An ACE2 Triple Decoy that neutralizes SARS-CoV-2 shows enhanced affinity for virus variants

The SARS-CoV-2 variants replacing the first wave strain pose an increased threat by their potential ability to escape pre-existing humoral protection. An angiotensin converting enzyme 2 (ACE2) decoy that competes with endogenous ACE2 for binding of the SARS-CoV-2 spike receptor binding domain (S RBD) and inhibits infection may offer a therapeutic option with sustained efficacy against variants. Here, we used Molecular Dynamics (MD) simulation to predict ACE2 sequence substitutions that might increase its affinity for S RBD and screened candidate ACE2 decoys in vitro. The lead ACE2(T27Y/H34A)-IgG1FC fusion protein with enhanced S RBD affinity shows greater live SARS-CoV-2 virus neutralization capability than wild type ACE2. MD simulation was used to predict the effects of S RBD variant mutations on decoy affinity that was then confirmed by testing of an ACE2 Triple Decoy that included an additional enzyme activity-deactivating H374N substitution against mutated S RBD. The ACE2 Triple Decoy maintains high affinity for mutated S RBD, displays enhanced affinity for S RBD N501Y or L452R, and has the highest affinity for S RBD with both E484K and N501Y mutations, making it a viable therapeutic option for the prevention or treatment of SARS-CoV-2 infection with a high likelihood of efficacy against variants.

SARS-CoV-2 variants have rapidly swept the globe 1-3 and very recent investigations reveal that several of these variants have shown the ability to escape neutralization by convalescent antibodies in recovered COVID-19 patients 4-10 and recombinant neutralizing antibodies (nAbs) developed as therapeutics 11,12 . There are also fears that current vaccines may not be as effective against some of the variants and early evidence suggests that for some vaccines, this risk may exist 13,14 . The latter is a particular concern, as the massive vaccine efforts currently underway employ vaccines designed to elicit immune responses against first-wave sequence SARS-CoV-2 spike (S) protein and specifically the S receptor binding domain (S RBD) that binds to angiotensin-converting enzyme 2 (ACE2) on the surface of human cells in the airway and gut that initiates viral entry and infection [15][16][17][18] . While one response to the threat of loss of vaccine efficacy might be to continually re-design vaccines to target specific new variants, this would be an ongoing game of catch-up because it can be expected that further novel variants will emerge, particularly since several recent reports have shown that antibodies elicited by infection and vaccination act as evolutionary forces that result in the predominance of viral variants that escape these immune defenses 19,20 .
While efforts to adapt vaccines should be encouraged, in parallel, new therapeutic approaches to neutralize viral infection that are not undermined by the presence of mutations should be advanced.
To address the need for a therapeutic and potentially prophylactic approach that has a low likelihood of being adversely affected by variant mutations, we have designed and tested ACE2 'decoys' that leverage the binding of the S RBD to ACE2. This is an approach that is also being pursued by others using a variety of fusion proteins and delivery methods [21][22][23][24] . Our ACE2 decoys under development are recombinant ACE2-IgG 1 F C or -IgAF C

Results
Wild type (WT) ACE2-IgG 1 F C and ACE2-IgAF C decoys show high affinity for S RBD. In initial studies to design an ACE2 decoy, we determined the affinity of both recombinant wild type (WT) ACE2(WT)-IgG 1 F C and -IgAF C fusion proteins for binding to S RBD by Biolayer Interferometry (BLI) analysis. The ACE2(WT)-IgG 1 F C decoy (Fig. 1a) showed moderate affinity for S RBD in 1:1 binding with a coefficient of dissociation (K D ) of 21.40 nM and high binding with avidity with a K D of 0.762 nM (Fig. 1c,d, respectively). The ACE2(WT)-IgAF C dimeric fusion protein (Fig. 1b) demonstrated even higher binding (with avidity) affinity for S RBD with a K D of 0.166 nM (Fig. 1e). Additional BLI kinetics values are shown in Table S1 in the "Supplementary Information S1".
An ACE2 decoy expressing T27Y and H34A mutations shows enhanced affinity for S RBD and improved neutralization of live SARS-CoV-2. Based on MD simulation-based predictions of mutations that may confer enhanced binding affinity of ACE2 for S RBD, several ACE2 variants were tested for binding affinity as ACE2-IgG 1 F C fusion proteins. A tyrosine (Y) substitution for threonine (T) at residue 27 ( Fig. 1f) and an alanine (A) substitution for histidine (H) at residue 34 ( Fig. 1g) of ACE2 resulted in 3-5-fold increases in binding affinities (T27Y K D = 8.01 nM; H34A K D = 4.09 nM). Combination of the T27Y and H34A substitutions Biolayer Interferometry (BLI) of ACE2-IgG 1 F C , ACE2-IgAF C, and mutant decoy binding to the spike receptor binding domain; and live virus neutralization. The (a) ACE2-IgG 1 F C decoy and (b) dimeric ACE2-IgAF C decoy fused via a J-chain are shown. BLI kinetics analysis of (c) 1:1 binding and (d) binding with avidity for the ACE2-IgG 1 F C decoy; (e) BLI binding with avidity for the ACE2-IgAF C decoy; (f) ACE2(T27Y)-IgG 1 F C , (g) ACE2(H34A)-IgG 1 F C , and (h) ACE2(T27Y/H34A)-IgG 1 F C decoys are shown with K D values. (i) Virus neutralization as percent for concentrations of decoy is shown with IC50s for ACE2(WT)-IgG 1 F C and ACE2(T27Y/H34A)-IgG 1 F C . Negative controls: media and higG1 (human IgG1).
Addition of an H374N mutation inhibits ACE2 enzyme activity. In addition to enhanced affinity for competitive binding of S RBD, we wanted to inhibit the enzymatic activity of ACE2 36 . Angiotensin-converting enzyme 2 has an important role in homeostasis of the renin-angiotensin system 37-39 by cleavage of its substrate angiotensin 1-9 40 and its activity affects a variety of systems. Addition of enzymatically active recombinant ACE2 to the system presents a high risk of unwanted side effects and since S RBD binding, but not substrate cleavage activity, is the key function for the ACE2 decoy, we tested a variety of mutations predicted to inhibit ACE2 enzymatic activity with a low likelihood of affecting S RBD binding affinity.
ACE2 triple mutant decoys comprising the S RBD binding affinity-enhancing T27Y/H34A mutations and the enzymatic activity-inhibiting mutations were produced and binding affinity assessed. Of the triple mutants, those with either the R273K or H374N mutations showed the highest S RBD affinity (Table S3 in the "Supplementary Information S1").
The final ACE2 Triple Decoy chosen for further testing was ACE2 (T27Y/H34A/H374N)-IgG 1 F C due to its more favorable biophysical characteristics as compared to the R273K-containing triple mutant, including a lower propensity to aggregate and a higher titer ( Fig. S2 and Table S4 in the "Supplementary Information S1").  Table S5 in the "Supplementary Information S1".
The ACE2 Triple Decoy showed higher binding affinity to all S RBD sequences as compared to the ACE2(WT) decoy. As compared to the ACE2 Triple Decoy binding affinity for S RBD WT, affinities for S RBD E484K/N501Y, N501Y alone and L452R were higher; affinities for S RBD E484K, K417N/N501Y, N417N/E484K/N501Y, K417K/ www.nature.com/scientificreports/ E484K, and K417N were lower. Findings were similar with the wild type ACE2 decoy, with the highest affinity seen for E484K/N501Y and N501Y alone, and the lowest affinities for variants expressing K417N. N501Y and L452R showed ~ 2-3-fold increase in binding affinity for both wild type ACE2 decoy and ACE2 Triple Decoy. E484K alone did not affect binding affinity to ACE2. K417N weakened binding affinity for ACE2(WT) and the Triple Decoy, but affinity was restored when combined with N501Y. The E484K, K417N and N501Y mutations occur together in the B.1.351 strain, whereas L452R alone is found in B.1.427/B.1.429, therefore assessment of ACE2 WT binding to these variants as they occur in nature may be considered the most physiologically relevant ( Fig. S3 and Table S6 in the "Supplementary Information S1"). Given the rapidly evolving landscape of variants and the risk of recombination events, to extend our analysis to 'theoretical' variants, in this instance combination of L452R with other mutations, we produced and tested S RBD L452R/K417N, L452R/E484K, L452R/N501Y, and L452R/K417N/E484K/N501Y. All but the construct expressing L452R/K417N showed higher affinity for ACE(WT) ( Table S7 in the "Supplementary Information S1"), which aligns with what might be predicted from the results in Table S5-K417N decreases affinity and N501Y increases affinity. Interestingly, affinity in combination appears to be additive, with the K D being 8.77 for E484K alone (as compared to 9.33 for RBD WT) and 5.58 for L452R alone; together the KD for binding to ACE2 (WT) is 4.10.
Inhibition of ACE2:S RBD binding in the surrogate virus neutralization assay correlates with binding affinity. The surrogate SARS-CoV-2 neutralization assay cPass™ 44 is based upon assessment of inhibition of binding of ACE2 (WT) to S RBD (WT). It is typically used to ascertain the presence of anti-S RBD antibodies in serum. Such antibodies inhibit binding of S RBD to ACE2 bound to an ELISA plate, and inhibition of ≥ 30% has been reported to correlate with neutralization of live virus. Here, the surrogate assay was used to determine if the ACE2 Triple Decoy could inhibit S RBD WT and variant binding to plate-bound ACE2, that is, compete with ACE2 (WT) for S RBD binding (see "Methods").
As shown in Fig. 3s, inhibition in the surrogate virus neutralization assay modified to assess competition by the ACE2 Triple Decoy for ACE2 (WT) binding to S RBD was similar for S RBD WT, E484K, N501Y, L452R, and K417N/E484K/N501Y. The percent inhibition with the ACE2 Triple Decoy and S RBD K417N was significantly lower than the other mutants, but still very high.

MD simulation accurately predicts the relative affinities confirmed by in vitro testing.
To predict binding affinities, we used Adaptively-biased MD (ABMD) simulations 45 , which allow observation and quantification of binding and unbinding, of both ACE2 WT and ACE2 (T27Y/H34A) binding to S RBD WT or B.1.351. For these simulations, the B.1.351 variant comprising the E484K, K417N, and N501Y mutations was used because these mutations occur together naturally and thus this combination has high physiological relevance. Furthermore, this variant has been reported to escape antibodies elicited by first-wave SARS-CoV-2 and some vaccines 4, 46-50 and is of particular interest. The ACE2 T27Y/H34A sequence without the additional H374N enzyme-deactivating mutation found in the ACE2 Triple Decoy was used because earlier simulations had been unable to detect a change in affinity due to the presence of the H374N mutation.
We predicted binding affinities by determining the Helmholtz binding free energy (ΔA bind ) using the ratio of the probability of the bound and unbound states based on the Free Energy Surfaces (FES) (Fig. 4) The predictive utility of these simulations is validated by the K D values determined in vitro and presented in Table S5 in the "Supplementary Information S1", where (for the combinations tested in MD simulations) the lowest affinity was also seen for ACE2 WT: RBD WT (K D = 9.33 nM), followed by ACE2 WT: RBD B.1.351 (K D = 5.28 nM), then ACE2 Decoy: RBD B.1.351 (KD = 0.465 nM), and ACE2 Decoy: RBD WT (KD = 0.315 nM). All affinities were high, and higher for Triple Decoy binding than ACE2 WT for all RBD sequences tested.

Discussion
To our knowledge, we are the first to report binding affinities of a recombinant mutant ACE2 decoy to the spike receptor binding domain expressing N501Y, E484K, N417Y, or L452R mutations; although we note Huang et al. reported previously on the affinity of their ACE2-F C to S RBD with the D614G mutation 51 . The greater affinity of ACE2 for S RBD with the N501Y substitution alone or in combination with E484K reported here is in alignment with our findings in Nelson et al. 52 , wherein we used MD simulation to predict that these mutations have a high probability of increasing affinity for ACE2.
The MD simulation data presented here was used to guide design of the ACE2 Triple Decoy and to predict affinities of the decoy as compared to ACE2 WT for a series of variants reveal again the merits of such simulations as a tool to inform therapeutic design.
Interestingly, widespread use of an ACE2 decoy has the potential itself to act as an evolutionary force; however, an ACE2 decoy largely recognizes the same residues as endogenous ACE2 and therefore it is highly unlikely a SARS-CoV-2 variant could emerge that 'escapes' the decoy yet still binds to endogenous ACE2. This phenomenon along with limited use of a decoy for therapy as compared to the spread of virus in a large population with opportunity for selection, makes the decoy approach less vulnerable to loss of efficacy due to mutation of the virus.
The enhanced binding affinity of our Triple Mutant ACE2 Decoy to S RBD with the variant mutations tested here supports continued pursuit of this therapeutic approach and further provides hope that even should the efficacy of vaccines currently in distribution or therapeutic neutralizing antibodies raised against WT spike be www.nature.com/scientificreports/ lessened by these variants, there will be an alternative therapeutic approach to successfully treat COVID-19 disease.
In our next steps in development of the ACE2 Triple Decoy, we will address the challenge of stability and successful delivery. Others developing ACE2 decoys have suggested use of intranasal 53 or nanoparticle/extracellular vesicle delivery 22,[54][55][56] . We anticipate going forward into our next studies using the dimeric IgA 57 fusion protein decoy expressed by the human adenovirus serotype 5 E1, E2b, E3 deleted (hAd5 [E1−, E2b−, E3−]) platform that we have used successfully in our vaccine development 58,59 . This platform can readily be used to generate oral and/or intranasal formulations to further facilitate delivery. Our ACE2 Triple Decoy delivered in vivo using the hAd5 platform is anticipated to overcome barriers to successful delivery and will be tested in animal models of SARS-CoV-2 infection in future studies. www.nature.com/scientificreports/ Methods MD simulation. System setup. The WT-ACE2/RBD complex was built from the cryo-EM structure, PDB 6M17 of full-length human ACE2 in the presence of the neutral amino acid transported B 0 AT1 with the S RBD as shown in Yan et al. 60 using RBD residues 336-518 and ACE2 residues 21-614. ACE2 residues 27 and 34 were mutated to tyrosine and alanine, respectively. The final simulation system was built using the Amber ff14SB force field 61 . The RISM program from AmberTools19 62 was used to determine optimal locations for water molecules in direct contact with the proteins. Bulk waters were added to create a sufficient octahedral water box and sodium ions were added at random locations to neutralize the system. The TIP3P force field was used for water and the Li/Merz parameters for sodium 63,64 . After introducing mutations at the relevant residues, the same procedure was used to generate the other three systems.
Simulation. Ten copies of each RBD:ACE2 complex were minimized, equilibrated and simulated. Minimization occurred in two phases. During the first, the protein and RISM-placed waters were restrained. The second phase minimized the entire system. Dynamics then began and the temperature was ramped from 0 to 300 K while restraining the protein and RISM-placed waters. All dynamics used SHAKE restraints on hydrogen-containing bonds and a 2 fs timestep. All restraints were then released and the system was equilibrated in the NPT ensemble for 2 ns. Finally, each system was equilibrated in the NVT ensemble for 100 ns. Steered MD was used to prepare the equilibrated systems for free energy calculation. Contacting residues from the adaptively biased MD (ABMD) simulations in Nelson et al. 28,45 were used. Starting from the NVT equilibrated structures and over a 10 ns simulation, the number of intermolecular contacts was linearly reduced to 0 using a 10 kcal/mol*Å steering bias. Structures were randomly selected from the steered MD simulations and used to seed ABMD simulations. Two dimensional ABMD simulations used intermolecular contacts and the center of mass distance as collective variables. Centers of mass were defined as the alpha carbons from all interfacial residues in each molecule. The well-tempered ABMD bias potential 65 was used for free energy calculations. Our goal with the simulations was to inform the experiment by ranking the affinities of each complex instead of predicting the magnitude of the binding affinity. Based on our experience with the ACE2 WT:RBD WT complex 28 Figure S4 in the "Supplementary Information S1". It is representative of the other three systems in that we see full coverage of both CVs and multiple transitions between the bound and unbound states. Production simulations were run in the NVT ensemble meaning the calculated free energy corresponds to the Helmholz free energy (ΔA).
ABMD produces a free energy surface (FES) that describes the relative free energy between any two points on the FES, ΔA FES . The binding free energy (ΔA bind ) is determined by the ratio of the probability of the bound and unbound states and can be determined from the FES: where β is the inverse of the Boltzmann constant multiplied by the temperature in Kelvins. More negative values of ΔA bind indicate a stronger association. The calculated ΔA bind values can be directly compared.
The "Bound" integral in Eq. (1) is defined to be over all ΔA FES (x,y) values with the number of contacts greater than 0.05 while the "Unbound" integral is over all values with fewer than 0.05 contacts. ΔA bind was calculated with different boundaries ranging from 0.0 to 1.0, inclusive. As expected, the resulting values of ΔA bind changed based on the chosen boundary. However, the relative ordering of the values did not. The value of 0.05 contacts was chosen as the boundary because it allowed for unambiguous categorization of points as either "unbound" (x = 0) or "partially" or "fully bound". All simulations were performed with the GPU-enabled version of pmemd from Amber20 62 . Multiple-walker ABMD simulations 66 used the MPI version of pmemd.cuda from Amber20. For each system, average ΔA and standard deviation values were calculated over the final ten percent of simulation data from each walker.
The RMSD for the RBD residues contacting ACE2 residue 34 were calculated from unbiased MD simulations. The ten 100 ns NVT WT RBD:WT ACE2 simulations were extended to a total of 5 μs. An A34 simulation system was set up and equilibrated following the procedure outlined above and 5 μs of unbiased NVT data was collected. The simulation trajectories were aligned to the alpha carbons of ACE2 residues 23-47. RMSD was then measured for heavy atoms of RBD residues 453, 455 and 493, which are in contact with residue H34 in the WT complex. Results were calculated using cpptraj 67 . www.nature.com/scientificreports/ was constructed similarly, with the exception that ACE2 R273Q was created by splice by overlap extension (SOE). A 1.86 kb SOE product was created by amplification with primers WH1043 and WH1044 of two PCR products:

Production of ACE2 Decoys and S RBD.
(1) 860 bp amplification of AO615ACE2 with primers WH1043 and WH1049, and (2) 1.059 bp amplification product of AO615ACE2 with primers WH1050 and WH1044. ACE2 T27Y/H34A-IgG 1 F C was constructed by the Gibson Assembly of: (1) a 9.041 kb NheI-PshA1 digestion fragment of ACE2-IgG 1 F C plasmid, and (2) a 0.773 kb SOE product of primers 5MutF and 5MutR of two PCR products. The first PCR product is a 0.154 kb amplification of plasmid SR9 with primers 5MutF and ACE-2T27YR). The second PCR product is a 0.642 kb amplification of plasmid SR9 with primers ACE2T27YF and 5MutR.
Most of the triple mutants were created by Gibson Assembly of three fragments: (1) the vector backbone from a 7.168 kb NheI-XhoI fragment of pWT35, (2) IgG 1 F C from a 0.701 kb PCR amplification of pXL159 with primers WH1045 and WH1046, and (3) the ACE2 variant from a 1.86 bp PCR product containing the three mutations. For the latter, the mutants were amplified with primers WH1043 and WH1044 with templates pWH230 (for  T27Y/ For transfection, cells in the exponential growth stage were pelleted by centrifugation at 1400 rpm for 10 min, re-suspended in 10 mL of electroporation buffer, and re-pelleted at 1400 rpm for 5 min. The cell pellet was resuspended at a density of 2 × 10 8 cells/mL in electroporation buffer, mixed with the plasmid harboring either the ACE2(WT)-IgG1Fc or ACE2(WT)-IgA sequence at a concentration of 150 µg/mL, and transfected using OC-400 processing assemblies in a Maxcyte ExPERT ATx Transfection System. Transfected cells were incubated for 30 min at 37 ℃, 5% CO 2 and then resuspended in Efficient Feed A ® Cocktail (CHO-CD EfficientFeed™ A + 0.2% Pluronic F-68 + 1% HT Supplement + 1% l-glutamine) at a density of ~ 4-6 × 10 6 cells/mL. This cell culture was incubated at 37 ℃ with 5% CO 2 and 125 rpm rotation overnight, 1 mM sodium buryrate was added, and the culture was further incubated at 32 ℃ with 3% CO 2 and 125 rpm for 13 more days; during this incubation period, Maxcyte ® Feed Cocktail (13.9% CD Hydrolysate, 69.5% CHO CD EfficientFeed™ A, 6.2% Glucose, 6.9% FunctionMax™ Titer Enhancer, 3.5% l-glutamine) was added at 10% of the culture volume on Days 3 and 8. Lipofectamine® transient transfection of RBD constructs. For transient expression of RBD wild-type and RBD mutants, HEK-293 T cells were cultured and incubated at 37 ℃ with 5% CO 2 . Plasmids harboring RBD constructs were mixed with lipofectamine with 1:1 (v:v) and incubated for 20 min at room temperature. The mixture was then added to cultures and incubated for 3-4 days.
Purification of ACE2 Decoy IgGs. The MaxCyte ® or FectoPRO ® transfection cell culture medium was centrifuged and filtered through a 0.22 µm filter to remove cells and debris, then loaded onto a HiTrap™ MabSelect SuRe™ column on the AKTA Pure system pre-equilibrated with 10 mM Na Phosphate and 150 mM NaCl at pH 7.0. After loading, the column was washed with ten column volumes of the same buffer. The protein was eluted with 100 mM sodium acetate, pH 3.6, then immediately neutralized using 2 M Tris pH 8.0. The elution fractions were pooled and dialyzed into 10 mM HEPES and 150 mM sodium chloride at pH 7.4. www.nature.com/scientificreports/ Purification of ACE2 Decoy IgAs. The MaxCyte ® transfection cell culture medium was centrifuged and filtered through a 0.22 µm filter to remove cells and debris, then loaded to a gravity column packed with CaptureSelect ® IgA resins (Thermo Fisher) pre-equilibrated with 10 mM Na Phosphate and 150 mM NaCl at pH 7.0. After loading, the column was washed with ten column volumes of the same buffer. The protein was eluted with 100 mM sodium acetate, pH 3.0, then immediately neutralized using 2 M Tris, pH 8.0. The elution fractions were pooled and dialyzed into 10 mM Hepes and 150 mM sodium chloride, pH 7.4.
Purification of RBD and RBD mutants. The Lipofectamine transfection cell culture medium was centrifuged and filtered through a 0.22 µm filter to remove cells and debris. A buffer of 50 mM Tris, 100 mM sodium chloride, and 10 mM imidazole was added to the supernatant then loaded to a gravity column packed with Ni-NTA resins (Qiagen) pre-equilibrated with 20 mM Tris, 300 mM sodium chloride, and 10 mM imidazole, pH8.0. After loading, the column was washed with 10 column volumes of the same buffer. The protein was eluted with 20 mM Tris, 150 mM sodium chloride, and 300 mM imidazole. The elution fractions were pooled and dialyzed into 10 mM HEPES and 150 mM sodium chloride, pH 7.4. cPass™ 44 surrogate SARS-CoV-2 neutralization assay. High BIND 96-well ELISA plates (Corning #3369) were coated with 50 ng/well ACE2 wild type decoy overnight at 4 ℃. After the antigen solution was removed, each well was blocked with 150 µL of 5% BSA/PBS for 1-2 h at room temperature with shaking. During the blocking step, 40 µL of 50 nM RBD and RBD variants were mixed with 40 µL of 25 µg/mL of ACE2 decoy were mixed in a 96-well plate and incubated at room temperature for 30 min with shaking. After blocking, the plate was then washed three times with 250 µL of PBS with 0.05% Tween 20 (PBS-T). To each well, 30 µL of 1:1667 diluted mouse anti-His, HRP and 60 µL of RBD/ACE2 decoy (or a no decoy control) were added and incubated at room temperature for 30 min. The plated was washed once with 250 µL of PBS-T. To develop the signal, 50 µL of TMB solution was added and incubated at room temperature in dark for 30 min, followed by addition of 50 µL of 2 M sulfuric acid; absorbance was the read at 450 nm. The percent inhibition was calculated using (1 − A450 (RBD + Decoy)/A450 (RBD only)) × 100.

RBD affinity determination of ACE2 decoys by bio-layer interferometry (BLI
Assay for ACE2 enzymatic activity. Enzymatic activity ACE2 decoys expressing a variety of mutations-R273Q, R273K, R273L, H245A, H505L, H374N, and H378N-selected to inhibit activity in combination with the S RBD affinity-enhancing mutations T27Y and H34A were assessed in the FRET based ACE2 activity assay.