471705 Computationally-Driven Design of a Heterobivalent and Gate to Improve Targeting Agent Selectivity
The AND gate molecule is a heterobivalent ligand comprising two binding domains linked via a flexible peptide. The two binding domains target two disease-associated biomarkers with a weak affinity that precludes monovalent binding to singly overexpressing cells while enabling robust bivalent binding to dually overexpressing cells. Optimal AND gate design – including individual affinities of the binding domains, linker length, and linker structure – depends on several properties of the both the tumor and the biomarker, such as average cell surface expression, internalization rate, homogeneity across cell membrane, and potential agonistic/antagonistic effects. To guide quantitative design, a geometric, cell-surface binding model was developed and integrated with a literature biodistribution model (Thurber and Wittrup J. Theor. Biol. 2012). The hybrid model indicates that, given a surface expression of 100,000 biomarkers per cell (each), a 100-fold increase in specificity can be achieved over cells expressing 100,000 of only one biomarker with individual ligand affinities of 125-575 nM. Stronger affinities hinder selectivity while weaker affinities preclude even bivalent binding. The optimal flexible linker length, to maximize local ligand concentration after binding of the first domain, is 22 amino acids for a receptor radius of 20 Å. Differential results are observed for flexible and rigid linkers.
These optimal ligand properties were then engineered into an experimental heterobivalent ligand. In our model system, the binding domains are two non-antibody protein scaffolds (10th type III domain of fibronectin (Fn3) domains) that have been previously engineered to bind either epidermal growth factor receptor (EGFR, clone A’) or carcinoembryonic antigen (CEA, clone C7) with initial affinities of 0.42 ± 0.06nM and 15 ± 15 nM, respectively. To obtain ligands with affinities within the 125-575 nM range, yeast surface display libraries, created by mutagenesis of each parental gene, were sorted by flow cytometry for decreased binding to the proper target. Mutants from the sorted sub-populations were produced and titrated on cancer cells, which revealed almost no correlation between yeast-displayed affinity to recombinant target and soluble affinity to the natively-expressed target. Nevertheless, several mutants for each target were identified; C7 mutants ranged from 2.5 to 5,600 nM and A’ mutants from 15 to 3,000 nM. To determine scaffold modularity, i.e., whether their affinities were affected by the fusion of a second Fn3 domain, the parental ligands A’ and C7 were fused on either the N- or C- termini with a non-binding Fn3 clone. The fusion of a non-binding Fn3 to the C-termini (distal to the binding site) decreased the affinity of A’ 50-fold to 21 ± 12 nM and C7 affinity held at 16 ± 10 nM. When the non-binding Fn3 was fused to the N-termini (near the binding site), A’ affinity decrease was similar (18 ± 6 nM), while C7 showed a decrease in affinity to 89 ± 97 nM. Two mutants of A’ were also tested in the fusion context; A’ mutants A’.1 and A’.2, which have affinities of 28 ± 3 nM and 17 ± 4 nM, respectively, as single domains, displayed affinities of 28 ± 12 nM and 125 ± 93 nM, respectively, when fused to the non-binding Fn3 via their C-termini. An additional high-affinity Fn3 binder to EGFR (clone D) demonstrated no effect when fused via its C-terminus (0.4 ± 0.1 nM) compared to the monovalent affinity of 0.25 nM. These variations in fusion modularity demonstrate the importance of quantifying the in vitro affinity of fusion components – in the fusion context – and highlight dependence on unknown binding paratope, receptor binding epitope, binding orientation, and structural folding variations. Ongoing efforts include identifying the structural effect of the fusion via circular dichroism and SEC analysis.
For initial AND gate testing, three heterobivalent ligands were synthesized: A’-C7, A’.1-C7, and A’.2-C7, where the N-terminus of C7 was fused to the C-terminus of the three EGFR binding Fn3 via a glycine-rich linker to test their selectivity on native and modified cell lines of varying EGFR and CEA expression. Preliminary data of the A’-C7, A’.1-C7, and A’.2-C7 fusions on A431 cells ( <2,000 CEA, 4,000,000 EGFR per cell) show that the fusion of C7 via the C-terminus further decreases the EGFR affinity of the fusion compared to the non-binding Fn3 fusion to 102 ± 31 nM, 443 ± 238 nM, and 1200 ± 1000 nM, respectively. However, titration of A’.1-C7 on lentiviral-transfected HEK-293T cells (250,000 CEA, <2,000 EGFR receptors per cell) results in an affinity of 69 ± 10 nM, showing that the C7 affinity is indistinguishable when fused to either A’.1 or the non-binding Fn3. This data shows that the effect of the fusion can be dependent on the clone you are fusing, which is crucial when accurately measuring the properties of the system, and the importance of cell lines with accurately characterized expression to quantify individual domain affinities within the heterobivalent ligand. Ongoing studies, evaluating binding selectivity and sensitivity on cell lines with varying levels of CEA and EGFR, will be discussed. Additional results to be shared include evaluation of ligands of varying affinities, additional scaffolds, and alternative fusion methods to create a more modular system.