Protein constructs

The constructs coding for Pangolin-CoV S ectodomains were based on coronavirus sequences reported by two independent groups, both of which isolated virus material from diseased Malayan pangolins (Manis javanica) likely smuggled into China’s Guangdong province in 2019. Pangolin-CoV S’ corresponded to residues 1–1200 (the equivalent of 1–1208 for SARS-CoV-2) of the S identified in the Pangolin-CoV genome (GISAID number EPI_ISL_410721) reported by Xiao et al.⁸ and Pangolin-CoV S corresponded to residues 1–1200 (also the 1–1208 equivalent in SARS-CoV-2) of the S (NCBI number QIG55945.1) from the Pangolin coronavirus MP789 isolate reported by Liu et al.⁹. Both constructs were made as “2 P” mutants for greater stability¹⁹, codon optimised for human expression and cloned by GenScript with the same expression and purification tags as described previously for RaTG13 and SARS-CoV-2 S⁶ viz. the N-terminal secretion sequence derived from μ-phosphatase and a C-terminal tag consisting of a TEV-cleavage site, the foldon trimerisation domain, and a hexahistidine. The RaTG13 and SARS-CoV-2 S (with its furin-cleavage site intact) constructs used here had the same overall architecture and were described previously⁶.

The construct coding for the human ACE2 ectodomain (residues 1–615, NCBI reference NM_021804.2) was codon optimised and made with a C-terminal Twin-strep tag preceded by a DYK-tag and cloned into pcDNA.3.1(+) by GenScript. The ACE2 ectodomains (residues 19–615) from the Malayan pangolin (Manis javanica, NCBI reference XP_017505746.1) and an archetypal horseshoe bat species, Greater horseshoe bat (Rhinolophus ferremequinum, Uniprot reference B6ZGN7) were also cloned by Genscript into pcDNA.3.1(+) with the same tags as described before for the human ACE2⁶ viz. DYK plus Twin-strep tag at the C-terminus and the secretion leader sequence derived from Ig-kappa at the N-terminus.

Protein expression and purification

The RaTG13 S, SARS-CoV-2 S, two Pangolin-CoV Spikes (S and S’) and ACE2 ectodomains were made as described before for the SARS-CoV-2 S and human ACE2⁶. Briefly, the proteins were expressed in in Expi293F cells (Gibco) grown in suspension in 37 °C humidified atmosphere with 8% CO₂. Cells were transfected with 1 mg of DNA per 1 L of cell culture and the protein expressed for 4 (in case of RaTG13 S) or 5 (for ACE2 ectodomains) days. The only difference with the method previously described was that, in case of the SARS-CoV-2 S and Pangolin-CoV S and S’, the cells were transferred to a 32 °C incubator 24 h after the transfection and harvested on the fifth day post transfection for increased yield²⁰.

Pangolin-CoV S and S’ were purified using affinity chromatography with TALON beads (Takara), followed by gel filtration into 50 mM MES pH 6.0, 100 mM NaCl buffer on a Superdex 200 Increase 10/300 GL column (GE Life Sciences). SARS-CoV-2 and RaTG13 spikes were made as described previously⁶. SARS-CoV-2 S was not treated with furin in vitro. All three ACE2 ectodomains were purified using Streptactin XT resin (iba) and gel filtered into a buffer containing 20 mM Tris pH 8.0 and 150 mM NaCl as described previously for the human ACE2 ectodomain⁶.

Biolayer interferometry assays

The biolayer interferometry assays were done as before⁶ using Octet Red 96 (ForteBio) and NiNTA (NTA, ForteBio) sensors in 20 mM Tris pH 8.0, 150 mM NaCl buffer at 25 °C. Spike proteins were immobilised at 20–70 µg/mL concentrations for 45–60 min and ACE2-binding measured using a 120–600 s association and 300–900 s dissociation stages.

Equilibrium dissociation constants (K_d) were determined from reaction amplitudes by analysis of the variation of maximum response with ACE2 concentration. K_d values were also determined using analysis of the kinetics of the reactions. Association phases were analysed as a single exponential function, and plots of the observed rate (k_obs) versus ACE2 concentration gave the association and dissociation rate constants (k_on and k_off) as the slope and intercept, respectively. The k_off values determined in this way were confirmed by analysis of the dissociation phase and K_d values were determined as k_off/k_on.

Cryo-EM sample preparation and data collection

Pangolin-CoV S at ~0.15 mg/mL concentration was applied on an R2/2 Quantifoil grid of 200 mesh covered with a thin layer of continuous carbon. The grid was glow discharged for 30 s at 45 mA prior to freezing; 4 uL of the sample was then applied to the grid before it was blotted for between 4 and 4.5 s and plunge frozen into liquid ethane using a Vitrobot MkIII. Data were collected using EPU software on a Titan Krios operating at 300 kV (Thermo Scientific), using a Gatan K2 detector mounted on a Gatan GIF Quantum energy filter operating in zero-loss mode with a slit width of 20 eV. Exposures were of 8 s with an accumulated dose of 51.8 e/Ų, which was fractionated into 32 frames. The calibrated pixel size was 1.08 Å and data were collected using a range of defoci between 1.5 and 3 µm.

Cryo-EM data processing

The frames of collected movies were aligned using MotionCor2²¹, implemented in RELION²², with Contrast Transfer Function fitted using CTFfind4²³. Particles were picked using RELION autopicking, and subjected to 2 rounds of RELION 2D classification, retaining classes with clear secondary structure features. An ab initio 3D model was generated using cryoSPARC²⁴ and used as a reference for RELION 3D classification. The particles contained in classes with clear secondary structure were subjected to Bayesian polishing²⁵ and refined using cryoSPARC homogeneous refinement, imposing C3 symmetry, with CTF refinement. This generated a map with a global resolution of 2.9 Å. The map had local resolution estimated using blocres²⁶ implemented in cryoSPARC, followed by local resolution filtering and global sharpening²⁷ in cryoSPARC.

Model building

The sequence of the Pangolin-CoV S was numbered as for SARS-CoV-2 S (NCBI YP_009724390.1) for the sake of simplicity of comparison. The model was built using our previous structure of SARS-CoV-2 spike (PDB 6ZGE)⁶ as a starting model, with adjustment of the sequence and manual fitting of the model carried out using Coot²⁸. Real-space refinement and model validation was carried out using PHENIX²⁹.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.