Determinants of antibody specificity

Abstract

High throughput screening methods such as yeast surface display (YSD) are frequently able to isolate high affinity antibodies against clinical targets; however, the success of these candidates depends on selecting for both on-target binding and desirable biophysical characteristics. Development liabilities, including antibody aggregation and nonspecificity, can lead to problems during production and poor pharmacokinetics (PK). The exact structural and sequence determinants causing this poor developability are unknown, which leads to inexact methods to correct otherwise promising clinical candidates. In this thesis we outline the development of high throughput methods to interrogate developability of candidate antibodies on the surface of yeast and apply these methods to both determine the causes of nonspecificity and create new libraries with improved biophysical properties. We first analyzed methods for early stage assessment of monoclonal antibodies, finding a polyspecificity reagent (PSR) binding assay on the surface of yeast which can accurately predict antibody clearance rates in mice. While robust, this assay relies on production of a poorly defined mixture of protein components, and thus, we next looked at potential alternatives to a multicomponent PSR reagent. We found that chaperone proteins may work as well-defined, easily producible reagents with similar broad predictive power to predict downstream antibody behavior. Next, we applied these assays to assess core determinants of nonspecificity. We first analyzed a case study of two antibodies with identical target antigens but vastly different performance on preclinical assessments of biophysical characteristics. Through this matched case, we found differences in clearance rates can be driven wholly by variable-region mediated effects independent of neonatal Fc receptor (FcRn) binding. Focused on the antibody variable region, we next utilized our nonspecificity assay as a sorting tool to look at a naive repertoire library. We found significant nonspecificity in the VH6 class of antibodies, driven by a poorly behaved complementarity determining region (CDR) H2 sequence. Subsequently, we applied a similar sorting technique to two synthetic library designs to identify a set of motifs that can drive nonspecificity. These included motifs containing tryptophan, valine, glycine and arginine located in CDR H3. We then applied these discoveries to the design of a new, semi-synthetic single chain variable fragment (scFv) library and demonstrated its ability to isolate high affinity, highly specific candidate clones against a panel of antigens. Finally, we explored the use of an alternate yeast display system capable of easily switching between scFv-Fc display and secretion, which may aid in the rapid development and testing of candidate antibodies. Taken in whole, the work in this thesis aids the clinical development of antibodies. We have presented both methods to assess nonspecificity at an early stage in the development process as well as a set of motifs to be eliminated in future library designs. With these combined findings, we hope to increase the utilization of in vitro screening methods such as yeast display for the isolation of clinical candidate antibodies with favorable biophysical characteristics.