Comparing force fields for biomolecular simulations

Cino EA, Choy WY, Karttunen M (2012) Comparison of Secondary Structure Formation Using 10 Different Force Fields in Microsecond Molecular Dynamics Simulations. J Chem Theory Comput 8:2725-2740. link to manuscript

Videos of our simulations of disordered proteins on YouTube and Flickr

Fig 1. Structures of the NRF2 hairpin from folding simulations and representative free energy landscape of the hairpin folding. The free energy landscape was constructed from a 3 dimensional histogram consisting of radius of gyration, backbone rmsd to bound state structure (PDB id: 2FLU) and distance between 2 hydrophobic residues on opposite strands of the hairpin that make close contacts (as determined by solution NMR for the peptide in the free state).

       A primary choice in performing MD simulations is which force field to use. Currently, specific force fields are employed depending on the system being investigated. For example, a certain force field may give good agreement with experimental data for a specific type of protein, but not necessarily for another. Even though modifications to biomolecular force fields have lead to improved transferability, further progress relies on continued testing. Ideally, these efforts will lead to the development of fully transferable force fields.

       A good method to test force field performance is by simulating protein folding and comparing the results to experimentally determined protein structures. However, most proteins fold on timescales unattainable by modern computer simulations. As a result, it can be challenging to find good test systems. One approach has been to extract amino acid sequences encoding self-folding motifs out of well-folded proteins. While this may be a viable approach to decrease system sizes and obtain folding events, care must be taken to ensure that the motif does indeed fold properly in the absence of the rest of the protein. Another approach has been to design small, fast folding proteins. However, protein design is not an easy task.

       Perhaps a better, in terms of being doable, approach for force field testing of protein folding is to use amino acid sequences encoding preformed structural elements (PSEs). As discussed in my January 11th post, intrinsically disordered proteins (IDPs) often contain PSEs to facilitate their interactions with other proteins. The benefits of using PSEs for folding simulations is that they are typically locally occurring features that do not rely as heavily upon long-range contacts as structural elements in well-folded proteins. Moreover, they often contain features that are found in well-folded proteins, such as hydrophobic clusters and electrostatic interactions. In many ways, PSEs can be though of as mini or micro proteins. These may be ideal candidates for testing of force fields.

Fig 2. Example of a hairpin motif. Hairpins are composed of two antiparallel beta strands connected by a turn. They are common structural elements found in many proteins.

       For this post, the folding of a PSE from the protein NRF2 with 10 commonly used biomolecular force fields is compared. This PSE has been studied experimentally and is known to form what is known as a ‘hairpin’ structure (Fig. 2). Starting from an extended conformation, the amino acid sequence encoding this hairpin has been shown to fold into a structure consistent with experimental data in < 1 µs. However, when comparing the folding of this structural element with commonly used force fields, differences were observed (Fig. 1). Although many of the force fields reproduced experimentally determined free state contacts and yielded hairpin structures, some did not (Fig. 1). As mentioned in my January 11th post, the hairpin appears to be stabilized by hydrogen bonds and hydrophobic contacts.

       The results from this investigation emphasize the importance of force field selection. Additionally, the work illustrates that PSEs may be ideal candidates for force field testing. The results obtained from folding simulations of such elements should be useful for improving biomolecular force fields.

Our related work references

1. Cino EA, Choy WY, Karttunen M (2012) Comparison of Secondary Structure Formation Using 10 Different Force Fields in Microsecond Molecular Dynamics Simulations. J Chem Theory Comput 8:2725-2740. link to manuscript

2. Cino EA, Wong-Ekkabut J, Karttunen M, Choy WY (2011) Microsecond molecular dynamics simulations of intrinsically disordered proteins involved in the oxidative stress response. PLoS One 6:e27371. link to manuscript

3. Cino EA, Karttunen M, Choy WY (2012) Effects of molecular crowding on the dynamics of intrinsically disordered proteins. PLoS One 7:e49876. link to manuscript

4. Cino E, Fan J, Yang D, Choy WY (2012) (1)H, (15)N and (13)C backbone resonance assignments of the Kelch domain of mouse Keap1. Biomol NMR Assign. In press. link to manuscript

5. Khan H, Cino, EA, Brickenden A, Fan J, Yang D, Choy WY (2013) Fuzzy Complex Formation between the Intrinsically Disordered Prothymosin α and the Kelch Domain of Keap1 Involved in the Oxidative Stress Response. J Mol Biol. In press. link to manuscript

Pilkington on thermodynamics and phase transitions

Confessions of Karl Pilkington, Science Man: The Early Years

Pilkington on the Big Bang

Confessions of Karl Pilkington, Science Man: The Early Years

Pilkington on snow

Confessions of Karl Pilkington, Science Man: The Early Years

Pilkington on superconductivity

Confessions of Karl Pilkington, Science Man: The Early Years

Preformed structural elements in intrinsically disordered proteins

Cino EA, Wong-Ekkabut J, Karttunen M, Choy WY (2011) Microsecond molecular dynamics simulations of intrinsically disordered proteins involved in the oxidative stress response. PLoS One 6:e27371. link to manuscript

Videos of our simulations of disordered proteins on YouTube and Flickr

          It was once thought that a protein must adopt a defined three-dimensional structure to function properly. The discovery of biologically active intrinsically disordered proteins (IDPs) illustrates that some proteins are able to carry out their functions through different mechanisms than well-folded proteins. IDPs comprise ~30% of the eukaryotic proteome. The abundance of IDPs in organisms suggests that they are essential for numerous functions. They are often found to be involved in crucial signaling and regulatory functions in cells. Therefore, it is not a surprise that IDPs are frequently associated with human diseases, in particular cancer and neurodegenerative diseases.

Fig 1. Structures of an IDP and a well-folded protein. The NMR ensemble structures of the IDP (Thylakoid soluble phosphoprotein TSP9, PDB id: 2FFT) do not overlay well because its intrinsic dynamic properties allow exchange between different conformations over time. On the other hand, the NMR ensemble structures of the well-folded protein (Ubiquitin, PDB id: 1D3Z) illustrate that a similar structure is maintained over time.

          Despite their name, IDPs do not adopt completely random structures. Many IDPs have considerable conformational propensities. Segments of IDPs that contain residual structure may act as recognition features for interacting with other proteins. There are two methods by which these interaction hot spots function. For some IDPs, the recognition features contain preformed structural elements (PSEs) that resemble the bound state, while others may couple conformational changes with target binding. For IDPs that bind using PSEs, the bound state structure is already formed in the unbound state. In the coupled folding and binding model, the IDP undergoes a disorder-to-order transition upon binding to a target. It is important to realize that these two interaction methods represent opposite ends of the binding mode continuum. In most cases, binding of IDPs is probably modulated by a combination of these two mechanisms.

Fig 2. Binding mechanisms of IDPs. An IDP can interact with binding partners by either folding into a bound state like conformation prior to binding (top), encountering the binding partner and then folding (bottom), or a combination of these two mechanisms (middle).

          

          Because preformed elements in unbound structural ensembles of IDPs often comprise protein-protein  interaction sites, their identification and characterization is an area of active investigations. The main approach is to identify preformed elements from sequence alone. Interaction hot spots in IDPs often have distinct sequence characteristics compared to their surroundings, with the primary difference being an increased hydrophobic content, which may promote local structure formations. The main problem with relying solely upon amino acid sequence properties to identify PSEs is the high number of false positives. In addition bioinformatics approaches, Nuclear Magnetic Resonance (NMR) spectroscopy has also proven to be a useful technique for detecting PSEs. The focus of this post, however, is on using Molecular Dynamics (MD) simulations to detect and characterize PSEs in IDP structures.

          Here, we used MD simulations to probe the free state structures and dynamics of two IDPs, PTMA and NRF2. These two proteins interact with a common partner, Keap1, in order to control the cellular response to oxidative stress. Misregulation of the oxidative stress response pathway can lead to neurodegenerative diseases, diabetes and cancer. Compounds that can disrupt the NRF2-Keap1 interaction have been proposed as potential therapeutic agents for enhancing the oxidative stress response. Development of drug candidates requires an understanding of the molecular basis of the interactions.

          By conducting microsecond timescale MD simulations, important PSEs were identified in the Keap1 binding regions of PTMA and NRF2. In the absence of Keap1, the PSEs had clear resemblance to their bound state structures. NRF2, which interacts with Keap1 with a higher affinity than PTMA formed PSEs with lower RMSDs to its bound state structure, compared to PTMA. It appears that the extents of bound state like structures that are formed in the absence of binding partner have important implications in dictating the binding thermodynamics of these proteins.

Fig 3. Formation of PSEs with different extents of bound state resemblance. Left: RMSDs to the bound state structures during the MD trajectories. Right: snapshots from the MD simulations (grey) overlaid with their bound state structures (pink).

          The MD simulations were analyzed to determine possible reasons to explain why NRF2 forms a more bound state like PSE compared to PTMA. The analysis suggested that NRF2 was able to form a more compact PSE compared to PTMA. This may be attributed to NRF2 forming more hydrogen bonds. Additionally, NRF2 contains more hydrophobic amino acids surrounding its PSE compared to PTMA, which may also promote structure formation.

Fig 4. Contributing factors to explain the different extents of preformed structure in PTMA and NRF2. Top: percentage of structures from the MD simulations with an end-to-end distance less than 0.7 nm. Bottom: percentage of MD structures with 1 or more hydrogen bond.

Our related references

1. Cino E, Choy WY, Karttunen M (2012) Comparison of secondary structure formation using 10 different force fields in microsecond molecular dynamics simulations J Chem Theory Comput 8:2725-2740. link to manuscript

2. Cino EA, Karttunen M, Choy WY (2012) Effects of molecular crowding on the dynamics of intrinsically disordered proteins. PLoS One 7:e49876. link to manuscript

3. Cino E, Fan J, Yang D, Choy WY (2012) (1)H, (15)N and (13)C backbone resonance assignments of the Kelch domain of mouse Keap1. Biomol NMR Assign. In press. link to manuscript

4. Khan H, Cino, EA, Brickenden A, Fan J, Yang D, Choy WY (2013) Fuzzy Complex Formation between the Intrinsically Disordered Prothymosin α and the Kelch Domain of Keap1 Involved in the Oxidative Stress Response. J Mol Biol. In press. link to manuscript

General references

1. Dunker AK, et al. (2001) Intrinsically disordered protein. J Mol Graph Model 19:26-59. link to manuscript

2. Uversky VN (2002) Natively unfolded proteins: a point where biology waits for physics. Protein Sci 11:739-56. link to manuscript

3. Uversky VN, Oldfield CJ, Dunker AK (2008) Intrinsically disordered proteins in human diseases: introducing the D2 concept. Annu Rev Biophys 37:215-46. link to manuscript

4. Wright PE, Dyson HJ (1999) Intrinsically unstructured proteins: re-assessing the protein structure-function paradigm. J Mol Biol 293:321-31. link to manuscript

5. Lambrughi M et al. (2012) Intramolecular interactions stabilizing compact conformations of the intrinsically disordered kinase-inhibitor domain of Sic1: a molecular dynamics investigation. Front Physiol 3:435. link to manuscript

New blog

Although I and SoftSimu have a long history wrt web presence, this blog is a new attempt. The intention is just to follow a relatively random walk through science and research related issues. The opinions here reflect only those of the respective authors and not our employers or other bodies any of us may be a member/part of. Let's see how this works out.