A computationally designed inhibitor of an Epstein-Barr viral Bcl-2 protein induces apoptosis in infected cells

Abstract

Because apoptosis of infected cells can limit virus production and spread, some viruses have co-opted prosurvival genes from the host. This includes the Epstein-Barr virus (EBV) gene BHRF1, a homolog of human Bcl-2 proteins that block apoptosis and are associated with cancer. Computational design and experimental optimization were used to generate a novel protein called BINDI that binds BHRF1 with picomolar affinity. BINDI recognizes the hydrophobic cleft of BHRF1 in a manner similar to other Bcl-2 protein interactions but makes many additional contacts to achieve exceptional affinity and specificity. BINDI induces apoptosis in EBV-infected cancer lines, and when delivered with an antibody-targeted intracellular delivery carrier, BINDI suppressed tumor growth and extended survival in a xenograft disease model of EBV-positive human lymphoma. High-specificity-designed proteins that selectively kill target cells may provide an advantage over the toxic compounds used in current generation antibody-drug conjugates.

Figure 1. Computationally-designed proteins incorporating elements of the Bim-BH3 motif bind BHRF1

(A) Computational models of individually-tested designed proteins (orange) that bind BHRF1 (green). The crystal structure of BHRF1 bound to Bim-BH3 (blue) is shown at left for comparison. BbpG designs are from side chain grafting, and BbpD designs are de novo assembled proteins. Apparent affinities (mean ± SE, n = 3–6) are from yeast display titrations. See Figure S1, and Tables S1 and S2. An archive file of designed structures is provided in Supplemental Information online. (B) Seventy-four computationally designed proteins without human modifications (Indexes-01 to 74) were included in a yeast display library. BbpD04 (Index-00) was included as a positive control. The library was sorted for cells expressing surface protein (lane 1), for the 2% of cells with highest expression (lane 2), and for cells showing binding signal after incubation with 100 nM (lane 3) or 400 nM BHRF1 (lane 4). The gene frequencies in the sorted population were divided by their frequencies in the naive library to calculate a log₂ enrichment ratio, plotted from −4 (i.e. depleted, red) to +4 (i.e. enriched, green). See Table S3. (C) Histogram of the mean RMSD between the ten lowest energy structures found in ab initio structure prediction calculations and the intended designed structure for each of the sets of designs in (B). Designs with computed energy minima near the designed target conformation have a higher probability of binding BHRF1. See Figure S2.

Figure 2. Affinity maturation of BbpD04

(A) Computational model of BHRF1 (light green ribbon) bound to design BbpD04 (transparent surface over a grey ribbon). The electrostatic potential from BHRF1 is shown on the BbpD04 surface (red, −2 kT/e to blue, +2 kT/e; calculated using the Adaptive Poisson-Boltzmann Solver built in ROSETTA). Mutations E48R and E65R (insets) decrease electrostatic repulsion in regions where the field from BHRF1 is negative. (B) Same as in (A), with Mcl-1 in place of BHRF1. E48 and E65 are now in regions that border positive potential from Mcl-1. The Mcl-1•BbpD04 model was generated by superposition of the BbpD04 binding site to Bim-BH3 bound to Mcl-1 in crystal structure 2PQK, followed by rotamer repacking and side chain/backbone minimization in ROSETTA. See Table S4. (C) The four additional mutations in BbpD04.2 (dark blue sticks) are shown on the computational model of BbpD04.1 (orange) bound to BHRF1 (green). Three mutations are distant from the interface. See Figure S3 and S4. (D) Sequence-fitness landscape. All single amino acid substitutions of BbpD04.3 were expressed in a yeast display library. The 1% of cells with highest binding signal for 400 pM biotinylated BHRF1 relative to surface expression were collected by FACS. Plotted for each substitution is the log₂ enrichment ratio from −3.5 (depleted, orange) to +3.5 (enriched, blue). Stop codons, *. The region of the incorporated Bim-BH3 motif is boxed with a broken line. Secondary structure and core residues (shaded grey) are indicated above. Substitutions to aspartate (tend to be depleted for core residues) and to proline (depleted for helical residues) are boxed. (E) As in (D), except the library was sorted for binding to 400 pM biotinylated BHRF1 in the presence of 8 nM competitor Bcl-2 proteins for specificity. Substitutions of N62 are boxed. (F) The modeled structure of BbpD04.3 is colored by sequence Shannon entropy from 2.8 (highly conserved, dark blue) to 4.3 (variable, red), based on the sequence-fitness landscapes. A broken line boxes the incorporated Bim-BH3 motif. (G) BbpD04.3 and its derivative BINDI were expressed as 6his-tagged proteins in E. coli, precipitated from cleared lysate with NiNTA-agarose and analyzed on a Coomassie-stained SDS-polyacrylamide electrophoretic gel. An arrowhead indicates the expected MW at 15 kD. See Figure S5. (H) The fraction of protein folded in the presence of guanidinium hydrochloride, based on the change in CD signal at 222 nm. (I) Summary of all mutations made to BbpD04 during affinity maturation.

Figure 3. BINDI binds BHRF1 with high affinity and specificity

(A) BINDI or knockout mutant BINDI L54E were mixed with BHRF1 and separated by SEC. (B) Biotinylated BHRF1 was immobilized to a BLI sensor and the interaction with BINDI was measured at the indicated concentrations. (C–E) BLI kinetic analysis of interactions between immobilized Bcl-2 proteins and soluble Bim-BH3 fused to the C-terminus of maltose-binding protein (C), BINDI (D) and BINDI N62S (E). Red labels on broken diagonal lines indicate the corresponding affinities / K_D. Plotted are means ± SD from 4–6 experiments.

Figure 4. The crystal structure of BINDI-bound BHRF1

(A) Electron density at 2.0 σ (blue mesh) in the region of the Bim-BH3 incorporation site. The crystallographic model of BINDI (orange) bound to BHRF1 (green) is superimposed on the original starting computational model of BbpD04•BHRF1 (grey). Residues of BINDI and BHRF1 are labeled black and green, respectively, and residues of BbpD04 prior to affinity maturation that differ are indicated in parentheses. See Figure S6. (B) The crystal structure of BINDI bound to BHRF1 superimposed on the original BbpD04•BHRF1 model. (C) Agreement between the crystal structure and computational model is represented from low (thin blue tubing) to higher Cα-Cα RMSD (thick red tubing). (D) Slice through the crystal structure of BINDI (orange ribbon) bound to BHRF1 (green ribbon with grey surface). The guiding scaffold 3LHP(S) (dark blue) is aligned to BINDI at the Bim-BH3 incorporation site. A direct graft of the BH3 motif into 3LHP(S) at this position causes clashes elsewhere with the BHRF1 surface. (E) Aligned ribbon traces of BINDI from the BHRF1•BINDI crystal structure in orange; the computational model of BINDI in grey; and the crystal structure of the guiding scaffold 3LHP(S) in dark blue. The three structures are aligned at the Bim-BH3 incorporation site on the middle helix in the back. The upper magnified insets show the agreement between core residue conformations in the BINDI computational model versus the crystal structure. The lower magnified insets show that the core residues of BINDI are distinct in both identity and position from those in the guiding scaffold. When aligned at the Bim-BH3 incorporation site, the crystal structure is closer to the design model than the guiding scaffold. (F) Crystal structure of BINDI (orange) bound to BHRF1 (green). (G) The surface of BHRF1, in the same orientation as in (F), with the buried contact surface in BHRF1•BINDI colored blue. (H) The surface of BINDI, rotated 180º compared to the orientation in (F), with the buried contact surface in BHRF1•BINDI colored. Buried residues whose identities were taken from the incorporated Bim-BH3 fragment are magenta. Buried residues that were designed are blue. (I) The crystal structure (PDB 2WH6) of Bim-BH3 (blue) bound to BHRF1 (green). Buried surface areas are calculated based on the 22 ordered residues of Bim built into the electron density (a 26-residue Bim-BH3 peptide was crystallized with BHRF1 (Kvansakul et al., 2010)). (J) The surface of BHRF1, in the same orientation as in (I), with the buried contact surface in BHRF1•Bim-BH3 colored blue. (K) The surface of Bim-BH3, rotated 180º compared to the orientation in (I), with the buried contact surface in BHRF1•Bim-BH3 colored blue.

Figure 5. BINDI triggers apoptosis in EBV-positive B cell lines

(A) Cytochrome c release from mitochondria harvested from Ramos (EBV-negative, grey) or Ramos-AW cells (EBV-positive, crimson) treated with Bim-BH3 peptide (broken line) or BINDI (solid line). Mean ± SD, n = 4, for panels A–C. See Figure S7. (B) Cytochrome c release from Ramos and Ramos-AW mitochondria treated with 10 μM BINDI or BINDI N62S. (C) Mitochondria were harvested from EBV-negative and positive lines and cytochrome c release was measured after treatment with 10 μM BINDI N62S. (D) Cells were incubated with sub-lethal doses (2 μM) of antennapedia peptide-fused BINDI or 3LHP(S). Diblock copolymer Pol300 was conjugated to the proteins for enhanced endosomal escape. Cell viability (mean ± SD, n = 3) was measured after 24 hours.

Figure 6. Treatment of EBV-positive B cell lymphoma xenograft tumors by intracellular delivery of BINDI in vivo

(A) Schematic representation of the copolymer-based treatment. Pol950 has stabilizing (green) and endosomolytic (red) blocks and forms a micelle at physiological pH. The stabilizing block couples to αCD19 and BINDI. Nude mice with subcutaneous Ramos-AW xenografts were treated on days 0, 3 and 6 with Pol950 (300 mg/kg) : αCD19 (15 mg/kg) : BINDI or 3LHP(S) (105 mg/kg). A maximum tolerated dose study determined this level of BINDI was nontoxic. Mice were injected 30 minutes prior to each treatment with CTX (35 mg/ml) and BTZ (0.5 mg/ml). (B–E) Tumor growth is plotted for each individual mouse until day 11 when the first mice are euthanized. (B) PBS control treatment, black, n = 8; (C) chemo-only, grey, n = 9; (D) 3LHP(S)-copolymer treatment, green, n = 9; (E) BINDI-copolymer treatment, orange, n = 10. (F) Kaplan-Meier survival plot. There is a significant increase in survival with treatment (log-rank test χ² = 46, P < 0.0001). See Figure S7E.