key: cord-326282-uxn64olw
authors: Lu, Maolin; Uchil, Pradeep D.; Li, Wenwei; Zheng, Desheng; Terry, Daniel S.; Gorman, Jason; Shi, Wei; Zhang, Baoshan; Zhou, Tongqing; Ding, Shilei; Gasser, Romain; Prévost, Jérémie; Beaudoin-Bussières, Guillaume; Anand, Sai Priya; Laumaea, Annemarie; Grover, Jonathan R.; Liu, Lihong; Ho, David D.; Mascola, John R.; Finzi, Andrés; Kwong, Peter D.; Blanchard, Scott C.; Mothes, Walther
title: Real-time Conformational Dynamics of SARS-CoV-2 Spikes on Virus Particles
date: 2020-09-13
journal: bioRxiv
DOI: 10.1101/2020.09.10.286948
sha: 
doc_id: 326282
cord_uid: uxn64olw

SARS-CoV-2 spike (S) mediates entry into cells and is critical for vaccine development against COVID-19. Structural studies have revealed distinct conformations of S, but real-time information that connects these structures, is lacking. Here we apply single-molecule Förster Resonance Energy Transfer (smFRET) imaging to observe conformational dynamics of S on virus particles. Virus-associated S dynamically samples at least four distinct conformational states. In response to hACE2, S opens sequentially into the hACE2-bound S conformation through at least one on-path intermediate. Conformational preferences of convalescent plasma and antibodies suggest mechanisms of neutralization involving either competition with hACE2 for binding to RBD or allosteric interference with conformational changes required for entry. Our findings inform on mechanisms of S recognition and conformations for immunogen design.

against the virus(1-4). S is synthesized as a precursor, processed into S1 and S2 by furin 15 proteases, and activated for fusion when human angiotensin-converting enzyme 2 (hACE2) engages the receptor-binding domain (RBD) and when the N-terminus of S2 is proteolytically processed (5) (6) (7) . Structures of soluble ectodomains and native virus particles have revealed distinct conformations of S, including a closed trimer with all RBD oriented downward, trimers with one or two RBDs up, and hACE2-stabilized conformations with up to three RBD oriented 20 up (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) ). Real-time information that connects these structures, however, has been lacking.

smFRET is well suited to inform on conformational dynamics of proteins reporting domain movements in the millisecond to second range, and has previously been applied to study HIV-1, influenza A, and Ebola spike glycoproteins, via measurements of the distance-dependent energy transfer from an excited donor to a nearby acceptor fluorophores in real-time(19-23).

To probe dynamics of SARS-CoV-2 spikes, we used available high-resolution structures of the SARS-CoV-2 S trimer to identify sites of fluorophore pair labeling that have the potential to 5 inform on distance changes expected to accompany conformational changes between the RBDdown and receptor hACE2-induced RBD-up trimer structures(14, 17) (Fig. S1 ). Accordingly, we engineered A4 and Q3 labeling peptides before and after the receptor-binding motif (RBM) to allow site-specific introduction of donor and acceptor fluorophores at these positions ( Fig. 1, A, B, and Fig. S1 ). We optimized retroviral and lentiviral pseudoviral particles carrying the SARS- 10 CoV-2 S protein (Fig. S2) to test the impact of these peptides on infectivity, and found that they were well tolerated, both individually and in combination (Fig. S1D ). To measure conformational dynamics of the SARS-CoV-2 S trimer on the surface of virus particles, we established two types of particles, lentiviral pseudoparticles carrying S, as well as coronaviruslike particles generated by expression of S, membrane (M), envelope (E) and nucleocapsid (N) 15 protein (S-MEN)(24, 25) (Fig. 1, A and B ). S-MEN particles co-express coronavirus surface proteins M and E. Particle quality and the presence of the corona-like S proteins on both particle surfaces were confirmed by cryo-electron microscopy ( Fig. 1, C and D) .

For smFRET, lentivirus particles and S-MEN particles were generated (see Materials and 20 Methods) by transfecting HEK293T cells with an excess of plasmid-encoding wild-type, doped with trace amounts of plasmid expressing labeling peptide-carrying S to ensure the production of virus particles that contain, on average, only a single engineered S protein. As for analogous investigations of HIV-1 Envelope protein(19, 21), donor (Cy3B(3S)) and acceptor fluorophore (LD650) were enzymatically conjugated to the engineered S proteins presented on the virus particle surface in situ (see Materials and Methods). A biotinylated lipid was then incorporated into the virus particle membrane to allow their immobilization within passivated microfluidic devices coated with streptavidin to enable imaging by total internal reflection microscopy 5 (Fig.1A) . Donor fluorophores on single, immobilized virus particles were excited by a singlefrequency 532 nm laser and fluorescence emission from both donor and acceptor fluorophores were recorded at 25 Hz ( Fig. 2A) . From the recorded movies, we computationally extracted hundreds of smFRET traces exhibiting anti-correlated donor and acceptor fluorescence intensities, the telltale signature of conformational changes within the S protein on individual 10 virus particles.

Analyses of smFRET data from ligand-free S protein on lentiviral particles revealed that the SARS-CoV-2 S protein is dynamic, sampling at least four distinct conformational states To identify the receptor-bound conformation of the SARS-CoV-2 S protein by smFRET, we measured the conformational consequences of soluble, monomeric hACE2 binding. Addition of 5 the monomeric hACE2 receptor to surface-immobilized virus particles lead to increased occupancy of the low-(~0.1) FRET S protein conformation (Fig. 2E) , which was observed at both the single-molecule and population level (Fig. 2F ). Similar hACE2 receptor impacts on the SARS-CoV-2 S protein were observed in both lentiviral particle and S-MEN coronavirus-like particle contexts (Fig. 2 , E to G). Dimeric hACE2, a more potent ligand (Fig. S4A )(26), 10 stabilized the low-(~0.1) FRET S protein conformation more efficiently (Fig. S4 , B and C), suggesting that the observed low-FRET state likely represents the receptor-bound state in which all three RBD domains are oriented upwards (RBD-up conformation).

A unique strength of single-molecule imaging is its capacity to reveal directly both the structural 15 and kinetic features that define biological function (27, 28) . To extract such information for the SARS-CoV-2 S protein, we employed Hidden Markov Modeling (HMM)(29) to idealize individual smFRET traces. These data allowed quantitative analyses of the thousands of discrete hACE2-binding -could be achieved spontaneously. 15 As expected, the binding of the hACE2 receptor modified the dynamic S protein conformational landscape towards the RBD-up conformation (~0.1 FRET), rendering it the most populated ( Fig.   2 , B, C, F, G). This change resulted from an increase in the transition rate from the RBD-down conformation (~0.5 FRET) towards the intermediate-(~0.3) FRET state and subsequently the RBD-up (~0.1 FRET) conformation, which was also modestly stabilized. The energy barriers for 20 reverse transitions towards the RBD-down conformation (~0.5 FRET), were also elevated, explaining receptor-bound conformation accumulation over time (Figs. S5). These analyses lead to a qualitative model for hACE2 activation of the SARS-CoV-2 S protein from the ground state to the receptor-bound state through at least one intermediate conformation (Fig. 2H ). The summary of relative state occupancies, transition rates among conformations and errors are listed in Tables S1 and S2, respectively.

In most cell types, the serine protease TMPRSS2 is required for pH-independent SARS-CoV-2 5 entry (5, 30, 31) . In vitro, the effect of TMPRSS2 is mimicked by the serine protease trypsin, which has similar cleavage specificity (5, 31) . smFRET analysis of trypsin-treated S protein on lentiviral particles in the absence of receptor revealed a clear shift towards activation (Fig. 3 , A, B). After trypsin treatment, the addition of hACE2 receptor was more effective at stabilizing the S protein in the RBD-up (~0.1 FRET) conformation (Fig. 3 , C and D, Fig. S6 ). To further 10 validate this finding, we measured the effects of trypsin pre-treatment in virus-cell and cell-cell fusion assays using split NanoLuc system consisting of LgBiT and HiBiT (Fig. 3, E and F, Fig.   S7 ). Here, membrane fusion restored luciferase activity between lentiviral particles carrying the S protein as well as a Vpr-HiBiT fusion protein with cells expressing the LgBiT counterpart fused to a PH domain. This assay revealed fusion to be strictly dependent on the hACE2 receptor 15 and to be stimulated by trypsin treatment (Fig. 3 , E and F). Nearly identical results were observed for cell-cell fusion between donor cells expressing S and target cells expressing hACE2 ( Fig. S7) , confirming the activating role of trypsin treatment.

We next explored the suitability of the smFRET assay to characterize the conformational 20 consequences of antibody binding to the SARS-CoV-2 S protein. Multiple studies on antibodies generated from COVID-19 patients have shown that one type of antibody often dominates immune responses(32-37). This prompted us to screen plasma from convalescent patients with 8 neutralizing activity that can bind to the S protein on lentiviral particles(38) using a modified virus-capture assay (VCA)(39). Cross-reactive CR3022(40), one of the very first reported antibodies from SARS-CoV-1 that also bind to SARS-CoV-2 spike RBD domain, served as a good indicator of RBD binding (Fig. 4A) . We identified two plasma samples (S002 and S006) able to specifically bind the RBD, recognize S expressed at the cell surface and to neutralize 5 viral particles (Fig. 4 , A to C, and Fig. S8 ). smFRET analysis of antibody-bound S revealed that both CR3022 and plasma from patient S006, stabilized S in the RBD-up (~0.1 FRET) conformation, in a similar fashion as receptor hACE2 (Fig. 4 , D and E). These data point to the presence of RBD-directed antibodies in patient S006. By contrast, smFRET indicated that plasma from patient S002 contained an activity that stabilized the RBD-down (~0.5 FRET) 10 conformation (Fig. 4F ). Plasma S002 antagonized hACE2 binding, but RBD competition did not affect its recognition of S, suggesting that its neutralization activity does not solely rely on blocking the receptor interface We then assessed the conformational preference of four RBDdirected antibodies: the potently neutralizing antibodies H4, 2-4 and 2-43, and the neutralization nanobody VHH72, each of which binds RBD in a different way(41-43). Antibody H4 and 15 nanobody VHH72 stabilized the S protein in an RBD-up (~0.1 FRET) conformation similar to hACE2, CR3022, and S006, whereas antibody 2-4 shifted the conformational landscape towards RBD-down (~0.5 FRET) conformation, similar to S002 (Fig. 4 , G to J). The very potent neutralizing antibody 2-43(43), meanwhile, showed a partial shift to the RBD-up (~0.1 FRET) conformation (Fig. S9) . The absence or presence of hACE2 did not appear to affect the RBD-up 20 stabilization evidenced for antibodies CR3022, S006, VHH72, or H4 (Fig. S9) . However, plasma S002, and to a lesser extent antibody 2-4, reduced the hACE2-dependent stabilization of the RBD-up (~0.1 FRET) conformation, suggesting that they may interfere with hACE2 receptor binding via an allosteric mechanism. These findings indicate that SARS-CoV-2 neutralization can be achieved in two ways: 1) antibodies that conformationally mimic hACE2 and directly compete with hACE2 receptor binding, or 2) by allosterically stabilizing the S protein in its RBD-down conformation. 5 The strength of the presented smFRET approach is revealed by the capacity to examine the dynamic properties of the S protein in real time, including: 1) the distinct conformational states that it spontaneously transits under physiological conditions; 2) the impact of sequence alterations on S protein dynamics; and 3) the responses of the S protein to cognate hACE2 receptor and antibody recognition. 10 The present analyses of dynamic S protein molecules provides three lines of evidence that indicate that the intermediate-(~0.5) FRET state observed represents the RBD-down, ground state conformation of the S protein, in which all three RBD domains are oriented towards the viral particle membrane. First, in line with previous electron microscopy (EM) investigations (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) 16), the RBD-down state is the most populated. In further agreement with recent EM studies, both the disulfide bridge (S383C, D985C)(9, 11) and antibody 2-4 stabilized the S protein in a conformation with all three RBD oriented down 42 . While our smFRET observations highlight considerable conformational flexibility in these contexts compared to EM of soluble trimers, we attribute these distinctions to a tendency of our analysis approach to over-emphasize dynamic 20 features, while EM may over-emphasize static conformations rigidified by cryogenic temperatures that may be more readily identified and classified(44). Multiple lines of evidence also facilitated assignment of the RBD-up (~0.1 FRET) conformation of the S protein with all three RBD domains oriented away from the virus particle membrane. For instance, this conformation was stabilized by soluble monomeric hACE2 receptor, and even further stabilized in the presence of soluble, dimeric hACE2 receptor as well as RBD-targeting antibodies, such as CR3022, that are known to access their epitopes when the S protein is in an activated, RBD-up conformation(40-42). The structure of the on-path (~0.3 FRET) intermediate observed during S1 5 opening is likely similar to the all-down ground state; cryo-EM structures of soluble SARS-CoV-2 S trimers(17) that engage one or two hACE2 molecules receptors(18) reveal that the distance between the two labeling sites increases in the ligand-free protomers adjacent to a protomer engaged to hACE2 (Fig. S1A ). The additional, highly compacted S conformation (~0.8 FRET) evidenced, which is also depopulated by activating ligands, remains unknown. 10 These smFRET analyses are in global agreement with the conformational states observed at the single particle EM and cryoET level (6, 8, 9, 11-17, 40-42, 45, 46) . The observed FRET changes are also are in good agreement with expected increase in the distance between the labeling peptide insertion sites that carry the fluorophores in the RBD-down and RBD-up conformations 15 of the S trimer. The capacity to examine the conformational preferences of RBD-directed antibodies to the S protein enabled us to identify conformational signatures of antibodies in patient plasma. This approach identified patients with antibody activities that either mimicked ACE2 (indicating anti-RBD activity) or stabilized the ground state of S, thereby interfering with Data and materials availability: All data is available in the main text or the supplementary materials. The data that support the findings of this study are available from the corresponding authors upon reasonable request. The full source code of SPARTAN, which was used for analysis of smFRET data, is publicly available. (http://www.scottcblanchardlab.com/software).

Some small customized Matlab scripts are available upon reasonable requests. 

A full-length wild-type pCMV3-SARS-CoV-2 Spike (S1+S2)-long (termed as pCMV-S, codon-optimized, Sino Biological, cat # VG40589-UT) plasmid was used as a template to generate tagged pCMV-S. The translated amino acid sequence of pCMV-S is identical to Each pair of inserted tags did not compromise S-dependent lentivirus infectivity. 15 Infectivity measurements

The infectivity of lentivirus particles carrying SARS-CoV-2 spike proteins was Cryo-electron tomography 6 nm gold tracer was added to the concentrated S-decorated HIV-1 lentivirus and S-MEN particles viruses at 1:3 ratio, and 5 µl of the mixture was placed onto freshly glow discharged holey carbon grids for 1 min. Grids were blotted with filter paper, and rapidly frozen in liquid 10 ethane using a homemade gravity-driven plunger apparatus.

Cryo-grids were imaged on a cryo-transmission electron microscope (Titan Krios, Thermo Fisher Scientific) that was operated at 300 kV, using a Gatan K3 direct electron detector in counting mode with a 20 eV energy slit. Tomographic tilt series between −51° and +51° were collected by using SerialEM(55) in a dose-symmetric scheme(56) with increments of 3°. The 15 nominal magnification was 64,000 X, giving a pixel size of 1.346 Å on the specimen. The raw images were collected from single-axis tilt series with accumulative dose of ~50 e− per Å 2 . The defocus was -3 µm and 8 frames were saved for each tilt angle.

Frames were motion-corrected using Motioncorr2(57) to generate drift-corrected stack files, which were aligned using gold fiducial makers by IMOD/etomo(58). Tomograms were 20 reconstructed by weighted back projection and tomographic slices were visualized with IMOD.

Virus particles were labeled through site-specifically enzymatic labeling, as previously The SARS-CoV-2 RBD ELISA assay used was recently described (38, 53) introduction of fluorophores (Cy3B, green; LD650, red) was guided by conformational changes in S1 induced by binding of the cellular receptor human angiotensin-converting enzyme 2 (hACE2) from the "RBD-down" to the "RBD-up" conformation (Fig. S1) . RBD, receptorbinding domain; NTD, N-terminal domain. Structures were adapted from RCSB Protein Data Bank accessories 6VSB ('Down' S1/S2 protomer: S1, light cyan; S2, dark blue) and 15 6VYB/6M0J ('Up' protomer S1/S2 engaged with hACE2: hACE2, magenta). Table S1 . (H) Relative free energy model of conformational landscapes of SARS-CoV-2 spikes in response to the hACE2 binding. The differences in free energies between states were roughly scaled based upon relative state occupancies of each state. (Fig. 2B) . FRET histograms represent mean ± s.e.m., determined from three randomly assigned populations of FRET traces. For state occupancies see Table S1 . FRET histograms represent mean ± s.e.m., determined from three randomly-assigned populations of all FRET traces. Evaluated state occupancies see Table S1 . Table S1 . Table S1 . Relative state-occupancy and fitting parameters in each of four FRET-defined conformational states of SARS-CoV-2 spike protein on the surface of virus particles. The FRET efficiency histograms were fitted into the sum of four Gaussian distributions (µ, the mean or expectation of the Gaussian distribution; σ, s.d. of the Gaussian distribution) for each conformational state. Parameters were based upon the observation of original FRET efficiency 5 data and were further determined using hidden Markov modelling. Relative conformational stateoccupancy of SARS-CoV-2 spike protein on viral particles are presented as mean ± s.e.m., determined from three independent measurements. R-squared values were evaluated to indicate the goodness of fit. Ligand-free 0.9886 26% +/-6% 23% +/-11% 39% +/-12% 12% +/-7% + hACE2 0.9977 49% +/-8% 24% +/-11% 20% +/-11% 7% +/-6% Table S2 . Transition rates between observed conformational states of SARS-CoV-2 spike on virus particles. The survival probability plots (Figs. S5 and S6) were derived from distributions of dwell times for each state-to-state transitions determined through Hidden Markov Modeling (HMM). Then plots were fitted by double-exponential distributions: y(t) = A 1 exp -k1t + A 2 exp -k2t ), where y(t) is the probability and t is the dwell time. The presented rates were the 5 weighted average of two rates derived from double-exponential decays. Rates were finally presented in the table as (weighted average +/-95% confidence intervals).

Development of an inactivated vaccine candidate for SARS-CoV-2

An mRNA Vaccine against SARS-CoV-2 -Preliminary Report

DNA vaccine protection against SARS-CoV-2 in rhesus macaques

Single-shot Ad26 vaccine protects against SARS-CoV-2 in rhesus macaques

SARS-CoV-2 cell entry depends on ACE2 and TMPRSS2 and is blocked by a clinically proven protease inhibitor

Structure, Function, and Antigenicity of the SARS-CoV

SARS-CoV-2 and bat RaTG13 spike glycoprotein structures inform on virus evolution and furin-cleavage effects

Distinct conformational states of SARS-CoV-2 spike protein

Controlling the SARS-CoV-2 spike glycoprotein conformation

Structure-based design of prefusion-stabilized SARS-CoV-2 spikes

Closing coronavirus spike 30 glycoproteins by structure-guided design. bioRxiv

Structural basis of receptor recognition by SARS-CoV-2

Structural and functional basis of SARS-CoV-2 entry by using human ACE2

Structure of the SARS-CoV-2 spike receptor-binding domain bound to the ACE2 receptor

Molecular architecture of the SARS-CoV-2 virus. bioRxiv

Structures and distributions of SARS-CoV-2 spike proteins on intact virions