An IgG-like fusion protein comprising an anti-spike S2 antibody and ACE2 exhibits potent and broad neutralization against SARS-CoV-2 and variants of concern

Abstract

Background

The continuous evolution of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) highlights the value of broad-spectrum antiviral strategies. Antibody-engineering approaches targeting conserved regions of the Spike protein may enhance neutralizing potency and breadth.

Methods

A human antibody (P23) against the Spike protein of SARS-CoV-2 was identified using a human antibody phage display library panning and screening for binding affinity and breadth against multiple coronavirus Spike proteins using surface plasmon resonance (SPR). The epitope of P23 was characterized using the S1 and S2 subunits of SARS-CoV-2 Spike protein and hydrogen–deuterium exchange mass spectrometry (HDX-MS). An IgG–like bispecific fusion protein (Bs-ACE2-P23) was engineered by fusing the extracellular domain (ECD) of human angiotensin-converting enzyme 2 (ACE2) to the N-terminus of the P23 heavy chain (HC). Neutralizing activity was evaluated against both pseudotyped and authentic SARS-CoV-2 variants.

Results

P23 cross-bound Spike proteins from SARS-CoV-2 wild type (WT), D614G and JN.1 variants, Pangolin-CoV, Bat coronavirus RaTG13, and SARS-CoV-1, recognizing an epitope on the S2 subunit adjacent to the fusion peptide (FP). While P23 was ineffective against D614G-containing SARS-CoV-2 variants, Bs-ACE2-P23 exhibited markedly enhanced neutralization potency. This bispecific architecture also improved the neutralizing activity of another FP-targeting antibody.

Conclusion

We developed a bispecific fusion protein with potent broad-spectrum neutralizing activity against SARS-CoV-2 variants. This architecture provides a promising strategy for next-generation coronavirus biologics.

Statement of Significance We obtained a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) neutralizing antibody, P23, binding adjacent to the FP in the S2 subunit. We engineered an IgG-like bispecific protein, Bs-ACE2-P23, combining P23 with the viral receptor ACE2, and demonstrated its broad and potent neutralization against SARS-CoV-2 variants, offering a promising strategy for broad-spectrum antiviral therapeutics.

Introduction

Coronaviruses remain a persistent threat to global health. Although severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has transitioned toward an endemic phase, continued evolution of SARS-CoV-2 and the potential emergence of related coronaviruses underscore the need for broadly applicable antiviral strategies [1]. Engineering approaches that enhance the breadth and neutralizing potency of SARS-CoV-2 antibodies may provide a generalizable strategy for developing next-generation antibody therapeutics against current and future coronavirus infections [2]. The Spike protein of coronaviruses is essential for binding to host cell receptors and facilitating membrane fusion for viral entry [3], and is the main target for the development of neutralizing antibodies (nAbs) [2]. The Spike protein of coronaviruses typically comprises two subunits within the ectodomain: S1, responsible for receptor binding, and S2, which facilitates membrane fusion. Specifically, using the wild type (WT) SARS-CoV-2 (Wuhan-Hu-1 strain) Spike protein as an example, its extracellular domain (ECD) is comprised of the S1 subunit (amino acids, aa 1–685) and the S2 subunit (aa 686–1211) [4, 5].

The S1 subunit contains the receptor-binding domain (RBD), which mediates binding to the angiotensin-converting enzyme 2 (ACE2) receptor on host cells [6]. The S2 subunit can be divided into several regions: a cleavage site at R815, fusion peptide (FP), heptad repeat 1 (HR1), central helix (CH), stem-helix (SH), and heptad repeat 2 (HR2) [5, 7]. Once the RBD attaches to ACE2, the S1 subunit is shed from the Spike protein [4, 6]. This triggers S2 subunit-mediated virus-cell membrane fusion. During the membrane fusion process, the FP inserts in the host cell membrane and triggers the refolding of the S2 subunit, enabling the virus to enter the host cell [8].

nAbs targeting the Spike protein have demonstrated potent neutralizing activity. Monoclonal antibodies (mAbs) that target the S1 subunit, such as REGN-10987 [9, 10], S309 [11], VIR-7229 [12], and 35B5 [13], can block RBD-ACE2 binding to exert their neutralization function. While mAbs targeting the S2 subunit—such as the FP-binding mAbs (76E1 [14], Fp.006 [15], and C77G12 [16]) and the SH-binding mAbs (CC40.8 [17], S2P6 [18], and CV3-25 [19])—inhibit viral-host membrane fusion, thereby exerting their neutralization function. Recently, the S2 apex has emerged as another conserved target for broadly reactive antibodies. 54043-5 was identified as a broadly reactive, non-nAb targeting a cryptic epitope at the apex of the S2 subunit. Despite lacking direct neutralizing activity, 54043-5 mediated Fc-dependent antiviral functions, including antibody-dependent cellular cytotoxicity (ADCC) and antibody-dependent cellular phagocytosis (ADCP), and its Fc-silencing variant provided protection in mouse models [20]. These findings highlight the functional diversity of S2-directed antibodies and the potential of conserved S2 epitopes as targets for developing broad-spectrum anti-coronavirus interventions.

Despite these advances, mutations in the Spike protein can either directly disrupt antibody binding or induce conformational changes in the Spike protein, leading to evasion of mAb neutralization [21]. As early as 2020, the D614G single-point mutation in the S1 subunit of the Spike protein emerged [22]. This mutation was found to increase virion Spike density and infectivity, as well as partially evade neutralization by some Spike-targeting antibodies [23, 24]. The currently circulating JN.1 sub-lineage of Omicron variant carries 31 Spike protein mutations, including D614G, relative to the WT Wuhan-Hu-1 strain [25]. These mutations impair the neutralizing activity of existing nAbs [26, 27]. Therefore, developing broadly nAbs (agents) and engineering strategies to enhance antibody breadth and neutralizing potency may provide promising approaches for blocking viral entry across diverse SARS-CoV-2 variants and guiding the development of next-generation antiviral therapeutics.

In this study, we first identified a human anti-S2 mAb, P23, from a non-immune phage display single-chain variable fragment (scFv) library that cross-binds to Spike proteins of both WT SARS-CoV-2 and Pangolin-CoV. P23 exhibited potent neutralizing activity and recognized an S2 epitope distinct from six previously reported S2 antibodies, but it failed to neutralize D614G-containing variants. To overcome this, we engineered a fusion protein (Bs-ACE2-P23) by fusing the hACE2 ECD to the N-terminus of the heavy chain (HC) of P23, which showed potent neutralization against SARS-CoV-2 variants of concern (VOCs), including multiple Omicron subvariants. Additionally, we found that anti-Spike mAbs—particularly mAbs targeting epitope adjacent to or directly with the FP—when fused to the hACE2 ECD using the Bs-ACE2-P23 architecture design_, exhibited potent neutralizing activity against the SARS-CoV-2 JN.1 variant, compared to their parent proteins. Collectively, our work expands the existing repertoire of S2-targeting antibodies and introduces a new ACE2-anti-S2 antibody bispecific fusion protein architecture may be used for future drug development for the prevention and treatment of infections caused by SARS-CoV-2 variants and potentially other coronaviruses.

Materials and methods

Cell lines

HEK293T (RRID: CVCL_0063) and the HEK293T-hACE2 stable cell line were cultured in DMEM (Gibco, Cat. 11965092) supplemented with 10% FBS (Gibco, Cat. A5669701) at 37°C with 5% CO₂. Expi293F cells (Thermo, Cat. A14527) and Expi293F GnTI⁻ cells [28] were maintained in SMM 293-TII Expression Medium (Sino Biological, Cat. M293TII) at 37°C with 8% CO₂, and shaking at 125 rpm.

Expression and purification of spike-associated proteins

For anti-Spike antibody panning and binding assessment, Spike ECD proteins of SARS-CoV-2 (Wuhan-Hu-1, aa 1–1211, GenBank YP_009724390.1, with R682S/R685S/K986P/V987P mutations), SARS-CoV-2 D614G (aa 1–1211), and Pangolin-CoV (aa 1–1203, GenBank QLR06867.1), as well as the SARS-CoV-2 RBD (aa 316–512) were C-terminally fused with a His₆-Avi tag for purification and biotinylation. The corresponding coding sequences were cloned into expression vectors and co-transfected with a BirA ligase-expressing vector into Expi293F cells using polyethyleneimine (PEI, Polysciences, Cat. 23966).

For surface plasmon resonance (SPR) and hydrogen–deuterium exchange mass spectrometry (HDX-MS) analyses, the Spike ECD of WT SARS-CoV-2 (Wuhan-Hu-1, aa 1–1211, with R682S/R685S/K98-6P/V987P mutations), Pangolin-CoV (aa 1–1203), and the SARS-CoV-2 JN.1 variant (aa 1–1211, Sino Biological, Cat. VG40986-UT), along with the WT SARS-CoV-2 S1 subunit (aa 1–685) and S2 subunit (aa 686–1137, with K986P/V987P mutations), were C-terminally fused with a His₈-Protein C tag. The coding sequences were cloned into expression vectors and transfected into Expi293F cells or GnT1⁻ Expi293F cells (for HDX-MS) using PEI. These Spike-associated proteins were affinity-purified using Ni-NTA agarose (Qiagen, Cat. 30210).

Antibody library panning and screening of spike-specific nAbs

For the phage display library panning, a non-immune human scFv phage display library was used for affinity selection [29]. Biotinylated Spike ECDs from WT SARS-CoV-2 and Pangolin-CoV were used as panning antigens in parallel. After two rounds of panning, individual phage clones were rescued and their initial binding to WT SARS-CoV-2 Spike protein was tested by flow cytometry, as previously described [30]. Phage clones exhibiting a mean fluorescence intensity greater than two-fold higher than the blank control were selected for sequencing and subsequent pseudovirus neutralization assays.

To evaluate the neutralizing activity of phage-scFv clones, phage-scFv particles were precipitated and concentrated from bacterial culture supernatants as previously described [29, 30]. The concentrated phage-scFvs were pre-incubated with a lentivirus-based WT SARS-CoV-2 pseudovirus harboring R682S/R685S mutations in the Spike protein for 30 min at room temperature, followed by infection of HEK293T-hACE2 stable cells as previously described [31]. Luciferase activity was measured to quantify pseudovirus infection. Clones showing potent neutralizing activity (>95% inhibition of pseudovirus infection) were selected for further characterization.

Expression and purification of spike-specific mAbs and bispecific fusion proteins

The variable heavy (VH) and variable light (VL) coding sequences of selected scFvs were cloned into human IgG1 (hIgG1) or mouse IgG2a (mIgG2a) HC and light chain (LC) expression vectors, respectively. The VH and VL sequences of six reported SARS-CoV-2 nAbs (S309 [11], REGN-10987 [9], 35B5 [13], S2P6 [18], Fp.006 [15], and CV3-25 [32]) were retrieved from the CoV-AbDab repository [33], codon-optimized for mammalian expression, synthesized (Rui Biotech, China), and subcloned into hIgG1 HC/LC vectors. The sequences of CC40.8 [17], 76E1 [14], and 54043-5 [20] were codon-optimized and synthesized as scFv-hFc expression plasmids (GenScript, China). Expi293F cells were transiently transfected with HC and LC plasmids (1:1) for IgG expression or with the corresponding scFv-hFc plasmids for scFv-hFc expression. Cell culture supernatants were harvested 5–7 days after transfection, and proteins were purified by Protein A affinity chromatography.

hACE2-hFc was constructed by fusing the hACE2 ECD (aa 18-615, H374N/H378N, UniProt Q9BYF1) to the hIgG1 hinge and Fc region as previously described [31]. Bs-Ab-P23 constructs were generated by fusing mAb scFv to the N-terminus of P23 hIgG1 HC via a 3 × G₄S linker. Bs-ACE2-Ab constructs were made by fusing hACE2 ECD to the VH N-terminus of seven hIgG1 antibodies (P23, S309, REGN-10987, 35B5, S2P6, Fp.006, CV3-25) via a 3 × G₄S linker. For Bs-ACE2-P23-V2 to V4, hACE2 ECD was fused to P23’s VH C-terminus, VL N-terminus, or VL C-terminus, respectively. Bs-ACE2-P23-V5 was constructed by fusing P23 scFv to the N-terminus of hACE2-hFc. For Bs-ACE2-P23-V6, hACE2 ECD was inserted into the “hole” Fc chain (Y349C, T366S, L386A, Y407A), and P23 VH was cloned into the “knob” Fc chain (S354C, T366W) as previously described [34]. All proteins were expressed in Expi293F cells and purified by Protein A affinity chromatography.

Enzyme-linked immunosorbent assay

For binding Enzyme-linked immunosorbent assay (ELISA), streptavidin-coated 96-well plates (Thermo, Cat. 436014) were incubated with biotinylated Spike ECD (2 μg/ml), followed by serial dilutions of mAbs. Detection was performed with anti-human Fc-HRP (Thermo, Cat. 31413).

For competitive ELISA, mAbs in the hIgG1 or scFv-hFc format were used at sub-saturating binding concentrations and pre-mixed with serial dilutions of competitor antibodies in the mIgG2a format. Mixtures were added to Spike-coated 96-well plates and incubated. Bound hIgG1 was detected with anti-human Fc-HRP as previously described. Optical density (OD) was measured at OD₄₅₀–OD₆₃₀ using a Bio-Rad iMark microplate reader.

Binding affinity analysis by SPR

SPR assays were performed using a Biacore T200 system (Cytiva). To assess the binding of 8 selected mAbs to Spike, S1, and S2 subunits, mAbs were diluted and captured at similar response unit (RU) levels on a CM5 sensor chip immobilized with anti-human IgG antibodies. Then, 200 nM of Spike, S1, or S2 fusion proteins were flowed over the captured mAbs for association, followed by dissociation.

To assess the binding affinity of mAbs and IgG-like bispecific fusion proteins to the Spike ECD of WT SARS-CoV-2 and SARS-CoV-2 JN.1 variant, mAbs or fusion proteins were captured on a CM5 sensor chip immobilized with anti-human IgG antibodies. Subsequently, two-fold serial dilutions (3.125–200 nM) of SARS-CoV-2 Spike ECD (WT and JN.1 variant) or S2 subunit were flowed over the captured proteins to measure association and dissociation. All SPR data were analyzed using Biacore T200 Evaluation Software (Cytiva) with a global fitting approach based on the 1:1 Langmuir binding model.

Flow cytometry assays

The binding of mAbs to cell surface-expressed Spike proteins was detected by flow cytometry. Plasmids encoding Spike proteins of various coronaviruses (WT SARS-CoV-2, SARS-CoV-2 JN.1, SARS-CoV-1 (GenBank UUW06557.1), Pangolin-CoV (GenBank QLR06867.1), Bat-CoV RaTG13 (GenBank QHR63300.2), MERS-CoV (HCoV-EMC/2012, GenBank AFS88936.1), and HCoV-HKU1 (GenBank YP_173238.1)) were cloned into an expression vector with a C-terminal IRES-RFP tag (GenBank UDF48714.1), then transfected into Expi293F cells using PEI.

Two days after transfection, cells were incubated with the indicated antibodies. Spike-expressing cells were gated by Allophycocyanin (APC) fluorescence. Binding of mAbs (10 μg/ml) was detected with FITC-conjugated anti-human IgG Fc (Sigma-Aldrich, Cat. F5512). For phage-displayed scFvs (phage-scFvs), binding was detected with anti-M13-mFc antibody (GE Healthcare, Cat. 27-9420-01), followed by FITC-conjugated anti-mouse IgG Fc (Sigma-Aldrich, Cat. F0257). Data were acquired on a BD FACS Aria III flow cytometer (BD Biosciences), and analyzed using FCS Express 6 (De Novo Software).

Protein thermal shift assays

Protein thermal shift assays were performed using the Protein Thermal Shift™ Dye Kit (Applied Biosystems, Cat. 44-611-46) according to the manufacturer’s protocol. Melting curves for individual mAbs were acquired on an Applied Biosystems 7500 Fast Real-Time Polymerase chain reaction (PCR) System with standardized parameters. Data were analyzed using Protein Thermal Shift™ Software v1.0, and melting temperatures (Tm) were determined by both Boltzmann and derivative methods.

Pseudovirus neutralization assays

Lentivirus-based SARS-CoV-2 pseudoviruses were produced as previously described [31]. HEK293T cells were transfected with a plasmid mixture of psPAX2 (Addgene, Cat. 12260), pHIV-luc (Addgene, Cat. 21375), and a Spike-encoding vector (WT SARS-CoV-2 [R682S/R685S], SARS-CoV-2 D614G, or Pangolin-CoV) at a 3:4:1 mass ratio, using Lipofectamine 3000 (Thermo, Cat. L3000015). The medium was replaced with DMEM containing 2% FBS 4 h post-transfection. Pseudovirus-containing supernatants were collected 48 h later. Vesicular stomatitis virus (VSV)-based pseudoviruses for WT SARS-CoV-2 (Cat. DD1502) and variants (XBB (Cat. DD1594), B.1.1.529 (Cat. DD1568), BA.2 (Cat. DD1569), XBB.1.5 (Cat. DD1597), BA.2.86 (Cat. DD15108), JN.1(Cat. DD15110), and XDV (Cat. DD15114)) were purchased from Vazyme Biotech.

Neutralization assays were performed as follows. HEK293T-hACE2 stable cells (8 × 10⁴) were seeded into 96-well white flat-bottom plates (Corning, Cat. 3917) 20 h before infection. Test samples (hACE2-hFc, mAbs, and fusion proteins) were serially diluted in DMEM containing 2% FBS, starting at 100, 50, or 10 μg/ml, and added in duplicate. For each well, 35 μl of the diluted sample was mixed with 75 μl of pseudovirus and incubated for 30 min at room temperature. The mixture was then added to cells for a final volume of 100 μl, followed by incubation at 37°C with 5% CO₂. For lentivirus-based pseudoviruses, the medium was replaced with fresh DMEM containing 2% FBS at 24 h post-infection, followed by an additional 24 h incubation before luciferase detection. For VSV-based pseudoviruses, luciferase was detected at 24 h post-infection. Luciferase activity was quantified using the Dual-Luciferase Reporter Assay System (Promega, Cat. E1910) on a PerkinElmer microplate reader. Neutralization percentage was calculated as: % neutralization = [1 – (RLU_sample – RLU_background)/ (RLU_virus_only – RLU_background)] × 100.

Authentic virus neutralization assays

Authentic virus neutralization assays were performed in a Biosafety Level 3 (BSL-3) laboratory at the Chinese Academy of Medical Sciences and Peking Union Medical College (Beijing, China). Delta (IPBCAMS_De0305SP_4/2021, GWH accession number: GWHBHTE01000000) and JN.1 (IPBCAMS-JN01–1/2024, GenBase accession number: C_AA082664.1) strains were used as authentic viruses, with Vero E6 cells (ATCC, CRL-1586) as the host cell line.

Neutralization assays were performed as follows. Vero E6 cells (1 × 10⁴) were seeded into 96-well plates 24 h before infection. Test proteins (Bs-ACE2-P23, hACE2-hFc, P23, and a 1:1 molar mixture of hACE2-hFc and P23) were serially diluted in DMEM starting at 1,000 nM. Each dilution (50 μl) was mixed with 50 μl of authentic virus at 100 TCID₅₀ and incubated for 1 h at room temperature. Cells were washed twice with PBS, incubated with the virus–protein mixture for 1 h at 37°C. The inoculum was subsequently removed, the cells were washed once, and then cultured with 100 μl of DMEM containing 2% FBS per well. The cytopathic effect (CPE) was monitored daily for a period of 5 days using light microscopy.

HDX-MS analysis

To analyze P23 binding to S2, deglycosylated S2 protein (8 μg/μl) and P23 hIgG1 (24 μg/μl) were buffer-exchanged into DPBS (pH 7.4). Complexes were formed at a 1:1 volume ratio and incubated for 1 h at 25°C; S2 alone served as the control. For P23 binding to Spike ECD (aa 1–1211), deglycosylated Spike ECD (3 μg/μl) and P23 hIgG1 (6 μg/μl) were buffer-exchanged into DPBS (pH 7.4), mixed at a 4:1 volume ratio, and incubated for 1 h at 25°C, with Spike ECD alone as the control.

For hydrogen–deuterium exchange, each sample was diluted 1:20 in DPBS prepared with 99% D₂O (pH 7.04) at 0 °C. Reactions were quenched after 30, 90, 300, or 1800 s (for P23-S2) or 30, 90, 300, or 600 s (for P23-Spike) with an equal volume of 200 mM glycine buffer (pH 2.5) at 0°C. Quenched samples were digested on ice. Peptic peptides were trapped and desalted on a C18 guard column, then separated on an Acquity UPLC BEH C18 (1.7 μm, 1.0 mm × 50 mm) column (Waters) using a 0–100% gradient of acetonitrile containing 1% (v/v) formic acid. Mass spectrometric analysis was performed on a Q Exactive™ Orbitrap mass spectrometer (Thermo, CA). Data analysis was conducted using the HDExaminer software (Sierra Analytics) and the Thermo Xcalibur Qual Browser (Thermo, CA). Deuterium uptake was quantified relative to the 0-s control for each peptide.

Pharmacokinetics analyses

Healthy 8-week-old female BALB/c mice were administered a single intravenous injection of 10 mg/kg of P23, hACE2-hFc, or Bs-ACE2-P23. Blood samples were collected at 15 min, 45 min, 3 h, and 8 h after administration. Serum concentrations of P23, hACE2-hFc, and Bs-ACE2-P23 were determined by ELISA using RBD- or S2-coated plates. Standard curves were generated using known concentrations of the corresponding proteins (P23, hACE2-hFc, or Bs-ACE2-P23), and serum protein concentrations were calculated based on the respective standard curves.

ADCC and ADCP assays

ADCC and ADCP assays were performed as previously described [35]. Expi293F cells transiently expressing full-length SARS-CoV-2 Spike (WT or JN.1) for 2–3 days were used as target cells in both assays. For the ADCC assay, NK92-MI^hCD16 cells were used as effector cells, and cytotoxicity was determined by measuring lactate dehydrogenase (LDH) release using the CytoTox 96® Non-Radioactive Cytotoxicity Assay Kit (Promega, Cat. G1780).

For the ADCP assay, bone marrow-derived macrophages (BMDMs) were generated from bone marrow cells isolated from adult male C57BL/6 N mice and used as effector cells. Target cells were labeled with CellTracker Red CMTPX (MKbio, Cat. MX4109), and BMDMs were stained with anti-mouse F4/80-FITC (clone BM8, BioLegend, Cat. 123108). Target cells were co-cultured with BMDMs for 2 h, and phagocytosis was assessed by confocal microscopy (Nikon A1 SIM).

Statistical analyses

Data from ELISA and neutralization assays were analyzed using GraphPad Prism 9.0. EC₅₀ and IC₅₀ values were determined by four-parameter logistic regression.

Results

Identification of a nAb against pangolin-CoV and SARS-CoV-2

To identify broadly nAbs against SARS-CoV-2, we used the Spike proteins of both WT SARS-CoV-2 (Wuhan-Hu-1) and Pangolin-CoV (GD/1/2019) as panning antigens. Pangolin-CoV shares 94.4% amino acid identity with the WT SARS-CoV-2 Spike but is more divergent from the WT Spike than the Spike proteins of the six major SARS-CoV-2 VOCs (Supplementary Fig. S1A–D) [36]. We reasoned that inclusion of Pangolin-CoV Spike would facilitate the isolation of antibodies recognizing evolutionarily conserved epitopes, thereby increasing the likelihood of identifying antibodies with broad neutralizing potential against diverse SARS-CoV-2 variants. We therefore generated recombinant fusion proteins containing the Spike ECDs of WT SARS-CoV-2 (aa 1–1211) and Pangolin-CoV (aa 1–1203) and used them to pan a non-immune human scFv phage display library (Supplementary Fig. S2) [29].

After two rounds of library panning with the WT SARS-CoV-2 or Pangolin-CoV Spike proteins, we used flow cytometry to identify monoclonal phage-scFvs that specifically bind to HEK293T cells expressing the full-length WT SARS-CoV-2 Spike protein. This screening identified 19 phage-scFv clones from panning against the WT SARS-CoV-2 Spike, designated NS2–NS48, and 15 phage-scFv clones from panning against the Pangolin-CoV Spike, designated P14–P56 (Supplementary Fig. S3A). These phage-scFvs were subsequently evaluated for their ability to neutralize lentivirus-based WT SARS-CoV-2 pseudovirus, with 8 phage-scFvs (P23, P28, P29, P48, NS2, NS14, NS28, and NS29) exhibited potent neutralizing activity (>95%, Supplementary Fig. S3B). These 8 scFvs were then expressed as hIgG1 proteins and assessed for their binding properties, neutralizing activity, and thermal stability to identify a lead antibody for further study.

Cross-binding reactivity of the eight hIgG1-format mAb to both Pangolin-CoV and WT SARS-CoV-2 Spike ECD was first analyzed using ELISA. They exhibited potent cross-binding activity, with EC₅₀ values of 36.3–138.4 pM for SARS-CoV-2 Spike ECD and 11.4–129.2 pM for Pangolin-CoV Spike ECD (Fig. 1A). Among the 8 mAbs tested, NS2, NS28, P23 and P28 exhibited strong binding activity to WT Spike ECD of SARS-CoV-2 with low EC₅₀ values <70 pM, and high maximal cross-binding level with the Spike proteins of both WT SARS-CoV-2 and Pangolin-CoV.

Figure 1.

Screening and identification of nAbs against the SARS-CoV-2. (A) ELISA-binding profiles of nAbs in hIgG1 format to the WT SARS-CoV-2 or pangolin-CoV spike ECD. EC₅₀ values are shown on the right. (B) Binding affinities of eight nAbs to the SARS-CoV-2 spike S2 subunit. Binding kinetics and affinities were measured using SPR using a Biacore T200. (C) Cross-neutralization activity analysis of nAbs against WT SARS-CoV-2 and pangolin-CoV in pseudovirus neutralization assays. IC₅₀ values are shown on the right. (D) Thermal stability of nAbs assessed by protein thermal shift assays. Baseline melting temperature (TmB) and denaturation melting temperature (TmD) were calculated using Protein Thermal Shift™ Software v1.0 (mean ± SD, n = 4).

To identify the epitopes targeted by the 8 mAbs on the SARS-CoV-2 spike protein, we employed SPR assays to evaluate their binding to the full-length WT Spike ECD, S1 subunit, and S2 subunit. We found that all 8 mAbs bound to both the Spike ECD and the S2 subunit, but not to the S1 subunit (Supplementary Fig. S4). Additionally, 5 of the 8 mAbs (NS14, NS28, NS29, P23, and P48) exhibited high affinity (K_D = 6.5–14.4 nM) to the S2 subunit (Fig. 1B). These results demonstrate that their binding epitopes are located within the S2 subunit.

We then evaluated neutralizing activity of these 8 mAbs against lentivirus-based WT SARS-CoV-2 and Pangolin-CoV pseudoviruses and found that all the tested mAbs exhibited cross-neutralizing activity against both WT SARS-CoV-2 (IC₅₀ value ranging from 107.0 to 344.5 pM) and Pangolin-CoV (IC₅₀ value ranging from 48.0 to 118.9 pM) (Fig. 1C). Among them, NS14, NS28, and P23 exhibited superior neutralizing activities against WT SARS-CoV-2, with IC₅₀ values of 107.0 pM, 109.7 pM, and 122.8 pM, respectively.

Given that antibodies with high thermal stability are advantageous for developing future therapeutics, we sought to further identify mAbs exhibiting superior thermostability. Using protein thermal shift assays, we evaluated the thermal stability of the 8 mAbs and found that two mAbs, P23 (Tm B = 78.52 ± 0.25°C, Tm D = 82.46 ± 0.09°C) and P28 (Tm B = 75.60 ± 0.25°C, Tm D = 83.15 ± 0.12°C), showed better thermal stability compared to the other six mAbs (Fig. 1D, Supplementary Fig. S5). Considering the binding ability, neutralizing activity and thermal stability, we selected P23 for further study.

P23 binds to an epitope adjacent to the FP in the S2 subunit

We next performed ELISA-based competition assays to determine whether P23 competes with six representative anti-S2 antibodies targeting distinct epitopes. Three antibodies (S2P6, CV3-25, and CC40.8) recognize the stem helix; two antibodies (Fp.006 and 76E1) target the FP; and one antibody (54043-5) binds the apex region [14, 15, 17–20]. P23 exhibited only partial competition with Fp.006 (~20%) and 76E1 (~30%) but showed no detectable competition with S2P6, CV3-25, CC40.8, or 54043-5 (Fig. 2A). These findings suggest that P23 recognizes an epitope distinct from those of S2P6, CV3-25, CC40.8, and 54043-5, while partially overlapping with the FP epitope recognized by Fp.006 and 76E1.

Figure 2.

Characterization of the binding and neutralizing properties of P23 compared with reported anti-S2 mAbs. (A) Competitive inhibition of S2P6, Fp.006, and CV3-25 (hIgG1 format) and CC40.8, 76E1, and 54043-5 (scFv-hFc format) binding to the SARS-CoV-2 spike protein by P23 mIgG2a was assessed using competitive ELISA. NC, negative control (PBS for S2P6, Fp.006, and CV3-25; control IgG for CC40.8, 76E1, and 54043-5). OD₄₅₀–OD₆₃₀ values are plotted against P23 concentration. (B) Cross-binding activity assessed by flow cytometry. Expi293F cells expressing full-length spike proteins from SARS-CoV-2 (WT, D614G, and JN.1), bat-CoV RaTG13, pangolin-CoV, MERS-CoV, HCoV-HKU1, or SARS-CoV-1 were stained with 10 μg/ml of each indicated antibody. (C) Neutralization activity of P23, S2P6, Fp.006, and CV3-25 against lentivirus-based pseudoviruses of WT SARS-CoV-2, SARS-CoV-2 D614G, and pangolin-CoV. (D) Neutralization activity of P23, S2P6, Fp.006, and CV3-25 against the VSV-based pseudoviruses of WT SARS-CoV-2. (E) IC₅₀ values corresponding to (C) and (D). n.m., not measurable.

We then used flow cytometry to examine the binding of P23, S2P6, Fp.006, and CV3-25 to Expi293 cells expressing the Spike protein of WT SARS-CoV-2, two SARS-CoV-2 variants (D614G and JN.1), and five other β-coronavirus (HCoV-HKU1, MERS-CoV, SARS-CoV-1, Pangolin-CoV, and RaTG13). These four nAbs displayed diverse cross-binding activity to cells expressing different Spike proteins: Fp.006 and S2P6 bound to all Spike proteins tested; CV3-25 and P23 bound to the Spike proteins of SARS-CoV-2 (WT and the two variants), SARS-CoV-1, Pangolin-CoV, and RaTG13, but not to the Spike proteins of MERS-CoV and HCoV-HKU1 (Fig. 2B).

We then compared neutralizing activities of P23, S2P6, Fp.006, and CV3-25 against pseudoviruses of WT SARS-CoV-2, SARS-CoV-2 D614G, and Pangolin-CoV. P23 exhibited superior neutralizing activity against lentivirus-based WT SARS-CoV-2 (IC₅₀ = 112.8 pM) and Pangolin-CoV (IC₅₀ = 107.1 pM), and VSV-based WT SARS-CoV-2 (IC₅₀ = 212.1 pM), compared to the other three tested nAbs (Fig. 2C–E). However, for the lentivirus-based SARS-CoV-2 D614G, S2P6, Fp.006, and CV3-25, but not P23, exhibited neutralizing activity (Fig. 2C). P23’s potent neutralizing activity against WT SARS-CoV-2 and Pangolin-CoV, yet lack of activity against the D614G variant, may reflect its distinct binding epitope and neutralization mechanism relative to S2P6, Fp.006, and CV3-25. To investigate this difference, we compared P23 binding to the WT SARS-CoV-2 Spike ECD and the D614G Spike ECD by ELISA. P23 bound the D614G Spike ECD less strongly than the WT Spike ECD (Supplementary Fig. S6), suggesting that the D614G substitution may alter the Spike conformation and thereby reduce the exposure or accessibility of the P23 epitope.

To define the P23 epitope, we employed an HDX–MS method. Briefly, reduced deuterium uptake in the antibody–antigen complex relative to antigen alone indicates protected binding interfaces. Analysis of the P23–S2 complex versus S2 alone identified four protected peptides: #18 (aa 824–833), #20 (aa 858–865), #41 (aa 931–945), and #43 (aa 946–961) (Fig. 3A–B, Supplementary Fig. S7A). Peptide #41 showed the largest deuterium exchange difference (ΔDeut >1 Da) across all time points, indicating a dominant contribution to P23 binding.

Figure 3.

Epitope mapping of P23 on the S2 subunit and spike protein with HDX-MS. (A) Heatmap of deuterium uptake differences (ΔDeut = the Deut of S2 + P23 – The Deut of S2) across all identified peptides at four timepoints (30, 90, 300, 1800 s). (B) Kinetic plots of deuterium incorporation over time for identified peptide #41, comparing free S2 with P23-S2 complex. (C) Heatmap of deuterium uptake differences (ΔDeut = the Deut of spike+P23 – The Deut of spike) for peptides 200–300 at four timepoints (30, 90, 300, 600 s). (D) Kinetic plots of deuterium incorporation over time for identified peptide #212, comparing free spike with P23-spike complex. (E) Mapping of the FP and residues 934–945 on the cryo-EM structure of spike (PDB 6XR8).

Further analyzing the P23–full-length Spike complex versus Spike alone using HDX–MS identified seven protected peptides: #65 (aa 757–770), #111 (aa 832–843), #182 (aa 907–913), #184 (aa 907–914), #212 (aa 934–948), #220 (aa 946–961), and #274 (aa 1014–1021) (Fig. 3C–D, Supplementary Fig. S7B and C). Peptide #212, overlapping ~80% with #41, exhibited the largest ΔDeut (>1 Da), indicating that P23 targets the region spanning their overlapping residues 934–945. Using a reported structure of the Spike trimer protein (PDB: 6XR8) [4], we mapped residues 934–945 alongside the FP (aa 816–834) and found they are in close spatial proximity, with an inter-helical distance of 6–8 Å (Fig. 3E). This observation supports that P23 recognizes an epitope located near the FP within the S2 subunit.

An IgG-like P23/ACE2 bispecific fusion protein exhibits potent neutralizing activity against the SARS-CoV-2 JN.1 variant

To overcome this limitation that P23 failed to neutralize D614G-containing SARS-CoV-2 variants, we generated IgG-like fusion proteins by combining P23 with the hACE2 ECD or RBD-targeting nAbs. The rationale for this design is that it can simultaneously target the two.

steps of viral infection: the RBD binding to ACE2 and the virus-cell membrane fusion mediated by the S2. We selected three nAbs, REGN-10987, S309, and 35B5, which bind to the S1 and have different binding epitopes [9, 11, 13], as well as the hACE2 ECD (aa 18-615, containing H374N/H378N mutations [31]) for constructing four IgG-like bispecific fusion proteins: Bs-REGN-10987-P23, Bs-S309-P23, Bs-35B5-P23 and Bs-ACE2-P23. Briefly, the scFvs of these nAbs and the hACE2 ECD were fused and expressed at the N-terminus of the HC of the P23 hIgG1 via a 3 × G₄S linker (Fig. 4A).

Figure 4.

Binding activity of IgG-like bispecific fusion proteins to spike proteins and their neutralizing activity against the SARS-CoV-2 JN.1 pseudovirus. (A) Schematic design of two representative IgG-like bispecific fusion proteins. (B and C) Binding affinity analysis of antibodies (in hIgG1 format) and bispecific proteins binding to the spike ECD of SARS-CoV-2 WT (B) or the JN.1 variant (C). (D) Comparison of neutralizing activities of the four IgG-like fusion proteins to hACE2-hFc and P23 hIgG1 against VSV-based JN.1 pseudovirus. n.m., not measurable.

After obtaining the four IgG-like bispecific fusion proteins, we first measured their binding to the Spike ECD proteins of WT SARS-CoV-2 and SARS-CoV-2 JN.1 using SPR. Upon fusion with P23, the binding affinities of Bs-REGN-10987-P23, Bs-35B5-P23, and Bs-ACE2-P23 to the WT Spike were enhanced by 53.0-fold, 13.0-fold, and 116.5-fold, respectively, compared to their parental forms (Fig. 4B, Supplementary Table S1). In contrast, the binding affinity of Bs-S309-P23 was reduced by 21.2-fold relative to S309. Compared to P23 (K_D = 10.0 nM), the binding affinities of Bs-REGN-10987-P23, Bs-S309-P23, Bs-35B5-P23, and Bs-ACE2-P23 to the WT Spike were enhanced by 4.3-fold, 2.0-fold, 5.5-fold, and 30.6-fold, respectively (Fig. 4B, Supplementary Table S1). For the binding affinity to JN.1 Spike, the binding affinity of Bs-REGN-10987-P23 decreased by 4.4-fold compared to P23 (K_D = 250.0 nM); the binding affinities of Bs-S309-P23, Bs-35B5-P23, and Bs-ACE2-P23 to the JN.1 Spike were comparable to that of P23 and hACE2-hFc (Fig. 4C, Supplementary Table S1).

To evaluate the neutralizing activity of these four IgG-like bispecific fusion proteins against the SARS-CoV-2 JN.1, we performed VSV-based pseudovirus neutralization assays against SARS-CoV-2 JN.1 variant. The neutralizing activities of Bs-REGN-10987-P23, Bs-S309-P23, Bs-35B5-P23, and P23 against the JN.1 pseudovirus were below the detection limit (Fig. 4D). Notably, Bs-ACE2-P23 neutralized the JN.1 pseudovirus with an IC₅₀ value of 34.1 pM, representing a 56.4-fold enhancement in neutralizing activity compared to hACE2-hFc (IC₅₀ = 1921.3 pM) (Fig. 4D), demonstrating that Bs-ACE2-P23 effectively neutralizes the D614G-containing SARS-CoV-2 JN.1 variant.

The Bs-ACE2-P23 bispecific protein exhibits synergistically enhanced neutralizing activity against the SARS-CoV-2 D614G and JN.1 variants

To investigate whether alternative architectures of ACE2 ECD-P23 IgG-like bispecific fusion proteins can exhibit superior neutralizing activity compared to Bs-ACE2-P23, we generated five additional distinct architectures of ACE2 ECD-P23 bispecific fusion proteins for comparative analysis with the Bs-ACE2-P23 (Fig. 5A, Supplementary Fig. S8). SPR analysis revealed that only Bs-ACE2-P23-V5 exhibited improved binding affinity compared with Bs-ACE2-P23, with ~93.8-fold and 3.2-fold lower K_D values for the WT SARS-CoV-2 Spike and SARS-CoV-2 JN.1 Spike, respectively (Fig. 5B–C, Supplementary Table S1).

Figure 5.

Bs-ACE2-P23 demonstrates enhanced neutralizing activity against SARS-CoV-2 variants. (A) Schematic illustration of six ACE2-P23 IgG-like bispecific fusion proteins. (B–C) Binding affinity analysis of antibodies (in hIgG1 format) and bispecific proteins binding to the spike ECD of SARS-CoV-2 WT (B) or the JN.1 variant (C). (D) Neutralizing activity of six fusion proteins against the pseudovirus of SARS-CoV-2 D614G and JN.1 variants. (E) Neutralization activity of Bs-ACE2-P23, hACE2-hFc, P23 hIgG1, and the hACE2-hFc/P23 hIgG1 combination against lentivirus-based WT SARS-CoV-2 and SARS-CoV-2 D614G pseudoviruses, as well as VSV-based SARS-CoV-2 JN.1 pseudovirus. n.m., not measurable. For combination treatment, hACE2-hFc and P23 hIgG1 were mixed at a 1:1 weight ratio. IC₅₀ values were calculated based on the total protein concentration of the combination using GraphPad Prism. The corresponding molar concentrations of each component were calculated according to their respective molecular weights.

We next compared their neutralizing activity against SARS-CoV-2 D614G and SARS-CoV-2 JN.1. Bs-ACE2-P23 exhibited the most potent neutralizing activity against both D614G and JN.1 pseudoviruses (IC₅₀ values of 31.9 pM and 43.6 pM, respectively) (Fig. 5D), suggesting that the architecture of Bs-ACE2-P23 is better than other tested architectures for neutralizing activity against SARS-CoV-2 variants.

We further evaluated whether the neutralizing activity of Bs-ACE2-P23 is better than the combination of P23 and hACE2-hFc. Bs-ACE2-P23 exhibited the best neutralizing activity against the three pseudoviruses tested, including WT SARS-CoV-2, D614G, and JN.1 variants, surpassing the activity of P23, hACE2-hFc, or their combination (Fig. 5E).

We next evaluated the biophysical stability and pharmacokinetic profiles of P23, hACE2-hFc, and Bs-ACE2-P23 in mice. Thermal shift analysis revealed that hACE2-hFc and Bs-ACE2-P23 exhibited reduced thermal stability compared with P23, with an ~25°C decrease in Tm B (Supplementary Fig. S9). Serum persistence was further evaluated using S2-based and RBD-based ELISAs to monitor P23- and ACE2-mediated binding activities, respectively. S2-based ELISA showed that Bs-ACE2-P23 was cleared more rapidly from serum than P23 (Supplementary Fig. S10A). Similarly, RBD-based ELISA revealed rapid loss of ACE2-mediated binding activity for both hACE2-hFc and Bs-ACE2-P23 within 8 h after administration (Supplementary Fig. S10B). These preliminary analyses provide an initial characterization of the biophysical stability and pharmacokinetic properties of Bs-ACE2-P23 in mice. Further studies will be required to comprehensively evaluate its developability and guide subsequent optimization.

Finally, we investigated whether ACE2 fusion affected the Fc-mediated effector functions of P23. S1-F4, a previously reported hIgG1 antibody targeting CD98 with potent ADCC and ADCP activity, was used as a positive control [35]. In the ADCC assay, S1-F4 induced cytotoxicity against Expi293F cells expressing either WT or JN.1 Spike, whereas P23, hACE2-hFc, and Bs-ACE2-P23 showed no detectable ADCC activity against either target cell compared with the NC group (Supplementary Fig. S11A–B). In the ADCP assay, S1-F4, P23, hACE2-hFc, and Bs-ACE2-P23 all mediated efficient phagocytosis of WT Spike-expressing cells compared with the NC group. Similarly, P23 also retained ADCP activity against JN.1 Spike-expressing cells, although its phagocytic activity was lower than that mediated by S1-F4, hACE2-hFc, and Bs-ACE2-P23 (Supplementary Fig. S11C–D). The reduced ADCP activity of P23 against JN.1 Spike-expressing cells was consistent with its decreased binding to JN.1 Spike-expressing cells (Supplementary Fig. S11A). Together, these results indicate that fusion of ACE2 to P23 does not impair Fc-mediated ADCP activity and Bs-ACE2-P23 retains the ADCP activity associated with both the ACE2 and P23 binding domains.

The Bs-ACE2-P23 bispecific protein exhibits broad-spectrum neutralizing activity against multiple SARS-CoV-2 VOCs

To further assess the breadth of neutralization mediated by Bs-ACE2-P23, we evaluated its activity against a panel of Omicron sub-variant pseudoviruses, including B.1.1.529, BA.2, XBB, XBB.1.5, BA.2.86, JN.1 and XDV (Supplementary Table S2). Bs-ACE2-P23 potently neutralized all of these pseudoviruses, with IC₅₀ values ranging from 9.4 to 255.7 pM (Fig. 6A), demonstrating the most potent activity against the BA.2.86 pseudovirus (IC₅₀ = 9.4 pM).

Figure 6.

Broad-spectrum neutralization of Bs-ACE2-P23 against SARS-CoV-2 variants. (A) Neutralizing activity of Bs-ACE2-P23 against a panel of pseudoviruses of SARS-CoV-2 variants, including omicron subvariants XBB, BA.2 (B.1.1.529.2), B.1.1.529, XBB.1.5, BA.2.86, JN.1, and XDV. (B) Neutralizing activity of Bs-ACE2-P23, hACE2-hFc, P23 hIgG1, and a 1:1 (molar ratio) combination of hACE2-hFc with P23 hIgG1 against authentic SARS-CoV-2 Delta (B.1.617.2) and JN.1 variants. n.m., not measurable.

Beyond evaluating the neutralizing activity with pseudoviruses, we conducted neutralization experiments using authentic SARS-CoV-2 Delta and JN.1 variants, which were isolated from respiratory samples of COVID-19 patients [37]. Briefly, 100 TCID₅₀ authentic viruses of SARS-CoV-2 Delta and JN.1 variants were incubated with serially diluted Bs-ACE2-P23, P23 alone, hACE2-hFc alone, or a combination of P23 and hACE2-hFc, followed by infection of Vero cells and CPE observation. Bs-ACE2-P23 demonstrated superior neutralizing activity against the authentic Delta variant (IC₅₀ = 17.4 nM), with a 4.9-fold increase, and against the JN.1 variant (IC₅₀ = 18.6 nM), with a 2.4-fold increase, compared to the combination of P23 and hACE2-hFc (Fig. 6B). Collectively, Bs-ACE2-P23 exhibited potent and broad-spectrum neutralizing activity against multiple of SARS-CoV-2 VOCs.

The Bs-ACE2-P23 architecture exerts neutralization enhancement for multiple anti-spike mAbs with distinct spike binding epitope

To investigate whether the Bs-ACE2-P23 architecture can enhance neutralizing activities of mAbs targeting distinct epitopes on Spike proteins, we selected antibodies binding to distinct S1 subunit epitopes (REGN-10987, S309, 35B5) and S2 subunit epitopes (S2P6, Fp.006, CV3-25) to produce IgG-like bispecific fusion proteins in the Bs-ACE2-P23 architecture for comparison of binding and neutralizing activity. For the WT Spike of SARS-CoV-2 and the Spike of SARS-CoV-2 JN.1, the fusion proteins all demonstrated improved binding affinity compared to hACE2-hFc, albeit with varying degrees of improvement (Fig. 7A–B, Supplementary Table S1).

Figure 7.

Neutralizing activities of ACE2-anti-spike bispecific IgG-like proteins in Bs-ACE2-P23 architecture against SARS-CoV-2 JN.1. (A–B) Binding affinity analysis of antibodies (in hIgG1 format) and bispecific proteins binding to the spike ECD of SARS-CoV-2 WT (A) or the JN.1 variant (B). (C–D) Neutralizing activities of ACE2-S1 IgG-like fusion proteins incorporating anti-S1 antibodies (REGN-10987, S309, and 35B5) (C) or ACE2-S2 IgG-like fusion proteins incorporating anti-S2 antibodies (Fp.006, S2P6, or CV3-25) (D) against the JN.1 pseudovirus. Data are shown as mean ± SD, n = 3. (E) IC₅₀ values corresponding to panels (C–D). IC₅₀ values for Bs-ACE2-Fp.006, FP.006, Bs-ACE2-S2P6, S2P6, Bs-ACE2-CV3-25, CV3-25 were calculated based on the results in panel (D), whereas IC₅₀ values for the other indicated fusion proteins and antibodies were calculated based on the results in panel (C). n.m., not measurable.

We then compared the neutralizing activity of these proteins against VSV-based SARS-CoV-2 JN.1 pseudovirus. All of these fusion proteins were more potent than hACE2-hFc; however, all except Bs-ACE2-Fp.006 showed less potent neutralizing activity than Bs-ACE2-P23, which exhibited an IC₅₀ value of 28.9 pM, representing a 64.2-fold compared to hACE2-hFc (Fig. 7C–E). As the epitopes recognized by P23 and Fp.006 were adjacent to the FP, indicating that hACE2 ECD fused with the antibody binds to the epitope adjacent to FP in the Bs-ACE2-P23 architecture, resulting in potent neutralizing fusion proteins against the SARS-CoV-2 JN.1 variant.

Discussion

In this study, we initially identified and characterized an anti-S2 nAb, P23, derived from a non-immune phage display human scFv library. P23 exhibited cross-neutralizing activity against the SARS-CoV-2 Wuhan-Hu-1 strain and Pangolin-CoV. P23 binds to an FP-adjacent epitope distinct from those targeted by six previously reported anti-S2 antibodies (S2P6, Fp.006, CV3-25, CC40.8, 76E1, and 54043-5). However, P23 failed to neutralize SARS-CoV-2 variants containing the D614G mutation. To overcome this, we engineered an IgG-like bispecific fusion protein (Bs-ACE2-P23) to simultaneously target two steps of viral infection: the interaction between the RBD and ACE2, and the virus-cell membrane fusion mediated by the S2 subunit. The Bs-ACE2-P23 fusion protein exhibited superior neutralizing activity against multiple D614G-containing SARS-CoV-2 variants as compared to hACE2-hFc alone and to the combination of hACE2-hFc with P23.

Emerging SARS-CoV-2 variants often carry mutations in the S1 subunit [38]. The S2 subunit, in contrast, is more evolutionarily conserved across coronaviruses, particularly within the β-Coronavirus genus (including SARS-CoV-2 and Pangolin-CoV), making it an attractive target for the development of broadly nAbs [14, 15, 17, 18, 39–42]. Competition analysis suggests that P23 recognizes a previously uncharacterized S2 epitope located proximal to the FP region. The identification of P23 expands the current landscape of S2-directed antibodies and highlights the diversity of conserved epitopes within the S2 subunit that can be exploited for coronavirus antibody development. Although HDX-MS identified a candidate binding region for P23, the precise binding epitope remains to be experimentally validated. Structural studies will be required to determine the detailed interaction interface between P23 and Spike and to further understand how this epitope contributes to broad recognition and neutralization.

The distinct antigen contexts used in different binding assays may also influence the apparent binding properties of S2-directed antibodies. For example, NS2 exhibited markedly different apparent binding profiles in ELISA and SPR. Because the ELISA assay used the trimeric Spike ectodomain, whereas SPR measured binding to the S2 subunit alone, the resulting EC₅₀ and K_D values reflect antibody recognition under distinct antigen contexts and assay formats and are therefore not directly comparable. The stronger apparent binding to the trimeric Spike ectodomain observed by ELISA likely reflects differences in antigen presentation, epitope accessibility, and avidity effects associated with the trimeric Spike ectodomain.

The limited activity of P23 against D614G-containing SARS-CoV-2 variants highlights challenges in S2-targeting antibody development. P23 failed to neutralize SARS-CoV-2 variants carrying the D614G mutation. Previous studies have shown that Spike mutations can impair the activity of anti-S2 antibodies either by directly disrupting antibody binding or by altering Spike conformation and epitope accessibility [43–45]. Consistent with this possibility, P23 showed reduced binding to the D614G Spike ECD. However, because detectable binding was still observed by both flow cytometry and ELISA, reduced binding alone is unlikely to fully explain the loss of neutralizing activity. The D614G substitution may therefore alter Spike conformational dynamics during membrane fusion, thereby impairing P23-mediated neutralization. Further studies are required to elucidate the underlying mechanism.

By simultaneously targeting RBD–ACE2 engagement and S2-mediated membrane fusion, Bs-ACE2-P23 achieved potent neutralizing activity against SARS-CoV-2 variants. The precise mechanism underlying the observed synergy between ACE2 and P23 remains unclear. It is possible that ACE2 binding to Spike induces conformational rearrangements that modulate the accessibility of FP-proximal epitopes, thereby enhancing the activity of antibodies such as P23 or Fp.006. Further structural and mechanistic studies will be required to define the molecular basis of this synergy.

Dual-targeting strategies have been successfully employed in recent studies developing bispecific biologics capable of potent neutralization of SARS-CoV-2 and its emerging variants [46, 47]. For example, Yuan et al. (2022) reported a bispecific antibody composed of the scFv of an anti-RBD nAb (35B5) and that of an anti-S2 antibody (47D10), and demonstrated that it exhibited potent neutralizing activity against multiple variants, including Delta, Omicron BA.1, and Omicron BA.2 [46]. Notably, because ACE2 is the conserved receptor for SARS-CoV-2, the RBD-ACE2 interaction is unlikely to be abolished by emerging Spike mutations. By incorporating the ACE2 ECD as a binding arm to block the RBD-ACE2 interaction, our Bs-ACE2-P23 bispecific fusion protein decreases the likelihood of mutational escape.

Miao et al. (2020) reported that an ACE2-based bispecific fusion protein generated by fusing ACE2 via a flexible linker to the C-terminus of an anti-S1-NTD antibody exhibited enhanced binding avidity and synergistic neutralization activity. However, fusion of ACE2 to the N-terminus of the same antibody abolished its binding to the S1 subunit [48]. In contrast, in our study, fusion of ACE2 to either the N-terminus or C-terminus of the P23 HC retained P23 binding to Spike. This difference may be attributed to the distinct epitope locations of the parental antibodies. S1- and S2-targeting antibodies may require different relative positioning of the ACE2 domain and the antibody binding site to facilitate optimal engagement with both RBD and the corresponding Spike epitopes, highlighting the importance of epitope selection and fusion orientation in the design of ACE2-based bispecific fusion proteins.

Notably, we used hACE2 ECD containing H374N/H378N mutations to deactivate the Zinc metalloenzyme activity as the binding arm. Yan et al. (2020) demonstrated that H374 and H378 are not involved in ACE2-RBD interaction [49]; thus, these mutations do not interfere with the binding affinity of ACE2 for RBD. Additionally, introducing additional affinity-enhancing mutations—including S19W, T27W, and N330Y—into hACE2 ECD has been reported to enhance ACE2 binding and neutralization [50]. We speculate that introducing such affinity-enhancing mutations into the Bs-ACE2-P23 architecture would enhance its neutralizing activity against SARS-CoV-2 variants. Finally, Liu et al. (2022) reported a hexavalent ACE2-hIgM-Fc fusion protein that potently neutralizes SARS-CoV-2 VOCs [31]. This finding suggests that generating an IgM-based hexavalent form of the P23-ACE2 ECD fusion protein, based on the Bs-ACE2-P23 architecture, may achieve superior neutralizing activity compared with the current bivalent IgG format.

In summary, our study identified a new anti-S2 nAb, P23, and engineered a novel IgG-like bispecific fusion architecture based on it. This architecture, which combines hACE2-ECD with antibodies binding adjacent to the FP region in the S2 subunit, represents a potential strategy to overcome the reduced neutralizing activity of S2 mAbs caused by mutations in the Spike protein. In addition to contributing to the development of broad-spectrum biologics against SARS-CoV-2 and its VOCs, this study also provides a framework for the rapid development of antiviral therapeutics capable of effectively neutralizing other coronaviruses.

Author contributions

Zhizhong Wei (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Project administration [lead], Validation [equal], Visualization [equal], Writing - original draft [equal], Writing—review & editing [supporting]), Ximing Liu (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Project administration [lead], Validation [supporting], Visualization [supporting]), Kailun Wang (Data curation [supporting], Formal analysis [supporting], Investigation [supporting], Validation [supporting], Visualization [supporting]), Lan Chen (Investigation [supporting]), Xia Xiao (Investigation [supporting]), Yinyan Sun (Data curation [supporting], Investigation [supporting], Methodology [supporting]), Wei Wang (Investigation [supporting], Methodology [supporting]), Yizhe Chen (Investigation [supporting]), Linqiang Fang (Investigation [supporting]), Yonghe Qi (Investigation [supporting], Methodology [supporting]), Fang Yang (Investigation [supporting]), Xinyan Hao (Investigation [supporting]), Ruihua Fan (Investigation [supporting]), Sanduo Zheng (Resources [supporting], Supervision [supporting]), Lili Ren (Resources [supporting]), Wenhui Li (Conceptualization [equal], Resources [equal], Supervision [equal]), Xinxin Tian (Conceptualization [supporting], Formal analysis [equal], Investigation [supporting], Methodology [supporting], Resources [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [lead]), and Jianhua Sui (Conceptualization [equal], Funding acquisition [lead], Methodology [equal], Resources [equal], Supervision [equal], Validation [equal], Writing—original draft [equal], Writing—review & editing [equal])

Conflicts of interest

None declared.

Funding

This work was supported by Beijing Municipal Science and Technology Commission (Beijing Key Laboratory of Pathogen Invasion and Immune Defense, Z171100002217064 to J. Sui).

Data availability

The data that support the findings of this study are available within the article and its supplemental information. The raw and analyzed datasets generated during the study are available from the corresponding author upon reasonable request.

Animal research statement

All animal experiments were conducted following the National Guidelines for Housing and Care of Laboratory Animals in China and performed under the approved Institutional Animal Care and Use Committee protocols at the NIBS (NIBS2025M0003). Mice were purchased from Charles River, China.

Ethics and consent statement

Not applicable.