Lipinski's rule of fives is a rule of thumb to predict whether a small-molecule lead is more likely to end up a drug or a dud, based on its pharmacological and biological properties.
To date, though, despite the fact that, in 2015, half of the 10 bestselling biopharma products were monoclonal antibodies (MAbs), a similar way of estimating how good a drug development candidate a given antibody is has been "missing," Dane Wittrup told BioWorld Insight. "And that's hurt the field."
In the Jan. 31, 2017, print issue the Proceedings of the National Academy of Sciences, Wittrup, who is the co-founder of antibody discovery and optimization company Adimab LLC, and his colleagues at Adimab took a first stab at developing such a set of rules.
The need for such rules is one sign of how far the field has come in terms of identifying antibodies that possess the primary key to success – ones that bind to their targets.
As technologies for finding good binders have improved, "you no longer have just one binder that you have to push forward with no matter what," Adimab's director of protein analytics, Yingda Xu, told BioWorld Insight.
But there is not a lot of understanding of how to sort that binder bounty into better or worse clinical development candidates – failure often comes the hard way, after significant effort and clinical development.
The team hopes to kick off a more general discussion about best practices for analyzing antibodies in terms of their clinical promise. "Part of the reason we started this kind of assessment is because of all the problems people reported during development," Xu said. "We want to move that forward to the discovery stage . . . the purpose [of the paper] is not to say, 'Oh, this drug is bad, that drug is bad.'"
In their work, the authors looked at a dozen biophysical characteristics of 137 antibodies that have progressed at least to phase II in the clinic, including nearly 50 approved antibodies.
For now, the answers are not cut and dried.
"We had expected to get a clear answer from this work, [but] . . . it's still impressionistic," Wittrup acknowledged of the dozen assays his team described. In fact, "some of the marketed drugs have really bad marks on some of these assays."
Xu stressed, though, that the purpose of the paper is to enable better antibody prioritization and ultimately better candidates to make their way forward. Nevertheless, "on any of these measures, there's a direction that's worse and a direction that's better." Individual assays could sort antibodies into broad categories of green, yellow and red flags, and the overall profile can be used to sort antibodies in terms of what the team termed "developability" in their paper.
And while each individual assay was not able to definitively predict failure, some combinations of assays appeared to be strongly predictive of problems. Antibodies that scored poorly on both the HIC assay, which measures an antibody's interaction with a nonbiological hydrophobic surface, and PSR, a measure of general reactivity, seem to be particularly likely to encounter problems.
"Not a single one of the approved drugs was double red against those two measures," Wittrup said.
The team also looked at how the different measures correlated with each other in terms of sorting antibodies into the best, middling and worst categories, and found that the 12 assays could be grouped into five clusters.
"If you're going to be efficient with your time, you probably shouldn't do two assays that are within the same cluster," Wittrup said.
The assays described in the current paper measure biophysical properties – characteristics such as how likely the antibody was to interact with other antigens besides its intended target. The Adimab scientists are also looking at biochemical properties that could be liabilities, such as oxidation, isomerization, deamidation and a number of others.
Like the ability to prioritize development candidates, the necessity of looking for biochemical liabilities is the antibody version of a First World problem, brought about because the technology available for characterization has improved massively.
"In the old days," Wittrup said, "those things just sailed through, because no one could see them."