Novel Polymyxin-Inspired Peptidomimetics Targeting the SARS-CoV-2 Spike:hACE2 Interface

Though the bulk of the COVID-19 pandemic is behind, the search for effective and safe anti-SARS-CoV-2 drugs continues to be relevant. A highly pursued approach for antiviral drug development involves targeting the viral spike (S) protein of SARS-CoV-2 to prevent its attachment to the cellular receptor ACE2. Here, we exploited the core structure of polymyxin B, a naturally occurring antibiotic, to design and synthesize unprecedented peptidomimetics (PMs), intended to target contemporarily two defined, non-overlapping regions of the S receptor-binding domain (RBD). Monomers 1, 2, and 8, and heterodimers 7 and 10 bound to the S-RBD with micromolar affinity in cell-free surface plasmon resonance assays (KD ranging from 2.31 μM to 2.78 μM for dimers and 8.56 μM to 10.12 μM for monomers). Although the PMs were not able to fully protect cell cultures from infection with authentic live SARS-CoV-2, dimer 10 exerted a minimal but detectable inhibition of SARS-CoV-2 entry in U87.ACE2+ and A549.ACE2.TMPRSS2+ cells. These results validated a previous modeling study and provided the first proof-of-feasibility of using medium-sized heterodimeric PMs for targeting the S-RBD. Thus, heterodimers 7 and 10 may serve as a lead for the development of optimized compounds, which are structurally related to polymyxin, with improved S-RBD affinity and anti-SARS-CoV-2 potential.


Introduction
Three years have passed since the beginning of the Corona Virus Disease COVID-19 pandemic caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), which left hundreds of millions of people infected, while claiming millions of victims and causing heavy economic and social burdens worldwide [1][2][3]. Widespread vaccination and current clinical management have strongly reduced both the mortality and severity of the disease [1, 2,4]; however, serious concerns remain, including the following: the rise of viral resistance and reduced sensitivity to vaccines [5,6]; difficulties in effectively protecting vulnerable segments of the population, including the elderly and patients with multiple comorbidities [7][8][9]; and the unclear mechanisms of chronic-lasting effects of the disease observed in a percentage of infected people (long-COVID effects) [10]. Thus, we are still in need of effective and safe therapeutic drugs to accompany vaccines, together with Figure 1. X-ray structure (PDB code 6M0J) of the complex between the human ACE2 (light gray ribbons, bottom) and SARS-CoV-2 S-RBD (dark gray ribbons, top). The S hot spot residues previously identified by computational alanine scanning [42] are represented in green (binding site BS1) and magenta (binding site BS2).   , top). The S hot spot residues previously identified by computational alanine scanning [42] are represented in green (binding site BS1) and magenta (binding site BS2). In a recent study aimed at in silico drug repurposing for SARS-CoV-2 that employed X-ray-derived structures of the S-RBD:hACE2 interface and computational alanine scanning, we determined the key residues in the S-RBD that majorly contributed to the binding-free energy of the protein interface: two clusters of hot spots were identified, binding site 1 (BS1) and binding site 2 (BS2), which were localized in two distinct and relatively limited regions on the border rim of the targeted interface ( Figure 1) [42]. In the same work, a virtual screening (VS) campaign revealed polymyxin B (PMX, Figure 2), a naturally occurring antibiotic, to be a potential binder for both BS1 and BS2 sub-regions. Figure 1. X-ray structure (PDB code 6M0J) of the complex between the human ACE2 (light gray ribbons, bottom) and SARS-CoV-2 S-RBD (dark gray ribbons, top). The S hot spot residues previously identified by computational alanine scanning [42] are represented in green (binding site BS1) and magenta (binding site BS2).  Based on these precedents, this work aimed to develop novel, medium-sized polymyxininspired peptidomimetics that could simultaneously bind to the sub-regions BS1 and BS2 of SARS-CoV-2 S and hinder the interaction with hACE2, thus possibly impeding the entry of SARS-CoV-2 into human cells.
Starting from the in silico design of polymyxin-based peptidomimetics, a collection of six novel heterodimeric PMs, crafted from the covalent conjugation of ten BS1-and BS2-targeted monomers, was identified and synthesized. The binding affinity of these compounds toward the S-RBD of SARS-CoV-2 was determined by surface plasmon resonance (SPR) experiments, and their ability to inhibit infection of authentic SARS-CoV-2 virus in in vitro cellular assays was determined, as described in the next paragraphs.

Results
As stated in the introduction, the design of our dimeric PM ligands was founded on a previously carried out VS of FDA-approved drugs for the spike RBD [42]. To be specific, starting from the crystal structure of the SARS-CoV-2 S-RBD bound to human ACE2 (PDB code 6M0J), two potential binding regions (BS1 and BS2) were identified by computational alanine scanning (Figure 1). The vVS identified several cyclopeptides as potential binders of either BS1 or BS2, amongst which polymyxin B (PMX, Figure 2) represented the best-ranked hit for both BS1 and BS2 and thus, it was selected for further evaluation. PMX is a naturally occurring amphiphilic lipopeptide produced by fungi (although first isolated by Bacillus polymyxa) as a secondary metabolite and it is currently used as a last-resort antibiotic drug for the treatment of Gram-negative multidrug-resistant bacteria, or in topical antibiotic applications [43][44][45]. Its structure features an N-terminating methyloctanoyl linear tripeptide anchored to a cycloheptapeptide ring; the presence of this cyclic restraint together with non-canonical amino acid residues, including six L-α,γdiaminobutanoic acid residues (Dab) and one D-series residue (D-Phe), qualify it as a natural peptidomimetic [40].
Preliminary SPR assays, conducted on commercially available PMX and using immobilized RBD of the Wuhan-Hu-1 SARS-CoV-2 S protein, detected a minimum of specific binding ( Figure S1 in the Supplementary Materials), supporting previous work on molecular modeling [42], and indicating PMX to be an interesting molecular starting platform for the rational design and construction of improved specific cyclic PMs. We hypothesized that the limited performance of natural PMX in recognizing the S-RBD was possibly due to both the high flexibility and lipophilicity of the 6-methyloctanoyl chain, and the highly charged nature of its cycloheptapeptide ring at physiological levels of pH. Moreover, we also hypothesized that a single monomeric ligand, in only partially targeting the S-RBD, might not be sufficient enough to cover the crucial broad spike:hACE2 interface. Based on this assumption, we sought to design more extended ligands capable of covering both BS1 and BS2.

In Silico Design of Polymyxin B-Based Peptidomimetics
The PMX structure was initially modified in silico by replacing the 6-methyloctanoyl chain with an acetyl (AcPMX) ( Figure 3A). AcPMX was then subjected to a conformational search, leading to 19 conformations within a range of about 7 kcal/mol. To maximize the docking sampling ability, each conformation was independently docked to the S-RBD model by focusing on both BS1 and BS2. The results were sorted and the top-ranked poses, within 1 kcal/mol range, were evaluated. One pose for BS1 and one pose for BS2 were selected, with the intent of linking the two monomers into a dimeric structure, without altering the optimal ligand-receptor interactions predicted by docking ( Figure 3B).
We envisaged the possibility of replacing Thr10 of AcPMX docked to BS2 with the non-natural amino acid Trz, namely, (S)-2-amino-3-[4-(carboxymethyl)-1H-1,2,3-triazol-1yl]propanoic acid ( Figure 3C). Trz was designed with a proper length and functionalization to be used as a linker. Moreover, the triazole ring is frequently used as an amide bond bioisostere due to its stability, low flexibility, and synthetic feasibility by click-reaction [46]. Subsequently, the linkage between the two monomers was obtained by an amide bond between the side chain carboxylate of Trz and the α-NH 2 group of Dab3 in BS1-bound AcPMX, following the removal of Dab1 and Thr2 , to obtain the starting dimer0. After a geometry minimization of the complex, all the ligand-receptor interactions predicted by docking remained unaltered. The Residue Scan (RS) tool implemented in the MOE (Molecular Operating Environment) was then used for optimizing the sequence of the dimer0.
bioisostere due to its stability, low flexibility, and synthetic feasibility by click-reaction [46]. Subsequently, the linkage between the two monomers was obtained by an amide bond between the side chain carboxylate of Trz and the α-NH2 group of Dab3′ in BS1-bound AcPMX, following the removal of Dab1′ and Thr2′, to obtain the starting dimer0. After a geometry minimization of the complex, all the ligand-receptor interactions predicted by docking remained unaltered. The Residue Scan (RS) tool implemented in the MOE (Molecular Operating Environment) was then used for optimizing the sequence of the dimer0.  RS computes the free energy difference upon the mutation of selected residues with any other natural or non-natural amino acid included in the MOE libraries. Mutations can be single (one residue is mutated at a time) or multiple (up to five residues are mutated at a time). After the mutation, the complex was subjected to short restrained molecular dynamics (MDs) to include the effect of side chain flexibility in the estimation of the change in affinity (∆Affinity) and in the overall complex stability (∆Stability). Since one of our objectives was to reduce the global positive charge of the ligand, we focused our attention on the Dab residues that were not involved in optimal interactions with the receptor. The apolar Leu7 and Leu7 residues were also included in the scan to improve the ∆Affinity and ∆Stability by maximizing the polar or apolar interactions. The full list of mutations evaluated by RS is given in Figure 3D. Two RS cycles were carried out, the first requesting single point mutations only. The second cycle evaluated multiple mutations, but only of those amino acids that provided improvements in ∆Affinity and ∆Stability in the first RS run. More than 200 mutants were obtained, exhibiting negative values for both ∆Affinity and ∆Stability. Due to the time constraints, after visual inspection, we selected 10 high-ranked peptides that had ∆Affinity comprised between −16.7 and −12.7 kcal/mol and ∆Stability in the range of −3.8 and −0.8 kcal/mol, compared to dimer0 (Table S1 in Supplementary Materials). The corresponding complexes were then subjected to a 200 ns MD simulation to evaluate the dynamic stability of the complex (Figures S2-S7 in Supplementary Materials). Three heterodimers maintained interactions with both the BSs for the whole simulation, namely peptidomimetics 3, 7, and 10 ( Figure 3E and Supplementary Materials Figures S5-S7), and these were the first set of compounds selected for the chemical synthesis and biochemical evaluation.

Chemical Synthesis of Polymyxin-Based Peptidomimetics
In keeping with the topic of this work, we sought to create new BS1/BS2-targeting peptidomimetic heterodimeric molecules by covalently connecting PMX-based monomers. The designed linker contained quite a short and robust triazolyl-amide moiety, which suggested possible disconnections either along the triazole ring via Huisgen [3+2] cycloaddition of azide-and alkynoyl-ending monomers [47], or along the amide bond between carboxyland amine-terminating monomers.
Although the triazole-forming cycloaddition usually proved to be an appealing strategy [47], in the specific case of "short" linkers, such as those of designed target compounds 3, 7, and 10 ( Figure 4), it was not applicable because the required 3-butinoyl terminus of the BS1-monomer portion would not have survived the standard solid-and/or in-solution peptide synthesis conditions. As consequence, the synthesis was conducted as shown in Schemes 1 and 2. It is known, in fact, that homopropargyl amides may easily furnish allenamides even under mild coupling conditions, which, in turn, are optimal Michael acceptors with nucleophiles such as thiols and others (see Scheme 2, dashed squared formulas) [48]. For this reason, the disconnection of compound 3 (7 or 10) called for the late-stage amide coupling between the appropriate, protected triazolyl carboxylic acid 2 (6 or 9) and the protected amine monomer 1 (5 or 8), as illustrated in Figure 4.
To access the monomeric constituents, we opted to apply uniformed, mixed solidand in-solution phase procedures that could be directly applied to obtain all the firstgeneration and subsequent-generation products required [49]. As shown in Scheme 1 for the prototypic BS2-targeted protected monomer prot-2, the synthesis started with commercially available γ-Boc-protected (S)-α,γ-diaminobutanoic acid supported on 2chlorotrityl-functionalized polystyrene resin (compound I), constituting the Dab9 residue of the original PMX structure. Standard Fmoc-based SPPS (solid-phase peptide synthesis) followed to afford the fully protected linear peptide II, which incorporated six out of seven residues of the cycloheptapeptide core, as well as the tripeptide sequence of the lateral chain. During SPPS, the deprotecting reagent for the Fmoc group was 20% piperidine in DMF, and the coupling conditions entailed the use of HATU and collidine in DMF in the presence of HOAt. Importantly, Dde (N-1-(4,4-dimethyl-2,6-dioxocyclohexyliden)ethyl) was used as the orthogonal protecting group of the γ-amine side chain of Dab4 due to its stability in Fmoc-SPPS [49,50]. the use of HATU and collidine in DMF in the presence of HOAt. Importantly, Dde (N-1-(4,4dimethyl-2,6-dioxocyclohexyliden)ethyl) was used as the orthogonal protecting group of the γ-amine side chain of Dab4 due to its stability in Fmoc-SPPS [49,50].  Filled circles and numbering denote amino acid residues as in the PMX structure in Figure 2. Magenta and green colors used to indicate BS2-and BS1-recognizing monomers, respectively, in line with Figure 3; black circles are amino acid residues modified with respect to the original PMX sequence.
the use of HATU and collidine in DMF in the presence of HOAt. Importantly, Dde (N-1-(4,4dimethyl-2,6-dioxocyclohexyliden)ethyl) was used as the orthogonal protecting group of the γ-amine side chain of Dab4 due to its stability in Fmoc-SPPS [49,50].   After capping the free amine terminus within II via acetylation, the Dde of Dab4 was selectively removed by 4% NH2NH2 in DMF. Subsequently, Fmoc-azidoAla-OH (Fmoc-Azal) was coupled under usual solid-phase conditions, followed by Fmoc removal. The linear peptide was released from the resin with the cleavage cocktail DCM:AcOH:TFE (3:1:1), furnishing linear peptide III in a 76% yield from I. This linear peptide was subjected to in-solution cyclization under HATU/HOAt/collidine in DCM/DMF at 0.7 mM concentration. A cyclic azido-terminating compound was obtained in a 74% yield, which was finally "clicked" with 3-butynoic acid under copper(I)-catalyzed [3+2] cycloaddition conditions. After purification by reverse phase chromatography, the protected monomer prot-2 was isolated as a colorless glassy solid in a 57% yield.
For BS1-targeted monomer prot-1, we again started from the solid-phase construction of the linear peptide sequence, comprising residues of the heptapeptide ring as well as the sole 3′ residue of the lateral chain (Scheme 2). Thus, commercial resin-supported protected threonine IV, representing the Thr10′ residue of PMX-like prot-1, was the starting residue for Fmoc-based SPPS, which gave linear peptide V via the above mentioned Fmoc-deprotection/coupling reagents. However, to preserve the orthogonal Fmoc protection at Dab3′, milder conditions than those adopted for the synthesis of prot-2 were required to remove the Dde protection at the γ-amino function of Dab4′ (NH2OH·HCl, After capping the free amine terminus within II via acetylation, the Dde of Dab4 was selectively removed by 4% NH 2 NH 2 in DMF. Subsequently, Fmoc-azidoAla-OH (Fmoc-Azal) was coupled under usual solid-phase conditions, followed by Fmoc removal. The linear peptide was released from the resin with the cleavage cocktail DCM:AcOH:TFE (3:1:1), furnishing linear peptide III in a 76% yield from I. This linear peptide was subjected to in-solution cyclization under HATU/HOAt/collidine in DCM/DMF at 0.7 mM concentration. A cyclic azido-terminating compound was obtained in a 74% yield, which was finally "clicked" with 3-butynoic acid under copper(I)-catalyzed [3+2] cycloaddition conditions. After purification by reverse phase chromatography, the protected monomer prot-2 was isolated as a colorless glassy solid in a 57% yield.
For BS1-targeted monomer prot-1, we again started from the solid-phase construction of the linear peptide sequence, comprising residues of the heptapeptide ring as well as the sole 3 residue of the lateral chain (Scheme 2). Thus, commercial resin-supported protected threonine IV, representing the Thr10 residue of PMX-like prot-1, was the starting residue for Fmoc-based SPPS, which gave linear peptide V via the above mentioned Fmocdeprotection/coupling reagents. However, to preserve the orthogonal Fmoc protection at Dab3 , milder conditions than those adopted for the synthesis of prot-2 were required to remove the Dde protection at the γ-amino function of Dab4 (NH 2 OH·HCl, instead of NH 2 NH 2 ). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material). To explore even further the chemical space covered by PMs, we next synthesized three additional heterodimers, namely, compounds 4, 14, and 16 ( Figure 5). Compound 4 differs from dimer 3 just in its spacer length, as it possesses one more methylene unit separating the triazole ring from the carboxyl moiety. From the previous in silico studies (vide supra), it emerged that the two PMX-ring moieties targeting the BS1/BS2 regions were spatially very close, and they tolerated little variation, if any, in the spacer length between them. An evaluation of "elongated" dimer 4 was expected to shed light on this matter.  instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).   instead of NH2NH2). The linear peptide VI was then detached from resin (overall 70% yield for the SPPS from IV) and subjected to in-solution cyclization (87% yield) to consign, after Fmoc-deprotection, the required amine prot-1 in a 79% yield (corresponding to 48% global yield for the overall sequence from IV).
The final amide coupling between carboxylic acid prot-2 and amine prot-1 proceeded using the usual HATU/HOAt reagent coupling to furnish, after global acidic deprotection, heterodimer 3 (Scheme 3). After purification by reverse-phase chromatography, heterodimer 3 was obtained in a 25% yield as a glassy solid. Compound 3 exhibited m/z (M+4H) 4+ 568.3075, which was consistent with the calculated exact mass m/z (M+4H) 4+ 568.3067 and corresponded to 2269.3 Da molecular weight (Table 1). Heterodimers 7 and 10 were obtained in an analogous manner by starting from monomers prot-6/prot-5 and prot-9/prot-8, respectively ( Figure 4, Table 1, and synthetic procedures in the Supplementary Material).  To explore even further the chemical space covered by PMs, we next synthesized three additional heterodimers, namely, compounds 4, 14, and 16 ( Figure 5). Compound 4 differs from dimer 3 just in its spacer length, as it possesses one more methylene unit separating the triazole ring from the carboxyl moiety. From the previous in silico studies (vide supra), it emerged that the two PMX-ring moieties targeting the BS1/BS2 regions were spatially very close, and they tolerated little variation, if any, in the spacer length between them. An evaluation of "elongated" dimer 4 was expected to shed light on this matter. Another issue of interest concerned the number of basic and, at physiological pH, protonated Dab residues and how this relative proportion influenced the ability of the PMs to recognize the spike protein. For this reason, compounds 14 and 16 featured fewer basic isosteres of Dab, namely, homoserine residues (Hse), instead of certain Dab moieties. In particular, these two compounds share the common BS2-targeted monomer 13 with two Hse residues instead of the original Dab residues (Hse8 and Hse9); however, they differ in the number of Hse moieties in the corresponding BS1-components (one Hse3′ in monomer 12 and two Hse3′/Hse9′ in monomer 15).
The chemical synthesis of products 4, 14, and 16 followed the same path as above, and arose from the ligation of the respective monomers prot-1/prot-2′, prot-12/prot-13, and prot-15/prot-13 ( Figure 5; for details on the synthesis and characterization of these products, see Table 1 and the Supporting Information). It must be pointed out that Hsecontaining heterodimers 14 and 16 exhibited a highly lipophilic character, with consequent issues in solubility and problems encountered in both the HPLC purification step and during the biological assays (vide infra). Nonetheless, our uniformed, modular synthetic strategy proved reliable, and this supported our claim that it can be applied to nu- Another issue of interest concerned the number of basic and, at physiological pH, protonated Dab residues and how this relative proportion influenced the ability of the PMs to recognize the spike protein. For this reason, compounds 14 and 16 featured fewer basic isosteres of Dab, namely, homoserine residues (Hse), instead of certain Dab moieties. In particular, these two compounds share the common BS2-targeted monomer 13 with two Hse residues instead of the original Dab residues (Hse8 and Hse9); however, they differ in the number of Hse moieties in the corresponding BS1-components (one Hse3 in monomer 12 and two Hse3 /Hse9 in monomer 15).
The chemical synthesis of products 4, 14, and 16 followed the same path as above, and arose from the ligation of the respective monomers prot-1/prot-2 , prot-12/prot-13, and prot-15/prot-13 ( Figure 5; for details on the synthesis and characterization of these products, see Table 1 and the Supporting Information). It must be pointed out that Hsecontaining heterodimers 14 and 16 exhibited a highly lipophilic character, with consequent issues in solubility and problems encountered in both the HPLC purification step and during the biological assays (vide infra). Nonetheless, our uniformed, modular synthetic strategy proved reliable, and this supported our claim that it can be applied to numerous structural variants.

Binding Affinity of Polymyxin-Based Peptidomimetics to Spike RBD by SPR Analysis
To evaluate the biological activity of the synthesized peptidomimetics, we first addressed the binding potential of both heterodimeric and monomeric compounds to the spike protein. More specifically, SPR was used to measure the interaction of compounds (analytes) with immobilized RBD of the Wuhan-Hu-1 SARS-CoV-2 spike protein, as previously reported [51]. This method is useful for measuring the kinetic parameters k a (association rate constant) and k d (dissociation rate constant), as well as the binding affinity K D (dissociation equilibrium constant). The RBD was immobilized on a sensor chip using standard amine coupling. Different concentrations of peptidomimetic compounds (in a 1:2 serial dilution range) were, in turn, passed over the chip surface, and the respective signal changes for each compound were recorded (in Resonance Units (RU)). The resulting data were seen to fit a 1:1 binding model, and the binding affinity and kinetic parameters were determined, as detailed in Figure 6, Table 2, and Supplementary Material Figure S8. To evaluate the biological activity of the synthesized peptidomimetics, we first addressed the binding potential of both heterodimeric and monomeric compounds to the spike protein. More specifically, SPR was used to measure the interaction of compounds (analytes) with immobilized RBD of the Wuhan-Hu-1 SARS-CoV-2 spike protein, as previously reported [51]. This method is useful for measuring the kinetic parameters ka (association rate constant) and kd (dissociation rate constant), as well as the binding affinity KD (dissociation equilibrium constant). The RBD was immobilized on a sensor chip using standard amine coupling. Different concentrations of peptidomimetic compounds (in a 1:2 serial dilution range) were, in turn, passed over the chip surface, and the respective signal changes for each compound were recorded (in Resonance Units (RU)). The resulting data were seen to fit a 1:1 binding model, and the binding affinity and kinetic parameters were determined, as detailed in Figure 6, Table 2, and Supplementary Material Figure S8. . Data are shown as black lines, and the best fit of the data to a 1:1 binding model is shown in red. One representative sensorgram from two independent experiments is shown for each compound. See Table 2 for kinetics values.
3.09 ± 0.59 6.83 ± 1.58 2.31 ± 0.95  Table 2 for kinetics values. For several compounds, we clearly measured binding to the RBD, albeit with a rather fast release (off rate). In particular, heterodimers 7 and 10 exhibited interesting K D values in the low µMolar range (7, 2.31 µM; 10, 2.78 µM), which proved superior, in terms of binding affinity, to those observed for their respective monomers. In fact, monomers 5 and 6-constituents of 7-showed a detectable yet returning binding ( Figure 6, Table 2), whereas monomers 8 and 9-constituents of 10-behaved differently, with 8 displaying an interesting 8.56 µM binding affinity and 9 being devoid of any meaningful RBD binding.
An SPR evaluation of dimers 3 and 4, on the other hand, did not produce reliable data because of the aspecific interaction with the matrix of the chip, despite the good affinity displayed by their monomeric components (1, 10.12 µM; 2, 9.62 µM) ( Figure S8, Supplementary Material).
Moreover, homoserine-containing compounds 12-16, chosen for their lower number of basic Dab residues, did not prove to be good RBD binders either. Among them, dimer 14 exhibited a jagged signal, possibly due to scarce solubility and its tendency to give aggregates in the medium used.

Antiviral Effect of PMX-Inspired Peptidomimetics against SARS-CoV-2
To further explore the therapeutic potential of the PMs, we next evaluated the dimeric and monomeric compounds in our antiviral assays for SARS-CoV-2. First, the highly permissive glioblastoma cell line U87.ACE2 + (that stably expresses the ACE2 receptor) was used for the infection with live SARS-CoV-2 (strain 20A.EU2), as previously reported [52]. As summarized in Table S2 (Supplementary Materials), the compounds did not exert cytotoxic effects up to 50 µM when tested in the U87.ACE2 + cells without virus (CC 50 > 50 µM), nor could they fully protect the cells from being infected with SARS-CoV-2 (IC 50 > 50 µM). However, dimer 10 had some minor antiviral effects at the highest concentration (50 µM) tested (11% and 2% inhibition of viral cytopathic effect in two independent experiments). Secondly, we investigated the antiviral effect of our compounds in lung epithelial A549 cells stably expressing ACE2 and TMPRSS2, accommodating a plasma membrane entry route of the virus. Similarly, treatment of the A549.ACE2 + .TMPRSS2 + cells with the compounds did not fully prevent the cells from being infected with a Delta strain of SARS-CoV-2 (Table S2 of Supplementary Materials). Interestingly, for compounds 7 and 8, a hint of antiviral effect was measured at 50 µM concentration, whereas for dimer 10, some detectable inhibition was recorded at 20 and 50 µM treatment (Figure 7).

Discussion
The present work was founded on a previous in silico study [42], which suggested that the naturally occurring antibiotic polymyxin B (PMX) was a prime fit for two discrete and non-overlapping regions of the S-RBD:ACE interface, namely, BS1 and BS2. The PMX structure was adopted as an ideal peptidomimetic (PM) starting platform, upon which new PM molecules could be built up for exploring the chemical space around the targeted PPI. Interestingly, PMX and analogues have long been the focus of intense research into the creation of new antibiotics that are effective at overcoming antimicrobial resistance [44,45]; however, to the best of our knowledge, this is the first example where PMX is used as a starting scaffold to construct PMs (resembling yet structurally different from PMX) to combat SARS-CoV-2.
In silico design entailed three levels of intervention to the original PMX structure at the (i) marked shortening of the lipidic side chain into an acetyl moiety to diminish the overall lipophilic character and reduce conformational freedom; (ii) the optimization of the calculated affinity and stability of the complex with the RBD via systematic mutation of non-crucial amino acid residues; and, importantly, (iii) the "heterodimerization" of the PMX structure by connecting two diverse PMX-based cores via a robust linker, with the clear intent of homing in on the two BS1 and BS2 subregions simultaneously [53].
Use of the so-called "discontinued peptides" which span non-contiguous yet spatially close hACE2 fragments, is a good tactic for identifying inhibitors of the SARS-CoV-2 S:hACE2 interface [30,32,54]; this offers small-or medium-sized molecular constructs the opportunity to effectively increase their coverage of the spatially extended targeted interface.
The designed PMs were chemically synthesized via mixed solid-phase and in-solution synthesis, which led to an unprecedented collection of PMX-based heterodimers and monomers with molecular weights ranging from 1 KDa to 2 KDa. Within this collection, two heterodimers, namely, compounds 7 and 10, exhibited the best binding affinity determined by SPR experiments toward S-RBD, respectively, 2.31 μM and 2.78 μM, which also proved at least three times greater than that of their constituting monomers. This result validated the overall design and molecular modeling studies and corroborated the hypothesis that the simultaneous coverage of the BS1 and BS2 S-RBD subregions by heterodimerization was beneficial for the binding interaction. None of the synthesized peptidomimetics were able to completely inhibit infection of authentic SARS-CoV-2 (strains 20A.EU2 and Delta) in human glioblastoma U87.ACE + cells or in human lung epithelial A549.ACE2.TMPRSS2 + cells up to 50 μM concentrations. However, a detailed analysis of the antiviral effect of the compounds revealed some minor protection of the dimers at the

Discussion
The present work was founded on a previous in silico study [42], which suggested that the naturally occurring antibiotic polymyxin B (PMX) was a prime fit for two discrete and non-overlapping regions of the S-RBD:ACE interface, namely, BS1 and BS2. The PMX structure was adopted as an ideal peptidomimetic (PM) starting platform, upon which new PM molecules could be built up for exploring the chemical space around the targeted PPI. Interestingly, PMX and analogues have long been the focus of intense research into the creation of new antibiotics that are effective at overcoming antimicrobial resistance [44,45]; however, to the best of our knowledge, this is the first example where PMX is used as a starting scaffold to construct PMs (resembling yet structurally different from PMX) to combat SARS-CoV-2.
In silico design entailed three levels of intervention to the original PMX structure at the (i) marked shortening of the lipidic side chain into an acetyl moiety to diminish the overall lipophilic character and reduce conformational freedom; (ii) the optimization of the calculated affinity and stability of the complex with the RBD via systematic mutation of non-crucial amino acid residues; and, importantly, (iii) the "heterodimerization" of the PMX structure by connecting two diverse PMX-based cores via a robust linker, with the clear intent of homing in on the two BS1 and BS2 subregions simultaneously [53].
Use of the so-called "discontinued peptides" which span non-contiguous yet spatially close hACE2 fragments, is a good tactic for identifying inhibitors of the SARS-CoV-2 S:hACE2 interface [30,32,54]; this offers small-or medium-sized molecular constructs the opportunity to effectively increase their coverage of the spatially extended targeted interface.
The designed PMs were chemically synthesized via mixed solid-phase and in-solution synthesis, which led to an unprecedented collection of PMX-based heterodimers and monomers with molecular weights ranging from 1 KDa to 2 KDa. Within this collection, two heterodimers, namely, compounds 7 and 10, exhibited the best binding affinity determined by SPR experiments toward S-RBD, respectively, 2.31 µM and 2.78 µM, which also proved at least three times greater than that of their constituting monomers. This result validated the overall design and molecular modeling studies and corroborated the hypothesis that the simultaneous coverage of the BS1 and BS2 S-RBD subregions by heterodimerization was beneficial for the binding interaction. None of the synthesized peptidomimetics were able to completely inhibit infection of authentic SARS-CoV-2 (strains 20A.EU2 and Delta) in human glioblastoma U87.ACE + cells or in human lung epithelial A549.ACE2.TMPRSS2 + cells up to 50 µM concentrations. However, a detailed analysis of the antiviral effect of the compounds revealed some minor protection of the dimers at the highest tested concentration. Especially for dimer 10, this limited antiviral effect was consistently detectable in both antiviral assays (in which a different combination of virus and cell line was used). This biological effect seems to correlate well with the binding potential determined by our SPR analysis. It seems plausible that, within the quite dynamic and complex cellular environment, the low micromolar binding affinity of our best candidates toward monomeric S-RBD is not enough to warrant efficient competitive neutralization of the trimeric spike in authentic virus particles. Of note, none of the synthesized PMs exerted cytotoxicity up to 50 µM when tested in the non-infected U87 and A549 cell lines (Supplementary Table S2).
In conclusion, heterodimeric compounds 7 and 10 provide a first map of the chemical space around a simplified PMX-resembling core needed to address the S-RBD:hACE2 interaction and they may serve as a useful guide in the search for structurally optimized molecules with an improved S-RBD affinity and anti-SARS-CoV-2 potential. Thus, the rational variation of the constituting amino acid residues, as well as turning the heterodimeric core into a multivalent entity (e.g., trimerization), could be promising options. Finally, such optimized S-RBD binders could serve as directing units within designed covalent conjugates for the targeted delivery of drugs or visualizing tools in the environment of SARS-CoV2-infected cells.

Molecular Modeling
The model of the spike receptor-binding domain (RBD) was prepared as previously described [42]. An initial ligand was designed in MOE [55] based on the structure of polymyxin B where the 6-methyloctanoyl group was replaced by acetyl (AcPMX). AcPMX was subjected to a low-mode conformational search [56] using the Amber10EHT force field and the Born solvation model for water, as implemented in MOE. A total of 19 unique conformations were obtained within an energy range of 6.9 kcal/mol. AcPMX was then docked to the RBD. Two separate runs were carried out, which directed the docking onto each of the two potential binding sites, BS1 and BS2, previously identified by computational alanine scanning [42]. The docking was performed with MOE, using the Triangle Matcher algorithm and the London dG scoring, followed by Rigid Receptor refinement of the top 1000 poses generated in the previous step and rescoring with the GBVI/WSA function [57]. To maximize the sampling, the docking was repeated with each of the previously obtained 19 conformations of AcPMX. The lowest energy poses for both BS1 and BS2 were used to design the dimeric ligand. The two monomers were modified for conjugation within the receptor, keeping all the ligand-receptor interactions generated by docking unaltered. With its optimal side chain length, flexibility and functionalization, and synthetic feasibility, the (S)-2-amino-3-[4-(carboxymethyl)-1H-1,2,3-triazol-1-yl]propanoic acid (Trz) amino acid was designed to be the linker. Thus, the BS2 monomer was designed from AcPMX by mutating the cyclic Thr to Trz (AcPMX-Trz). The BS1 monomer was obtained by removing the first diaminobutyric acid (Dab) and Thr from the linear peptide chain, thus leaving the α-amino group of the remaining Dab (Dab3) as the anchoring point (Dab-PMX). The starting dimer (dimer0) was then obtained by linking AcPMX-Trz and Dab-PMX through an amide bond between the Trz sidechain carboxylate and the Dab α-NH 2 . The dimer0 sequence was Ac-Dab1-Thr2-Dab3-γDab4 ‡-Dab5-(D)-Phe6-Leu7-Dab8-Dab9-Trz ‡-Dab3 -γDab4 #-Dab5 -(D)-Phe6 -Leu7 -Dab8 -Dab9 -Thr10 #, where the ‡ and # symbols identify residues involved in cyclization. More than 200 mutants were obtained having negative ∆Affinity and ∆Stability. The 10 best-ranked mutants for both ∆Affinity and ∆Stability were selected for further evaluation by molecular dynamics (MD). Restricted electrostatic potential (RESP) charges were computed for non-natural amino acids (Trz, linear Dab, and cyclic γDab) using the R.E.D. tool [58] by following a well-established protocol [59][60][61]. Each complex was solvated by an octahedral TIP3P water box [62], extending up to 10 Å from the protein, and chlorine ions were added to neutralize the total charge. The ff14SB force field [63] was used for the simulations, and missing parameters for Trz were taken from gaff2 force field [64]. Multiple equilibration steps at constant volume and temperature (NVT) and constant pressure and temperature (NPT) ensembles, were carried out at a final temperature of 300 K, as detailed in previous works [65][66][67]. A production run of 200 ns was then performed, and trajectories were analyzed by computing the root mean squared deviation vs. time. MD simulations and trajectory analyses were performed with the Amber20 and AmberTools21 packages [68]. Three peptides remained stable within the receptor-binding site, namely compounds 3, 7, and 10, which were selected as the first set for synthesis and biochemical evaluation. Variants of these peptides were later designed by following the same strategy as above.
In-solution cyclization. The cyclization was performed as described for compound prot-1, starting from linear peptide III (75.5 mg, 0.049 mmol, and 1 eq). After completion, the solution was concentrated under vacuum. The crude product was purified by reverse phase chromatography (eluent: from 90: 10  Click reaction. To a solution of the previous compound (20.5 mg, 0.013 mmol, and 1 eq) and 3-butynoic acid (3.2 mg, 0.038 mmol, and 3 eq) in DMF (2.4 mL), a solution of Cu(OAc) 2 (0.75 mg, 0.004 mmol, and 0.3 eq) and sodium ascorbate (1.5 mg, 0.008 mmol, and 0.6 eq) in water (1 mL) was added. The reaction was left under stirring under argon atmosphere after 3 cycles of argon/vacuum. After 6 h, the solvent was removed under reduced pressure and the residue was washed with water (3×) and diethyl ether (3×). The crude product was purified by reverse phase chromatography (eluent: from 90:10 H 2 O + 0.1% TFA:ACN to 100% ACN) affording the cyclopeptide prot-2 as a colorless glassy solid ( To a solution of compound prot-2 (6.4 mg, 0.004 mmol, and 1 eq), HATU (2.9 mg, 0.008 mmol, and 2 eq), HOAt (1.0 mg, 0.008 mmol, and 1 eq), and 2,4,6-collidine (1.8 µL, 0.011 mmol, and 3 eq) in DMF (0.4 mL), a solution of prot-1 (7.4 mg, 0.005 mmol, and 1.2 eq) in DMF (0.1 mL) was added. The reaction was left to stir at room temperature under nitrogen atmosphere for 6 h. Subsequently, the solvent was removed under reduced pressure and the crude was washed with water (5×). The resulting crude was then treated with a solution of TFA/TIS/H 2 O (95:2.5:2.5) (209 µL). After 1 h, the solvent was removed under reduced pressure and the solid residue was washed with Et 2 O (3×). The resulting crude was purified by reverse phase chromatography (eluent: from 90:10 H 2 O + 0.1% TFA:ACN to 100% ACN) affording the cyclopeptide BS1/BS2-dimer 3 as a colorless glassy solid (2.8 mg, 25% yield for 2 steps). HRMS (ES + ) m/z calcd for C 110  SPR technology (Biacore T200, Cytiva, Marlborough, MA, USA) was used to determine the binding kinetics and affinity of the peptidomimetics to the RBD of the wild-type Wuhan-Hu-1 SARS-CoV-2 spike protein (2019-nCoV spike RBD, cat n • 40592-VNAH, SinoBiological, Beijing, China). The RBD was immobilized on a CM5 sensor chip using standard amine coupling in 10 mM HEPES (pH 7.0) at a level between 5000 and 7000 RU. Interaction studies between peptidomimetics and the RBD were performed at 25 • C in HBS-P + (10 mM HEPES, 150 mM NaCl, and 0.05% surfactant P20; pH 7.4). Two-fold serial dilutions of peptidomimetics were injected at 30 µL/min using multiple cycle kinetics. Totals of 10 mM NaOH and 10 mM Glycine-HCl (pH 1.7) were used to regenerate the RBD and antibody surface, respectively. Several buffer blanks were included for double referencing. Apparent binding kinetics (K D , k a , and k d ) were derived after fitting the experimental data to the 1:1 Langmuir binding model using the Biacore T200 Evaluation Software 3.1. The experiments were performed at least in duplicate.
Viruses. All virus-related work was conducted in the high-containment biosafety level 3 facilities of the Rega Institute at the Katholieke Universiteit (KU) Leuven (Leuven, Belgium), in accordance with institutional guidelines. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) isolates were recovered from nasopharyngeal swabs of RT-qPCR-confirmed human cases obtained from the University Hospital (Leuven, Belgium). SARS-CoV-2 viral stocks were prepared by inoculation of confluent Vero E6 cells or A549.ACE2 + .TMPRSS2 + , as described in detail [69] Titers were determined by tissue culture infectious dose 50 (TCID 50 ) method of Reed and Muench on A549.ACE2 + .TMPRSS2 + and U87.ACE2 + cells. Viral genome sequence was verified, and all infections were performed with passage 2 to 5 virus.

Wild-Type Virus Infection and Antiviral Assays
One day prior to the experiment, A549.ACE2 + .TMPRSS2 + (Invivogen, San Diego, CA, USA) or U87.ACE2 + cells [52] were seeded in 96-well microtiter plates. The following day, 2.5-or 5-fold serial dilutions of the test compounds were prepared in virus infection media (similar to cell culture medium, but with reduced FBS content), overlaid on cells, and virus was added to each well. Cells were incubated at 37 • C under 5% CO 2 for the duration of the experiment. At endpoints (4 or 5 days p.i.), the virus-induced cytopathic effect (CPE) was microscopically evaluated. Inhibition was calculated by comparison to virus control wells with no inhibitor added. IC 50 values were determined by interpolation. Four (U87.ACE2 + cells) or five (A549.ACE2 + .TMPRSS2 + cells) days after infection, the cell viability of mock-and virus-infected cells was assessed spectrophotometrically via the in situ reduction of 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxy-methoxyphenyl)-2-(4sulfophenyl)-2H-tetrazolium inner salt, using the CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega, Madison, WI, USA). The optical density (OD) of the samples was compared to appropriate cell control replicates (cells without virus and drugs) and virus control wells (cells with virus but without drugs). The concentration that inhibited SARS-CoV-2-induced cell death by 50% (IC 50 ) was calculated from interpolation.