Molecular switches regulating the potency and immune evasiveness of SARS-CoV-2 spike protein

SARS-CoV-2 spike protein plays a key role in viral entry and host immune responses. The conformation of the spike protein can be either open or closed, yet it is unclear how the conformations affect the protein’s functions or what regulate the conformational changes. Using SARS-CoV-1 and bat RaTG13-CoV as comparisons, we identified two molecular switches that regulate the conformations of SARS-CoV-2 spike protein: (i) a furin motif loop turns SARS-CoV-2 spike from a closed conformation to a mixture of open and closed conformations, and (ii) a K417V mutation turns SARS-CoV-2 spike from mixed conformations to an open conformation. We showed that the open conformation favors viral potency by exposing the RBD for receptor binding and viral entry, whereas the closed conformation supports viral immune evasion by hiding the RBD from neutralizing antibodies. Hence SARS-CoV-2 spike has evolved to reach a balance between potency and immune evasiveness, which may contribute to the pandemic spread of SARS-CoV-2. The dynamics between viral potency and invasiveness is likely to further evolve, providing insights into future evolution of SARS-CoV-2.


Introduction 43
Coronaviruses have a long history of infecting humans and animals, yet none had 44 caused the same devastation as produced by SARS-CoV-2 (1, 2). For example, a virulent 45 and lethal coronavirus, SARS-CoV-1, yielded a much smaller outbreak in humans in 46 2002-2003 (3, 4). Numerous human coronaviruses such as NL63-CoV cause common 47 colds annually (5, 6). With an intermediate virulence, SARS-CoV-2 causes a fatality rate 48 that is significantly lower than that of SARS-CoV-1, but much higher than that of NL63-49 CoV. SARS-CoV-2 carriers show clinical signs that facilitate the spread of the virus: they 50 may develop mild or no symptoms, experience delayed onset of symptoms, develop low 51 levels of neutralizing antibodies, or endure prolonged virus shedding period (7-11). These 52 features contribute to the wide spread of SARS-CoV-2 and severe health outcomes, 53 triggering a global COVID-19 pandemic that is unprecedented in the era of modern 54 medicine. Understanding the molecular determinants of COVID-19 provides important 55 clues to the evolution and cross-species transmission of coronaviruses. A dangerous 56 feature of coronaviruses is their propensity to cross species barriers (12, 13). In fact, 57 coronaviruses similar to human coronaviruses such as SARS-CoV-1 and NL63-CoV have 58 been identified in bats and other animals (14-16). RaTG13-CoV, a coronavirus with 59 ~96% genomic sequence homology with SARS-CoV-2, has been identified in bats (17). 60 Thus, coronaviruses that originate from bats or other animals pose a long-term threat to 61 humans. A comparison of the molecular mechanisms of SARS-CoV-2 and other 62 coronaviruses not only facilitate an understanding of the COVID-19 pandemic, but also 63 shed light on the evolution of coronaviruses, including their cross-species transmission 64 and adaptation to humans. 65 6 significant cleavage, suggesting that inactivation of FnM successfully suppressed furin 112 cleavage of the spikes. Second, we performed a protein pull-down assay using 113 recombinant human ACE2 as the bait and the cell-surface-anchored spikes as the target. 114 For cross validation, both His-tagged ACE2 and Fc-tagged ACE2 were used. We 115 previously showed that this pull-down assay is a reliable method to probe the RBD 116 conformation in cell-surface-anchored spikes, with higher pull-down levels of the spikes 117 associated with more spike molecules in the RBD-open conformation (27). Our results 118 showed that the wild type and FnM-point spikes had similar affinities for ACE2, and both 119 demonstrated much higher affinities for ACE2 than the two FnM-deletion spikes (Fig.  120 1C). Third, we performed a pseudovirus entry assay where retroviruses pseudotyped with 121 SARS-CoV-2 spike (i.e., SARS-CoV-2 pseudoviruses) were used to enter cells 122 expressing human ACE2 (Fig. 1D). The result showed that the FnM-point spikes 123 mediated pseudovirus entry slightly worse than the wild type spike, suggesting that furin 124 pre-activation had small, albeit significant, impact on SARS-CoV-2 spike's capability in 125 mediating viral entry. In contrast, both of the FnM-deletion spikes mediated pseudovirus 126 entry much worse than both the wild type spike and FnM-point spike, suggesting that the 127 closed conformation of the spike substantially reduced its capability to mediate viral 128 entry. The data from protein pull-down and pseudovirus entry assays revealed that FnM 129 deletion resulted in decreased potency of SARS-CoV-2 spike, as demonstrated in reduced 130 ACE2 binding and reduced capability of mediating viral entry. These results suggest that 131 due to the FnM deletion, more SARS-CoV-2 spike molecules switched to the closed 132 conformation with reduced potency.

7
Next we directly visualized the conformation of SARS-CoV-2 spike containing 134 the FnM deletion using cryo-EM. To this end, we expressed and purified the ectodomain 135 of SARS-CoV-2 spike containing the FnM deletion (it also contained a C-terminal foldon 136 trimerization tag and two proline mutations in S2, both of which stabilize the pre-fusion 137 structure). As a comparison, we also prepared the ectodomain of SARS-CoV-2 spike 138 containing the FnM point mutation (in addition to the foldon tag and proline mutations). 139 We then collected cryo-EM data on both of these proteins and performed 3-D showed that all of the FnM-deletion spike molecules were in the closed conformation 148 with all three RBDs hidden ( Fig. 2A). Therefore, consistent with our biochemical data, 149 our cryo-EM data confirmed that the FnM deletion caused SARS-CoV-2 spike to switch 150 to the closed conformation. 151 We further determined the cryo-EM structures of SARS-CoV-2 FnM-deletion 152 spike ectodomain at 3.8 Å and FnM-point spike ectodomain at 4.4 Å ( Fig. 2A, 2B; Fig.  153 S2A, Fig. S2B). Overall, the two structures are similar to each other and to the previously 154 determined cryo-EM structures of FnM-point spike ectodomain and virus-surface wild 155 type full-length spike (28, 29). In the trimeric spike structures, each S1 subunit contains 8 an N-terminal domain (NTD), an RBD, and two subdomains (SD1 and SD2); the RBD 157 from one S1 subunit packs against the NTD from another S1 subunit and it also packs 158 against the two RBDs from the other two S1 subunits ( Fig. S3A)  with the hinge region and the NTD (Fig. S3A) (28). Detailed structural analysis revealed 162 that compared to the FnM-point spike, the RBD and NTD in each S1 subunit of the FnM-163 deletion spike rotated towards each other by ~2.5 o (Fig. S4A). Because of this movement, 164 compared to the FnM-point spike, the RBD/NTD interface, the RBD/RBD interface and 165 hence the total interface in trimeric S1 all increased significantly in the FnM-deletion 166 spike, leading to enhanced S1 packing (Fig. S3B). As a comparison, the corresponding 167 interfaces in a previously determined FnM-point spike were similar to those in our FnM-168 point spike ( overall similar structures (Fig. S4B), we combined the structural information from these 174 two spikes, which revealed an interaction network involving the FnM loop, anchor loop, 175 the hinge region, and the NTD (Fig. S4B). Hence, one possibility is that the FnM deletion 176 disturbed this interaction network and caused the movements of the RBD and NTD, 177 which subsequently led to enhanced S1 packing, reduced dynamics of the RBD and 178 hence the closed spike. Thus, as supported by the biochemical data and 3D classification 9 data, the physical presence of the FnM, instead of furin cleavage per se, leads to open 180 spike molecules by reducing S1 packing. 181 To further understand the relationship between the presence of FnM and the 182 conformation of the spikes, we inserted FnM into RaTG13-CoV spike (i.e., FnM-insert) 183 (Fig. 3A). As a comparison, we also inserted a random sequence, glycine-serine-glycine-184 serine, into the same location as the inserted FnM in RaTG13-CoV spike (i.e., GSGS-185 insert) (Fig. 3A). When expressed on cell surfaces, FnM-insert spike, but not wild type 186 spike or GSGS-insert spike, was cleaved by furin ( Fig. 3B), confirming the introduction 187 of FnM. We could not obtain recombinant RaTG13 spike ectodomains (wild type or 188 mutants) that were stable enough for cryo-EM analysis (recombinant spike ectodomains 189 are generally less stable than full-length membrane-anchored spikes). Instead, we CoV-2 and SARS-CoV-1 spikes in the context of their tertiary structures. We identified 200 residue 417 as potentially a key difference between the two spikes: in the closed SARS-201 CoV-2 spike, Lys417 in the RBD forms a salt bridge with the RBD from another subunit, 202 stabilizing the RBD in the closed conformation and hence enhancing S1 packing; it 203 becomes a valine in SARS-CoV-1 spike, losing its capability to interact with the other 204 subunit and hence reducing S1 packing (Fig. 4A). We introduced the K417V mutation 205 into SARS-CoV-2 spike, and examined its impact on the conformation of SARS-CoV-2 206 spike. Both the protein pull-down and pseudovirus entry assays demonstrated that 207 compared to the wild type spike, the K417V mutation allowed more spike molecules to 208 open up for binding ACE2 and mediating viral entry (Fig. 4B, 4C). We could not obtain 209 recombinant SARS-CoV-2 K417V spike ectodomain that was stable enough for cryo-EM 210 analysis. Instead, we prepared recombinant SARS-CoV-2 spike ectodomain containing 211 the K417V mutation and FnM deletion (in addition to proline mutations) (K417V/FnM-212 deletion). Cryo-EM analysis at 4.6 Å revealed that 91% of the K417V/FnM-deletion 213 spike molecules were open and 9% were closed (Fig. 2C). In comparison, as presented 214 earlier, 100% of the recombinant FnM-deletion spike molecules were closed ( Fig. 2A). 215 Therefore, despite lacking FnM, SARS-CoV-1 spike is open due to Val417 and 216 potentially other residues that destabilize the closed conformation of the RBD and reduce 217 S1 packing. 218 To understand how the RBD conformations of SARS-CoV-2 spike affect host 219 immune responses targeting the RBD, we immunized mice with one of the following 220 three recombinant SARS-CoV-2 spike ectodomains: FnM-deletion spike, FnM-point 221 spike, and K417V/FnM-deletion spike (in addition to the proline mutations in all of 222 them). Four weeks after the initial immunization, the mice were further boosted with the 223 same immunogen. Ten days after the second immunization, mouse sera were collected. 224 We measured the amounts of RBD-specific antibodies in the mouse sera using ELISA. 225 The result showed that K417V/FnM-deletion spike and FnM-point spike induced 226 significantly more RBD-specific antibodies than FnM-deletion spike (Fig. 5A). We 227 further measured the amounts of neutralizing antibodies in the mouse sera using 228 pseudovirus entry inhibition assay. The result showed that K417V/FnM-deletion spike 229 and FnM-point spike induced significantly more neutralizing antibodies than FnM-230 deletion spike (Fig. 5B). These data confirm that more molecules of K417V/FnM- (29) and a study that used the same protein construct and similar protein preparation to 246 the current study (28). Importantly, our cryo-EM analysis is consistent with our three proportion of closed spikes, making the virus more immune evasive but less potent. If 317 that happens, SARS-CoV-2 may become an endemic (but milder) virus like NL63-CoV 318 (NL63-CoV RBD binds to human ACE2 with high affinity, but is hidden in the closed 319 spike) (25, 30, 40). This study showed that just one or a few structural changes in the 320 spike protein can significantly impact the dynamics between viral potency and 321 evasiveness. This makes coronaviruses a current and future danger to human health. 322 Understanding the molecular determinants that regulate the potency and evasiveness of 323 coronaviruses is critical not only for our understanding the current COVID-19 pandemic, 324 but also for monitoring and preparing for potential future coronavirus pandemics. 325

Acknowledgements 326
This work was supported by NIH grants R01AI089728 and R01AI110700 (to 327  Relative light unites (RLUs) were measured using EnSpire plate reader (PerkinElmer). In 372 the meanwhile, the amounts of pseudovirus-packaged spikes were measured by western 373 blot using an anti-c9 antibody and then were quantified using Fiji (https://imagej.net/). 374 The RLUs were then normalized against the amounts of pseudovirus-packaged spikes. 375 All of the measurements were carried out in quadruplicates. 376 For pseudovirus entry inhibition, mouse sera were serially diluted in DMEM 377 media and then mixed with SARS-CoV-2 pseudoviruses. Subsequently the mixtures were 378 added to HEK293T cells expressing human ACE2 for the pseudovirus entry assay. The 379 fitted curves and the 50% neutralizing antibody titers (NT 50 ) were calculated using the 380 Graphpad Prism program. All the measurements were carried out in triplicates. 381

Western blot 382
Pseudoviruses were mixed with SDS loading buffer and then were incubated at 383 95°C for 10 min. Samples were run in a 10% SDS Tris-Glycine Gel and transferred to a 384 PVDF membrane. An anti-c9 or anti-His 6 monoclonal primary antibody (1:1000 dilution, 385 Santa Cruz Biotech) and a horseradish peroxidase-conjugated mouse secondary antibody 386 (1:10,000 dilution, Jackson Laboratory) were used for Western blotting. A LAS-4000 387 imager was used to develop images. 388

Cryo-electron microscopy (cryo-EM) 406
For sample preparation, aliquots of recombinant SARS-CoV-2 spike ectodomain 407 (3 µl; 0.35 mg/ml; in buffer containing 10 mM Tris pH7.4 and 100 mM NaCl) were 408 applied to glow-discharged CF-2/1-4C C-flat grids (Protochips). The grids were then 409 plunge-frozen in liquid ethane using a Vitrobot system (FEI Company).  Table S2. 418 For data processing, whole frames in each movie were corrected for motion and 419 dose compensation using MotionCor2 (43). The final images were bin-averaged to reach 420 a pixel size of 1.04 Å. The parameters of the microscope contrast transfer function were 421 estimated for each micrograph using GCTF (44). Particles were automatically picked and 422 extracted using Gautomatch (http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch/) and 423 RELION (45) with a box size of 300 pixels. For FnM-deletion spike ectodomain, 728,804 424 particles were initially extracted and subjected to 2D alignment and clustering using 425 RELION. The best classes were then selected for an additional 2D alignment. ~5,000 best 426 particles were selected for creating the initial 3D model using RELION. 107,268 particles 427 selected from 2D alignment were then subjected to 3D classification. The best class with 428 65,302 particles was subjected to 3D refinement to generate the final density map with 429 C3 symmetry. For FnM-point spike ectodomain, 583,127 particles were initially 430 extracted and subjected to 2D alignment and clustering using RELION. The best classes 431 were then selected for an additional 2D alignment. ~5000 best particles were selected for 432 creating the initial 3D model using RELION. 52,134 particles selected from 2D 433 alignment were then subjected to 3D classification for open and closed conformations. The best open-conformation class with 21,894 particles and the best closed-conformation 435 class with 23,849 particles were subjected to 3D refinement to generate the final density 436 maps with C1 symmetry and C3 symmetry, respectively. For K417V/FnM-deletion spike 437 ectodomain, 1,267,763 particles were initially extracted and subjected to 2D alignment 438 and clustering using RELION. The best classes were then selected for an additional 2D 439 alignment. 124,721 best particles were selected for creating the initial 3D model using  Table S2. 462

Calculation of interface area 463
The buried surface areas between NTD and RBD and between RBD and RBD in 464 the trimeric spike ectomains were calculated using the PISA server at the European 465 Bioinformatics Institute (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html) (51). For each 466 trimeric spike ectodomain, a PDB file containing the coordinates from the pair of the 467 corresponding domains was submitted to the PISA server, and the buried surface area for 468 each pair was individually calculated. 469

Calculation of angle between domains 470
The rotation angle between the S1 domains in SARS-CoV-2 spike structures was 471 calculated using the angle_between_domains script in the Psico program 472 After another incubation at 37°C for 1 h, ELISA substrate (Sigma-Aldrich) was added. 490 The ELISA reaction was stopped using 1N H 2 SO 4 , and the ELISA signal was read at the 491 450 nm wavelength using an ELISA plate reader (Tecan). 492

Ethics statement 493
Mouse work was performed in strict accordance with the guidance and 494