Population-specific design of de-immunized protein biotherapeutics

Benjamin Schubert, Charlotta Schärfe, Pierre Dönnes, Thomas A. Hopf, Debora S. Marks, Oliver Kohlbacher

TL;DR

This paper introduces a novel computational method for designing protein biotherapeutics that are less likely to trigger an immune response in patients. By simultaneously optimizing for reduced immunogenicity and preserved protein function, and crucially, by tailoring the design to specific patient populations, this approach promises safer and more effective treatments.

What Problem Does This Paper Address?

Biotherapeutics, like protein-based drugs, are powerful tools for treating a wide range of diseases. However, a major hurdle in their development is immunogenicity: the tendency of the body's immune system to recognize these drugs as foreign and mount an attack. This can lead to rapid clearance of the drug, rendering it ineffective, and can even cause adverse reactions. A key driver of this immune response is the interaction of the drug with Human Leukocyte Antigen (HLA) molecules on antigen-presenting cells, which then present fragments (epitopes) of the drug to T-cells.

Current methods for reducing immunogenicity, known as de-immunization, often focus on removing these T-cell epitopes. While moderately successful, these approaches have limitations: they may not account for the varying frequencies of different HLA alleles across diverse human populations, and they can sometimes compromise the drug's essential function or stability. This necessitates costly experimental screening to identify viable candidates. The authors aim to address these limitations by developing a more sophisticated, population-specific de-immunization strategy.

Methods

The core of this work is a bi-objective combinatorial optimization framework. This means the method simultaneously optimizes two competing goals:

  1. Minimizing Immunogenicity: This objective is novel in that it considers the specific HLA allele frequencies of a target population, rather than treating all alleles equally. It uses a state-of-the-art continuous immunogenicity prediction method (TEPITOPEpan) to score potential epitopes.
  2. Maximizing Protein Fitness: This objective approximates the protein's functional fitness using a maximum entropy model. This model is inferred from multiple sequence alignments (MSAs) of protein families and captures evolutionary constraints, allowing it to predict the impact of mutations on protein stability and function without requiring explicit 3D structural information.

The problem is formulated as a Bi-Objective Mixed Integer Linear Program (BOMILP), which is solved using a novel parallel two-phase rectangle-splitting approach. This algorithm efficiently explores the vast search space of possible mutations to find a set of Pareto-optimal solutions – sequences that represent the best possible trade-offs between immunogenicity and fitness.

Datasets

The study utilizes data from the Factor VIII variant database to train and validate their models for predicting hemophilia A severity. For model inference, multiple sequence alignments (MSAs) of the Factor VIII C2 domain were used. Experimental validation involved synthesizing overlapping 15-mer peptides containing introduced mutations and their wild-type counterparts.

Key Insights and Ideas

  • Population Specificity is Crucial: Recognizing that HLA allele frequencies vary significantly across populations is key. Designing de-immunized therapeutics tailored to these specific frequencies can lead to more effective and safer treatments for targeted patient groups.
  • Sequence-Based Fitness is Sufficient: The authors demonstrate that a sequence-based approach, leveraging evolutionary information, can effectively predict protein fitness and guide de-immunization, potentially bypassing the need for complex and sometimes incomplete structural information.
  • Balancing Act: The bi-objective optimization framework elegantly handles the inherent trade-off between reducing immunogenicity and maintaining protein function, providing a spectrum of optimized designs rather than a single solution.
  • Evolutionary Couplings as Guides: Identifying regions enriched in evolutionary couplings (ECs) highlights functionally important sites, allowing the method to avoid disrupting critical protein features while de-immunizing.

Key Results

  • The developed method successfully designed 21 Pareto-optimal sequences for the Factor VIII C2 domain with up to three simultaneous point mutations.
  • These designed sequences showed a significant reduction in predicted immunogenicity (up to 45%) while maintaining high predicted protein fitness, with 20 out of 21 designs falling within the 95th percentile of wild-type fitness.
  • Experimental validation showed a strong correlation between predicted and measured immunogenicity for both individual peptides (r=0.76) and the reconstructed protein region (r=0.85).
  • The best designed mutant (Design-11) achieved a 45% immunogenicity reduction with only a 1.7% predicted fitness loss.

Limitations and Future Work

  • MSA Availability: The reliance on high-quality, diverse sequence alignments for inferring maximum entropy models means the approach might be limited when such alignments are not available.
  • Experimental Validation Scope: While experimental validation was performed on peptides and a specific domain, full validation of the designed protein constructs in vivo would be the next logical step.
  • Future Directions: The authors suggest that their approach can be further validated using mouse models or T-cell proliferation assays on PBMCs from HA patients. They also highlight the potential for bioengineers to use this method to explore design spaces and select promising candidates for experimental evaluation.

Summary updated

Abstract

Immunogenicity is a major problem during the development of biotherapeutics since it can lead to rapid clearance of the drug and adverse reactions. The challenge for biotherapeutic design is therefore to identify mutants of the protein sequence that minimize immunogenicity in a target population whilst retaining pharmaceutical activity and protein function. Current approaches are moderately successful in designing sequences with reduced immunogenicity, but do not account for the varying frequencies of different human leucocyte antigen alleles in a specific population and in addition, since many designs are non-functional, require costly experimental post-screening. Here we report a new method for de-immunization design using multi-objective combinatorial optimization that simultaneously optimizes the likelihood of a functional protein sequence at the same time as minimizing its immunogenicity tailored to a target population. We bypass the need for three-dimensional protein structure or molecular simulations to identify functional designs by automatically generating sequences using probabilistic models that have been used previously for mutation effect prediction and structure prediction. As proof-of-principle we designed sequences of the C2 domain of Factor VIII and tested them experimentally, resulting in a good correlation with the predicted immunogenicity of our model.
BibTeX
@article{schubert2018population,
  author = {Benjamin Schubert and Charlotta Schärfe and Pierre Dönnes and Thomas A. Hopf and Debora S. Marks and Oliver Kohlbacher},
  title = {{Population-specific design of de-immunized protein biotherapeutics}},
  journal = {PLOS Computational Biology},
  year = {2018},
  doi = {10.1371/journal.pcbi.1005983},
}
Figure