Abstract
Energy landscape theory suggests that native interactions are primary determinants of protein folding mechanisms. Consequently, structure-based (Gō) models have successfully captured folding dynamics by focusing on the native state. However, atomic details and non-native interactions are often essential for a complete description of protein dynamics. This paper presents ProfasiGo, a hybrid all-atom model that integrates the implicit solvent physics-based Profasi force field with a structure-based Gō potential. Implemented in the PHAISTOS and ProFASi software packages, the model utilizes Monte Carlo sampling to explore the folding free energy landscapes of four topologically similar proteins. Our results demonstrate that ProfasiGo captures the nuanced physical details of explicit solvent simulations while maintaining the sampling efficiency of Gō models.
Introduction: The Simulation Trade-off
Simulating the process by which a protein folds into its functional three-dimensional structure remains a central challenge in biophysics. Researchers typically choose between two computational extremes:
- All-atom physics-based models: These provide high accuracy by accounting for every atom and interaction. However, their computational cost is immense, often requiring specialized hardware or months of supercomputer time to observe a single folding event.
- Structure-based (Gō) models: These models are computationally efficient because they assume the protein is biased toward its known native state. While effective for large-scale transitions, they often fail to capture the "roughness" of the energy landscape—the non-native interactions and intermediate traps that define real-world folding pathways.
ProfasiGo addresses this gap by creating a "middle ground" that retains atomic-level physical detail while employing a minimal native bias to guide the simulation toward the native state.
Methodology
The ProfasiGo model integrates two distinct components to balance accuracy and speed:
- Profasi Force Field: An all-atom, implicit solvent model that calculates physical interactions, including electrostatics, hydrogen bonding, and excluded volume.
- Gō Potential: A structure-based bias that adds attractive forces between residue pairs known to be in contact in the native state.
To explore the energy landscape, the authors utilized Monte Carlo (MC) sampling rather than standard Molecular Dynamics (MD). MC sampling allows the simulation to efficiently "jump" across energy barriers. This was further enhanced by Parallel Tempering (Replica Exchange) using the MUNINN software to prevent the simulation from becoming trapped in local energy minima.
Benchmarking and Results
The model was tested on four three-helix bundle proteins to evaluate its ability to distinguish between similar topologies:
- α3W and α3D: Designed proteins with similar topologies but opposite chiralities.
- EnHD (Engrailed Homeodomain) and UVF: Proteins with nearly identical topologies but different sequences and stabilities.
Sequence and Chirality Effects
ProfasiGo successfully identified that α3W and α3D fold via different mechanisms despite their structural similarity. α3W follows a "diffusion-collision" model, where helices form independently before packing together. In contrast, α3D exhibits "nucleation-condensation," where helix formation and packing occur simultaneously.
For EnHD and UVF, the model demonstrated that sequence-specific hydrophobic interactions significantly alter the folding pathway. Pure Gō models often fail to capture these nuances because they rely almost exclusively on the final topology. ProfasiGo, however, revealed that the physics-based terms provide the necessary "roughness" to identify intermediate states missed by simpler models.
Computational Efficiency and Validation
The hybrid model's results were qualitatively similar to "gold standard" explicit solvent MD simulations using the CHARMM22* force field. Crucially, ProfasiGo achieved converged results in a few weeks on a single CPU, a fraction of the resources required for explicit MD. The model also produced clear heat capacity peaks, allowing for the precise determination of melting temperatures ($T_f$). The folding mechanisms remained robust even when the strength of the Gō bias ($\epsilon_{Go}$) was varied, suggesting that the physics-based Profasi terms drive the mechanistic details.
Discussion and Limitations
While ProfasiGo is a powerful tool for thermodynamic analysis, it has specific limitations:
- Kinetics: Because it relies on Monte Carlo sampling, the model cannot provide real-time kinetic data (e.g., microseconds to fold). It is designed to map the energy landscape rather than track time-dependent trajectories.
- Temperature Calibration: The simulated folding temperatures do not perfectly align with experimental values, requiring a temperature scan to identify the transition point.
- Future Applications: While validated on small 3-helix bundles (40-80 residues), future work will focus on scaling the model to larger proteins containing β-sheets or intrinsically disordered regions. Additionally, the model shows promise for integrating sparse experimental data, such as NMR chemical shifts, to refine structural ensembles.
Conclusion
ProfasiGo represents a significant step forward in protein simulation, offering a computationally accessible way to study complex folding landscapes. By combining the physical realism of all-atom models with the efficiency of structure-based biases, it allows researchers to explore the delicate balance between sequence, topology, and folding mechanism.
Summary updated