Structural basis for the recognition of SARS-CoV-2 by full-length human ACE2

Structural determination of the ACE2-B⁰AT1 complex

Full-length human ACE2 and B⁰AT1, with Strep and FLAG tags on their respective N termini, were coexpressed in human embryonic kidney (HEK) 293F cells and purified through tandem affinity resin and size exclusion chromatography. The complex was eluted in a single monodisperse peak, indicating high homogeneity (Fig. 1A). Details of cryo-sample preparation, data acquisition, and structural determination are given in the materials and methods section of the supplementary materials. A three-dimensional (3D) reconstruction was obtained at an overall resolution of 2.9 Å from 418,140 selected particles. This immediately revealed the dimer of heterodimers’ architecture (Fig. 1B). After applying focused refinement and C2 symmetry expansion, the resolution of the extracellular domains improved to 2.7 Å, whereas the TM domain remained at 2.9-Å resolution (Fig. 1B, figs. S1 to S3, and table S1).

The high resolution supported reliable model building. For ACE2, side chains could be assigned to residues 19 to 768, which contain the PD (residues 19 to 615) and the CLD (residues 616 to 768), which consists of a small extracellular domain, a long linker, and the single TM helix (Fig. 1C). Between the PD and TM helix is a ferredoxin-like fold domain; we refer to this as the neck domain (residues 616 to 726) (Fig. 1C and fig. S4). Homodimerization is entirely mediated by ACE2, which is sandwiched by B⁰AT1. Both the PD and neck domains contribute to dimerization, whereas each B⁰AT1 interacts with the neck and TM helix in the adjacent ACE2 (Fig. 1C). The extracellular region is highly glycosylated, with seven and five glycosylation sites on each ACE2 and B⁰AT1 monomer, respectively.

During classification, another subset with 143,857 particles was processed to an overall resolution of 4.5 Å. Whereas the neck domain still dimerizes, the PDs are separated from each other in this reconstruction (Fig. 1D and fig. S1, H to K). We therefore define the two classes as the open and closed conformations. Structural comparison shows that the conformational changes are achieved through rotation of the PD domains, with the rest of the complex left nearly unchanged (movie S1).

Homodimer interface of ACE2

Dimerization of ACE2 is mainly mediated by the neck domain, with the PD contributing a minor interface (Fig. 2A). The two ACE2 protomers are hereafter referred to as A and B, with residues in protomer B followed by a prime symbol. Extensive polar interactions are mapped to the interface between the second (residues 636 to 658) and fourth (residues 708 to 717) helices of the neck domain (Fig. 2B). Arg⁶⁵² and Arg⁷¹⁰ in ACE2-A form cation-π interactions with Tyr^641′ and Tyr^633′ in ACE2-B. Meanwhile, Arg⁶⁵² and Arg⁷¹⁰ are respectively hydrogen-bonded (H-bonded) to Asn^638′ and Glu^639′, which also interact with Gln⁶⁵³, as does Asn^636′. Ser⁷⁰⁹ and Asp⁷¹³ from ACE2-A are H-bonded to Arg^716′. This extensive network of polar interactions indicates stable dimer formation.

The PD dimer interface appears much weaker, with only one pair of interactions between Gln¹³⁹ and Gln^175′ (Fig. 2C). Gln¹³⁹ is in a loop that is stabilized by a disulfide bond between Cys¹³³ and Cys¹⁴¹ as well as multiple intraloop polar interactions (Fig. 2C). The weak interaction is consistent with the ability to transition to the open conformation, in which the interface between the neck domains remains the same while the PDs are separated from each other by ~25 Å (Fig. 2D and movie S1).

Overall structure of the RBD-ACE2-B⁰AT1 complex

To gain insight into the interaction between ACE2 and SARS-CoV-2, we purchased 0.2 mg of recombinantly expressed and purified RBD-mFc of SARS-CoV-2 (for simplicity, hereafter referred to as RBD; mFc, mouse Fc tag) from Sino Biological Inc., mixed it with our purified ACE2-B⁰AT1 complex at a stoichiometric ratio of ~1.1 to 1, and proceeded with cryo-grid preparation and imaging. Finally, a 3D EM reconstruction of the ternary complex was obtained.

In contrast to the ACE2-B⁰AT1 complex—which has two conformations, open and closed—only the closed state of ACE2 was observed in the dataset for the RBD-ACE2-B⁰AT1 ternary complex. The structure of the ternary complex was determined to an overall resolution of 2.9 Å from 527,017 selected particles. However, the resolution for the ACE2-B⁰AT1 complex was substantially higher than that for the RBDs, which are at the periphery of the complex (Fig. 3A). To improve the local resolution, focused refinement was applied; this allowed us to reach a resolution of 3.5 Å for the RBD, supporting reliable modeling and analysis of the interface (Fig. 3, figs. S5 to S7, and table S1).

Interface between the RBD and ACE2

As expected, each PD accommodates one RBD (Fig. 3B). The overall interface is similar to that between SARS-CoV and ACE2 (7, 8), mediated mainly through polar interactions (Fig. 4A). An extended loop region of the RBD spans the arch-shaped α1 helix of the ACE2-PD like a bridge. The α2 helix and a loop that connects the β3 and β4 antiparallel strands, referred to as loop 3-4, of the PD also make limited contributions to the coordination of the RBD.

The contact can be divided into three clusters. The two ends of the bridge interact with the N and C termini of the α1 helix as well as small areas on the α2 helix and loop 3-4. The middle segment of α1 reinforces the interaction by engaging two polar residues (Fig. 4A). At the N terminus of α1, Gln⁴⁹⁸, Thr⁵⁰⁰, and Asn⁵⁰¹ of the RBD form a network of H-bonds with Tyr⁴¹, Gln⁴², Lys³⁵³, and Arg³⁵⁷ from ACE2 (Fig. 4B). In the middle of the bridge, Lys⁴¹⁷ and Tyr⁴⁵³ of the RBD interact with Asp³⁰ and His³⁴ of ACE2, respectively (Fig. 4C). At the C terminus of α1, Gln⁴⁷⁴ of the RBD is H-bonded to Gln²⁴ of ACE2, whereas Phe⁴⁸⁶ of the RBD interacts with Met⁸² of ACE2 through van der Waals forces (Fig. 4D).

Comparing the SARS-CoV-2 and SARS-CoV interfaces with ACE2

Superimposition of the RBD in the complex of SARS-CoV (SARS-CoV-RBD) and ACE2-PD [Protein Data Bank (PDB) 2AJF] with the RBD in our ternary complex shows that the SARS-CoV-2 RBD (SARS-CoV-2-RBD) is similar to SARS-CoV-RBD with a root mean square deviation (RMSD) of 0.68 Å over 139 pairs of Cα atoms (Fig. 5A) (8). Despite the overall similarity, a number of sequence variations and conformational deviations are found in their respective interfaces with ACE2 (Fig. 5 and fig. S8). At the N terminus of α1, the variations Arg⁴²⁶→Asn⁴³⁹, Tyr⁴⁸⁴→Gln⁴⁹⁸, and Thr⁴⁸⁷→Asn⁵⁰¹ at equivalent positions are observed between SARS-CoV-RBD and SARS-CoV-2-RBD (Fig. 5B). More variations are observed in the middle of the bridge. The most prominent alteration is the substitution of Val⁴⁰⁴ in the SARS-CoV-RBD with Lys⁴¹⁷ in the SARS-CoV-2-RBD. In addition, from SARS-CoV-RBD to SARS-CoV-2-RBD, the substitution of interface residues Tyr⁴⁴²→Leu⁴⁵⁵, Leu⁴⁴³→Phe⁴⁵⁶, Phe⁴⁶⁰→Tyr⁴⁷³, and Asn⁴⁷⁹→Gln⁴⁹³ may also change the affinity for ACE2 (Fig. 5C). At the C terminus of α1, Leu⁴⁷² in the SARS-CoV-RBD is replaced by Phe⁴⁸⁶ in the SARS-CoV-2-RBD (Fig. 5D).

Discussion

Although ACE2 is a chaperone for B⁰AT1, our focus is on ACE2 in this study. With the stabilization by B⁰AT1, we elucidated the structure of full-length ACE2. B⁰AT1 is not involved in dimerization, suggesting that ACE2 may be a homodimer even in the absence of B⁰AT1. Further examination suggests that a dimeric ACE2 can accommodate two S protein trimers, each through a monomer of ACE2 (fig. S9). The trimeric structure of the S protein of SARS-CoV-2 was recently reported, with one RBD in an up conformation and two in down conformations (PDB 6VSB) (14). The PD clashes with the rest of the S protein when the ternary complex is aligned to the RBD of the down conformation. There is no clash when the complex is superimposed on RDB in the up conformation, with a RMSD of 0.98 Å over 126 pairs of C_α atoms, confirming that an up conformation of RDB is required to bind to the receptor (fig. S9) (14).

Cleavage of the S protein of SARS-CoV is facilitated by cathepsin L in endosomes, indicating a mechanism of receptor-mediated endocytosis (10). Further characterization is required to examine the interactions between ACE2 and the viral particle as well as the effect of cofactors on this process (25, 33). It remains to be investigated whether there is clustering between the dimeric ACE2 and trimeric S proteins, which may be important for invagination of the membrane and endocytosis of the viral particle, a process similar to other types of receptor-mediated endocytosis.

Cleavage of the C-terminal segment, especially residues 697 to 716 (fig. S4), of ACE2 by proteases, such as transmembrane protease serine 2 (TMPRSS2), enhances the S protein–driven viral entry (34, 35). Residues 697 to 716 form the third and fourth helices in the neck domain and map to the dimeric interface of ACE2. The presence of B⁰AT1 may block the access of TMPRSS2 to the cutting site on ACE2. The expression distribution of ACE2 is broader than that of B⁰AT1. In addition to kidneys and intestine, where B⁰AT1 is mainly expressed, ACE2 is also expressed in lungs and heart (27). It remains to be tested whether B⁰AT1 can suppress SARS-CoV-2 infection by blocking ACE2 cleavage. Enteric infections have been reported for SARS-CoV, and possibly also for SARS-CoV-2 (36, 37). B⁰AT1 has also been shown to interact with another coronavirus receptor, aminopeptidase N (APN or CD13) (38). These findings suggest that B⁰AT1 may play a regulatory role for the enteric infections of some coronaviruses.

Comparing the interaction interfaces of SARS-CoV-2-RBD and SARS-CoV-RBD with ACE2 reveals some variations that may strengthen the interactions between SARS-CoV-2-RBD and ACE2 and other variations that are likely to reduce the affinity compared with SARS-CoV-RBD and ACE2. For instance, the change from Val⁴⁰⁴ to Lys³¹⁷ may result in a tighter association because of the salt bridge formation between Lys³¹⁷ and Asp³⁰ of ACE2 (Figs. 4C and 5C). The change from Leu⁴⁷² to Phe⁴⁸⁶ may also result in a stronger van der Waals contact with Met⁸² (Fig. 5D). However, replacement of Arg⁴²⁶ with Asn⁴³⁹ appears to weaken the interaction by eliminating one important salt bridge with Asp³²⁹ on ACE2 (Fig. 5B).

Our structural work reveals the high-resolution structure of full-length ACE2 in a dimeric assembly. Docking the S protein trimer onto the structure of the ACE2 dimer with the RBD of the S protein bound suggests simultaneous binding of two S protein trimers to an ACE2 dimer. Structure-based rational design of binders with enhanced affinities to either ACE2 or the S protein of the coronaviruses may facilitate development of decoy ligands or neutralizing antibodies for suppression of viral infection.