Multiple Myeloma (MM) is a cancer of mature, antibody-producing B cells (plasma cells) that grow uncontrollably in the bone marrow. They thereby disturb the bone formation process and hematopoietic equilibrium, leading to characteristic bone lesions, hypercalcemia and general anemia (1). They also produce excessive amounts of dysfunctional immunoglobulin molecules (M- or paraprotein), causing renal problems (2). Worldwide, MM accounts for 14% of all hematological cancers, making it the third most observed one. It is considered a treatable but incurable malignancy, with a 5-year overall survival rate of 54% (3).

Standard induction therapy for newly diagnosed MM patients usually comprises a combination of immunomodulatory drugs (e.g., lenalidomide) and corticosteroids (e.g., dexamethasone), most often combined with proteasome inhibitors (e.g., bortezomib) (1, 4). This regimen is supplemented with anti-CD38 monoclonal antibody (mAb)-therapy (daratumumab) in some cases. For all eligible patients, induction therapy is followed by autologous stem cell transplantation (SCT). Subsequent maintenance therapy may be bortezomib- or lenalidomide-based (4). Although these regimens are often initially successful, relapse with an increased tolerance to previous treatment regimens is a commonly observed phenomenon. Upon relapse, combination treatment options are diverse and adjusted to the patient specifically (2).

It is established that the MM tumor microenvironment is highly immunosuppressive, among others due to disturbed cytokine production by malignant, stromal and immune cell populations and an outbalanced programmed death-1 (PD-1)/PD-1 ligand (PD-L1) immune checkpoint axis (5). This leads to a malfunctioning innate and adaptive immune environment, involving both myeloid and lymphoid actors. Increased understanding about these aberrances, together with observed graft-versus-myeloma effects in early allogenic SCT trials, highlight the potential added value of immune therapy for MM (6). Forms of adoptive cell transfer currently being evaluated for the treatment of MM include T-cell receptor (TCR)-modified T-cells, (allogenic) chimeric antigen receptor (CAR)-T cells, (CAR)-natural killer (NK) cells and tumor-infiltrating lymphocytes. Furthermore, immune checkpoint inhibitors, bi-specific T-cell engagers and cancer vaccination strategies are under (clinical) evaluation (6). Of these, adoptive cell transfer, and more particularly CAR-T cell therapy holds great promise due to the unseen curative outcomes observed in other hematological malignancies, namely high-grade lymphomas and leukemias (7).

CAR-T cells are patients’ own T-cells that are genetically modified to express a transmembrane CAR (8). This receptor can recognize a tumor antigen, expressed on the surface of tumor cells, with its extracellular domain. Classically, a monoclonal antibody (mAb)-derived single-chain variable fragment (scFv) is incorporated to achieve tumor antigen recognition (9, 10). Upon antigen encounter, the intracellular T-cell co-stimulatory (most often CD28- and/or 4-1BB-derived) and CD3ζ T-cell activation domains are responsible for engaging a cytotoxic T-cell response toward the malignant cell. After ex vivo modification and expansion of these patient-derived T-cells, they are administered back to the patient, where they thus act as a living drug (11).

Historically, most of the evolution of CAR design has been focused on the optimization of the intracellular portion of the receptor, in order to achieve maximal T-cell activation (10). The extracellular antigen-binding moiety has received less attention, as many mAbs against relevant tumor antigens are (clinically) available. However, the use of the artificial mAb-derived scFv format has been linked to some limitations in CAR-T cell efficacy (9, 12). Firstly, scFvs lack natural stability and therefore need to be artificially linked. This instability has been associated with aggregation, antigen-independent (tonic) signaling and subsequent premature T-cell exhaustion (13, 14). Secondly, their non-human nature limits CAR-T cell persistence in vivo, as anti-CAR immune responses have been observed in patients upon relapse (15). Thirdly, because these molecules are usually derived from clinically validated mAbs, there is often no step of (structural) optimization of the scFv domain included in the development of these CARs, as a scFv selection procedure is not required (9, 11). In recent years however, the importance of optimizing this extracellular protein domain has become increasingly recognized, and alternatives to the classical scFv format are rapidly emerging, as reviewed elsewhere (9). One example of such alternative CAR design incorporates a Variable Heavy domain of Heavy chain (VHH) molecule, derived from camelid-found heavy-chain-only antibodies (HCAbs) as an antigen-binding moiety. HCAbs compare to mAbs by lacking light chains and constant heavy-1 (C_H1) domains (16). The antigen-binding part therefore consists of only one protein domain, the VHH. Evolutionarily, this has ensured that the VHH domain behaves as a monomer and is therefore intrinsically more stable compared to a scFv (16). It also allows for more straightforward VHH screening and selection, as the availability of (immune) libraries is more evident (16). Furthermore, VHH immunogenicity is expected to be lower in comparison to scFvs (17), as there is a high sequence resemblance with human VH sequences of family III. At this point, no reports of anti-VHH CAR immune responses with a neutralizing effect on CAR-T cell therapy have been reported from clinical trials (18), and VHH humanization protocols are available (9, 19, 20).

In the last decades, CD19-directed CAR-T cell therapy has astonished the medical landscape with its curative outcomes in certain forms of leukemia and lymphoma (21). This raises hope for its potential in other hematological malignancies, including MM (11). However, as CD19 is usually not expressed by plasma cells, clinical success of these well-established CAR-T cell products in MM has been limited (11). Higher success rates have been achieved with B-cell maturation antigen (BCMA)-targeted CAR-T cells, with currently two FDA-approved CAR-T cell products (i.e., idecabtagene vicleucel and ciltacabtagene autoleucel) as a result (7). Although the observed overall response rates are high, a progression-free survival period of more than 1 year is observed in under 50% of patients treated (22, 23). The most commonly reported causes for this are antigen shedding and CAR immunogenicity, which can lead to anti-CAR immune responses with a therapy-neutralizing effect (11, 18, 22). Hence, other target proteins are under investigation. Of these, SLAMF7/CS1, GPRC5D, CD138 and CD38 are showing different clinical success rates, while data for newer tumor antigens like CD70, NKG2DL and κ-light chain are to be expected (11). Particularly of interest as an alternative cancer antigen for MM CAR-T cell therapy is CD2 subset 1 (CS1, CD319 or signaling lymphocytic activation molecule family member 7 [SLAMF7]). Indeed, CS1 is highly expressed by > 95% of both healthy as well as malignant plasma cells (11, 24), and expression is retained after multiple lines of therapy (25). Expression on healthy tissue is lower, limited to hematopoietic cell lineages (NK, T, B, and dendritic cells; monocytes and macrophages) and absent on hematopoietic stem cells (11, 26).

These observations have provided the rationale for this study, which consists of developing a new form of CAR-T cell therapy for MM, targeting CS1. To this end, we aimed to develop an optimized CAR through variation in the antigen-binding part of the receptor, as well as determine which specific parameters of this extracellular part are crucial to achieve potent CAR-T cell activation. The influence of differences in affinity, cell-binding capacity, epitope location, cell-to-cell distance and binding orientation were examined as possible determinant factors.

In this study, a side-by-side comparison was made of different VHHs, their characteristics and their ability to lead to T-cell activation when incorporated into a CAR. This was evaluated in the context of MM, a highly relevant cancer type for the optimization of CAR-T cell therapy. We opted to target CS1, often mentioned as a promising next-generation MM antigen, and generated a panel of CS1-specific VHHs, which were fully characterized as soluble compounds and as components of VHH-based CARs.

Although the affinity proved to be predictive for the in vivo MM tumor-tracing ability of the VHHs to a certain extent, no link was discovered between the affinity and the ability of CAR-T cell activation. Remarkably even, the relatively weak binding VHH-2 proved to be among the most capable CAR-T cell activating VHHs, while – in line with its affinity for CS1 -, MM cell binding and in vivo MM tumor tracing capacity are inferior. A possible link between the targeted epitope of the VHH and CAR-T activation was suggested by competition CAR-T cell activation studies with excess soluble VHH added to the co-culture, but this relationship could not be confirmed by in silico structure modeling of the VHH-antigen interactions. The latter was used to estimate whether the intercellular distance and/or the orientation of CAR-target binding influence the T-cell activation potential. Here too, no cause- and-effect relationship was discovered. Of note is that these conclusions were drawn based on the read-out of an eGFP-based reporter T-cell line.

In particular, we have worked with 2D3 cells to assess the influence of the antigen-recognizing domain on T-cell activation. This is an immortalized cell line that is genetically engineered to specifically map (differences in) T-cell activation kinetics (40, 41). In order to make a critical comparison of different binding domains while excluding all other possible factors- biological or other -, a universal screening platform was required. 2D3 cells served that purpose, as they allow the variation of one CAR component solely. Yet, they come with the limitation that no direct estimate can be made of the in vivo therapeutic effect and/or possible cytotoxicity of the designed CARs; which are highly important parameters when a clinical translation of the CAR-T cells is envisaged. To further develop the VHH-CARs for therapeutic purposes, in vivo evaluation of the compounds is primarily required. However, to estimate the therapeutic potential of the designed CARs in a highly immune-complex disease such as MM, this must be investigated in clinically relevant animal models. These should be fully immune competent and representative of human disease in terms of tumor cell localization, behavior, disease progression and immune system involvement. Such models exist for murine MM disease (42, 43), but are currently not available for the human variant.

Indeed, by making use of a reporter cell line, our data can only state that it is of utmost importance to evaluate a (sub)library of different antigen-binding compounds side-by-side, while already in the structure of a CAR. Therefore, this should be regarded as a proof-of-concept comparison study, aimed at conveying a fundamental message about the importance of multiple candidate-screening when designing CAR-T cells.

In summary, our data simultaneously demonstrate the importance of carefully selecting the extracellular part of a CAR and the serendipity to which this design is subject. Multiple classical VHH parameters were investigated, including affinity, cell binding potential, in vivo behavior, epitope location, intercellular T-cell to target cell distance and orientation of the CAR-antigen interaction. Even though important differences were observed in the extent to which different VHHs can activate a CAR-T cell, none of the VHH parameters investigated had a predictive value for this. Therefore, the classical workflow of VHH selection does not match the needs for optimal VHH-CAR design.

Within the medical context, VHHs are mainly used as smaller, soluble, antigen-specific alternatives for the classically large mAbs (44). Therefore, it is important that the compound shows highly discriminative binding to target tissue versus non-target tissue. When certain conditions are met, such as sufficient (in vivo) specificity and (thermo)stability, this is generally obtained by compounds that display a high affinity (ideally < 10 nM) for the (cell-expressed) target protein (44, 45). The general flow of the VHH selection protocol, as described here, is considered standard, both by our research group (27, 30, 35, 46, 47), as by related research groups (48, 49), and other groups in the world (50), as well as by industry (45). It is optimized for finding highly stable, soluble pharmaceuticals with strong affinities for target proteins of all kinds, and has resulted in moieties binding different types of proteins, both extracellular (27, 35, 47) and intracellular (51). In fact, this selection procedure is optimized for identifying and selecting in vivo tracers. However, the workflow also contains a few bias points for selecting such soluble compounds. Firstly, the initial phase of the selection is performed on crude periplasmic extract, which automatically filters for the highly produced and aqua-stable VHHs in ELISA and flow cytometry. Secondly, an off-rate screening creates bias toward high-affinity compounds with a slow off-rate. Thirdly, thermostability assays and production yield measurements, as well as sequence-dependent parameters are often considered when selecting VHHs. These parameters are less relevant when applications such as VHH-based CAR-T cells are intended.

Instead, it is believed that in order to obtain maximal CAR-T cell activation, the formation of a functional immunological synapse between the T-cell and the target cell is important. The kinetic segregation model for T-cell activation states that the balance of presence of kinases (mainly Lck) versus dephosphorylases (mainly CD45) in the T-cell:antigen-presenting cell interaction zone (immunological synapse) is determinant for TCR downstream signaling and thus for T-cell activation. This balance is mainly influenced by spatial restrictions that appear upon T-cell:target cell interactions.

CD45 enzymes are abundantly present on the membrane of resting T-cells, resulting in sufficient dephosphorylation of the TCR immunoreceptor tyrosine-based activation motifs (ITAMs) to maintain a basal, inactive T-cell status. However, these molecules are relatively bulky and are excluded from the immunological synapse upon TCR-pMHC (peptide major histocompatibility complex) and CD58-CD2 cell adhesion interactions. Consequently, the ITAM phosphorylation equilibrium is outbalanced, and T-cells become activated as a result of downstream intracellular phosphorylation events (52, 53).

This model highlights the importance of the formation of a spatially correct immunological synapse, which is equally relevant in CAR-T cells, as similar signal transduction pathways are determinant for T-cell activation (53). The immunological synapse has been described to have an ideal intermembrane distance of around 15 nm in the center (where TCR-pMHC interactions are mainly taking place) to 100 nm at the edges, where granules and cytokines are exchanged and receptor-ligand pairs are the main interaction molecules (54). As the VHH molecule is anchored to the CAR via a (large) flexible linker, it is difficult to make a concrete estimate of the effective intercellular distance. However, in silico modeling of the various VHH-CS1 interactions shows that there are relevant differences between different VHH-CAR:CS1 pairs at this level. These observed differences did not translate directly to differences in CAR-T cell activation potential, and therefore it was assumed that neither location of binding, nor VHH binding orientation are predictive parameters for CAR-T cell activation.

These conclusions are based on sequence-driven computational estimates for the protein structures and interactions. It should be noted here that these are artificial intelligence-based structure interaction predictions, which are not supported by experimental data. Although the AlphaFold2 algorithm has already proven its predictive power for more evolutionary conserved protein interactions (37, 39), it is noteworthy that the program is considered slightly less powerful and therefore less trustworthy for the prediction of antigen-antibody binding. This is because less evolutionary background (on which the modeling is based) is available for these types of interactions (55). As these data do not coincide with the experimental data retrieved from the competition activation assay, one might argue on the accuracy of these modeled interactions. To precisely know the orientation of interaction, experimental data on the interaction structure would be necessary, which could for example be obtained via X-ray crystallography. This is however a technically challenging method that is not recommended to screen complete VHH libraries. All in all, our data demonstrate that the VHH:antigen interaction, as currently computationally modeled, has no predictive value for the VHH-CAR-T cell activation potential. Whether this is due to a lack of accuracy of the prediction or due to a missing link between binding orientation and activation potential, remains an open question.

Within the immunological synapse, the affinity of a TCR for its target pMHC molecule is relatively low, i.e., 1 μM to 100 μM. As such, cells with low target pMHC expression do not trigger T-cell activation and T-cell overstimulation (which may lead to premature exhaustion) is avoided (56). CARs harboring a mAb-derived scFv usually have a (sub)nanomolar affinity for their target (57). This while various studies have shown that an unnecessary increase in affinity of a CAR is more likely to lead to higher toxicity and a shorter T-cell life span (56–58).

Chmielewski and colleagues have identified a ceiling of K_D = 10^-8 M, below which the affinity no longer contributes to enhanced T-cell function (58). It has also been reported that lower affinity CARs are more capable of correctly distinguishing cancerous tumor antigen-overexpressing cells from healthy cells with normal expression levels of a tumor antigen (59). However, this is a balance of sensitivity and selectivity in which (the degree of) antigen expression is also important. Indeed, different studies in different contexts show different results. For example, an optimal CAR affinity in the micromolar range was found for target antigens ICAM-1, CD38 and HER2 (56, 60, 61). Meanwhile, CARs with a nanomolar affinity appeared optimal for tumor antigens EGFR, EGFRvIII and CD123 (62–64). This of course is also related to the avidity of the CAR-antigen interaction, which is subject to not only the affinity, but also to the expression levels of both the CAR on the T-cell surface and the antigen on the cancer cell. Clinical data suggest that the importance of generating an interaction with adequate avidity may prevail over attaining a desired affinity (65).

In any case, these studies highlight the importance of case-by-case fine-tuning of the affinity of a CAR for its target antigen, and the advantage of using the weakest possible affinity binder that is still sufficiently sensitive. Moreover, the balance between selectivity and sensitivity will be even more important in solid tumors, where antigen expression often also occurs on vital healthy tissue. In a hematological context, antigen expression is more likely to be limited to hematopoietic cell lineages, rendering selectivity less critical. All in all, these studies demonstrate that the antigen-binding portion of a CAR has a major impact on its functionality, and that case-by-case selection is needed. When classical scFvs are incorporated, this is less obvious, as they are usually adopted from existing and clinically validated mAbs. Furthermore, the lack of (immune) libraries makes it less straightforward to evaluate multiple scFvs side-by-side.

VHHs have this advantage over scFvs, as they are monomeric by nature and selection usually starts from relatively small immune libraries. Furthermore, VHH-based CARs have already proven their functionality in the clinic, in particular in the context of MM. In the 2022 FDA-approved ciltacabtagene autoleucel (cilta-cel) CAR, antigen-recognition is ensured by two (bi-epitopic) BCMA-specific VHHs (66). This is the second clinically approved CAR-T cell product for the treatment of MM. Idecaptagene vicleucel (ide-cel), also directed against BCMA, was previously approved. In the latter, antigen recognition is provided by a classic scFv format.

Despite the high initial success rates of these therapies, MM relapse remains an important problem, mainly caused by BCMA shedding. These findings demonstrate the power of CAR-T cell therapy for MM, but also its unfulfilled potential due to suboptimal choice of antigen. Therefore, other target antigens are under investigation. Of these, CS1 is of particular interest due to its retained expression after several lines of therapy and its expression pattern that is limited to hematological cell (sub)types. Several CS1-directed CAR-T cell therapies are under investigation, both in a preclinical and clinical stage (11, 67).

Our study combines both optimization points of current CAR-T cell research for MM. In particular, we use a CAR directed against the promising target CS1, and achieve this using a VHH molecule selected out of a library as being optimal in the tested CAR design. Furthermore, we have developed several CS1-specific VHHs that may be complementary to each other. Indeed, antigen expression throughout therapy is an important parameter for its success rate. The combination of therapy and diagnostic imaging is therefore an interesting concept that is increasingly gaining attention in the field of personalized medicine. Since VHHs are ideally suited as radiodiagnostic tools, it would be possible to monitor the status of CS1 expression before and during CS1-targeted (CAR-T cell) therapy. To this end, however, it is important that competition for target binding between the diagnostic and the therapeutic tool is avoided. Thus, compounds that bind a different epitope are needed. Our data lend themselves ideally to such a scenario, as the compounds that perform well in vivo do not necessarily correspond to those that provide good CAR-T cell activation, and two different compounds can be optimized according to their intended use.

Finally, it is important to note that although we demonstrate the importance of the incorporated VHH for T-cell activation, this is not the only point of optimization of CAR design and several other factors that may affect CAR-T cell behavior are under investigation (68). These include the other CAR subunits, such as the hinge and transmembrane region (69), as well as the incorporated co-stimulatory molecule(s) (70). The best-known CARs to date are second generation, implying that a single co-stimulation domain is incorporated, mostly 4-1BB or CD28-derived (9, 68). For example, third-generation CARs harbor a combination of two co-stimulatory domains and are currently under clinical evaluation (71). Fourth-generation CARs seek their added value through the local secretion of immune stimulatory cytokines, mainly IL-12 (72).

Another potential influencing parameter that is increasingly gaining attention is the composition and phenotype of the harvested and infused T-cells. As such, the ideal composition of CD4⁺/CD8⁺ T-cell fractions and the added value of eliminating regulatory T-cells are important fields of research (73, 74). Additionally, data show that younger, more stem cell-like phenotypes of T-cells may exhibit superior in vivo efficacy, owing to reduced or delayed exhaustion (75, 76). Furthermore, there are many variables in ex vivo modification protocols, genetic CAR constructs and manner of cell modification that might influence T-cell behavior (68). In particular, the benefit of reducing CAR-T cell manufacturing time is also gaining interest, as this may not only save patients with rapidly progressing disease and reduce overall costs, but could also potentially be of therapeutic added value, as reduced culturing time might result in a younger (i.e., more stem cell-like) T-cell phenotype (68, 77, 78).

All in all, there are many different variables that determine the behavior and efficacy of CAR-T cells. Our data show that although it has historically received little attention, the antigen-binding component certainly is one of them and that adequate screening is needed.

Taken together, our results demonstrate that the antigen-binding part of a CAR has a major influence on its functionality, as well as on the activation kinetics of modified CAR-T cells. This emphasizes the importance of selecting, or in the case of scFvs, optimizing this protein domain. We scrutinized several parameters of VHHs trying to link them – without success-, to a superior CAR-T cell response. Consequently, CAR-T cell research remains subject to serendipity and case-by-case selection of the optimal antigen-binding moiety is still needed. Hence, the importance of screening multiple candidate binding domains directly in a CAR configuration cannot be underestimated. In this regard, VHHs have an important advantage over mAb-derived scFvs, due to the easy availability of immune libraries.

The original contributions presented in the study are included in the article/ Supplementary Materials . Further inquiries can be directed to the corresponding author.

Ethical approval was not required for the studies on humans in accordance with the local legislation and institutional requirements because only commercially available established cell lines were used. The animal study was approved by ethical committee for use of laboratory animals, Vrije Universiteit Brussel. The study was conducted in accordance with the local legislation and institutional requirements.

HH: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Software, Visualization, Writing – original draft. FM: Data curation, Formal analysis, Investigation, Writing – review & editing. YDV: Conceptualization, Data curation, Formal analysis, Methodology, Writing – review & editing. QL: Conceptualization, Data curation, Methodology, Writing – review & editing. JP: Methodology, Writing – review & editing. PD: Data curation, Investigation, Methodology, Writing – review & editing. TDG: Data curation, Methodology, Software, Writing – review & editing. CG: Investigation, Writing – review & editing. KDV: Conceptualization, Funding acquisition, Methodology, Resources, Supervision, Writing – review & editing. KB: Conceptualization, Funding acquisition, Methodology, Resources, Supervision, Writing – review & editing. ND: Conceptualization, Funding acquisition, Methodology, Resources, Supervision, Writing – review & editing.