Computational design of SARS-CoV-2 peptide binders with better predicted binding affinities than human ACE2 receptor

SARS-CoV-2 is coronavirus causing COVID-19 pandemic. To enter human cells, receptor binding domain of S1 subunit of SARS-CoV-2 (SARS-CoV-2-RBD) binds to peptidase domain (PD) of angiotensin-converting enzyme 2 (ACE2) receptor. Employing peptides to inhibit binding between SARS-CoV-2-RBD and ACE2-PD is a therapeutic solution for COVID-19. Previous experimental study found that 23-mer peptide (SBP1) bound to SARS-CoV-2-RBD with lower affinity than ACE2. To increase SBP1 affinity, our previous study used residues 21–45 of α1 helix of ACE2-PD (SPB25) to design peptides with predicted affinity better than SBP1 and SPB25 by increasing interactions of residues that do not form favorable interactions with SARS-CoV-2-RBD. To design SPB25 with better affinity than ACE2, we employed computational protein design to increase interactions of residues reported to form favorable interactions with SARS-CoV-2-RBD and combine newly designed mutations with the best single mutations from our previous study. Molecular dynamics show that predicted binding affinities of three peptides (SPB25Q22R, SPB25F8R/K11W/L25R and SPB25F8R/K11F/Q22R/L25R) are better than ACE2. Moreover, their predicted stabilities may be slightly higher than SBP1 as suggested by their helicities. This study developed an approach to design SARS-CoV-2 peptide binders with predicted binding affinities better than ACE2. These designed peptides are promising candidates as SARS-CoV-2 inhibitors.

www.nature.com/scientificreports/ between residue 21-45 of ACE2 and SARS-CoV-2-RBD 29,37 and then combine the newly designed single mutations with the best designed single mutations from our previous study to further enhance the binding affinities of the designed peptides so that their predicted binding affinities are better than ACE2; our previous work designed the residues that have not been reported to form favorable interactions with SARS-CoV-2-RBD to increase favorable interactions of these residues and avoid disrupting existing favorable interactions. In this study, Rosetta was employed to design SARS-CoV-2-RBD peptide binders, and the designed positions were selected from the residues that were previously reported to form favorable interactions with SARS-CoV-2-RBD 29,37 and their side chains could potentially form favorable interactions upon mutations with SARS-CoV-2-RBD. In this study, Q4 (24), T7 (27), D10 (30), K11 (31), H14 (34), E15 (35), E17 (37), D18 (38), Y21 (41) and Q22 (42) were selected based on these criteria. Each designed position was allowed to be any of standard amino acids except G and P because G and P occur infrequently in an α-helix. P also can cause a destabilizing kink in a helix structure 38 . The total of 156 designed peptides with single mutation were obtained from Rosetta (Table S1). Ten designed peptides with better ΔG bind (Rosetta) than SPB25 (ΔΔG bind (Rosetta) < 0 REU) were selected for MD simulations to validate whether their predicted binding affinities by the more accurate Molecular Mechanics-Generalized Born Surface Area (MM-GBSA) method [39][40][41] (ΔG bind (MM-GBSA) ) were better than that of SPB25 (ΔΔG bind (MM-GBSA) < 0 kcal/ mol). These designed peptides are SPB25 T7I, SPB25 T7V, SPB25 K11F, SPB25 K11W , SPB25 H14V, SPB25 E15L, SPB25 E17F, SPB25 E17W , SPB25 D18E and SPB25 Q22R .
The binding free energy components of designed peptides with predicted binding affinity to SARS-CoV-2-RBD better than or similar to ACE2. Figure 2 shows binding energy components of the four designed peptides with predicted binding affinities better than or similar to ACE2 (Fig. 2) as compared to those of ACE2, SBP1 and SPB25. The electrostatic interaction terms are the main components contributing to the favorable predicted binding affinities of SPB25 K11F, SPB25 Q22R , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R to SARS-CoV-2-RBD. The van der Waals energy and non-polar solvation terms also contributes favorably to the predicted binding affinity. However, the polar solvation terms have unfavorable contribution to the predicted binding affinity.
As shown in Fig. 2 and Table 1, SPB25 Q22R is the designed peptide with the best predicted binding affinity with the ΔG bind (MM-GBSA) value of − 75.3 ± 0.5 kcal/mol. Its predicted binding affinity is better than those of ACE2, SBP1 and SPB25 by − 4.1 ± 0.6, − 20.2 ± 0.6 and − 15.0 ± 0.6 kcal/mol, respectively. The favorable binding of SPB25 Q22R to SARS-CoV-2-RBD is mostly caused by the increase in the favorable van der Waals energy and non-polar solvation terms as well as the decrease in unfavorable polar solvation term as compared to those of SBP1 and SPB25. The favorable electrostatic interaction term of SPB25 Q22R is worse than those of SBP1 and SPB25. The predicted binding affinity of SPB25 K11F is better than those of SBP1 and SPB25 and similar to that of ACE2. The favorable binding of SPB25 K11F to SARS-CoV-2-RBD is mostly caused by the increase in the favorable van der www.nature.com/scientificreports/ Waals energy, electrostatic interaction terms and non-polar solvation terms as compared to those of SBP1 and SPB25. The unfavorable polar solvation term of SPB25 K11F is worse than that of SBP1 and SPB25. The predicted binding affinities of SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R are better than those of SBP1, SPB25 and ACE2. The favorable binding of SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R to SARS-CoV-2-RBD is mostly caused by the increase in the favorable van der Waals energy and non-polar solvation terms as well as the decrease in unfavorable polar solvation terms as compared to those of SBP1 and SPB25. However, the favorable electrostatic interaction terms of these two designed peptides are worse than those of SBP1 and SPB25. The predicted binding affinities to SARS-CoV-2-RBD of the three designed peptides are better than that of ACE2 because their unfavorable polar solvation terms are substantially lower than that of ACE2 although their favorable van der Waals, electrostatic interaction and non-polar solvation terms are worse than that of ACE2. www.nature.com/scientificreports/ Identification of important binding residues of designed peptides with predicted binding affinity to SARS-CoV-2-RBD better than or similar to ACE2. To identify important binding residues to SARS-CoV-2-RBD of four designed peptides with predicted binding affinities better than or similar to ACE2, per-residue free energy decomposition was calculated and shown in Fig. 3. An important binding residue was defined to be a residue with the total energy contribution better than − 1.0 kcal/mol 42 . Overall, the number of important binding residues of SPB25 K11F (11), SPB25 Q22R (8), SPB25 F8R/K11W/L25R (12) and SPB25 F8R/K11F/Q22R/L25R www.nature.com/scientificreports/ (12) were predicted to be relatively similar or more than those of SBP1 (8), SPB25 (9) and residues 21-45 of the α1 helix of ACE2 (7) 36 . Overall, four residues of all designed peptides were predicted to have high binding affinity (better than − 2.0 kcal/mol) such as Y21 (the best binding residue), Q4, T7, and K11/F11/W11. Additionally, H14 of SPB25 Q22R , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R , R22 of SPB25 Q22R as well as S24 of SPB25 F8R/K11W/L25R were also predicted to have high binding affinity to SARS-CoV-2-RBD.
In terms of the designed peptides with quadruple mutation, the F8R/K11F/Q22R/L25R mutation was predicted to favorably increase the total energy contributions of residues 8 and 25 from − 1.5 and − 0.6 kcal/mol in ACE2 and − 1.4 and − 0.4 kcal/mol in SPB25 to − 3.3 and − 1.3 kcal/mol in SPB25 F8R/K11F/Q22R/L25R , respectively, while this quadruple mutation was predicted to unfavorably decrease the total energy contributions of residues 11 and 22 from − 3.  www.nature.com/scientificreports/

Hydrogen bond and pi interactions of designed peptides with predicted binding affinities to SARS-CoV-2-RBD better than or similar to ACE2. To identify important hydrogen bonds and pi
interactions for the binding to SARS-CoV-2-RBD of four designed peptides with predicted binding affinities better than or similar to ACE2, hydrogen bond occupations, pi-pi, cation-pi and sigma-pi interactions were analyzed as shown in Table 2 and Table S2. Key binding interactions are shown in Fig. 4. Overall, the binding positions and orientations of all designed peptides to SARS-CoV2-RBD are relatively similar to those of ACE2.
In terms of the designed peptides with quadruple mutations, the total number of predicted hydrogen bonds of SPB25 F8R/K11F/Q22R/L25R is higher than that of SBP1, lower than that of SPB25, and similar to that of ACE2, but Table 2. Numbers of hydrogen bond and pi interactions of ACE2, SBP1, SPB25 and designed peptides contributing to SARS-CoV-2-RBD binding.

Peptide helicities of designed peptides with predicted binding affinities to SARS-CoV-2-RBD better than or similar to ACE2. The RMSD plots and the percent helicities of designed peptides in water
with predicted binding affinities to SARS-CoV-2-RBD better than or similar to ACE2 are shown in Figure S2 and Fig. 5, respectively. Because of their high flexibilities, the percent helicities of the N terminus and C terminus of each peptide are lower than those of the residues in the middle. Overall, the trends of percent helicities in water of SPB25 K11F , SPB25 Q22R , SPB25 F8R/K11W//L25R SPB25 F8R/K11F/Q22R/L25R and SPB25 are slightly higher than those of SBP1 (the experimentally proven peptide binder of SARS-CoV-2).

Discussion
COVID-19 pandemic has caused large numbers of cases and deaths globally, and it is caused by SARS-CoV-2 that initially uses its SARS-CoV-2-RBD to bind to ACE2-PD to enter human cells. Therefore, inhibiting the binding between SARS-CoV-2-RBD and ACE2-PD is a promising therapeutic solution for COVID-19. As alternatives to small molecules that are often ineffective in inhibiting large protein binding interfaces 43 , peptides can potentially be used as SARS-CoV-2 inhibitors because their surfaces are larger and they have more functional groups and similar interactions to the native protein-protein interactions than small molecules 20 . Designed based on residues 21-43 of the α1 helix of ACE2-PD, the 23-mer peptide (SBP1) was experimentally found to bind to SARS-CoV-2-RBD with lower binding affinity than ACE2 and could potentially be used as a SARS-CoV-2 inhibitor 32 . To further enhance the binding affinity of SBP1, our previous study employed computational protein design (Rosetta) and MD (AMBER) to design 25-mer peptide binders of SARS-CoV-2-RBD, based on residues 21-45 of the α1 helix of ACE2-PD (SPB25), by using residues that have not been reported to form favorable interactions with SARS-CoV-2-RBD to increase favorable interactions of these residues and avoid disrupting existing favorable interactions. Five designed peptides such as SPB25 F8N , SPB25 F8R , SPB25 L25R , SPB25 F8N/L25R and SPB25 F8R/L25R were predicted to bind to SARS-CoV-2-RBD with better binding affinities than SBP1 and SPB25. However, their predicted binding affinities to SARS-CoV-2-RBD are still lower than human ACE2 receptor. The aim of this study is to further increase the binding affinities of 25-mer peptides so that their predicted binding affinities are better than human ACE2 receptor using computational protein design (Rosetta) and MD (AMBER). Using SPB25 as a designed template and reference, our design strategy is to enhance the binding affinity of residues that were previously reported to form favorable interactions between residue 21-45 of ACE2-PD and SARS-CoV-2-RBD 29,37 and combine the newly designed single mutations with the best designed single mutations (SPB25 F8N , SPB25 F8R and SPB25 L25R ) from our previous study to further increase the binding affinities of designed peptides so that their predicted binding affinities are better than human ACE2 receptor. In this study, designed positions were selected from residues that were previously reported to form favorable interactions with SARS-CoV-2-RBD 29,37 and their side chains could potentially form favorable interactions upon mutations with SARS-CoV-2-RBD. Q4 (24), T7 (27), D10(30), K11(31), H14(34), E15 (35), E17(37), D18 (38), Y21 (41) and Q22(42) of SPB25 were selected for design based on our criteria, and they were allowed to be any of standard amino acids except G and P. The total of 156 designed peptides with single mutations were obtained from Rosetta, and the values of ΔG bind (Rosetta) of ten designed peptides are better than that of SPB25 (ΔΔG bind (Rosetta) < 0 REU). These ten designed peptides were selected for MD, and their binding free energies (ΔG bind (MM-GBSA) ) were calculated by the more accurate MM-GBSA method to determine whether their predicted binding affinities were better than that of SPB25. Our results show that three designed peptides with single mutations such as SPB25 K11F, Figure 5. The percent helicities in water of SBP1 36 , SPB25 36 and designed peptides with predicted binding affinities to SARS-CoV-2-RBD better than or similar to ACE2. www.nature.com/scientificreports/ SPB25 K11W and SPB25 Q22R were predicted to bind to SARS-CoV-2-RBD better than SPB25 with ΔΔG bind (MM-GBSA) of − 11.3 ± 0.7, − 2.9 ± 0.6 and − 15.0 ± 0.6 kcal/mol, respectively. These three designed single mutations were combined with the three best designed single mutations (SPB25 F8N , SPB25 F8R and SPB25 L25R ) from our previous work to construct 11, 12 and 4 designed peptides with double, triple and quadruple mutations using Rosetta (SPB25 F8N/L25R and SPB25 F8R/L25R were not included in this study because their predicted binding affinities were already reported in our previous study). MD was performed on these designed peptides, and their values of ΔG bind (MM-GBSA) were computed. In terms of designed peptides with double mutations, SPB25 F8N/K11W , SPB25 F8R/K11F , SPB25 F8R/K11W , SPB25 F8R/Q22R , SPB25 K11F/L25R and SPB25 K11W/L25R were predicted to bind to SARS-CoV-2-RBD better than SPB25 with ΔΔG bind (MM-GBSA) of − 6.5 ± 0.6, − 9.4 ± 0.6, − 9.2 ± 0.6, − 3.6 ± 0.5, − 2.8 ± 0.6 and − 4.0 ± 0.6 kcal/mol, respectively. For designed peptides with triple mutations, SPB25 F8N/K11F/L25R , SPB25 F8N/K11W/L25R , SPB25 F8R/K11W/L25R and SPB25 K11W/Q22R/L25R were predicted to bind to SARS-CoV-2-RBD better than SPB25 with ΔΔG bind (MM-GBSA) of − 0.3 ± 0.6, − 1.4 ± 0.6, − 14.7 ± 0.5 and − 7.5 ± 0.6 kcal/mol, respectively. In terms of designed peptides with quadruple mutation, SPB25 F8R/K11F/Q22R/L25R and SPB25 F8R/K11W/Q22R/L25R were predicted to bind to SARS-CoV-2-RBD better than SPB25 with ΔΔG bind (MM-GBSA) of − 11.9 ± 0.6 and − 7.1 ± 0.6 kcal/mol, respectively. All designed peptides were also predicted to bind to SARS-CoV-2-RBD better than SBP1 (the experimentally proven peptide binder of SARS-CoV-2-RBD), suggesting that they should be able to bind to SARS-CoV-2-RBD better than SBP1, experimentally. Most importantly, three designed peptides (SPB25 Q22R , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R ) were predicted to bind to SARS-CoV-2-RBD better than ACE2 by − 4.1 ± 0.6, − 3.8 ± 0.5 and − 1.0 ± 0.6 kcal/mol, respectively, suggesting that they should bind to SARS-CoV-2-RBD better than ACE2, experimentally. Moreover, one designed peptide (SPB25 K11F ) was predicted to bind to SARS-CoV-2-RBD with relatively similar binding affinity (− 71.6 ± 0.6) to ACE2 (− 71.2 ± 0.4), suggesting that it should bind to SARS-CoV-2-RBD with relatively similar K D to ACE2. The ranking of the predicted binding affinities of the designed peptides, SPB25, SBP1 and ACE2 (best to worst) is SPB25 Q22R ≈ SPB25 F8R/K11W/L25R > SPB25 F8R/K11F/Q22R/L25R > SPB25 K11F ≈ ACE2 > SPB25 > SBP1. Although ACE2 is markedly larger and has more residues interacting with SARS-CoV-2-RBD, including residues in the α2 helix and the linker of the β3 and β4 antiparallel strands in addition to residues 21-45 6,9 , than our best designed 25-mer peptides, our approach was able to design 25-mer peptides with better predicted binding affinity than ACE2, suggesting the effectiveness of our approach and the high efficacies of our best designed peptides. Moreover, the binding positions and orientations of all designed peptides to SARS-CoV2-RBD are relatively similar to that of residues 21-45 of the α1 helix of ACE2-PD, suggesting that they could potentially disrupt the binding interactions between SARS-CoV2-RBD and ACE2-PD. SPB25 Q22R is the most promising designed peptide because its predicted binding affinity is better than ACE2, SPB25, SBP1 and all designed peptides. This result is supported by the fact that its total numbers of predicted hydrogen bonds (involving E3, Q4, D10, H14, E15, E17, Y21, R22, S23, S24 and L25) and pi interactions (involving K11, H14, Y21 and R22) are higher than those of SPB25, SBP1 and ACE2. The per-residue free energy decomposition results suggest Q4, T7, F8, K11, H14, E17, Y21 and R22 as important binding residues. Additionally, the Q22R mutation was predicted to cause substantial favorable increase in the total energy contribution of this residue and the total energy contributions of other residues such as Q4, F8, H14, E17, and Y21 as compared to those of SPB25 and ACE2. SPB25 F8R/K11W/L25R was predicted to bind better to SARS-CoV2-RBD than SBP1, SPB25 and ACE2. This result is supported by the fact that its total numbers of predicted hydrogen bonds (involving I1, E3, Q4, R8, D10, H14, E17, Y21, Q22, S24 and R25) and pi interactions (involving W11, R8 and Y21) are higher than those of SBP1 and ACE2, and the number of predicted strong hydrogen bonds of SPB25 F8R/K11W/L25R is higher than that of SBP1, SPB25 and ACE2. The predicted binding affinity of SPB25 F8R/K11W/L25R is lower than SPB25 Q22R , and this result is supported by the fact that its total numbers of predicted hydrogen bonds and pi interactions are lower than those of SPB25 Q22R . The results from per-residue free energy decomposition suggest E3, Q4, T7, R8, D10, W11, H14, E17, Y21, Q22, S24 and R25 as important binding residues. Furthermore, the F8R/K11W/L25R mutation was predicted to cause substantial increase in the total energy contribution of residue 8 and 25 as well as other residues such as E3, Q4, H14, E17, Y21 and S24 as compared to those of SPB25 and ACE2.
The binding affinity of SPB25 K11F was predicted to be better than those of SBP1, SPB25 and relatively similar to ACE2. The enhanced binding affinity of SPB25 K11F is probably caused by the increase in the total numbers of predicted hydrogen bonds (involving E2, Q4, D10, H14, E17, D18, Y21, Q22, S23, S24 and L25) and pi interactions (involving F11, F20 and Y21) of SPB25 K11F as compared to those of SBP1, SPB25 and ACE2. SPB25 K11F has the worst predicted binding affinity among the four best designed peptides, and this finding is supported by the fact that its total numbers of predicted strong, medium and weak hydrogen bonds as well as pi interactions are the lowest among these four best designed peptides. The results from per-residue free energy decomposition www.nature.com/scientificreports/ suggest Q4, T7, F8, D10, F11, H14, E17, F20, Y21, Q22 and S24 as important binding residues. Moreover, the K11F mutation caused significant increase in the total energy contributions of other residues such as F8, E17, F20, Y21 and S24 as compared to those of SPB25 and ACE2. In terms of peptide helicities, the trends of percent helicities in water of SPB25 Q22R , SPB25 K11F , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R are slightly higher than that of SBP1. These results suggest that their stabilities in water may be slightly better than that of SBP1 (the experimentally proven binder of SARS-CoV-2-RBD), and these designed peptides should be stable enough to be used as peptide binders of SARS-CoV-2.
Employing computational protein design and MD, we designed three 25-mer peptides (SPB25 Q22R , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R ) and one 25-mer peptide (SPB25 K11F ) with predicted binding affinities better than and similar to that of human ACE2 receptor, respectively. Although their sizes are markedly smaller than human ACE2 receptor, they were predicted to bind to SARS-CoV-2-RBD with better or similar binding affinities, suggesting their high efficacies. These four designed peptides are promising candidates that could potentially be employed as inhibitors to prevent the binding of SARS-CoV-2-RBD and ACE2. One potential application is to use these designed peptides as inhaled therapeutics for topical lung delivery to prevent the binding of SARS-CoV-2-RBD and ACE2 in the lung 44 . Moreover, these 25-mer peptide binders are approximately 40-fold smaller than a full antibody molecule; therefore, they have roughly 40-fold more potential neutralizing sites than a full antibody molecule at the same equal mass, thereby enhancing their potential efficacies. Furthermore, since they do not require expression in mammalian cells for proper folding like antibodies, the cost of scale-up and increase production volumes of these peptides should be lower than those of antibodies. Their small sizes should also allow them to be formulated in a gel for nasal application as well as to be delivered to the respiratory system as a dry powder or by nebulization 15 .
In conclusion, we developed an approach to design 25-mer peptide binders of SARS-CoV-2 with predicted binding affinities better than human ACE2 receptors, using computational protein design and MD. Employing SPB25 (residue 21-45 of ACE2-PD) as a designed template, our design strategy is to enhance the binding affinity of residues that were previously reported to form favorable interactions between residue 21-45 of ACE2-PD and SARS-CoV-2-RBD and combine the newly designed single mutations with the best designed single mutations (SPB25 F8N , SPB25 F8R and SPB25 L25R ) from our previous study to further increase the binding affinities of designed peptides so that their predicted binding affinities are better than human ACE2 receptor. Using this strategy, we designed three 25-mer peptides (SPB25 Q22R , SPB25 F8R/K11W/L25R and SPB25 F8R/K11F/Q22R/L25R ) and one 25-mer peptide (SPB25 K11F ) with predicted binding affinities to SARS-CoV-2-RBD, by the MM-GBSA method, better than and similar to human ACE2 receptor, respectively. Moreover, their predicted helicities in water are slightly higher than SBP1 (the experimentally proven 23-mer peptide binder of SARS-CoV-2-RBD), suggesting that their stabilities may be slightly better than SBP1. These four peptides are promising candidates as SARS-CoV-2 inhibitors.

Methods
Structure preparation. The 25-mer peptide of SPB25 (21 IEEQAKTFLDKFNHEAEDLFYQSSL 45) bound to SARS-CoV-2-RBD complex was obtained from our previous work 36 and it was constructed from the crystal structure of α1 helix of ACE2 peptidase domain (ACE2-PD) bound to SARS-COV-2-RBD (PDB ID: 6M0J 37 ). The complex was protonated at the physiological pH (pH 7.4) using H ++ server 45 . The LEaP module of AMBER18 46 was used to build the final structure of the complex.
Computational protein design. The structure of SPB25/SARS-CoV-2-RBD complex was employed as a template to design the SARS-CoV-2-RBD peptide binders using Rosetta. Our design strategy is to increase the binding affinity of residues that were previously reported to form favorable interactions between residue 21-45 of ACE2 and SARS-CoV-2-RBD 29,37 and further combine the newly designed single mutations with the best designed single mutations from our previous study to further increase the binding affinities of designed peptides so that their predicted binding affinities are better than ACE2. Obtained from our previous study, these best designed mutations were designed from the residues that have not been reported to form favorable interactions with SARS-CoV-2-RBD to increase favorable interactions of these residues and avoid disrupting existing favorable interactions. In this study, designed positions were selected from residues that were previously reported to form favorable interactions with SARS-CoV-2-RBD 29,37 and their side chains could potentially form favorable interactions upon mutations with SARS-CoV-2-RBD. The structure of designed residues were designed, repacked and minimized using the CoupledMoves protocol 47,48 in RosettaDesign module of Rosetta3.11 49 with beta_nov16 energy function. The designed positions were allowed to be any of standard amino acids except G and P, and the neighboring residues within 10 Å of designed position were also repacked and minimized. 400 independent runs were performed, and the total of 400 conformation of designed sequences were obtained for each design (some sequences may have multiple conformations). The binding free energy [ΔG bind (Rosetta) ] of each designed conformation was calculated in Rosetta Energy Unit (REU) using Interface Analyzer 50,51 module of Rosetta3.11. ΔΔG bind (Rosetta) upon mutation was computed by subtracting the values of ΔG bind (Rosetta) between the designed conformation and SPB25 conformation. The designed conformations with the best binding free energy and ΔΔG bind (Rosetta) < 0 REU of each design position were selected for MD simulations to validate their predicted binding affinities by the MM-GBSA method 39-41 . MD simulations and analyses. Using protein.ff14SB 52 and GLYCAM06j-1 force field parameters 53 in AMBER18 46 , the structures of designed peptides/SARS-CoV-2-RBD complexes were constructed in isomeric truncated octahedral boxes of TIP3P water molecules with the buffer distance of 13 Å. Each system was minimized using the five-step procedure 42,54-59 . All minimization steps include 2500 steps of steepest descent and www.nature.com/scientificreports/ 2500 steps of conjugate gradient with different restraints on the proteins to remove unfavorable interactions. In the first step, the hydrogen atoms and water molecules were minimized, while the heavy atoms of proteins were restrained with a force constant of 10 kcal/(mol Å 2 ). The backbones of the proteins were then restrained with the force constants of 10, 5 and 1 kcal/(mol Å 2 ) in the second, third and fourth steps of minimizations, respectively. Finally, no restraining force was applied in the system. All systems were simulated with the periodic boundary condition, using the GPU (CUDA) version of PMEMD module [60][61][62] . The SHAKE algorithm 63 was employed to constrain all bonds involving hydrogen atoms, allowing the time step of 0.002 ps. The Langevin dynamics technique was applied to control the temperatures of all systems with a collision frequency of 1.0 ps −1 . All systems were heated from 0 K to the physiological temperature of 310 K in the NVT ensemble for 200 ps, and a force constant of 10 kcal/(mol Å 2 ) was applied to restrain the backbones of the proteins. All systems were then equilibrated without restraint at 310 K in the NVT ensemble for 300 ps. Finally, they were subsequently simulated at 310 K and 1 atm in the NPT ensemble for 100 ns.
To analyze the stability of each system, the Root Mean Square Deviation (RMSD) values with respect to the minimized structure were calculated. The 80-100 ns trajectories of all systems with stable RMSD values were chosen for further analyses. To predict the binding affinities between designed peptides and SARS-CoV-2-RBD, the MM-GBSA method was used to calculate the total binding free energies [ΔG bind (MM-GBSA) ] of all systems. The designed peptides with better predicted binding affinity than ACE2 were further analyzed in terms of per-residue free energy decomposition and binding interactions. Hydrogen bond occupations were computed to analyze hydrogen bond interactions. In this study, a hydrogen bond was considered to occur if the following criteria were met: (1) a proton donor-acceptor distance ≤ 3.5 Å and (2) a donor-H-acceptor bond angle ≥ 120°4 2,54,55,64 . Hydrogen bond occupations were defined into four levels: (1) strong hydrogen bonds (hydrogen bond occupations > 75%), (2) medium hydrogen bonds (75% ≥ hydrogen bond occupations > 50%), (3) weak hydrogen bond interactions (50% ≥ hydrogen bond occupations > 25%) and (4) very weak hydrogen bond interactions (25% ≥ hydrogen bond occupations > 5%) 42,55,56 . To compute peptide helicities, Define Secondary Structure of Protein (DSSP) was employed. Percent helicity was calculated from the summation of the percentage of α-, 3 10and pi-helix structures 65 .

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information files). www.nature.com/scientificreports/