Computational refinement and multivalent engineering of complementarity-determining region-grafted nanobodies on a humanized scaffold for retaining antiviral efficacy

Abstract

Recently, nanobody-based therapeutics have emerged as a highly effective strategy for COVID-19 treatment. However, camelid-derived nanobodies often require humanization engineering to reduce immunogenicity in clinical applications while simultaneously preserving their target-binding affinities. Here, we employed a computational and engineering approach to optimize the binding affinities of complementarity-determining region (CDR)-grafted humanized variants of the camelid-derived nanobody Nb2–67, which exhibits potent SARS-CoV-2 neutralization. By grafting the three CDR loops of Nb2–67 onto the humanized scaffold of the approved therapeutic nanobody Caplacizumab and refining the target-binding interface, we generated five nanobody variants with improved computational humanness scores. Three of these variants (Nb491, Nb273, and Nb1052) retained neutralizing activity. To further enhance their potency, we fused these variants to a self-assembling scaffold, generating three multivalent constructs with higher humanness scores. Pseudovirus assays showed that all the trivalent nanobodies exhibited picomolar neutralizing potency comparable to the original trivalent Nb2–67. Our study presents a novel computational and multivalent engineering strategy that effectively restores the antiviral efficacy of humanized CDR-grafted nanobody variants, offering a valuable approach for developing nanobody-based therapeutics against COVID-19 and other diseases.

Introduction

Antibody-based therapies have emerged as a cornerstone for combating viral infections, particularly during the COVID-19 pandemic, by blocking the interaction between the viral Spike (S) protein and the human angiotensin-converting enzyme 2 (ACE2) receptor [1–4]. Among these therapies, single-domain heavy chain antibodies (also known as nanobodies) [5] have garnered significant interest due to their compact size (~15 kDa), exceptional stability, and superior tissue penetration compared with conventional antibodies [6]. For example, camelid-derived nanobodies, such as Nb2–67 and Nb19, have demonstrated ultrapotent neutralization of SARS-CoV-2 variants through high-affinity binding to the receptor-binding domain (RBD) of the S protein [7–9]. Similarly, multivalent nanobodies isolated from immunized alpacas have shown cross-reactive neutralization against diverse variants of concern, including Omicron subvariants (BA.1, BA.2, and BA.4/5), SARS-CoV-1, and sarbecoviruses [10]. Despite their therapeutic potential, the clinical translation of camelid nanobodies is hindered by a critical limitation: non-human amino acid residues in their framework regions (FRs) can trigger immunogenic responses in humans [11]. This highlights the need for engineering strategies that balance humanization with the preservation of functional efficacy.

To address immunogenicity, complementarity-determining region (CDR) grafting—grafting CDR loops from camelid nanobodies onto humanized FRs—has been widely explored [12]. However, structural incompatibilities between non-human CDRs and human FRs often destabilize the antigen-binding interface, leading to insoluble expression or loss of function [13–16]. For example, CDR3-grafted nanobodies with frameworks of neural cell adhesion molecule and FK506-binding protein failed to express in soluble forms, while other grafted variants exhibited dramatically reduced binding affinity (e.g. >100-fold weaker K_D for hen egg white lysozyme) [14]. Similarly, grafted nanobodies targeting Botulinum neurotoxin A and fluorescein showed only micromolar affinity prior to labor-intensive affinity maturation [16]. These cases highlight the inherent challenge of balancing humanization and functionality through conventional CDR grafting.

To overcome these limitations, traditional optimization approaches have relied on experimental methods such as yeast-display screening, random mutagenesis libraries, and site-directed mutagenesis [15–19]. While these strategies have achieved notable successes, e.g. a 15-fold potency increase in a SARS-CoV-2 neutralizing nanobody via yeast-display screening [17] or enhanced PD-1/EGFR-targeting nanobodies through site-directed mutagenesis [18, 19], they remain time-consuming, labor-intensive, and prone to reintroducing immunogenic residues [20, 21]. In this context, computational strategies offer an ideal alternative by enabling systematic optimization of FR–CDR compatibility. Molecular docking and dynamics simulations have been successfully applied to optimize CDR conformations, stabilize FR interactions, and predict immunogenic hotspots without iterative experimental trials [18, 22]. For example, Heidari et al. [18] validated FR–CDR synergies in anti-CD20 nanobodies using molecular docking, while Devasani et al. [22] restored the binding affinity of a PD-1-targeting nanobody by simulating mutation-induced conformational changes. These studies highlight the potential of computational tools to rationalize nanobody optimization. However, these methods primarily refine existing experimentally generated mutants and lack the ability to design completely novel nanobody grafts. A promising solution lies in computational systems that directly design functional nanobody grafts, using predictive models to assess structural fits and binding strength. Although computational methods have been widely applied in protein and peptide design [23, 24], their application to CDR grafting for retaining nanobody function remains underexplored.

Beyond monomeric optimization, multivalent nanobody engineering has emerged as a potent strategy to amplify neutralizing efficacy, particularly against viruses with homomultimeric surface proteins like SARS-CoV-2 [25–29]. Koenig et al. [26] showed that constructing a trimer of a monovalent anti-SARS-CoV-2 nanobody increased its neutralizing activity by >100-fold. Huo et al. [27] reported four nanobodies (C5, H3, C1, and F2) engineered as homotrimers with picomolar affinity for the RBD of the SARS-CoV-2 S protein. Similarly, Qin et al. found that trimerization of a nanobody against the SARS-CoV-2 Omicron BA.1 variant greatly enhanced its neutralizing power. Their trimerization also enabled this nanobody to neutralize the Omicron XBB subvariant (XBB) variant, which the monomeric form could not achieve, highlighting the potential of multivalency to overcome viral evolution and resistance [29]. These studies collectively underscore the transformative potential of multivalent nanobody engineering in combating rapidly evolving pathogens.

In this study, we present a computational modeling and multivalent engineering pipeline to retain the anti-SARS-CoV-2 activity of humanized CDR-grafted nanobodies. Using the camelid-derived Nb2–67 as a CDR donor, we first grafted its CDR loops onto a humanized FR scaffold and then refined the nanobody–RBD interface through iterative interface design and molecular docking. To further enhance potency, we engineered trivalent forms of the optimized nanobodies. The resulting candidates achieved high computational humanness scores while retaining picomolar neutralizing efficacy against pseudotyped SARS-CoV-2 Omicron variant, comparable to the original trivalent camelid nanobody (Tr67). Thus, this work establishes a framework for deimmunizing nanobodies without compromising antiviral function, bridging a critical gap in therapeutic antibody development.

Results

Complementarity-determining region grafting and computational screening

To construct highly humanized nanobodies targeting SARS-CoV-2 Omicron, we selected Caplacizumab (Protein Data Bank, PDB  code 7EOW) as the grafting framework. Caplacizumab, an EU-approved nanobody drug for treating acquired Thrombotic Thrombocytopenic Purpura, was chosen due to its high stability and low immunogenicity (3% anti-drug antibody incidence in clinical trials) [30]. Importantly, experimental studies have shown that Caplacizumab does not bind to the SARS-CoV-2 Omicron BA.1 S protein [31], making it an ideal candidate for CDR grafting to develop Omicron-specific nanobodies.

We began by comparing the FRs of Nb2–67 and Caplacizumab using IMGT numbering [32] (Fig. 1). Sequence alignment revealed 16 amino acid differences distributed as follows: 2 in FR1, 5 in FR2, and 9 in FR3. The high divergence in FR3 (9/16 differences) suggested potential structural and functional impacts on the grafted nanobody. To quantify the humanization level, we calculated the humanness scores using the Hu-mAb algorithm [33] (Table 1). Caplacizumab showed higher humanness score than Nb2–67, with score of 0.935 versus 0.875 for Nb2–67, indicating that Nb2–67 requires further optimization to achieve sufficient humanization.

Figure 1.

Sequence alignment of the frameworks of Nb2–67 and Caplacizumab.

Table 1.

Humanness scores of nanobodies

Nanobody	Humanness score
Nb2–67	0.875
Caplacizumab	0.945
Cab67	0.915
Nb491	0.935
Nb273	0.930
Nb1052	0.935
Nb3154	0.900
Nb329	0.885
Tr67	0.875
Tr491	0.930
Tr273	0.935
Tr1052	0.935

Next, we performed CDR grafting to transfer the antigen-binding regions of Nb2–67 onto the Caplacizumab framework. Using IMGT numbering, we defined the CDR loops of Caplacizumab as follows: CDR1 (residues 26–33), CDR2 (residues 51–58), and CDR3 (residues 97–117) (Fig. 2A). The CDR sequences of Nb2–67 (Fig. 2B) were grafted onto the corresponding regions of Caplacizumab, resulting in the new nanobody Cab67 (Fig. 2C, left). We then generated a structural model of Cab67 using homology modeling (Fig. 2C, right).

Figure 2.

The grafting of the CDR loops into a humanized nanobody scaffold. (A) Sequence of the Caplacizumab nanobody. (B) CDR loops of Nb2–67. (C) Sequence and structure of the grafted nanobody Cab67.

To evaluate whether Cab67 could bind to the Omicron S protein, we performed molecular docking using HDOCK [34], due to its efficient sampling algorithm and accurate scoring function. We docked Cab67 to the up-state RBD of the Omicron S protein (PDB code 7TGW), generating the top 100 binding conformations with the highest docking scores (Fig. 3A). Since the up-state RBD on the S protein is known to be linked with the N-terminal domain (NTD) and can sterically collide with the adjacent RBDs, we hypothesized that Cab67 binding to some part of the RBD might face similar spatial constraints. To systematically assess this, we defined collision sites as residues on the RBD that are either directly linked to the NTD or within 4 Å of other chains [35]. The remaining epitopes were classified as target binding sites [31]. Using these criteria, we screened the 100 docking conformations and selected those where: the CDRs formed the binding interface, and the distance to collision sites was >4 Å (Supplementary Table S1). The top two docking complexes (Cab67_model1 and Cab67_model2) exhibited distances to collision sites of <4 Å, suggesting potential steric clashes with the S protein. Structural alignment of Cab67_model1 and Cab67_model2 with the S protein (PDB code 7TGW) confirmed these clashes (Supplementary Fig. S1), particularly in regions where the NTD and the adjacent RBDs interact. These results indicate that Cab67 does not bind to the Omicron S protein under the tested conditions, likely due to steric hindrance or conformational incompatibility with the S protein.

Figure 3.

Computational design of nanobodies targeting SARS-CoV-2 omicron S protein. (A) Docking of the Cab67 nanobody to RBD. (B) Interface design of the Cab67–RBD complexes with the largest delta solvent accessible surface areas (∆SASA). (C) Binding interface score analysis of nanobody–RBD candidates. (D) Functional validation.

Optimization of the grafted complementarity-determining region sequences

To optimize the directly grafted sequences in Cab67, we first identified conformations matching the initial criteria (∆S_CDR > 1 and distance from collision sites > 4 Å) (see Materials and Methods), then selected the Cab67–RBD complex showing maximal change in solvent-accessible surface area (∆S_CDR). This optimal candidate, termed Cab67_model39, served as the starting point for interface redesign (Fig. 3B).

We applied Rosetta interface design [36] to optimize interface residues of Cab67_model39, including key interface residues such as CDR1, CDR2, and CDR3. A total of 10 000 mutated sequences were generated, and the top 20 sequences were selected based on their Rosetta energy scores. These sequences were then subjected to structural modeling (see Materials and Methods) and HDOCK docking, resulting in the top 100 docking conformations with the most negative docking scores for each sequence. The HDOCK score, which reflects the binding possibility of the nanobody to RBD, was used to rank the conformations. A more negative HDOCK score indicates a higher likelihood of binding. We assessed whether the highest-scoring conformations met the initial criteria. For conformations meeting these criteria, we used Rosetta Interface Analyzer [37] (Fig. 3C) to further filter high-quality complexes using two important indicators: (i) binding energy efficiency () and (ii) interface packing quality (packstat >0.65). After several iterations screening, we identified 16 nanobody–RBD complexes with an optimized interface (Supplementary Table S2). To better visualize the effects of interface design, we displayed the top 1 of docking conformations of all 16 nanobody–RBD complexes in Supplementary Fig. S2. With the exception of a few nanobodies (Nb5001, Nb522, Nb624, and Nb2154), most optimized nanobodies adopt binding poses similar to that of the original Cab67_model39. Importantly, interface design led to broader interface contacts involving both the antigen and the nanobody. Moreover, many mutated residues were directly involved in the antibody–antigen interface, suggesting that the introduced mutations may have enhanced the binding interactions.

To validate the binding specificity of the 16 selected nanobodies, we performed Rosetta global docking [38] using the HDOCK docking results as the initial conformation. Each nanobody was docked to RBD in five parallel sets, with 400 independent runs per set, resulting in a total of 2000 docking simulations (Fig. 4). The docking results were analyzed by comparing the root-mean-square deviation (RMSD) and interface score (I_sc) of each conformation relative to the initial one. Two distinct binding patterns were emerged from the docking analysis (Fig. 4A and B). The irregular pattern shown in Fig. 4A indicates that the nanobodies bind to multiple epitopes with similar probabilities (similar I_sc), reflecting non-specific binding across various sites. In contrast, the more convergent pattern in Fig. 4B shows that several low-energy conformations exhibit RMSD values close to zero, suggesting stable and conformation-specific interactions. Among the 16 screened nanobodies, eight exhibited the favorable funnel conformation. Five candidates (Nb491, Nb1052, Nb273, Nb3154, and Nb329) showed superior interface stability (I_sc < −40). These findings confirm their high binding specificity and the structural stability of the complexes.

Figure 4.

Docking interface score landscapes of 16 nanobody candidates. (A) Irregular distributions of docking interface scores, showing multiple low-score complexes with diverse RMSD values. (B) More convergent distributions of docking interface scores, where several low-score complexes have RMSD values close to zero, indicating better agreement with the reference conformation. Data points are represented with a continuous scale of interface scores, where larger points indicate more favorable scores.

Finally, we calculated the humanness scores of Cab67 and the five optimized nanobodies (Table 1). The humanness score of Cab67 was 0.915, which is less than that of Caplacizumab (0.935), indicating that the inclusion of non-human CDR sequences reduced its humanization potential. However, the interface-optimized nanobodies showed significant improvements. The humanness scores were 0.935 (Nb491), 0.935 (Nb1052), 0.930 (Nb273), 0.900 (Nb3154), and 0.885 (Nb329). Among them, four nanobodies (Nb491, Nb1052, Nb273, and Nb3154) outperformed the originally grafted Cab67 (0.915), likely making them more suitable for clinical applications. Although the score of Nb329 was lower than that of Cab67, it exhibited interface stability and its score was higher than that of Nb2–67. Thus, we selected these five optimized nanobodies as the candidates targeting SARS-CoV-2 Omicron for further experimental verification (Fig. 3D).

Experimental validation of complementarity-determining region-grafted nanobodies

To evaluate the ability of the five optimized nanobody sequences to bind to the Omicron S protein, we conducted a series of functional assays. First, the five nanobodies were expressed in Escherichia coli Rosetta (DE3) cells. All nanobodies, except for Nb329, were successfully expressed in soluble form (Supplementary Fig. S3). Then, the four soluble nanobodies (Nb491, Nb273, Nb1052, and Nb3154) were purified using Ni-NTA affinity chromatography under native conditions.

The binding affinity of the nanobodies to the Omicron RBD was measured using Bio-Layer Interferometry (BLI). As shown in Fig. 5A–C and Supplementary Fig. S4, the best-fit equilibrium dissociation constants (K_D) for Nb491, Nb273, Nb1052, and Nb3154 were 259, 503, 582, and 554 nM, respectively. As a lower K_D value indicates stronger binding affinity, these results confirmed that all four nanobodies bind to the Omicron RBD, with Nb491 exhibiting the highest binding affinity.

Figure 5.

Experimental validation of designed nanobodies targeting SARS-CoV-2 Omicron. (A–C) BLI measurements of binding kinetics between three designed nanobodies and Omicron RBD. (D) Pseudovirus neutralization assay results demonstrating neutralization efficacy against SARS-CoV-2 Omicron BA.1 variant.

The neutralizing activity of the nanobodies against SARS-CoV-2 pseudovirus (Omicron BA.1) was assessed using a cell-based assay. The half-maximal inhibitory concentration (IC₅₀) values for Nb491, Nb273, and Nb1052 were 2.80, 5.96, and 10.84 μM (Fig. 5D), respectively, while Nb3154 failed to achieve any detectable neutralization (Supplementary Fig. S5), despite its measurable binding to Omicron RBD (K_D = 554 nM) in the BLI assay. This suggests that Nb3154 either binds to a non-neutralizing epitope or has insufficient affinity to effectively block the interaction between S and ACE2. The IC₅₀ value is a key indicator of neutralizing potency, with lower values indicating stronger activity. These results demonstrate that Nb491, Nb273, and Nb1052 can neutralize Omicron BA.1, with Nb491 showing the strongest neutralizing activity.

The experimental results indicate that the interface-optimized nanobodies retain their binding and neutralizing abilities, although their potency is lower than that of the original camelid nanobody. The strong correlation between binding affinity (K_D) and neutralizing activity (IC₅₀) suggests that the binding strength to the RBD is a critical factor for neutralization. However, the lack of neutralizing activity in Nb3154, despite its moderate binding affinity, might be due to its binding to a non-neutralizing epitope or conformational changes upon the binding. In the case of Nb329, its CDR loops may not be compatible with the FR, so it could not fold properly for a soluble expression.

Although the redesigned monomeric nanobodies retained a certain degree of neutralizing activity, their potency was markedly reduced compared with the original Nb2–67. To explore the structural basis for this difference, we performed a basic analysis of the nanobody–RBD complexes (Supplementary Fig. S6). As shown in Supplementary Fig. S6A, Nb2–67 binds directly to the top of the RBD, overlapping with the ACE2-binding site, which likely explains its high neutralization potency. In contrast, structural docking models of the redesigned nanobodies (Nb491, Nb273, and Nb1052) revealed a shift in binding orientation—from the ACE2-binding region to the lateral surface of the RBD—leading to engagement with non-neutralizing or weakly neutralizing epitopes. This epitope shift also resulted in a distinct set of interface residues on the RBD, as quantified in Supplementary Fig. S6B.

Such a change in binding mode may be attributed to the altered paratope geometry introduced by CDR grafting and interface redesign. While these variants retained stable docking poses and moderate affinity, their reduced capacity to directly compete with ACE2 likely contributed to the diminished neutralization efficacy.

Multivalent engineering of the complementarity-determining region-grafted nanobodies

To further enhance the functional activity of the three nanobodies, we employed a trimerization strategy [29] to increase their binding affinity and neutralizing activity for the S protein. Specifically, we linked the N-terminus of each nanobody to a scaffold protein via a glycine-serine (GS) linker (Supplementary Table S3), enabling self-assembly into trivalent nanobodies in solution (Fig. 6A). With multiple binding sites, the trivalent nanobodies can simultaneously engage three RBDs on the S protein (Fig. 6B). This multivalent interaction significantly strengthens the binding and stabilizes the complex (Supplementary Fig. S7).

Figure 6.

Construction and characterization of the multivalent nanobody constructs. (A) Schematic representation of nanobody self-assembly into the trivalent forms by N-terminal GS-linker conjugation to the scaffold protein. (B) Structural model of a trivalent nanobody binding to the SARS-CoV-2 Omicron BA.1 S protein. (C–E) SEC profiles for the purification efficiency of the trivalent nanobody variants. (F) Native-PAGE analysis of SEC-purified trimer fractions.

Using this approach, we constructed three trivalent nanobodies. All three nanobodies were expressed in soluble form in E. coli Rosetta (DE3). After purification with Ni-NTA affinity chromatography, their trimerization efficiency reached 90%, as confirmed by size exclusion chromatography (SEC) and native polyacrylamide gel electrophoresis (native-PAGE) analysis (Fig. 6C–F). These results indicate that the three nanobodies exist in the expected trivalent form and have a strong self-assembly capability.

We then evaluated the binding and neutralizing activities of these trivalent nanobodies against Omicron BA.1 (Fig. 7A–F). As shown in Fig. 7A–C, the K_D values for Tr491, Tr273, and Tr1052 were 22.7, 41.6, and 93.8 nM, respectively, showing more than a 10-fold improvement in affinity compared with their monomeric forms. Figure 7D–F shows that their IC₅₀ values were 2.69, 31.1, and 116 pM, respectively, significantly higher than those of the monomeric nanobodies. Tr491 exhibited the highest neutralizing capacity, with an IC₅₀ 106 times lower than its monomeric counterpart. These results demonstrated that the trimerization has an enhanced potency and thus greater potential to combat viral infections compared with its monovalent counterpart.

Figure 7.

Functional assays of the trivalent nanobodies. (A–C) BLI measurements of the binding kinetics of three trivalent nanobody variants against the Omicron RBD. (D–F) Pseudovirus neutralization assay results demonstrating neutralization efficacy against SARS-CoV-2 Omicron BA.1 variant.

Surprisingly, although Nb491 and Nb273 had much lower neutralizing activity than Nb2–67, Tr491 and Tr273 exhibited inhibitory activity comparable to or even stronger than that of Tr67. This suggests that trivalent modification greatly improves the neutralizing efficacy of the CDR-grafted nanobodies. Given their picomolar IC₅₀ values, Tr491 may have broader neutralizing activity, potentially making it more resilient to viral mutations. Importantly, the trivalent nanobodies retained humanness scores identical to their monomeric counterparts (Table 1), demonstrating that the engineered linker and self-assembling scaffold did not introduce immunogenic sequences or compromise humanization—crucially preserving their clinical safety profile.

Discussion

In this study, we developed a two-step computational and multivalent engineering strategy to address the challenge of preserving antiviral efficacy during CDR-grafted humanization engineering of camelid-derived nanobodies. By grafting the CDRs of the potent camelid-derived Nb2–67 onto the humanized Caplacizumab scaffold and computationally refining its target-binding interfaces, we generated four nanobody variants with significantly improved humanness scores (0.930–0.935 versus 0.875 for Nb2–67). Three of these variants (Nb491, Nb273, and Nb1052) retained submicromolar neutralizing activity against the Omicron BA.1 pseudovirus (IC₅₀ = 2.80–10.84 μM). To further enhance their potency, we fused these variants to a self-assembling scaffold, generating trivalent constructs with high humanness scores (0.930–0.935). Pseudovirus assays showed that all trivalent nanobodies achieved picomolar neutralization (IC₅₀ = 2.69–116 pM), comparable to the original trivalent Nb2–67 (IC₅₀ = 55 pM). Thus, our study successfully restored the antiviral efficacy of humanized CDR-grafted variants of the camelid-derived nanobody, highlighting the potential application of our method for the development of nanobody-based therapeutics against COVID-19 and other diseases.

Our computational strategy overcomes two major limitations of traditional CDR grafting. First, unlike framework-modification approaches such as FR2 hydrophobicity engineering by Vincke et al. [39] or complete framework replacement by Computational Universal Monoclonal Antibody design software (CUMAb) [40], we preserved 100% of the humanized Caplacizumab scaffold. Using HDOCK-guided collision mapping and Rosetta interface design, we resolved structural conflicts through CDR modification alone, achieving humanness scores of 0.930–0.935 without altering any FR residues (Supplementary Table S4–S5). In contrast, deep learning tools such as AbNatiV [41] risk destabilizing nanobodies by replacing structurally essential FR residues during humanization. Second, our trimerization design eliminates immunogenic linkers prevalent in Fc-fusion methods. By employing a self-assembling scaffold similar to the clinically validated GS-linker of Ozoralizumab [42], we achieved a 10–100-fold increase in affinity (K_D = 22.7–93.8 nM versus 259–582 nM for monomers) without introducing non-human sequences. Crucially, this dual innovation decoupled humanization from functional loss, overcoming the persistent trade-off observed in global optimization approaches. While the trimerization strategy itself has been used in prior studies, its application in restoring function to CDR-grafted humanized nanobodies represents a practical and effective solution to the potency-immunogenicity trade-off, and distinguishes our approach from more conventional multivalent designs.

The strategy presented holds promise for the development of next-generation nanotherapeutics against rapidly evolving pathogens. The computational pipeline could potentially be extended to humanize nanobodies against other targets, such as influenza HA, HIV Env, or RSV F proteins, while the trimerization method, which does not require a specific scaffold, may allow the creation of molecules that bind multiple viral sites simultaneously, further preventing immune escape. However, two limitations warrant attention: first, the monomeric variants exhibited reduced potency compared with Nb2–67 (IC₅₀ = 2.80 μM versus 0.492 nM), suggesting residual energetic penalties at optimized CDR–FR interfaces. Second, although the humanness scores predict reduced immunogenicity, in vivo validation remains essential. Future work should utilize other computational methods or deep learning methods to refine CDR geometries [43–47], and evaluate cross-neutralization against emerging variants such as XBB.15. In addition, comparative immunogenicity studies between traditional Fc-fusions and our connector-free trimers could quantify clinical safety benefits.

In summary, this study presents a computational and multivalent engineering approach to graft the CDR loops of the camelid-derived nanobody Nb2–67 onto the humanized scaffold of Caplacizumab. This strategy produced three nanobody variants (Nb491, Nb273, and Nb1052) with improved humanness scores and retained neutralizing activity. By fusing these variants to a self-assembling scaffold, we created trivalent constructs (Tr491, Tr273, and Tr1052) that achieved humanness scores >0.930 and exhibited picomolar neutralization potency against SARS-CoV-2 pseudoviruses, matching the efficacy of the original trivalent Nb2–67. These results address the critical challenge in antibody engineering—balancing deimmunization with functional preservation. Our work highlights the potential of combining computational design with multivalent engineering to develop potent therapeutics against rapidly evolving viruses, offering a promising pipeline for nanobody-based treatments targeting COVID-19 and other diseases.

Materials and methods

Molecular docking and binding interface screening

Rigid-body docking between Cab67 and the Omicron RBD was performed using the HDOCK software. In this docking protocol, both Cab67 and the RBD are treated as rigid structures; no conformational sampling of the nanobody CDR loops is performed. HDOCK explores different relative orientations and positions between the two molecules, and a total of 4397 distinct docking poses were generated. For Cab67-RBD, we selected the top 100 docking poses ranked by docking score for further evaluation. For other nanobody–RBDs, the docking pose corresponds to model_1, the top-ranked complex based on the HDOCK docking score. We selected model_1 as the most likely binding conformation and used it as the basis for further analysis and optimization. To assess binding interface engagement, we calculated the average change in solvent accessible surface area (∆S_CDR) of the nanobody CDRs before and after binding using the Biopython Bio.PDB.SASA module. A ∆S_CDR value of 0 indicated no contact between the CDRs and the RBD. Additionally, we calculated the minimum distance between the nanobody and predefined steric clash regions on the RBD. Poses with distances ≤4 Å were considered to involve steric clashes with the S trimer.

The initial filtering criteria were: ∆S_CDR > 1 and distance >4 Å. Among all docking poses that passed these thresholds, the pose with the highest ∆S_CDR and no steric clashes (distance >4 Å) was selected as the optimal structure for further analysis and interface optimization.

Structural prediction by iterative homology modeling and energy minimization

The structural prediction of nanobodies with low-homology templates was achieved through a perturbation-based protocol integrating homology modeling and energy minimization. The workflow was implemented by Python 3.8, which consisted of two cyclic phases: (i) iterative homology modeling using MODELLER (v10.6) [48] and (ii) structural refinement through Rosetta FastRelax [49]. The first template structure was Caplacizumab (PDB code 7EOW). Our perturbation approach introduced single-point mutations through sequential cycles: at each iteration, one target-specific amino acid substitution (e.g. A12P) was introduced to the template sequence, generating an intermediate sequence with >95% identity to the current template. Homology modeling was conducted using MODELLER. Then the resulting models were optimized by energy minimization using Rosetta FastRelax. This protocol performed five cycles of repack–relax steps retaining the lowest-energy conformation. The optimized structure served as the new template for subsequent mutation cycles. All intermediate structures and mutation records (including physicochemical classifications of substitutions) were archived in mutation_info files.

Computational design of nanobody–receptor-binding domain interface

Rosetta interface design was used to optimize the nanobody sequences at the complex interface. The RosettaScripts settings were as follows: the “ScoreFunction name” was set to “r15,” the “weights” set to “ref2015.” The TASKOPERATIONS field was set as: <ProteinInterfaceDesign name = “design” repack_chain1 = “1” repack_chain2 = “1” design_chain1 = “1” design_chain2 = “0” interface_distance_cutoff = “8”/>. Here, the nanobody chains were designated as the first chain, and only the nanobody chain was designed. For each interface design, 1000 or 10 000 constructs were generated and saved as PDB files. The I_sc parameter was used to sort the complexes for further analysis.

Evaluation of nanobody–receptor-binding domain interface

The quality of the interface in the complexes was assessed using the Rosetta Interface Analyzer. A blueprint file was prepared to specify computations, including the packstat calculation. Parameters such as “-compute_packstat,” “-tracer_data_print,” “-pack_input,” “-pack_separated,” and “-add_regular_scores_to_scorefile” were set to true. The output score file provided the necessary information to evaluate the interface quality, with thresholds set as: and packstat >0.65.

Self-consistency global docking

For the 16 nanobody–RBD complexes with favorable interface scores, we performed global docking using HDOCK-generated top 1 conformations as a starting point. Five parallel docking sets were run, with each set consisting of 400 docking simulations (totaling 2000 dockings). In the Rosetta docking blueprint, the -nstruct parameter was set to 400, and -partners A_B (where A corresponds to the nanobody and B to the RBD) was specified. Additional parameters included -dock_pert 3 8 and -spin -randomize1 -ex1 -ex2aro. The binding energy landscape of each complex was generated, and the I_sc and Irms parameters from the output score file were used to identify the best complexes for experimental validation.

Nanobody expression and purification

Monovalent nanobodies (Nb491, Nb1052, Nb273, Nb3154, and Nb329) were expressed in E. coli using codon-optimized genes encoding the designed protein sequences. The genes were cloned into the pET-26b (+) vector, which contained an N-terminal PelB sequence and a C-terminal 6 × His tag. The recombinant plasmids were transformed into E. coli Rosetta (DE3) cells. Cells were cultured at 37°C in Lysogeny broth (LB) until an OD₆₀₀ of ~0.6–0.8, then induced with 0.5 mM Isopropyl β-d-1-thiogalactopyranoside (IPTG) at 25°C for 20 h. Periplasmic extracts were obtained by osmotic shock. For protein purification, the cell supernatant was filtered and applied to gravity-flow columns with 2 mL of Ni-NTA resin, equilibrated in buffer A (1× PBS, 20 mM imidazole, pH 7.4). The column was washed with buffer A containing 80 mM imidazole, and proteins were eluted with buffer B (1× PBS, 300 mM imidazole, pH 7.4). Protein integrity was confirmed by SDS-PAGE. To remove the His tag, Tobacco Etch Virus (TEV) protease was used to cleave the tag from the purified proteins. Prior to cleavage, the protein solutions were desalinated using a HiTrap desalting column. The reaction mixture was incubated overnight at 4°C to ensure complete tag removal, and the tag-free proteins were further purified by Ni-NTA chromatography.

Multivalent nanobody construction and purification

To construct trivalent nanobodies, the N-terminal PelB sequence was replaced with a 6 × His tag and TEV cleavage sequence, followed by the appropriate nanobody-Foldon constructs, e.g. Nb273-(GGGGS)₅-foldon. The recombinant plasmids were transformed into E. coli Rosetta (DE3) cells. To generate trivalent nanobodies, the N-terminal PelB sequence was substituted with a 6 × His tag and a TEV cleavage site, followed by the appropriate nanobody-Foldon constructs, e.g. Nb273-(GGGGS)₅-foldon. The recombinant plasmids were introduced into E. coli Rosetta (DE3) cells, which were subsequently harvested, sonicated, and lysed on ice using lysis buffer (pH 7.4, 1 × PBS, 20 mM imidazole). The soluble fraction was obtained by centrifugation at 12 000 rpm for 30 min. The purification process followed the same procedure as that used for the monovalent nanobodies.

Size exclusion chromatography

SEC was performed using a Superdex 200 increase 10/300 GL column (GE Healthcare) connected to an ÄKTA avant system. The column was equilibrated with 1 × PBS (pH 7.4), and 500 μL of concentrated trivalent protein was loaded onto the column. Proteins were eluted at a flow rate of 0.5 mL/min, and the elution profile was monitored by UV absorbance at 280 nm. Oligomeric fractions were identified based on the integrated peak areas, and the trimer fraction was collected for further analysis.

Native polyacrylamide gel electrophoresis

Native-PAGE was performed to assess the conformational homogeneity of the trimeric proteins [50]. The purified trimeric protein fraction from SEC was diluted in native sample buffer and loaded onto a 15% native gel. Electrophoresis was carried out at 120 V and 4°C for ~1.5 h using native running buffer. The gel was stained to visualize protein bands.

Biolayer interferometry measurements

Binding kinetics of monovalent and trivalent nanobodies to RBD were measured using the Octet RED96e system (ForteBio) as describe by Qin et al. [29]. RBD was immobilized on Streptavidin (SA) biosensors by incubating the sensors in 100 μg/mL RBD solution. Nanobody binding was assessed by immersing the sensors in wells containing different concentrations of protein samples (1000–31.25 nM for Nb273 and 2000–62.5 nM for Nb1052). After an equilibration period in running buffer (PBS, 0.02% Tween-20, 0.1% Bovine serum albumin (BSA)), the sensors were transferred to fresh wells to determine the dissociation rate. Data were analyzed using Octet Data Analysis software (V10.0), fitting the binding curves to a 1:1 binding model. The K_D values were then determined from curves with R² > 0.95. All measurements were performed in biological triplicates (n = 3).

Pseudovirus neutralization assays

SARS-CoV-2 pseudoviruses carrying the firefly luciferase reporter gene were generated using the vesicular stomatitis virus (VSV) pseudotyped system, as described by Nie et al. [51]. Plasmids encoding Omicron BA.1 S proteins were transfected into HEK 293 T cells with VSVΔG pseudovirus particles. After 24 h, viral supernatants were harvested, filtered, and stored at −80°C. The pseudovirus titer was calculated by measuring the median tissue culture infectious dose (TCID₅₀) using a human ACE2-overexpressing cell line. For neutralization assays, pseudoviruses were diluted to ~10⁴ TCID₅₀/mL. Nanobody samples were incubated with the pseudoviruses for 1 h at 37°C, followed by a 24-h incubation at 37°C, 5% CO₂. Infection was quantified by measuring luminescence using a luminometer (PerkinElmer). Neutralization data were fitted using non-linear regression, and IC₅₀ values were calculated using a four-parameter regression equation in GraphPad Prism. All measurements were performed in biological triplicates (n = 3).

Key Points

• We developed a computational and multivalent engineering strategy to graft the complementarity-determining region (CDR) loops of the camelid-derived nanobody Nb2–67 onto the humanized scaffold of Caplacizumab. This approach generated three nanobody variants (Nb491, Nb273, and Nb1052) with improved humanness scores while retaining neutralizing activity against SARS-CoV-2.

• We engineered three trivalent constructs by fusing the optimized variants to a self-assembling scaffold (Tr491, Tr273, and Tr1052). These trivalent nanobodies achieved humanness scores >0.930 and demonstrated picomolar neutralization potency against SARS-CoV-2 pseudoviruses, comparable to the efficacy of the original trivalent Nb2–67.

• Our study addresses a critical challenge in antibody engineering—balancing deimmunization with functional preservation. Our work provides a useful approach for developing potent nanobody-based therapeutics against rapidly evolving viruses, offering promising applications for COVID-19 and other diseases.

Author contributions

Q.H. supervised the project. L.H. and Q.H. conceived the project; L.H. performed the computational study; L.H., Q.Q., X.Z., Y.C., T.T., X.H., T.Z., and Y.X. performed the experimental validation; L.H., Q.Q., and X.Z. performed experimental data analysis; L.H. and T.T. drafted the manuscript; L.H., T.T., and Q.H. revised the manuscript.

Conflict of interest: None declared.

Funding

This work was supported by the National Key Research and Development Program of China (2021YFA0910604), the National Natural Science Foundation of China (31971377), the Beijing Municipal Science & Technology Plan (Z221100007922016), and the Beijing Science and Technology New Star Program (Z211100002121005).

Data availability

The data supporting this study are available from the corresponding author upon reasonable request.

Consent for publication

All authors have consented to publication of the manuscript. No additional consent required.