Engineering AvidCARs for combinatorial antigen recognition and reversible control of CAR function

T cells engineered to express chimeric antigen receptors (CAR-T cells) have shown impressive clinical efficacy in the treatment of B cell malignancies. However, the development of CAR-T cell

therapies for solid tumors is hampered by the lack of truly tumor-specific antigens and poor control over T cell activity. Here we present an avidity-controlled CAR (AvidCAR) platform with

inducible and logic control functions. The key is the combination of (i) an improved CAR design which enables controlled CAR dimerization and (ii) a significant reduction of antigen-binding

affinities to introduce dependence on bivalent interaction, i.e. avidity. The potential and versatility of the AvidCAR platform is exemplified by designing ON-switch CARs, which can be

regulated with a clinically applied drug, and AND-gate CARs specifically recognizing combinations of two antigens. Thus, we expect that AvidCARs will be a highly valuable platform for the

development of controllable CAR therapies with improved tumor specificity.

CARs are synthetic proteins consisting of antigen-binding domains linked to T-cell-activating signaling domains (Fig. 1a)1. T cells genetically engineered to express CARs (CAR-T cells)

efficiently eliminate antigen-expressing target cells, which is reflected by the impressive efficacy of CD19-directed CAR-T cell therapies in patients with B cell malignancies2. A major

challenge in translating this success to other malignancies is the fact that CAR antigens are almost exclusively tumor-associated antigens (TAAs), which are not truly tumor specific. Since

most TAAs are also expressed in vital healthy tissues, efficient TAA targeting is expected to result in unacceptable on-target/off-tumor toxicity, which has indeed been observed in several

clinical studies3,4,5,6,7. Ultimately, this could preclude increasing the efficacy of CAR-T cell therapies to levels required for sustained remission of solid tumors. The problem of

on-target/off-tumor toxicity is further aggravated by the fact that the activity of CAR-T cells is difficult to control in a reversible manner.

a Schematic of a conventional CAR (high-affinity CAR) typically containing a high-affinity antigen-binding domain (usually a single-chain variable fragment, scFv) fused via a dimerizing

hinge region (derived from, e.g., CD8α) to the cytoplasmic domains of a co-stimulatory receptor (mostly 4-1BB or CD28) and CD3ζ. Due to high-affinity binding, monovalent interaction with the

target antigen is sufficient for CAR activation. b Schematic of a low-affinity CAR in which a low-affinity antigen-binding domain is fused to the same conventional CAR backbone shown in

(a). We hypothesized that such low-affinity CARs require bivalent interaction with the antigen for efficient activation. We further hypothesized that this avidity-based concept (i.e., the

dependency on bivalent antigen recognition) can be exploited for various applications: c for generating ON-switch AvidCARs whose function is regulated by a dimerizing small molecule; d for

constructing AND-gate AvidCARs which are dimerized by a soluble antigen B, leading to avidity-based recognition of antigen A; e, f and for the design of heterodimeric AND-gate AvidCARs

specifically recognizing a combination of two surface bound antigens with two different recognition domains, respectively. This AND-gate concept can be realized in a nonswitchable format by

constitutive CAR heterodimerization (e) or in a switchable version by inducing heterodimerization with a small molecule (f).

To address the lack of control and tumor specificity associated with CAR-T cells, several promising strategies have been developed including CARs for combinatorial antigen

recognition8,9,10,11,12,13,14, affinity-tuned CARs to minimize killing of healthy cells expressing low levels of target antigen15,16,17,18,19,20 and CARs whose function can be regulated by

administration of small molecules or soluble proteins10,21,22,23,24,25,26,27,28,29. Despite these important advances, there is still a need for improvement. For example, currently available

systems for the control of CAR-T cells are either based on small molecules which are suboptimal for clinical application or on suicide switches which irreversibly destroy the CAR-T cells.

Furthermore, previously described AND-gate CARs suffer from low specificity12,13,14 or their inability to differentiate between a double-positive cell and two single-positive cells, which

express the antigens A and B complementarily and are located in close proximity to each other21,30. That is, in this SynNotch strategy, expression of the CAR can be induced by healthy cells

expressing antigen A, enabling recognition of nearby healthy cells expressing antigen B. Furthermore, the SynNotch system is based on xenogeneic and thus potentially immunogenic proteins.

One critical parameter defining the potency and function of CARs is the affinity of their antigen-binding domains. Importantly, the interaction with a target antigen not only depends on the

affinity of an individual antigen-binding domain, but also on the valency of the interaction. That is, a multivalent interaction amplifies the individual affinities through a phenomenon

known as avidity. Many types of natural immune reactions are based on low affinities, which are amplified by avidity effects, such as antigen binding by IgM during early humoral immune

responses. However, the influence of avidity effects on CAR function has mostly been neglected in the CAR field so far, even though most currently used CARs are based on naturally dimeric

components31,32,33.

Here, we demonstrate that this aspect can be exploited for the integration of inducible and logic control functions into CAR molecules. We show that it is possible to generate

avidity-controlled CARs (AvidCARs) which are highly potent and dependent on bivalent antigen engagement (Fig. 1). These AvidCARs are based on two main design principles: controlled CAR

dimerization and low-affinity antigen binding. We exemplify the potential of the AvidCAR platform (1) by generating a controllable ON-switch CAR that can be regulated by CAR homodimerization

with a clinically applied small molecule (Fig. 1c); (2) by constructing an AND-gate CAR that is triggered only in the presence of both a surface molecule on tumor cells and a soluble

protein found in tumor stroma (Fig. 1d); and (3) by generating a heterodimeric AND-gate CAR which enables the combinatorial recognition of two different antigens exclusively when

co-presented on the surface of the same cell. This latter AND-gate AvidCAR is designed either in a constitutive (Fig. 1e) or in a switchable version regulated by a dimerizing small molecule

(Fig. 1f). Finally, we also investigate the suitability of different antigen-binding modules and various co-stimulatory domains for the design of efficient AvidCARs. Together, our study

illustrates the potential of exploiting avidity effects and demonstrates that the AvidCAR platform is a versatile concept for inducible and combinatorial CAR control.

Since a critical design principle of AvidCARs is the control over CAR dimerization, the usage of CAR components potentially causing uncontrolled dimerization or oligomerization needs to be

avoided. Given the recent report on scFv-mediated CAR tonic signaling (probably caused by scFv-mediated clustering)34, we looked for an antigen-binding domain based on a single protein

domain, which prevents intermolecular dimerization as was observed with several scFvs35,36,37. Therefore, we chose the human epidermal growth factor receptor (hEGFR)-specific binder E11.4.1,

which was previously engineered based on the single-domain protein rcSso7d38. As described above, we hypothesized that the construction of an AvidCAR also requires the incorporation of

low-affinity binding domains (Fig. 1b). For this purpose, we reduced the affinity of the high-affinity binder E11.4.1 by performing an alanine scan, resulting in eight different E11.4.1

mutants. To determine the affinities of those mutants, they were expressed as monomeric proteins (Supplementary Fig. 1a) and titrated on Jurkat cells expressing truncated hEGFR (hEGFRt,

Supplementary Fig. 1c). These titration experiments yielded an affinity of 13 nM for the high-affinity binder E11.4.1, which is in close agreement with the previously published value (17 

nM)38. Among the mutants we identified two (E11.4.1-G25A and E11.4.1-G32A) with substantially decreased, yet measurable affinities of 125 and 877 nM, respectively (Fig. 2a). Complementary

affinity determination by surface plasmon resonance (SPR) analysis with the entire extracellular domain of hEGFR yielded highly comparable Kd values (Supplementary Fig. 1b).

a Flow cytometric determination of the affinities of hEGFR-specific binder mutants. Soluble binder proteins (sfGFP fused to E11.4.1 or mutants thereof) were titrated on hEGFRtpos Jurkat

cells. Shown are the median fluorescence intensities (MFI) of bound E11.4.1-variants (three independent experiments; mean values ± SD). Colored lines indicate the high-, intermediate-, and

low-affinity binders used in the following experiments. The insert in the diagram shows the structure of the original Sso7d binder scaffold with its mutated binding surface in pink (PDB

1SSO). b, c Dependence of CAR function on binder affinity and CAR dimerization. b Schematic of the tested CARs. Selected rcSso7d-based binders were fused to a CAR backbone in which the CD8α

hinge contained either two cysteine- or two serine residues instead (see SEQ-IDs 1 and 2 in Supplementary Data 1, respectively). c The figure shows the cytolytic activity of mock-T cells (no

CAR) and T cells expressing the monovalent (Ser-BBz) and bivalent (BBz) CARs with different affinities against hEGFRtpos Jurkat target cells. (Luciferase-based assay; four independent

experiments with four different donors; **p