Crystal structure of SARS-CoV-2 main protease provides a basis for design of improved α-ketoamide inhibitors

In December 2019, a new coronavirus caused an outbreak of pulmonary disease in the city of Wuhan, the capital of Hubei province in China, and has since spread globally (1, 2). The virus has been named severe acute respiratory syndrome–coronavirus 2 (SARS-CoV-2) (3) because the RNA genome is about 82% identical to that of the SARS coronavirus (SARS-CoV); both viruses belong to clade b of the genus Betacoronavirus (1, 2). The disease caused by SARS-CoV-2 is called coronavirus disease 2019 (COVID-19). Whereas at the beginning of the outbreak, cases were connected to the Huanan seafood and animal market in Wuhan, efficient human-to-human transmission led to exponential growth in the number of cases. On 11 March 2020, the World Health Organization (WHO) declared the outbreak a pandemic. As of 9 April, there were >1,500,000 cumulative cases globally, with a ~5.9% case fatality rate.

One of the best-characterized drug targets among coronaviruses is the main protease (M^pro, also called 3CL^pro) (4). Along with the papain-like protease(s), this enzyme is essential for processing the polyproteins that are translated from the viral RNA (5). The M^pro operates at no fewer than 11 cleavage sites on the large polyprotein 1ab (replicase 1ab, ~790 kDa); the recognition sequence at most sites is Leu-Gln↓(Ser, Ala, Gly) (↓ marks the cleavage site). Inhibiting the activity of this enzyme would block viral replication. Because no human proteases with a similar cleavage specificity are known, such inhibitors are unlikely to be toxic.

Previously, we designed and synthesized peptidomimetic α-ketoamides as broad-spectrum inhibitors of the main proteases of betacoronaviruses and alphacoronaviruses as well as the 3C proteases of enteroviruses (6). The best of these compounds (11r; Fig. 1) showed an half-maximal effective concentration (EC₅₀) of 400 pM against Middle East respiratory syndrome–coronavirus (MERS-CoV) in Huh7 cells as well as low-μM EC₅₀ values against SARS-CoV and a whole range of enteroviruses in various cell lines, although the antiviral activity seemed to depend to a great extent on the cell type used in the experiments (6). To improve the half-life of the compound in plasma, we modified 11r by hiding the P3-P2 amide bond within a pyridone ring (Fig. 1, green ovals) in the expectation that this might prevent cellular proteases from accessing this bond and cleaving it. Further, to increase the solubility of the compound in plasma and to reduce its binding to plasma proteins, we replaced the hydrophobic cinnamoyl moiety by the somewhat less hydrophobic Boc group (Fig. 1, red ovals) to give 13a (see scheme S1 for synthesis).

To examine whether the introduced pyridone ring is compatible with the three-dimensional structure of the target, we determined the crystal structure, at 1.75 Å resolution, of the M^pro of SARS-CoV-2 (Fig. 2). The three-dimensional structure is highly similar to that of the SARS-CoV M^pro, as expected from the 96% sequence identity (see fig. S8); the root mean square deviation between the two free-enzyme structures is 0.53 Å for all Cα positions [comparison between SARS-CoV-2 M^pro structure and SARS-CoV M^pro, PDB entry 2BX4 (7)]. The chymotrypsin-like and picornavirus 3C protease–like domains I and II (residues 10 to 99 and 100 to 182, respectively) are six-stranded antiparallel β barrels that harbor the substrate-binding site between them. Domain III (residues 198 to 303), a globular cluster of five helices, is involved in regulating the dimerization of the M^pro, mainly through a salt-bridge interaction between Glu²⁹⁰ of one protomer and Arg⁴ of the other (8). The tight dimer formed by SARS-CoV-2 M^pro has a contact interface of ~1394 Ų, predominantly between domain II of molecule A and the NH₂-terminal residues (“N-finger”) of molecule B, with the two molecules oriented perpendicular to one another (Fig. 2). Dimerization of the enzyme is necessary for catalytic activity, because the N-finger of each of the two protomers interacts with Glu¹⁶⁶ of the other protomer and thereby helps shape the S1 pocket of the substrate-binding site (9). To reach this interaction site, the N-finger is squeezed in between domains II and III of the parent monomer and domain II of the other monomer.

Interestingly, in the SARS-CoV but not in the SARS-CoV-2 M^pro dimer, there is a polar interaction between the two domains III involving a 2.60-Å hydrogen bond between the side-chain hydroxyl groups of residue Thr²⁸⁵ of each protomer, supported by a hydrophobic contact between the side chain of Ile²⁸⁶ and Thr²⁸⁵ Cγ2. In SARS-CoV-2, the threonine is replaced by alanine (indicated by the black spheres in Fig. 2) and the isoleucine by leucine (fig. S8). It was previously shown that replacing Ser²⁸⁴, Thr²⁸⁵, and Ile²⁸⁶ by alanine residues in SARS-CoV M^pro leads to enhancement of the catalytic activity of the protease by a factor of 3.6, concomitant with a slightly closer packing of the two domains III of the dimer against one another (10). This was accompanied by changes in enzyme dynamics that transmit the effect of the mutation to the catalytic center. Indeed, the Thr²⁸⁵ → Ala replacement observed in the SARS-CoV-2 M^pro also allows the two domains III to approach each other more closely (the distance between the Cα atoms of residues 285 in molecules A and B is 6.77 Å in SARS-CoV M^pro and 5.21 Å in SARS-CoV-2 M^pro, and the distance between the centers of mass of the two domains III shrinks from 33.4 Å to 32.1 Å). However, the catalytic efficiency of SARS-CoV-2 M^pro is only slightly higher, if at all [turnover number (k_cat)/Michaelis constant (K_m) = 3426.1 ± 416.9 s^–1 M^–1] than that of SARS-CoV M^pro (k_cat/K_m = 3011.3 ± 294.6 s^–1 M^–1). Further, the estimated dissociation constant of dimerization is the same (~2.5 μM) for the two enzymes, as determined by analytical ultracentrifugation (fig. S10).

We used this crystal structure to dock the α-ketoamide 13a; this suggested that the pyridone ring might have some steric clash with the side chain of Gln¹⁸⁹. However, in our previous work (6), we had found Gln¹⁸⁹ to be quite flexible, and therefore we went ahead with 13a as a lead. The plasma half-life of this compound in mice was increased by a factor of ~3 relative to 11r (from 0.3 hours to 1.0 hours), the in vitro kinetic plasma solubility was improved by a factor of ~19 (from 6 μM for 11r to 112 μM for 13a), and the thermodynamic solubility increased by a factor of ~13 (from 41 μM to 530 μM). Binding to mouse plasma protein was reduced from 99% to 97% [many drugs have plasma protein binding of >90% (11)]. However, relative to 11r (IC₅₀ = 0.18 ± 0.02 μM), the structural modification led to some loss of inhibitory activity against the main protease of SARS-CoV-2 (IC₅₀ = 2.39 ± 0.63 μM) as well as the 3C proteases (3C^pro) of enteroviruses. 11r was designed for broad-spectrum activity, with the P2 cyclohexyl moiety intended to fill a pocket in the enterovirus 3C^pro. The S2 pocket of the betacoronavirus M^pro (Fig. 3) features substantial plasticity, enabling it to adapt to the shape of smaller inhibitor moieties (6). To enhance the antiviral activity against betacoronaviruses of clade b (SARS-CoV-2 and SARS-CoV), we sacrificed the goal of broad-spectrum activity and replaced the P2 cyclohexyl moiety of 13a by the smaller cyclopropyl in 13b (Fig. 1, blue circles). Here, we present x-ray crystal structures in two different crystal forms, at 1.95 and 2.20 Å resolution, of the complex between α-ketoamide 13b and the M^pro of SARS-CoV-2. One structure is in space group C2 (Fig. 3), where both protomers of the M^pro dimer are bound by crystal symmetry to have identical conformations; the other is in space group P2₁2₁2₁, where the two protomers are independent of each other and free to adopt different conformations. Indeed, we find that in the latter crystal structure, the key residue Glu¹⁶⁶ adopts an inactive conformation in protomer B (as evidenced by its distance from His¹⁷² and the lack of H-bonding interaction with the P1 moiety of the inhibitor), even though compound 13b is bound in the same mode as in molecule A. This phenomenon has also been observed with the SARS-CoV M^pro (12) and is consistent with the half-site activity described for this enzyme (13). In all copies of the inhibited SARS-CoV-2 M^pro, the inhibitor binds to the shallow substrate-binding site at the surface of each protomer, between domains I and II (Fig. 3).

Through the nucleophilic attack of the catalytic Cys¹⁴⁵ onto the α-keto group of the inhibitor, a thiohemiketal is formed in a reversible reaction. This is clearly reflected in the electron density (Fig. 3, inset); the stereochemistry of this chiral moiety is S in all copies of compound 13b in these structures. The oxyanion (or hydroxyl) group of this thiohemiketal is stabilized by a hydrogen bond from His⁴¹, whereas the amide oxygen of 13b accepts a hydrogen bond from the main-chain amides of Gly¹⁴³, Cys¹⁴⁵, and partly Ser¹⁴⁴, which form the canonical “oxyanion hole” of the cysteine protease. It is an advantage of the α-ketoamides that their warhead can interact with the catalytic center of the target proteases through two hydrogen-bonding interactions (6) rather than only one, as with other warheads such as aldehydes (14) or Michael acceptors (15).

The P1 γ-lactam moiety, designed as a glutamine surrogate (15, 16), is deeply embedded in the S1 pocket of the protease, where the lactam nitrogen donates a three-center (bifurcated) hydrogen bond to the main-chain oxygen of Phe¹⁴⁰ (3.20/3.10/3.28 Å; values for the structure in space group C2/space group P2₁2₁2₁ molecule A/space group P2₁2₁2₁ molecule B) and to the Glu¹⁶⁶ carboxylate [3.35/3.33/(3.55) Å], and the carbonyl oxygen accepts a 2.57/2.51/2.81 Å hydrogen bond from the imidazole of His¹⁶³. The P2 cyclopropyl methyl moiety fits snugly into the S2 subsite, which has shrunk by 28 ų relative to the complex between compound 13a with P2 = cyclohexyl methyl and the SARS-CoV M^pro (17). The pyridone in the P3-P2 position of the inhibitor occupies the space normally filled by the substrate’s main chain; its carbonyl oxygen accepts a 2.89/2.99/3.00 Å hydrogen bond from the main-chain amide of residue Glu¹⁶⁶. Further, the P3 amide donates a 2.83/2.96/2.87 Å hydrogen bond to the main-chain oxygen of Glu¹⁶⁶. Embedded within the pyridone, the P2 nitrogen can no longer donate a hydrogen bond to the protein (the H-bond prevented from forming would connect the P2 nitrogen and the side-chain oxygen of Gln¹⁸⁹; these two atoms are highlighted in fig. S9). However, our previous crystal structures showed that the P2 main-chain amide of the linear α-ketoamides does not make a hydrogen bond with the protein in all cases, so this interaction does not seem to be critical (6). The protecting Boc group on P3 does not occupy the canonical S4 site of the protease [in contrast to the protecting groups of other inhibitors in complex with the SARS-CoV M^pro (18)] but is located near Pro¹⁶⁸ (3.81/4.17/3.65 Å) (Fig. 3); as a result of this interaction, the latter residue moves outward by more than 2 Å (relative to the structure of the free enzyme). This contact explains why removing the Boc group as in compound 14b (Fig. 1, purple ovals) weakens the inhibitory potency of this compound by a factor of ~2. Interestingly, there is a space between the pyridone ring of 13b, the main chain of residue Thr¹⁹⁰, and the side chain of Gln¹⁸⁹ (smallest distance: 3.6 Å), which is filled by a dimethyl sulfoxide (DMSO) molecule in the C2 crystal structure and a water molecule in the P2₁2₁2₁ structure. This suggests that P3 moieties more bulky than pyridone may be accepted here.

Compound 13b inhibits the purified recombinant SARS-CoV-2 M^pro with IC₅₀ = 0.67 ± 0.18 μM. The corresponding IC₅₀ values for inhibition of the SARS-CoV M^pro and the MERS-CoV M^pro are 0.90 ± 0.29 μM and 0.58 ± 0.22 μM, respectively. In a SARS-CoV replicon (19), RNA replication is inhibited with EC₅₀ = 1.75 ± 0.25 μM. In human Calu-3 cells infected with SARS-CoV-2, an EC₅₀ of 4 to 5 μM was observed, whereas compound 14b lacking the Boc group was almost inactive (Fig. 4). This suggests that the hydrophobic and bulky Boc group is necessary to cross the cellular membrane and that an even more hydrophobic moiety might be advantageous here, although this may again lead to increased plasma protein binding, as observed for the cinnamoyl-containing 11r.

To assess the absorption-distribution-metabolism-excretion (ADME) properties of the pyridone-containing α-ketoamides, we first investigated compound 13a. Metabolic stability in mouse and human microsomes was good, with intrinsic clearance rates Cl_{int_mouse} = 32.0 μl min^–1 (mg protein)^–1 and Cl_{int_human} = 21.0 μl min^–1 (mg protein)^–1. This means that after 30 min, ~80% and 60% (for mouse and human, respectively) of residual compound remained metabolically stable. Pharmacokinetic studies in CD-1 mice using the subcutaneous route at 20 mg/kg showed that 13a stayed in plasma for up to 4 hours but was excreted via urine for up to 24 hours. The maximum plasma concentration (C_max) was determined at 334.5 ng ml⁻¹ and the mean residence time was ~1.6 hours. Although 13a seemed to be cleared very rapidly from plasma, at 24 hours it was found at 135 ng/g tissue in the lung and at 52.7 ng ml⁻¹ in bronchio-alveolar lavage fluid (BALF), which suggests that it was mainly distributed to tissue. Next, we investigated 13b for its pharmacokinetic properties in CD-1 mice using the subcutaneous route as well, but at 3 mg kg⁻¹. The ADME parameters of 13b were similar to those of 13a; in addition, binding to human plasma proteins was found to be 90%. The C_max of 13b was determined at 126.2 ng ml⁻¹. This is around 37% of the C_max detected for 13a, although the 13b dosage was lower by a factor of ~7. The mean residence time for 13b was extended to 2.7 hours and the plasma half-life in mice was 1.8 hours. In addition, 13b showed a less rapid clearance relative to 13a (table S3). During the pharmacokinetic study with 13b, we monitored its lung tissue levels. After 4 hours, 13b was still found at ~13 ng g⁻¹ in lung tissue. This lung tropism of 13a and 13b is beneficial given that COVID-19 affects the lungs. In addition to subcutaneous administration, 13b was nebulized using an inhalation device at 3 mg kg⁻¹. After 24 hours, 13b was found at 33 ng g⁻¹ in lung tissue. Inhalation was tolerated well and mice did not show any adverse effects, which suggests that direct administration of the compound to the lungs would be possible. Given these favorable pharmacokinetic results, our study provides a useful framework for the development of the pyridone-containing inhibitors toward anticoronaviral drugs.