Iterative Epitope Expansion Enables Structure-Guided <i>de novo</i> Design of IgE-Binding Miniproteins Targeting the Fc <i>ε</i> RI Interface

Abstract Designing de novo protein binders to multidomain assemblies remains challenging due to their structural complexity, conformational diversity, and broad functional epitopes. Here, we present an iterative epitope-expansion strategy that enables structure-guided design of compact miniproteins targeting human immunoglobulin E (IgE). Direct attempts to design binders against the full Fc ε RI-binding interface of IgE produced candidates with poor in silico confidence and limited experimental success. Instead, by first targeting a nearby, geometrically tractable seed epitope on the IgE C ε 3 domain, we obtained initial binders that served as anchors for progressive, structure-guided expansion toward the full receptor-binding interface. Iteration through an integrated AI–human–wet-lab design loop yielded a second-round miniprotein with approximately 30-fold improved affinity and substantial structural overlap with the Fc ε RI-binding site. Alanine-scanning mutagenesis confirmed that loss-of-function substitutions mapped precisely to the designed binding interfaces, validating the structural models. Polyspecificity profiling revealed that two of three binders exhibited low off-target binding comparable to antibody benchmarks, whereas one design displayed elevated polyspecificity that was strongly amplified upon Fc fusion. Notably, a previously published high-affinity miniprotein benchmark exhibited similar format-dependent liabilities, underscoring an underappreciated developability risk in the de novo miniprotein literature. Competition assays further demonstrated that two binders occupy regions of IgE that overlap with the Fc ε RI binding interface, as evidenced by loss of miniprotein binding when Fc ε RI is pre-bound to IgE Fc, consistent with the intended design geometry. Together, these results establish epitope expansion as a generalizable strategy for steering generative protein design toward challenging multidomain epitopes and highlight the necessity of tightly coupled AI–human–experimental workflows for achieving both functional specificity and developability in de novo binder discovery.