Early antibody drug development mainly relied on antibodies derived from mouse immunity. However, these antibodies can induce human anti-mouse antibody reactions (HAMA) in humans, which may lead to a rapid decline in efficacy, a shortened half-life, and potentially serious immunotoxicity, making it difficult for early antibody drugs to be clinically translated[1].
The core goal of antibody humanization is to retain or even optimize antigen-binding affinity, structural stability, expression, and developability as much as possible while reducing immunogenicity. Therefore, humanization has gradually transformed from a post-processing step into a core element of the antibody design process.
CDR grafting is a classic strategy for antibody humanization, based on the key assumption that the antigen-binding specificity of an antibody is mainly determined by the complementarity-determining region (CDR), while the framework region only provides structural support. Therefore, humanization is usually achieved by selecting a human germline framework and embedding the CDR of a murine antibody into it [2]. However, this assumption is not entirely valid in practice. The spatial conformation of the CDR depends on the support of the surrounding framework region, including Vernier region residues, VH/VL interface interactions, and the internal hydrophobic core. Once these structural environments change, the CDR conformation may shift, resulting in a significant decrease in antibody affinity, often by 10 to 1000 times.
To restore antibody activity, back-mutation is usually required, that is, retaining or restoring murine residues at key framework sites. These sites are mostly located in the vicinity of the CDR, at the VH/VL interface, or at locations that affect the canonical structure, and their role is to maintain the correct conformation of the CDR.
Fig1. Schematic illustration of the complementarity determining region (CDR) grafting and framework (FR) shuffling humanization strategies. [3]
In addition, framework shuffling offers an alternative approach. This method constructs a library of combined human frameworks and selects the most compatible combination with CDR, thereby finding the optimal solution in a larger design space. Studies have shown that this strategy is superior to traditional CDR porting in some cases, and its essence is a structure adaptation process based on experimental screening [3]. However, these processes still largely rely on empirical judgment and trial-and-error, thus they are inefficient and difficult to systematize.
Relying solely on experience to reverse mutations is inefficient, while the introduction of structural biology has enabled humanization design to move from trial and error to targeted optimization. With the development of high-resolution X-ray crystallography and Cryo-EM, researchers can identify key factors affecting CDR conformation at the atomic level, thereby avoiding the replacement of functions in a targeted manner [4].
Fig 2. Sequence and structural alignment of mouse antibody 3UJT. [4]
In structure-guided humanization, the core is to identify and process key structural residues, including hydrophobic core residues, hydrogen bond network-related sites, and VH/VL interface residues. If these positions differ significantly between the human framework and the parent antibody, direct substitution often disrupts structural stability and affects antigen binding. Therefore, practical designs usually adopt a selective humanization strategy, that is, retaining animal-derived residues at key structural sites while humanizing exposed surface regions to achieve a balance between immunogenicity and function. In a study on the humanization of an antibody against the multiple myeloma marker BCMA, researchers performed targeted mutations through structural analysis, avoiding amino acid substitutions that disrupt the antigen-binding interface, and ultimately restoring antigen-binding affinity[5].
Fig 3. Strategy for humanization of a mouse antibody based on in silico homology modeling and energy minimizations (simulated annealing). [4]
At the same time, humanization no longer focuses solely on binding activity; it also requires simultaneous evaluation of the molecular development characteristics, including thermal stability, anti-aggregation ability, and stability under different pH conditions. These properties directly determine the feasibility of the antibody in subsequent development and formulation. [6]
The establishment of large-scale natural antibody sequence databases, represented by Observed Antibody Space (OAS), is the data foundation for deep learning. [7] In this context, a number of humanization tools based on deep learning have emerged. Among them, the BioPhi platform integrates two modules, Sapiens and OASis, to evaluate the humanization of antibodies from different perspectives. Sapiens learns the statistical characteristics of human antibody sequences through neural networks, thereby scoring the humanization of each residue; while OASis evaluates the potential T cell epitope risk based on short peptide fragment matching. This multi-dimensional evaluation method makes humanization no longer dependent on simple sequence homology, but based on the probability judgment of whether it conforms to the distribution of human antibodies. [8] Similarly, AbNatiV et al. use generative models (such as variational autoencoders or Transformer architectures) to model antibody sequences, thereby evaluating their nativeness. This indicator can be regarded as a proxy variable for immunogenicity risk to some extent. More importantly, such models can not only be used for evaluation, but also for sequence optimization, that is, to make the antibody sequence closer to the distribution of human natural antibodies while maintaining function. Compared to traditional methods, AI-driven humanization has significant advantages. It can simultaneously consider the synergistic effects of multiple sites across the entire sequence, rather than modifying them point by point; it also does not rely on explicit structural information, thus making it applicable even when high-resolution structures are unavailable.
In addition to learning-based models, another important method is the global design strategy based on energy functions. Among them, the CUMAb method is representative. The core idea of this method is not to select the most similar human framework, but to systematically graft the CDR of animal antibodies onto a large number of different human frameworks and to rank these candidate structures by energy functions. [9]
The advantage of this method is that it can find the optimal solution in a larger structural space, rather than being limited to a few candidate frameworks. In many cases, this strategy can not only obtain variants with functions comparable to the original antibody, but may also perform better in terms of stability or expression level. Its essence is to transform the humanization problem into a multi-objective optimization problem and find the equilibrium solution in the complex design space through computational methods.
Fig 4. Key steps in CUMAb antibody humanization based on energy function.
A. Schematic diagram of antibody domains.
B. CUMAb first generates more than 20,000 candidate source frames for each antibody to be humanized by combining the V and J regions of the human germline heavy and light chains.
C. The CDR of the parent antibody is transplanted into each human receptor frame, and Rosetta is used for structural modeling and energy assessment. Cluster analysis is then performed on the resulting designs to screen representative candidate molecules from low-energy clusters for subsequent experimental screening and validation. [9]
The humanization of nanobodies (VHH) varies significantly in methodology. Due to their camel origin and the absence of light chains, their framework regions, particularly the FR2 region, possess unique hydrophobic characteristics. These characteristics are crucial for maintaining the solubility and stability of single-domain antibodies.
Therefore, directly applying humanization strategies to IgG antibodies often leads to problems. For example, simply replacing the FR2 region with human sequences may disrupt solubility or even cause protein aggregation. To address this issue, researchers typically adopt a more conservative strategy, prioritizing the preservation of key structural residues while gradually introducing human characteristics. Furthermore, specialized tools such as Llamanade have been developed to analyze the differences between VHHs and human antibodies, guiding more rational humanization designs. Overall, the humanization of nanobodies requires a more nuanced trade-off between structural integrity and humanization.
As antibody humanization gradually enters the "multi-objective optimization" stage, relying solely on a single algorithm or traditional experience is no longer sufficient to meet the needs of drug development. In practical projects, humanization not only needs to reduce immunogenicity but also needs to simultaneously consider:
● Maintaining antigen-binding activity
● Thermal stability and anti-aggregation ability
● Expression level and manufacturability
● VHH-specific solubility and structural integrity
● Compatibility with subsequent ADCs, multivalent constructions, or cell therapy
Therefore, humanization is evolving from a "sequence substitution problem" to a "systems engineering problem."
As a CRO platform focused on custom VHH/sdAb discovery & development, AlpVHHs has built a complete platform covering nanobody discovery, optimization, humanization, and development evaluation. Based on large-scale camel immunization systems, phage display, and yeast display technologies, the platform supports continuous development from early antibody discovery to candidate molecule optimization.
CDR grafting is the main method to be used in humanization, the main process is as follows:
● A state-of-the-art deep learning-based method will be utilized to predict the precise structure.
● Templates will be chosen based on general sequence similarity, loop structural similarity, expression level, and other criteria.
● Critical binding sites will be identified through structural analysis.
● Immunogenicity will be calculated using in-house developed software, and post-translational modifications (PTM) will be annotated based on real-world data.
● After the initial humanization round, regions exhibiting high immunogenicity, PTM, proline (PI), hydrophobicity, andaggregation propensity will be further engineered.
The primary goal of antibody humanization design is to reduce the immunogenicity of the antibody rather than striving for a higher homology with human sequences.
Our tool exhibits strong predictability for the lower limit of antibody immunogenicity, with a Pearson correlation coefficient of 0.73.
Reference
[1]. J Biomed Sci. 2020 Jan 2;27(1):1. doi: 10.1186/s12929-019-0592-z.
[2]. Sci Rep 8, 14820 (2018). https://doi.org/10.1038/s41598-018-32986-y
[3]. Front Immunol. 2024 Jul 15;15:1395854. doi: 10.3389/fimmu.2024.1395854
[4]. Bioinformation. 2014 Apr 23;10(4):180–186. doi: 10.6026/97320630010180
[5]. J Mol Med (Berl). 2024 Jul 25;102(9):1151–1161. doi: 10.1007/s00109-024-02470-4
[6]. Proc Natl Acad Sci U S A. 2017 Jan 17;114(5):944–949. doi: 10.1073/pnas.1616408114
[7]. Protein Sci. 2021 Oct 29;31(1):141–146. doi: 10.1002/pro.4205
[8]. MAbs. 2022 Jan-Dec;14(1):2020203. doi: 10.1080/19420862.2021.2020203
[9]. Nat Biomed Eng. 2023 Aug 7;8(1):30–44. doi: 10.1038/s41551-023-01079-1
[10]. MAbs. 2015 May 27;7(4):693–706. doi: 10.1080/19420862.2015.1046648
[11]. J Biol Chem. 2009 Jan 30;284(5):3273-3284. doi: 10.1074/jbc.M806889200
