Biophysical fitness landscape design traps viral evolution

This paper introduces biophysical fitness landscape design (FLD), a computational framework that uses optimized antibody ensembles to quantitatively reshape viral fitness landscapes, thereby proactively suppressing the evolution of escape variants like those of SARS-CoV-2 before they emerge.

The Gist

Imagine evolution as a hiker trying to climb a mountain. The higher the hiker climbs, the "fitter" the virus becomes—it replicates faster, spreads more easily, and causes more trouble. In the past, scientists have tried to predict where the hiker will go next by studying the mountain's shape. But this paper introduces a revolutionary idea: What if we could redesign the mountain itself?

The authors, Vaibhav Mohanty and Eugene Shakhnovich, propose a method called Fitness Landscape Design (FLD). Instead of just watching the virus evolve, they want to build a "trap" that forces the virus to stay in a low valley, unable to climb to the top.

Here is how it works, using simple analogies:

1. The Problem: The Virus is a Master Escapist

Think of a virus like a pickpocket. When you vaccinate against it (like putting a security guard on the pickpocket), the pickpocket changes their clothes (mutates) to slip past the guard.

• Current Strategy: We usually react. We see a new "outfit" (variant), make a new vaccine, and hope it works. But by the time we make the new vaccine, the virus has often already changed again. It's a game of "whack-a-mole" where the mole is always one step ahead.
• The Goal: Instead of chasing the mole, we want to build a cage that makes it impossible for the mole to grow big enough to escape, no matter what outfit it wears.

2. The Solution: Redesigning the Mountain

The authors treat the virus's evolution like a hiker on a mountain.

• The Old Way: We assume the mountain is fixed. The virus just climbs the highest peak available.
• The New Way (FLD): We use computer algorithms to reshape the mountain. We can dig deep valleys where the virus wants to go, and we can flatten the peaks so there's nowhere high to climb.

How do we do this? By designing antibodies (the "security guards") that act like the terrain itself.

• If the virus tries to mutate to escape one antibody, our new design ensures that the "escape route" leads to a dead end or a steep cliff where the virus dies out.
• We aren't just blocking one path; we are redesigning the entire map so that any path the virus takes leads to a dead end.

3. The "Chess Master" Approach

The paper describes a method called Iterative FLD-A. Imagine playing chess against a very smart opponent (the virus).

• Standard Vaccination: You make a move to protect your King (the wild-type virus). The opponent finds a way to checkmate you anyway.
• The FLD Strategy: You think three moves ahead. You ask, "If I make this move, what will my opponent do? And if they do that, what will I do next?"
  • The computer simulates millions of future moves.
  • It finds a specific "target" (a specific part of the virus) that, if vaccinated against, forces the virus into a position where its best possible move is still a losing one.
  • It essentially builds a "fitness trap." Even if the virus evolves, it can only reach a low, weak version of itself, not a dangerous super-variant.

4. The "Lego" Analogy

Think of the virus's surface as a set of Lego bricks.

• Old Science: We tried to build a wall to stop the Lego structure from moving.
• This Paper: We realized we can change the shape of the Lego bricks themselves. By designing a specific set of "anti-Lego" antibodies, we ensure that if the virus tries to snap on a new brick to escape, the whole structure collapses or becomes too heavy to move.

5. Why This Matters

The authors tested this on SARS-CoV-2 (the virus that causes COVID-19).

• They showed that they could computationally design a set of antibodies that would force the virus to evolve into a "low-fitness" state.
• Instead of the virus finding a way to become more dangerous, the vaccine would force it to become less dangerous, effectively trapping it in a corner where it can't spread effectively.

The Big Picture

This paper is like moving from reactive defense (fixing the roof after the rain starts) to proactive architecture (designing a house that never leaks, no matter how hard it rains).

By using physics and math to "sculpt" the evolutionary landscape, we can stop pandemics before they start. Instead of waiting for the next scary variant to appear, we can design vaccines that anticipate the virus's moves and trap it before it ever gets the chance to escape. It's the difference between chasing a runaway train and building a track that leads to a gentle, safe stop.

Technical Summary

1. Problem Statement

Current approaches to vaccine and therapeutic design are largely reactive, targeting currently prevalent viral strains. However, rapidly evolving viruses (e.g., SARS-CoV-2, influenza) quickly mutate to escape immune pressure, leading to "escape variants" that render vaccines less effective. This creates a vicious cycle where vaccines must be updated frequently.

The core challenge identified is the lack of methodology to proactively control the evolutionary trajectory of a virus. While evolutionary biology has long used "fitness landscapes" to visualize how genotypes map to reproductive success, these landscapes have traditionally been viewed as fixed or externally determined. The authors propose solving the "inverse problem" of Fitness Landscape Design (FLD): Instead of predicting evolution on a given landscape, can we design the biophysical interactions (specifically antibody repertoires) to reshape the fitness landscape itself? The goal is to engineer a landscape that suppresses the fitness of escape variants, effectively "trapping" the virus in a low-fitness state before it can evolve.

2. Methodology

The authors developed a comprehensive computational framework called Fitness Landscape Design with Antibodies (FLD-A). The methodology consists of three main pillars:

A. Biophysical Fitness Modeling

The authors derived analytical models linking viral genotype to fitness based on microscopic chemical reaction dynamics:

• In Vitro Model: Derived from reversible binding of viral antigens to host receptors and competitive inhibition by antibodies, coupled with irreversible viral replication. The fitness $F(s)$ is proportional to the probability of the viral protein binding to the host receptor in the presence of competing antibodies.
  $$F(s) \approx k_{rep}(N_0 - 1)N_{ent} p_b(s)$$
  Where $p_b(s)$ is the equilibrium probability of host binding, dependent on binding free energies ($\Delta G$) of host-antigen and antigen-antibody interactions.
• In Vivo Model: Extended the model to include immune-mediated clearance (Fc$\gamma$R-mediated removal of virus-antibody complexes) and multiple epitopes. This model accounts for both the reduction in cell entry (competitive inhibition) and the active removal of virions.

B. Validation of the Model

Before applying the design algorithms, the authors validated the biophysical models against independent data sources:

• MNV-1 (Murine Norovirus): Used serial passage experimental data (with and without antibodies) to show the model accurately predicts empirical fitness changes using EvoEF force field calculations.
• SARS-CoV-2: Used global epidemiological sequencing data (PyR0 inferred fitness) and Tite-Seq binding affinity data to show the in vivo model accurately fits the fitness of real-world variants (e.g., Omicron sub-lineages).
• In Silico Simulations: Used BioNetGen to simulate chemical reaction dynamics and serial dilution, confirming that the derived fitness equations accurately predict the observed evolutionary trajectories in stochastic environments.

C. Optimization Algorithms

• oFLD-A (Optimal FLD-A): A stochastic optimization protocol using simulated annealing (Metropolis-Hastings algorithm). Given a user-defined target fitness landscape, the algorithm iteratively adjusts antibody paratope sequences and concentrations to minimize the least-squares error between the target landscape and the biophysical landscape generated by the antibodies.
• iFLD-A (Iterative FLD-A): A proactive vaccine design protocol. It simulates a "chess game" against the virus:
  1. Identify the current wildtype.
  2. Find the "escape peak" (highest fitness mutant) in the presence of the current antibody.
  3. Design a new antibody to target this escape peak.
  4. Repeat until the global fitness peak is minimized or a cycle is detected.
     The output is a target antigen and corresponding antibody that minimizes the maximum possible fitness of any future escape variant.

3. Key Contributions

1. Definition of Fitness Landscape Design (FLD): The paper formally defines FLD as an inverse problem in evolutionary biology, moving from observing evolution to actively controlling it.
2. Designability Phase Diagrams: The authors mapped the "codesignability" of fitness landscapes, demonstrating that for a set of viral genotypes, there exists a specific region of "designable" fitness assignments achievable by some antibody ensemble. They showed that distinct viral strains can be assigned independent fitness values if the antibody repertoire is sufficiently diverse and selective.
3. Quantitative Reshaping: The study proves that antibodies can not only suppress overall viral fitness but also mold relative fitnesses. They demonstrated the ability to:
   • Stretch: Exaggerate fitness differences between neutral networks.
   • Invert: Turn a fitness-increasing mutation (e.g., Q493R in SARS-CoV-2) into a deleterious one.
   • Suppress: Lower the absolute fitness of all variants while maintaining relative rankings.
4. Proactive Vaccine Strategy: The iFLD-A protocol introduces a method to design vaccines that target a "future" global fitness peak, effectively lowering the ceiling for all potential escape variants, rather than reacting to the current dominant strain.

4. Key Results

• Validation: The biophysical models achieved high correlation ($r > 0.86$ for MNV-1; $r > 0.91$ for SARS-CoV-2) with empirical fitness data, validating the use of chemical kinetics to predict evolutionary outcomes.
• Landscape Manipulation: Using oFLD-A, the authors successfully designed antibody ensembles for 256 SARS-CoV-2 RBD variants that matched a user-specified target landscape with near-exact quantitative accuracy ($r = 0.994$ in silico).
• Neutral Network Control: The study successfully manipulated two interconnected neutral genotype networks (G485 variants with and without Q493R). They demonstrated the ability to invert the fitness hierarchy, making the Q493R mutation deleterious rather than beneficial.
• Escape Suppression: In the iFLD-A simulations:
  • The iFLD-A antibody reduced the peak fitness of escape variants significantly more than a standard wildtype-targeting (WTT) antibody.
  • In Wright-Fisher evolutionary simulations (both polymorphic and near-monomorphic regimes), populations under iFLD-A pressure exhibited slower fitness growth and were trapped under a lower fitness ceiling compared to WTT pressure.
  • The iFLD-A antibody maintained strong neutralization of the wildtype while being optimized to suppress future peaks, achieving a "best of both worlds" scenario.

5. Significance

• Pandemic Preparedness: This work offers a paradigm shift from reactive to proactive vaccine design. By anticipating escape mutations and designing vaccines that suppress the fitness of all potential escape trajectories, it could break the cycle of annual vaccine updates required for viruses like influenza and SARS-CoV-2.
• Generalizability: While focused on viruses, the FLD-A framework is applicable to other rapidly evolving systems, including:
  • Cancer Therapeutics: Designing CAR-T cells or peptide drugs to reshape the fitness landscape of tumor cells, preventing immune escape.
  • Antimicrobial Resistance: Controlling the evolution of drug-resistant bacteria.
• Theoretical Advancement: It bridges the gap between biophysics (binding affinities) and evolutionary dynamics, providing a quantitative, mechanistic tool to control evolutionary outcomes. It transforms the fitness landscape from a theoretical visualization tool into a designable engineering parameter.

In conclusion, the paper establishes that by leveraging biophysical principles and stochastic optimization, it is possible to computationally design antibody ensembles that quantitatively reshape viral fitness landscapes, offering a robust strategy to trap viral evolution and improve long-term pandemic preparedness.