Introduction

Protein sequence-structure relationships are of fundamental importance to the field of protein physical chemistry1,2,3,4. Intrinsic backbone conformational preferences of 20 amino acids determine the local structure of unfolded protein chains; these intrinsic preferences might guide the folding processes at early stages of protein folding. From this respect, the intrinsic backbone conformational preferences of different amino acids are part of the "folding mechanism" that remains poorly understood after more than 50 years since the protein folding question was raised1,2,3,4. Currently, to predict protein structure from amino acid sequences, database-based strategies are more successful than the physics-based algorithms. Advances in the physics-based algorithms demand continuous improvements in force field accuracy. The intrinsic backbone conformational preference data are crucial for this purpose.

Among three major backbone conformations, α-structure is relatively minor compared to polyproline II (PII) and β-conformations in model unfolded peptides as demonstrated by recent lines of independent evidence5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29. Reliable derivation of populations of the minor α-conformers in model peptides that exist in multiple equilibriums of different backbone conformations remains a challenge11,18,19,23,24,25,27. NMR measurements can only be carried out on a slow time scale as compared to backbone conformers’ lifetimes which lie in the range of 10–200 ps, conformational averaging over different conformers occurs during NMR measurements. The optical spectroscopy results are measured on a fast time scale and various optical spectra can be used to detect different backbone conformations6,8,9,16,17,23. However, most optical techniques suffer from their resolutions: band overlapping in the measured spectra generally cause significant uncertainties during quantitative analysis, particularly for accurate derivations of minor conformers. In our previous work, quantitative account of the sampled conformation in AcGGXGGNH2 and XAO by NMR 3J(Hα-HN) (3JαN) coupling constants was carried out through a two-state analysis for an equilibrium mainly between PII and β conformations; α-population was ignored completely as an approximation11,18. Here we have designed two series of peptides: AcGXPNH2 and AcGXGNH2 (X ≠ Gly, Pro). Combined analyses on data from both series allow us to derive the populations of three major conformers including PII, β and α in AcGXGNH2.

Proline is unique among the amino acids in that it has a five-membered ring which has a dramatic effect on the conformational preferences of the preceding residue. In AcGXPNH2 peptides, X can only adopt PII or β conformations as steric clashes between the Cδ of proline and both the Cβ and amide nitrogen of residue X make α-conformation inaccessible to residue X30,31,32. With the measured 3JαN coupling constants of X, previous procedure through a two-state analysis for the equilibrium between PII and β is justified for AcGXPNH2 peptides11,18. PII to β population ratio for each of AcGXPNH2 can be determined; assuming the ratio for X in AcGXPNH2 and AcGXGNH2 is approximately the same, we can derive the population of α-conformer in AcGXGNH2 peptides through equation (1), see Supplementary Information for derivation of the equation in which xα(GXG) denotes the percentage of α-conformer in AcGXGNH2; 3JαN(GXP) and 3JαN(GXG), measured 3JαN coupling constants of X in AcGXPNH2 and AcGXGNH2; 3JαN(α), standard 3JαN coupling constant of a residue in α-helices.

Further, we can derive the populations of PII and β in AcGXGNH2 through equations (2) and (3) in which 3JαN(PII),3JαN(β) and 3JαN(α) denote standard 3JαN coupling constant of a residue in PII, β-, and α-conformations, respectively; xPII(GXG), xβ(GXG) and xα(GXG) denote the percentage of PII, β- and α-conformations in AcGXGNH2, respectively. With the derived percentage values, the free energy-conformation diagrams of AcGXGNH2 in aqueous solution can be constructed for the three major backbone conformations.

Derived results show that on average residue X in AcGXGNH2 adopt PII, β, and α 44.7%, 44.5% and 10.8% of time, respectively. Importantly, minor populated α-conformations of different amino acids in AcGXGNH2 determine their varying α-helix nucleation capabilities33. According to Zimm-Bragg theory34, helix-to-coil transition can be described by a nucleation constant σ and helix propagation constants s, the product σ∙s represents the probability of formation of an α-helical segment comprising three residues34,35,36. From our derived vales of xα, we can estimate the probability for Ala peptides, σ∙s = (xα)3 = 4.29 × 10−3 (xα = 0.1625 for Ala), this value is very close to those reported37. Our free energy-conformation diagrams would set a foundation for physics-based algorithmic developments for protein structure predictions38,39.

Results and Discussions

Model peptides AcGXPNH2 and AcGXGNH2 and their CD spectra

Our previous study on AcGGXGGNH2 peptides showed that these peptides are present predominantly in the extended PII or β structure, around 10% α or turn structures could be present, but the exact percentage of α or turn conformation could not be determined. In AcGXGNH2, X is expected to sample all three major backbone conformations, with PII or β structure being dominant and α basin being minor; in AcGXPNH2, however, X can sample only PII or β conformations. To avoid end and charge effects, two peptide series of this study have both ends blocked27. CD spectra for most AcGXGNH2 peptides except those with ring side chains (His, Trp, Tyr, Phe) show the characteristic far-UV CD signature of a mixture of PII and β conformations, with a strong negative band at ≈198 nm and a weak positive band or shoulder at ≈215 nm18,28,40(Fig. S1). CD spectra of AcGXGNH2 are very similar to those of AcGGXGGNH218. CD spectra of AcGXPNH2 are obscured by the contributions from Pro (Fig. S1). Small populations of Pro could exist in cis configurations; typical CD spectra of Pro peptides in PII helix usually shift to a longer wavelength as compared to those of non-Pro peptides. As a result, interpretation of CD spectra for AcGXPNH2 is not very obvious. Differential spectra between AcGXPNH2 and AcGXGNH2 reveal that Pro exists as a mixture of PII and PI (polyproline I) helices in AcGXPNH241; thus CD spectra of AcGXPNH2 reflect contributions from both X and Pro, contributions from X are expected to show the characteristic far-UV CD signature of a mixture of PII and β conformations, similar to those observed for AcGXGNH2.

Contents of α-conformers in AcGXGNH2 correlate with α-helix nucleation capabilities of X

3JαN coupling constant is directly related to the backbone ϕ angle by Karplus equations42,43. Measured 3JαN values at 25 °C (pH = 4.0) for AcGXGNH2 and AcGXPNH2 peptides are shown in Table 1 (see Fig. S2 for the NMR spectra and results of fitting). In AcGXPNH2, there is a slow trans-to-cis equilibrium for Pro, 3JαN for both cis- and trans- species are well resolved in 1D 1H NMR spectra, here only 3JαN values of X corresponding to trans-Pro are reported. Measured 3JαN coupling constants for AcGXGNH2 are compared to those for dipeptides (blocked amino acids)19 at 30 °C (pH = 4.9) in Fig. 1. The plot reveals a good agreement between two sets of coupling constants (R = 0.86).

Table 1 Experimentally determined 3JαN (298K) of AcGXGNH2, AcGXPNH2 and amino acid dipeptides and derived α, PII and β-contents for X in AcGXGNH2.
Full size table
Figure 1
figure 1

The 3JαN coupling constants measured for AcGXGNH2 peptides are plotted against those for amino acid dipeptides.

Full size image

3JαN values for AcGXGNH2 are smaller than those for AcGXPNH2 for most amino acids except for residues Asp (pH = 2.0 and 6.0), Asn and Thr. Excluding Thr, Asn and Asp’s, 3JαN values for AcGXGNH2 are on average 0.41 Hz smaller than those for AcGXPNH2. The smaller 3JαN values for AcGXGNH2 are consistent to X samples all three major backbone conformations in AcGXGNH2, while X samples only PII and β conformations in AcGXPNH2 (Thr, Asn and Asp are excluded). For AcGXPNH2 (X = Thr, Asn and Asp), X is expected to form turn structures44; it explains smaller observed 3JαN values for these residues in AcGXPNH2 compared to those in AcGXGNH2. For all other amino acids, contents of α conformations in AcGXGNH2 can be calculated from equation (1), in which 3JαN(α) is assigned to be 4.11 Hz, corresponding to a ϕ value of −60° (Table 1). For Thr, Asn and Asp (pH = 2.0 and 6.0) in AcGXGNH2, their contents of α conformations cannot be determined. It is a conservative and proximate practice to assign the values to be 0.04, 0.025, 0.02 and 0.05 for Thr, Asn and Asp (pH = 2.0 and 6.0), respectively, corresponding to the values from dipeptides by Grdadolnik et al.23 (Table 1). Contents of α conformations derived from blocked amino acids are significantly smaller than our values, 5.2 % vs. 12.6 % on average with Thr, Asn and Asp being excluded.

Our results indicate that xα values for hydrophobic or aromatic amino acids are significantly larger than those for polar amino acids, 14.9% vs. 7.4% on average. The differences among different non-polar residues are marginal (Table 1). Contents of minor populated α-conformations of different amino acids in AcGXGNH2 determine their varying α-helix nucleation propensities. Our results suggest that: for non-polar amino acids, the nature or the size of side chains, being aromatic ring or β-branching, do not have strong steric impact on helix nucleation, in contrast to their strong effects on helix propagation due to different steric constraints. The xα values observed show no correlation to any α-helix propensity scales45,46,47 that report mainly the propensity of amino acid residues to propagate on a preformed helix; the observation corroborates the conclusion by Miller et al.33 Effects of individual side chains on helix nucleation are difficult to deconvolute from those of helix propagation. Recently, Miller et al. have successfully separated the effects through studying a synthetic model and found that amino acid side chains contribute in a completely different manner to nucleation than to propagation33. In this study, the relative rates of disulfide formation serve as indirect indicators for different residues’ α-helix nucleation capabilities. Our derived populations of α conformations in AcGXGNH2 are compared to the relative rates of disulfide formation for limited amino acids by Miller et al.33 (Table S1); a good correlation is revealed (Fig. 2, R = 0.88).

Figure 2
figure 2

The correlation of determined α-contents for AcGXGNH2 and the relative rates of disulfide formation in a synthetic model.

Full size image

From derived vales of xα, we can calculate the probability of forming an α-helical segment comprising three residues, σ∙s = (xα)3 = 1.26 × 10−3 if we use the average value of xα for all amino acids. For Ala peptides, we can determine the probability, σ∙s = (xα)3 = 4.29 × 10−3 (xα = 0.1625 for Ala). The value is very close to those reported for Ala-rich peptides (the measured σ = 0.004 ± 0.002 with sAla = 1.4–1.6)37. As parameters, products of (xα1 • xα2 • xα3) for a combination of three different amino acids would be sensitive indicators to uncover the potential helix nucleation sites within sequences that form α-helices. From the derived xα values (Table 1), we predict sequences comprised of Val, Trp, Ile, His, Glu (pH = 2.0) and Ala are most likely the nucleation sites at early stages of α-helix formation; whereas sequences comprised of Asp, Cys, Asn and Thr (Pro and Gly are not considered here) are least likely the nucleation sites. Fast folding kinetic studies on model protein/peptides are expected to validate or invalidate our predictions.

Contents of PII and β conformations in AcGXGH2 and construction of free energy-conformation diagrams for three major backbone conformations

Contents of PII and β conformations in AcGXGNH2 can be calculated using equations (2) and (3) (Table 1). We assign standard 3JαN values for PII and β conformations to be 5.42 and 9.30 Hz, respectively. The value of 5.42 Hz for 3JαN(PII) corresponds to a ϕ value of −70°; the value of 9.30 Hz for 3JαN(β) is the result from fitting measured 3JαN values on blocked dipeptides to their β-populations derived from optical spectroscopic bands23. X in AcGXGNH2 adopts predominantly the extended PII or β conformations; on average, X samples about the same amount of time in PII or β basin, 44.7% vs. 44.5%. Our analysis indicates that β-contents or ΔG values for corresponding PII to β equilibriums show weak or reasonable correlations with β propensity scales (weak with β-contents and reasonable with ΔG), consistent to the observation in AcGGXGGNH2 peptides18. Correlations between ΔG and the β-sheet scale by Kim and Berg48 are shown in Fig. S3.

A more relevant comparison is between our data to those from blocked amino acids (dipeptides). Grdadolnik et al. have determined populations of the three major backbone conformations in 19 amino acid dipeptides (N-acetyl-X-N′-methylamide) by using the amide III region of the peptide infrared and Raman spectra23. The work by Grdadolnik et al. represents a major advance in band assignments of the peptide infrared and Raman spectra to different backbone conformations23. This advance made determination of backbone conformational distribution possible. If we compare our derived ΔG values for PII to β transitions to those derived for dipeptides, we find a reasonably strong correlation (Fig. 3, R = 0.84). Comparison of this correlation to the one in Fig. 1 (R = 0.86) indicates that the correlation between ΔG values is limited to that between 3JαN values. Given totally independent strategies on different systems were used, the correlation provides validations for both methods.

Figure 3
figure 3

The correlation of ΔG(β to PII) derived for AcGXGNH2 and that for dipeptides.

Full size image

The average length of β-strands in β-sheets is about 6 residues, the probabilities of forming a β-strand of 6 residues is (xβ)6 = 7.77 × 10−3 if we use the average value of xβ for all amino acids. Considering strands of 3–6 amino acids long might all play important roles in the early stages of β-hairpin folding, the population of a preformed β-strand of 3 residues long would reach as high as 20% (corresponding to xβ = 0.585). Following the procedure for α-helices, products of (xβ1 • xβ2 • xβ3) for a combination of three different amino acids might be used to locate the potential sites that form β-strands at early stages of protein folding. Similarly, from the derived xβ values (Table 1), we predict sequences comprised of Thr, Asp (pH = 2), Asn, His, Ile and Val are most likely the sites that tend to form nascent β-strands; whereas sequences comprised of Ala, Glu (pH = 6) and Trp (Pro and Gly are not considered here) are least likely the sites to form nascent β-strands. Nascent β-strands then initiate a productive or non-productive collision.

With the derived PII, β and α-contents, we can construct a free energy-conformation diagram on each AcGXGNH2 in aqueous solution for the three major backbone conformations (Fig. 4). The diagrams clearly show that the free energy level for α-basin is the highest among three for all amino acids; the free energy level for PII basin is the lowest for most amino acids except for Ile, Val, Asn, His, Thr, Glu (pH = 2.0) and Asp (pH = 2.0). Together with the results on 19 amino acid dipeptides from the optical spectroscopic data23, it is our believe that the derived free energy-conformation diagrams would provide a bench mark for testing predicting calculations of conformational energy maps of flexible model peptides38,39.

Figure 4
figure 4

Derived free energy-conformation diagrams for AcGXGNH2.

Full size image

Turn conformations in AcGNPNH2, AcGTPNH2 and AcGDPNH2 (pH = 2 and 6) and effects of different 3JαN(PII) and 3JαN(β) values on data analysis

We have detected significant turn structures in AcGNPNH2, AcGTPNH2 and AcGDPNH2 (pH = 2 and 6) as shown in Table 2. This observation is consistent with the findings by Hagarman et al.44 In this study, we assign standard 3JαN values for PII and β conformations to be 5.42 and 9.30 Hz, respectively. In our previous study on AcGGXGGNH2 peptides, a set of residue-specific 3JαN reference values for PII and β conformations were used18,49 (See Table 1 of reference 18). If we use the previous set of reference values to analyze the data in this study, slightly different PII, β and α-contents are obtained. Comparison of two sets of results indicates they are matched to each other overall with derived conclusions being the same. (See Supplementary Information for details). Regardless, the choice of different 3JαN(PII) and 3JαN(β) values has no effects on our derived xα values for X in AcGXGNH2 as implied by equation (1) (see Supplementary Information for derivation of the equation).

Table 2 Derived turn, PII and β-contents of X in AcGXPNH2 for Thr, Asn and Asp (pH = 2 and 6).
Full size table

NOE data and error analysis

NOEs can be used to analyze the conformations. Amide region of NOESY spectra for AcGXGNH2 peptides are shown in Fig. S10. Strong dαN(i, i + 1) NOE cross peaks are observed for X residues in AcGXGNH2 peptides, while the intensities of dαN(i, i) NOEs are weakened by about two- to fourfold relative to those of dαN(i, i + 1) NOEs; the dNN(i, i + 1) NOEs are not measurable due to their weak intensities and being very close to the diagonal peaks. These results indicate that AcGXGNH2 peptides are present predominantly in the extended PII or β-conformations that are consistent with our conclusion through analyzing coupling constant data. Figure S2 shows the amide region of 1D NMR spectra for all AcGXPNH2 and AcGXGNH2 peptides. The coupling constants were measured by a peak-fitting procedure to Lorentzian line shape, the fitting results are also shown in the figure. The derived coupling constants can be reproduced within 0.02 Hz if we fit a certain spectrum multiple times independently. In this and our previous studies, we used the Karplus equation by Vuister and Bax43 with coefficients: A = 6.51, B = −1.76 and C = 1.60; another parametrization for the Karplus equation with A = 6.98, B = −1.38 and C = 1.72 by Wang and Bax50 is believed to be more accurate. Calculated 3JαN(α) values for ϕ = −60° are coefficient dependent: 4.11 vs. 4.16 Hz for two sets of parameters; as a result, the derived α-population differs by ~2%. Given the average difference between 3JαN of AcGXPNH2 and AcGXGNH2 is about 0.41 Hz, plus a maximal uncertainty of 0.2 Hz on 3JαN(α) due to the uncertainties on the Karplus equation coefficients, we estimate the error of the derived α-population being around 10% for the majority of residues with non-overlapping amide signals, the estimated error could reach to 15–20% for those residues with overlapping peaks.

The relative population ratio between PII and β for AcGXGNH2 and AcGXPNH2

In this study, we assume that the population ratio between PII and β is approximately the same for AcGXGNH2 and AcGXPNH2. It is a known fact that there are secondary neighboring residue effects; we consider the effects from the side chain of residue X itself the primary effects. To our knowledge, Pro as a neighboring residue will make X favoring PII as compared to other neighboring residues. As a result, the population ratio between PII and β cannot be exactly the same for AcGXGNH2 and AcGXPNH2; it is most likely that the ratio for AcGXPNH2 is relatively larger than that for AcGXGNH2. Unfortunately, our current understanding on neighboring residue effects remains poor. To investigate the effects, first we define a parameter for the ratio of ratios, RR = GXGPII/β/GXPPII/β = [xPII(GXG)/xβ(GXG)]/[xPII(GXP)/xβ(GXP)], then we analyze our data systematically with the parameter RR setting from 0.80–1.10 in a step function of 0.05. (Table S2). It is clear that the derived content values shift in the same direction for all residues upon changing the value of RR. Specifically, average contents of PII increase by 1.8%, while average contents of β and α decrease by 0.5% and 1.3%, respectively, upon increasing the parameter RR by 0.05. To our gratification, the correlations and the conclusions hold really well upon changing the value of the parameter RR from 0.80–1.10 (Figs S11–S13).

Conclusion

We have determined the populations of three major conformers in AcGXGNH2 through analyzing 3JαN coupling constants of AcGXPNH2 and AcGXGNH2; the free energy-conformation diagrams are constructed for AcGXGNH2 peptides in aqueous solution. Our derived results show that on average residue X in AcGXGNH2 adopt PII, β, and α 44.7%, 44.5% and 10.8% of time, respectively. Minor populated α-conformations of different amino acids in AcGXGNH2 determine their varying α-helix nucleation capabilities. The contents of α-conformations for different amino acids define an α-helix nucleation propensity scale. There are no correlations observed between the xα values and any α-helix propensity scales45,46,47. Based on our derived β-contents, ΔG values for the corresponding PII to β equilibriums show a reasonable correlation with the β-sheet scale by Kim and Berg48, consistent to the observation in AcGGXGGNH2 peptides18. Derived ΔG values for PII to β transitions show a good correlation to those derived for dipeptides23. We have detected significant turn structures in AcGNPNH2, AcGTPNH2 and AcGDPNH2 (pH = 2 and 6)44. Results from this study have broad implications on the early-stage events of protein folding. Together with the results on 19 amino acid dipeptides23, our results would provide a bench mark for force field developments and for testing predicting calculations of conformational energy maps of flexible model peptides38,39.

Methods

Equation (1) was derived by assuming the PII to β population ratio of X in AcGXPNH2 and AcGXGNH2 being approximately the same. Peptides were synthesized and characterized as described27, by using an automated peptide synthesizer with standard Fmoc chemistry. CD spectra were recorded on a J-810 spectrometer with about 100–500 μM peptides in 10 mM phosphate buffer at 25 °C. The concentrations of peptides were determined from a combination of UV absorbance and NMR peak integration27. 1D and 2D (TOCSY and NOESY) 1H NMR spectra were collected on Bruker AVANCE 400/600 MHz spectrometers at 25 °C. 3JαN coupling constants were determined from high resolution 1D spectra. Details are described in Materials and Methods of Supplementary Information.

Additional Information

How to cite this article: Zhou, Y. et al. Populations of the Minor α-Conformation in AcGXGNH2 and the α-Helical Nucleation Propensities. Sci. Rep. 6, 27197; doi: 10.1038/srep27197 (2016).