Computational Recipe for Designing Antibodies against the Ebola Virus

Amir Barati Farimani, N. R. Aluru, Emad Tajkhorshid, Eric JakobssonbioRxiv 2020 · DOI

TL;DR

This paper presents a computational "recipe" that combines bioinformatics and molecular dynamics simulations to predict how viruses like Ebola will mutate to escape antibodies and to design new synthetic antibodies that can effectively neutralize these mutated viruses. This approach offers a powerful way to proactively develop antiviral therapies against rapidly evolving pathogens.

What Problem Does This Paper Address?

Viruses, particularly those like Ebola with high fatality rates and a propensity for rapid mutation, pose a significant threat to public health. When a virus mutates, it can evade the immune system's antibodies, rendering existing treatments ineffective. The challenge is to anticipate these viral escape mutations and design antibodies that can neutralize them before they emerge and cause widespread outbreaks. Traditional methods of antibody development are often reactive and time-consuming, making it difficult to keep pace with viral evolution.

Methods

The authors developed a multi-pronged computational strategy:

  • Bioinformatics and Structural Biology: They started with existing crystal structures of Ebola virus glycoprotein (GP) bound to neutralizing antibodies (KZ52 for Zaire strain, 16F6 for Sudan strain).
  • Molecular Dynamics (MD) Simulations: Extensive MD simulations were performed on both wild-type and computationally generated mutated GP-antibody complexes. This allowed them to analyze the stability and interactions at the binding interface.
  • Contact List Generation: MD trajectories were used to identify "thermodynamically stable residues" at the GP-antibody interface, creating a "contact list" of key interacting amino acids.
  • Mutation Prediction (SIFT & PAM):
    • The Sorting Intolerant From Tolerant (SIFT) algorithm was used to predict mutations in the viral GP that are likely to preserve its function, thus identifying potential escape mutations.
  • Position Acceptable Mutations (PAM) scores were then used to rank these predicted mutations based on their natural frequency in protein databases, prioritizing the most probable ones.
  • Combinatorial Mutagenesis: The identified GP mutations and potential antibody binding site mutations were combined to generate a large matrix of possible mutated GP-antibody complexes.
  • Interaction Energy Analysis: MD simulations were run on these combinatorial mutants to calculate interaction energies, with lower (more negative) energies indicating stronger, more stable binding.

Datasets

The primary data used were:

  • Crystal structures of Ebola virus glycoprotein (GP) bound to antibodies: Specifically, the Zaire Ebola virus GP-AB complex (PDB: 3CSY) and the Sudan GP-engineered antibody complex (PDB: 3S88).
  • Databases of mutations and their functional consequences: Used by the SIFT algorithm.
  • Sequence and structure databases: Used by the PAM algorithm.

The paper does not introduce a new benchmark dataset but rather utilizes existing structural data and computational tools to generate and analyze a vast number of potential mutations.

Key Insights and Ideas

  • The "Evolutionary Arms Race" in silico: The study effectively models the dynamic interplay between viral evolution and antibody response, allowing for proactive design rather than reactive development.
  • Identifying Critical Contact Residues: MD simulations proved crucial in pinpointing the specific amino acids at the GP-antibody interface that are most important for binding stability, going beyond static crystal structures.
  • Prioritizing Mutations: The combination of SIFT and PAM provides a robust method for filtering down the astronomical number of possible viral mutations to the most likely and functionally significant ones.
  • Antibody Redesign for Broad Efficacy: The methodology demonstrated the ability to redesign antibody sequences to effectively neutralize a wide range of predicted viral mutations, suggesting a path towards broad-spectrum antivirals.

Key Results

  • The computational recipe successfully predicted likely viral escape mutations and identified candidate synthetic antibodies capable of inhibiting them.
  • MD simulations of wild-type and mutated GP-AB complexes revealed that while Zaire and Sudan GP-AB interactions are stable within their own strains, cross-strain binding is weak.
  • The study simulated 1,270 mutated GP-AB complexes, with 64 mutation pairs showing predicted successful inhibitory responses, exhibiting more favorable interaction energies than the wild-type antibody.
  • The method was able to predict mutations that could make current antibodies bind even stronger than the wild-type structure.
  • An alanine-scanning study on Ebola Zaire GP results matched well with the predicted viral escape mutations.

Limitations and Future Work

  • Computational Cost: While the method aims to be efficient, the sheer number of potential mutations still requires significant computational resources. The authors acknowledge that constraints on their search algorithm might have prevented finding even more effective antibodies in some cases.
  • In vivo Efficacy: The study focuses on computational predictions. The ultimate effectiveness of the designed antibodies in a biological context (i.e., in vivo) remains to be experimentally validated. The paper suggests that an interaction energy of 10.5 kcal/mol might be sufficient, but this is a point of speculation.
  • Experimental Validation: The authors explicitly state their eagerness to collaborate with experimental projects to test their computational methods.
  • Broader Applicability: The authors believe their approach can be widely applied to other viruses where structural data of viral coat protein-antibody pairs are available, suggesting future work in applying this methodology to different viral targets.
  • Proactive Antibody Design Pathway: They envision a future where this method is used proactively: 1) obtain viral coat protein-antibody structures, 2) project future viral mutations, 3) compute counteracting antibody sequences, and 4) engineer synthetic antibodies.

Summary updated

Abstract

Abstract A conceptual basis for antiviral therapy is to deliver a synthetic antibody that binds to a viral surface protein, and thus prevents the virus from deploying its cell-entry mechanism. The fast and untraceable virus mutations take lives of thousands of people before the immune system can produce the inhibitory antibody. In this paper, we devised a computational recipe to predict both the viral escape mutations and the possible inhibitory synthetic antibodies. We combined bioinformatics, structural biology, and molecular dynamics (MD) simulations to explore the most likely viral mutations and the candidate antibodies that can inhibit those escape mutations. Specifically, using the crystal structures of the Sudan and Zaire Ebola viral GPs in complex to their respective antibodies (ABs), we have performed an extensive set of MD simulations, both on the wild-type structures and on a large array of additional complexes designed and generated through combinatorial mutations. We discovered that our methods enabled the successful redesign of antibody sequences to essentially all likely glycoprotein mutations. Our findings and the computational methodology developed here for general antibody design can facilitate therapy of current and possibly next generations of viruses. Significance of the Manuscript This manuscript has high significance both methodologically and in potential biomedical application. In methodology, the manuscript combines molecular dynamics, Monte Carlo calculations, and bioinformatics in a novel way to simulate the evolutionary arms race between an evolving viral coat protein and a counter-evolving antibody against the virus. This simulation is shown to provide a method for designing a synthetic antibody against the newly emerging viral strains. This work is done in the context of ongoing work in other laboratories in which cells can be induced to produce synthetic antibodies and those synthetic antibodies can be edited (via, for example, CRISPR) to have an arbitrary sequence in the region that binds the viral coat protein. Putting those experimental methods together with the computational methods we present in this paper has the potential to provide a important approach to produce antibodies-on-demand against evolving viruses.
BibTeX
@misc{farimani2020computational,
  author = {Amir Barati Farimani and N. R. Aluru and Emad Tajkhorshid and Eric Jakobsson},
  title = {{Computational Recipe for Designing Antibodies against the Ebola Virus}},
  year = {2020},
  doi = {10.1101/2020.04.04.024430},
  howpublished = {bioRxiv},
}
Figure