High throughput development of TCR-mimic antibody that targets survivin-2B80-88/HLA-A*A24 and its application in a bispecific T-cell engager

Intracellular tumor-associated antigens are targeted by antibodies known as T-cell receptor mimic antibodies (TCRm-Abs), which recognize T-cell epitopes with better stabilities and higher affinities than T-cell receptors. However, TCRm-Abs have been proven difficult to produce using conventional techniques. Here, we developed TCRm-Abs that recognize the survivin-2B-derived nonamer peptide, AYACNTSTL (SV2B80-88), presented on HLA-A*24 (SV2B80-88/HLA-A*24) from immunized mice by using a fluorescence-activated cell sorting-based antigen-specific plasma cells isolation method combined with a high-throughput single-cell-based immunoglobulin-gene-cloning technology. This approach yielded a remarkable efficiency in generating candidate antibody clones that recognize SV2B80-88/HLA-A*24. The screening of the antibody clones for their affinity and ability to bind key amino-acid residues within the target peptide revealed that one clone, #21-3, specifically recognized SV2B80-88/HLA-A*24 on T2 cells. The specificity of #21-3 was further established through survivin-2B-positive tumor cell lines that exogenously or endogenously express HLA-A*24. A bispecific T-cell engager comprised of #21-3 and anti-CD3 showed specific cytotoxicity towards cells bearing SV2B80-88/HLA-A*24 by recruiting and activating T-cells in vitro. The efficient development of TCRm-Ab overcomes the limitations that hamper antibody-based immunotherapeutic approaches and enables the targeting of intracellular tumor-associated antigens.

We recently developed a fluorescence-activated cell sorting (FACS)-based antigen-specific plasma cells (PCs) isolation method, which we termed (ERIAA), for rapid and scalable automation in monoclonal antibody (mAb) generation 10 . ERIAA is based on features of PCs, abundant cytoplasmic rough endoplasmic reticulum (ER) that can be used for PC identification with ER-specific florescent dye (ER-tracker) and weakly expressed cell surface IgG that can be used as a tag for a complementary fluorescently labeled antigen. By applying ERIAA with fluorescently labeled HLA-tetramers and a high-throughput single-cell-based immunoglobulin-gene-cloning technology, we succeeded to generate mAb clones that bind to the SV2B 80-88 /HLA-A*24 from SV2B 80-88 / HLA-A*24-immunized mice. The mAb clones were analyzed for their affinity and ability to bind key amino acid residues within the SV2B 80-88 , which revealed that one mAb clone (#21-3) showed the highest binding specificity with the equilibrium dissociation constant (K D ) = 7.4 nM. A bispecific T-cell engager (BiTE) comprised of #21-3 and CD3 showed specific cytotoxicity towards cells bearing the SV2B 80-88 /HLA-A*24 in vitro by crosslinking T-cells to the target cells. To the best of our knowledge, this is the first TCRm-Ab that targets the antigenic peptide derived from intracellular survivin-2B in the context of HLA-A*24.

Results
Isolation of candidate TCRm-Abs that target SV2B 80-88 /HLA-A*24. First, we attempted to develop TCRm-Abs by conventional B-cell hybridoma from splenocytes of mice that were immunized with SV2B 80-88 / HLA-A*24 as an antigen. Upon screening of 1,000 hybridoma clones by ELISA, five clones showed a positive signal with the SV2B 80-88 /HLA-A*24 monomer. Additional ELISA screening of these clones was conducted with a panel of irrelevant peptide/HLA-A*24 monomers, HIVgp160 (HIV), NY-ESO, SOX2-1, SOX2-2, MAGE3A-1 and MAGE3A-2; this analysis revealed that only one clone (5FG) showed the required specificity for SV2B 80-88 / HLA-A*24 (Fig. 1a). The low probability of obtaining a candidate TCRm-Ab by hybridoma prompted us to use our recently developed ERIAA and high-throughput single-cell-based immunoglobulin-gene-cloning technology. The splenocytes from the immunized mice were stained with anti-mouse IgG, ER-tracker and SV2B 80-88 / HLA-A*24 tetramer; the SV2B 80-88 /HLA-A*24-specific PCs gated as IgG Medium ER-tracker High and SV2B 80-88 / HLA-A*24 High (R3 gate) were single-sorted by FACS (Fig. 1b). Single cell-based immunoglobulin heavy chain variable (V H ) and light chain variable (V L ) gene amplification was conducted by PCR of the R3-gated cells, followed by the DNA transfection of cognate pairs of immunoglobulin heavy and light chain gene into 293FT-cells, which resulted in the production of recombinant mAbs. The screening of 96 clones by ELISA with a panel of peptide/HLA-A*24 monomers revealed that 47 clones bound to the SV2B 80-88 /HLA-A*24 monomer, among which six clones (#33-3, #34-23, #21-3, #21-34, #1-5 and #2-41) did not bind to six irrelevant peptide/HLA-A*24 monomers (Fig. 1a). The DNA sequencing revealed that these mAb clones were divided into three phylogenetic clusters (Fig. 1c). We selected four mAb clones from each cluster (5FG, 21-3, 21-34 and 1-5) and analyzed their specificity with T2 cells stably expressing HLA-A*24 (T2/A24). As shown in Fig. 1d, all mAbs appeared to bind to SV2B 80-88 -pulsed T2/A24 cells but not to HIV-pulsed cells. To map key amino-acid residues that are involved in antibody interactions, each residue on the SV2B 80-88 was replaced with glycine (except the canonical anchor residues on positions 2 and 9), and mAb binding was assessed on T2/A24 cells. As shown in Fig. 2a, #21-3 showed the widest epitope coverage; substitutions on positions 4, 5, 6 and 8 abrogated the binding; and position 7 reduced the binding by 52%. However, substitutions on either position 1 or 3 did not abrogate the binding. On #21-34, substitutions on position 4 abrogated the binding and that on any of positions 5, 6, 7 or 8 reduced the binding by 52~75%; however, substitutions on either position 1 or 3 did not abrogate the binding. #5FG and #1-5 were relatively insensitive to substitution at all positions except position 4; their recognition patterns correspond to the phylogenetic data shown in Fig. 1c. Based on the glycine substitution analysis, #21-3, reacting with the C-terminus of SV2B 80-88 , was selected and subjected to further analysis.

#21-3BiTE directs T-cells to kill human tumor cell lines in vitro.
To establish the ability of the #21-3 to inhibit tumor growth in vitro, we generated a BiTE by fusing an anti-CD3 single-chain variable fragment (scFv) to the C terminal of the full length of #21-3 (#21-3BiTE) (Fig. S1a,b). To ascertain that #21-3BiTE binds simultaneously to CD3 and SV2B 80-88 /HLA-A*24, PBMC from a HLA-A*24 − healthy donor and WiDr cells were stained with #21-3BiTE. The anti CD3 ScFv arm of #21-3BiTE stained CD4 + and CD8 + T-cells, but the binding was 10-fold lower than that of the parent anti-CD3 antibody (OKT3) (Fig. S1c,d). #21-3BiTE retained a similar binding specificity to WiDr cells as the parent antibody (Fig. S1e). These results demonstrate that #21-3BiTE contains the antigen-binding capacity of each of the parent antibodies.
To assess if #21-3BiTE activates naïve T-cells by bridging T-cells and target cells, PBMC from a HLA-A*24 + healthy donor were cultured with #21-3BiTE, #21-3 or control mouse IgG in the presence or absence of WiDr cells. T-cell activation was measured by assessing the expression of the activation marker CD69. Upregulated CD69 expression on both CD4 + and CD8 + T-cells was observed only when PBMC were incubated with #21-3BiTE and WiDr cells (Fig. 4a). To determine if the BiTE drives T-cell proliferation, CellTrace-Blue-stained T-cells expanded from a HLA-A*24 + healthy donor were cultured with #21-3BiTE, #21-3 or control mouse IgG in the presence or absence of WiDr for four days. Approximately 50% of the T-cells proliferated in the presence of #21-3BiTE and the tumor, whereas T-cells cultured with or without WiDr cells in the presence of #21-3 or control mouse IgG did not show potent proliferation (Fig. 4b).
When T-cell activation was further evaluated based on secretion of the interferon (IFN)-γ, only T-cells incubated with WiDr cells in the presence of #21-3BiTE showed an elevated expression of IFN-γ. T-cells incubated without WiDr cells in the presence of #21-3BiTE secreted low level of IFN-γ, which was probably due to the non-specific T-cell activation mediated through the anti CD3 arm of #21-3BiTE (Fig. 4c).
Next, the ability of #21-3BiTE to crosslink T-cells to target cells was analyzed. Incubation of SV2B 80-88 -pulsed T2/A24 cells with T-cells expanded from a HLA-A*24 + healthy donor in the presence of #21-3BiTE induced potent T-cell cytotoxicity in a dose-dependent manner. In contrast, T-cell-mediated cytotoxicity of HIV-pulsed T/A24 cells by #21-3BiTE was relatively ineffective (Fig. 5a). #21-3BiTE also showed specific cytotoxicity against HeLa/A24 cells in co-cultures of the T-cells (Fig. 5b). Next, we tested the therapeutic efficacy of #21-3BiTE by using cell lines endogenously expressing both surviving-2B and HLA-A*24. WiDr and Colo320DM cells stably expressing luciferase were mixed with the T-cells expanded from a HLA-A*24 + healthy donor in the presence of #21-3BiTE, #21-3 or control mouse IgG, and cytotoxicity was measured. As shown in Fig. 5c,d, specific cytotoxicity was only observed on #21-3BiTE but neither on #21-3 nor on control mouse IgG. #21-3BiTE-induced T-cell cytotoxicity against WiDr cells was inhibited by the addition of the SV2B 80-88 /HLA-A*24 monomer in a dose-dependent manner (Fig. 5e). The BiTE-dependent T-cell-mediated cytotoxicity of Colo201 cells was relatively ineffective (Fig. 5f). Similar results were also observed when using T-cells expanded from a HLA-A*24 − donor ( Fig. S2a-d). These results indicate that reduced tumor cell viability is induced by specific binding of T-cells to SV2B 80-88 /HLA-A*24 on tumor cells through #21-3BiTE, supporting the conclusion that #21-3BiTE has a potential for therapeutically targeting tumor cells that express SV2B 80-88 /HLA-A*24.

Discussion
Therapeutic antibodies are one of the most successful biological drugs used to treat cancers. However, their application has been limited to extracellular or cell-surface proteins because antibodies are too large to cross the cell membrane to target intracellular proteins. TCRm-Abs can recognize intracellular antigens in the form of peptides loaded on the HLA, thereby increasing the opportunity to expand the repertoire of therapeutic antibodies. Several groups have attempted to generate TCRm-Abs by conventional hybridoma techniques; however, these attempts have shown a very low probability of success because many antibodies bind epitopes that are not directly involved in the target peptide presented on HLA [11][12][13][14][15][16] . Our attempts to isolate TCRm-Ab by hybridoma also resulted in low efficiency. Overall, obtaining specific TCRm-Ab has been proven to be difficult by conventional hybridoma techniques. Our FACS-based strategy combined with single-cell immunoglobulin-gene-cloning technology enables to isolate rare antigen-specific PCs, resulting in a remarkable efficiency in generating candidate mAbs that recognize SV2B 80-88 in complex with HLA-A*24 from antigen-immunized mice. One selected mAb clone, #21-3, was capable of recognizing SV2B 80-88 /HLA-A*24, discriminating normal PBMC from tumor cells. #21-3-derived BiTE selectively mediates anti-tumor reactivity, indicating a potential therapeutic tool capable of targeting malignancies.
TCRm-Abs have been known to recognize off-target epitopes that share homologous peptide sequences 4,17 . Given that #21-3 primarily contacts the C-terminal half of SV2B 80-88 (positions 4~8), this TCRm-Ab may cross-react with antigenic peptides sharing the key amino acids essential for antibody binding. In addition to the off-target binding, the risk of on-target/off-tumor recognition caused by target antigen expression on normal cells is also a major obstacle for antibody-based therapies 18 . Survivin-2B has been found to be expressed by several stem cells, thereby #21-3 might have off-tumor's reactivity that target stem cells 19 . We have shown that #21-3 did not bind to PBMC from a healthy HLA-A*24 + donor, which suggested there was no broad expression of off-target and off-tumor epitopes on normal cells, at least in differentiated haematopoietic cells.
represent the sequential gating strategy. (I) FSC vs SSC with gate R1 represent lymphocytes. (II) The antimouse IgG Low ER-Tracker High fraction was defined as PCs. (R2). (III) The SV2B 80-88 /HLA-A*24-specific PCs were defined as IgG Medium ER-Tracker High SV2B 80-88 /HLA-A*24 High (R3 gate). Numbers indicate the percentages of cells in the gated area. 50,000 events were recorded. (c) Phylogenetic analysis of V H and V L amino-acid sequences of SV2B 80-88 /HLA-A*24-specific mAb clones. (d) FACS analysis of the candidate mAbs by T2/A24 cells. T2/A24 cells (1 × 10 5 ) pulsed with either SV2B 80-88 or HIV were stained with crude candidate mAbs. The binding ability of each mAb was evaluated by mean fluorescent intensity (MFI) of stained T2/A24 cells.
Immunization. Groups of three mice were immunized at 3-week intervals for a total of 3 times by intraperitoneal injection of SV2B 80-88 /HLA-A*24:02 monomer antigen (50 µg) plus TiterMax adjuvant, followed by an intraperitoneal injection of antigen alone 7 days before harvesting splenocytes.
Generation of tCRm antibodies. Isolation of TCRm-specific PCs was performed as previously described, with slight modifications. Mouse splenocytes (1 × 10 7 /mL) were suspended in 1 mL of PBS-BSA and stained with PE-labeled SV2B 80-88 /HLA*A24:02-tetramer (0.1 µg/mL) and fluorescently labeled antibody against mouse IgG at 4 °C for 30 min with gentle agitation 23 . After washing with phosphate buffered saline (PBS), the cells were suspended in 4 mL of PBS containing ER-tracker (0.25 µM) and, subsequently, analyzed by FACS. The forward-versus-side-scatter (FSC vs SSC) lymphocyte gate (R1) was applied to exclude dead cells. The PCs (IgG Medium ER-tracker High , R2 gate) were further subdivided into fractions according to their binding of fluorescently labeled HLA tetramer to define the antigen-specific PCs (IgG Medium ER-tracker High SV2B 80-88 / HLA-A*24 High ). Single-cell sorting was performed using a JSAN Cell Sorter that was equipped with an automatic cell deposition unit (http://baybio.co.jp/english/top.html) with fluorescently labeled antibody against IgG monitored in the FL-l channel, PE-labeled SV2B 80-88 /HLA-A*24-tetramer in the FL-2 channel and ER-tracker in the FL-7 channel. Molecular cloning of V H and V L genes from single cells and recombinant antibody expression were performed as previously described 24 . Antibodies were produced via the Expi293 cell culture system according to the manufacturer's protocol (Thermo Fisher Scientific). Antibodies were purified via protein G column chromatography and then analyzed by SDS-PAGE. Hybridoma production was outsourced to TransGenic Inc. (http://www.transgenic.co.jp/en/). Large-scale production of #21-3BiTE was outsourced to Absolute Antibody Ltd. (https://absoluteantibody.com).
eLIsA. Streptavidin-coated 96-well plates (SUMITOMO BAKELITE, https://www.sumibe.co.jp) were coated with 50 ng of biotinylated peptide/HLA-monomers in 50 μl of PBS at 4 °C overnight. Plates were washed with PBS, and 100 µl of crude mAbs were added to each well. After 1 h incubation at room temperature, bound mAbs were detected by alkaline phosphatase-conjugated goat anti-mouse IgG (Sigma) with KPL BluePhos Microwell Phosphatase Substrate System (Roche, https://www.roche.com) and then quantified with a Varioskan LUX multimode reader (Thermo Scientific).
Antibody affinity and kinetic assay. SPR single-cycle kinetic experiments were performed by an indirect capture method using a Biacore T100 instrument (GE Healthcare, https://www.gelifesciences.com/en/us). Briefly, two adjacent channels on a CM5 sensor chip were immobilized with IgG binder using Mouse Antibody Capture Kit according to the manufacturer's recommendations. Purified TCRm-Ab was injected at a flow rate of 10 μl/min to achieve a ligand immobilization level of 200 to 400 relative units. Five different concentrations of each of the HLA monomer (ranging from 0 nM to 30 nM) were then injected at a flow rate of 30 μl/min for 1 min, followed by 5 min of washing with HBS-P buffer. The single-cycle kinetic curves were fitted with 1:1 binding stoichiometry for ka, kd and K D .