Conduit CAR: Redirecting CAR T-Cell Specificity with A Universal and Adaptable Bispecific Antibody Platform

The success of chimeric antigen receptor (CAR) T-cell therapy against hematologic malignancies has altered the treatment paradigm for patients with these diseases. Nevertheless, the occurrence of relapse due to antigen escape or heterogeneous antigen expression on tumors remains a challenge for first-generation CAR T-cell therapies as only a single tumor antigen can be targeted. To address this limitation and to add a further level of tunability and control to CAR T-cell therapies, adapter or universal CAR T-cell approaches use a soluble mediator to bridge CAR T cells with tumor cells. Adapter CARs allow simultaneous or sequential targeting of multiple tumor antigens, control of immune synapse geometry, dose control, and the potential for improved safety. Herein, we described a novel CAR T-cell adapter platform that relies on a bispecific antibody (BsAb) targeting both a tumor antigen and the GGGGS (G4S) linker commonly used in single-chain Fv (ScFv) domains expressed on CAR T-cell surfaces. We demonstrated that the BsAb can bridge CAR T cells to tumor cells and potentiate CAR T-cell activation, proliferation, and tumor cell cytolysis. The cytolytic activity of CAR T-cells was redirected to different tumor antigens by changing the BsAb in a dose-dependent manner. This study highlights the potential of G4S-displaying CAR T cells to be redirected to engage alternative tumor-associated antigens (TAA). Significance: New approaches are needed to address relapsed/refractory disease and manage potential toxicities associated with CAR T-cell therapy. We describe an adapter CAR approach to redirect CAR T cells to engage novel TAA-expressing cells via a BsAb targeting a linker present on many clinical CAR T-cell therapeutics. We anticipate the use of such adapters could increase CAR T-cell efficacy and reduce potential CAR-associated toxicities.


Introduction
New methods for redirecting and reengineering T cells have revolutionized the cancer treatment paradigm by harnessing the vast potential of the human immune system (1). Chimeric antigen receptor (CAR) T-cell immunotherapy has shown tremendous success in the treatment of acute lymphocytic leukemia (ALL) and refractory diffuse large B-cell lymphoma (DLBCL) led by the FDA-approved CD19 targeting CARs Kymriah (tisagenlecleucel) and more modular and adaptable. Although CAR design can vary, the majority of CARs in the clinic include an extracellular domain which recognizes a tumor-associated antigen (TAA). The extracellular binding domain is usually composed of a single-chain Fv (ScFv) region derived from an antibody. ScFv molecules are composed of a variable heavy (VH) and a variable light (VL) domain derived from an antibody connected by an unstructured synthetic linker [typically made of repeating GGGGS (G 4 S) repeats]. Attached to the extracellularly accessible CAR ScFv are the hinge region and transmembrane domain that are commonly derived from the CD8 extracellular domain. Intracellularly, the CAR protein contains one more costimulatory domains and the intracellular signaling domain from CD3ζ . A drawback of this mono-specific CAR T-cell design is that antigen escape can render CAR T cells unable to recognize the tumor and may lead to relapse.
The CAR-adapter approach (13) is designed to allow for controllable tumor engagement to reduce toxicity and increase adaptability to engage multiple antigens (14). This approach involves decoupling antigen recognition the CAR T-cell itself and using a soluble mediator to act as a bridge or "conduit" between the engineered T-cell and the targeted tumor cell. This design relies on recognition of the soluble adapter by the extracellular portion of the CAR. A central feature of the CAR-adapter approach is the introduction of an engineered antigen on T cells that can be engaged by the soluble adapter molecule (to bridge the T cell with tumor cells). Multiple designs for CAR-adapter pairs have been put forward with a variety of engineered antigens and respective adapters including an antifluorescein CAR with fluorescein-labeled antibodies (15)(16)(17), CD16-based CARs designed to bind to antibody Fc regions (18)(19)(20), CAR-adapter pairs based on leucine zippers (21), and a CAR-adapter pair using the peptide neo-epitope (PNE) derived from the yeast transcription factor GCN4 and an antibody recognizing it (17,22).
In this study, we describe a versatile CAR-adapter pair that does not rely on the introduction of any novel engineered antigen on T cells, but rather exploits an existing feature already present in most clinical CAR T-cell therapies, the flexible ScFv linker. (Fig. 1). We generated an antibody targeting the (GGGGS) n or (G 4 S) n linker present on most existing CARs and have used BsAbs targeting both this linker and a TAA to demonstrate killing of cells expressing the TAA. We showed that CAR T cells expressing either a germline antibody ScFv (with no known specificity) or a CD19 targeting CAR can be redirected to target prostate tumor cells via a bispecific adapter molecule. We anticipate this approach may be clinically advantageous due to its adaptability to target novel TAAs, control of potential toxicity via dosing adjustment, and its potential compatibility with existing clinical CARs.

Reagents and Cell Lines
Human prostate tumor cell lines were purchased from ATCC; these include PC3 (RRID:CVCL_0035; obtained and banked in 2019) and LNCaP (RRID:CVCL_0395; obtained and banked in 2018) cells. The PC3 cells used in this article were sorted and enriched for prostate-specific membrane antigen (PSMA) surface expression prior to use. HEK293T (ATCC; obtained and banked in 2018) cells were maintained in DMEM supplemented with 10% heatinactivated FBS in purpose to generate lentiviral particles. Cell lines used in this study were banked immediately after receiving and experiments occurred within approximately 2 months of resuscitation; cell lines were not authenticated. All cells were periodically tested for mycoplasma at the using MycoAlert The extracellular portion of the designed CAR construct contains an exposed (G 4 S) 4 linker sequence joining the VH and VL regions of the TAA targeting ScFv. The (G 4 S) 4 -containing CAR molecules can act as a functional epitope and be recognized by anti-G 4 S binding antibody. Using a bispecific antibody, tumor cells expressing a TAA can be targeted via the adapter bispecific. This bispecific adapter molecule then redirects CAR T cells to a new tumor cell target and consequently activate CAR T cells to further engage tumor cells.
(Lonza) and found to be negative. G 4 S peptides of various lengths were generated by peptide synthesis techniques at New England Peptide company. A biotin moiety was appended to the C-terminal end of the peptides to allow for immobilization on streptavidin coated ELISA plates or beads.
Protein A-coated magnetic beads were purchased from Spherotech, and recombinant human CD19-Fc fusion (rhCD19) was purchased from BioLegend. Protein A-coated beads were loaded with Fc-containing bispecifics or human CD19-Fc according to manufacturer's instructions and excess soluble protein was washed before use.

Antibody Generation
Antibodies were generated by immunizing New Zealand white rabbits. Rabbits were manually restrained and the dorsal area shaved and cleaned with chlorhexidine. Immunogen (G 4 S peptide-BSA) was injected in 4-8 sites subcutaneously (usually 0.25 mL/site) or intradermally. The animals were immunized 1-2 times based on serum titer. Subsequently, the animals were anesthetized, blood collected, and the spleen and secondary lymph nodes aseptically removed for fusion for hybridoma generation or subjected to single B-cell sorting via FACS. Hybridoma supernatants were screened by MSD electrochemiluminescence for binding to biotinylated peptide. Hits with the desired binding profile were selected and sequenced as described below.

ELISA Methods
ScFvs with (G 4 S) 4 and modified Bird and colleagues linker (23) were coated on 96-well maxisorp plates (Nunc) overnight at 4°C (2 μg/mL in PBS). Plates were then blocked with 5% milk in PBS for 1 hour at room temperature. After washing (3×), titrated antibodies were added and allowed to incubate for 1.5 hour. After three washes in PBS, peroxidase-conjugated goat anti-human Fc (Jackson Immunoresearch) was used for detection (1-hour incubation). The wells were washed three times, and TMB substrate solution was added. The reaction was stopped with 2 N H 2 SO 4 and the absorbance at 450 nm was measured. Data analysis was performed using GraphPad Prism 7.0 (GraphPad Prism, RRID:SCR_002798).

Flow Cytometry
HEK-293 cells expressing CARs with either (G 4 S) 3  with FACS buffer, Flow cytometry was performed using Intellicyt IQue2 (Sartorius). Analysis was performed using Intellicyt software and data were graphed and analyzed using GraphPad Prism 7.0.

Biolayer Inferometry
Biolayer Inferometry was performed using the Octet Red system (ForteBio). Streptavadin Dip and Read biosensors were prewetted by dipping them into Octet Kinetic Buffer (ForteBio) for 10 minutes before use and they were incubated with various Biotin-G 4 S peptides and immobilized on sensor tips. Tips were then dipped into wells with varying concentrations of anti-G 4 S antibodies after 170-second loading time, tips were dipped into Octet Kinetic Buffer for a disassociation step (170 seconds). Sensograms were recorded and analyzed using Octet Software to determine on-rate and off-rate.

BsAb Cloning, Expression, and Purification
Bispecific mAbs targeting two different prostate cancer antigens were generated in the study [anti-(G 4 S) n × anti-PSMA and anti-(G 4 S) n × anti-TMEFF2] using the bispecific format described in Coloma and colleagues (24). ScFv portions of the bispecific antibodies were generated using 20 amino acidlong linkers modified from Bird and colleagues (23) with the following amino acid sequence: GGSEGKSSGSGSESKSTGGS. G 4 S linkers were not used in the construction of these bispecifics to avoid intramolecular binding and selfaggregation. DNA gBlocks were synthesized containing the sequence of anti-(G 4 S) n scFv or anti-PSMA scFv or anti-TMEFF2 scFv. These gBlocks were inserted into a mammalian expression vector using InFusion method. Human CD4 signal peptides were encoded to allow for efficient secretion of antibodies into culture supernatant. All constructs were sequence verified and scaled up using Endotoxin free maxi preparation kits. ExpiCHO mammalian expression system was used for protein expression (Invitrogen). To ensure proper light chain loading in the mature protein, a 3:1 light chain:heavy chain DNA ratio was used. The DNA mixture was incubated with Expifectamine and immediately added to the culture. ExpiCHO suspension cultures were harvested after 10 days by centrifuging at 3,000 × g for 10 minutes to pellet cells. The supernatants were stored at 4°C until purification.
HiTrap MabSelect Sure (GE Healthcare) columns were used to purify BsAb protein. Supernatants from transfected cells were applied to the column at a flow rate of 1 mL/minute for maximum capture. Columns were washed using 20 col-

CAR Construction, Lentiviral Production, and CAR T-Cell Generation
Human codon-optimized DNA comprising the 3-23/B3 ScFv sequence, CD8α hinge, and transmembrane domains, 4-1BB, and CD3ξ domain were cloned into the lentiviral vector. An additional CD19 targeting CAR with an N-terminal MYC-tag and a (G 4 S) 3 ScFv linker was also constructed as above. To produce high-titer replication-defective lentiviral vectors, 293-T human embryonic kidney cells were transfected with lentiviral packaging plasmids [pVSV-G (RRID:Addgene_138479), pRSV.REV (RRID:Addgene_106453), pMDLg] and CAR-encoding lentiviral vector using Lipofectamine 2000 (Invitrogen). The viral supernatant was harvested at 24 and 48 hours posttransfection. Viral particles were concentrated using Lenti-X concentrator (Takara). Concentrated viral particles were resuspended in PBS and stored frozen at −80°C. Primary human CD4 + and CD8 + T cells (purchased from HemaCare) were isolated from healthy volunteer donors following leukapheresis by negative selection. T cells were cultured in complete media (RPMI1640 supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, 10 mmol/L HEPES), stimulated with anti-CD3 and anti-CD28 mAb-coated beads (Invitrogen). Twenty-four hours after activation, T cells were transduced with recombinant lentiviral particles at multiplicity of infection of approximately 5-10. Human recombinant IL2 (PeproTech) was added every other day to 50 IU/mL final concentration and 0.5-1 × 10 6 cells/mL cell density was maintained. CAR surface expression was verified by flow cytometry using GLPB30 IgG as primary staining followed by PE-labeled anti-human Fc antibody as secondary antibody. Anti-MYC antibody labeled with AF-647 (BioLegend) was used to stain anti-CD19 MYC-tagged CAR T cells and assess expression levels.

CAR T-Cell CD107a Assay and Proliferation Assay
CD107a (also known as LAMP-1) is a marker of CD8 + T-cell degranulation following stimulation and activation (25). For CD107a assay, CAR T cells were cocultured with PC3 prostate tumor cells in a 96-well plate at an effectorto-target ratio (E:T) equal to 5:1 in the presence or absence of anti-PSMA × anti-G 4 S BsAbs (5 mg/mL). Phycoerythrin-labeled anti-CD107a antibody was added 1 hour before adding protein transport inhibitor, Golgi Stop (BD Biosciences) and the plate was incubated for 3 hours. The anti-CD8 antibody was added and incubated at 37°C for 30 minutes. After incubation, the samples were  Cytotoxicity of the CAR-expressing T cells was also tested by using the IncuCyte zoom living cell imaging system. Coculture was set up the same way as above in xCELLigence. Assay images were taken every 30 minutes and the number of dead cells was quantified.

Cytokine Assay (Intellicyt iQue)
The Intellicyt human T-cell activation and cytokine profiling kit was applied for T-cell activation and cytokine profile. Briefly, CAR T cells were cocultured with PC3 prostate tumor cells at an E:T ratio of 1:1 in 96-well round-bottom plate in 200 μL RPMI complete media. The BsAbs of anti-PSMA × anti-G4S (5 mg/mL) was added. Coculture without BsAb as well as target and CAR T cells alone were used as controls. Seventy-two hours later, T-cell activation was assessed by the TCA kit from a 30 μL cell/supernatant mixture sample following the protocol. Samples were acquired on the Intellicyt iQue Screener PLUS. Standard curves were generated to quantitate the levels of secreted cytokines.
Data were analyzed with ForeCyt software. In another experiment, BsAb-or rhCD19-coated beads were substituted for PC3 tumor cells and used to activate CAR T cells (5:1 bead:cell ratio was used).

Data Availability Statement
Data were generated by the authors and are included in the article or available upon request.

Generation of an Anti-(G4S) n Linker Antibody
To develop a modular T-cell adapter that can bind to a variety of TAA-binding domains, we sought to generate antibodies that recognize the poly-Glycine-Serine linker peptide because of its ubiquitous use in ScFv constructs.
Hybridoma screening of clones against (G 4 S) 4  linker with an EC 50 of 0.57 nmol/L but did not bind to a ScFv (with the same variable domains) containing the modified Bird and colleagues' linker ( Fig. 2A). Similar results were observed when these ScFvs were generated as CAR constructs and binding to GLPB30 was assessed by flow cytometry (Fig.  2B). GLPB30 bound to HEK 293 cells expressing the (G 4 S) 4

linker containing
ScFv with an EC 50 of 0.78 nmol/L. No binding was observed to HEK 293 cells expressing CAR ScFvs with a different linker (Fig. 2B). The minimal length of (G 4 S) n peptide that GLPB30 could recognize was then determined. G 4 S peptides of differing lengths (from 5 to 20 AA in length) were generated and binding was assessed via biolayer interferometry (BLI; Table 1).

AACRJournals.org
No binding was observed to (G 4 S) peptides below 10 AA in length [i.e., shorter than (G 4 S) 2 ]. Binding could be observed to the 10 amino acid-long linker (G 4 S) 2 albeit with lower affinity than to the (G 4 S) 4 peptide. At approximately 16 AA in length, binding was similar to that of the 20 amino acid (G 4 S) 4 peptide and leveled off. BsAbs generated with GLPB30 variable domains exhibited a similar binding pattern to that of GLPB30 IgG.

Generation of a Germline ScFv CAR Construct
We hypothesized that a BsAb targeting both the (G4S) n linker on T cells expressing CARs and a TAA could facilitate a cytotoxic T-cell response similar to traditional CAR T cells that directly engage TAAs. To assess this hypothesis, we generated a lentiviral CAR construct encoding the human germline antibody

Design and Characterization of Bispecific Conduit Molecule
BsAbs were then generated to test whether targeting the G 4 S linker in ScFvs could act as effective adapter molecule and facilitate CAR T-cell signaling and cytotoxicity. ScFv fusions linked directly to the C-terminus of IgG heavy chains (24) were employed to generate the bispecific adapters. These molecules were generated both with the anti-TAA binding moiety as the antibody Fab arm and the GLPB30 antibody as the ScFv arm and in the reverse orientation (with the GLPB30 variable domains formatted as a Fab arm and the anti-TAA variables as ScFvs; Fig. 1). In either format, the linker used for ScFv generation was not the G 4 S linker but rather the modified Bird and colleagues linker (described in Methods; ref. 23) that does not bind to GLPB30 (Fig. 2). For the TAA-binding arms, we utilized variable domains binding to either the PSMA or another prostate cancer TAA, transmembrane protein with EGF like and two Follistatin like domains 2 (TMEFF2).

BsAb and Tumor Cells are Required for Expansion, Proliferation, and CD107a Expression In 3-23/B3 CAR T Cells
We next examined whether the presence of BsAb targeting G 4 S linker affected CAR surface expression in 3-23/B3 ScFv-expressing CAR T cells in the absence of tumor cells. TAA-independent adapter CAR activation could lead to unfavorable conditions such as ligand-independent exponential expansion, constitutive cytokine release, and phenotypical changes not conducive to tumor killing (28). Cultured CAR T cells were divided and BsAb (5 μg/mL) was added into cultured CAR T cells while control wells were untreated. Cells were then extensively washed, and CAR surface expression was observed at 48 hours. As shown in Fig. 4A, incubation with soluble BsAb did not alter the surface expression level of CAR, indicating that binding of the BsAbs do not alter surface expression of the CARs.
As incubation with the BsAb alone did not alter CAR surface expression, we next examined whether the addition of tumor cells (in the presence and absence of bispecific molecule) could induce proliferation. We compared proliferation of CFSE-labeled CAR T cells in the presence or absence of BsAb with PSMA-expressing tumor cells. Our results demonstrate increased proliferation of 3-23/B3-bearing CAR T cells (observed in two donors) when BsAbs were added into coculture (Fig. 4B). Notably, some proliferation was observed in the absence of BsAb, but this is likely due to allogeneic reactions with the tumor cells.
CD107a is an effective biomarker of T-cell activation, degranulation, and cytolytic function (25). Using 3-23/B3-expressing CAR T cells, we sought to determine whether the presence of the BsAb targeting G 4 S linker and PSMA could activate CAR T cells and induce CD107a expression in the presence of PSMA + tumor cells. 3-23/B3 CAR T cells were cocultured with PSMAexpressing tumor cells in the presence or absence of BsAb (5 μg/mL). After 5-hour coculture, increased CD107a expression was observed in the total cell population only in the presence of BsAb, suggesting that degranulation occurred in response to BsAb addition (Fig. 4C). GLPB30 antibody was used to detect G4S-bearing CAR T cells (after washing). For both populations of CD8 + cells (CAR + CD8 + and CAR − CD8 + ), we compared CD107a expression. As shown in Fig. 4C

Soluble BsAbs Do Not Induce Cytokine Production in Anti-CD19 CAR T Cells
To further demonstrate that BsAbs targeting the G 4 S linker did not activate CAR T cells in the absence of cells or immobilized antigen, we utilized

anti-CD19 CAR T cells that can be activated by beads immobilized with CD19
antigen. Anti-CD19 CAR T cells were incubated with either soluble recombinant human CD19 extracellular domain-Fc fusion (rhCD19), BsAb, or protein A beads loaded with rhCD19 or BsAb. After 72-hour incubation, IFNγ, IL6, and GM-CSF levels produced by CAR T cells were quantified. We observed that only bead immobilized bispecific molecules or rhCD19 antigen resulted in significantly increased cytokine production in anti-CD19 CAR T cells, whereas soluble BsAb or rhCD19 did not increase cytokine levels above the untreated control ( Supplementary Fig. S2). This result is consistent with 3-23/B3 CAR T-cell data showing soluble BsAbs do not induce proliferation or activation.

Cytotoxic Activity of 3-23/B3 CAR T Cells Is Mediated by BsAbs
We next examined the ability of 3-23/B3 ScFv bearing CAR T cells to lyse tumor cells in the presence of BsAbs. First, BsAbs targeting PSMA and the G 4 S linker were utilized as conduit (adapter) molecules. Anti-PSMA BsAb1 contains GLPB30 Fab arms with an anti-PSMA ScFv appended to the heavy-chain C-terminus. Anti-PSMA BsAb2 uses a reverse orientation, with anti-PSMA Fab arms and a C-terminal GLPB30 ScFv. When these anti-PSMA × anti-G 4 S BsAbs were titrated in the presence of 3-23/B3 CAR T cells and PSMA + PC3 cells (E:T ratio = 5:1), tumor cell lysis was observed using xCELLigence monitoring (Fig. 5A). In wells lacking BsAb, tumor cell growth continued unabated.

3-23/B3 CAR T cells mediated tumor cell-specific lysis in the presence of either
BsAb. Further analysis of this cytotoxicity experiment can be found in Supplementary Fig. S3 where cytotoxicity at the 96-hour timepoint is quantified and statistically significant cytotoxicity is observed at all concentrations of BsAb (compared with the untreated cells).
observed using an anti-TMEFF2 BsAb containing anti-TMEFF2 Fab arms with the GLPB30 ScFv appended to the heavy-chain C-terminus. In this case, 3-23/B3 CAR T cells were shown to kill LNCaP (a prostate-derived tumor cell line) in the presence of BsAb (Fig. 5C). Cytotoxicity and statistical significance of these TMEFF2-positive LNCaP cells is also shown (at the 96-hour timepoint) in Supplementary Fig. S3.
Having demonstrated the cytotoxic potential of conduit CAR T cells in the presence of BsAbs, we next sought to quantify whether cytokines commonly observed upon CAR T-cell activation were also being produced in the presence of BsAbs. IFNγ, IL6, and GM-CSF levels produced by CAR T cells were Data are shown as the mean ± SD. Significance was calculated using one-way ANOVA with multiple comparisons (comparing to CAR T + PC3, no BsAbs; Dunnett test). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.
concentration of 5 μg/mL. The addition of bispecific molecules significantly increased cytokine production by CAR T cells in the presence of target cells (Fig. 5D). We observed IFNγ levels more approximately double in the presence of BsAb1 and triple in the presence of BsAb2. GM-CSF levels increased similarly (compared with CAR T and PC3 alone). IL6 levels were significantly increased in the presence of BsAb2, but not BsAb1.

Cytotoxic Activity of Anti-CD19 CAR T Cells Redirected to Target PC3 Cells
After demonstrating G 4 S-targeting bispecifics could redirect 3-23/B3-bearing CAR T cells, we next examined whether they could be used to redirect CD19targeting CAR T cells to target a novel antigen. We again utilized BsAb1 targeting PSMA and the G 4 S linker as conduit or adapter molecules. BsAb1 and 2 were again titrated in the presence of MYC-tagged anti-CD19 CAR T cells [bearing a (G 4 S) 3 ScFv linker] and PSMA + PC3 cells (E:T ratio = 5:1), tumor cell lysis was again observed using xCELLigence monitoring (Fig. 6A). In wells lacking BsAb1 or 2, tumor cell growth continued unabated, whereas cells treated with BsAb1 were able to lyse the PSMA + PC3 cells. Interestingly, BsAb1 seemed to show better killing using the CD19-targeting CAR T cells than BsAb2. Further analysis of this cytotoxicity experiment can be found in Supplementary  Fig. S3.
Having demonstrated the cytotoxic potential of conduit CAR T cells in the presence of BsAbs, we next sought to quantify whether cytokines commonly observed upon CAR T-cell activation were also being produced in the presence of BsAbs. IFNγ, IL6, and GM-CSF levels produced by CAR T cells were quantified. CAR T cells were cocultured with tumor cells at E:T ratio of 5:1 and bispecific molecules were added at a final concentration of 5 μg/mL. The addition of bispecific molecules significantly increased cytokine production by CAR T cells in the presence of target cells (Fig. 6B). We observed significant increases in IFNγ, GM-CSF levels, and IL6 levels in the presence of BsAb1 and 2.

Discussion
CAR T-cell therapy has proven remarkably effective in treating certain hematopoietic cancers (7). To further refine and advance this transformative therapy, researchers now face the multipronged challenges of understanding and reducing relapse rates, expanding this cell therapy to solid tumors, managing serious adverse events, and lowering costs. One approach that has been proposed to address some of these challenges is to use soluble "adapter" molecules to act as a conduit between CAR T cells and target tumor cells. Most current clinical CAR T-cell candidates, and all the FDA-approved CAR T-cell therapies target one TAA via a fixed, membrane-bound antibody fragment. A drawback of this fixed design is tumors that can downregulate expression or alter the structure or localization of the targeted antigen (antigen escape) can experience relapse when their exogenous CAR T-cell population can no longer engage tumor cells (8). An adapter CAR T-cell approach could offer more flexibility in which TAAs could be targeted and allow for dosing control to reduce toxicity. Our conduit CAR T-cell adapter described herein targets the (G 4 S) n linker present on many existing clinical CAR molecules. The (G 4 S) n linker consists of a repeated G 4 S sequence and is commonly used to connect VH and VL domains in ScFvs. CAR constructs and other biologic therapeutics utilizing this small (∼25 kDa) antibody domain usually contain at least one exposed G 4 S linker [typically (G 4 S) 3 or (G 4 S) 4 ]. Indeed, clinically approved CARs Kymriah (tisagenlecleucel); and Yescarta (axicabtagene ciloleucel) utilizing the FMC63 ScFv contain (G 4 S) 3 linkers.
In this study, we generated and characterized a mAb (GLPB30) that can specifically target the G 4 S linkers and generated bispecific molecules targeting both TAAs and this ScFv linker. These bispecific antibodies were capable of inducing CAR T-cell proliferation, and activating CAR T cells in the presence of BsAb-coated beads or BsAb-treated tumor cells (but not soluble BsAb alone). In addition, we were able to demonstrate redirected killing of PSMA + cells using both CD19-targeting CAR T cells as well as 3-23/B3-bearing CAR T cells (lacking inherent TAA specificity).
Use of this G 4 S-targeting adapter approach could allow for existing CAR T-cell therapies to be redirected to different TAAs upon relapse. Patients experiencing relapse due to loss or downregulation of CAR-targeted antigen (e.g., CD19), could be treated with G 4 S binding adapters to target additional TAAs. Alternatively, bispecific adapters could be utilized during initial CAR treatment, prior to any potential relapse. Utilizing adapter bispecifics while CARs are newly primed and activated may expand the target cell population and reduce the odds of relapse due to antigen loss or downregulation. Currently, all CD19targeting CARs approved by the FDA have G 4 S linkers that can be targeted via an anti-G 4 S linker adapter molecule. If other ScFv linkers are used for future CARs, applying this approach to those CARs would of course require the generation of new reagents specific for those linkers. A methodology has been described using CD19 antigen fused to TAA-targeting moieties to redirect CD19-targeting CARs (29). This approach is of course limited to use with CD19 CARs. A G 4 S linker targeting adapter molecule may be more broadly applicable to any G 4 S-containing CAR T-cell therapeutics.
In addition to use in conjunction with existing CAR T-cell therapies, G 4 Stargeting adapter molecules could be used with CARs lacking affinity to any TAA (as described herein with the 3-23/B3 bearing CAR). An advantage of this approach is reduced safety concerns as the presence of an adapter or conduit bispecific molecule controls CAR T-cell proliferation and activation. By halting dosing of the adapter molecule, CAR proliferation and activation will stop (once the adapter is cleared), thus eliminating the need for a suicide gene switch engineered into some novel CAR T-cell therapeutics (30).
Several adapter CAR designs have been proposed, each with various potential advantages and disadvantages (14,(31)(32)(33)(34). The reliance on nonhuman proteins for an adapter or CAR molecule such as the yeast transcription factor GCN4 (22) or Streptococcus pyogenes derived SpyCatcher/SpyTag systems (31) could be disadvantageous due to potential immunogenicity. Although the G 4 S linker itself (a sequence not found in the human proteome) could potentially pose an immunogenicity risk as well, this has yet to be observed in multiple approved clinical therapeutics bearing this linker. Nevertheless, there is of course the chance of immunogenicity or an anti-drug antibody (ADA) response to the linker that may interfere with this approach. Indeed, preexisting ADA to G 4 S linker-containing ScFvs has been observed before (although it is unclear whether the G 4 S linker was the epitope; ref. 35) and O-linked glycans on G4S linkers (a heterogeneity concern) could also pose an immunogenicity risk (36).
Prior to clinical studies, further refinement of the conduit CAR approach will be necessary. In this study, we used a rabbit-derived anti-G 4 S variable domain, but for clinical purposes, this molecule would likely be humanized to further reduce immunogenicity. Another step before clinical validation, would be to confirm a lack of off-target binding of the G 4 S binding antibody. This will be needed to ensure the adapter BsAb will have specificity for only linkers displayed on the CAR surface. Furthermore, a conduit CAR adapter approach will likely be limited clinically to patients not being cotreated with additional G 4 S-containing therapeutics as these may interfere with conduit-CAR T-cell targeting. While this study provides initial proof of concept for an ScFv-linker targeting adapter CAR in vitro, multiple in vivo studies using appropriate xenograft or syngeneic models may be required to refine and validate this concept further. COVID-19related resource constraints and new prioritizations prevented the completion of lengthy in vivo studies for this project. Nevertheless, due to the rapid evolution and impact of work within the CAR T field, we strongly believe it is very timely to present this emerging strategy for improving the specificity and utility of CAR T cells for treating cancer. We believe comprehensive in vivo models are required to advance this concept and prior to clinical development of programs using this technology, yet that was unfortunately not possible due to current constraints mentioned above.
Conduit CAR (along with other adapter or universal CAR formats), offer the advantages of being able to fine tune and control CAR function to quickly meet emerging clinical understanding of optimal CAR properties. It has recently been shown that decreasing affinity of CAR molecules can increase proliferation and antitumor activity in a clinical setting using anti-CD19 CAR T cells (37). Adapter CARs, such as the system described in this study, allow for tuning of affinity and avidity of adapters to explore optimal properties for the cytotoxic immune synapse via changing intrinsic affinity, stoichiometry, and valency.