Characterizing the denatured state ensemble of ubiquitin under native conditions using replica exchange molecular dynamics

Abstract

The folding of peptides into three-dimensional structures has been correlated to the occurrence of biological phenomena. In order to fully understand the pathway of protein folding, the structure of the denatured state, the reference state, under native conditions should be well-defined. However, very few of the denatured proteins have been studied under native conditions because the denatured state cannot be determined under exact native conditions using known experimental approaches. Herein, we characterize the denatured state ensembles (DSE) of ubiquitin under native conditions starting from a fully extended sequence resembling a peptide generated from a ribosome using replica exchange molecular dynamics. The representative structures of DSE with specific secondary structures distinct from the native ones are discussed. This result allows us to define the protein folding pathway more unambiguously and opens the door to biomedical studies of diseases correlated to protein misfolding. Potentially, the insights into the DSE can lead to targeted design of structure-based drug.

---

Introduction

Proteins made by translation in the ribosome are initially in a disordered, denatured state before they attain an ordered conformation. How freshly translated amino-acid chains fold into their unique three-dimensional structures with specific functions is an essential topic in understanding protein functions at the molecular level and in understanding diseases related to protein folding.¹ It is highly evident that these denatured proteins play a critical role in protein interaction networks, and the ordered/disordered transition might represent a regulation of numerous cellular processes.² As in a chemical reaction, the denatured state is the reactant, whereas the native state is the product. The ratio of the native proteins to the denatured proteins is constant at a given temperature under thermodynamic equilibrium. In other words, proteins exist in both their denatured state and native state. The structure of the denatured state, the reference state of protein folding, has been extensively studied under non-native conditions. However, very few of the structures have been studied under native conditions because the denatured state cannot be determined under exact native conditions using known experimental approaches. When the unfolded state of a protein is not well defined, how can the protein folding pathway be described precisely? Some studies unambiguously pointed out that, under physiological conditions, many proteins are highly dynamic with structures resembling those of denatured proteins more than those of proteins having globular folded states.^3,4 Proteins that depart from physiologically functioning native proteins are actually misfolded proteins,⁵ and can be the starting state of amyloid aggregation,⁶ the potential cause of many neurodegenerative diseases.⁷ Ubiquitin has even been identified as a component of helical filaments in Alzheimer's disease.⁸ Therefore, elucidation of the structure of the denatured state is critical for understanding the mechanism of amyloid diseases.

Ubiquitin is a small globular protein composed of 76 amino acids without disulfide crosslinks, and plays a critical role in protein recycling in a post-translational modification.⁹ To facilitate the process of ubiquitination, ubiquitin has to have conformational flexibility and intramolecular motions in order to interact with specific enzymes. Decreasing conformational flexibility, e.g., by incorporation of a disulfide bridge, largely weakens the activity of ubiquitin in signaling proteolysis.¹⁰ An alternate folded native state of ubiquitin upon binding to ubiquitin-activating enzyme E1 has been identified.¹¹ Ubiquitin has been a model protein in the study of protein folding for its reversible two-state conversion with rapid folding-unfolding kinetics.¹² The structural features of partially folded ubiquitin (designated A-state) in methanol/water cosolvent at pH 2 have been studied by experimental^13,14 and computational¹⁵ approaches. However, the denatured state of ubiquitin under native conditions has never been well characterized.

It is evident that numerous proteins are intrinsically unstructured or unfolded even in their functional states.² Unlike the native state, the denatured/unfolded state is typically populated with an ensemble of diverse structures.^16,17 These diverse structures are commonly referred to as denatured state ensembles (DSE) or unfolded ensembles. However, due to the structural diversity and the intense mobility of DSE, experimental characterization of DSE is a significant challenge. Previous experimental approaches have only provided an ensemble average measure of global physical properties of denatured proteins,¹⁸ or the structure of the denatured proteins under non-native conditions, such as denaturation with urea^19–22 or acid.²² Because of the lack of structural information of the denatured state under native conditions, the reactant of the folding reaction has never been well-defined. To obtain a clear description of the denatured state for fully understanding protein folding, atomic-level characterization of DSE is required. Compared to current experimental methods, computer simulation with enhanced sampling efficiency could provide more insights into the DSE of proteins under native conditions.

Many attempts have been made to simulate protein folding, including enhanced sampling at physiological temperature.²³ Replica exchange molecular dynamics (REMD) is a state-of-art algorithm promoting a system for escaping from local energy traps to travel new conformational space through exchanging non-interacting replicas at different temperatures.²⁴ The thermodynamic parameters derived from the simulation is highly consistent with empirical measurement,²⁵ indicating the suitability of this method and the applied parameters. Taking advantage of this highly improved computer simulation method, REMD was used to characterize the DSE of ubiquitin starting from a fully extended sequence of ubiquitin that resembles the peptide freshly synthesized from the ribosome. Concurrently, to absolutely prove the appropriateness of this simulation, we have some calibration of bench mark in comparison with experimental results. Based on the trajectories of REMD simulation, the representative structures of DSE were analyzed based on clustering and principal component analysis (PCA).

Methods

Simulation procedure

The AMBER 11 simulation package²⁶ was used in REMD simulation and data analysis. The all-atom point-charge force field (also known as ff03)²⁷ was applied to represent the protein, as this force field in implicit solvent has been extensively investigated for the capability of folding protein into structures that resemble those obtained from multidimensional NMR spectroscopy and X-ray study with only 0.5 Å difference.²⁸ The denatured state of ubiquitin was computed starting from heating the X-ray solved structure (PDB code: 1ubq²⁹) to 600 K to fully open the rigid conformation.³⁰ The structure of the fully extended sequence of ubiquitin was optimized at 300 K. Prior to the REMD simulation, 32 replicas were equilibrated to reach the target temperatures at 275.2, 281.3, 287.3, 293.6, 300.0, 306.5, 313.2, 320.0, 326.9, 334.1, 341.3, 348.7, 356.2, 364.0, 371.8, 379.9, 388.1, 396.4, 404.9, 413.6, 422.6, 431.7, 440.9, 450.4, 460.1, 470.0, 480.1, 490.4, 500.9, 511.6, 522.6 and 533.8 K. The temperatures were set based on the estimation of exchange probabilities at 0.5.³¹ Applying a 16 Å force-shifted non-bonded cutoff and generalized Born solvent models,³² the 32 replicas were simulated for 100 ns with an integration step of 2 fs, based on the fully extended structure.

Structural analysis

According to the crystallographic structure,²⁹ native ubiquitin has an extremely compact structure including five β-strands, one α-helix (residues 23–34) and one 3₁₀ helix (residues 56–59). The five β-strands are β1 (residues 1–7), β2 (residues 10–17), β3 (residues 40–45), β4 (residues 48–50) and β5 (residues 64–72) as illustrated in Fig. 1(A). The structural information of the unfolded ubiquitin can be illustrated as radius of gyration (R_g) of the entire protein.

Fig. 1 The structures of (A) the native ubiquitin (PDB code: 1ubq) and (B) the fully extended conformation as the starting structure of this REMD simulation. In (A), the α-helix and 3₁₀ helix are labeled in red, and magenta, respectively. Five β-strands are colored in blue, purple blue, skyblue, deep blue and cyan for β1, β2, β3, β4 and β5, respectively.

Clustering analysis

Simulated REMD trajectory at 300 K was clustered into distinct groups based on average linkage methods³³ with 3 Å-cutoff. The top 10 populated structures were compared with the following PCA results based on the C^α-RMSD of two structures.

Principal component analysis (PCA)

To characterize the population of explored structures in more detail, principal component analysis was carried out on the trajectory simulated at 300 K. The landscape of structure population was prepared from the first two PCA vectors generated from 228 coordinates in each eigenvector. The top 8 populated structures were chosen to be compared with the clustering analysis.

Results and discussion

The native structure of ubiquitin contains 34% of β-structure and 23% of helical conformations²⁹ as shown in Fig. 1(A). The starting structure of this simulation is a fully extended conformation lacking any ordered structure, as illustrated in Fig. 1(B).

Thermodynamics analysis

Ubiquitin is a well-known model protein for two-state folding-unfolding. The melting temperature (T_m) can be obtained from the REMD simulation using van't Hoff analysis.²⁸ To deduce the melting transition, the heat capacity profile was calculated since the heat capacity is a direct measure of the thermodynamic properties in protein folding. Therefore, we performed a calculation of the heat capacity (C_p) with the equation C_p = (〈E²〉 − 〈E〉²)/RT, where E is the energy, R is the gas constant, and T is the temperature. As shown in Fig. 2, the resulting heat capacity profile exhibits one broad peak centered at 334.6 K (61.1 °C), indicative of one distinct melting transition. Remarkably, the T_m value of this simulation is similar to T_m measured by temperature-jump unfolding experiment (64.0 °C).³⁴ The calculated denaturation enthalpy (ΔH_m) for the melting transition is 47.9 kcal mol⁻¹. The change in the entropy at T_m, ΔS_m, is 143.4 cal mol⁻¹ K⁻¹. Our analyzed thermodynamic parameters (ΔH_m and ΔS_m) are also very close to the experimental results (ΔH_m = 40.9 kcal mol⁻¹, ΔS_m = 121 kcal mol⁻¹ K⁻¹).³⁴

Fig. 2 Analysis of REMD simulation on heat capacity as a function of temperature.

Analysis of radius of gyration

To evaluate the compactness of the protein, the radius of gyration of a specific replica started at 300 K over the simulation time was plotted, as shown in Fig. 3(A). The initial structure is fully extended with the radius of gyration (R_g) at ∼56 Å. Quickly, R_g decreases in the first 10 ns and then remained relatively steady at ∼33 Å, which is larger than R_g of urea-unfolded ubiquitin at 24.6 Å.³⁵ The fast decrease of R_g indicates the rapid formation of the specific structure. This gyration radius is smaller than that of native α-synuclein (40 Å),³⁶ which can potentially misfold into amyloid fibrils deposited in Lewy bodies typically implicated as the cause of Parkinson's disease.³⁷ Given that the deposition of some natively unfolded proteins may lead to the development of several neurodegenerative disorders,³⁸ it is necessary to pay more attention to the DSE of proteins having important physiological functions.

Fig. 3 Analysis of a specific replica started at 300 K with subsequent temperature change at high temperatures. (A) The radius of gyration. (B) The most populated structures recorded at time-course. The representative structures are generated from clustering analysis and colored according the structural features of the native structure described in Fig. 1.

Calibration of bench mark

The extended structure of ubiquitin initially equilibrated at 300 K experienced multiple exchanges to various set temperatures. While under thermodynamic equilibrium at 300 K, about half of the molecules carry higher heat, and such molecules experienced high-temperature conditions, which resulted in a heat-induced denatured state. The representative structures of this heating trajectory illustrated in Fig. 3(B) indicates that α-helical conformation was built in the first 10 ns and the construction/destruction of the helices continues in the subsequent folding process. Ubiquitin has experience conformational change dynamically, while the α-helical structure is highly conserved along the simulation. To be noted, β-sheets in the native ubiquitin are not presented in these structures. The representative structures of this 300 K-replica are highly comparable to partially folded A-state ubiquitin observed at pH 2 in methanol/H₂O cosolvent by NMR^13,14 and simulated with explicit-solvent MD,¹⁵ although ubiquitin was simply partially denatured from the compact native-structure in the previous works.

Time-course of the secondary structural features

The specific conformations were formed as judged by R_g. The distribution of the secondary structure of the entire protein at 300 K along the 100 ns of simulation was analyzed, as shown in Fig. 4. Clearly, formation of specific inter-residual interactions started in the first few nanoseconds. During the course of the simulation, α-helical conformation was widely presented, especially in the β5 region. Similarly, α-helical conformation is built in part of β2 and β3 regions. In the α-helix region, most residues remained α-helical. Remarkably, antiparallel and parallel β-strands were very limited.

Fig. 4 The change of the secondary structure of one replica in the time-course of the simulation starting at 300 K. The assignments of the secondary structure of the native state are marked in the right.

Clustering analysis and the most populated structure of DSE

Given the greatly enhanced sampling in REMD, a great number of structures were generated. To find representative structures of DSE, we performed a clustering analysis based on the similarity of the C^α-RMSD of all structures. As a result, the 10 most populated structures (cluster 1–cluster 10) are shown in Fig. 5(A). The top 10 clusters occupy 24% of all DSE. The three most populated structures of DSE (clusters 1–3) are compared in Fig. 6(A). All three structures contain a large proportion (57–67%) of α-helices, especially in their C-terminal ends, but they lack β-strands.

Fig. 5 Representative structures of DSE from the REMD simulations. (A) The top 10 clusters from clustering analysis. (B) The top 8 structures from PCA analysis. The structures are numbered in the order of their population. The population of each structure is listed in the parentheses.

Fig. 6 Clustering analysis of representative structures of DSE. (A) Superposition of three most populated DSE colored in green, blue and magenta in the order of the population. (B) The superposition of the most populated structures generated by clustering and by PCA shown in green and magenta, respectively.

PCA analysis and the most populated structure of DSE

Eight highly populated structures (PCA1–PCA8, occupy 13% of all DSE, Fig. 5(B)) analyzed with PCA were compared with clusters 1–10 for structural similarity judged by their C^α-RMSD as listed in Table 1. Obviously, some PCA structures highly resemble those of the clustering results. For example, cluster 1 is very similar to PCA2 with high superposition with 2.9 Å of C^α-RMSD shown in Fig. 6(B), and cluster 2 is highly comparable to PCA8 with 0.9 Å of C^α-RMSD. Structural similarity is also present between cluster 3/PCA4 (1.7 Å), cluster 4/PCA1 (1.6 Å), cluster 5/PCA1 (3.3 Å) and so on (see Table 1 for detail). Therefore, comparable structures of highly populated DSE are independently revealed by these two analysis methods.

Table 1 Comparison of populated DSE structures analysed with clustering and PCA. C^α-RMSD values of two structures are listed in angstrom

	PCA1	PCA2	PCA3	PCA4	PCA5	PCA6	PCA7	PCA8
Cluster 1	9.9	2.9	9.3	12.3	14.5	15.8	14.3	9.8
Cluster 2	10.8	10.4	2.2	12.8	16.0	15.1	14.9	0.9
Cluster 3	11.2	12.9	12.7	1.7	14.6	13.0	12.8	12.8
Cluster 4	1.6	10.9	11.2	11.4	17.4	18.7	17.1	11.2
Cluster 5	3.3	11.1	11.8	11.0	17.3	18.6	16.8	11.9
Cluster 6	9.8	7.7	9.6	12.1	14.6	15.4	13.3	10.2
Cluster 7	10.4	1.2	10.0	13.0	13.8	15.8	14.3	10.4
Cluster 8	10.5	13.0	12.5	8.7	13.3	13.7	12.1	13.0
Cluster 9	12.2	11.0	10.5	8.6	14.3	13.3	13.3	10.5
Cluster 10	3.9	11.2	11.2	11.8	17.2	18.2	16.5	11.3

---

Contrary to the general prediction that proteins become structure-less in the denatured state, it is evident that protein unfolding can generate a compact denatured state with specific residual interaction.³⁹ Rather than preservation of the residual structure of the native state, the highly populated DSE can assemble structures to some order that are distinct from those of the native state. Formation of distinct secondary structure in the folded and unfolded states has been observed experimentally under non-native conditions. For example, NMR characterization of FK506 binding protein (FKBP), a protein folding chaperone involved in many cellular processes,⁴⁰ indicated that two segments of β-strand structure in the native state turned into a more helical structure upon urea-denaturation.⁴¹ Similarly, acid-denatured staphylococcal nuclease formed a two-strand β-hairpin below 5 M urea, and this conformation turned into a three-stranded antiparallel sheet when urea was diluted to 3 M.⁴² In comparison to the partially folded A-state of ubiquitin,¹⁴ the content of α-helical conformation of ubiquitin DSE found in this study is similar to that of A-state ubiquitin. High tendency of helical formation was similarly observed in sodium dodecyl sulfate induced unfolding.⁴³ These comparable helical structures indicate the propensity of ubiquitin to fold into α-helical structure in the denatured state.

Conclusions

Molecular recognition in cellular process is determined by the protein structure. Protein folding relies on the equilibrium between the native and denatured states, the DSE being largely populated. The result of this study allows us to define the protein folding pathway more unambiguously. Resolving the structure of the denatured state opens the door to biomedical studies of diseases correlated to protein misfolding, and potentially builds a solid ground on which structure-based drugs may be designed.⁴⁴ In addition, formation of β-sheet is an entropy-unfavored process more than the construction of α-helix.⁴⁵ How to drive development of highly entropy-unfavored conformations such as parallel and antiparallel β-sheets in protein folding is next problem that we need to clearly elucidate.

Acknowledgements

We are grateful to the National Center for High-performance Computing for computer time and facilities, and Ministry of Science and Technology for financial support (MOST 105-2113-M-194-005). We also thank Dr Raymond Chung for editing the manuscript.

Notes and references

1. S. B. Prusiner, Science, 1997, 278, 245.
2. P. E. Wright and H. J. Dyson, J. Mol. Biol., 1999, 293, 321.
3. V. N. Uversky, Protein Sci., 2002, 11, 739–756.
4. H. J. Dyson and P. E. Wright, Nat. Rev. Mol. Cell Biol., 2005, 6, 197.
5. J. D. Bryngelson and P. G. Wolynes, Proc. Natl. Acad. Sci. U. S. A., 1987, 84, 7524.
6. P. Polverino de Laureto, N. Taddei, E. Frare, C. Capanni, S. Costantini, J. Zurdo, F. Chiti, C. M. Dobson and A. Fontana, J. Mol. Biol., 2003, 334, 129.
7. F. Chiti and C. M. Dobson, Annu. Rev. Biochem., 2006, 75, 333.
8. H. Mori, J. Kondo and Y. Ihara, Science, 1987, 235, 1641.
9. A. Hershko and A. Ciechanover, Annu. Rev. Biochem., 1982, 51, 335.
10. D. J. Ecker, T. R. Butt, J. Marsh, E. Sternberg, A. Shatzman, J. S. Dixon, P. L. Weber and S. T. Crooke, J. Biol. Chem., 1989, 264, 1887.
11. S. Kitazawa, T. Kameda, A. Kumo, M. Yagi-Utsumi, N. J. Baxter, K. Kato, M. P. Williamson and R. Kitahara, Biochemistry, 2014, 53, 447.
12. D. N. Woolfson, A. Cooper, M. M. Harding, D. H. Williams and P. A. Evans, J. Mol. Biol., 1993, 229, 502.
13. M. M. Harding, D. H. Williams and D. N. Woolfson, Biochemistry, 1991, 30, 3120.
14. B. Brutscher, R. Bruschweiler and R. R. Ernst, Biochemistry, 1997, 36, 13043.
15. D. B. Kony, P. H. Hunenberger and W. F. van Gunsteren, Protein Sci., 2007, 16, 1101.
16. S. L. Kazmirski, K. B. Wong, S. M. Freund, Y. J. Tan, A. R. Fersht and V. Daggett, Proc. Natl. Acad. Sci. U. S. A., 2001, 98, 4349.
17. B. Zagrovic, C. D. Snow, S. Khaliq, M. R. Shirts and V. S. Pande, J. Mol. Biol., 2002, 323, 153.
18. A. J. Petrescu, V. Receveur, P. Calmettes, D. Durand, M. Desmadril, B. Roux and J. C. Smith, Biophys. J., 1997, 72, 335.
19. D. Neri, M. Billeter, G. Wider and K. Wuthrich, Science, 1992, 257, 1559.
20. D. Shortle and C. Abeygunawardana, Structure, 1993, 1, 121.
21. D. Shortle and M. S. Ackerman, Science, 2001, 293, 487.
22. W. Meng, B. Luan, N. Lyle, R. V. Pappu and D. P. Raleigh, Biochemistry, 2013, 52, 2662.
23. Y. M. Rhee and V. S. Pande, Biophys. J., 2003, 84, 775.
24. Y. Sugita and Y. Okamoto, Chem. Phys. Lett., 1999, 314, 141.
25. J. L. Tang, P. J. Wu, S. C. Wang and C. I. Lee, J. Phys. Chem. B, 2012, 116, 3305.
26. D. A. Case, et al., AMBER 11, University of California, San Francisco, CA, 2010.
27. Y. Duan, C. Wu, S. Chowdhury, M. C. Lee, G. M. Xiong, W. Zhang, R. Yang, P. Cieplak, R. Luo, T. Lee, J. Caldwell, J. M. Wang and P. Kollman, J. Comput. Chem., 2003, 24, 1999.
28. H. Lei, C. Wu, H. Liu and Y. Duan, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 4925.
29. S. Vijay-Kumar, C. E. Bugg and W. J. Cook, J. Mol. Biol., 1987, 194, 531.
30. C. I. Lee and N. Y. Chang, Biophys. Chem., 2010, 151, 86.
31. A. Patriksson and D. van der Spoel, Phys. Chem. Chem. Phys., 2008, 10, 2073.
32. A. Onufriev, D. Bashford and D. A. Case, Proteins, 2004, 55, 383.
33. J. Y. Shao, S. W. Tanner, N. Thompson and T. E. Cheatham, J. Chem. Theory Comput., 2007, 3, 2312.
34. H. S. Chung, A. Shandiz, T. R. Sosnick and A. Tokmakoff, Biochemistry, 2008, 47, 13870.
35. V. N. Uversky, Eur. J. Biochem., 2002, 269, 2.
36. J. Li, V. N. Uversky and A. L. Fink, Biochemistry, 2001, 40, 11604.
37. L. S. Forno, J. Neuropathol. Exp. Neurol., 1996, 55, 259–272.
38. V. N. Uversky, J. R. Gillespie and A. L. Fink, Proteins, 2000, 41, 415.
39. A. L. Fink, L. J. Calciano, Y. Goto, T. Kurotsu and D. R. Palleros, Biochemistry, 1994, 33, 12504.
40. Y. Harrar, C. Bellini and J. D. Faure, Trends Plant Sci., 2001, 6, 426.
41. T. M. Logan, Y. Theriault and S. W. Fesik, J. Mol. Biol., 1994, 236, 637.
42. Y. Wang and D. Shortle, Biochemistry, 1995, 34, 15895.
43. G. F. Schneider, B. F. Shaw, A. Lee, E. Carillho and G. M. Whitesides, J. Am. Chem. Soc., 2008, 130, 17384.
44. A. N. Bullock and A. R. Fersht, Nat. Rev. Cancer, 2001, 1, 68.
45. M. C. Baxa, E. J. Haddadian, A. K. Jha, K. F. Freed and T. R. Sosnick, J. Am. Chem. Soc., 2012, 134, 15929.

---