Viruses traverse the human proteome through peptide interfaces that can be biomimetically leveraged for drug discovery

Abstract

We present a drug design strategy based on structural knowledge of protein-protein interfaces selected through virus-host coevolution and translated into highly potential small molecules. This approach is grounded on Vinland, the most comprehensive atlas of virus-human protein-protein interactions with annotation of interacting domains. From this inspiration, we identified small viral protein domains responsible for interaction with human proteins. These peptides form a library of new chemical entities used to screen for replication modulators of several pathogens. As a proof of concept, a peptide from a KSHV protein, identified as an inhibitor of influenza virus replication, was translated into a small molecule series with low nanomolar antiviral activity. By targeting the NEET proteins, these molecules turn out to be of therapeutic interest in a nonalcoholic steatohepatitis mouse model with kidney lesions. This study provides a biomimetic framework to design original chemistries targeting cellular proteins, with indications going far beyond infectious diseases.

Keywords: biomimetism; chemoinformatics; interactomics; kidney; viruses.

Fig. 1.

Principle of the drug discovery process inspired by viruses. (A) Viruses and human as networks of interacting proteins (Left). Following infection, a viral protein modulates a cellular function by interacting with a human protein (Middle). Interactions can be mapped to the level of small interfaces of 5 to 20 amino acids (Right). These peptides can be used as new chemical entities or translated into small molecules. Overall, viruses highlight manipulable human proteins and provide starting points to modulate their functions. (B) Sequential steps of the drug discovery process. The viral peptide library is screened in a cell-based assay. The pharmacophores of a peptide hit are translated into a library of small molecules by computer-assisted drug design. After the screening of the library, each hit becomes a matrix for a structure–activity relationship (SAR) process for the selection of a lead. The target identification is then a prerequisite to select the therapeutic indication.

Fig. 2.

Description of the viral infection landscape in the Vinland database. (A) Key figures of the curated virus–human protein interactome integrated into Vinland. (B) Representation of interactomes between human and viruses of clinical importance. Blue dots: Human proteins. Red dots: Viral proteins. Gray edges: Physical protein interaction. (C) Venn diagram comparing the papers reporting at least one interaction between a human protein and a viral protein in Vinland, IntAct (12), VirHostNet (13), and HVIDB (14). (D) Topological analysis. Left: Degree distributions of human proteins (black) and human proteins targeted by viruses (red); P(k) is the probability of a node to connect to k other nodes in the network. Degree: Wilcoxon test, P = 3.2 × 10⁻¹⁹¹. Right: Betweenness distributions of human proteins (black) and human proteins targeted by viruses (red); P(b) is the probability for a node to have a betweenness value of b in the network. Betweenness: Wilcoxon test, P = 1.3 × 10⁻¹⁷⁶. Solid lines are linear regression fits. Vertical dashed lines indicate the mean degree/betweenness values for each distribution. (E) Distribution of the length of interacting sequences on viral proteins.

Fig. 3.

Viral peptide library screens. (A) Anti-influenza virus activity of CPEP31 peptide. Neuraminidase (NA) activity measured in the supernatant of A549 cells infected with influenza A H1N1 virus and treated with CPEP31 and derivative peptides. Cell viability is measured by the resazurin assay in infected and noninfected cells. Values are normalized to those of cells treated by vehicle (N = 3). CPEP31-mut3Ala is CPEP31 where LFL amino acids are mutated in AAA. CPEP31-EVOL1 and CPEP31-EVOL2 are CPEP31-similar peptides in vFLIP proteins from HVS and BoHV-4 viruses (YCLLFLINGC and FVMYFLLDPY respectively, in fusion with the cell-penetrating sequence). (B) Relative peptide activities in EBV screen. Late gene reporter transcriptional activation is quantified 48 h post-TPA and BA treatment in Hone-1 cells treated with 20 µM of peptide. (C) Relative peptide activities in HIV1 screen. Replication is monitored by measuring β-galactosidase activity in HeLa P4 cells carrying a Tat-inducible LacZ gene and treated with 20 µM of peptide. (D) Relative peptide activities in HCV screen. The HCV replicon RNA copy number in cells is evaluated by measuring the firefly luciferase activity in Huh7 and treated with 20 µM peptide. (E) Relative peptide activities in Francisella novicida screen. The AUC is calculated from PI incorporation signal in function of time in F. novicida infected BMDMs treated with 20 µM of peptide. (B–E) Data are normalized to signal with vehicle only. Arrows indicate peptides for which targets are commented on in SI Appendix.

Fig. 4.

Ligand-based drug design of a library of small molecules from a bioactive peptide. (A) 3D structure of the peptide CPEP31 (displayed as balls and sticks) as a part of the protein vFLIP, (PDB entry 3CL3) represented as cartoon ribbon. (B) Energy-minimized conformation of peptide CPEP31, its shape is displayed in gray and major pharmacophoric points as colored spheres (yellow: hydrophobic, green: ring, solid red: negative ionizable, mesh red: H-bond acceptor, and mesh blue: H-Bond donor). (C) Overlay of five different rocs (32) queries with their shape highlighted in color (yellow, red, blue, green, and gray) and pharmacophoric points as colored spheres (same color code as in B). Each individual query covers only a part of the external surface of the peptide. Taken all together, they reproduce the overall shape of the full peptide. Six other queries are not represented in this picture for the sake of clarity. (D) Overlay of two rocs queries (shapes in gray and yellow) with 3D structures of an identified molecular hit shown as gray sticks.

Fig. 5.

SAR, ADME, and interaction with NEET proteins during optimization. (A) Chemical structures of molecules inspired by CPEP31 from dEF384 to dEF3122 including main intermediates with key chemical modifications. (B) Antiviral activity and ADME (absorption, distribution, metabolism, and excretion) data of main compounds: IC50 from the neuraminidase activity assay in the supernatant of A549 cells infected by H1N1 influenza A virus (nM), kinetic solubility at pH 7.4 in µM, CL_int (intrinsic clearance) in hLM/mLM (human/murine liver microsomes) in µL/min/mg protein, CL_int (intrinsic clearance) in hHep/mHep (human/murine hepatocytes) in µL/min/million cells, plasma protein binding (PPB) (% bound) murine, mouse T_1/2 (h-hours, iv-intravenous, po-per os), AUC 0-∞ (area under the curve) in h*µg/mL, C_max (maximum concentration) po in ng/mL and po bioavailability in %, (ND-not determined). (C) Tissue distribution of dEF3122 in mouse and rat. (D and E) [2Fe–2S] cluster release from the three recombinant NEET proteins at 37 °C was determined by following the absorbance of the cluster at 460 nm as a function of time for dEF3122 (D) and CPEP31 (E).

Fig. 6.

Effect of the lead compound in a diet-induced obese model of NASH. Mice fed with a high fat, high fructose, and cholesterol supplemented diet for 38 wk were then treated in the last 8 wk with vehicle, dEF3122, or OCA (obeticholic acid) at indicated concentrations (n = 12 animals per group). (A and B) Histopathological mean score changes between the beginning and end of the treatment for liver portal inflammation (A) and liver portal fibrosis (B) are shown with representative stained liver sections of hematoxylin and eosin (H&E) and picrosirius red stains, respectively. (C–E) (C) Areas of kidney fibrosis at the end of the treatment are shown as % of the total area of the selected fields with representative sections of the kidney stained with picrosirius red. (D) % CD3 and (E) % F4/80 positive areas in kidneys with representative sections of CD3 and F4/80 immunostainings. *P < 0.05, **P < 0.01, and ***P < 0.001 from the T test with a preliminary nonsignificant Shapiro–Wilk test or from Wilcoxon signed–rank test otherwise.