What are VHH Antibodies?

VHHs, also known as single-domain antibodies (sdAbs) or nanobodies, are the smallest antigen-binding fragments derived from heavy-chain-only antibodies found in camelids, such as alpacas, llamas and camels.

First discovered in 1993, VHH antibody fragments offer unique characteristics which are particularly advantageous in therapeutics, diagnostics, and research tools. This is due to their small size and structure which allows them to penetrate tissues and reach targets that may be challenging for conventional antibodies, in addition to their stability and ability to bind with high affinity to specific targets.¹

The VHH domain consists of approximately 120–135 amino acids, forming a characteristic structure with four framework regions (FRs) and three complementarity-determining regions (CDRs). The absence of a light chain and the unique arrangement of CDRs enable VHHs to bind to epitopes that are often inaccessible to conventional antibodies, such as protein-protein interaction sites and small molecules. This unique binding capability makes VHHs valuable in therapeutic applications, diagnostics, and research tools.

Applications in Therapeutics

VHHs have emerged as promising therapeutic agents, particularly in oncology and immunology. Their modularity allows for the construction of bi- and multispecific antibodies, enhancing their potential as therapeutic candidates. Examples of VHH-based therapeutics include bispecific antibodies targeting multiple antigens, trifunctional constructs, and immuno-oncology agents that engage immune effector cells. 

Bispecific and Trispecific VHHs

VHHs can be engineered to create bispecific or trispecific constructs, allowing simultaneous targeting of different antigens or pathways. For instance, the VHH-based bispecific antibody Erfonrilimab (KN046) targets PD-L1 and CTLA4, while M1095/ALX0761 is a trispecific VHH targeting IL-17A, IL-17F, and human serum albumin (HSA). The presence of HSA in these constructs extends their plasma half-life, making them suitable for therapeutic applications.

Additionally, novel constructs such as TriTACs (trispecific T-cell activating constructs) have been developed, which incorporate VHHs to enhance T cell recruitment. These constructs target tumor-associated antigens (TAAs) while binding to CD3, facilitating the activation of T cells in the tumor microenvironment.

Clinical Trials

Several VHH-based therapeutics are undergoing clinical evaluation. M1095 is being investigated for its efficacy in moderate-to-severe psoriasis, hidradenitis suppurativa, and psoriatic arthritis (Phase 2 trials). BI836880 targets vascular endothelial growth factor (VEGF), angiopoietin-2, and HSA, with potential applications in solid tumors. Inhibrx’s INBRX105, a bispecific tetravalent antibody targeting PD-L1 and CD137, is also in Phase 2 trials in combination with pembrolizumab. 

Machine Learning in VHH Design

The advent of machine learning techniques has significantly impacted VHH design and optimization. Computational approaches are increasingly utilized for structural prediction, lead identification, and sequence optimization. 

Structural modeling is crucial for the design of VHHs, particularly when experimentally determined structures are unavailable. Machine learning tools such as AlphaFold and its successors have revolutionized protein structure prediction. VHH-specific tools like NanoBodyBuilder2 and ABodyBuilder2 allow for accurate predictions of VHH structures. The integration of molecular dynamics simulations with these predictions can enhance the design of multivalent VHHs, demonstrating improved binding affinities and biological activity. 

Generative machine learning models, including deep learning frameworks, are employed to design novel VHHs. Models such as AntiBERTY, AbLang, and IgLM focus specifically on antibody sequences, learning from vast datasets to identify patterns for lead identification. Recent advances have enabled these models to generate diverse VHH sequences and optimize binding properties. 

One notable example is DiffAb, which uses antibody-antigen structures to generate novel CDRs while preserving target binding capability. Similarly, LLMs have been employed for affinity maturation campaigns, resulting in significant improvements in binding affinities for VHHs. 

Machine learning approaches like Bayesian optimization (BO) have shown promise in navigating the vast sequence space of VHHs. BO enables efficient exploration of the design space while optimizing multiple properties, such as affinity, stability, and developability. For instance, AntBO and LaMBO2 have demonstrated success in designing high-affinity scFv libraries and optimizing VHH constructs. 

Humanization of VHHs is often necessary to reduce immunogenicity and improve compatibility with human therapeutics. While several machine learning tools have been developed for antibody humanization, their applicability to VHHs remains uncertain due to unique residues present in VHH frameworks. Tools like AbNatiV allow for fine-tuning humanization while preserving crucial camelid residues.²

References: 

1. Hamers-Casterman, C., Atarhouch, T., Muyldermans, S., Robinson, G., Hamers, C., Songa, E. B., Bendahman, N., & Hamers, R. (1993). Naturally occurring antibodies devoid of light chains. Nature, 363(6428), 446–448. https://doi.org/10.1038/363446a0

2. Mullin, M., McClory, J., Haynes, W., Grace, J., & Robertson, N. (2024). Applications and challenges in designing VHH-based bispecific antibodies: Leveraging machine learning solutions. MAbs, 16(1). https://doi.org/10.1080/19420862.2024.2341443