Main

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has infected over 628 million people worldwide resulting in over 6.6 million deaths as of October 2022 (ref. 1). Multiple SARS-CoV-2 variants with increased infectivity have emerged, which jeopardize the utility of current vaccines and therapeutic antibodies. Although several monoclonal antibody (mAb) therapeutics have received emergency use authorization and demonstrated efficacy in patients, they are limited by high production cost, global supply issues and inconvenient routes of administration2. In addition, many human-derived mAbs have shown reduced efficacy against newly evolved viral variants3.

In this article, we report the engineering and characterization of two highly potent and broadly neutralizing synthetic proteins, FSR16m and FSR22, as candidates for preventing and treating coronavirus disease 2019 (COVID-19). The active domains of FSR16m and FSR22 are designed ankyrin repeat proteins (DARPins) engineered to mimic human angiotensin-converting enzyme 2 (hACE2) binding to the receptor-binding domain (RBD) of the spike protein. DARPin is a versatile synthetic binder scaffold that has high thermostability and has been engineered to bind an array of targets with pico- to nanomolar affinities4,5, including the SARS-CoV-2 spike protein6. Trimerization of RBD-binding DARPins SR16m and SR22 with a T4 foldon7 molecule increased their neutralization potency by >300-fold. Remarkably, and despite using the Wuhan-1 historical isolate spike protein as the target of engineering, both FSR16m and FSR22 exhibit substantially enhanced neutralization potency against a panel of SARS-CoV-2 variants of interest (VOI) and variants of concern (VOC) relative to the ancestral virus in pseudovirus assays. Cryo-electron microscopy (cryo-EM) studies confirmed that both SR16m and SR22 target an essential ACE2-binding epitope on the RBD, providing support for the idea that these DARPins may remain effective against future SARS-CoV-2 variants. Although the improvement in neutralization potency against some newly emerged viral variants was not built into our DARPins design, the broad-spectrum activity is a direct result of our engineering approach that combines protein trimerization to increase avidity with the selection of DARPins that can compete with the viral receptor (that is, hACE2).

Results

Engineering of FSR16m and FSR22

Employing an in-house DARPin library with >109 distinct clones displayed on M13 phage particles, we sequentially enriched binders to biotinylated RBD and the full-length spike protein (Wuhan-1 strain) over four rounds of phage panning (Supplementary Fig. 1). The enriched DARPin library pool from the final round was cloned into an expression vector with 6xHis and Myc tags at the N-terminus and transformed into BL21(DE3) Escherichia coli cells. Three hundred colonies were picked and grown in 96-well plates, and the cell lysate was subjected to enzyme-linked immunosorbent assay (ELISA)-based screens. Eleven unique clones bound strongly to both the RBD and the full-length spike protein and were purified by nickel-affinity chromatography. A competition ELISA was used to identify DARPin molecules that could inhibit binding between hACE2 and spike proteins and yielded two unique clones: SR16 and SR22. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS–PAGE) analysis of these proteins revealed a homogenous band for SR22 and an additional band for SR16 (Supplementary Fig. 2). The N-terminal 6xHis tag in SR16 was moved to the C-terminus to give rise to SR16m.

Because the SARS-CoV-2 spike protein adopts a trimeric structure, we trimerized both SR16m and SR22 through fusion to a T4 foldon7, a small trimeric protein, with a flexible (GGGGSLQ)x2 linker to form FSR16m and FSR22 (Supplementary Fig. 3). These trimeric DARPins should bind the spike protein with much higher avidity. Subsequently, using lentiviruses pseudotyped with the Wuhan-1 spike protein, we determined the neutralization activity of SR16m and SR22. Monomeric SR16m and SR22 showed weak inhibitory activity against the historical Wuhan-1 virus with 50% inhibitory concentration (IC50) values of 1 and 0.7 μM, respectively, which were dramatically enhanced upon trimerization (Fig. 1a and Table 1). Despite the use of the Wuhan-1 spike protein during DARPin engineering, both FSR16m and FSR22 exhibited greater neutralization activity toward viruses pseudotyped with spike proteins from VOCs. As an example, the IC50 values of FSR16m and FSR22 against pseudoviruses displaying the RBD from the Omicron variant are 8.5 and 7.3 pM, respectively, which are substantially more potent than toward viruses displaying Wuhan-1 spike protein (331 pM and 1.8 nM, respectively).

Fig. 1: Neutralization of spike protein pseudotyped lentiviruses.
figure 1

a, Illustration of the engineered monomeric and trimeric DARPin molecules and their neutralization of pseudotyped lentiviruses. B.1.1.529* harbors a chimeric spike protein in which the RBD region (residue, 338–514) of B.1.351 was replaced with that from B.1.1.529 (BA.1). Residues in red were randomized during DARPin engineering. Data points are shown as mean ± s.d. from at least two independent experiments. b, SDS–PAGE gel analysis of the DARPin molecules under nonreducing and reducing conditions. The gel image was representative of three independent experiments. c, Size-exclusion chromatography of DARPin molecules under nonreducing conditions. The chromatograph was representative of two experiments.

Source data

Full size image
Table 1 Summary of the neutralization IC50 values in Fig. 1a
Full size table

All monomeric and trimeric DARPin proteins were efficiently expressed in E. coli and easily purified by one-step immobilized metal affinity chromatography. The presence of surface Cys residues (Supplementary Fig. 3) did not impact the tertiary structure of these proteins (Fig. 1b). Monomeric DARPin SR16m migrated slightly faster than SR22 on SDS–PAGE gel and eluted later on size exclusion chromatography, while their respective trimers had nearly identical size (Fig. 1b,c). FSR16m was homogenous in solution, whereas a small percentage of FSR22 appeared to aggregate under nonreducing conditions (Fig. 1c).

FSR16m and FSR22 bind diverse RBD variants with high avidity

FSR16m exhibited superior binding avidity than FSR22 against a panel of spike proteins from VOC and VOI (Fig. 2a and Supplementary Fig. 4) in a biolayer interferometry binding assay. Due to the negligible dissociation rate, the KDapp values of FSR16m for the RBD of Wuhan-1 and six viral variants were below the detection limit (KDapp < 1 pM). In the case of Omicron RBD, a faster dissociation rate was observed, resulting in KDapp of 3.65 nM. This result contrasted with our neutralization assay in which the FSR16m exhibited improved IC50 values against lentiviruses displaying RBD from the Omicron variant compared to Wuhan-1 virus. The difference in the apparent binding avidity may be due to differences in the tertiary conformation of the protein in solution and on the virus. The binding study used dimeric Fc-tagged RBD. FSR22 maintained a similar binding avidity toward all tested RBDs with a slight increase in avidity to the RBD of Omicron (KDapp = 2.63 nM) compared to Wuhan-1 (KDapp = 12.3 nM). Overall, a faster dissociation rate was observed for FSR22 than FSR16m.

Fig. 2: FSR16m and FSR22 maintain high avidity toward SARS-CoV-2 variants.
figure 2

a, Binding kinetics to selected variant RBD proteins. Each data curve is from one biosensor at indicated time points, and six concentrations were tested in one representative experiment. b, ELISA binding titration to a panel of 24 RBD variants. Data points are shown as mean ± s.d. from one representative experiment with two technical replicates. c, Summary of the 50% maximum effective binding concentrations (EC50) to RBD proteins with indicated variants. The EC50 values are the means of duplicate wells from two independent experiments.

Source data

Full size image

To assess the breadth of FSR16m and FSR22 interaction with other SARS-CoV-2 VOC and VOI, we determined their binding to a panel of 24 RBD mutants by ELISA. Remarkably, FSR16m maintained nanomolar EC50 values toward all variants except for the ones with K417E and F486V/F486S substitutions (Fig. 2c). While the K417E variant was predicted to escape antibody neutralization based on the pseudotyped virus studies8, this amino acid change also resulted in reduced binding affinity of spike for hACE2 (17.8% of Wuhan-1 control)9 due to the loss of a critical salt bridge interaction between the positively charged K417 in the spike protein and the negatively charged Asp in hACE2. K417N/K417T substitutions are present in B.1.351 (β) and P.1 (γ) variants (Supplementary Table 1). The affinity of FSR16m for the B.1.351 RBD (K417N + E484K + N501Y, EC50 0.43 nM) is similar to Wuhan-1 RBD (EC50 of 0.59 nM) but slightly weaker than that for a variant with only E484K + N501Y (EC50 of 0.23 nM), indicating that the K417N mutant weakens the interaction slightly between FSR16m and the spike protein. Similarly, the variants F486V and F486S also exhibit decreased hACE2-binding affinity (37% and 57%, respectively, of Wuhan-1 control9). Consequently, variants with these mutations (for example, BA.4/.5) may exhibit reduced viral fitness and virulence. Mutations G476S and G446V enable resistance to neutralization by antibodies REGN-10933 and REGN-10987, respectively10. Nonetheless, and despite a greater than 10-fold increase in EC50 values, FSR16m still maintained nM binding avidity to spike proteins with both of these variants (Fig. 2c). However, no detectable interaction was observed between FSR16m and the RBD proteins of sarbecoviruses of clade 1a (SARS-CoV-1, RaTG13, WIV1), clade 2 (BtkY72) and clade 3 (Rs4081) using the same binding assay (Supplementary Fig. 5), indicating that the epitope of FSR16m is not conserved among other sarbecoviruses.

FSR22 retained high-avidity binding to most of the tested variants with EC50 < 10 nM although its binding strength was generally lower than FSR16m. Similar to FSR16m, FSR22 also exhibits lower avidity to RBDs harboring variants K417E, F486V/F486S, G446V or G476S. Beyond these, FSR22 had reduced avidity to RBDs containing a T478K substitution, pointing to its engagement of a slightly different binding interface than FSR16m. The ability of FSR16m and FSR22 to bind diverse RBD variants with high avidity may stem from our engineering approach, which aimed to identify binders that mimic hACE2 engagement, a step that the virus is obligated to maintain as high-affinity binding for efficient infection.

FSR16m and FSR22 neutralize authentic viruses

We next tested the capacity of FSR16m and FSR22 to neutralize authentic SARS-CoV-2 in Vero-hACE2-TMPRSS2 cells11. Both FSR16m and FSR22 DARPins efficiently inhibited infection of several authentic SARS-CoV-2 strains (Fig. 3a and Table 2). The IC50 values of FSR16m against B.1.351, B.1.617.2 and B.1.617.2.AY1 viruses were 3.4, 2.2 and 3.3 ng ml−1, respectively, values comparable to that of the Regen-COV antibody cocktail12. While both DARPins neutralized authentic Omicron strains, FSR16m was more potent. The IC50 values of FSR16m and FSR22 against BA.1, BA.1.1 and BA.2 strains were 44.7 and 169.2 ng ml−1, 7.4 and 41.2 ng ml−1, and 33.3 and 216.2 ng ml−1, respectively (Table 2).

Fig. 3: Neutralization of authentic SARS-CoV-2 viruses.
figure 3

a, FSR16m and FSR22 potently neutralized authentic SARS-CoV-2 variants. DARPins were incubated with 102 FFU of SARS-CoV-2 for 1 h followed by addition to Vero-hACE2-TMPRSS2 cells. One representative experiment with the mean of two technical replicates is shown. b–e, Intranasally administered FSR16m protected mice against infection by SARS-CoV-2 B.1.617.2. b, Weight change following infection with 103 FFU of SARS-CoV-2 B.1.617.2 via intranasal administration. Data points are shown as the mean ± s.e.m. from one representative experiment with six technical replicates; Welch’s two-tailed t-test of the AUC (c–e) Viral RNA levels at 7 dpi in the lung, heart and nasal wash (n = 6, two experiments). Two-tailed Mann–Whitney test, significant P values are denoted in the individual panel. f, Heat map of cytokine and chemokine protein expression levels in lung homogenates from the indicated groups. Data are presented as log2-transformed fold-change (FC) over naive mice. Blue, reduction; red, increase. Results are from two experiments.

Source data

Full size image
Table 2 Summary of neutralization IC50 values against authentic SARS-CoV-2 viruses in Fig. 3a
Full size table

Given the potency of FSR16m, we investigated the in vivo efficacy of this DARPin in mice. Eight-week-old female heterozygous K18-hACE2 C57BL/6J13 mice were administered 103 focus-forming units (FFU) of SARS-CoV-2 B.1.617.2 on day 0. This dose of SARS-CoV-2 was determined to cause severe lung infection and inflammation in previous studies14,15,16. On days 1 and 4 post infection, FSR16m (50 μg per mouse in PBS) was administered via an intranasal route. FSR16m-treated mice had less weight loss and 10- to 100-fold lower levels of viral RNA in the lung, heart and nasal wash than mice treated with PBS (Fig. 3b–e). Mice treated with FSR16m also showed lower levels of several proinflammatory cytokines compared to the control-treated, infected mice (Fig. 3f and Supplementary Fig. 6). Thus, FSR16m showed post exposure therapeutic efficacy against SARS-CoV-2 as an intranasally administered protein.

Cryo-EM structures of DARPin in complexes with SARS-CoV-2 spike

To understand the broadly neutralizing activity of FSR22 and FSR16m against SAR-CoV-2 variants, we determined their cryo-EM structures in complex with SARS-CoV-2 S 6P spike proteins17. Ab initio reconstructions of cryo-EM images of the complexes followed by heterogeneous refinements revealed trimeric SARS-CoV-2 spikes with a variety of RBD conformations, ranging from all RBD-down to one, two or three RBD-up conformations—with SR22 or SR16m binding to the RBD-up forms in all cases (Supplementary Fig. 7). As FSR22 and FSR16m, the trimeric forms of three SR22 or SR16m linked by a foldon were the most inhibitory, we focused on refining the three RBD-up conformations of SARS-CoV-2 spikes with the FSR22 (or FSR16m) bound to the tip of the RBD in its open conformation (Fig. 4a,b, Supplementary Figs. 8, 9 and Supplementary Table 2). While the trimeric spike complexes were generally well-defined, the tip of the RBD along with FSR22 (or FSR16m) showed substantial conformational heterogeneity. As a result, we could not delineate the FSR22 (or FSR16m)-RBD interface in atomic detail, even after extensive heterogeneous refinement followed by local refinement (Supplementary Figs. 8 and 9). The overall fold of the DARPins and the tip of the RBDs did, however, fit well, and these revealed both FSR22 and FSR16m to bind at the ‘tip’ of the RBD in their ‘RBD-up’ conformation (Fig. 4a–d).

Fig. 4: Cryo-EM structures of DARPins in complexes with SARS-CoV-2 spike.
figure 4

a, Cryo-EM map of FSR22 (magenta)-bound SARS-CoV-2 S 6P spike in its RBD-open conformation. b, cryo-EM map of FSR16m (orange)-bound SARS-CoV-2 S 6P spike. c, SR22-bound RBD in cartoon representation (left). The footprint of SR22 on the surface of RBD (gray) was shown in magenta (right). RBD residues that underwent mutations in VOCs were highlighted in cyan (right). d, SR16m-bound RBD in cartoon representation (left). The footprint of SR16m on the surface of RBD (yellow) was shown in orange (right). e, Amino acid residues of RBD contacting FSR22, FSR16m and ACE2 were highlighted in magenta, orange and green, respectively. Amino acid residues of RBD that underwent mutations in VOCs were denoted with an asterisk (*). ‘**’ and ‘***’ are multiples of ‘*’. Amino acid residues of RBD that form a core footprint of monomeric and trimeric DARPins were highlighted with rectangular boxes. f, ACE2-bound RBD was in cartoon representation (PDB: 7C8D) (left panel). The ACE2 footprint on RBD was highlighted in green (right). g, Overlay of SR22-bound RBD and ACE2-bound RBD structure (left). Overlay of footprints of ACE2, SR22 and SR16m core and RBD residues underwent mutations in the VOCs (right).

Full size image

Although the FSR22 structure in complex with SARS-CoV-2 spike was determined at slightly higher resolution than the FSR16m structure (Supplementary Figs. 8 and 9), the conformational mobility of these structures reduced their resolution. To corroborate our findings, we determined structures of these DARPin molecules in complex with RBD by itself. To provide sufficient mass for cryo-EM structure determination, we added the antigen-binding fragments (Fabs) from two antibodies (S309 and CR3022) that neither compete with each other nor with either DARPin molecule (Supplementary Figs. 10–12 and Supplementary Table 2). We obtained a structure of the quaternary complex with SR22 at 4.11 Å resolution, with density for the DARPin of similar quality as RBD and Fab variable domains (Supplementary Fig. 11); we also obtained a structure of the quaternary complex with SR16m at 4.26 Å resolution, where the slightly reduced resolution related to the propensity of SR16m to self-aggregate (Supplementary Fig. 12). Despite differences in trimerization (with either spike/RBD or trimeric/monomeric DARPin), the overall fold of the two DARPins and their binding mode to SARS-CoV-2 were nearly identical.

The four structures suggested that the DARPins recognize a subset of RBD surface composed of residues F456, A475, F486, N487 and Y489, which overlap closely with the hACE2-binding surface18,19 (Fig. 4e). Notably, most of these residues did not overlap with residues that changed in VOC, including N501Y in B.1.1.7 (Alpha), K417N, E484K and N501Y in B.1.351 (Beta), L452R and E484Q in B.1.617.2 (Gamma), and K417N, S477N, Q498R and N501Y in B.1.1.529 (Omicron) (Fig. 4e–g), suggesting that the surface recognized by FSR22 (or FSR16m) is important for ACE2 binding, and changes in these residues may compromise viral infectivity and fitness19. Comparison of the two DARPin-RBD complexes and the ACE2-RBD structure revealed a steric clash between a monomer of SR22 (or SR16m) and ACE2 bound to RBD, as the DARPins and ACE2 recognize overlapping binding surfaces (Fig. 4f,g). However, SR22 and SR16m only buried approximately 500 Å2 of surface area on RBD, whereas ACE2 binding buried approximately 970 Å2 of the surface on RBD, explaining the structural basis for the relatively poor neutralizing ability of SR22 and SR16m (monomers) as they could be outcompeted by ACE2 for RBD binding (Fig. 4f,g). Conversely, FSR22 and FSR16m, composed of three SR22 and SR16m, respectively, can overcome the lower affinity of monomeric DARPins to RBD by binding three RBDs in their open conformation (Fig. 4a,b). Collectively, the cryo-EM structures of FSR22 and FSR16m-bound SARS-CoV-2 spike revealed that FSR22 and FSR16m potently neutralize SARS-CoV-2 including VOCs by targeting a minimal but essential subset of ACE2-binding residues and taking advantage of avidity gained by trimerization.

Discussion

By using phage panning coupled with functional screening, we report the engineering of DARPins with potent and broad neutralization activity against SARS-CoV-2. Monomeric DARPin SR16m and SR22 exhibited weak viral neutralization potency, but this was dramatically enhanced upon trimerization by fusion with a T4 foldon. Although the Wuhan-1 spike protein was used for engineering, FSR22 and FSR16m exhibited substantially increased neutralization potency against newer viral variants.

Our cryo-EM structural analysis revealed that both FSR22 and FSR16m recognize a distinct region of the ACE2-binding surface on RBD, particularly residues 421, 456 and 485–489. Shang et al. demonstrated that mutation of residues 481–489 reduced ACE2 binding substantially19, and Starr et al. reported that muations in residues 421, 475 and 487 reduced RBD expression and possibly viral fitness, whereas mutations in residues 421, 456, 475, 486, 487 and 489 reduced ACE2-binding affinity9. The precise targeting of essential ACE2 interacting residues on RBD by FSR22 and FSR16m may explain their pan-neutralization of naturally derived SARS-CoV-2 variants, and our two DARPins may effectively prevent infection of variants that require hACE2 for entry.

The ability of these DARPins to exhibit increased neutralization potency toward SARS-CoV-2 variants contrasts with many human-derived anti-SARS-CoV-2 mAbs, which lose neutralization potency3. The following reasons may account for this phenomenon: (a) naturally derived viral variants are selected in part by their ability to evade population immunity (either through vaccination or natural infection)3. This contrasts with in vitro engineered DARPins whose potency appears to be less affected by the virus evolution history. (b) The dimeric structure of an antibody limits a maximum of two spike proteins within a spike trimer to be simultaneously engaged by a single antibody, necessitating a high affinity between the antibody and the spike protein and a close match of the binding interface. Even small changes at or near the binding interface could disrupt antibody-spike protein interactions and reduce the antibody binding and neutralization potency. In contrast, due to its much smaller size, a trimer DARPin can easily engage all three monomers in a spike trimer concurrently. The strong avidity effect from the trimer interaction can compensate for the weak binding affinity of the monomer and enable the trimers to tolerate variants at or near the binding interface without compromising potency. (c) DARPins SR16 and SR22 were selected based on their ability to compete with hACE2 for binding to the spike protein rather than their ability to inhibit viral infection. An effective way to inhibit ACE2 binding to spike is to occupy the interface for hACE2 binding, and indeed both DARPins engage key residues within the hACE2-binding interface. As emerging natural VOCs tend to exhibit higher infectivity and ACE2 binding affinity20,21,22, these variants may become more susceptible to binding and neutralization by our DARPin molecules.

The upper airway epithelium is the first site of infection by SARS-CoV-2 due in part to its robust expression of hACE2 (ref. 23). Infected upper airway epithelium cells release progeny viruses, which lead to infection of other organs including the lung. Antibody therapeutics have been an effective weapon for treating COVID-19, with several virus-neutralizing IgG antibodies approved for emergency use by the Food and Drug Administration24,25,26. However, antibodies have several limitations as therapeutics for treating SARS-CoV-2: (a) antibody production requires sophisticated mammalian cell culture and is expensive and suffers from limited global manufacturing capacity; (b) IgG antibody requires subcutaneous or intravenous routes of administration that may be inconvenient or impractical for some patients and care providers; (c) intravenously and subcutaneously administered antibodies have poor access to mucosal compartments with an estimated 50- to 100-fold lower antibody levels than in blood27,28,29, necessitating a high therapeutic dose and (d) many current COVID-19 antibody therapeutics have a narrow neutralization spectrum and require cocktails to maintain efficacy toward newly emerged SARS-CoV-2 variants.

Through the use of the stringent K18-hACE2 mouse model of SARS-CoV-2 pathogenesis, we showed that intranasal administration of FSR16m on days 1 and 4 post infection reduced viral burden and weight loss. As several other studies have demonstrated14,30,31,32, overexpression of hACE2 in these mice results in severe lung inflammation and disease upon infection with SARS-CoV-2. Therefore, the level of protection we observed by intranasal administration of FSR16m is significant. However, the current format of intranasal delivery likely is not directly translatable to humans due to limited lung exposure. Several protein-based antivirals are currently under development against COVID-19 (refs. 29,33,34,35). However, unlike FSR16m, which potently neutralized all tested VOCs, many of these have narrower spectrum of neutralization. FSR16m is efficiently expressed in E. coli (>200 mg per liter of shaker flask culture, on par with that reported previously36), enabling its cost-effective production on a large scale. In addition, FSR16m exhibits remarkable stability with <10-fold loss in activity after storage at room temperature for 6 weeks (Supplementary Fig. 13), consistent with the previously reported DARPin storage stability37 and making the molecule a promising therapeutic candidate.

Previously, two highly potent anti-SARS-CoV-2 hetero-trimeric DARPin molecules, MM-DC (MP0420, ensovibep) and MR-DC (MP0423), were reported6. Both DARPins exhibited picomolar IC50 values against spike protein pseudotyped vesicular stomatitis virus particles and were 10–50-fold more potent than their constituent monomer DARPins. This level of potency enhancement contrasts with both FSR16m and FSR22, whose IC50 values improved >300-fold upon homo-trimerization. When administered via intraperitoneal route at 10 mg kg−1, MR-DC reduced the viral load in nasal turbinates and lung by ~10 and ~100-fold, respectively38. A similar level of viral load reduction in these tissues was achieved by intranasally administered FSR16m at 2.5 mg kg−1 per animal. Both MM-DC and MR-DC are pentameric molecules composed of two albumin-binding DARPins and three spike-binding DARPins and have a half-life of 4–6 days in hamsters. Intravenously administered MM-DC reduced the risk of hospitalization by 78% in phase 2 clinical trial39. Although FSR16m and FSR22 in their current form are expected to have a short half-life in vivo and are not suitable for systematic application, the same albumin binding DARPin molecules can be fused to FSR16m to extend its in vivo half-life and make it suitable for intravenous application.

In summary, we report the engineering of two homotrimeric DARPin molecules, FSR16m and FSR22, with potent and broad SARS-CoV-2 neutralization activity. Both proteins are amenable to rapid and cost-effective production in E. coli, and intranasally administered FSR16m protected mice previously infected with SARS-CoV-2, suggesting that these proteins might be effective countermeasure candidates for COVID-19. Finally, our work suggests that homo-multimerization of monomeric proteins with moderate antiviral activity may represent an effective strategy for generating broadly neutralizing antiviral agents.

Methods

Selection and screening of spike protein-binding DARPin molecules

An in-house N3C DARPin library with >109 diversity was used in the phage panning essentially as described previously40. Purified full-length spike protein (BEI NR-52724) and RBD (BEI NR-52306) were biotinylated via EZ-link-sulfo-NHS-LC-biotin (Thermo Fisher Scientific, 21335) and used as target proteins. RBD was used as the target protein in Rounds 1, 2 and 4, while the full-length spike protein was used as the target in Round 3 to ensure the enrichment of DARPins that recognize RBD present on the full-length spike protein. Round 1 used the target protein (100 nM) in solution, whereas Rounds 2–4 employed decreasing concentrations of the target protein immobilized on streptavidin or neutravidin-coated ELISA plates (100, 50 and 20 nM). The enrichment of RBD binding DARPins was confirmed by phage ELISA against both RBD and full-length spike protein following a published protocol40 (Supplementary Fig. 1).

The enriched DARPin pool from the fourth round was cloned into the pET28a vector for high-level DARPin expression as described previously41. The resulting DARPin contains a Myc tag and a 6xHis tag at the N-terminus. After transformation, a total of 300 individual E. coli BL21 (DE3) clones were picked and grown in deep 96-well plates (1 ml per well) at 37 °C in Lysogeny broth (LB) and induced with isopropyl-β-d-thiogalactopyranoside (IPTG; 0.5 mM). The next day cell pellets were collected, resuspended in 200 μl of PBS (1.8 mM KH2PO4, 10 mM Na2HPO4, 137 mM NaCl, 2.7 mM KCl, pH 7.4) and supplemented with lysozyme (200 μg ml−1; Amresco 0663-5G)41. An ELISA was used to identify the target binding DARPins. Briefly, Nunc MaxiSorp plates (Thermo Fisher Scientific, 50-712-278) were coated with 4 μg ml−1 neutravidin (Thermo Fisher Scientific, 31000) in PBS at room temperature for 2 h. The wells were washed with PBS and then blocked with PBS supplemented with 0.5% BSA (Thermo Fisher Scientific, BP9706100; PBS-B) at 4 °C overnight. The next day, after washing with PBS-T (PBS with 0.1% Tween 20), the wells were incubated with biotinylated RBD or full-length spike protein (20 nM in PBS) at room temperature for 1 h. The wells were washed again with PBS-T before the addition of 100 μl of cell lysate (fivefold diluted in PBS) and incubated at room temperature for 2 h. The amount of plate-bound DARPin in each well was quantified using mouse anti-myc antibody (Invitrogen, 13-2500, 1:2500 diluted in PBS-B) and HRP-conjugated goat antimouse antibody (Jackson ImmunoResearch, 115-035-146; 1:1000 diluted in PBS-B) as the primary and secondary antibodies, respectively, and BioFx TMB (VWR, 100359-154) for color development. In total, 20 and 23 clones showed significant ability to bind to RBD and full-length spike protein, respectively. Sequencing revealed 11 unique clones with significant ability to bind both RBD and full-length spike protein.

To identify DARPin clones able to block spike protein and hACE2 interaction, a competitive ELISA was used42. Briefly, the Maxisorp plates were first coated with full-length spike protein (BEI 52308, 2 nM in PBS) at room temperature for 2 h. The wells were blocked with PBS-B at 4 °C overnight, washed with PBS-T, and then incubated with mixtures of hACE2-HRP (prepared in-house, 0.5 nM) and different IMAC-purified DARPins (50 mM)41,43 at room temperature for 1 h. The amount of hACE2-HRP in each well was quantified using BioFx TMB.

For preparation of hACE2-HRP, hACE2 (Raybiotech, 230-30165; 1.76 mg ml−1) was first biotinylated as described above and then incubated with an equal molar amount of streptavidin-HRP (JIR, 016-030-084) in PBS at room temperature for 20 min and stored at −20 °C in 50% glycerol until use.

Plasmids

Plasmids encoding the wild type Δ19 spike protein were obtained from Addgene (145780). The plasmids for the Δ19 spike protein of B.1.617.2 and C.37 were generously provided by Nathaniel Landau (New York University)44. DNA fragment encoding the Δ19 spike protein of strain B. 1.351 and the RBD (residues 339-501) of B.1.1.529 was synthesized by Gene Universal and inserted into the pCG1 plasmid45. Paul Bieniasz (The Rockefeller University) provided the 293T cell clone 22 (293T.c22) with high expression efficiency of human ACE2 and the lentiviral reporter plasmid pHIV-1NL4-3-ΔEnv-NanoLuc46. The plasmid encoding the chimeric B.1.1.529 spike protein (B.1.1.529*) was constructed by replacing the RBD region (residues 338–514) in B.1.351 with that from B.1.1.529. Briefly, the gBlock fragment of B.1.1.529 RBD was digested with BsaI and BlpI, and ligated to (1) pCG1-B.1.351 backbone digested with BamHI and BlpI and (2) PCR product amplified from the same backbone with primers Spike_F and Spike_R and digested with BamHI and BsaI in a three-fragment-ligation reaction.

gBlock of B.1.1.529 RBD:

Aaaaaaggtctcacttcgatgaggtgttcaatgccaccagattcgcctctgtgtacgcctggaaccggaagcggatcagcaattgcgtggccgactactccgtgctgtacaacctggcccctttcttcaccttcaagtgctacggcgtgtcccctaccaagctgaacgacctgtgcttcacaaacgtgtacgccgacagcttcgtgatccggggagatgaagtgcggcagattgcccctggacagacaggcaacatcgccgactacaactacaagctgcccgacgacttcaccggctgtgtgattgcctggaacagcaacaagctggactccaaagtctctggcaactacaattacctgtaccggctgttccggaagtccaatctgaagcccttcgagcgggacatctccaccgagatctatcaggccggcaataagccttgtaacggcgtggccggcttcaactgctacttcccactgagatcctactcctttagacccacatatggcgtgggccaccagccctacagagtggtggtgctgagcttcgaa

Spike_F: cgaattcggatccgccacca (Italic letters denote BamHI recognition sequence)

Spike_R: ttttttggtctctgaaggggcacagattggtga (Italic letters denote BsaI recognition sequence)

To construct trimeric DARPin molecules, the DNA fragment encoding a codon-optimized T4 foldon (Protein Data Bank (PDB): 1RFO) was synthesized by Gene Universal (Newark, DE). T4 foldon was fused to the N-terminus of a DARPin molecule via a flexible (GGGGSLQ)x2 linker and cloned into the pET28a expression vector. DARPin SR16, SR22, FSR16 and FSR22 contain a 6xHis tag and a Myc tag at the N terminus, while SR16m and FSR16m contains only a 6xHis tag at the C terminus (Supplementary Fig. 3).

SARS-CoV-2 spike lentiviral pseudoviruses

Lentiviral pseudoviruses with different SARS-CoV-2 spike proteins (CoV2pp) were produced as previously reported46. Briefly, plasmids encoding the Δ19 spike protein and reporter pHIV-1NL4-3-ΔEnv-NanoLuc46 (1:3 molar ratio, 10 μg total) were mixed with 500 μl of serum-free DMEM medium and 44 μl of PEI (1 mg ml−1; Polysciences transporter 5, 26008-5) and used to transfect 5 × 106 293T cells seeded the night before. Twenty-four-hour post-transfection, the medium was replaced with fresh DMEM supplemented with 10% FBS and 48-h post-transfection the viral supernatant was collected, aliquoted and stored at −80 °C until use.

To determine the neutralization efficiency, serially diluted DARPin molecules were incubated with CoV2pp (final 500-fold diluted) at 37 °C for 30 min before being added to 293T.c22 cells seeded the night before at 104 cells per well in 96-well plates. The plates were incubated at 37 °C/5% CO2 for 48 h, and the NanoLuc signal from each well was quantified using the Nano-Glo Luciferase Assay kit (Promega, N1120) and a Cytation 5 plate reader equipped with Gen5 3.05 software.

DARPin protein production and characterization

All DARPin molecules were expressed in E. coli Bl21 (DE3) cells in LB medium supplemented with 50 μg ml−1 of kanamycin. Protein expression was induced with IPTG (0.5 mM) when the culture reached OD600 = 0.5. The protein expression was continued at 37 °C for 5 h, and the cells were collected by centrifugation. The proteins were purified via immobilized metal affinity chromatography (IMAC) using gravity Ni-NTA agarose columns. Protein purity was determined using 12% SDS–PAGE gels.

For in vivo studies, the IMAC-purified FSR16m was sterilized by filtration through a 0.22 μm filter, concentrated and buffer exchanged into PBS via ultrafiltration (Amicon column MWCO 10 KDa, UFC801024) before endotoxin removal using High-Capacity Endotoxin Removal Spin Columns (Pierce, 88274). The endotoxin level in the protein sample (1.5 mg ml−1) was quantified to be <30 U ml−1 using the Pierce Chromogenic Endotoxin Quant Kit (Thermo Fisher Scientific, A39552).

For size exclusion chromatography studies, DARPin samples (0.9 mg ml−1 × 0.25 ml) were loaded onto an Enrich SEC 70 × 300 Column equilibrated with PBS (GE AKTApure, with Unicorn 6.3 software). Vitamin B12 (Sigma, V2876-100MG) was used at a final concentration of 1 mg ml−1 as an internal control.

Spike RBD and DARPin avidity measurement

The human IgG1 Fc-tagged RBD proteins were made in-house as previously described33,47. The avidity measurement was performed on the ForteBio Octet RED 96 system (Sartorius). Briefly, the RBD proteins (20 µg ml−1) were captured onto protein A biosensors for 300 s. The loaded biosensors were then dipped into the kinetics buffer for 10 s for adjustment of baselines. Subsequently, the biosensors were dipped into serially diluted (from 0.13 to 300 nM) DARPin proteins for 200 s to record association kinetics and then dipped into kinetics buffer for 400 s to record dissociation kinetics. Kinetic buffer without DARPin was used to correct the background. Octet Data Acquisition 9.0 software was used to collect affinity data. For fitting of KD values, Octet Data Analysis software v11.1 was used to fit the curve by a 1:1 binding model using the global fitting method.

ELISA binding assay

ELISA plates were coated with recombinant DARPin protein (2 μg ml−1) at 4 °C overnight and blocked with 5% skim milk at 37 °C for 2 h. One hundred microliters of serially diluted human IgG1 Fc-tagged RBD proteins in 1% skim milk was added to each well, and the plates were incubated at room temperature for 3 h before the addition of 100 μl per well of HRP-conjugated goat anti-human IgG antibody (Jackson ImmunoResearch, 109-035-088; diluted 1:5,000), and the plates were incubated at room temperature for another hour. The plates were washed three to five times with PBS-T (0.05%, Tween 20) between incubation steps. TMB (3,3′,5,5′-tetramethylbenzidine) substrate was added at 100 μl per well for color development. The reaction was stopped by adding 50 μl per well 2 M H2SO4. The OD450 nm was read by a SpectraMax microplate reader using Softmax Pro 6.5.1 and analyzed with GraphPad Prism 8.

Cells and authentic viruses

Vero-TMPRSS2 (ref. 48) and Vero-hACE2-TMPRSS2 (ref. 49) (a gift of A. Creanga and B. Graham, NIH) were cultured at 37 °C in Dulbecco’s Modified Eagle medium (DMEM) supplemented with 10% FBS, 10 mM HEPES pH 7.3, 1 mM sodium pyruvate, 1× nonessential amino acids, 100 U ml−1 of penicillin–streptomycin and 5 μg ml−1 of puromycin. The B.1.617.2, B.1.617.2.AY1, B.1.351, BA.1, BA.1.1 and BA.2 SARS-CoV-2 strains were obtained from infected individuals and have been described previously15,50. Infectious stocks were propagated by inoculating Vero-hACE2-TMPRSS2 cells. Supernatant was collected, aliquoted and stored at −80 °C. All work with infectious SARS-CoV-2 was performed in Institutional Biosafety Committee-approved BSL3 and A-BSL3 facilities at Washington University School of Medicine using positive pressure air respirators and protective equipment. All virus stocks were deep-sequenced after RNA extraction to confirm the presence of the anticipated substitutions.

Mouse experiments

Animal studies were carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Institutional Animal Care and Use Committee at the Washington University School of Medicine (assurance number A3381-01). Virus inoculations were performed under anesthesia that was induced and maintained with ketamine hydrochloride and xylazine, and all efforts were made to minimize animal suffering.

Heterozygous K18-hACE C57BL/6J mice (strain: 2B6.Cg-Tg(K18-ACE2)2Prlmn/J) were obtained from The Jackson Laboratory. Animals were housed in groups and fed standard chow diets. Eight-week-old female mice were administered 103 FFU of SARS-CoV-2 via intranasal administration.

Measurement of viral burden

Tissues were weighed and homogenized with zirconia beads in a MagNA Lyser instrument (Roche Life Science) in 1,000 μl of DMEM media supplemented with 2% heat-inactivated FBS. Tissue homogenates were clarified by centrifugation at 10,000g for 5 min and stored at −80 °C. RNA was extracted using the MagMax mirVana Total RNA isolation kit (Thermo Fisher Scientific) on a Kingfisher Flex extraction robot (Thermo Fisher Scientific). RNA was reverse transcribed and amplified using the TaqMan RNA-to-CT 1-Step Kit (Thermo Fisher Scientific). Reverse transcription was carried out at 48 °C for 15 min followed by 2 min at 95 °C. Amplification was accomplished over 50 cycles as follows: 95 °C for 15 s and 60 °C for 1 min. Copies of SARS-CoV-2 N gene RNA in samples were determined using a previously published assay11,51. Briefly, a TaqMan assay was designed to target a highly conserved region of the N gene (forward primer, 5′-ATGCTGCAATCGTGCTACAA-3′; reverse primer, 5′-GACTGCCGCCTCTGCTC-3′; probe, 5′-/56-FAM/TCAAGGAAC/ZEN/AACATTGCCAA/3IABkFQ/-3′). This region was included in an RNA standard to allow for copy number determination down to 10 copies per reaction. The reaction mixture contained final concentrations of primers and probe of 500 and 100 nM, respectively.

Cytokine and chemokine protein measurements

Lung homogenates were incubated with Triton-X-100 (1% final concentration) for 1 h at room temperature to inactivate SARS-CoV-2. Homogenates were analyzed for cytokines and chemokines by Eve Technologies Corporation using their Mouse Cytokine Array/Chemokine Array 31-Plex (MD31) platform.

Authentic virus neutralization assay

Serial dilutions of DARPins were incubated with 102 FFU of the indicated SARS-CoV-2 strains for 1 h at 37 °C. DARPin-virus complexes were added to Vero-hACE2-TMPRSS2 cell monolayers in 96-well plates and incubated at 37 °C for 1 h. Subsequently, cells were overlaid with 1% (wt/vol) methylcellulose in MEM supplemented with 2% FBS. Plates were collected 24 h later by removing overlays and fixed with 4% PFA in PBS for 20 min at room temperature. Plates were washed and sequentially incubated with an oligoclonal pool of SARS2-2, SARS2-11, SARS2-16, SARS2-31, SARS2-38, SARS2-57 and SARS2-71 antispike protein antibodies52 and HRP-conjugated goat antimouse IgG in PBS supplemented with 0.1% saponin and 0.1% bovine serum albumin. SARS-CoV-2-infected cell foci were visualized using TrueBlue peroxidase substrate (KPL) and quantitated on an ImmunoSpot microanalyzer (Cellular Technologies). Data were processed using Prism software (GraphPad Prism 8.0).

Cryo-EM sample and grid preparation

SARS-CoV-2 S6P spike protein was produced in 293F cells and purified from cell culture supernatant using a Ni-NTA agarose column followed by Superdex S200 16/600 (GE Healthcare) size-exclusion column chromatography as described53.

The subdomain SARS-CoV-2 RBD protein used in Cryo-EM was produced as previously described54 with minor modification. Briefly, the DNA sequence encoding RBD residues 330–526 was cloned in a mammalian expression vector with an HRV3C cleavable single-chain Fc tag before the coding sequence. The tagged RBD protein was expressed by transient transfection in FreeStyle 293 for 6 d. The desired protein in culture supernatant was captured by protein A resin and liberated by HRV3C cleavage. Finally, the subdomain RBD protein was purified on a Superdex 200 16/600 gel filtration column equilibrated with PBS.

The S309 IgG32 was expressed by transient transfection in Expi293F cells for 5 d at 37 °C and purified with Protein A Sepharose Fast Flow resin (Cytiva). The DNA encoding CR3022 IgG55 was cloned into a pVRC8400 vector with the addition of the HRV3C protease-cleavage site in the hinge region of IgG1 and expressed by transient transfection in Expi293F cells for 5 d at 37 °C and purified with protein A resin. The S309 Fab was cleaved from the S309 IgG using endoproteinase LysC (New England Biolab), and the CR3022 Fab was cleaved from CR3022 IgG with altered hinge with HRV3C protease. Protein A resin was then added to each mixture to remove Fc, and Fab was collected in the flowthrough and further purified on a Superose 6 10/300 SEC column (Cytiva) with X1 PBS (Gibco).

To generate SARS-CoV-2 S6P and FSR16m (or FSR22) complexes, SARS-CoV-2 S6P and FSR16m (or FSR22) were incubated in 1:3 molar ratio and the complexes were purified by Superdex S200 GL 10/300 (GE Healthcare) using 10 mM HEPES, 7.4, 150 mM NaCl as running buffer and were confirmed by SDS–PAGE and negative stain EM. To make RBD:SR22:S309Fab:CR3022Fab and RBD:SR16m:S309Fab:CR3022Fab ternary complexes, we incubated RBD (330–526) with molar excess of either SR22, C309 Fab and CR3022 Fab, or SR16m, S309 Fab and CR3022 Fab. 2.7 µl of the complexes at 1 mg ml−1 concentration in 10 mM HEPES, 7.4, 150 mM NaCl were deposited on Quantifoil R2/2 grids (quantifoil.com). Grids were vitrified using an FEI Vitrobot Mark IV (Thermo Fisher Scientific) with a wait time of 30 s, blot times of 1.5–4.5 s and blot force of 1.

Cryo-EM data collection and processing

The FSR22/SARS-CoV-2 S 6P and FSR16m/SARS-CoV-2 S 6P complexes grids were imaged using a Titan Krios electron microscope equipped with a Gatan K3 Summit direct detection device. Movies were collected at ×105,000 magnification over a defocus range of −1.0 to −2.25 µm for 2.3 s with the total dose of 40.02 e–/Å2 fractionated over 40 raw frames. Single particle cryo-EM data were collected with Latitude S. All data processing was done with cryoSPARCv3.3.1 (ref. 56). Motion correction and CTF estimation in patch mode, blob particle picking and particle extraction with the box size of 500 Å for the FSR22 (or FSR16m) : SARS-CoV-2 spike complexes and 230 Å for the RBD:SR22 (or SR16m):S309 Fab: CR3022 Fab complexes were performed followed by 2D classifications, ab initio 3D reconstruction, and multiple rounds of 3D heterogeneous refinement. C3 symmetry was applied for the final reconstruction of the FSR22/FSR16m: SARS-CoV-2 spike complex after the initial 3D heterogeneous refinement using C1 symmetry identified a trimer with three RBD-up conformation bound three SR22 (or SR16m) molecules. To define RBD–FSR22 interface (Supplementary Figs. 12 and 13), local refinement was performed using a soft mask covering one SR22 and one RBD molecule. The parameters for data collection and processing were summarized in Supplementary Table 2.

Model building and refinement

Coordinates from PDB 7BNO and the initial models of SR22/SR16m generated using AlphaFold were used for the initial fit to the reconstructed maps. To build S309 Fab and CR3022 Fab, PDB 7TN0 and 6YLA were used, respectively. Then, the models were manually built using Coot.v0.9.8.1 (ref. 57) followed by simulated annealing and real space refinement in Phenix v1.20.1-4487 (ref. 58) iteratively. Geometry and map fitting were evaluated throughout the process using Molprobity59 and EMRinger v1.0.0 (ref. 60). Figures were generated using PyMOL v2.4.2 (www.pymol.org) and UCSF ChimeraX.v1.3.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.