Multispecific Targeting with Synthetic Ankyrin Repeat Motif Chimeric Antigen Receptors

The adoptive transfer of chimeric antigen receptor–modified T cells (CAR-T) targeting single B-cell lineage markers CD19 and CD22 has induced durable remissions in some patients with B-cell–derived malignancies that have failed all conventional therapies (1–4). However, treatment failure occurs in some patients because of the outgrowth of tumor cells that have lost or express low levels of the target antigen; or, in the case of acute leukemia, that have undergone a lineage switch to a myeloid phenotype (1, 5–8). Solid tumors present an even greater therapeutic challenge because few surface molecules that could serve as CAR targets are solely expressed on tumors and not on normal tissues and because antigen expression is often more heterogeneous on solid tumors than on B-cell malignancies (9, 10). Indeed, clinical trials in which CAR-T cells targeting EGFRvIII were administered to treat glioblastoma or targeting c-MET in triple-negative breast cancer have resulted in the outgrowth of antigen-negative variants (11, 12).

In an attempt to reduce antigen escape, bispecific CAR-T cells with 2 scFv moieties separated by G₄S linkers have been designed to target two antigens simultaneously. Bispecific constructs have been described that target CD19/CD123, CD19/CD22, or CD19/CD20 in B-cell malignancies, or HER2/IL13Rα2 in glioblastoma (1, 13–17). Linking more than 2 CAR encoding transgenes in tandem can also be achieved (18), although vector size becomes a barrier for efficient gene delivery, and obtaining the correct stoichiometry to efficiently recognize all target molecules can be challenging. Furthermore, tandem antibody domains are prone to alternative association and aggregations (19).

Advances in protein engineering have led to the development of alternative antigen-binding proteins that are smaller in size, more thermodynamically stable, and less aggregation prone than antibodies (20). Designed Ankyrin Repeat proteins (DARPins) are one class of such antibody mimetic proteins with such properties and are comprised of a variable number of ankyrin repeat motifs where each ankyrin repeat consists of 33 amino acids, forming 2 α-helices followed by a β-turn (20). DARPin libraries have been generated by consensus design and by randomizing 6 surface interaction residues per repeat without compromising the stability of the hydrophobic core. A library of 2 or 3 internal repeats, flanked by N- and C-terminal hydrophilic capping repeats, has a theoretical diversity of 5.2 × 10¹⁵ or 3.8 × 10²³, respectively. Using ribosome display, practical diversities of 10¹¹ to 10¹² can be routinely reached, and with phage display about one order of magnitude less.

Library screening against many targets has enabled the selection of DARPin binders that have low nanomolar or picomolar affinity, are extremely specific and stable, and have been developed as protein therapeutics for cancer and ocular diseases (21). Previous work has demonstrated that DARPins can function as a single or bispecific recognition motifs in CAR-T cells and confer antigen-specific recognition (22–24). Here, we extend these observations to demonstrate the utility of multiple DARPins for constructing a single highly multispecific CAR that is capable of efficiently recognizing and eliminating heterogeneous tumors expressing disparate antigens.

Blood samples were obtained from donors who provided written-informed consent for research protocols approved by the Institutional Review Board at the Fred Hutchinson Cancer Center. Peripheral blood mononuclear cells were isolated by centrifugation using Ficoll–Histopaque (Sigma) or Lymphocyte Separation Media (Corning). CD8⁺ and CD4⁺ T cells were selected with EasySep Direct Human CD8⁺ and CD4⁺ T Cell Isolation Kit (Stem Cell Technologies).

The tumor cell lines K562 (ATCC CCL-243), MDA-MB-231 (ATCC HTB-26), NCI-H1975 (ATCC CRL-5908), and Raji (ATCC CCL-86) were purchased from the American Type Culture Collection, grown in RPMI media with 10% FCS and 100 U/mL penicillin–streptomycin (Gibco), and transduced with firefly luciferase (ffluc). 293T (ATCC CRL-11268) were cultured in DMEM with 10% FCS and 100 U/mL penicillin–streptomycin (Gibco). K562 cells were transduced with a lentivirus expressing full-length EGFR. Raji tumor cells were transduced with lentivirus expressing full-length EGFR, EpCAM, and/or HER2 alone and in combination to derive cells that express one, two, or all three of the target antigens (EGFR, EpCAM, and HER2). All tumor cell lines were tested routinely for mycoplasma contamination using the MycoAlert Mycoplasma Detection kit (Lonza) as per the manufacturer's instructions.

Monospecific CARs were designed using codon-optimized nucleotide sequences for the cetuximab scFv and DARPins E01 (EGFR), Ec1 (EpCAM), and G3 (HER2) linked to sequences encoding a modified IgG4 hinge with or without a (G₄S)₂ linker between the DARPin and hinge, followed by the CD28 transmembrane domain and the 4-1BB costimulatory and intracellular CD3ζ domains (25–28). Multispecific DARPins were linked in tandem each separated by a (G₄S)₂ or (G₄S)₄ linker from the neighboring DARPin followed by a (G₄S)₂ linker, the IgG4 hinge, and transmembrane and signaling domains as described above. CAR sequences were cloned into an epHIV7 lentiviral vector downstream of a GM-CSF receptor leader sequence and separated by a T2A sequence from a truncated CD19 tag (tCD19) used for detection and enrichment of transduced T cells (29, 30). A MYC tag (amino acid sequence EQKLISEEDL) was fused to the N-terminus of monospecific and multispecific DARPins CARs, and these constructs were used solely to assess surface expression of the CARs. Lentivirus was generated by transient transfection of psPAX2 and pMD2G packaging plasmids in HEK293-T cells. Primary CD8⁺ and CD4⁺ T cells were separately activated by anti-CD3/CD28 beads (Life Technologies), transduced 1 day after activation by centrifugation at 800 g for 90 minutes at 32°C with lentiviral supernatant and 1 μg/mL polybrene (Millipore). Transduced cells were then sorted for expression of tCD19, restimulated using the rapid expansion protocol as described (31), and tested in functional assays on day 11 or 12 after restimulation.

T-cell cytotoxicity was analyzed in a 4-hour chromium release assay, cytokine secretion was detected by ELISA of supernatants collected after 24 hours, and proliferation of T cells was measured by dilution of carboxyfluorescein succinimidyl ester or Cell Trace Violet–labeled cells by flow cytometry as described previously (32). Activation-induced cell death (AICD) of T cells was detected by Annexin V and propidium iodide staining after incubation with target cells for 24 hours (BD Biosciences# 556547).

T cells were stained with mAbs against human CD8 (SK1-BD Biosciences), CD4 (RPA-T4-BD Biosciences), human PD-1 (J105-eBioscience), and corresponding isotype antibody. Flow analysis was performed on a FACSCantoII or FACSARIA II, and data were analyzed with FlowJo (Treestar). Transduced T cells were detected by staining with anti-human CD19 mAbs (BD Biosciences #555415) to detect tCD19. A MYC tag fused to the N-terminus of DARPin-based CARs was used to determine CAR surface expression on the T cells by staining with the anti-MYC antibody (Cell Signaling Technology #2233S; ref. 33). Flow analysis of target expression on tumor cells was performed using mAbs against EGFR, EpCAM, and HER2 (Biolegend #352910, Biolegend #324208, Biolegend #324420). Immunoblot analysis was performed against anti-human CD247 pTyr142 (K25–407.69, BD Biosciences) and pan CD247 to detect CAR expression as previously described (25).

The Institutional Animal Care and Use Committee approved all mouse experiments. Six- to eight-week-old NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ [NOD/SCID/γc−/− (NSG)] female mice were procured from The Jackson Laboratory or bred in-house. A mixture of CD8⁺ and CD4⁺ T cells (1:1 ratio), unless otherwise specified, was transferred via tail vein injection 1 week after subcutaneous implantation of MDA-MB-231 or after intravenous implantation of Raji. Tumor growth was monitored by bioluminescence imaging as previously described (32). Transferred T cells were detected in blood or tumors harvested at specified intervals after therapy using mAbs specific for CD8 and CD4. Analysis of EGFR, EpCAM, and HER2 antigen expression on tumor cells harvested at euthanasia was performed by flow cytometry.

Approximate binding affinities of 9-mer peptides derived from the DARPins and scFvs to MHC were predicted using NetMHC (34).

Log-transformation was conducted for the data with large value before analysis. To compare two paired groups, two-sample paired t test was used. The two-way ANOVA was used if there were two factors in group comparisons. Kaplan–Meier curves were generated for survival analysis and the curves compared by the Log-rank test. Results with a P value of <0.05 were considered significant. To compare multiple groups with a single control, Dunnett test was used for multiplicity adjustment. For survival curve comparison, an unadjusted P value was reported.

Normal donors were enrolled for peripheral blood collection and provided written-informed consent for research protocols approved by the Institutional Review Board of the Fred Hutchinson Cancer Research Center (FHCRC). Blood sample collection was conducted in accordance with the Declaration of Helsinki. The in vivo mouse experiments were approved by the FHCRC Institutional Animal Care and Use Committee.

The E01 DARPin has been described previously and recognizes a similar region on domain III of the EGFR protein as cetuximab (35). We designed a second-generation 4-1BB/CD3ζ CAR with monospecific E01 DARPin as the recognition motif and compared its function in T cells with a similarly designed CAR in which the DARPin was replaced with the cetuximab scFv (Fig. 1A). Separate aliquots of CD8⁺ T cells were transduced with lentivirus encoding the E01 DARPin or the cetuximab scFv CARs and transduced tCD19⁺ T cells were sorted to high purity (Supplementary Fig. S1A). The E01 DARPin and cetuximab scFv CAR-T cells were tested for recognition of a panel of target cells expressing EGFR including K562 cells that were transduced with full-length human EGFR (K562/EGFR), as well as MDA-MB-231 breast cancer and NCI-H1975 lung cancer cell lines that endogenously expressed EGFR (Supplementary Fig. S1B). K562/EGFR cells but not the EGFR⁻ K562 parental line, and endogenous EGFR⁺ tumor cell lines were efficiently lysed by both the E01 DARPin CAR and cetuximab CAR (Fig. 1B and C).

DARPins are small (15–18 kDa), and we hypothesized that the addition of a flexible linker between the DARPin and the hinge and/or an increase in spacer length might optimize the geometry of the T-cell/tumor cell interaction and improve CAR function. We constructed the E01 DARPin CAR with and without the addition of a (G₄S)₂ linker between the DARPin and the hinge in a short (IgG4 hinge) and modified long (hinge-CH2-CH3) spacer format (Supplementary Fig. S1C; ref. 36). The expression of these different DARPin CARs was evaluated by Western blotting using an antibody against the CD3ζ chain, with the endogenous CD3ζ chain (16 kDa) serving as a loading control. Both the short and long spacer E01 CARs were expressed in T cells at the expected size of 42 and 67 kDa respectively and appeared to be at comparable levels with the cetuximab scFv CAR (Fig. 1D). Each DARPin CAR sample had doublet bands at the predicted size, which is likely due to incomplete unfolding of the DARPin in SDS-PAGE gels due to their high thermodynamic stability (37).

We evaluated T cells expressing these modified E01 DARPin CARs in functional assays. CD8⁺ T cells expressing E01 DARPin CARs in the long and short spacer format with the (G₄S)₂ linker lysed EGFR⁺ MDA-MB-231 comparable with CD8⁺ T cells expressing the cetuximab CAR. In contrast, the short spacer without the (G₄S)₂ linker exhibited a minor reduction in cytolysis not observed with the long spacer (Fig. 1E). Interestingly, CD8⁺ T cells expressing the short spacer E01 DARPin CAR without the (G₄S)₂ linker showed markedly lower production of IFNγ and IL2, and decreased proliferation after coculture with MDA-MB-231 target cells compared with all other EGFR CAR-T cells tested (Fig. 1F and G). CD8⁺ T cells expressing the (G₄S)₂ linker containing short and long spacer E01 CARs proliferated similarly to T cells expressing the cetuximab CAR (Fig. 1G), and produced IFNγ and IL2, although the level of cytokines produced by these E01 DARPin CAR-T cells was less than that produced by the cetuximab CAR-T cells (Fig. 1F). A similar hierarchy in cytokine production was observed when CD4⁺ T cells were transduced with the E01 DARPin and cetuximab scFv CARs. CD4⁺ T cells expressing the short spacer E01 CAR lacking (G₄S)₂ produced significantly less IFNγ and IL2 and proliferated poorly after stimulation with MDA-MB-231, and the addition of the (G₄S)₂ linker or the CH2-CH3 sequences improved function of the short spacer construct (Supplementary Fig. S2A and S2B). The long spacer formats of the DARPin CAR showed equivalent proliferation and slightly less cytokine production than the cetuximab CAR in CD4⁺ T cells. Because of the high affinity of E01 and cetuximab for EGFR, we also measured whether engagement of tumor cells might induce AICD of CAR-T cells. We found that the cetuximab CARs showed slightly higher AICD (Annexin V⁺, PI⁺, CD8⁺ cells) in vitro than the E01 DARPins CARs after 24 hours of incubation with MDA-MB-231 (Supplementary Fig. S2C).

We next examined whether the monospecific DARPin CAR-T cells were effective in treating NSG mice engrafted with MDA-MB-231. Tumor cells were inoculated subcutaneously and 1 week later mice received a 1:1 ratio of CD8⁺ and CD4⁺ CAR-T cells (38). T cells expressing all of the E01 DARPin CARs with various spacer lengths with or without the (G₄S)₂ linker exhibited superior tumor reduction compared with the nontransduced T cells; however, the effects of spacer design on CAR-T cell function were more evident in vivo than suggested by the in vitro functional assays. E01 DARPin CAR-T cells with the short spacer CAR lacking the (G₄S)₂ linker induced only modest tumor regression and improvement in survival compared with control T cells (Fig. 1H–J). This reduced tumor control by E01 short spacer–specific cells lacking (G₄S)₂ was predicted based on in vitro functional assessments, but it was unexpected that the E01 long spacer formats [±(G₄S)₂] were less effective in tumor control compared with the (G₄S)₂ linker containing short E01 CAR, despite equivalent in vitro function (Fig. 1 H–I). Only E01 DARPin–specific cells with the short spacer and (G₄S)₂ linker and the cetuximab scFv–specific cells completely eradicated tumor and cured all mice (Fig. 1H–J). CAR-T cells were present in similar levels in the blood of mice with MDA-MB-231 tumors in all groups early after T-cell transfer (Supplementary Fig. S2D). However, the frequency of E01 (+(G₄S)₂) short spacer CAR-T cells was higher than the long spacer (+(G₄S)₂) at the tumor site but not in the spleen, suggesting enhanced local survival or proliferation of T cells expressing the E01 (+(G₄S)₂) short spacer CAR (Supplementary Fig. S2E). Together, these results indicate that DARPin binders can function efficiently as recognition motifs in CARs in vitro and in vivo and show that spacer and linker modifications can be decisive in achieving optimal efficacy.

Ankyrin repeats are more stable and aggregation-resistant than scFvs, and it is feasible to link DARPins in tandem repeats in one molecule (27). To evaluate DARPins in a multispecific CAR format, we first linked EGFR-specific (E01) and EpCAM-specific (Ec1) DARPins in tandem either separated from each other by a (G₄S)₂ or (G₄S)₄ linker, followed by a (G₄S)₂ linker, the hinge, and the 4-1BB and CD3ζ signaling domain (26, 27). We designed the CARs to have either E01 or Ec1 as the membrane-proximal DARPin, expressed them in T cells, and tested recognition of Raji tumor cells transfected to express EGFR or EpCAM alone, or both EGFR and EpCAM (Fig. 2A; Supplementary Fig. S3A and S3B). T cells expressing each of the bispecific constructs recognized single EGFR or EpCAM-positive target cells, but there was a hierarchy in cytolytic activity and cytokine production. CAR-T cells expressing the Ec1-(G₄S)₄-E01 bispecific DARPin exhibited slightly higher cytotoxicity and cytokine secretion against the single positive targets compared with the other bispecific CAR formats (Fig. 2B), and showed the highest cytotoxicity and cytokine secretion against Raji target cells expressing both EGFR and EpCAM (Fig. 2B and C).

We then examined if DARPins could be used for designing targeting structures of even higher multispecificity by fusing a third DARPin (G3) that recognizes HER2 either upstream or downstream of the optimal bispecific DARPin CAR (Ec1-(G₄S)₄-E01; Fig. 2D; refs. 28, 39). Evaluation of T cells that expressed these constructs against single EpCAM-, EGFR-, or HER2-positive Raji targets (Supplementary Fig. S3A and S3B) demonstrated that the trispecific DARPin CAR produced IFNγ and mediated cytotoxicity in response to HER2⁺ cells only when G3 was membrane-distal (G3-Ec1-E01), and not when in a membrane-proximal (Ec1-E01-G3) position (Fig. 2D and E). This was in line with published data that G3 DARPin binds to a membrane-proximal epitope on HER2, and therefore might require a membrane-distal location in the trispecific CAR format for effective T-cell synapse formation (39). T cells expressing the trispecific CAR with the E01 DARPin in a membrane-proximal position also exhibited better IFNγ production against single EGFR-expressing target cells than when E01 was placed between the G3 and Ec1 DARPins (Fig. 2D). Cytokine production by the Ec1 DARPin against single EpCAM expressing target cells was equivalent either in a membrane-distal position or when placed between the G3 and E01 DARPin (Fig. 2D). Based on these results, we chose the G3-Ec1-E01 construct as a trispecific CAR to characterize further.

The function of trispecific G3-Ec1-E01 CAR against Raji cells expressing either EGFR or EpCAM or HER2 was then compared with the respective monospecific CAR constructs to determine whether tandem linking might reduce function against single antigen-positive target cells. We first assessed the level of expression of each CAR on T cells by fusing an N-terminal MYC tag to the monospecific and trispecific CARs to allow direct measurement of CAR surface expression by flow cytometry. We observed lower surface expression of the trispecific G3-Ec1-E01 construct compared with each monospecific CAR (Fig. 3A). Consistent with the reduction in expression, trispecific CAR-T cells exhibited reduced cytotoxic activity against Raji tumor cells expressing EGFR, EpCAM, or HER2 alone, and produced lower levels of IL2 against single-positive targets compared with the corresponding monospecific CAR-T cells (Fig. 3B and C). T cells expressing the trispecific CAR proliferated after incubation with Raji tumor cells expressing only EGFR, EpCAM, or HER2 at levels similar to or slightly less than T cells expressing the corresponding monospecific CARs (Fig. 3D). We did not observe basal CAR phosphorylation in the monospecific, bispecific, or trispecific DARPin CAR T cells as observed in previously published tonic signaling CARs such as the ROR1- CD28/CD3ζ CAR T cells (Supplementary Fig. S4A; ref. 40). The monospecific, bispecific, or trispecific DARPin CAR T cells also did not demonstrate cell surface expression of PD-1 prior to activation with target cells, and only upregulate PD-1 with activation as shown for the E01 CAR stimulated with MDA-MB-231 tumor cells (Supplementary Fig. S4B; ref. 41). Comparison of function between the trispecific CAR (G3-Ec1-E01) and the optimal bispecific CAR construct (Ec1-(G₄S)₄-E01) demonstrated equivalent IFNγ production against Raji cells expressing EGFR or EpCAM or both EGFR and EpCAM (Supplementary Fig. S4C). Further, the bispecific and trispecific CAR recognized endogenously expressed antigens on the MDA-MB-231 breast tumor line (Supplementary Fig. S4D).

Heterogeneity of antigen expression in tumors is a barrier to effective immunotherapy and might be overcome by multispecific targeting, either by infusing T cells expressing different monospecific CARs or expressing a single multispecific CAR. We evaluated the efficacy of CD4⁺ and CD8⁺ T cells expressing the trispecific G3-Ec1-E01 DARPin CAR, each monospecific CAR, or a mixture of T cells expressing each monospecific CAR for treating a heterogeneous tumor in NSG mice. Mice were injected with 0.5 million Raji tumor cells expressing either EGFR, EpCAM, or HER2 in a 1:1:1 mix and treated 7 days later with CAR-T cells against one of the targets, a mix of monospecific CAR-T cells or trispecific (G3-Ec1-E01) CAR-T cells at the same total cell dose (Fig. 4A). The monospecific CAR-T cells did not exert measurable tumor control compared with mock T cells; however, a mixture of T cells expressing each monospecific CAR and T cells expressing the trispecific CAR exhibited a delay in tumor growth and improved survival (Fig. 4B–D). CD8⁺ and CD4⁺ expressing the trispecific and each monospecific DARPin CAR were detectable in peripheral blood for over 3 weeks after T-cell transfer (Fig. 4E). The CD8⁺ and CD4⁺ trispecific CAR T cells demonstrated an early peak expansion higher than the mix of monovalent CARs followed by a decline at later time points that correlated with decreased tumor burden. On examining tumors from mice that met the euthanasia criteria, we observed that in the heterogeneous tumor setting, monospecific specific cells eliminated tumor cells that expressed their cognate antigen and outgrowth represented tumor cells that expressed only the nontargeted antigens or were antigen null (Fig. 4F). In mice treated with trispecific CAR-T cells or a mix of monospecific CAR-T cells, most of the tumor cells were antigen null indicating outgrowth of tumor cells that had either lost cognate antigen or rare tumor cells that were not transfected (Fig. 4F). We observed Raji tumor cells that were antigen null in all groups, but the percentage was low in mice treated with mock T cells or with the E01 or G3 monospecific DARPin CARs where the harvested Raji tumors were predominantly EpCAM⁺, possibly due to more aggressive growth of tumor cells transfected with EpCAM (Fig. 4F).

The goal of multispecific targeting is to eliminate tumor cells that express any constellation of target antigens. Therefore, we examined potential differences in in vivo antitumor efficacy of the trispecific construct against Raji tumor cells expressing just a single target, a combination of 2 targets or all 3 targets. Expression of the target antigens (EpCAM and HER2) was similar in the single, double, or triple transfectants, but expression of EGFR in the triple-transfected Raji cells was 0.6-fold of the level in single or double EGFR⁺ targets (Supplementary Fig. S3B). We observed that a dose of 9 million trispecific CD4⁺ and CD8⁺ CAR-T cells was insufficient to clear Raji tumors that expressed only HER2 or EpCAM but was curative against Raji tumor cells expressing EGFR, suggesting that the potency of the trispecific CAR is different against tumor cells expressing different single target antigens (Fig. 5A). However, at the same CAR-T cell dose, we observed more rapid tumor clearance and more prolonged tumor control in mice engrafted with Raji tumor cells coexpressing both EpCAM and HER2 compared with mice engrafted with single antigen (Raji/EpCAM and Raji/HER2) tumors, although the survival difference was not significant (Fig. 5A and B). The superior antitumor activity did not reflect differences in CAR-T cell expansion, suggesting that efficacy was due to superior recognition by the trispecific CAR of tumor cells expressing these two antigens (Supplementary Fig. S5A).

To assess synergy of trispecific CAR T cells against tumor cells expressing EGFR and additional antigens, we lowered the dose of T cells because the higher dose was curative against Raji tumors expressing EGFR. At a T-cell dose of 5 million CAR-T cells, we saw equivalent antitumor activity from trispecific CAR-T cells in mice engrafted with Raji tumors expressing only EGFR and those engrafted with Raji expressing EGFR and EpCAM (Fig. 5C and D). However, the trispecific CAR cleared tumor and improved the survival in mice with Raji tumors expressing all 3 antigens compared with tumors expressing only one or two antigens (Fig. 5C and D). T-cell expansion was also observed in blood in all treatment groups (Supplementary Fig. S5B). We did not observe significant differences in short-term cytotoxicity, cytokine secretion, or proliferation assays of the trispecific DARPin CAR against target cells expressing single, dual, or triple antigens (Supplementary Fig. S5C–S5E; ref. 14).The trispecific CAR also did not show an increase in AICD on encountering single versus multiple antigens (Supplementary Fig. S5F). These results illustrate the ability of trispecific CARs to target heterogeneous tumors and exhibit synergistic activity in vivo against tumor cells that express multiple antigens.

A barrier that could limit the use of antibody mimetic proteins in CARs is the potential for immunogenicity of non–scFv-based motifs. MHC-restricted CD8⁺ and CD4⁺ T-cell responses to transgene-encoded epitopes have been observed in some patients treated with T cells expressing a foreign protein, including CAR-T cells designed from murine scFvs (30, 42). Because mouse models are poor predictors of immunogenicity in humans due to differences in MHC proteins, we used an in silico approach to identify candidate epitopes for human class I MHC. We focused on the immunogenicity of the core monospecific and trispecific DARPins and compared them with the cetuximab scFv (25). We identified 9-mer binding sequences with a predicted affinity of less than 100 nmol/L to the human MHC I supertype alleles using the NetMHC prediction algorithm (Supplementary Table S1). We found that both monospecific DARPins and the cetuximab scFv have between 1 and 2 peptides, predicted to bind to the MHC I supertype. The trispecific DARPin has an increase in the number of peptides (n = 5) predicted to bind MHC I alleles over the monospecific construct (Supplementary Table S1). However, on screening the predicted immunogenic peptides against the human proteome, we found that most of the DARPin but not mouse cetuximab scFv hits matched highly similar sequences in the human proteome. These results indicate that DARPin sequences could potentially be even less susceptible to T–cell–mediated immunogenicity than murine scFvs because of the abundance of natural ankyrin repeat proteins in the human proteome.

Antigen loss as a mechanism of escape after T-cell therapy for cancer can limit the efficacy of this approach (1, 5–7, 11, 12). Strategies to target multiple antigens are likely to be necessary for optimal efficacy, and multiple groups are working to develop multispecific CARs using scFv-based technologies (1, 13–17). In this study, we show that single domain antibody mimetic proteins such as DARPins can serve as multispecific recognition motifs and may offer advantages over scFvs due to their stable nature and aggregation resistance (20).

We tested multispecific recognition using three previously characterized DARPins that target EGFR (E01), EpCAM (Ec1), and HER2 (G3; refs. 26–28). We acknowledge that these antigens are expressed on normal tissues in addition to high expression on tumors; however, they serve as model antigens for proof-of-principle experiments to evaluate multispecific targeting. In the design of monospecific E01 DARPin CARs, we found that the addition of a (G₄S)₂ linker between the DARPin and the hinge improved CAR function most likely due to a more optimal spacer length or to the added flexibility that improved target binding (35). The design of multispecific DARPin CARs, including the trispecific CAR, was empirically determined. Evaluation of the functionality of each DARPin in all three positions of the trispecific CAR identified G3-Ec1-E01 construct as the optimal design which was in agreement with the published crystal structures of the G3 DARPin recognizing a membrane-proximal epitope of HER2 and E01 recognizing a membrane-distal epitope of EGFR (35, 39). Although a published crystal structure of the Ec1 DARPin and its target is not available, Ec1 functioned in all locations in the trispecific design, thus no inferences can be made about the optimal location.

Functional testing of the trispecific DARPin CAR-T cell in both in vitro and mouse studies revealed a hierarchy in antitumor efficacy against the different targets. A CAR-T cell dose that was ineffective in clearing tumors expressing only EpCAM or HER2 was sufficient to clear an EGFR-expressing tumor. The divergence may be partly explained by differences in growth kinetics between the transfected Raji targets or the efficiency of engagement of individual DARPins with their target on tumor cells. However, the potency of the trispecific CAR was lower than the corresponding monospecific CARs against tumor cells expressing a single target, revealing a potential limitation of multispecific designs, possibly requiring refinement in the geometric presentation of the binders to various epitopes. The optimal level of CAR expression on T cells is not known; however, the trispecific CAR was expressed at lower levels than the monospecific CARs, which might explain the decreased activity of the trispecific CAR against single antigen targets. Alternatively, positioning of the G3 and Ec1 DARPin may not be optimal due to the additional length of the extracellular domain of the trispecific CAR.

Importantly, in the context of a mixture of heterogeneous antigen-positive tumor cells, the trispecific CAR-T cells were as effective as a mix of monospecific CARs. The trispecific CAR also exhibited increased in vivo antitumor activity at suboptimal T-cell doses against tumors that expressed multiple (2 or 3) antigens compared with activity against tumor cells expressing only a single antigen. The synergistic increase in trispecific CAR activity could be due to enhanced T-cell activation due to binding of multiple antigens simultaneously, which may potentiate signaling at the immune synapse (14). This could also possibly generate a safety window for targeting a combination of antigens on tumors when any of the individual antigens is expressed on normal tissue. These data suggest that for treatment of a heterogeneous solid tumor, a single T-cell product expressing a multispecific CAR is an alternative that avoids the complexity of engineering multiple monospecific vectors and multiple T-cell products.

Our data also reaffirm the obstacle of antigen escape. We consistently observed immune selection when a heterogeneous mixture of tumor cells expressing the three target antigens were engrafted into mice and then treated with monospecific or trispecific CAR-T cells. Tumors that grew out in mice treated with the trispecific CAR or with a mixture of monospecific CAR-T cells were mostly comprised of antigen-null variants. In our system, the target molecules were transfected into Raji cells and were not essential for the malignant phenotype. Furthermore, the studies were performed in immunodeficient mice where engagement of an endogenous host response cannot occur. For clinical translation, multispecific receptors should target molecules that are involved in tumor cell maintenance or possibly be combined with checkpoint inhibitors to engage endogenous immune response at the tumor site (43–45). An additional challenge with multispecific CARs is the potential for enhanced immunogenicity of complex-binding domains.

Notwithstanding the aforementioned challenges, our results demonstrate that in principle DARPin motifs can serve as effective monospecific and multispecific tumor recognition domains for CARs and function effectively in T cells to reduce antigen escape. Refinements such as improved cell surface expression, computational modeling of ligand binding, and evaluation of immunogenicity can further contribute to the use of antibody mimetic proteins in CAR design and facilitate clinical translation.

A.I. Salter is an employee/paid consultant for Lyell Immunopharma. S.R. Riddell is an employee/paid consultant for Lyell Immunopharma and Juno Therapeutics (a Celgene company), reports receiving commercial research grants from Juno Therapeutics (a Celgene company), and holds ownership interest (including patents) in Juno Therapeutics (a Celgene company) and Lyell Immunopharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Balakrishnan, A. Rajan, A. Plückthun, S.R. Riddell

Development of methodology: A. Balakrishnan, A. Rajan, M.C. Jensen, S.R. Riddell

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Balakrishnan, A. Rajan, A.I. Salter, S.R. Riddell

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Balakrishnan, A. Rajan, Q. Wu, J. Voutsinas, S.R. Riddell

Writing, review, and/or revision of the manuscript: A. Balakrishnan, A. Rajan, A.I. Salter, Q. Wu, J. Voutsinas, A. Plückthun, S.R. Riddell

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Balakrishnan, A. Rajan, P.L. Kosasih, M.C. Jensen, S.R. Riddell

Study supervision: A. Plückthun, S.R. Riddell

The study was supported by grants from the NIH (R01 CA114536; R01 CA136551; P50 CA138293 to S.R. Riddell) and the FHCRC Walker immunotherapy fellowship in support of A. Balakrishnan. We would like to thank Melissa Comstock, Don Evan S. Parrilla, and Mercedez Jess (Shared Resources, Fred Hutchinson Cancer Research Center) for performing mouse husbandry and tumor xenograft experiments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.