Tushar Jain, Tingwan Sun, Stéphanie Durand, Amy Hall, Nga Rewa Houston, Juergen H. Nett, Beth Sharkey, Beata Bobrowicz, Isabelle Caffry, Yao Yu, Yuan Cao, Heather Lynaugh, Michael Brown, Hemanta Baruah, Laura T. Gray, Eric M. Krauland, Yingda Xu, Maximiliano Vásquez, K. Dane Wittrup

PNAS 114(5), 944–949. DOI: 10.1073/pnas.1616408114

January 31, 2017

This study by Adimab scientists aimed to empirically define boundaries for developability in therapeutic antibodies. We hypothesized that analyzing the biophysical properties of antibodies that had successfully advanced through clinical development or received approval, would reveal acceptable ranges for various biophysical metrics and help us understand the implications of developability risks.

The overarching objectives of this work were to:

• Establish analogous guiding principles for antibody drugs, similar to Lipinski's Rule of Five for small molecules.
• Quantify the historical limits of acceptability for various biophysical metrics related to developability by evaluating a large panel of clinical-stage antibodies. For antibodies, critical characteristics include high-level expression, low polyspecificity, and low immunogenicity.
• Provide practical guidelines for prioritizing future antibody drug candidates.
• Offer a prescription to filter or rank order candidates for therapeutic development.
• Generate a reference dataset that encompasses the full set of antibody drugs tested in advanced clinical trials (Phase 2 or 3) or approved by the FDA, along with their biophysical properties. This dataset could then be used for sequence/function correlation analyses by others.

We also aimed to reconcile the observation that some clinical-stage antibodies exhibited purportedly unfavorable biophysical properties with their successful progression by reframing these as "red flags" that might have been resolved during bioprocess and formulation development. We were particularly interested in seeing how specific high-throughput assays delineate the space of antibody drugs in clinical development.

Methodology and technical approach

To understand how biophysical properties correlate with clinical success, our scientists analyzed 137 antibodies sequences drawn from advanced clinical-stage programs, including many that have received FDA regulatory approval. For consistency, all antibodies were expressed in the same IgG1 format and produced in mammalian cells, creating a standardized platform for comparison across different molecules.

Each antibody was put through a suite of 12 biophysical assays covering key developability characteristics such as self-interaction, polyspecificity, hydrophobicity, thermal stability, expression, and aggregation. The selected assays are widely used in therapeutic antibody development and were chosen for their compatibility with early-stage screening in the antibody discovery cascade due to their low material requirements and scalability.

For data analysis, we first examined the distributions of assay results across all antibodies. Since many measurements were non-normally distributed, we employed Spearman's rank correlation to assess relationships among biophysical properties. Unsupervised hierarchical clustering was used to explore how both antibodies and assays grouped based on shared characteristics.

To assess how specific biophysical attributes tracked with clinical success, we empirically defined a “red flag” threshold for each assay, marking the worst-performing 10% of values observed among approved antibodies. This conservative benchmark was modeled after the approach used in Lipinski’s Rule of Five for small molecules. Antibodies were then scored by the number of red flags they carried.

Statistical tests, including the Fisher’s exact test (for categorical comparisons) and Anderson-Darling test (to compare distribution shapes), were used to determine whether differences between approved and non-approved antibodies were statistically significant. This allowed us to link specific developability traits to the likelihood of clinical advancement and to identify potential liabilities earlier in the development pipeline. 

Key findings

Our scientists observed that biophysical properties relevant to antibody developability often exhibit long-tailed, non-normal distributions, with some molecules showing markedly unfavorable values. Despite this, many such antibodies still progressed through clinical development, indicating that poor early biophysical scores do not always predict failure. This suggests that downstream process optimization, such as improved manufacturing processes or formulation, can help overcome certain liabilities, though overall drug pharmacokinetic performance may still be affected due to other causes, such as target-mediated clearance.

A key trend emerged: antibodies with fewer biophysical “red flags” were more likely to succeed in clinical development. Approved antibodies had the fewest red flags overall, and a distinct “clean” cluster, defined by low risk across multiple properties, contained the majority of approved therapeutics (34 out of 48). Clustering analysis also showed that certain biophysical assays captured related molecular behaviors, providing insight into how different developability metrics interrelate. Notably, antibodies derived from phage display platforms were more likely to exhibit high-risk characteristics than those from mammalian systems, reinforcing previous observations.

Implications for therapeutic antibody development

This study provides analogous guiding principles for antibody drugs by defining empirical boundaries of drug-like behavior based on extensive biophysical analysis of clinical-stage molecules. These findings have several implications for antibody development:

• Prioritization of candidates: The reference dataset and the established cutoffs for acceptable biophysical properties can serve as practical guidelines for prioritizing future antibody drug candidates for development. This allows for early-stage assessment of developability risks, which is crucial given the high cost of late-stage failures.
• Resource management: Candidates with more favorable developability profiles are more likely to fit within existing platform approaches for process development and formulation. Conversely, molecules with suboptimal characteristics may demand additional resources and efforts for successful development and formulation.
• Holistic assessment: The study emphasizes the need for a holistic approach, involving weighted consideration of a broad panel of biophysical assays with complementary and overlapping characteristics. No single assay definitively predicted failure to advance to clinical trials; therefore, a comprehensive evaluation is necessary for making informed decisions on lead molecule selection and determining the need for further protein engineering or formulation efforts.

Like Lipinski's Rule of Five for small molecules, our study offers a data-driven guide to streamline discovery, highlight molecules that fit standard platforms, and flag those needing further engineering, ultimately improving decision-making, resource allocation, and the chances of clinical success.

For more details, read the full article in PNAS.