Rosetta FunFolDes - A general framework for the computational design of functional proteins

Jaume Bonet, Sarah Wehrle, Karen Schriever, Che Yang, Anne Billet, Fabian Sesterhenn, Andreas Scheck, Freyr Sverrisson, Barbora Veselkova, Sabrina Vollers, Roxanne Lourman, Mélanie Villard, Stéphane Rosset, Thomas Krey, Bruno E. CorreiaPLOS Computational Biology 2018 · DOI

TL;DR

This paper introduces Rosetta FunFolDes, a computational framework that allows for the design of functional proteins by embedding specific functional motifs into existing protein scaffolds. It addresses the challenge of designing proteins with novel functions by enabling the repurposing of protein folds, even when the functional motif and scaffold have low structural similarity.

What Problem Does This Paper Address?

Designing proteins with specific functions (like binding or catalysis) computationally is a major challenge. While we can design proteins that fold well, imbuing them with precise biochemical functions is much harder. Current methods often struggle when the desired functional part of a protein (a "motif") doesn't structurally match the "host" protein scaffold where we want to place it. This "designability" problem limits our ability to create new proteins for applications ranging from therapeutics to biosensors. The authors aim to overcome this by developing a more flexible and robust computational approach that can effectively "graft" functional motifs onto diverse protein scaffolds, even those with significantly different structures.

Methods

The core of this work is the Rosetta FunFolDes protocol, an evolution of a previous method called Rosetta Fold From Loops (FFL). FunFolDes integrates protein folding and sequence design, allowing the protein scaffold to adapt its structure to accommodate the inserted functional motif. Key improvements include:

  • Multi-segment functional sites: FunFolDes can now handle functional motifs composed of multiple, discontinuous segments, which is common for real-world protein functions.
  • Target-biased design: The protocol can incorporate a "binding target" (another protein, DNA, or small molecule) into the simulation. This "functional constraint" biases the design process towards sequences and structures that are compatible with binding, even if it means deviating from the lowest energy state.
  • Region-specific constraints: Users have finer control over which parts of the protein are flexible and which are constrained, allowing for targeted structural adjustments.
  • On-the-fly fragment picking: Instead of relying on pre-generated fragment libraries, FunFolDes can generate these "building blocks" based on the input scaffold's structure, improving efficiency and applicability.
  • Integration with RosettaScripts: This allows for flexible customization and integration with other Rosetta tools.

Datasets

The paper uses several datasets for benchmarking and validation:

  • CATH database: Used to identify and select 14 benchmark targets (T01-T14) representing small protein domains with natural sequence insertions/deletions and conformational changes.
  • BINDI-BHRF1 system: A previously studied protein-protein interaction system, used to test FunFolDes's ability to design binders in the presence of a target.
  • RSVF site II epitope: A viral epitope used to functionalize the 1kx8 protein scaffold.
  • RSVF site IV epitope: Another viral epitope used to functionalize the de novo designed TOP7 protein.
  • Saturation mutagenesis data for BINDI: Provided by Procko et al., this data was used to evaluate the sequence sampling quality of FunFolDes.

Key Insights and Ideas

  • Functionality can trump global energy: Designs optimized for function (e.g., binding) may not reside at the global energetic minimum of the protein. This suggests that "functional constraints" are crucial for designing proteins with specific tasks, potentially requiring a trade-off with overall stability.
  • Repurposing distant folds is feasible: FunFolDes can successfully graft functional motifs onto protein scaffolds with low structural similarity, significantly expanding the pool of potential host proteins. This opens doors to re-engineering existing protein structures for entirely new purposes.
  • Flexibility is key to designability: Allowing the protein scaffold to adapt its conformation (backbone flexibility) is essential for accommodating functional motifs, especially when structural compatibility is low.
  • "Functional constraint" is a powerful design principle: Incorporating the binding target directly into the simulation guides the design process towards more functional outcomes, even if it leads to less stable structures.

Key Results

  • Benchmark performance: FunFolDes outperformed constrained ab initio simulations in recovering target structures and sequences across 14 benchmark cases, particularly in scenarios involving structural changes. Structure-based fragments generally led to better structural recovery.
  • BINDI binder design: Simulations including the binding target (especially with side-chain repacking and minimization) yielded designs with significantly better binding energies and higher scores in saturation mutagenesis assays compared to designs made without the target.
  • 1kx8 functionalization: A viral epitope was successfully grafted onto the 1kx8 protein scaffold, resulting in a stable, monomeric protein (1kx8_d2) that bound a monoclonal antibody with a K<sub>D</sub> of 1.14 nM, a significant improvement over previous designs.
  • TOP7 functionalization: A de novo designed, "functionless" protein (TOP7) was successfully functionalized with a viral epitope, creating a protein (TOP7_full) that bound its target antibody with a K<sub>D</sub> of 24.2 nM. Notably, this functionalization led to a decrease in thermal stability compared to the original TOP7.

Limitations and Future Work

  • Stability-function trade-off: The paper highlights that achieving function might come at the cost of reduced protein stability, a phenomenon observed in both the BINDI and TOP7 designs. Future work could explore strategies to balance these competing requirements.
  • Computational cost: While FunFolDes aims to be more accessible, complex simulations can still be computationally intensive.
  • Experimental validation: The paper demonstrates successful experimental characterization of designed proteins, but further validation across a wider range of functions and scaffolds would be beneficial.
  • Broader applicability: The authors suggest that FunFolDes's ability to repurpose diverse protein folds could be applied to a vast range of translational applications, including biotechnology, biomedicine, and biomaterials. The integration with RosettaScripts also allows for further customization and expansion of its capabilities.

Summary updated

BibTeX
@article{bonet2018rosetta,
  author = {Jaume Bonet and Sarah Wehrle and Karen Schriever and Che Yang and Anne Billet and Fabian Sesterhenn and Andreas Scheck and Freyr Sverrisson and Barbora Veselkova and Sabrina Vollers and Roxanne Lourman and Mélanie Villard and Stéphane Rosset and Thomas Krey and Bruno E. Correia},
  title = {{Rosetta FunFolDes - A general framework for the computational design of functional proteins}},
  journal = {PLOS Computational Biology},
  year = {2018},
  doi = {10.1371/journal.pcbi.1006623},
}
Figure