The human ACE2 extracellular domain (aa 18-740; Gene ID: 59272) and the variable domain sequences of the REGN 10933 antibody (14) were codon-optimized for expression by Chinese hamster cells using the GeneArt GeneOptimizer tool (Thermo Fisher Scientific, Regensburg, Germany). VH, VL, and ACE2 sequences (ACE2 wild-type (ACE2) or the indicated mutants, ACE2-RR, ACE2-K, ACE2-M) were synthesized de novo at GeneArt (Thermo Fisher Scientific, Regensburg, Germany). ACE-2 coding sequences were fused at their C-terminus to a human Igγ1 Hinge- Fc domain via a flexible (GGGGS)³ linker. Modifications in the CH2 domain consisting of the amino acid substitutions and deletions E233P; L234V; L235A; ΔG236; D265G; A327Q; A330S (EU index), which abrogate FcR binding and complement fixation, were introduced as described (23). REGN 10933 variable sequences were inserted into a human Igγ1 backbone comprising CH1‐CH2‐CH3‐ or CΚ‐constant domain sequences as described (23). All constructs were transiently transfected and produced using the ExpiCHO™ Expression System (Thermo Fisher Scientific, Regensburg, Germany) according to the manufacturer’s instructions and were then purified by HiTrap™ MabSelect™ SuRe columns (Cytiva, Freiburg, Germany), before being subjected to preparative and analytical size exclusion chromatography (SEC) using HiLoad™ 16/600 Superdex 200 pg and Superdex™ 200 Increase 10/300 GL columns (Cytiva Freiburg, Germany), respectively. Endotoxin levels of samples, as determined by a limulus amebocyte lysate assay (Endosafe^®Charles River, Charleston, SC), were < 0.5 EU/ml. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was performed as previously described (23).

SARS-CoV-2 full-length trimeric spike proteins corresponding to the B.1 (B.1.126), Alpha (B.1.1.7), Beta (B1.351), Gamma (P.1), and Delta (B.1.617.2) were purchased from BioServ, (Sheffield, UK). The Trimeric S-protein corresponding to Omicron (B.1.1.529) or Omicron subvariant BA.5 and BQ1.1 were obtained from Sino Biological (Beijing, China).

Enzymatic activity of ACE2 fusion proteins was measured using the ACE2 Activity Assay Kit (Fluorometric) (BioVision, Milpitas, CA) according to the manufacturer’s instructions. The proteins were diluted in assay buffer to 22.7, 4.54, and 0.91 nM final concentration. Fluorescence was measured using a Wallac 1420 Victor 2 Multi-Label Counter (Perkin-Elmer, Waltham, MA).

The indicated S trimeric proteins were coated on 96-well plates at 1µg/ml, 4°C overnight. After washing, wells were blocked with PBS containing 3% BSA for 1 hour at room temperature. Next, a serial dilution of the indicated ACE2 fusion proteins, REGN 10933 or serum antibodies were pre-mixed with 150 nM of His-tagged ACE2 wild-type protein (BioLegend, San Diego, CA) and added to the plates. In case of the Omicron variants (Sino Biological), the His-tagged ACE2 protein was additionally biotinylated using the One-Step Biotinylation Kit (Miltenyi, Cologne, Germany) according to the manufacturer’s instructions. For visualization, a Penta-His HRP conjugate (1:1000) (Qiagen, Hilden, Germany) or mouse anti-Biotin HRP conjugate (1:1000) (Invitrogen, Waltham, MA) were used. Unbound HRP-conjugated antibodies were removed by washing, TMB substrate was added, and absorbance was measured at 450nm.

Sera were collected from 8 healthy donors (25-65 years of age). All donors received a first dose of Vaxzevria vaccine and then a second dose of Comirnaty or Spikevax. Sera were collected 40-45 days after the second dose and anti-SARS-CoV-2 antibody concentrations were measured using the Euroimmun Anti-SARS CoV-2 ELISA IgG kit (Euroimmun, Luebeck, Germany). Briefly, serum samples were diluted at 1:100 and 1:1000 and ELISA was performed following manufacturer’s instructions. A titration of a reference neutralizing antibody (REGN 10933) was used to calibrate the assay.

Trimeric S-proteins were analyzed for their binding to ACE2-RR or ACE2-M fusion proteins using an Octet HTX system (Sartorius, Goettingen, Germany). Assays were run with a sensor offset of 3 mm and an acquisition rate of 5 Hz on AHC biosensors in 16-channel mode. Microplates were loaded with 60 μL per well of assay buffer consisting of PBS with 0.05% Tween-20 and 0.1% BSA. Sensors were equilibrated in assay buffer for 10 min. Following a baseline step of 60s, the analyte S-proteins were loaded for 120 s. Association was measured for 150s and dissociation for 300s. Regeneration of the sensors was performed using 10 mM Glycine pH 1.5. Data evaluation was done using Octet Analysis HT Software. The reference subtraction was performed to consider the potential dissociation of analyte loaded onto the biosensor. Data traces were aligned to the baseline, followed by an inter-step correction for the dissociation step. Savitzky-Golay filtering was applied to the data and the curves were fitted globally using a 1:1 binding model (with Rmax unlinked by sensor).

All experiments with SARS-CoV-2 viruses were conducted in a Biosafety Level 3 laboratory at the University Hospital Tübingen. The SARS-CoV-2 strain icSARS-CoV-2-mNG (24) was obtained from the World Center for Emerging Viruses and Arboviruses (WRCEVA) of the UTMB (University of Texas Medical Branch, Galveston, TX, USA). SARS-CoV-2 B.1.126 (parental D614G), referred as B.1, and SARS-CoV-2 B.1.351 (Beta), were isolated from patient samples and variant identity was confirmed by next-generation sequencing of the entire viral genome as described before (25, 26). SARS-CoV-2 B.1.1.529 (Omicron) was isolated from a throat swab collected in December 2021 at the Institute for Medical Virology and Epidemiology of Viral Diseases, University Hospital Tübingen, from a PCR-positive patient. Fifty microliters of patient material were diluted in medium and used directly to inoculate 150,000 Caco-2 cells in a six-well plate. 48 hours post-infection (hpi), the supernatant was collected, centrifuged, and stored at -80°C. After two consecutive passages, an RNA sample from the supernatant was prepared, and NGS confirmed that the clinical isolate belongs to the lineage B.1.1.529. All virus stocks were generated in Caco-2 cells collecting supernatants 48-72 hpi. Multiplicity of infection determination (MOI) was conducted by titration using serial dilutions of both virus stocks. The number of infectious virus particles per ml was calculated as (MOI × cell number)/(infection volume), where MOI = -ln (1-infection rate).

Caco-2 (Human Colorectal adenocarcinoma, ATCC HTB-37) cells were cultured at 37°C with 5% CO2 in DMEM containing 10% FCS, 2 mM L-glutamine, 100 mg/ml penicillin-streptomycin and 1% NEAA. Neutralization assays using clinical isolates were performed as described in Wagner et al., 2021. Briefly, cells were co-incubated with the clinical isolate SARS-CoV-2 B.1.126, SARS-CoV-2 B.1.351 (Beta), or SARS-CoV-2 B.1.1.529 (Omicron), at MOI of 0.7-4.0, and serial dilutions of the ACE2 protein designs. 48 hpi, cells were fixed with 80% acetone, and immunofluorescence (IF) staining was performed using an anti-SARS-CoV-2 nucleocapsid antibody (GeneTex, Cat No. GTX135357) and goat anti-rabbit Alexa594-conjugated secondary antibody. Cells were counterstained with DAPI solution and images were taken with the Cytation3 (BioTek). Infection rates were calculated as the ratio of Alexa594-positive over DAPI-positive cells, which were automatically counted by the Gen5 software (BioTek). Inhibitory concentration 50 (IC50) was calculated as the half-maximal inhibitory dose using four-parameter nonlinear regression (GraphPad Prism).

For binding experiments, 3×10⁶ Caco-2 cells were seeded in a T75 flask the day before infection, in a medium containing 5% FCS. Cells were infected with SARS-CoV-2-mNG, and 48 hpi cells were detached from the flask using Accutase, fixed with 2% PFA for 10 minutes at 37°C, and resuspended in FACS buffer (PBS, 1%FCS). 1×10⁶ cells in 100 µL in FACS buffer were distributed in a U-shape 96-well plate. The plate was centrifuged at 600 g for 5 min and the buffer was removed by a fast decant. Cells were incubated for 1h at 4°C using 50 µl of 3-fold serial dilutions of ACE2 protein or REGN 10933, tested from 40 µg/ml following 12 dilution points. Cells were washed with 150 µl of FACS buffer/well, centrifuged, and the supernatant decanted. The washing step was repeated using 200 µl of FACS buffer/well. Subsequently, cells were incubated with 50 µl of a 1:200 dilution of the Secondary AB- R-Phycoerythrin (PE) conjugated affinity pure F(ab’)2 Fragment Goat-anti Human IgG-Fc gamma fragment (Jackson-Immuno) for 30 minutes at 4°C. The two washing steps were repeated, and the cells were resuspended using 100 µl of FACS buffer/well. Controls included: mock-infected cells incubated with the highest and lower protein concentrations; infected cells non-incubated as well as infected cells stained only with the secondary antibody. Alternatively, Caco-2 cells were infected with SARS-CoV-2 parental or Omicron variants and the same protocol described above was followed. After incubation with PE-secondary antibody, cells were permeabilized with 80% Acetone for 5 minutes at room temperature, washed as described before and immunofluorescence (IF) staining was performed using an anti-SARS-CoV-2 nucleocapsid antibody (GeneTex, Cat No. GTX135357) (1:1000, 1 h) and goat anti-rabbit Alexa594-conjugated secondary antibody (1:2000, 1 h). After the final washing steps, the cells were analyzed using a MACSQuant VYB (Miltenyi). FACS analysis was performed with MACS Quantify Software (Miltenyi) and Flowlogic (Miltenyi–Inivai).

ACE2 fusion proteins were constructed by fusing the ACE2 extracellular domain to a human IgG1 Fc fragment that was originally developed to prevent binding to FcRs by T cell activating bispecific antibodies. This modification consists of 6 mutations and one deletion in the CH2 domain of the human IgG1 Fc domain that completely abrogates binding to FcRs and ablates complement activation (23).

To avoid the introduction of undesired peptidase activity, two catalytically-inactive forms of ACE2 were generated. The first mutant, ACE2-RR, contains two substitutions in the catalytic pocket, H374R, and H378R, that prevent zinc binding within the active site of the protein (27). The second mutant, ACE2-K, has a single mutation at position 273 (R273K), which is critical for enzymatic activity (28). Fc-fusion proteins constructed with either of the two variants show the size expected for a dimeric molecule as demonstrated by SDS-PAGE and size exclusion chromatography ( Supplementary Figures 1A–C ). In contrast to ACE2-RR, ACE2-K retained some residual enzymatic activity and showed a slight loss in binding affinity to recombinant trimeric parental S-protein as measured by enzyme-linked immunosorbent assay (ELISA) ( Supplementary Figures 1D, E ). Thus, the ACE2-RR version of the fusion protein was selected for further modification.

To enhance binding to the S-protein, we introduced three additional mutations to the ACE2-RR-Fc fusion protein that are located at the S-protein binding surface of ACE2: T27Y, L79Y, and N330Y ( Figures 1A, B ). These mutations are similar to an ACE2 variant described by Chan and collaborators (29). ACE2-M, was produced as a stable dimer and had no detectable enzymatic activity ( Figures 1C–E ).

We next measured the binding affinity of recombinant trimeric S-protein and ACE2-M by ELISA, which was benchmarked against an Fc-fusion protein comprising the wild type ACE2 protein, as well as against the reference antibody REGN10933. In this assay, ACE2-M showed a 4-fold higher binding affinity than wild-type ACE2 fusion protein ( Figure 1F ). Interestingly, the apparent binding affinity measured by flow cytometry of Caco-2 cells infected with a recombinant parental virus resulted in a greater than 18-fold affinity enhancement ( Figure 1G ). In both assays, the S-protein apparent binding affinity of ACE2-M and REGN 10933 was similar. The differences between the affinities of antibodies and fusion proteins to S-proteins coated to an ELISA plate compared with those embedded in the membrane of a virus-infected cell are noteworthy. Obviously, binding to infected cells resembles the “physiological state” more closely, and may predict different neutralization capacities of the various fusion proteins more reliably as demonstrated below.

To test whether ACE2-M retains binding to the globally spreading SARS-CoV-2 VOCs Alpha (B.1.1.7), Beta (B1.351), Gamma (P.1), Delta (B.1.617.2), and Omicron (B.1.1.529), structural analysis of ACE2-M and the respective S-RBD variants was performed ( Supplementary Figure 2 ). This analysis revealed that the affinity-enhancing mutations introduced into the ACE2-M protein appear to be spared by the viral escape mechanism, probably because they form part of the RBD-binding interface. Those observations were further confirmed by measuring the binding kinetics of ACE2-M to all variants by biolayer interferometry technology (BLI). The results depicted in Supplementary Figure 3 demonstrate a superior apparent binding affinity of ACE2-M to spike variants. Again, the differences in affinities were less pronounced compared to those observed by flow cytometry of infected cells ( Figure 1G ).

Next, we tested the ability of ACE2-RR, ACE2-M, and REGN 10933 to inhibit binding of ACE2 to various S-proteins by competition ELISA. To this end, the fusion proteins and REGN 10933 were mixed with a saturating amount of His-tagged-ACE2 protein before binding to immobilized S-proteins was determined. In line with the BLI results, we observed an improved inhibition of binding by ACE2-M to all S-proteins ( Figure 2A ). In contrast, REGN 10933 failed to inhibit ACE2 binding to Beta, Gamma, and Omicron S-proteins. These results are consistent with previously published data, suggesting that the reduced binding of REGN 10933 to certain S variants is likely due to mutations in amino acids K417 and E484 of the S-protein (30, 31), found in the Beta, Gamma, and Omicron variants.

Likewise, binding of ACE2 fusion proteins and REGN 10933 antibody to Caco-2 cells infected with authentic SARS-CoV-2 natural isolates corresponding to the parental strain (D614G) and Omicron (B.1.1.529) showed similar results ( Supplementary Figure 4 ). REGN 10933 bound with high affinity to Caco-2 cells infected with the parental virus, in line with the results obtained with the recombinant virus in Figure 1G . However, REGN 10933 failed to bind to Caco-2 cells infected with the Omicron virus. In contrast, ACE2-M retains high-affinity binding to Caco-2 cells infected with either isolate. Unexpectedly, we observed a 4-fold lower MFI for binding to cells infected with Omicron vs. parental virus. This observation could be explained by the reduced replication capacity recently described for Omicron variant (32–34). The reduced replication seems to be due to the inefficient use of the cellular protease TMPRSS2, which promotes cell entry through plasma membrane fusion (33). We speculate that reduced replication in Caco-2 cells may result in a decreased expression of S-proteins on the cell surface.

To investigate the efficacy of ACE2-M in preventing viral infection, we performed a neutralization assay in Caco-2 cells using SARS-CoV-2 natural isolates from the parental B.1 as well as Beta, and Omicron strains. ACE2-M neutralized all SARS-CoV-2 infected cells at picomolar concentrations ( Figure 2B and Supplementary Figure 5 ). ACE2-M activity was greater against Omicron than against the Beta variant or the parental B.1. In contrast, the activity of REGN 10933 against the Beta and Omicron variants was reduced and undetectable, respectively. Altogether, our results demonstrate that binding of ACE2-M to S-variants (vs. parental) is preserved or enhanced (in case of Omicron), while it is weakened or lost in the case of REGN 10933. Thus, these results confirm our “founding hypothesis” for the construction of ACE2 fusion proteins, namely that viral variants will mutate “away” from recognition by antibodies but “towards” recognition by ACE2 fusion proteins and hence to neutralization by such proteins.

To evaluate the binding efficiency of antibodies generated by active immunization to parental, Beta, Delta, and Omicron trimeric S-proteins, sera from 8 healthy fully vaccinated donors were used and evaluated in a competitive ELISA similar to that described above in Figure 2A . Similar to REGN 10933, antibodies present in the post-vaccination sera showed a reduction in binding to Beta, Delta, and, to a greater extent, to Omicron S-proteins ( Figure 3A ). These results suggest that antibodies generated after active immunization against the parental S-protein are less effective against new variants in accordance with recent reports (35–40).

Next, we defined the required amount of neutralizing antibodies and ACE2-M protein, to achieve a complete inhibition of binding to S-proteins. To this end, SARS-CoV-2 antibody concentrations present in the sera of vaccinated donors were quantified, adjusted, and compared in a competitive ELISA to the ACE2-M protein. Our results, depicted in Figure 3B , show that ACE2-M, at a concentration of 1 µg/ml, achieved almost complete binding inhibition of ACE2 wild type to all S-variants, however, only a partial binding inhibition could be obtained with antibodies in the various sera. Of note, a serum concentration of 1 µg/ml is easily reached during treatment of antiviral disease with suitable monoclonal antibodies (41, 42).

We have demonstrated that ACE2 fusion proteins are not subject to immune escape exerted by variants of the S-protein in line with recent publications (43–46). In fact, variants with increased affinity such as Omicron variants (BA.1, BA.5 and BQ1.1), are neutralized more effectively than the parental B.1. Recently, alternative receptors have been reported to interact with the SARS-CoV-2 virus (47–51). In cells with low ACE2 expression, it is thought that SARS-CoV-2 can enter the cells via several alternative receptors, but the entry mechanism remains to be defined. Although there is no direct evidence that SARS-CoV-2 escapes ACE2 treatment, the fact that the highly transmissible variant Omicron BA.1 has evolved to be less dependent on TMPRSS2, raises the possibility that alternative SARS-CoV-2 mutations may also contribute to viral evolution and may cause ACE2 immune escape.

To our knowledge, ACE2-M is the first engineered ACE2 fusion molecule combining modifications for Fc-attenuation, enzymatic depletion, and ACE2 affinity enhancement. Although animal studies are required to conclusively evaluate the importance of enzymatic depletion and Fc-attenuation, this molecule provides an important additional option for treatment of COVID-19 and other coronaviruses that use the ACE2 protein as entry receptor.