1), but these states are dynamic and interconvert readily. These simulations illustrate the potential problems associated with assuming a single or even dominant conformation based on experimental observables over large ensembles, and they call into question fitting procedures to “determine” populations, because different population distributions may reproduce the observables. Given the lack of repeating secondary structure in this peptide, the lack of side chain interactions influencing the local conformations populated, and the convergence reached both within and among independent simulations, we believe that the distributions obtained reflect the intrinsic conformational propensities of Ala. Consequently, we now expand upon this work to investigate the other 19 amino acids.

GGXGG pentapeptides.

Ramachandran plots were created for each of the amino acids from the distribution of instantaneous Φ and Ψ angles in the GGXGG simulations (Fig. 2). The plots revealed that all of the residues, except for Pro, exhibit a strong propensity for local turns, as reflected in the populations of the α_R and/or near-α_R regions. The α-helix peak is shifted from the strictly defined α-helical region (Fig. 2), which occurs because what we observe is not repeating structure, but the conformational properties of a single residue. The β-sheet, α_L, and P_II conformations were also populated to varying degrees (Table 1).

The most highly populated (Φ, Ψ) angles for GGAGG were Φ = (−175 to −50) and Ψ = (−55 to −5), which we define as near-α_R (Figs. 1 and 2). The average Φ and Ψ angles were −110° ± 41° and −12° ± 64°, respectively (calculated by using circular statistics). Consideration of the average in light of the actual conformational space sampled (Table 1 and Fig. 2) illustrates the problem of describing mixed ensembles in this way. As with the GGAGG peptide, we calculated proton and heavy atom chemical shifts over our ensembles for all GGXGG peptides (Table S2). Overall, the agreement between MD and experiment (14) is very good, with a correlation coefficient of 0.9998 for 151 values (Table S3). The agreement for N_H is lower than for the other atoms, possibly because of the difference in pH: MD was done at neutral pH and the experiments were at pH 2.3 in 8 M urea. N_H chemical shifts are known to be very sensitive to pH, temperature, and environment (13). The C_α chemical shifts are very sensitive to conformation, and the good agreement suggests that the MD-generated ensembles are reasonable (R = 0.9879 for the 20 aa, Table S3). Another noteworthy point is that the proton chemical shifts have been determined for the GGXGG peptide series under a variety of pH, temperature, and solvent conditions, and the experimental results are in good agreement with each other and with our MD-derived values (R ≥ 0.996) (Table S4).

For many of the residues (Asp, His, Phe, Trp, and Tyr), the near- α_R region was the most populated conformational state (Table 1). As mentioned above, Ala sampled 51% of the (Φ,Ψ) space (Table 1). Not surprisingly, the highest value was for Gly at 75% coverage. The lowest was 13% for Pro, followed by ≈30% for Asp and Ile. For comparison, the values for Ala and Gly at 498 K are 68 and 82%, respectively (22).

Free energies were calculated for α_R relative to all other conformations (Table S5). For example, for Ala, the P_II content was ≈15% in four independent simulations. Comparison of the populations yields ΔG_PII→αR = −0.3 kcal/mol. In addition, the free energy for helical conformers can be compared with experimentally derived consensus helical propensities from experimental host–guest studies of different peptide and protein “hosts” (Table S5) (23). Contrary to most experiments, Ala did not have the highest helical propensity; instead, Asn and then Leu displayed the strongest helical propensities in the GGXGG peptides. In fact, 14 of the amino acids had higher helical propensities than Ala.

Comparison of amino acid conformational preferences in peptides and proteins.

Fig. 2 shows a comparison of (Φ,Ψ) populations for three systems: (i) the GGXGG pentapeptides; (ii) MD simulations at 298 K of 188 native proteins with varying architecture from our Dynameomics project (24); and (iii) 5,626 structures from the nonredundant Astral40 Protein Data Bank (PDB)-derived database (25). In the native simulations and the PDB, the amino acid in question was not required to have flanking Gly residues. The native-state MD simulations and the experimental structures sample very similar (Φ,Ψ) regions, whereas the pentapeptides can have quite different preferences (Fig. 2). The Pro and Thr distributions are the most similar in the various simulations, or in other words, they are the least sensitive to environmental influence. Most other residues experienced shifts from the pentapeptide distributions, reflecting the importance of context in determining conformation. The large nonpolar residues (for example, Leu, Met, and Trp) showed the greatest difference between the pentapeptide and protein environments (Fig. 2). That these residues are more sensitive makes sense, given that they are subject to conformational constraints due to their burial in hydrophobic cores.