From discovery to the clinic: structural insights, engineering options, clinical, and ‘next wave’ applications of camelid-derived single-domain antibodies

ABSTRACT

Camelid-derived single domain antibodies (sdAbs), referred to as VHHs (variable domains of the heavy chain of a heavy chain-only antibodies), recently emerged as promising building blocks for the construction of therapeutic molecules. As of October 2025, on a global perspective, five therapeutics harboring VHH-based paratopes have been granted marketing access from different health authorities. VHHs possess several favorable attributes, such as classical antibody-like affinities and specificities, adequate stabilities and a relatively small size (i.e., low molecular weight) as independent paratopes. Moreover, due to the lack of -light chain association, these sdAbs afford the benefit of multiple reformatting options for engineering of bi- and multifunctional antibodies. In this review, we summarize the structural features of VHHs, discuss sequence diversities of sdAb repertoires found in different camelid species commonly used for VHH generation, and describe different platform technologies for the isolation of VHH paratopes and their humanization. Moreover, we briefly review the composition as well as mechanisms of action of marketed VHH-based therapeutics and present novel biological concepts that harness sdAbs as targeting compounds. Finally, we provide an in silico property analysis of 49 VHH sequences that have progressed into clinical studies, which may inform future development of VHH-based therapeutics.

Introduction

Antibody (Ab) therapeutics have revolutionized the treatment of many diseases. This is exemplified by the fact that, according to the Antibody Society, as of September 2025 nearly 200 Ab-based therapeutics have been approved for treatment by different health authorities (YAbS database: https://db.antibodysociety.org/., accessed on September 26, 2025).¹ In this regard, of the Top10 selling drugs in 2024, five therapeutic molecules were Ab-derived, pembrolizumab, dupilumab, risankizumab, daratumumab, and ustekinumab.² Main therapeutic areas of approved Abs are cancer indications followed by immune-mediated disorders and infectious diseases, but Abs have also been approved for the treatment of cardiovascular, metabolic, neurological, ophthalmic, and skeletal disorders.¹ While most of the approved Ab-like therapeutics have the classical monospecific immunoglobulin G (IgG) composition, there is a clear trend toward more complex architectures, such as antibody-drug conjugates (ADCs) and bi- or multispecific antibodies.^3–7 In addition to this, chimeric antigen receptor (CAR) modified T cells, which are genetically engineered T cells expressing synthetic Ab-based paratopes as antigen-targeting receptors, have proven to be successful as therapeutics, especially in hematologic malignancies.^8,9

While classical IgGs are heterotetramers composed of two identical heavy chains and two identical light chains (Figure 1(A)), the adaptive immune systems of camelids and cartilaginous fish generate homodimeric Abs consisting of heavy chains only, referred to as heavy chain antibodies (HcAbs, Figure 1(B)). In camelids, the antigen-binding site is a single variable domain, termed VHH (commercially also known as a “nanobody”), which functions as an autonomous paratope without a light chain variable domain (VL) partner. Here, we focus on camelid HcAbs and their VHH domains. Information about HcAbs found in cartilaginous fish is available in other reviews^10–15. Throughout this review, we will use the term VHH for paratopes that are clearly derived from camelids, whereas sdAbs also encompass single-domain paratopes derived from human heavy chain variable domains (VH) or transgenic animals.

Figure 1.

Architecture and structural determinants of camelid VHHs compared with conventional IgG. (A) Schematic of a human IgG1: two heavy chains and two light chains form the paratopes by VH–VL pairing. (B) Schematic of a camelid heavy chain–only antibody (HCAb): a single VHH is followed by hinge, CH2, and CH3. Light chains and CH1 are absent. (C) Classical Fv view: a VH domain requires a VL partner to complete the paratope, and the VH FR2 VGLW motif is buried at the VH–VL interface. (D) VHH domain as an autonomous paratope: FR2 hallmark residues at IMGT® positions 42, 49, 50, and 52 (e.g., FERF, YQRL, FERG, YERW) are solvent‑exposed and can contact the antigen or stabilize the CDR3 conformation. (E) Overlay of representative VHH structures with CDRs highlighted: the frameworks superimpose to a conserved Ig Fold. (F) Binding‑site–focused overlay: CDR3 shows broad conformational variation and, together with CDR1/2 and framework residues, can create large paratopes. (G, H) Examples with non‑canonical disulfides linking CDR3 to Cys38 (CDR1) or Cys55 (FR2), respectively (PDB codes 2X6M, 4LAJ), mirroring species‑enriched positions in the repertoire logo plots (see Figure 2). Although distant in sequence, both positions are proximal in 3D, enabling CDR3 linkage.

In the 1990s, naturally occurring heavy chain-only antibodies devoid of light chains were first described in camelids.¹⁶ These HcAbs, which also lack constant heavy-chain domain CH1, are composed (from the N- to the C-terminus) of a variable domain (VHH), followed by a hinge region as well as domains CH2 and CH3 (Figure 1(B)). Phylogenetic analyses have shown that γ genes encoding for HcAbs originated from conventional Ab encoding genes.^17,18 Likewise, germline VHH genes derived from classical VH genes and the D and J(H) genes are even shared in the creation of both VH and VHH domains.¹¹ Interestingly, a point mutation within the intron between CH1 and the hinge exons, inactivating the splicing recognition sequence, elicits removal of the CH1 region during mRNA splicing, ultimately resulting in the formation of HcAbs in camelids.^14,19,20 Here, the antigen-binding site is composed of one single variable domain (VHH), mediating high-affinity and highly specific binding to a given antigen.

Due to their simple architecture, low aggregation propensities, high thermostability, and high human-likeness, VHH domains emerged as auspicious building blocks for the generation of therapeutic Abs.^10,12,13 This is evidenced by the fact that five marketed therapeutics are composed of VHH domains, which are described within this work. Furthermore, patent landscape analyses revealed a rapid growth in VHH innovation within the last decade, as indicated by the annual growth rate of globally filed patent applications related to VHHs of approximately 20%.^21–23 Since these small paratopes do not associate with VLs, VHHs can be readily incorporated into multispecific Ab architectures. Indeed, Garces and coworkers’ review of the clinical landscape of multispecific Abs in 2022 showed that approximately 10% of multispecifics within Phase 1 to 3 clinical trials harnessed VHHs as building blocks, suggesting a prominent role for camelid-derived sdAbs for the next wave of multispecific biologics.²⁴

General architecture, sequence, and structural features of VHHs

Unlike classical antibody variable heavy (VH) domains, which require pairing with a variable light (VL) domain to form a complete paratope (Figure 1(B)), camelid VHHs function autonomously. They possess so-called hallmark signatures and structural features that enable solubility and stability in the absence of a light chain partner (Figure 1(C, D)). While often described as adaptations to compensate for the loss of VL pairing, it has been discussed that VHHs may also reflect a more ancient or parallel mode of antigen recognition, akin to the independently evolved IgNAR antibodies in cartilaginous fish.^11,17 Their paratope is formed by three complementarity-determining regions (CDRs), with CDR3 often being longer and more variable than in conventional VH domains, enabling recognition of otherwise more difficult epitopes.^12,25–34 Large-scale structural overlays of VHHs from the Protein Data Bank (Figure 1(D-F)) show a highly conserved Ig-fold across frameworks, while the antigen-binding site, particularly CDR3, exhibits substantial conformational diversity.

Framework region 2 (FR2) of VHHs contains characteristic hallmark residues (positions 42, 49, 50, and 52, IMGT numbering) essential for stability and solubility.^35–37 Whereas classical VH typically presents a VGLW motif, buried by the VL chain, VHHs display diverse motifs (e.g., FERG, FERF, YQRL, YERW)^38,39 that can directly contribute to antigen binding and help pre‑organize the bioactive CDR3 conformation.^40–44 Due to the genetic organization of the camelid heavy‑chain locus, both VHH and conventional VH genes contribute to heavy chain‑only antibodies. Consequently, a minority of HcAbs retain the VH‑like VGLW motif.⁴⁵ For clarity, throughout this review we refer to all variable domains derived from camelid heavy chain-only antibodies as “VHHs,” regardless of whether they contain VHH-specific hallmark residues or retain VH-like sequence features, as their defining characteristic is their ability to function autonomously without light chain pairing.

VHHs also frequently harbor non‑canonical disulfide bonds, either within CDR3 or linking CDR3 to framework residues, that stabilize CDR3, reduce conformational entropy, and expand paratope diversity.^{17,30,41,12,46–49} Repertoire analyses indicate that more than 25% of VHH sequences contain such CDR3‑associated disulfides.^38,41 Comparative analyses across camelid repertoires reveal conserved principles but species‑specific cysteine placement shaping disulfide connectivity. As is evident from logo-plot analysis (Figure 2), non-canonical cysteines are enriched in CDR1 (preferentially IMGT position 38) in camels and dromedaries versus FR2 position 55 in llamas or alpacas.^{17,30,33,50,51} Although these positions are distant from a sequence perspective, they are structurally proximal (Figures 1(G, H)), both enabling disulfide linkage with CDR3. Notably, VHHs carrying an additional non-canonical disulfide bond tend to exhibit longer CDR3 sequences that often adopt pre-organized conformations, thereby broadening the range of epitope geometries they can recognize.^41,42 Of note, non-canonical cysteines can also be found in human VH domains, mainly in antibodies with long CDR-H3 regions, potentially forming various disulfide bond patterns within CDR-H3 or between CDR-H3 and other CDRs and framework regions.⁵²

Figure 2.

Sequence-logo style summaries of V-region segments highlight conserved and divergent features across repertoires. CDR1 residues are shown in blue and CDR2 in magenta. FR2 hallmark positions (IMGT 42, 49, 50, 52) are highlighted in orange, illustrating camelid-enriched motifs (e.g., FERF/FERG/YQRL/YERW) versus the human VH VGLW motif (bottom row). Yellow marks non‑canonical cysteine hotspots. Two species‑biased positions are evident: CDR1 IMGT 38 (camel/dromedary‑enriched) and FR2 IMGT 55 (llama/alpaca‑enriched). Although distant in sequence, these positions are proximal in 3D and can form stabilizing disulfides with CDR3 (see Figure 1G,H). IMGT numbering is provided at the bottom to facilitate position identification. Overall, camelid repertoires show diverse FR2 motif usage and characteristic cysteine placement alongside variable CDR1/2 composition.

Overall, logo-plot analysis reveals that the four camelid V-gene repertoires (Figure 2) are highly similar. In comparison, the human VH repertoire is less similar but still shows substantial framework conservation, supporting straightforward humanization of camelid VHHs (see below).

Generation of antigen-specific sdAbs

Antigen-specific sdAbs can be generated in several ways, for instance, from immunized camelids and transgenic sdAb animals, naïve and synthetic or semi-synthetic libraries.^53–62 Most recently, de novo design approaches harnessing artificial intelligence have been applied for the discovery of VHHs.⁶³

Likewise, sdAb selections can be conducted using multiple different platform technologies, ranging from phage display over yeast surface display or ribosome display to mammalian technologies exploiting microfluidics.^53,64–70 Since different display technologies were extensively discussed elsewhere, we will focus the following on the different sources of sdAb diversities.^71–74

Camelid immunization

Camelids represent ideal hosts for immunization against human antigens for two complementary reasons: 1) their distant phylogenetic relationship to humans, which increases the foreignness, i.e., antigenicity of human proteins when injected into camelids; and 2) their high sequence similarity in terms of immunoglobulin variable genes, particularly between camelid VHH domains and human VH3 family genes (Figure 2), making VHHs readily amenable to humanization.⁷⁵ Consequently, the most obvious procedure to obtain high-affinity and highly specific VHH domains against a given target is camelid immunization, leveraging the animal’s adaptive immune system, e.g., somatic hypermutation for Ab generation and affinity maturation. Immunization of various different camelid species such as llamas, alpacas, camels and dromedaries has been described, as elegantly reviewed by Muyldermans.⁵⁷ Within a timeframe of typically 6–12 weeks,^66,76 animals are immunized several times with antigen, most commonly recombinant proteins. For example, our standard immunization protocol comprises four injections in total (d0, d14, d28, and d35) using 300 µg of antigen per immunization (subcutaneous injection) and animal with blood collection (100 mL) seven days post final administration. In this regard, several different adjuvants can be used, e.g., Freund’s Adjuvant or Gerbu fama adjuvant.^77–79 Of note, immunization can also be performed in a cocktail manner, using several different antigens at once.⁵⁷ Besides protein antigens, other sources can also be harnessed for immunization, such as bacteria, viruses, or cancer cell extracts.^80–83 For proteins that might be difficult to purify in their native conformation, genetic immunization might pose a viable strategy.⁸⁴ A differentiating characteristic to conventional mouse immunization protocols relies on the fact that immunized camelids usually remain alive after the procedure, a clear benefit from an animal welfare perspective.

After blood collection, RNA is typically isolated from peripheral blood mononuclear cells (PBMCs), followed by cDNA synthesis and PCR-based VHH amplification for, for example, phage or yeast display library construction.^66,68 Selection and panning can usually be conducted within 1–3 rounds and, ultimately, antigen-specific VHH sequences can be obtained in about 3–4 weeks after the immunization procedure.

Immunization of transgenic sdAb animals

As an alternative to camelid immunization, transgenic animals that produce sdAbs as part of their humoral immune response have been generated. While several platforms were described decades ago,^85,86 we focus here on recent developments in this field.

Casellas and coworkers engineered the ‘Nanomouse’ via CRISPR-Cas9-based integration of 18 alpaca, 7 dromedary and 5 Bactrian camel VHH genes into the genome of mouse embryonic stem cells.⁶¹ The 25 kb insertion cassette replaced a 2.5 mb gene fragment in the mouse IgH locus. Additionally, the CH1 exons of IgM and IgG1 were deleted. In this mouse strain, the VHH-derived V genes recombine with mouse D and J segments and the hinge and Fc regions are based on mouse genes.⁸⁷

The group of Koch-Nolte generated a VHH-producing transgenic mouse line referred to as ‘LamaMouse’.^54,87 To this end, an engineered llama immunoglobulin heavy chain (IgH) locus was randomly inserted into IgH-deficient mice by exploiting a bacterial artificial chromosome (BAC). Resulting V-domains also contain llama D and J segments, encoding for CDR3 and FR4. ‘LamaMice’ exclusively produce camelid IgM and IgG heavy chain-only antibodies.

Clarke et al. described the creation of human sdAb producing transgenic rats called ‘UniRats’.⁸⁸ These animals produce fully human VH-based sdAbs that pair with rat CHγ genes lacking the CH1 domain. Due to the large size of the full VH repertoire, the group generated two separate UniRat strains with 22 and 23 different V genes, respectively. Both strains harbor the full suite of human D and JH genes. The strains were created via DNA microinjection of overlapping BACs into fertilized rat oozytes. The endogenous rat IgH, Igκλ loci were inactivated.

In addition to transgenic murine species, Leighton and colleagues generated transgenic chickens capable of producing human sdAbs, named ‘OmnidAb’.⁶⁰ For this, the group designed a ‘camelized’ human VH domain based on human VH3-23 and JH4 harboring 10 stabilizing mutations that enable the expression as autonomous VH. These mutations were ‘inspired’ by camelid VHH domains and serve to reshape the former interface regions with VL and CH1 to decrease hydrophobicity. The resulting variable region was inserted into the chicken IgH locus in which the chicken V, D, and J segments were deleted. To facilitate sufficient diversification via gene conversion,⁸⁹ the authors also inserted a set of 14 pseudogenes.

Naïve, synthetic, and semi-synthetic library approaches for VHH generation

Numerous libraries, either derived from non-immunized camelids, or (semi-)synthetic libraries (either camelid or humanized) have been developed for VHH discovery in the past.^56,90–99

In naïve libraries, blood is collected from non-immunized animals and PBMC-derived RNA is usually used as a starting material to amplify the VHH repertoire. To increase library diversity and ultimately enhance the chance to select for adequate paratopes against a given target, these libraries are often based on multiple camelids.^100,101 In semi-synthetic approaches, parts of the library are of camelid origin. For instance, a naïve library can serve as a starting point that is subsequently further diversified in a synthetic manner.⁵⁵ In synthetic libraries, VHHs involving framework regions and CDRs are created in an artificial manner involving framework design and diversification of CDRs.¹⁰

As summarized by Valdés‑Tresanco et al.,¹⁰² synthetic libraries potentially offer several practical advantages: they enable discovery against non‑immunogenic or toxic targets, can be reused across multiple programs, shorten timelines to obtain binders, and allow upfront optimization of physicochemical properties. These benefits are tempered by trade‑offs: selected VHHs may exhibit weaker biophysical behavior, very large library sizes are typically required, and affinities and specificities are often lower than those obtained from immune (immunization‑derived) libraries.

Erasmus et al.¹⁰³ recently reported a semi-synthetic/semi-naive VHH discovery platform optimized toward humanness, stability, affinity, diversity, developability, and Fc‑free, Protein A – enabled purification. Four therapeutic humanized VHHs (retaining their hallmark regions) served as backbones. For each scaffold, human VH3‑family CDR1 and CDR2 sequences were mined from next‑generation sequencing datasets, computationally purged of liabilities, synthesized as array‑based oligos, and cloned as single‑CDR libraries. These were filtered by yeast display for correct folding and Protein A binding. The filtered CDR1/2 sets were then combined with amplified CDR-H3s from human donors to assemble full libraries. A subsequent iteration removed long consecutive tyrosine tracts in CDR3 and enriched CDR1/2 diversity for elevated heat tolerance, yielding a broad diversity of high‑affinity (KD ∼100 pM-10 nM) binders against different targets with favorable profiles in early developability screening assays.

Our group³⁸ described a semi-synthetic/semi-immune approach for generating humanized and in silico-optimized VHHs following camelid immunization. In this strategy, CDR3 regions are amplified from the PBMC repertoire of an immunized llama and grafted onto two different humanized VHH backbone libraries, each featuring distinct hallmark motifs (FERF and VGLW). The humanized frameworks were derived from human germline sequences and were specifically designed to enhance developability by maximizing human-likeness, removing chemical liability motifs and glycosylation sites, and minimizing predicted immunogenicity. Importantly, both backbone libraries contained CDR1 and CDR2 regions that were moderately sequence-diversified similar to natural immunized and naïve antibody repertoires. The resulting libraries were screened using yeast surface display, enabling the rapid isolation of NKp46-targeting sdAbs with favorable biophysical and developability profiles and high affinities (KD values in the low nanomolar range), closely matching the affinities of wild-type VHHs.

In addition to classical screening, libraries and repertoires obtained from selections can be deeply mined using next-generation sequencing and artificial intelligence/machine learning (AI/ML) approaches.¹⁰⁴ As demonstrated in our recent work,¹⁰⁵ NGS of sorted display libraries enables quantitative analysis of sequence enrichment and abundance across selection rounds. By clustering enriched sequences and applying long-short term memory deep generative models, sdAb variants can be in silico sampled and further triaged for developability. This combined strategy allows rapid identification and design of promising, humanized, and sequence-optimized VHHs directly from immune repertoires. Recent advances go even further by integrating massively parallel functional screening with sequencing and ML: Porebski et al.⁶⁵ described a “deep screening” approach that combines Illumina sequencing, ribosome display, and affinity screening directly on the flow cell, enabling the simultaneous measurement of binding properties for millions of antibody variants. These rich datasets can then be used to train large language models, which are capable of generating new antibody sequences within these sequence spaces with improved affinities.

Target-specific de novo design of VHHs

While successful cases have been reported for the de novo design of miniproteins in literature,^106,107 de novo design of VHHs and classical antibodies is still in an early but rapidly advancing stage.

In a preprint, Bennett et al.⁶³ tailored RFdiffusion¹⁰⁸ for VHH and antibody design by training predominantly on antibody – antigen complexes, constraining frameworks while focusing sampling on CDR loops, incorporating epitope-targeting features to drive loop-mediated interactions. Designs were triaged with a fine‑tuned RoseTTAFold2¹⁰⁹ self‑consistency filter and Rosetta ddG, then screened either at higher throughput by yeast surface display (≈9,000 designs per target for RSV sites I/III, SARS‑CoV‑2 RBD, and influenza HA) or at lower throughput via expression in E. coli followed by binding assessment using single‑concentration surface plasmon resonance technology (SPR, 95 designs per target for TcdB, IL‑7 R, and HA). Confirmed binders generally showed modest initial affinities (e.g., KD = 78 nM for influenza HA; 260 nM for TcdB), which could be improved to the low-nanomolar range by affinity maturation. Cryo‑electron microscopy validated that designed VHHs bound in the predicted orientations, with near-atomic agreement with the models.

In a recently published study on 4-1BB,¹¹⁰ Poddiakov et al. applied a complementary workflow that combined knowledge‑based CDR sampling on a 4‑1BB – VHH template with AlphaFold2¹¹¹ complex evaluation, interface surface‑area calculations, and Rosetta interface energies, followed by iterative sequence redesign using ProteinMPNN¹¹² and Rosetta FastRelax¹¹³. From 80 top‑ranked designs, 65 were assembled and 35 sequence‑validated. Only weak binding signals were detected for one clone and several constructs showed expression challenges, underscoring the complexity of sampling diverse CDRs and providing important benchmarks for improving foldability and developability filters.

Most recently, Mille-Fragoso et al.¹¹⁴ introduced “Germinal,” a generative framework that integrates AlphaFold-Multimer with an antibody-specific protein language model (IgLM)¹¹⁵ to co-optimize structure and sequence during de novo CDR design. The method incorporates custom loss functions to ensure CDR-mediated binding and flexible loop conformations while biasing designs toward antibody-like sequences. Designs are filtered using AlphaFold3 confidence scores and PyRosetta¹¹⁶-derived interface metrics before experimental validation. Across the four diverse targets (PD-L1, IL3, IL20, and BHRF1), Germinal achieved experimental success rates of 4–22% after testing 43–101 designs per target, yielding VHHs with nanomolar binding affinities (KD range of best binders: 140–560 nM). In parallel, Swanson et al. developed mBER¹¹⁷, an open-source framework that builds on ColabDesign¹¹⁸ to achieve antibody-format binder design without requiring additional training of underlying folding and language models. The method combines ESM-2¹¹⁹-derived sequence guidance for CDR regions with NanoBodyBuilder2¹²⁰ generated structural templates to direct AlphaFold-Multimer toward functional VHH designs against specific epitopes. Using this approach, mBER designed two libraries comprising over 1 million VHH sequences against 436 diverse targets, achieving significant design success against 45% of experimentally tested targets (65 of 145), with binding rates reaching up to 38% for favorable epitopes after strict filtering.

Together, these studies indicate that the de novo VHH design is already capable of generating structurally accurate, epitope-focused binders and validating them experimentally. While further optimization of affinity and developability remains necessary, the pace of progress suggests that de novo approaches will increasingly complement traditional discovery and maturation strategies, potentially offering a new avenue for VHH discovery and engineering.

VHH humanization and sequence optimization

VHH domains usually require humanization and sequence optimization to be suitable for therapeutic use.^25,41 As recently reviewed (Gordon et al., 2024), protocols developed for conventional antibodies cannot be applied verbatim to VHHs: FR2 hallmark residues (IMGT positions 42, 49, 50, 52), reliance on long, often pre‑organized CDR3 loops, and frequent non‑canonical disulfides tightly couple the framework composition to the paratope geometry, making direct transfer of VH–VL humanization rules non‑trivial. A pragmatic strategy, supported by multiple humanization studies,^37,41,121 is a framework-based humanization that deliberately preserves sensitive positions to generate a “low-risk” variant: retain the parental CDRs on a human (or humanized) scaffold, back-mutate critical Vernier-zone and hallmark positions, and preserve any cysteines engaged in non-canonical disulfides with CDR3. This ‘low-risk’ variant typically maintains parental affinity unless the VHH uses an atypical binding mode. From this baseline sequence, additional variants can be derived selectively, guided by position-specific risk assessments and ranked not only by human-likeness but also by orthogonal criteria, including predicted physical/chemical stability and “human nativeness” or related MHC-II liability proxies along the sequence^122,123

VHH developability assessment, in silico prediction, and optimization

Recent approvals of VHH‑based drugs, together with a steady stream of VHH candidates entering clinical evaluation, demonstrate that VHH domains are clinically developable building blocks. However, experimental and in silico developability assessment for VHHs presents distinct challenges compared to conventional antibodies.

For conventional antibodies, decades of experience with numerous approved products have enabled the development of empirically defined drug-like boundaries and widely accepted developability blueprint criteria.^124–130 In silico and ML predictors were trained on these large datasets and typically predict antibody Fv domain properties.^{126,131–136} This approach is effective for classical IgGs because constant regions are largely identical, enabling reliable predictions from variable-domain features alone.

Several in silico property prediction approaches are transferable from Fv domains to VHHs, such as hydrophobicity, chemical stability, and immunogenicity risk assessments. In addition, general Fv‑level in silico approaches are continuously being adapted to VHHs.^{21,39,137–140} Emerging tools include polyreactivity predictors trained on VHH data,¹⁴¹ VHH-specific melting temperature models,^140,142 tailored structure prediction methods like NanobodyBuilder2, and sequence-based language models for context-aware design, such as nanoBERT.¹⁴³

However, in contrast to classical IgGs, large and systematic experimental datasets are lacking for VHHs, with only five marketed products available for calibration of “drug-like” boundaries. A further practical challenge in VHH developability assessments is the diversity of VHH-based antibody formats. Unlike IgG1s where architectural uniformity enables cross-program benchmarking and Fv-based in silico developability assessment, VHH therapeutics use diverse formats even among the five approved products (see below). The mapping from domain-level properties to whole-construct developability for non-classical IgG1s remains poorly understood,^144–146 as it depends on complex interactions between architecture, inter-domain interfaces, valency, dipole moments, linker design, and formulation context. Consequently, predictive models that can bridge domain-level predictions to construct-level developability across diverse VHH formats are urgently needed.

Current best practice for developability assessment of full constructs employs early experimental screening of the (often multispecific) Ab derivatives supported by predictive modeling using simple, format-agnostic, sequence-based descriptors, such as isoelectric points, charge distribution, and hydrophobicity. In our recent systematic case study on bispecific IgG1-VHH constructs, aligning both Fab and VHH domains to slightly basic pI values, approximately 7.5 to 9.0, mitigated charge asymmetries and yielded variants with improved colloidal stability and lower viscosity.¹⁴⁷

Approved VHH-based therapeutics

As outlined above, as of 2025, five VHH-based therapies reached marketing approval by various healthcare authorities for the treatment of different diseases (Figure 3). Even within this small set, Ab architectures are appreciably diverse, ranging from IgG-like molecules with a VHH replacing the VH and paired to a ‘non-functional’ VL, over homodimeric VHH-Fc fusions or standalone multivalent/tandem VHHs in a ‘beads-on-a-string’ manner (including bispecific, human serum albumin (HSA) binding constructs), to a CAR whose ectodomain is composed of biparatopic VHHs assembled in tandem. Figure 3 illustrates how VHHs (blue) are deployed as IgG variable domains, Fc-fused dimers, soluble multivalent nanobodies, or the antigen-recognition module of engineered cells, with valency, specificity, and half-life tuned by linkers, Fc, and HSA binders. This architectural breadth underscores the plug-and-play nature of VHHs and their compatibility with both protein- and cell-therapy scaffolds.

Figure 3.

Architectures used by the five approved VHH-based therapeutics. VHH domains are shown in blue; other components are given in gray. From left to right: caplacizumab, a monospecific bivalent tandem VHH; netakimab, an IgG-like antibody in which a VHH replaces the VH and pairs with a VL; envafolimab, a VHH–Fc homodimer; ozoralizumab, a trivalent “beads-on-a-string” construct with two anti‑TNF VHHs flanking an HSA binding VHH for half‑life extension; and ciltacabtagene autoleucel, a CAR whose ectodomain harbors biparatopic anti‑BCMA VHHs arranged in tandem. These formats illustrate how VHHs enable tuning of valency, specificity, and half-life via linkers, Fc, and HSA engagement or cellular display. Figure partially generated with bioRender (www.biorender.com).

Caplacizumab

Caplacizumab (Cablivi), developed by Ablynx (now a Sanofi affiliate), was first approved in Europe in 2018 for the treatment of acquired thrombotic thrombocytopenic purpura (aTTP)¹⁴⁸. The disease is caused by large aggregates of platelets on ultralarge von Willebrand factor (vWF) multimers, resulting in clinical manifestations such as thrombocytopenia, hemolytic anemia, fever, and organ dysfunction.¹⁴⁹ Caplacizumab targets vWF and inhibits the interaction between ultralarge vWF and platelets.¹⁵⁰ From a structural perspective, the therapeutic is a humanized anti-vWF monospecific bivalent tandem VHH of two identical paratopes, separated by a short triple alanine linker (Figure 3).

Netakimab

Netakimab (Efleira) was developed by Biocad and was first approved in 2019 in Russia for the treatment of moderate-to-severe plaque psoriasis.¹⁵¹ The humanized monoclonal antibody neutralizes interleukin (IL)-17, a cytokine involved in the pathogenesis of several immune-related diseases.^152,153 Structurally, netakimab is a modified IgG1 in which the VH was replaced by an anti-IL17 VHH that was paired with a light chain belonging to the VK3 family^154,155 (Figure 3).

Envafolimab

Envafolimab (Enweida), co-developed by Alphamab Oncology and 3D Medicines, was approved in China in 2021 for the treatment of microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR) advanced solid tumors.¹⁵⁶ As a programmed cell death ligand 1 (PD-L1) inhibitor, the molecule belongs to the class of immune checkpoint inhibitors. Envafolimab is composed of a humanized, camel-derived VHH fused to the hinge and Fc region of an IgG1¹⁵⁷ (Figure 3).

Ozoralizumab

Ozoralizumab (Nanozora) was first approved in 2022 in Japan for rheumatoid arthritis (RA) inadequately managed by currently available treatments. The compound was initially developed by Ablynx (a Sanofi affiliate) but later licensed to and further developed by Taisho Pharmaceuticals in Japan.¹⁵⁸ Ozoralizumab potently neutralizes tumor necrosis factor (TNF), a pleiotropic and proinflammatory cytokine playing a prominent role in the pathology of several autoimmune diseases, for instance, RA, psoriasis, or Crohn’s disease.^159–161 From a design perspective, ozoralizumab is a trivalent bispecific antibody composed of two humanized anti-TNF VHHs, as well as a humanized VHH binding to HSA for half-life extension with the tandem orientation VHH_TNF-VHH_HSA-VHH_TNF¹⁶² (Figure 3). The individual building blocks are separated by two nine amino acid flexible linkers (Gly₄SerGly₃Ser).¹⁶³

Ciltacabtagene autoleucel

Ciltacabtagene autoleucel (CARVYKTI) received its first approval in the USA in 2022 for the treatment of adult patients with relapsed/refractory multiple myeloma after at least four lines of therapy.¹⁶⁴ Ciltacabtagene autoleucel is a CAR T cell therapy co-developed by Legend Biotech and Johnson & Johnson.¹⁶⁵ Here, the patient’s T cells are harvested and genetically edited ex vivo with artificial receptors enabling T cell activation after antigen recognition, which subsequently are re-infused into the respective patient. The therapy is directed against B cell maturation antigen (BCMA), which is preferentially expressed on mature B cells. Its overexpression is associated with multiple myeloma.¹⁶⁶ The extracellular BCMA-targeting region of the CAR of ciltacabtagene autoleucel contains two different (biparatopic) BCMA-targeting VHH domains^167,168 arranged in tandem and separated by a flexible linker, enabling high binding avidity (Figure 3). The signaling portion of the CAR construct consists of a human CD8α hinge and a transmembrane region, followed by a CD137 (4-1BB) costimulatory cytoplasmic domain that is fused to a human CD3ζ cytoplasmic region.

Sequence features of clinical sdAbs and VHHs

To provide insights into the properties of sdAbs including VHHs that have progressed to the clinic, we queried TheraSAbDab¹³⁵ (version 19.02.2025) and identified 46 variable sdAb domains. Manual literature curation added three further VHH domains (two from ciltacabtagene autoleucel and one from netakimab), yielding a total of 49 sdAb domains incorporated into clinical-stage or marketed therapeutics. For each sdAb, we performed in silico analyses comprising sequence clustering (by CDR3 and full-length sdAb), human-likeness (closest human IGHV/IGHJ match and identity for full variable domain and framework region only), detection of non-canonical cysteines, potential N-glycosylation sites in CDRs and frameworks, and computed isoelectric points (pI) for the full sdAbs and CDRs (Figure 4). Clinical sdAbs span a moderate to high human-likeness (58.7–89.1% identity to the nearest human germline). As expected, framework identities are generally higher than full-domain identities. Of note, the sdAbs with the highest human‑likeness originate from transgenic animal platforms engineered to produce human sdAbs: porustobart’s anti‑CTLA‑4 paratope derives from a transgenic mouse HCAb platform that yields fully human HCAbs,¹⁶⁹ and surovatamig’s anti-CD19 paratope was originally discovered using Teneobio’s UniRat transgenic rat platform, which expresses fully human heavy chain antibodies (UniAbs).¹⁷⁰ Sequence clustering revealed considerable re-use of sdAb building blocks across different multispecifics, either identically or with minor sequence changes. This modularity is most evident for HSA binders, where identical or highly similar sequences occur across ozoralizumab, sonelokimab, vobarilizumab, brivekimig, lunsekimig, isecarosmab, gocatamig, and podentamig, reflecting a common strategy to prolong exposure by engaging endogenous HSA and leveraging neonatal Fc receptor (FcRn) recycling to reduce renal clearance.¹⁷¹ Likewise, identical anti-PD-L1 paratopes are found in envafolimab and erfonrilimab, underscoring the role of validated sdAbs as interchangeable components across different drug discovery and development programs. The most common target classes include HSA (half-life extension), immune checkpoints (PD-L1, CTLA-4), and cytokines (TNFα, IL-17A/F, IL-13). Notably, 6 of 49 sdAbs carry an additional disulfide bond that links CDR3 to a cysteine in CDR1 (IMGT 38) or FR2 (IMGT 50 or 55). As outlined in the general architecture section and Figures 1(G) and (H), these linkages are enriched at species-biased positions and help stabilize and shape the CDR3 loop. Notably, no N‑glycosylation motifs (N‑X‑S/T) are present in either CDRs or frameworks in this clinical set. Clinical sdAbs are frequently basic: 34/49 (≈69%) have pI (VHH) >7.5 (blue in Figure 4). FR2 hallmark motifs are dominated by FERF (15/49; 30.6%) and VGLW (12/49; 24.5%), with the remaining 22 motifs distributed across a diverse set (e.g., FERG, FERR, FGLG, YQRL/YQRW, YERL, VGPW).

Figure 4.

Sequence features of clinical and marketed sdAbs (n = 49). Sequences were collected from TheraSAbDab (version 19.02.2025). IMGT numbering was used for region delineation and for FR2 hallmark positions. “CDR3 (85% id)” and “VHH (100% id)” report cluster identifiers assigned at 85% sequence identity for CDR3 and at 100% identity for the full‑length sdAb, respectively; the numeric entry is the cluster id, and repeated IDs across therapeutics indicate re‑use of identical (full VHH) or clustered (CDR3) sequences. “Most similar germline” lists the nearest human IGHV/IGHJ pair, while “SEQ‑ID VHH” and “SEQ‑ID fw” give percent identity over the full variable domain and over frameworks only (IMGT FR1–FR4), respectively. PTMs denote post-translational modifications, either “non‑standard Cys” indicating the predicted presence (2) or absence (0) of cysteines beyond the conserved intradomain cysteines; “N‑GLYCOSYLATION” indicates absence of the canonical N‑X‑S/T motif (X ≠ pro) for all sdAbs scrutinized. The isoelectric points “pI VHH” and “pI CDR” are computed for the full VHH and for the combined CDRs; cells with pI > 7.5 are colored blue, otherwise light red. “hm” lists the FR2 hallmark motif at IMGT positions 42, 49, 50, and 52.

Antibody engineering harnessing camelid-derived sdAbs as building blocks

Since VHH domains do not associate with light chains, and hence function as autonomous paratopes, these sdAbs can be readily integrated into an unprecedented diversity of Ab architectures, involving multispecific or multivalent designs (Figure 5(A)).^172–174 This is exemplified by the fact that all five marketed VHH-harboring therapeutics comprise different Ab formats (Figure 3), as described above. It has been shown that the Ab format and ultimately paratope valencies, as well as the spatial orientation of individual antigen-binding sites within the overall architecture, significantly determine the functionality. In this regard, VHHs afford the benefit of a ‘plug-and-play’ approach for generating and testing different Ab formats to identify suitable designs that fulfill a prescribed mode-of-action.

Figure 5.

Antibody design examples harnessing VHHs as building blocks. (A) Tandem design architectures in a beads-on-a-string manner enable the facile incorporation of different paratope valencies and specificities into a given molecule. Also, other payloads such as cytokines can be integrated in this manner. (B) Due to the absence of a light chain, VHHs can easily be fused to a pre-existing IgG via a peptide linker to generate bispecificity. (C) The use of a heavy chain heterodimerization technique allows for the generation of bi- and multispecific VHH-based IgG-like bi- and multispecifics with varying valencies. (D) For half-life extension, VHHs can either be fused to an IgG Fc region or to an HSA binder. Figure generated with bioRender (www.biorender.com).

VHHs can be easily combined with already existing and well-characterized paratopes to engineer bifunctionality or even multifunctionality.^{78,79,175–181} To this end, the sdAb paratope (or several different ones) can simply be fused via a flexible linker to each of the free N- or C-termini of the parental antibody to generate a bispecific Ab (bsAb) that enables bivalent targeting of both paratopes (Figure 5(B)). If monovalent binding of one or even both paratopes is required, a heavy chain heterodimerization technique can be used, such as the knob-into-hole (KiH) or the strand-exchange engineered domain technology (SEED) (Figure 5C).^182–185 An important consideration is that both classical VH domains within Fab-based paratopes and VHH domains have their N-terminus positioned near CDR3, which is typically the most critical region for antigen binding. Consequently, the orientation and positioning of VHHs in multivalent or multispecific fusions can significantly impact binding, e.g., via steric hindrance, with some constructs showing marked preferences for N-terminal versus C-terminal positioning.^{77,176,180,181} These positional effects potentially affect the accessibility of the paratope when a VHH is placed C-terminal to another protein domain, and appropriate linker design can help mitigate suboptimal positioning effects.

sdAb-only based molecules typically display a short half-life, requiring frequent administration into patients.^148,186 In order to prolong the half-life of VHH domains, sdAbs can be genetically fused to the IgG Fc domain enabling recycling via FcRn (D, Figure 5(d) left). This strategy has been applied for the engineering of envafolimab, which is administered subcutaneously and has a first-dose half-life of 14 days and a steady state half-life of 23 days ¹⁵⁷. Alternatively, a commonly applied strategy to achieve favorable pharmacokinetics relies in fusing the therapeutic VHH ‘payload’ to an sdAb that targets HSA¹⁷¹, eliciting a half-life extension via FcRn in a similar manner to IgG Fc fusions (Figure 5(d), right). Ozoralizumab, a bivalent TNF blocking VHH therapeutic that incorporates such an HSA-directed sdAb is given subcutaneously and displays a half-life of approximately 18 days.¹⁵⁹ In fact, multiple VHH-based bispecific molecules recently evaluated in clinical trials integrated one of the half-life extension strategies discussed.¹⁸⁰

Future perspectives

The therapeutic landscape of VHH-based biologics continues to expand rapidly. While the first wave of clinical VHH-derived molecules mainly encompassed blockers, either for soluble disease mediators or membrane-bound receptors as well as ‘binding arms’ directed against tumor-associated antigens (TAAs), the field has clearly matured beyond the five currently approved therapeutics, with novel modes of action on the horizon. In this respect, multivalent VHH-based agonists against TNF receptor superfamily members have progressed into clinical development. Ozekibart, for instance, is a tetravalent (VHH–VHH)-Fc fusion protein that targets and clusters death receptor 5 (DR5), eliciting strong DR5 agonism eventually resulting in tumor cell apoptosis.^187,188 Moreover, INBRX-106, a hexavalent (VHH-VHH-VHH)-Fc Ab derivative, targets and agonizes OX40, a costimulatory receptor important for T cell survival and differentiation.¹⁸⁹ Beyond, VHHs are increasingly being incorporated into next-generation treatment modalities. These include but are not limited to ADCs, where VHHs are thought to enable enhanced tumor penetration and ultimately, payload delivery, and CAR-T cell therapies with multifunctional VHH-based targeting receptors, with ciltacabtagene autoleucel representing the first approved VHH-based CAR-T cell therapy.^{164,190–192} In the following section, we briefly discuss some emerging therapeutic mechanisms of action that are mediated by sdAbs.

VHHs as building blocks for effector cell redirection

VHHs (and sdAbs in general) are valuable building blocks for effector cell redirection.^193,194 For instance, VHHs have been generated for redirecting and engaging γδ T cells by targeting the γδ T cell receptor as well as a TAA in a bispecific manner. These γδ T cell engagers have been developed for different tumor indications ranging from hematologic malignancies to solid tumors, by targeting different TAAs such as CD1d, CD33, CD123, or EGFR.^195–199 VHHs have also been described targeting CD3 for the construction of conventional T cell engagers.^200–202 In addition to redirecting T cells, VHHs have been used for the construction of natural killer (NK) cell engagers. Prominent examples are (next generation) TriKEs (Tri-specific Killer Engager molecules), which are based on a VHH targeting CD16 for NK cell redirection, an IL-15 moiety, and a TAA-specific paratope that can either be a single-chain variable fragment (scFv) or VHH.²⁰³ Furthermore, VHHs targeting Natural Cytotoxicity Receptors NKp30 and NKp46 were generated for the construction of potent NK cell engagers, and it has been shown that the spatial orientation of individual paratopes as well as paratope valencies significantly impact tumor cell killing efficiencies.^78,175,177

VHHs as cytokine mimetics

Cytokines are (typically) secreted signaling proteins essential for regulating inflammation, immune responses, cell growth, tissue repair, and apoptosis.²⁰⁴ Their potent immunomodulatory properties have led to the approval of several cytokines for treating diseases such as cancer, multiple sclerosis, anemia, neutropenia, or thrombocytopenia.^205,206 However, the therapeutic use of cytokines is limited by pleiotropic, context-dependent effects, short half-life, and suboptimal biodistribution when administered systemically.^207,208 Moreover, dose-limiting toxicities might be a consequence, in part due to the initiation of cytokine cascades.²⁰⁹

In addition to engineering the cytokines themselves, an alternative strategy to fine-tune a given biological functionality relies on the generation of cytokine mimetics, also referred to as surrogate agonists.²¹⁰ Here, bi- or multispecific sdAbs are commonly used to target a given (most often) heterodimeric cytokine receptor to trigger receptor agonism. As such, the cytokine functionality can be highly modulated, for instance, in terms of signaling strength or cell subset bias.^77,211 To this end, different paratope combinations for the respective receptor subunits can be screened in a bi- or multispecific format inducing distinct receptor geometries, eventually resulting in differential downstream signaling. Moreover, the use of VHHs or sdAbs in general allows for engineering paratope valencies, as well as the spatial orientation of individual paratopes within the overall design architecture to fine-tune the functional effect.⁷⁹ sdAb-based cytokine mimetics have been described for various cytokines such as IL-2, IL-10, IL-12, IL-18 or type-I interferons.^{77,79,211–213} Additionally, artificial cytokine receptor agonists were generated, as well as surrogate agonists with conditional activity, clearly emphasizing the plethora of engineering options of sdAbs to tailor-make cytokine biology.^211,214

VHHs as allosteric modulators

Allostery is a fundamental mechanism of protein regulation, enabling modulation of activity through conformational and dynamic changes triggered at sites distant from the active site. While traditionally explored in the context of small molecules, antibodies, particularly VHHs, are now emerging as powerful allosteric modulators.²¹⁵ By engaging epitopes distinct from native ligand-binding regions, VHHs can reshape conformational landscapes of their targets, yielding therapeutic profiles that differ from conventional inhibitors.

This potential is well illustrated by studies on the epidermal growth factor receptor (EGFR). Allosteric VHHs were discovered to inhibit receptor activity by stabilizing signaling-incompetent conformations rather than blocking ligand engagement.^216,217 Mechanistic insights were obtained by integrating structural and functional assays with computational models of allosteric communication,²¹⁸ which revealed how antibody binding perturbs distal regions of the receptor. These combined approaches illustrate how experimental and computational methods can converge to rationalize and ultimately guide the design of allosteric antibodies. On a different target, VHHs acting as silent and positive allosteric modulators of the α7 nicotinic acetylcholine receptor were reported by Li and colleagues.²¹⁹ Importantly, VHH-mediated allosteric modulation has now entered translational neuroscience: a recent study demonstrated that a VHH acting as a positive allosteric modulator of the metabotropic glutamate receptor mGlu2 rescued behavioral deficits associated with NMDA receptor hypofunction, a model of cognitive impairment in schizophrenia.²²⁰

Beyond their therapeutic implications, VHHs have also become indispensable allosteric tools in structural biology. Steyaert and colleagues pioneered their use to stabilize G protein – coupled receptors (GPCRs) in discrete conformational states, enabling landmark structures of the β2-adrenergic receptor bound to G proteins.²²¹ Building on this foundation, Manglik and coworkers developed VHHs such as Nb80 and Nb60, which trap distinct β2-adrenergic receptor states and reveal mechanisms underlying agonist efficacy.²²² Additional examples are from Scheerer’s group, who has further contributed to GPCR structural biology, including active-state structures of the melanocortin-4 receptor in complex with the heterotrimeric stimulatory G protein Gs, where a stabilizing VHH (Nb35) was essential to trap the receptor-G protein assembly for high-resolution cryo-EM.²²³

Together, these advances establish VHHs as versatile allosteric modulators: therapeutic agents capable of tuning protein activity, and research tools that illuminate the dynamic nature of complex signaling proteins.

Oral delivery and alternative administration routes

Due to their intrinsic physicochemical stability, VHHs are potentially attractive modalities for oral delivery targeting gastrointestinal (GI) pathogens in humans and animals.²²⁴ In this respect, several approaches to further optimize resistance to extreme pH, heat, and proteolysis, but also through formats tailored for avidity and target coverage, have been described aiming to improve the functionality of VHHs within gastric and intestinal environments. In this regard, different multivalent constructs, such as bivalent VHH-VHH and Fc/secretory IgA fusions, as well as synthetic pentameric VHH-verotoxin subunit B fusions, have shown enhanced pathogen agglutination in animals, improving control of Campylobacter, Shiga toxin-producing Escherichia coli, and related pathogens.^225–227 To retain protein functionality after passaging the GI tract, Arbabi-Ghahroudi and colleagues generated protease-resistant VHHs inhibiting the motility of Campylobacter jejuni by combining disulfide-bond engineering with a panning approach of an error-prone polymerase-chain reaction library in the presence of GI proteases.²²⁸ Beyond reducing bacterial infection risk, orally administered VHHs have also been evaluated for neutralizing viral GI infections as well as in the context of autoimmune-related GI diseases. In this regard, a VHH specific for a capsid protein of rotavirus demonstrated protection of piglets against diarrhea and a significant reduction of stool output in infants.²²⁹ Furthermore, Ota and colleagues presented an engineered protease-stable VHH that targets the IL-23 receptor (IL-23 R) for the treatment of inflammatory bowel diseases which exhibited effective pathway inhibition in murine models and sustained IL-23 R inhibition in the circulation of non-human primates following oral administration.²³⁰

Despite numerous promising examples, preserving the stability of orally administered antibody-based therapeutics remains a major obstacle. Consequently, alternative delivery routes have been explored. One attractive approach is the in situ delivery of VHHs using organisms classified as “generally regarded as safe” (GRAS), such as Lactobacillus/Lactococcus strains or cyanobacterium Arthrospira platensis (spirulina), as well as the use of plant hosts.^231–236 In this regard, Tokuhara et al. showed that mice receiving genetically modified rice delivering multimeric anti‑rotavirus VHHs achieved efficacious protection against rotavirus infection even after more than a year of storage and boiling prior to administration.²³⁷

Although substantial progress has been made regarding oral application of VHHs and alternative delivery methods, successful implementation will depend on precise GI dosing and release, scalable low‑cost manufacturing, improved stability formulations, and clear regulatory pathways for GMO‑containing products.²²⁴

Funding Statement

The author(s) reported there is no funding associated with the work featured in this article.

Disclosure statement

All authors are affiliated with Merck Healthcare KGaA, Darmstadt, Germany. Besides, this work was conducted in the absence of any further commercial interest.

Data availability statement

This study includes no data deposited in external repositories.