TL;DR
This paper introduces the Therapeutic Antibody Profiler (TAP), a computational tool that provides five in silico guidelines to predict potential developability issues in therapeutic antibodies. By analyzing a large dataset of clinical-stage antibodies, TAP identifies sequences that deviate from typical profiles, helping researchers flag problematic candidates early in the drug discovery process.
What Problem Does This Paper Address?
Developing therapeutic monoclonal antibodies (mAbs) is a complex process. Beyond simply binding to a target, these molecules must also possess favorable "developability" characteristics. These include good stability, low aggregation propensity, acceptable viscosity, and efficient expression. Unlike small-molecule drug discovery, which benefits from established rules like Lipinski's Rule of Five, there hasn't been a similar in silico framework for guiding antibody design. This lack of predictive tools means that potential developability issues are often discovered late in development, leading to costly setbacks. This paper aims to fill that gap by creating a computational system to flag antibodies with characteristics that are rare among successful clinical-stage therapeutics.
Methods
The core of this work involves developing and applying a computational tool called the Therapeutic Antibody Profiler (TAP). TAP utilizes homology modeling to generate 3D structures of antibody variable domains from their amino acid sequences. These models are then analyzed using five key metrics that have been linked to poor developability:
- Total CDR Length: The combined length of the complementarity-determining regions (CDRs), which influences antigen binding site shape.
- CDR Vicinity PSH (Patchy Surface Hydrophobicity): A metric that quantifies the extent and clustering of hydrophobic residues on the antibody surface, particularly in the CDR regions, as hydrophobicity is a known driver of aggregation.
- CDR Vicinity PPC (Positive Charge Patch) and PNC (Negative Charge Patch): Metrics that identify dense patches of positive or negative charges within the CDR vicinity, which can lead to issues like rapid clearance or poor expression.
- Structural Fv Charge Symmetry Parameter (SFvCSP): A measure of the asymmetry in net charge between the heavy and light variable domains, which can impact viscosity and self-association.
The guidelines for these metrics were derived by comparing the distributions of these properties in clinical-stage therapeutics (CSTs) against a snapshot of the natural human antibody repertoire. A flagging system (amber for unusual, red for previously unobserved) is proposed to highlight non-conforming candidates.
Datasets
The study utilized several key datasets:
- Clinical-Stage Therapeutics (CSTs): A dataset of 137 post-phase-I clinical-stage antibody sequences was used as the primary reference for "good" developability. A further set of 105 phase II+ CSTs was used for validation.
- Human Antibody Repertoire:
- Observed Antibody Space (OAS) database: Provided a snapshot of the human antibody gene repertoire (human VdH Ig-seq).
- UCB Pharma Ltd. proprietary dataset: A larger dataset of human antibody sequences (human UCB Ig-seq).
- Protein Data Bank (PDB): Used for structural modeling and validation, including a set of 33 non-engineered human mAb crystal structures.
Key Insights and Ideas
- Human antibodies aren't all "developable": While clinical-stage antibodies share many features with the natural human repertoire, the study suggests that not all naturally occurring human antibodies would be suitable therapeutics. This is likely due to the stresses (pH, temperature, concentration) that therapeutic antibodies face during development and storage, which natural antibodies do not experience.
- Structure matters, but models can suffice: The authors found that homology models generated by ABodyBuilder were sufficiently accurate for their developability metric calculations, even when compared to crystal structures. This is crucial as crystal structures are not always available.
- Hydrophobicity is a key culprit: The PSH metric revealed that clinical-stage antibodies tend to avoid large, clustered hydrophobic patches in their CDR vicinity, suggesting this is a significant factor in aggregation.
- Charge distribution is also critical: Both positive and negative charge patches, as well as charge asymmetry between heavy and light chains, were found to be important indicators of potential developability issues.
- A practical tool for early screening: The development of TAP provides a much-needed in silico tool that can be integrated into early discovery pipelines, potentially saving significant time and resources.
Key Results
- The TAP guidelines were developed based on five metrics: total CDR length, CDR vicinity PSH, CDR vicinity PPC, CDR vicinity PNC, and SFvCSP.
- Thresholds for these metrics were established by analyzing the distributions in 242 clinical-stage therapeutic antibodies.
- On two independent drug discovery sets, TAP successfully flagged antibodies with known developability issues (aggregation and poor expression) with amber or red flags. For example, a problematic antibody variant was red-flagged for high PSH, while its engineered version with resolved issues returned to normal values.
- The TAP tool is made freely available as a web application.
Limitations and Future Work
- Scope of Developability: The authors acknowledge that TAP does not capture all potential developability issues, such as immunogenicity or more subtle stability problems.
- Threshold Updates: The authors plan to regularly update the threshold values as new clinical-stage antibodies become available, reflecting the evolving landscape of therapeutic antibodies.
- Subclass Stratification: With more data, it may be possible to stratify guidelines for subclasses of antibodies (e.g., based on kappa vs. lambda light chains), as preliminary analysis suggests differences in hydrophobicity.
- Advancements in Formulation: The authors note that future advances in formulation and process development might shift the boundaries of what is considered acceptable for these developability metrics.
Summary updated