Effects of common mutations in the SARS-CoV-2 Spike RBD domain and its ligand the human ACE2 receptor on binding affinity and kinetics

2 The interaction between the SARS-CoV-2 virus Spike protein receptor binding domain (RBD) 3 and the ACE2 cell surface protein is required for viral infection of cells. Mutations in the RBD 4 domain are present in SARS-CoV-2 variants of concern that have emerged independently 5 worldwide. For example, the more transmissible B.1.1.7 lineage has a mutation (N501Y) in 6 its Spike RBD domain that enhances binding to ACE2. There are also ACE2 alleles in humans 7 with mutations in the RBD binding site. Here we perform a detailed affinity and kinetics 8 analysis of the effect of five common RBD mutations (K417N, K417T, N501Y, E484K and 9 S477N) and two common ACE2 mutations (S19P and K26R) on the RBD/ACE2 interaction. 10 We analysed the effects of individual RBD mutations, and combinations found in new SARS-11 CoV-2 variants first identified in the UK (B.1.1.7), South Africa (B.1.351) and Brazil (P1). Most 12 of these mutations increased the affinity of the RBD/ACE2 interaction. The exceptions were 13 mutations K417N/T, which decreased the affinity. Taken together with other studies, our 14 results suggest that the N501Y and S477N mutations primarily enhance transmission, the 15 K417N/T mutations facilitate immune escape, and the E484K mutation facilitates both 16 transmission and immune escape.


Introduction
Since its identification in 2019, the second coronavirus able to induce a severe acute 20 respiratory syndrome in humans, SARS-CoV-2, has resulted in the most severe global 21 pandemic in 100 years. To date more than 135 million people have been infected, resulting 22 in the deaths from the resulting disease, COVID-19, of more than 3 million people ("WHO 23 Coronavirus  Dashboard," 2021), and measures introduced to control spread 24 have had harmful social and economic impacts. Fortunately, effective vaccines have been 25 developed, and a global vaccination programme is underway (Mahase, 2021). New  CoV-2 variants of concern are emerging that are making containment of the pandemic more 27 difficult, by increasing transmissivity of the virus (Davies and Edmunds, 2021;Korber et al., 28 2020;Volz et al., 2021aVolz et al., , 2021bWashington et al., 2021) and/or its resistance to protective 29 immunity induced by previous infection or vaccines (Darby and Hiscox, 2021;Dejnirattisai et 30 al., 2021;Garcia-Beltran et al., 2021;Madhi et al., 2021aMadhi et al., , 2021bMahase, 2021). (Volz et al.,31 2021a, 2021b) 32 The SARS-CoV-2 virus enters cells following an interaction between the Spike (S) protein on 33 its surface with angiotensin-converting enzyme 2 (ACE2) on cell surfaces (V'kovski et al., 34 2021). The receptor binding domain (RBD) of the Spike protein binds the membrane-distal 35 portion of the ACE2 protein. The S protein forms a homotrimer, which is cleaved shortly 36 after synthesis into two fragments that remain associated non-covalently: S1, which 37 contains the RBD, and S2, which mediates membrane fusion following the binding of Spike 38 to ACE2 (V'kovski et al., 2021). During the pandemic mutations have appeared in the Spike 39 protein that apparently increase transmissivity (Davies and Edmunds, 2021;Korber et al., 40 2020;Volz et al., 2021aVolz et al., , 2021bWashington et al., 2021). One that emerged early in Europe, 41 D614G, and quickly became dominant globally (Korber et al., 2020), increases the density of 42 intact Spike trimer on the virus surface by preventing premature dissociation of S1 from S2 43 following cleavage (Zhang et al., 2021. A later mutant, N501Y, which has appeared in 44 multiple lineages, lies within the RBD domain, and increases its affinity for ACE2 (Starr et al., 45 2020;Supasa et al., 2021). These findings suggest that mutations that directly or indirectly 46 enhance Spike binding to ACE2 will increase transmissivity. 47 the effects of the mutations K417T, K417N and E484K on kon. K417 forms a salt bridge with 142 D30 on ACE2 (Lan et al., 2020) while E484 is ~9 Å from E75 on ACE2 (Lan et al., 2020). Thus 143 the mutations K417N/T and E484K would decrease and increase, respectively, long-range 144 electrostatic forces that may accelerate association (Schreiber and Fersht, 1996). 145 We also examined the effect on ACE2 binding of combinations of RBD mutations, including 146 combinations present in VOC-202102-02, a subset of the B.1.1.7 lineage (N501Y) with the 147 E484K mutation("SARS-CoV-2 Variants of concern and variants under investigation -148 GOV.UK," 2021), and the B.1.351 and P.1 variants (Fig. 3C, Table 1). In the case of VOC-149 202102-02, the addition of the E484K mutation to N501Y further increased the affinity, to 150 ~15 fold higher than WT RBD (KD ~5 nM), by further increasing the kon. Because the higher 151 kon could result in mass transfer limiting binding, we confirmed that the kinetic 152 measurement for this variant was not substantially affected by varying levels of 153 immobilization (Fig. S4). The affinity of the B.1.351 (K417N/ E484K/N501Y) and P.1 154 (K417T/E484K/N501Y) RBD variants for ACE2 increased by 3.7 and 5.3 fold, respectively, 155 relative to wild type RBD, by both increasing the kon and decreasing the koff rate constants. 156 We next examined whether the effects of the mutations were additive, as is typically the 157 case for multiple mutations at protein/protein interfaces (Wells, 1990). To do this we 158 converted the changes in KD to changes in binding energy (G, Table 2) and examined 159 whether the G measured for RBD variants with multiple mutations was equal to the sum 160 of the G values measured for the individual RBD mutants. This was indeed the case ( Fig.  161 3D), indicating that the effects on each mutation are independent. This is consistent with 162 them being spaced well apart within the interface (Fig. 1C), and validates the accuracy of 163 the affinity measurements. 164

The effects of ACE2 mutations 165
We next examined the effects of mutations of ACE2 (S19P and K26R) on binding to both wild 166 type and common variants of RBD ( Fig. 4 and Table 1). Both S19P and K26R increased the 167 affinity of WT RBD binding by ~3.7 and ~2.4 fold (Fig. 4A). These increases in affinity were 168 the result of both increases in the kon and decreases in the koff. 169 Finally, we looked for interactions between RBD and ACE2 mutations by measuring the 170 effects of the ACE2 mutations on binding to all mutant forms of RBD (Table 1). After 171 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 converting changes in KD to G (Table 2) we examined whether G measured for a given 172 ACE2 variant/RBD variant interaction was equal to the sum of the G measured for ACE2 173 variant/RBD WT and ACE WT/RBD variant interactions. This is depicted as the difference 174 between the measured and predicted G for interactions between ACE2 and RBD variants 175 (G in Figs. 4B and C). In most cases G values were close to zero, indicating that the 176 effects of these mutations were largely independent. The one exception was the 177 combination of ACE2 S19P and RBD S477N variants, where the measured value was 178 significantly lower than the predicted value ( Fig. 4B), indicating that these mutations were 179 not independent. This is consistent with the fact that the ACE2 residue S19 is adjacent to 180 RBD residue S477 in the contact interface (Fig. 1C). An important consequence of this is that 181 the S477N mutation increased the affinity of RBD for ACE2 WT but decreased its affinity for 182 ACE2 S19P. 183

184
While our finding that the SARS-CoV-2 RBD binds ACE2 with an affinity of KD 74 nM at 37°C 185 is consistent with previous studies (KD 11 to 133 nM) (Lei et al., 2020;Shang et al., 2020;186 Supasa et al., 2021;Wrapp et al., 2020;Zhang et al., 2021Zhang et al., , 2020, the rate constants that we 187 measured (kon 0.9 M -1 .s -1 and koff 0.067 s -1 ) were more than 3 fold faster than all previous 188 reports. One likely reason for this is that previous measurements were performed at a lower 189 temperature, which almost always decreases rate constants. While one study stated that 190 binding constants were measured at 25°C , most studies did not report 191 the temperature, suggesting that they were performed at room temperature or the 192 standard instrument temperature (20-25°C). A second likely reason is that previous kinetic 193 studies were performed under conditions in which the rate of diffusion of soluble molecule 194 to the sensor surface limits the association rate, and rebinding of dissociated molecules to 195 the surface reduces the measured dissociation rate. These are known pitfalls of both 196 techniques used in these studies, surface plasmon resonance (Myszka, 1997) and bilayer 197 interferometry (Abdiche et al., 2008). In the present study we avoided these issues by 198 immobilizing a very low level of ligand on the sensor surface. A third possible reason is that 199 the proteins were aggregated, which can cause problems even when aggregates are a very 200 minor contaminant (van der Merwe and Barclay, 1996). The presence of aggregates results 201 in complex binding kinetics, which can be excluded if the simple 1:1 Langmuir binding model 202 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 fits the kinetic data. While this was demonstrated in the present study, and some previous 203 studies (Shang et al., 2020;Wrapp et al., 2020;Zhang et al., 2021), such fits were not shown 204 in all studies, one of which reported more than 20 fold slower kinetics than reported here 205 (Lei et al., 2020;Supasa et al., 2021). 206 The RBD mutants that we selected for analysis have all emerged independently and become 207 dominant in a region at least once in different lineages, suggesting that they provide a 208 selective advantage. Our finding that the N501Y, E484K, and S477N all increase the binding 209 affinity of RBD for ACE2 raises the question as to whether this contributed to their selection. which has the N501Y mutation (Volz et al., 2021b;Washington et al., 2021). Finally, a SARS-215 CoV-2 variant with the Spike mutation D614G, which increases its activity by stabilizing it 216 following furin cleavage (Zhang et al., 2021, rapidly became dominant globally after it 217 emerged (Korber et al., 2020;Volz et al., 2021a). Taken together, these findings suggest that 218 the WT Spike/ACE2 interaction is limiting for transmission, and that mutations which 219 enhance it, including the N501Y, E484K, and S477N mutations, would provide a selective 220 advantage by increasing transmissibility. This raises two questions. Firstly, will other RBD 221 mutations appear in SARS-CoV-2 which further enhance transmission? This seems likely, 222 given that a large number of RBD mutations have been identified that increase the 223 RBD/ACE2 affinity (Starr et al., 2020;Zahradník et al., 2021). Secondly, will combinations of 224 existing mutations be selected because they further increasing the affinity? While the 225 appearance E484K together with the N501Y in three lineages (B.1.1.7, B.1.351 and P.1) 226 supports this, it is also possible that E484K was selected because it disrupts antibody 227 neutralization, as discussed below. 228 Studies of other enveloped viruses, including SARS-CoV-2, suggest that increases in affinity 229 of viral fusion ligands for their cellular receptors can increase cell infection and disease 230 severity (Hasegawa et al., 2007;Li et al., 2005). One study found that increasing this affinity 231 enabled the virus to infect cells with lower receptor surface density (Hasegawa et al., 2007). 232 It follows that increases in affinity could increase the number of host tissues infected, which 233 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 could increase the severity of disease (Cao and Li, 2020) and/or increase the viral load in the 234 upper respiratory tract el (Hoffmann et al., 2020;Wölfel et al., 2020), thereby increasing 235 spread. 236 Another mechanism by which mutations of RBD could provide a selective advantage is 237 through evasion of immune responses. This is supported by the observation that 238 neutralizing antibodies present in those infected by or vaccinated against SARS-CoV-2 239 primarily target the RBD domain (Garcia-Beltran et al., 2021;Greaney et al., 2021a;Rogers 240 et al., 2020). Furthermore, two variants with RBD mutations that abrogate antibody 241 neutralization, B.1.351 and P1, became dominant in regions with very high levels of prior 242 SARS-CoV-2 infection (Cele et al., 2021;Hoffmann et al., 2021;243 Sabino et al., 2021;Tegally et al., 2021;. Both lineages include the 244 N501Y mutation, but this appears to have modest effects on antibody neutralization 245 (Greaney et al., 2021a(Greaney et al., , 2021b. In contrast, the E484K mutation, also present in both 246 lineages, potently disrupts antibody neutralization (Greaney et al., 2021a(Greaney et al., , 2021b. Our 247 finding that the K417N/T mutants present in B.1.351/P.1 lineages decrease the affinity of 248 RBD for ACE2 suggests that they were selected because they facilitate immune escape. 249 Indeed, mutations of K417 can block antibody neutralization, albeit less effectively than 250 E484K (Greaney et al., 2021a(Greaney et al., , 2021b. It is notable that these affinity-251 reducing K417N/T mutants have only emerged together with mutants (N501Y and E484K) 252 that increase the affinity of RBD for ACE2, suggesting a cooperative effect between 253 mutations that enhance immune escape and mutations that increase affinity. 254 The effect of the increased affinity for SARS-CoV-2 Spike RBD of the K26R and S19P ACE2 255 mutants are less clear. The evidence summarised above that WT RBD/ACE2 binding is 256 limiting for SARS-CoV-2 transmission, suggest that carriers of these ACE2 variants will be at 257 greater risk of infection and/or severe disease. However, in contrast to SARS-CoV-2 RBD 258 mutations, the effects of ACE2 variants are primarily relevant to the carriers of these 259 mutations. A preliminary analysis (MacGowan et al., 2021) suggests that the carriers of the 260 K26R ACE allele might be at increased risk of severe disease, but the findings did not reach 261 statistical significance, and further studies are required. 262 The interaction that we identified between the RBD S477N and ACE2 S19P mutants 263 highlights the importance of considering variation in the host population when studying the 264 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ; https://doi.org/10.1101/2021.05.18.444646 doi: bioRxiv preprint evolution of viral variants. In this case, the opposite effect of the RBD S477N mutation on its 265 affinity for ACE2 S19P (decreased) compared with ACE2 WT (increased), suggests that this 266 RBD variant may have a selective disadvantage amongst carriers of the ACE2 S19P variant, in 267 contrast to those with ACE2 WT, where it appears to be advantageous. However, the low 268 frequency of this variant means that this is unlikely to be important at a population level 269 and will be difficult to detect. 270 It is noteworthy that the two most common ACE2 variants are in positions on ACE2 with no 271 known functional activity. This raises the question as to whether these mutations are a 272 remnant of historic adaption to pathogens that utilised this portion of ACE2. The fact that 273 ACE2 S19P mutation is largely confined to African/African-American populations, suggests 274 that it is more recent than K26R and/or selected by pathogen (s)  The carboxy-terminal end has a biotin acceptor peptide (underlined) followed by an 291 oligohistidine tag. 292 The WT RBD construct, which was kindly provided by Quentin Sattentau (Sir William Dunn 293 School of Pathology), encoded the following protein: 294 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is with 8% CO2 on a shaking platform at 130 rpm. Cells were passaged every 2-3 days with the 305 suspension volume always kept below 33.3% of the total flask capacity. The cell density was 306 kept between 0.5 and 2 million per ml. Before transfection cells were counted to check cell 307 viability was above 95% and the density adjusted to 1.0 million per ml. For 100 ml 308 transfection, 100 µl FreeStyle™ MAX Reagent (16447100)  (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ; https://doi.org/10.1101/2021.05.18.444646 doi: bioRxiv preprint concentrated, and buffer exchanged into size exclusion buffer (25 mM NaH2PO4, 150 mM 325 NaCl at pH 7.5) using a protein concentrator with a 10,000 molecular weight cut-off. The 326 protein was concentrated down to less than 500 μl before loading onto a Superdex 200 327 10/300 GL size exclusion column (Fig. S2). Fractions corresponding to the desired peak were 328 pooled and frozen at -80 °C. Samples from all observed peaks were analysed on an SDS-329 PAGE gel (Fig. S2). 330 Surface plasmon resonance (SPR) 331 RBD binding to ACE2 was analysed on a Biacore T200 instrument (GE Healthcare Life 332 Sciences) at 37°C and a flow rate of 30 µl/min. Running buffer was HBS-EP (BR100669). 333 Streptavidin was coupled to a CM5 sensor chip (29149603)  Such double-referencing improves data quality when binding responses are low as needed 345 to obtain accurate kinetic data (Myszka, 1999). At the end of each experiment an ACE2-346 specific mouse monoclonal antibody (NOVUS Biologicals, AC384) was injected at 5 µg/ml for 347 10 minutes to confirm the presence and amount of immobilized ACE2. 348

Data analysis 349
Double referenced binding data was fitted using GraphPad Prism. The koff was determined 350 by fitting a mono-exponential decay curve to data from the dissociation phase of each 351 injection. The koff from four to six RBD injections was averaged to give a value for the koff 352 (Fig. S3A). The k on was determined by first fitting a mono-exponential association curve to 353 data from the association phase, yielding the kobs. The kon was be determined by plotting the 354 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ; https://doi.org/10.1101/2021.05.18.444646 doi: bioRxiv preprint kobs vs the concentration of RBD and performing a linear fit of the equation kobs = kon*[RBD] 355 + koff to this data (Fig. S3B), using the koff determined as above to constrain the fit. 356 The KD was either calculated (calculated KD = koff/kon) or measured directly (equilibrium KD) 357 as follows. Equilibrium binding levels at a given [RBD] were determined from the fit above of 358 the mono-exponential association phase model to the association phase data. These 359 equilibrium binding levels were plotted against [RBD] and a fit of the simple 1:1 Langmuir 360 binding model to this data was used to determine the equilibrium KD (Fig. 2D). 361 ΔG for each affinity measurement was calculated the relationship G =R*T*lnKD, where R = 362 1.987 cal mol -1 K -1 , T = 310.18 K, and K D is in units M. ΔΔG values (Table 2  between the ACE2 and RBD mutants was calculates as G = measured G -predicted 367 G (Fig. 4B) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021     from the Nextstrain Global (Hadfield et al., 2018) sample of SARS-CoV-2 sequences from the GISAID database (Shu and McCauley, 2017) (Accessed 15th April 2021, N = 4,017). (B) Alignment illustrating the Spike residues that differ between SARS-CoV-2 variants, with the RBD mutants boxed. The variants are labelled with their clade designation from Nextstrain (Hadfield et al., 2018) and/or PANGO lineage (Rambaut et al., 2020) where relevant. The RBD mutations were collated from CoVariants CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101 doi: bioRxiv preprint (Hodcroft, 2021) and Nextstrain. (C) The structure of human ACE2 (green) in complex with SARS-CoV-2 Spike RBD (cyan). The area enclosed by the box is shown enlarged on the right, with the residues mutated in this study labelled. Drawn using UCSF Chimera (Pettersen et al., 2004) using coordinates from PDB 6m0j (Lan et al., 2020).
. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021   . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ; https://doi.org/10.1101/2021.05.18.444646 doi: bioRxiv preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ; https://doi.org/10.1101/2021.05.18.444646 doi: bioRxiv preprint right panels show an expanded view of the dissociation phase. The blue lines show fits used for determining the kon and koff. (C) The fold change relative to WT RBD of the calculated KD, kon, and koff for the indicated RBD variants binding to immobilised WT ACE2 (Error bars show SD, n = 3). Representative sensorgrams from all mutants shown in Fig. S5, and the mean values from multiple repeats are in Table 1. (D) The blue lines show the measured ΔΔG for indicated RBD variants. The red lines show the predicted ΔΔG for the RBD variants with multiple mutations, which were calculated by adding ΔΔG values for single mutation variants (Error bars show SD, n = 3).
. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021

Figure 4. Effect of mutations in ACE2
(A) The fold change relative to WT ACE2 of the calculated KD, kon, and koff for the interaction of WT RBD and the indicated ACE2 variants (Error bars show SD, n = 3). (B-C) Show the difference (G) between the measured and predicted G for S19P (B) and K26R (C) ACE2 variants binding to the indicated RBD variants, calculated from data in Table 2. The predicted G values for each variant RBD/variant ACE2 interaction were calculated from the sum of the G for the ACE2 variant binding WT RBD and the G for the RBD variant binding WT ACE2 (Table 2).   . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 Figure S1. Emergence of the same RBD mutations in multiple SAR2-CoV-2 clades. The figure highlights the SARS-CoV-2 clades containing RBD mutations investigated in this study. The phylogenetic trees were constructed as in Fig. 1A from SARS-CoV-2 sequences accessed on the 22nd April 2021 (N = 3,914). (A) N501Y has emerged independently of the three clades 501Y.V1, 501Y.V2, and 501Y.V3. Mutation to T at this position has also occurred frequently. (B) E484K has also been observed independently of its main progenitor clades 501Y.V2 and 501Y.V3. E484Q and E484G have also been observed. (C) S477N has been observed beyond clades 20F and 20A.EU2. Mutations to I and R have also been occasionally observed at this position. (D) Mutations of K417 to N and T have been observed almost exclusively in the 20H.501Y.V2 and 20J.501Y.V3 clades. Figure S1 A B C D . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is  . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is   Figure S4 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021.    . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 24, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021