Retaining binding characteristics and strong biophysics during VHH humanisation

Introduction to humanisation

Single-domain antibodies (VHHs), also known as nanobodies, are the variable domains of heavy-chain-only antibodies naturally found in camelids such as alpacas, llamas, and camels. Their small size (~15 kDa), high stability, ability to recognise cryptic epitopes inaccessible to conventional antibodies, ease of expression in low-cost production systems (e.g. yeast), tuneable blood clearance (ISOXTEND®) and effective tissue penetration have made them increasingly attractive therapeutic, diagnostic and imaging modalities. Several VHH-based therapeutics have now entered clinical development, with five achieving marketing approval, as of 2025, (caplacizumab, netakimab, envafolimab, ozoralizumab and ciltacabtagene autoeucel), demonstrating the clinical relevance of this antibody format (Muyldermans, 2013; Arbabi-Ghahroudi, 2017; Evers et al., 2025).

Because VHHs are derived from camelid immune repertoires, they contain framework sequences that differ from those found in human immunoglobulins. While VHHs generally exhibit relatively low immunogenicity compared with larger biologics, such as classical immunoglobulin G (IgG) antibodies, when administered therapeutically these non-human sequence elements may have an impact on the eliciting of anti-drug antibodies (ADAs). Although immunogenicity is influenced by multiple factors such as, route of administration, dosing regimen, aggregation, and patient-specific immune status (fully human Abs can elicit ADA responses (Getts et al., 2010)), reducing non-human sequence content remains one of the strategies for lowering immunogenic risk during antibody optimisation (Rossotti et al., 2021).

Figure 1A. Structural features of VHHs and a rational approach to humanisation. Comparison of a VHH with a conventional human IgG antibody, highlighting key structural features relevant to VHH humanisation.

Humanisation is the process of modifying a non-human antibody sequence to more closely resemble human germline antibodies while preserving the molecular features responsible for antigen recognition and favourable developability. Unlike conventional VH domains, VHHs have evolved to function independently of a paired light chain and consequently possess several hallmark structural features that are essential for maintaining stability and function (Vincke et al., 2009; Rossotti et al., 2021). One of the most characteristic differences lies within framework region 2 (FW2), where conserved camelid-specific residues (typically Phe42, Glu49, Arg50 and Gly52, IMGT numbering) replace the hydrophobic amino acids that form the VH–VL interface in conventional antibodies. These substitutions create a more hydrophilic surface, improving solubility in the absence of a light chain. Many VHHs also possess hallmark cysteine residues that form an additional disulphide bond, most commonly linking complementary-determining region (CDR) 1 or framework region 2 to CDR3. This extra disulphide bridge can increase conformational stability, particularly in VHHs containing long CDR3 loops, and helps maintain the architecture of the antigen-binding site (Nguyen et al., 2000; Nguyen, Desmyter & Muyldermans, 2001; Vincke et al., 2009). Together, these structural adaptations contribute to the exceptional stability and developability of VHHs while also necessitating a considered approach to humanisation, as many of the residues that distinguish camelid VHHs from human VH domains play critical structural roles (Figure 1).

Successful VHH humanisation therefore requires a balanced approach that minimises immunogenic risk while preserving the characteristics that make the parental molecule an attractive therapeutic candidate. Modern humanisation strategies typically combine sequence analysis, structural modelling, and targeted framework engineering to identify substitutions that maximise human sequence identity without compromising molecular performance.

Figure 1B. Structural features of VHHs and a rational approach to humanisation. Schematic overview of an iterative, structure-guided VHH humanisation workflow.

The objective of this project was to humanise an existing alpaca-derived VHH binder while maintaining its functional and developability characteristics. Specifically, the goal was to increase the human-likeness of the framework sequence without adversely affecting affinity, expression, stability, or other key biophysical properties. Achieving this balance would generate a molecule better suited for downstream therapeutic development while retaining the performance of the original lead.

Design – How did we decide which changes to make?

The objective of the humanisation campaign was to increase the human sequence identity of the parental alpaca-derived VHH while preserving its antigen-binding affinity and favourable biophysical properties. Rather than maximising sequence identity alone, the design strategy focused on introducing only those framework substitutions predicted to reduce immunogenic risk without disrupting structural integrity or molecular function (Rossotti et al., 2021).

Humanisation designs were generated using a combination of bioinformatic analyses and expert knowledge of VHH structure. The parental sequence was first compared against human germline repertoires to identify the most appropriate human VH framework to use as a reference for subsequent variant generation. Rather than selecting a single germline sequence directly, the reference framework was generated using statistical analysis of residue frequencies across the closest human germlines, combined with expert knowledge, published scientific literature on VHH humanisation strategies, and insights gained from previous internal humanisation campaigns.

Potential substitutions were assessed in the context of sequence conservation, structural location, solvent accessibility, and proximity to the antigen-binding loops. Particular attention was paid to residues known to influence VHH folding, stability, and developability. Hallmark camelid residues, including the characteristic framework region 2 (FW2) substitutions that replace the conventional VH–VL interface, were carefully evaluated before any changes were introduced. Likewise, cysteine residues involved in conserved or non-canonical disulphide bonds were retained where they were predicted to contribute to structural stability. Framework residues likely to influence complementarity-determining region (CDR) conformation or antigen recognition were also preserved to minimise the risk of affinity loss (Vincke et al., 2009; Fernández-Quintero et al., 2024).

While computational analyses provided a systematic framework for prioritising substitutions, final design decisions were informed by extensive experience in antibody engineering and VHH developability. Knowledge of residue-specific effects on expression, aggregation propensity, stability, and manufacturability allowed substitutions to be assessed beyond sequence similarity alone. This combination of computational prediction and empirical expertise enabled informed decisions where purely algorithmic approaches may have recommended changes carrying an unacceptable functional risk (Marchetti et al., 2024).

Table 1. Summary of 1^st iteration mutations. Shown are the mutation combinations and respective impact on affinity and biophysical characteristics. Highlighted in pink are the 2 clones with the higher OASis % (A17-013) and higher number of mutations (A17-020).

Multiple humanised variants were subsequently designed as part of an iterative optimisation strategy, representing different levels of framework humanisation. This iterative strategy enabled the contribution of individual or grouped substitutions to be experimentally evaluated while balancing increased human-likeness against retention of the parental molecule’s functional and biophysical characteristics.

Results 1 – first- generation humanisation

A panel of 23 humanised variants was designed to evaluate the impact of humanisation on the functional and biophysical properties of the parental alpaca-derived VHH. Each variant incorporated a subset of framework mutations, which represent increasing degrees of humanisation. Residues predicted to be important for structural integrity, solubility and antigen recognition were retained, with no changes introduced to the CDR regions.

Following sequence design, all variants (and the parental VHH) were recombinantly expressed and purified using a standard production workflow. The majority of the constructs expressed at levels comparable to the parental VHH, yielding highly monomeric purified protein, as determined by analytical SEC (>85% purity) which was suitable for downstream characterisation (Figure 2). This demonstrated that the introduced framework substitutions were generally well tolerated.

The panel was evaluated further using titration ELISA to determine whether humanisation had affected the binding properties in comparison to the parental VHH. Overall, most of the variants retained antigen binding comparable with the parental VHH, with modest differences in performance for individual substitutions (Figure 3). Importantly, favourable function and biophysical characteristics were retained not only in variants containing a limited number of framework substitutions, but also in those incorporating a substantially higher degree of humanisation. This demonstrates that the rational, structure-guided design strategy enabled extensive framework engineering while preserving the key molecular properties of the parental VHH.

Figure 2. Analytical SEC traces for the parental VHH and 23 variants, showing monomeric purity and aggregation profiles. Superdex 75 Increase 5/150 GL column; flow rate 0.3 mL/min in PBS; 10 µL injection volume.

Figure 3. Titration ELISA curves showing binding to the target antigen for the VHH variants with highest humanness (A17-013) and most mutations (A17-020). Curves represent a concentration series for each variant; left-shifted curves indicate higher apparent affinity (lower EC₅₀), whilst right shifted curves indicate lower affinity. The nomenclature P4 or P7 next to the Parental indicate that the data comes from different experimental plates to match the same conditions under which the mutants were tested (representative of technical consistency between plates).​

Overall, these findings suggest that the sequence-guided humanisation strategy successfully increased human sequence identity while preserving the molecular characteristics required for further development. Analysis of this humanised panel also provided valuable insight into the tolerance of individual framework positions to humanise. For instance, individual changes (mutations 7 and 12) appeared to be mildly detrimental, whilst mutation 4 appeared detrimental when introduced alone, but well tolerated when combined with other changes predicted to stabilise the construct by creating the canonical DNSK motif (Table 1). This finding reinforces the value of combining computational analysis with expert knowledge of VHH architecture.To further increase human sequence identity, the best performing variants from the first generation were selected for a second round of rational optimisation. Candidate selection was based on an integrated assessment of functional and biophysical data, prioritising variants that balanced low-risk, with higher risk (increased human-likeness), and retained antigen affinity and biophysical properties.

One low-risk variant designed with three framework substitutions (A17-013) and one high-risk variant with six framework substitutions (A17-020), highlighted in Table 1, exhibited characteristics closely matched to that of the parental VHH. These constructs provided an optimal starting point for further engineering and were selected for a second round of humanisation.

Results 2 – second-generation humanisation 

A second-generation design strategy was implemented to introduce additional framework substitutions predicted to further increase human sequence identity while maintaining the characteristics observed during the initial screen. The resulting second-generation panel demonstrated a high success rate.

The parental VHH, the 2 first-generation humanised lead candidates, and 19 second-generation humanised variants were recombinantly expressed and purified. The majority of the constructs expressed well and yielded predominantly monomeric purified protein, as determined by analytical SEC (>85% purity). Only four variants showed measurable reductions in production and protein purity, indicating that the additional humanising substitutions were generally well tolerated (Figure 4). This was further supported by the functional data where most of the variants retained antigen-binding affinity comparable to the parent VHH (Figure 5).

Figure 4. Analytical SEC traces for the parental VHH and 19 variants, showing monomeric purity and aggregation profiles. Superdex 75 Increase 5/150 GL column; flow rate 0.3 mL/min in PBS; 10 µL injection volume.​

Figure 5. Titration ELISA curves for binding of top 7 VHH variants to the target antigen. Curves represent a concentration series for each variant; left-shifted curves indicate higher apparent affinity (lower EC₅₀), whilst right shifted curves indicate lower affinity. Panels A–E show variants with affinity comparable to wildtype, while Panel F displays all variants from the Second Generation (A17-026 to 051) alongside controls A17-001, A17-013 and A17-020. The nomenclature P1, 3, 5, 6 or P7 next to the Parental indicate that the data comes from different experimental plates to match the same conditions under which the mutants were tested (representative of technical consistency between plates).

Table 2. Summary of 2^nd iteration mutations. Shown are the mutation combinations and respective impact on affinity and biophysical characteristics. Highlighted in pink are the top 7 variants.

Importantly, a number of the second-generation variants achieved substantially improved T20 humanness scores (a well-established humanisation scoring system used in industry) while preserving the functional performance of the parental molecule. Antibodies with high T20 scores (>80-85%) have been shown to have lower immunogenicity in the clinic (Gao et al., 2013). Most second-generation variants exceeded 80% human-likeness. Together these findings demonstrate that the iterative humanisation strategy successfully increased human-likeness without comprising key developability characteristics, resulting in multiple candidates suitable for downstream development. A final panel of seven variants were selected to progress for further assessment.

Further Characterisation of the top 7 humanised variants 

The top seven humanised variants were progressed for further biophysical and functional analysis.

To further assess the impact of humanisation on the biophysical properties of the parental VHH, the thermal stability of lead panel was determined using a SYPRO Orange-based thermal shift assay. Most of the variants retained or improved thermal stability relative to the parental VHH (Table 1).

To further evaluate the impact of humanisation on antigen recognition, the lead panel were characterised by BLI. Kinetic analysis demonstrated that the humanised variants produced binding profiles within three-fold of the parental VHH. Two clones which produced KD values similar to the parental VHH (within two-fold) had higher humanness scores of 83% and 82% (T20 %).

Figure 6. BLI sensorgrams for top seven VHH variants (and wildtype) binding to immobilised target antigen. Curves represent the concentration series, with fits to a 1:1 binding model used to derive K_D values.

Table 3. Summary of biophysical and functional characteristics for the top seven VHH variants (and parental wildtype). Values are colour-coded, with darker green indicating increased thermal stability. Kinetic binding parameters (k_on, k_off, K_D) for top seven VHH variants (and wildtype A17-001) determined by BLI. K_D fold difference is reported relative to the parental control (A17-001).

The observed association rates were broadly consistent across the panel, with some differentiation observed in the dissociation rates, which were faster for all mutants except one. While these findings suggest there is some effect on the molecular interactions responsible for antigen binding there is no appreciable loss in overall affinity (Table 3), showing that

the rational humanisation strategy successfully preserved the functional properties of the parental VHH while substantially increasing human-likeness.

Want to avoid humanisation after Discovery? 

Explore our pre-humanised synthetic library technology for discovery of novel pre-humanised VHH delivery of humanised clones with identification of unique binders in as little as 4-6 weeks. We combine multiple highly diverse and pre-humanised VHH discovery libraries, also designed to limit liabilities, in each discovery project. Our libraries range in humanisation approaches from being built on a humanised framework to also including human-like content in CDR3 and all VHH hits are then confirmed and characterised in a humanised format from the very start. Downstream characterisation and developability testing is available to confirm specificity, as well as desirable features and functionality. Starting with pre-humanised VHHs saves months of time and cost and avoids the serious risk that downstream humanisation attempts might break a good binder.

References 

1. Arbabi-Ghahroudi, M. (2017) ‘Camelid single-domain antibodies: Historical perspective and future outlook’, Frontiers in Immunology, 8, p. 1589. https://doi.org/10.3389/fimmu.2017.01589
2. Evers, A., Guarnera, E., Pekar, L. and Zielonka, S. (2025) ‘From discovery to the clinic: structural insights, engineering options, clinical, and ‘next wave’ applications of camelid-derived single-domain antibodies’, mAbs, 17(1), doi: 10.1080/19420862.2025.2583210.
3. Fernández-Quintero, M.L., Seidler, C.A., Liedl, K.R. and colleagues (2024) ‘On the humanization of VHHs: Prospective case studies, experimental and computational characterization of structural determinants for functionality’, Protein Science, 33(9), e5176. https://doi.org/10.1002/pro.5176
4. Gao, S.H., Huang, K., Tu, H. and Adler, A.S. (2013) ‘Development of a T20 score to assess antibody humanness’, BMC Biotechnology, 13, p. 55. https://doi.org/10.1186/1472-6750-13-55
5. Getts, D.R., Getts, M.T., McCarthy, D.P., Chastain, E.M.L. and Miller, S.D. (2010) ‘Have we overestimated the benefit of human(ized) antibodies?’, mAbs, 2(6), pp. 682–694. doi: 10.4161/mabs.2.6.13601.
6. Marchetti, F., Fernández-Quintero, M.L., Georges, G. and colleagues (2024) ‘Prospects for the computational humanization of antibodies and nanobodies’, Antibodies, 13(2), p. 26. https://doi.org/10.3390/antib13020026
7. Muyldermans, S. (2013) ‘Nanobodies: Natural single-domain antibodies’, Annual Review of Biochemistry, 82, pp. 775–797. https://doi.org/10.1146/annurev-biochem-063011-092449
8. Nguyen, V.K., Hamers, R., Wyns, L. and Muyldermans, S. (2000) ‘Camel heavy-chain antibodies: Diverse germline VHH and specific mechanisms enlarge the antigen-binding repertoire’, The EMBO Journal, 19(5), pp. 921–930. https://doi.org/10.1093/emboj/19.5.921
9. Nguyen, V.K., Desmyter, A. and Muyldermans, S. (2001) ‘Functional heavy-chain antibodies in Camelidae’, Advances in Immunology, 79, pp. 261–296. https://doi.org/10.1016/S0065-2776(01)79008-7
10. Rossotti, M.A., Bélanger, K., Henry, K.A. and Tanha, J. (2021) ‘Immunogenicity and humanization of single-domain antibodies’, The FEBS Journal, 288(15), pp. 4304–4327. https://doi.org/10.1111/febs.15809
11. Vincke, C., Loris, R., Saerens, D., Martinez-Rodriguez, S., Muyldermans, S. and Conrath, K. (2009) ‘General strategy to humanize a camelid single-domain antibody and identification of a universal humanized nanobody scaffold’, The Journal of Biological Chemistry, 284(5), pp. 3273–3284. https://doi.org/10.1074/jbc.M806889200