A Complementarity-Based Approach to De Novo Binder Design

Abstract

De novo design of binders capable of targeting arbitrarily selected epitopes remains a substantial challenge. Here, a generalizable computational strategy is presented to design site-specific protein binders, obviating steps of extensive empirical optimization or in vitro screening. The dock-and-design pipeline retrieves complementary scaffolds from a protein structure database to a given query epitope, where the scaffold is mutated to carve a binding site de novo. The docking step utilizes a novel fingerprint that greatly simplifies and accelerates the surface complementarity evaluation. As proof-of-concept, we designed protein binders to target three distinct epitopes on two different oncogenic targets; vascular endothelial growth factor (VEGF) and interleukin-7 receptor-α (IL-7Rα). Experimental characterization of only 24 candidates identified nanomolar binders against each of the target epitopes, where the binders belonged to five different folds. Several designs were active in vitro. Moreover, anti-VEGF designs showed tumor-inhibiting activity in vivo, highlighting their therapeutic potential.

Keywords: de novo binder design; IL‐7R binders; VEGF inhibitors; complementarity evaluation; protein‐protein docking.

Figure 1

Strategy to design epitope‐directed binders using the HECTOR software. A) The design workflow primarily relies on simplifying the complexity of the docking process and decoupling it from the design stage. The docking step particularly, is based on a novel surface mesh fingerprinting protocol that analytically defines a complementary surface fingerprint and trivially matches it to a query surface. This is followed by sequence design and a two‐step molecular dynamics‐based testing of binding interface stability, utilizing serial tempering simulations (left pane) and steered MD (right pane). B) Histograms show distributions of R‐factor for close and far subject maps, where close maps are located within 1 Å and far maps are located more than 1 Å from their corresponding query maps selected at the interface of split proteins. As the R‐factor decreases, the probability of identifying a close subject map increases. C) The GPU implementation of R‐factor calculation greatly accelerates map‐to‐map comparisons, compared to CPU. As the number of query maps scales up, the evaluation time decreases even further. Evaluations were performed for 9 randomly selected proteins, where n maps from a query protein were compared versus all maps from a subject protein. Error bars represent the standard deviations across 9 analyzed proteins. D) The HECTOR algorithm allows to identify the correct pair of subject maps corresponding to the pair of query maps at the protein‐protein interfaces with high shape complementarity. The violin plot shows the distribution of the correct maps' rankings within the top 100 HECTOR hits.

Figure 2

Design and biophysical characterization of the VEGF binders. A) VEGFR subunit is composed of 7 extracellular domains, a transmembrane segment, and an intracellular kinase domain. VEGFR is activated when dimeric VEGF binds, and in turn, dimerizes two receptor subunits. This dimeric receptor configuration triggers the intracellular domain kinase activity. VEGFR signaling can be inhibited via: 1) protein‐based quenching of the ligand binding site on the receptor surface, 2) protein‐based sequestration of VEGF itself at its receptor binding site, or 3) inhibition of the kinase activity through the use of small molecules. B) Anti‐VEGF binders were designed based on two different scaffolds, Sam (blue) and Sima (coral), that showed a high shape complementarity to the receptor‐binding site of VEGF (teal). Amino acids mutated during binder pocket design are colored in yellow. C) Thermal unfolding curves show melting temperatures (T_m ) of both Sam0.7 and Sima3.2 designs to be higher than 60 °C. D) The designed proteins bind VEGF with nanomolar affinity as measured by SPR. E) The crystal structure of Sam0.7 (yellow) matches the computational design model (blue) with atomic‐level accuracy. The VEGF epitope, in its modeled orientation relative to Sam0.7, is shown as a teal surface. The scatterplot displays RMSD between coordinates of the C_α atoms in the Sam0.7 design model and corresponding C_α atoms in the crystal structure. Residues in gaps are assigned an RMSD value of 0. Regions in the design model with RMSD values higher than 3 Å are highlighted in red.

Figure 3

The designs exhibit anti‐VEGF activity in vitro and in vivo. A) VEGF‐dependent survival of HUVEC primary cells was significantly reduced in a dose‐dependent manner by the designed binders. Yellow bars indicate the survival of the cells in an endothelial cell growth basal medium without VEGF, whereas green bars correspond to the survival of the cells in a basal medium with 30 nM of VEGF. Blue and red bars show results on cell survival in a basal medium with 30 nm of VEGF and increasing concentrations of Sam0.7 and Sima3.2, respectively. B) Treatment of U937 acute myeloid leukemia cell line with Sam0.7 and Sima3.2 proteins at low micromolar concentrations inhibited the cell growth. In contrast, the unmutated scaffold Sima_cntrl showed much weaker inhibitory activity, while Sam_cntrl did not inhibit proliferation at all. Error bars represent the standard deviations across nine replicates from three experiments. Statistical significance was calculated using Fisher's one‐sided t‐test (^* p ≤ 0.05, ^** p ≤ 0.01, ^*** p ≤ 0.001, ^**** p ≤ 0.0001 vs. the PBS group). C) Quantification of the engrafted LNZ308‐GFP glioma cells in zebrafish embryos that were injected with PBS, inactive protein (mvn_cntrl) as a negative control, bevacizumab as a positive control, Sam 0.7, or Sima 3.2. Each dot indicates one embryo. p‐value was calculated by the Mann‐Whitney two‐tailed test. d – Cohen's d value. D) Representative zebrafish xenografts treated with PBS and Sam0.7. Arrowheads indicate transplanted LNZ308‐GFP cells in the brain. The scale bar is 200 µm. E) Schematic representation of in vitro microvasculature formation and analysis: iPSC‐derived endothelial cells (ECs) and pericytes (PCs) were co‐cultured in the fibrin gel with Sam0.7 or Sam_cntrl as negative control. The formed microvasculature was imaged at day 7, and images were analyzed to calculate microvasculature parameters. F) Quantitative analysis of microvasculature formation. Plots show the percentage of the area covered with blood vessels, and the blood vessel diameter. Statistical significance was calculated using the one‐way ANOVA test (^** p ≤ 0.01, ^*** p ≤ 0.001, ^**** p ≤ 0.0001 treated Sam0.7 vs. Sam_cntrl group). G) Representative images showing in vitro microvasculature formation in the presence of Sam0.7 or Sam_cntrl at two working concentrations (50 and 100 µg mL⁻¹). The scale bar is 500 µm.

Figure 4

Design and experimental validation of the IL‐7Rα binders. A) IL‐7Rα binds IL‐7, leading to the recruitment of the γ_c receptor. The interaction between these three components is crucial for signaling cascades involved in the development and homeostasis of T and B cells. The designed binders are intended to target two distinct sites on the surface of IL‐7Rα. B) IL‐7Rα binders were designed using three different scaffolds that showed high shape complementarity to either target site 1 or site 2. The designs are shown in green, IL‐7Rα in orange, and the query patches in blue. C) SPR sensograms show the designed proteins to bind IL‐7Rα with nanomolar affinities. D) des03, des06, and des07 inhibited IL‐7 signaling in HEK‐Blue^TM IL‐7 reporter cells, similarly to lusvertikimab, an anti‐IL‐7Rα antibody. Statistical significance was calculated using Fisher's one‐sided t‐test (^* p ≤ 0.05, ^** p ≤ 0.01, ^*** p ≤ 0.001, ^**** p ≤ 0.0001 vs. the group treated with IL‐7 only).