Quantitative determination of mechanical stability in the novel coronavirus spike protein

Rodrigo A. Moreira *a, Mateusz Chwastyk b, Joseph L. Baker c, Horacio V. Guzman d and Adolfo B. Poma *a
aBiosystems and Soft Matter divison, Institute of Fundamental Technological Research, Polish Academy of Sciences, Pawińskiego 5B, 02-106 Warsaw, Poland. E-mail: rams@ippt.pan.pl; apoma@ippt.pan.pl
bInstitute of Physics, Polish Academy of Sciences, al. Lotników 32/46, 02-668 Warsaw, Poland
cDepartment of Chemistry, The College of New Jersey, 2000 Pennington Road, Ewing, NJ 08628, USA
dDepartment of Theoretical Physics, Jožef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia

Received 23rd May 2020 , Accepted 22nd July 2020

First published on 23rd July 2020

We report on the novel observation about the gain in nanomechanical stability of the SARS-CoV-2 (CoV2) spike (S) protein in comparison with SARS-CoV from 2002 (CoV1). Our findings have several biological implications in the subfamily of coronaviruses, as they suggest that the receptor binding domain (RBD) (∼200 amino acids) plays a fundamental role as a damping element of the massive viral particle's motion prior to cell-recognition, while also facilitating viral attachment, fusion and entry. The mechanical stability via pulling of the RBD is 250 pN and 200 pN for CoV2 and CoV1 respectively, and the additional stability observed for CoV2 (∼50 pN) might play a role in the increasing spread of COVID-19.


Since the recent outbreak of the 2019 novel coronavirus (CoV2) and the fast spread of the disease (COVID-19) around the globe, a rapid and very well-coordinated scientific research machinery has been put in place all over the world. In the past 5 months several scientific groups have pursued a comprehensive structural characterization of the main protein components of CoV2 to fight against the disease. Importantly, the four main structural proteins are well-resolved at atomic scale, including the spike (S), envelope (E), membrane (M) and nucleocapsid (N) proteins.1–3 Each component plays a crucial role in cell-recognition (S), generating ionic channel (E), defining the shape of the viral envelope (M), and binding the positive-stranded RNA that is made up of about 30[thin space (1/6-em)]000 nucleotides (N). These components are either in contact with the viral membrane (S, E and M) or in the inner part (N). In order to fight against CoV2 we need to determine the molecular weaknesses in the structure or along the processes that involve CoV2 proteins. In a recent article by Wang et al.,4 the high affinity of CoV2 to the human angiotensin-converting enzyme (ACE2), which is an enzyme attached to the outer membrane surface, compared to CoV1 was shown in vitro (94.6 nM vs. 408.7 nM for CoV2 and CoV1, respectively). Such a difference in KD has been suggested to be the reason for high spreading of the disease. However, Walls et al.3via in vivo studies has shown there is not a substantial change in KD and the same conclusion has been reached by in silico studies.5 It has also been suggested that the S1/S2 furin-like cleavage in the sequence Q677TNSPRRAR↓SV687 could enhance its transmissibility and enable fusion machinery in CoV2, and that the cleavage event may lead to a destabilized structure which can facilitate viral entry.6 Recently it was shown the possibility of a combined effect given by the lying-down conformation of RBD and the high ACE binding affinity as a mechanism to evade immune surveillance.7 These features are the cornerstone of traditional antiviral development for inhibition of CoV2 against cell recognition,8 while also enhancing the biophysical understanding of another component contained in the complete virus theory and simulation at different scales.9,10 In this study, we use a combination of tools from structural biology and molecular dynamics simulation to unveil the mechanical forces that the spike proteins may withstand before losing their function. It has been shown in bacteriophage T4 long tail fibers that thermal Brownian fluctuations of the virus can exert large forces11 such as 190 ± 70 pN at the level of the viral receptor. Here we determine a substantial mechanical stability of CoV2 and CoV1, which is about that of the Ig27 domain of titin,12,13Fmax(Ig27) = 200 pN. This information allows us to envision the molecular-scale scenario during the diffusion of viral particles under Brownian motion and upon their first encounter with the cell membrane receptor (ACE2). This mechanical picture has emerged from extensive molecular dynamics and coarse-grained simulations, and it enables the description of a dynamical process of recognition and confirms that the RBD makes a significant contribution to the mechanical stability of the full spike timer (see Fig. 1). We also interrogated the contribution of the current mutations on the dynamics of the CoV2 spike protein.
image file: d0nr03969a-f1.tif
Fig. 1 Schematic diagram of the nanomechanical stability that renders CoV2 more stable than CoV1 (k2 > k1). The spike protein for CoV2 and the human ACE2 are represented in blue and green color respectively.


COVID-19 spike glycoprotein + ACE2 models

We modeled the spike (S) protein from CoV2 and CoV1, both bound to ACE2 at the prefusion geometry and related structures (PDB IDs 6ACG, 6M0J, 6VSB, 6CRV and 2XY9), with one RBD up and two RBD at down positions and in complex with ACE2. The sequences that describe the spike proteins come from QIQ50172.1 and AAR86775.1 stored at GenBank database for CoV2 and CoV1, respectively. The standard Needleman–Wunsch algorithm was used as implemented by Chimera visualization software to align the sequences and the missing loops were modeled by homology using MODELLER14 and its energy minimization algorithms. The disulfide bonds were the ones described by the PDB files, including 3 for ACE2 and 14 and 15 per chain of the spike homotrimer for CoV2 and CoV1, respectively.

Contacts maps

The OV + rCSU contact maps used in this work was successfully used before to describe proteins.15,16 Here we have employed this methodology to scan through the MD trajectory and determine contacts between amino acids in a dynamics form. The overlap of enlarged spheres was used to define the OV contact map. The rCSU approach places the chemical character of each atom, and respective bonds, into categories and counts the number of stabilizing and destabilizing contacts per residue, defining a contact when both residues have a net stabilizing character. We implemented our own contact map software, as detailed by Wołek et al.17 The data set of all-atom MD trajectories, contact map methods, and scripts of CoV2 and CoV1 spikes can be accessed via the Zenodo repository.18

Molecular dynamics

All-atom simulations were carried out with Amber18,19 and system components (protein, ions, water) were modeled with the included FF14SB20 and TIP3P21 parameter sets. Energy minimization used CPU pmemd, while later simulation stages used GPU pmemd. CoV2 and CoV1 systems with one RBD up (with/without ACE2) were solvated in 12 Å octahedral water shells. Cysteine residues identified in the initial models as having a disulfide bond (DB) were bonded using tLeap. All simulations used 0.150 M NaCl. Hydrogen mass repartitioning was applied only to the protein to enable a 4 fs timestep.22 The SHAKE algorithm was applied to hydrogens, and a real-space cutoff of 8 Å was used. Periodic boundary conditions were applied and PME was used for long-range electrostatics. Minimization was by steepest descent (2000 steps) followed by conjugate gradient (3000 steps). Heating used two stages: (1) NVT heating from 0 K to 100 K (50 ps), and (2) NPT heating from 100 K to 300 K (100 ps). Restraints of 10 kcal mol−1 Å−2 were applied during minimization and heating to Cα atoms. During 6 ns of equilibration at 300 K Cα restraints were gradually reduced from 10 kcal mol−1 Å−2 to 0.1 kcal mol−1 Å−2. Finally, restraints were released and 320 ns unrestrained production simulations were carried out for CoV2 and CoV1 systems. Production simulations began from the final equilibrated snapshots, and five copies of each system were simulated. As unrestrained systems can freely rotate we monitored simulations for any close contacts and found that in one copy of the CoV1 simulation without ACE2 and one RBD up that a few contacts close to 8 Å occur near the end of the 320 ns between the RBD and a different subdomain of the spike complex in a periodic image. However this did not influence analyzed structural properties which is verified by comparing results across simulations. The Monte Carlo barostat was used to maintain pressure (1 atm), and the Langevin thermostat was used to maintain 300 K temperature (collision frequency 1 ps−1), as implemented in Amber18.19 In aggregate, 6.4 μs of all-atom simulation of systems ranging from 396[thin space (1/6-em)]147 to 879[thin space (1/6-em)]100 atoms was carried out for this work. Snapshots from the MD simulations can be found in the Zenodo repository.18

Nanomechanics of proteins

The nanomechanical simulations are based on the Gō-like model23,24 that has been used to sample conformational changes in proteins and calculate the elastic parameters under force deformation in single proteins, protein filaments, cellulose, and protein–protein, protein–polysaccharide and protein–lipid interfaces.15,25–28 At first we pull each chain of the trimer to identify the mechanostable protein domains. Our results for CoV2 show the RBD to be last domain to unfold. The S2 and NTD unfold first than RBD (FmaxS2 = 160 pN, FmaxNTD = 125 pN and FmaxRBD = 250 pN). In order to quantify the difference in mechanical stability of the RBD between CoV2 and CoV1 we perform pulling simulations of the RBD which is pulled along the end-to-end vector connecting the Cα-atoms from the N- and C-terminus and the reaction coordinate is the displacement of the pulling spring. Moreover, additional beads have been attached to those Cα-atoms with the spring constant being 0.1 kcal mol−1 Å−2, which is a typical value of the AFM cantilever stiffness in protein stretching studies. The type of coarse-grained methodology employed here offers the flexibility to use much lower pulling speeds than in all-atom MD simulations and keep it applied for a longer time. Each system was pulled over the course of 25 × 107 ps with a velocity of 10−6 Å ps−1 which equals a total simulation time of 250 μs per pulling trajectory. Although this value is still far from the experimental values of cantilever velocities (10–8 Å ps−1 (ref. 29)), it represents a significant computational improvement to access the experimental time scale. In experiments, multiple proteins are linked sequentially, and one can observe a number of corresponding peaks, which signal the full unfolding of one protein module. Because of the space resolution, intermediate unfolding states are not detected in AFM experiments. However, in the case of our model we can access these intermediate states with a better resolution and assign to each of them a force peak. The largest of these force peaks, Fmax, defines the characteristic largest rupture force in the system before unfolding. We construct a set of 50 independent trajectories to assess the mechanostability of the RBD component.

Results and discussion

Our current understanding of both CoV1 and CoV2 has been centered in the structural analysis of the spike (S) glycoprotein. Recently, the structure of the CoV2 S glycoprotein was determined by cryo-EM at high resolution (in the range of 2.8–3.5 Å) and it has been characterized in up (PDB ID: 6VYB3 and 6VSB1 entry) and down (PDB ID: 6VXX3) conformations. The community is still debating the link between the high spread of COVID-19 and the 30% difference (due to mutations) in CoV2 S protein compared to CoV1, and also whether the concurrent existence of both up and down conformations in vivo could be a mechanism for enhancing receptor recognition and subsequently facilitating infection. Only the RBD in the up conformation has been found in close contact with ACE2, indicating the important role of this conformation. In this work we employ extensive molecular dynamics simulations to characterize the molecular features which can differentiate both the CoV1 and CoV2 spikes, and in particular we examine the conformational space of the RBD which is responsible for binding ACE2. A process mostly dominated by the interplay between elastic, electrostatic interactions and environmental properties.5,30 Here, we highlight structural differences at the level of the most conserved part of the protein. In this regard we define the consensus residues as the set residues common in CoV2 and CoV1 after a sequence alignment and full reconstruction. Then we investigate the role of each single conserved residue in terms of the ‘native and created non-native contacts’ during the dynamics using a contact map determination for proteins.17 Our analysis indicates that there is a differential change in the set of contacts between consensus residues present most of the time in the MD simulations of the full spike that is primarily related to the RBD and its receptor binding module (RBM) subunit (see Fig. 2(a)). In this case the frequency fres of contacts was defined by fres = Cres/N, where N is the total number of frames used in the analysis (7500), and Cres counts the number of contacts computed per residue. We show in Fig. 2 the per residue signed difference between the frequencies from CoV2 and CoV1, dfres = (fCoV2resfCoV1res), such that a positive number indicates that CoV2 has effectively more contacts than CoV1, while a negative number indicates that CoV1 has more contacts. In Fig. 2(b) we depict the additional strong set of consensus contacts present in the conserved part of the protein. This result shows that the CoV2 RBD is structurally more stable compared to CoV1 and this is further supported by the smaller deviation of the root-mean-square fluctuations (RMSF) as we show in Fig. 2.
image file: d0nr03969a-f2.tif
Fig. 2 Panel (a) shows for CoV2 and CoV1 with ACE2 (similar results w/o ACE2) the difference between frequencies dfres = (fCoV2resfCoV1res) of chains from CoV2 and CoV1 in contact to ACE2 at each consensus residue, shown in blue color. Panel (b) shows the structure of the mechanostable protein domain in the spike (i.e. RBD) based on the computational nanomechanics, where red (blue) residues represent positive (negative) values of dfres at consensus residues and lines the additional set of high-frequency contacts of CoV2 not present in CoV1. On the right side is depicted the reduced flexibility as gauged by the RMSF all over the RBD sequence both in the presence of ACE2 (dashed line) and without ACE2 (solid line), which agrees with a larger number of consensus contacts, as shown between brackets in panel (a).

The observed differences in contacts strongly suggest a larger stability of the CoV2 RBD compared to the CoV1 RBD. To further demonstrate the mechanical stability of the RBD, we employed a validated structure-based modelling approach.15,28,29,31,32 Our computational structure-based study (see Fig. 3) shows that the RBD is a mechanically stable component in the spike (see next section) and also identifies the structural motif (see Fig. 3) that helps to hold together the structure before the largest force of rupture of intermolecular interactions (Fmax = 250 ± 11 pN for CoV2 and Fmax = 200 ± 13 pN for CoV1). The observed difference of ≈50 pN is the equivalent of non-invasive AFM indentation of proteins33,34 and it suggests that random mutations of the RBD have led to increased stability of the CoV2 RBD compared to the CoV1 RBD. Some deviation from the mean values of our result via Single Molecule Force Spectroscopy (SMFS) are expected, but the conclusive differential effect as this point is mostly due to the structural differences.

image file: d0nr03969a-f3.tif
Fig. 3 Pulling simulation of the RBD in up conformation. Top panel shows snaphots of the deformation process at cantilever distance equal to: (a) d = 0, (b) d = 100 and (c) d = 105 and direction of pulling (arrows). Panel (d) shows the pulling force and cantilever distance in simulation for CoV2 (blue) and CoV1 (red). The meaning of Fmax or rupture force is associated with three protein segments highlighted in red, blue and pink color in panel (a) which also lists 11 native contacts associated with Fmax in CoV2. Four disulphide bonds are represented as red sticks.

We also identify the amino acids in CoV2 that contribute to the largest stability in the RDB which turns out to be the same for CoV1. The protein segments involved simultaneously in the maximum force (see Fig. 3) are (A348–A352), (F400–R403)and (N450–R454) which corresponds to a hairpin loop that couples two beta strands together (β4 and β5). The total number of contacts that stabilize those regions are 11 interactions or native contacts which are mostly hydrophobic in character (see Fig. 3(a)). These structural elements are in close contact with loops that interact directly with ACE2. Based on this analysis we would suggest to target new experiments and molecular modelling in antiviral drug design that may disrupt those interactions in the RBD and as a consequence perhaps destabilize the process of cellular recognition. This result highlights the need to launch new rapid research not only in the development of antibodies, entry inhibitors, and antivirals as mechanism of action, but also in new therapies aimed at destabilizing certain key contacts responsible for the increased stability. Also, we expect to motivate nanomechanical studies via single molecule force spectroscopy35,36 of the spike proteins with focus on the RBD in coronavirus system as well as other experiments which can be relevant to elucidate other weaknesses in the protein structure of CoV2 and connect this information with its biological role during the cell recognition process.

Author contribution

A. B. P and R. A. M designed the research; R. A. M., M. C., H. V. G., J. L. B., and A. B. P performed research; R. A. M., H. V. G., J. L. B., H. V. G., M. C. and A. B. P. analyzed data; and R. A. M, H. V. G, J. L. B and A. B. P wrote the paper; and A. B. P supervised the research.

Conflicts of interest

There are no conflicts to declare.


A. B. P. and R. A. M. thank the financial support from the National Science Centre, Poland, under grant No. 2017/26/D/NZ1/00466. H. V. G thanks the financial support by the Slovenian Research Agency (Funding No. P1-0055). J. L. B. acknowledges use of the ELSA high performance computing cluster at The College of New Jersey. This cluster is funded in part by the National Science Foundation under grant numbers OAC-1826915 and OAC-1828163. M. C. has received support from the National Science Centre (NCN), Poland, under grant No. 2018/31/B/NZ1/00047. The authors gratefully acknowledge the computing provided by the Jülich Supercomputing Centre on the supercomputer JURECA at Forschungszentrum Jülich and additional computer resources were supported by the PL-GRID infrastructure. R. A. M. thanks for source codes gently provided by Marek Cieplak.


  1. D. Wrapp, N. Wang, K. S. Corbett, J. A. Goldsmith, C.-L. Hsieh, O. Abiona, B. S. Graham and J. S. McLellan, Science, 2020, 367, 1260–1263 CrossRef CAS .
  2. R. Yan, Y. Zhang, Y. Li, L. Xia, Y. Guo and Q. Zhou, Science, 2020, 367, 1444–1448 CrossRef CAS .
  3. A. C. Walls, Y.-J. Park, M. A. Tortorici, A. Wall, A. T. McGuire and D. Veesler, Cell, 2020, 180, 281–292 CrossRef .
  4. Q. Wang, Y. Zhang, L. Wu, S. Niu, C. Song, Z. Zhang, G. Lu, C. Qiao, Y. Hu and K.-Y. Yuen, et al. , Cell, 2020, 181, 894–904 CrossRef CAS .
  5. C. C. Giron, A. Laaksonen and F. L. B. da Silva, Virus Res., 2020, 198021 CrossRef .
  6. B. Coutard, C. Valle, X. de Lamballerie, B. Canard, N. Seidah and E. Decroly, Antiviral Res., 2020, 176, 104742 CrossRef CAS .
  7. J. Shang, Y. Wan, C. Luo, G. Ye, Q. Geng, A. Auerbach and F. Li, Proc. Natl. Acad. Sci. U. S. A., 2020, 117, 11727–11734 CrossRef CAS .
  8. S. Boopathi, A. B. Poma and P. Kolandaivel, J. Biomol. Struct. Dyn., 2020, 1–14 CrossRef .
  9. M. Martínez, C. D. Cooper, A. B. Poma and H. V. Guzman, J. Chem. Inf. Model., 2019, 60, 974–981 CrossRef .
  10. R. Zandi, B. Dragnea, A. Travesset and R. Podgornik, Phys. Rep., 2020, 847, 1–102 CrossRef CAS .
  11. P. Ares, C. Garcia-Doval, A. Llauro, J. Gomez-Herrero, M. Van Raaij and P. De Pablo, Phys. Rev. E: Stat., Nonlinear, Soft Matter Phys., 2014, 89, 052710 CrossRef CAS .
  12. M. Carrion-Vazquez, A. F. Oberhauser, S. B. Fowler, P. E. Marszalek, S. E. Broedel, J. Clarke and J. M. Fernandez, Proc. Natl. Acad. Sci. U. S. A., 1999, 96, 3694–3699 CrossRef CAS .
  13. H. Lu, B. Isralewitz, A. Krammer, V. Vogel and K. Schulten, Biophys. J., 1998, 75, 662–671 CrossRef CAS .
  14. A. Fiser and A. Šali, Methods Enzymol., Elsevier, 2003, vol. 374, pp. 461–491 Search PubMed .
  15. A. B. Poma, M. Cieplak and P. E. Theodorakis, J. Chem. Theory Comput., 2017, 13, 1366–1374 CrossRef CAS .
  16. M. Chwastyk, A. P. Bernaola and M. Cieplak, Phys. Biol., 2015, 12, 046002 CrossRef .
  17. K. Wołek, À. Gómez-Sicilia and M. Cieplak, J. Chem. Phys., 2015, 143, 243105 CrossRef .
  18. R. A. Moreira, M. Chwastyk, J. L. Baker, H. V. Guzman and A. B. Poma, Zenodo, 2020,  DOI:10.5281/zenodo.3817447 .
  19. D. A. Case, I. Y. Ben-Shalom, S. R. Brozell, D. S. Cerutti, T. E. Cheatham, III, V. W. D. Cruzeiro, T. A. Darden, R. E. Duke, D. Ghoreishi, M. K. Gilson, H. Gohlke, A. W. Goetz, D. Greene, R Harris, N. Homeyer, Y. Huang, S. Izadi, A. Kovalenko, T. Kurtzman, T. S. Lee, S. LeGrand, P. Li, C. Lin, J. Liu, T. Luchko, R. Luo, D. J. Mermelstein, K. M. Merz, Y. Miao, G. Monard, C. Nguyen, H. Nguyen, I. Omelyan, A. Onufriev, F. Pan, R. Qi, D. R. Roe, A. Roitberg, C. Sagui, S. Schott-Verdugo, J. Shen, C. L. Simmerling, J. Smith, R. SalomonFerrer, J. Swails, R. C. Walker, J. Wang, H. Wei, R. M. Wolf, X. Wu, L. Xiao, D. M. York and P. A. Kollman, AMBER 2018, University of California, San Francisco, 2018 Search PubMed .
  20. J. A. Maier, C. Martinez, K. Kasavajhala, L. Wickstrom, K. E. Hauser and C. Simmerling, J. Chem. Theory Comput., 2015, 11, 3696–3713 CrossRef CAS .
  21. W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey and M. L. Klein, J. Chem. Phys., 1983, 79, 926–935 CrossRef CAS .
  22. C. W. Hopkins, S. Le Grand, R. C. Walker and A. E. Roitberg, J. Chem. Theory Comput., 2015, 11, 1864–1874 CrossRef CAS .
  23. J. I. Sułkowska and M. Cieplak, Biophys. J., 2008, 95, 3174–3191 CrossRef .
  24. C. Clementi, H. Nymeyer and J. N. Onuchic, J. Mol. Biol., 2000, 298, 937–953 CrossRef CAS .
  25. M. Sikora, J. I. Sułkowska, B. S. Witkowski and M. Cieplak, Nucleic Acids Res., 2010, 39, D443–D450 CrossRef .
  26. M. Sikora, J. I. Sułkowska and M. Cieplak, PLoS Comput. Biol., 2009, 5, e1000547 CrossRef .
  27. A. B. Poma, M. S. Li and P. E. Theodorakis, Phys. Chem. Chem. Phys., 2018, 20, 17020–17028 RSC .
  28. A. B. Poma, H. V. Guzman, M. S. Li and P. E. Theodorakis, Beilstein J. Nanotechnol., 2019, 10, 500–513 CrossRef CAS .
  29. A. Valbuena, J. Oroz, R. Hervás, A. M. Vera, D. Rodríguez, M. Menéndez, J. I. Sulkowska, M. Cieplak and M. Carrión-Vázquez, Proc. Natl. Acad. Sci. U. S. A., 2009, 106, 13791–13796 CrossRef CAS .
  30. M. Hernando-Pérez, A. Cartagena-Rivera, A. L. Božič, P. J. Carrillo, C. San Martín, M. G. Mateu, A. Raman, R. Podgornik and P. De Pablo, Nanoscale, 2015, 7, 17289–17298 RSC .
  31. S. Senapati, A. B. Poma, M. Cieplak, S. Filipek and P. S.-H. Park, Anal. Chem., 2019, 91, 7226–7235 CrossRef CAS .
  32. A. B. Poma, M. Chwastyk and M. Cieplak, Phys. Chem. Chem. Phys., 2017, 19, 28195–28206 RSC .
  33. H. V. Guzman, A. P. Perrino and R. Garcia, ACS Nano, 2013, 7, 3198–3204 CrossRef CAS .
  34. H. V. Guzman, Beilstein J. Nanotechnol., 2017, 8, 968–974 CrossRef CAS .
  35. M. Krieg, G. Fläschner, D. Alsteens, B. M. Gaub, W. H. Roos, G. J. Wuite, H. E. Gaub, C. Gerber, Y. F. Dufrêne and D. J. Müller, Nat. Rev. Phys., 2019, 1, 41–57 CrossRef .
  36. B. Yang, Z. Liu, H. Liu and M. A. Nash, Front. Mol. Biosci., 2020, 7, 85 CrossRef .

This journal is © The Royal Society of Chemistry 2020