Target Antigen Density Governs the Efficacy of Anti–CD20-CD28-CD3 ζ Chimeric Antigen Receptor–Modified Effector CD8⁺ T Cells Free

This article is featured in In This Issue, p.845

Chimeric Ag receptor (CAR)–transduced T (CAR-T) cell therapy is an emerging therapeutic strategy for refractory acute lymphoblastic leukemia (ALL) and chronic lymphocytic leukemia (CLL) (1, 2). Second- and later-generation CARs generally consist of a single-chain variable fragment (scFv) from a mAb fused to the signaling domain of CD3ζ, and contain one or two costimulatory endodomains, respectively (3–5). This technology has two main potential benefits over TCR gene insertion. One is that Ag recognition by CAR is independent of HLA, meaning that CAR therapy can be used to treat all Ag-positive patients regardless of their HLA. The other is that once CARs ligate to target molecules, full activation signals, including costimuli such as CD28 or 4-1BB, are transmitted to CAR-T cells (3–5). A superior effector function and proliferation following activation have been reported in second- and third-generation CAR-T cells (6–9).

In contrast, CAR-T cells may induce adverse effects by recognizing low expression levels of the target Ag in an off-target organ. This activity has been referred to as the “on-target/off-tumor effect.” A serious adverse event induced by CAR-T cells, which recognize very low expression levels of ERBB2 on lung epithelial cells, was reported with CAR therapy targeting ERBB2 based on trastuzumab (Herceptin) (10). Although ERBB2 is expressed at low levels in various normal tissues, including lung, the anti-ERBB2 humanized mAb trastuzumab has been used safely in clinical settings (11), indicating that ERBB2 expression levels on lung cells are negligible in terms of trastuzumab therapy (12). However, ERBB2–CAR-T cells induce significant Ag-specific responses against this low expression of ERBB2 (10, 11). Therefore, selection of a target Ag is critical for both efficacy and avoiding adverse effects. TCRs recognize very low numbers of peptide/HLA complexes, whereas a relatively high number of target molecules are required for mAbs to induce cytotoxic activity (13, 14). However, the range of Ag density in which CAR-T cells can recognize and induce cytotoxicity has not been investigated in detail. Furthermore, research has not yet clarified the number of Ag molecules expressed that could be candidates for targets when expressed at low levels or that should be avoided owing to the on-target/off-tumor effect (15).

CD20 is an activated glycosylated phosphoprotein that is expressed on the surface of B lymphocytes. An anti-CD20 mAb is an effective therapeutic option for various B cell malignancies such as ALL (16), CLL (17), and malignant lymphoma (18, 19). Although combination chemotherapies with rituximab have achieved favorable results in CD20-positive B cell lymphoma patients, acquired resistance to rituximab has become a problem, with a suggested mechanism of reduced expression of CD20 (20–24). Accordingly, a therapeutic option that efficiently eradicates target cells expressing low levels of CD20 that survive rituximab or ofatumumab (ofa) therapy needs to be developed. Therefore, we developed a novel CD20-CAR and investigated the minimum threshold Ag expression level required for lysis of target cells and activation of CAR-T cells. To avoid possible immunological rejection against anti-mouse Abs, we used a humanized anti-CD20 mAb to construct CD20CAR (25). We also assessed its effects against tumor cell lines and primary cells isolated from mAb therapy-refractory, CD20-downregulated B cell tumors (24, 26, 27).

K562, CCRF-CEM, SU-DHL-4, SU-DHL-6, SU-DHL-10, Raji, RRBL1, and WILL2 cells were cultured in RPMI 1640 medium. OCI-Ly3 and OCI-Ly10 cells were kind gifts from Dr. K. Takeyama (Dana-Farber Cancer Institute, Boston, MA) and were cultured in IMDM (Sigma-Aldrich, St. Louis, MO). Each type of medium contained 10% FBS, 0.8 mM l-glutamine, and 1% penicillin-streptomycin. RRBL1 and WILL2 cells are cell lines established from a B cell lymphoma patient who exhibited CD20-negative phenotypic changes after repeated chemotherapy with rituximab (26, 27). CD20-transduced CCRF-CEM cell lines (CD20-CEMs) expressing various levels of CD20 were described elsewhere (28). CD20-transduced K562 (CD20-K562) cells were generated by retroviral transduction with the full-length CD20 molecule, as described (29).

Primary B cell tumor cells were obtained from PBMCs (CLL patient) or pleural effusion (lymphoma patient) according to protocols approved by the Institutional Review Board of Nagoya University School of Medicine, and written informed consent was obtained from each patient in accordance with the Declaration of Helsinki.

CD20 molecules expressed on the surface of CD20-CEMs or other cell lines were quantified using quantitative immunofluorescence indirect assay (QIFIKIT; Dako, Glostrup, Denmark). Briefly, cells were stained with unlabeled anti-CD20 mouse mAb (BD Bioscience, San Jose, CA) or purified mouse IgG-κ (BioLegend, San Diego, CA) as an isotype control. The cells of interest and calibration beads from the kit were then simultaneously labeled with primary mAb, followed by FITC-conjugated goat anti-mouse secondary Ab staining. Labeled cells and calibration beads were analyzed on a flow cytometer, and a standard regression line between fluorescence intensity and Ag density that was expressed as Ab-binding capacity (ABC) in molecules per cell was calculated. Finally, the specific ABC (sABC) was determined by subtracting the background Ab equivalent of the isotype control from ABC (30).

CD20-binding scFv was constructed based on the reported sequences of the humanized anti-CD20 mAb (OUBM mAb) (25). OUBM mAb exhibits high CD20 binding affinity (K_D, 10.09 nM). H chain and L chain V region segments were linked with an 18-aa linker. scFv was then fused to a human IgG₄ hinge, a CD3-ζ chain, a CD28 costimulatory domain, and a truncated version of the epidermal growth factor receptor (tEGFR) that lacked epidermal growth factor binding and intracellular signaling domains downstream of the self-cleaving T2A sequence (31–33). By inserting the T2A sequence between CD20CAR and tEGFR, the two proteins were coexpressed at equimolar levels from a single transcript. Cell-surface tEGFR was detected using the biotinylated anti-EGFR mAb Erbitux (Bristol-Myers Squibb, New York, NY). The CD20CAR transgene was assembled by overlap extension PCR (34). CD20CAR was inserted into LZRS-pBMN-Z, using HindIII and NotI sites, and the CD20CAR-encoding retrovirus was produced using the Phoenix-Ampho system (Orbigen, San Diego, CA) and concentrated with Retro-X Concentrator (Clontech Laboratories, Mountain View, CA).

The PBMCs of a normal donor were isolated by centrifugation of whole blood using Ficoll-Paque (GE Healthcare, Wauwatosa, WI). CD8⁺ lymphocytes were then purified with immunomagnetic beads (Miltenyi Biotec, Bergisch Gladbach, Germany), activated with anti-CD3/CD28 beads (Invitrogen, Carlsbad, CA), and transduced on day 3 after activation with the recombinant human fibronectin fragment (RetroNectin, Takara Bio, Otsu, Japan) by centrifugation at 2100 rpm for 45 min at 32°C with the retroviral supernatant (multiplicity of infection = 3). T cells were expanded in RPMI 1640 medium containing 10% human serum, 0.8 mM l-glutamine, 1% penicillin-streptomycin, and 0.5 μM 2-ME and supplemented with recombinant human IL-2 to a final concentration of 50 IU/ml. CAR-positive cells were enriched using immunomagnetic selection with biotin-conjugated anti-EGFR mAb and streptavidin beads (Miltenyi Biotec). The transduced T cells were expanded in culture by plating with γ-irradiated EBV-transformed lymphoblastoid cell line (LCL) at a T cell to LCL ratio of 1:7 and supplemented with IL-2 to 50 IU/ml (29).

All samples were analyzed with flow cytometry (FCM) on the FACSAria instrument (BD Biosciences), and data were analyzed using FlowJo software (Tree Star, Ashland, OR). Biotinylated Erbitux and streptavidin-PE were used to identify T cells that expressed tEGFR.

For the [⁵¹Cr] release assay, target cells were labeled for 2 h with [⁵¹Cr] (PerkinElmer, Waltham, MA), washed twice, dispensed at 2 × 10³ cells per well into triplicate cultures in 96-well round-bottom plates, and incubated for 4 h at 37°C with CD20CAR–T cells at various E:T ratios. Percent of specific lysis was calculated using a standard formula [(experimental − spontaneous release)/(maximum load − spontaneous release) × 100 (%)] and expressed as the mean of triplicate samples. Regarding the coculture assay, CEMs were labeled with 0.1 μM CFSE (Invitrogen), washed, and plated with CD20 CAR-T cells at a ratio of 1:1 without IL-2 supplementation. After a 72-h incubation, cells were stained with anti-CD8 mAb and analyzed with FCM. The percentages of CAR-T cells and CEMs within the live cell gates were assessed.

CD20CAR–T cells and K562 or CCRF-CEM cells that expressed CD20 were mixed at a 1:1 ratio in the presence of brefeldin A (Sigma-Aldrich) and then fixed and permeabilized with Cell Fixation/Permeabilization Kits (BD Biosciences) for intracellular cytokine assay. After fixation, T cells were stained with anti–IFN-γ and anti–CD8-allophycocyanin mAb (BD Biosciences). As a positive control for cytokine production, cells were stimulated with 10 ng/ml PMA and 1 μg/ml ionomycin (Sigma-Aldrich). CD20CAR–T cells and CEMs for the cytokine secretion assay were plated at an E:T ratio of 1:1, and IFN-γ, TNF-α, and IL-2 in the supernatant were measured with ELISA (BD Biosciences) after 16 h of incubation.

CD20CAR–T cells were labeled with 0.2 μM CFSE, washed, and then plated with stimulator cells at a ratio of 1:1 without IL-2 supplementation. After a 72- or 96-h incubation, cells were stained with the anti-CD8 mAb, samples were analyzed with FCM, and the division of live CD8⁺ T cells was assessed with CFSE dye dilution.

CD20CAR–T cells and CD20-CEM cells expressing various levels of CD20 were mixed at a 1:5 ratio, centrifuged briefly, and incubated for various times at 37°C. Cells were then fixed by the addition of BD Cytofix Fixation Buffer at 37°C for 10 min, permeabilized in ice-cold BD Phosflow Perm Buffer III, and incubated on ice for 30 min (BD Biosciences). P-p44/42 MAPK (T202/Y204) or P-Zap-70 (Y319)/SyK(Y532) Rabbit Ab (Cell Signaling Technology, Danvers, MA) and bovine anti-rabbit IgG-FITC as a secondary Ab (Santa Cruz Biotechnology, Dallas, TX) were used for phospho-specific staining.

Differences among results were evaluated with one-way or two-way ANOVA analysis and the Bonferroni test, as appropriate. Differences were considered significant when p < 0.05. Statistical analysis was performed using GraphPad Prism Version 5 software.

To develop functional CD20CAR, we constructed CD20CAR consisting of anti–CD20-scFv linked to CD3ζ, a CD28 costimulatory domain, and a tEGFR; CD8⁺ T cells were then retrovirally transduced with CD20CAR (Fig. 1A). After one course of stimulation and transduction, the expression of CD20CAR generally reached 40–80%. To determine transduction efficiency, CD20CAR and tEGFR were labeled with an anti-Fc Ab and biotinylated Erbitux, respectively. The expression of tEGFR reflected that of CAR on the transduced T cells, and we verified that the expression of CAR and tEGFR was similar after each transduction experiment (Fig. 1B) (32). Transduction efficiency could be monitored with tEGFR with high reproducibility (Fig. 1B, right panel). Using intracellular staining, we assessed the ability of CAR-T cells to produce IFN-γ in response to CD20. Stimulation with CD20-K562 cells induced robust production of IFN-γ, whereas mock-transduced K562 cells did not (Fig. 1C). These results demonstrated that CD20CAR–T cells recognized CD20 in an Ag-specific manner. After the transduction culture, CD20CAR-positive cells were enriched to a purity of >95% with biotinylated Erbitux and anti-biotin immunomagnetic beads (32), expanded by stimulating with a γ-irradiated LCL, and then used for subsequent experiments. The expression of CAR/tEGFR before and after LCL stimulation was sufficiently maintained (Fig. 1D). The ability of CD20CAR–T cells to lyse CD20⁺ target cells was assessed after one course of transduction and expansion. CD20CAR–T cells specifically lysed CD20-K562 cells (Fig. 1E) in a highly reproducible manner (Fig. 1E, right panel). To examine background cytotoxicity, CD19CAR (non–target-specific CAR)–transduced T cells were examined for cytotoxicity against K562 or CD20-K562. Both experiments demonstrated almost the same range of cytotoxicity by the CD20CAR–T cells against K562 as in Fig. 1E. The range of cytotoxicity was 7–11% at an E:T ratio of 10:1 (n = 4). Two repeated LCL stimulations caused a log-scale expansion that resulted in 10,000-fold expansion of CD20CAR–T cells (Supplemental Fig. 1). The CD20CAR–T cells almost uniformly demonstrated effector phenotype (CD28⁻, CD62L⁻, CD45RO⁺) after LCL stimulation (data not shown). In all subsequent experiments, CD20CAR^– T cells were selected with tEGFR and expanded with one course of LCL stimulation; thus the transduction level of CD20CAR was uniformly >95% (Fig. 1D).

Although CAR-T cells very efficiently recognize targets, the range of target molecule expression to which CAR-T cells can respond remains unknown (3–5). To assess this range more precisely, the number of CD20 molecules expressed on the cell surface of various cell lines was quantified as the CD20-specific Ab-binding capacity (CD20-sABC) on a per cell basis. We obtained 30 clones of CD20-CEMs expressing various levels of CD20 for use as target cells or stimulators (28). Of these, expression of CD20 by four representative clones was depicted, and the cells were used for several subsequent experiments as stimulators [CD20–very low CEM (VL-CEM) (CD20-mean fluorescence intensity [MFI]: 126/sABC: 240 molecules); CD20–low CEM (L-CEM) (CD20-MFI: 576/sABC: 5320 molecules); CD20–medium CEM (M-CEM) (CD20-MFI: 2396/sABC: 26,900 molecules); and CD20–high CEM (H-CEM) (CD20-MFI: 11,388/sABC: 142,722 molecules)] (Fig. 2A, Table I). The CD20-sABC values were 500,000 molecules for the germinal center B cell–type diffuse large B cell lymphoma (DLBCL)–derived cell lines; SU-DHL-4, -6, and -10, and 100,000 molecules for the non–germinal center B cell–type DLBCL-derived cell lines Ly-3 and -10 (Fig. 2B, Table I). RRBL1 and WILL2 are cell lines established from patients who experienced a relapse in B cell lymphoma with very weak expression of CD20 and who became resistant to rituximab (26, 27). The expression levels of CD20 by RRBL1 and WILL2 cells were 15,632 and 4869 molecules per cell, respectively (Fig. 2B, ◇ and ×, respectively). Relative to other cell lines, CD20-CEMs represented a very wide range of CD20 expression, from 240 to 230,546 molecules per cell, which was considered very low to high (Fig. 2B, Table I).

To evaluate the potential influence of costimulation, inhibitory signals, and adhesion molecule, the expression of CD80, CD86, CD54 (ICAM-1), CD58 (leukocyte function–associated molecule-3), and PD-L1 on target tumor cells was investigated. CEM cells demonstrated a tolerogenic phenotype, expressing low levels of CD80 and CD86 and relatively high levels of the inhibitory ligand PD-L1. CD54 was positive in all examined cell lines, whereas CD58 was negative in WILL2 cells (Supplemental Fig. 2) (35).

The level of CD20 Ag expression for rituximab-induced complement-dependent cytotoxicity (CDC) was determined using the same set of CD20-CEMs (28). We performed the rituximab-induced CDC assay and obtained almost the same results using human complement (Supplemental Fig. 3A). As demonstrated previously, Ab-dependent cellular cytotoxicity (ADCC) with rituximab against CD20-CEMs did not show a clear threshold of CD20 expression (data not shown) (28). CD20-CEMs with an MFI <1000 (equivalent to sABC of 10⁴) did not induce significant CDC, whereas CD20-CEMs with an MFI of 1000–3000 (equivalent to sABC of 10⁴–10⁵) did. CD20-CEMs with an MFI >3000 effectively induced cytotoxicity, and maximal CDC was obtained at an MFI >5000 (sABC of 10⁵) (Supplemental Fig. 3A). CDC induced by the humanized anti-CD20 mAb OUBM was also examined. OUBM mAb, with which CD20CAR was constructed, induced marked CDC with half-maximum cytotoxicity at a CD20 expression level similar (MFI of 3000) to that of rituximab (MFI of 3000) (Supplemental Fig. 3).

In contrast to the weak CDC caused by rituximab and OUBM mAb, CD20-CEMs were more efficiently lysed by CD20CAR–T cells, with the exception of VL-CEMs, which underwent a significantly lower degree of lysis (Fig. 2C).

To determine the threshold expression level of the CD20 Ag required to induce CAR-T cytotoxicity, we performed a [⁵¹Cr] release assay with CD20CAR–T cells against the clones of CD20-CEMs expressing various levels of CD20 (CD20-MFI: 126–6924/CD20-sABC: 240–230,546 molecules). CD20CAR–T cells lysed VL-CEMs, which had the lowest level of CD20 (MFI: 126/sABC: 240 molecules, 22.8 ± 2% lysis). In addition, CD20CAR–T cells induced similar lysis (40–60% lysis) of various CD20-CEMs with higher expression of CD20 (CD20-MFI: 157/CD20-sABC: ≥5172 molecules, E:T ratio of 10:1) (Fig. 2D). CD20CAR–T cells exhibited efficient cytotoxicity against CD20-CEMs with an MFI <1000; at this level, rituximab and OUBM mAb did not induce significant CDC (Supplemental Fig. 3A, 3B, Fig. 2D). Half-maximum cytotoxicity by CD20CAR–T cells was observed at an MFI of ∼200–300 (equivalent to sABC of 10³). Therefore, the minimum threshold number of surface target molecules that CAR-T recognized and lysed was markedly low, at approximately a few hundred molecules.

A coculture assay was performed as a more physiological model. In this assay, CD20CAR–T cells partially, but not completely, eradicated VL-CEMs. Conversely, CAR-T cells completely eradicated L-, M-, and H-CEMs after a 72-h coculture (Fig. 2E).

An advantage of CAR-T cell therapy over mAb therapy is that CAR-T cells can become activated and proliferate upon specific stimulation of the target Ag, enabling CAR-T cells to exhibit long-lasting efficacy in vivo (1, 3–5, 9). Although we titrated the threshold Ag density for CAR-T–induced lysis, the threshold for cytotoxicity and full activation, including cytokine production and proliferation, are uncoupled in Ag-specific T cells (36). Thus, we examined the threshold Ag density for CAR-T activation. To define the minimum threshold of CD20 expression that was needed for effective activation and expansion of CAR-T cells, we examined phosphorylation of the signaling molecules ERK and ZAP70 after stimulation with the four representative CD20-CEMs. The CD20-CEMs, except for VL-CEMs, induced similar phosphorylation of ERK (pERK) and ZAP70 in CAR-T cells (Fig. 3A and data not shown). pERK was equally upregulated when CAR-T cells were stimulated with L-, M-, and H-CEMs, but not with VL-CEMs, after 10 min (Fig. 3A). Time-course analysis showed that the pERK MFI responses were almost equal after L-, M-, and H-CEM stimulation, and the peak time was 5–10 min after stimulation. Nevertheless, VL-CEM induced only minimal phosphorylation of ERK in CAR-T cells, similar to that of parental CEMs (Fig. 3A, 3B).

Cytokine production and proliferation were evaluated following different stimuli. Stimulation with VL-CEM did not induce the production of cytokines from CAR-T cells. Conversely, L-, M-, and H-CEMs induced equivalent production of IFN-γ (Fig. 3C–E), IL-2 (Fig. 3F, 3G), and TNF-α (Fig. 3H). IL-2 production after H-CEM stimulation was approximately half that after M-CEM stimulation in repeated experiments (n = 3). Because we observed no significant difference in intracellular IL-2 production (Fig. 3F), the low IL-2 concentration after H-CEM stimulation may have reflected an increase in cytokine consumption. Regarding proliferation, VL-CEM did not induce cell division of CAR-T cells, whereas other CEMs induced efficient cell division 72 and 96 h after stimulation (Fig. 3I and data not shown). The kinetics of CD20CAR–T cell division increased with higher CD20 expression on CD20-CEMs, but the percentages of proliferating cells were equivalent among L-, M-, and H-CEM stimulation (Fig. 3I). The kinetics of division appeared to be partly dependent on target Ag density (Fig. 3I).

Taken together, the minimum threshold required to induce activation and proliferation of CAR-T cells was between the levels expressed by VL-CEMs and L-CEMs. This threshold was very low: less than the CD20 expression level of L-CEMs (CD20-MFI: 576/CD20-sABC: 5320). CD20 expression above the threshold significantly activated CAR-T cells.

Because we demonstrated that CD20CAR–T cells recognized markedly low expression of CD20, we examined the effectiveness of CD20CAR–T cell therapy against CD20^lo tumor cells. First, the cytotoxicity of CD20CAR–T cells against CD20^lo tumor cell lines was investigated. CD20CAR–T cells lysed both CD20^lo cell lines, RRBL1 and WILL2, very efficiently (Fig. 2B, 4A, 4B, lower panel).

We also evaluated cytotoxicity against CD20^lo primary cells from a patient with DLBCL (double-hit lymphoma). The patient exhibited disease recurrence after a full course of R-hyper CVAD (37), and lymphoma cells were obtained from pleural effusion. At the time of relapse, CD20 expression was reduced in most cells, and only ∼20% of cells showed low CD20 expression (Fig. 4C, upper panel). CD20CAR–T cells efficiently lysed CD20^lo primary DLBCL cells. This lytic activity was higher than the percentage of the CD20⁺ cell fraction, suggesting that CD20CAR–T cells partially lysed the cell fraction expressing low levels of CD20 (Fig. 4C, lower panel).

CLL is a chronic lymphoproliferative disease in which the anti-CD20 mAb is a choice for standard care (38). The expression of CD20 by CLL cells is generally lower than that of other CD20⁺ lymphoid malignancies such as ALL and lymphoma (39). To compare the potency of CD20 recognition by anti-CD20 mAb, we examined cytotoxicity and cytokine production following stimulation of CD20-downregulated CLL cells. Before starting mAb therapy, the expression of CD20 by CLL cells was intact, and the lytic activity of CD20CAR–T cells was remarkable (Fig. 5A). The patient then became chemorefractory following repeated administration of rituximab (the clinical course of this patient is summarized in Supplemental Fig. 4). In this patient, CLL cells could not be controlled with rituximab-combined chemotherapy. The expression of CD20 by CLL cells decreased, and the MFI showed two peaks: the nearly negative fraction and the CD20 low fraction (Fig. 5B, upper panel). However, cytotoxicity by CD20CAR–T cells was maintained (Fig. 5B, lower panel). The patient was then treated with the novel anti-CD20 mAb ofa (40). A marked decrease in the number of CLL cells and regression of lymphadenopathy were observed after a single course of ofa, whereas the CD20 very low fraction, which was confirmed to consist of CD5⁺ CLL cells (data not shown), remained in the peripheral blood. We obtained residual CLL cells from the peripheral blood of the patient after the 10th course of ofa. After ofa treatment, the CD20 relatively low fraction disappeared, and CD20 expression by CLL cells was almost uniformly nearly negative (Fig. 5C, upper panel). The residual cells were exposed once to the mAb therapy and survived. Therefore, the CD20 expression level of the residual cells was considered to be below the effective range of rituximab or ofa. In the [⁵¹Cr] release assay using these primary CLL cells, CD20CAR–T cells efficiently recognized and lysed not only CLL cells before ofa but also CLL cells after ofa (Fig. 5B, 5C, lower panel). With an intracellular IFN-γ assay, after ofa, stimulation of CD20CAR–T cells with CLL cells, which were nearly CD20-negative, induced production of IFN-γ by CD20CAR–T cells (Fig. 5D).

In the current study, we generated a novel CD20CAR based on a humanized anti-CD20 mAb (25). CD20CAR–T cells specifically and effectively lysed CD20-positive target cells. The expression of CD20CAR was precisely evaluated using the anti-Fc Ab and biotinylated Erbitux. Although we did not directly evaluate the copy number of the CD20CAR transgene, the variation observed in lytic activity against K562-CD20 cells was very low following tEGFR selection, suggesting that the expression of CD20CAR was similar among the CD20CAR–T cell lines. With cytotoxicity analysis of CD20CAR–T cells against CD20-CEMs expressing various levels of CD20, we first titrated the minimum threshold of CD20 expression that CAR-T cells could recognize and lyse. We demonstrated that CD20CAR–T cells lysed CD20-CEMs with CD20-ABC = 240 molecules, which was the lowest CD20 level in this set. This level was 1000-fold lower than that required to induce CDC with rituximab and OUBM mAb. The difference in cytolytic activity between CDC and CAR should mostly depend on the presence or absence of effector cells. Although CDC and CAR activity is similar against CD20-high CEM cells, CD20CAR–T cells demonstrated far better lytic activity than CDC against CD20-low CEM cells (Fig. 2D, Supplemental Fig. 3). This finding suggested that CAR-T therapy might show better effect in the case of only a limited number of target Ags on the tumor cells. The correlation between CD20-ABC and specific lysis was also represented with a saturation curve, which had a sharp inclination against CD20^lo targets. This phenomenon was attributed to CAR technology providing full activation with CD3ζ and the simultaneous costimulation of CD28 (3–6).

We next determined the threshold of CD20 expression that could activate and expand CD20CAR–T cells upon stimulation with the representative CD20-CEMs. Although cytotoxicity analysis revealed that CD20CAR–T cells lysed VL-CEMs, these cells did not induce downstream signaling, production of IFN-γ, or proliferation of CAR–T cells. Stimulation with L-, M-, and H-CEMs (CD20-ABC: ≥5320 molecules) effectively and equally activated CD20CAR–T cells. Taken together, these results indicated that the threshold of CD20 expression for recognition and lysis by CD20CAR–T cells, which we termed the “lytic threshold,” was a few hundred molecules, and the threshold required for activation and expansion of CAR-T cells, termed the “activating threshold,” was slightly higher, at a few thousand molecules. These results are consistent with previous findings in which the lytic threshold and activating threshold were different in TCR activation (36). Because endogenous T cells such as melanoma-specific T cells and virus-specific T cells require 10–100 epitope molecules per target cell to trigger specific lysis (13), both the lytic threshold and the activating threshold were slightly lower in endogenous T cells compared with CAR-T cells. Obviously, the thresholds are affected by the affinity of the mAb or TCR for the ligand or peptide/HLA complex. In our study, the affinity of the humanized anti-CD20 mAb (OUBM mAb), which was used to construct CD20CAR, was within the same range as that of rituximab (K_D value: OUBM mAb, 10.09 nM; rituximab, 5.35 nM) (25). Using a mAb with this range of affinity, both thresholds of CAR-T cells were close to those of endogenous T cells. Furthermore, CD20CAR–T cells recognized and lysed CD20-downregulated target cells that survived after mAb therapy, indicating that manufacturing a CAR with a mAb may reinforce target recognition more than the mAb itself.

The epitope location of the mAb is another important issue. Ofa exposure before sample collection may account for the apparent CD20 downregulation. However, we confirmed CD20 downregulation using another CD20 mAb, a B9E9 clone in which the epitope location is distinct from that of ofa (23). This confirmation indicated that CD20 downregulation after ofa treatment was not caused by competition between the analytical Ab and ofa. The epitope location targeted by OUBM mAb and ofa partially overlaps, but that of OUBM mAb and rituximab does not (25). Therefore, ofa can theoretically block the ligation of CD20CAR-T but rituximab cannot. We observed that CD20CAR–T cells indeed lysed CD20-downregulated target cells both after rituximab and ofa, suggesting that the potential effect of epitope blocking was minor in the current study.

The results of the current study led us to propose a novel concept for future searches for target Ag in CAR-T therapy. Suitable target Ags for CAR-T cell therapy are considerably different from those for mAb therapy in terms of their expression profiles and levels. Higher expression levels on the surface of tumor cells have been considered in target Ag searches for mAb therapy because off-tumor expression is usually negligible (41). However, for the target of CAR-T cell therapy, off-tumor expression of the target molecules must be strictly negative or at a very low level that is below the lytic threshold, at a few hundred molecules. Otherwise, severe adverse effects could occur as a result of off-tumor effects (10). The target Ag safety in the context of mAb therapy does not necessarily translate into the safety of Ag in the context of markedly more sensitive CAR-T cell therapy (10). Conversely, even if the threshold was below the mAb therapy range, low Ag expression above the activating threshold, such as at a few thousand molecules, could be considered a candidate for the target Ag of CAR-T cell therapy.

Acquired resistance to rituximab has become a problem in the treatment of patients with CD20-positive B cell tumors (20, 23). One suggested mechanism is downregulation of CD20 (20, 22, 26). A total of 15–20% of relapsed patients exhibit CD20 Ag loss, as observed with immunohistochemistry analysis in samples taken at relapse (20, 23). Our CD20CAR–T cells recognized and lysed primary cells isolated from patients with mAb therapy–refractory lymphoma and CLL, although the expression level of CD20 was very low. We also analyzed CD20CAR-T recognition against CD20-downregulated, mAb-refractory CLL in detail (21). The residual cells after CD20 mAb therapy expressed significantly low levels of CD20, and this expression level must have been below the effective range of the mAb in principle. CD20CAR–T cells lysed both postrituximab and postofa residual CLL cells, indicating that CD20CAR–T cells have a greater potential to recognize the target than mAbs. Even residual postofa CLL cells stimulated CD20CAR–T cells, and thus we conclude that the very low expression of CD20 on CLL cells could still efficiently provide stimulation for the further repopulation of CD20CAR–T cells.

In other CAR therapies targeting CD19, several patients were reported to have relapsed despite the completely negative conversion of the Ag molecule after CAR-T cell therapy (2). Although CAR-T cells recognize very low levels of the target Ag, they cannot recognize completely Ag-negative cells. The strategy of administering CAR-T therapy as a first-line treatment, or in earlier phases with the aim of earlier eradication of target cells, may prevent immunological escape by negative conversion of the Ag.

One limitation of the current study is that we assessed the threshold using only CD20 and CD20CAR systems. Because the threshold may be influenced by many other factors, such as affinity (42), structure (43), epitope localization of individual CAR-Ag pairs (44, 45), and the expression of a coreceptor on target cells (46), the threshold may vary among mAbs and target Ags. We also could not investigate the relationship between the expression of CD20 and the ADCC activity of mAbs because NK cell activity is predominant in the CD20-CEM system, and a clear threshold has not been observed (28). Although the potential relationship between target Ag density and ADCC activity has been investigated in other experimental systems, >10⁴ Ag molecules per cell are needed to demonstrate significant ADCC (47). In the current study, the minimum threshold of CAR recognition was 3-log units lower than that of mAbs to trigger CDC. CAR can also directly mobilize T cells to target cells, whereas mAb therapy mainly depends on indirect cytotoxicity such as CDC or ADCC (3–5, 19, 28). Thus, the lytic and activating thresholds of CAR are considered significantly lower than those of mAbs.

We concluded that CAR-T cells can recognize and lyse cells expressing considerably low levels of the target Ag and were activated and expanded upon such stimulation. CD20CAR–T cell therapy may also be applicable for the treatment of CD20-positive lymphoid malignancies.

We thank Yoko Matsuyama, Asako Watanabe, and Chika Wakamatsu for technical assistance.

H.K. has received research funding from Bristol-Myers Squibb, Chugai Pharmaceutical, Kyowa Hakko Kirin, Dainippon Sumitomo Pharma, Zenyaku Kogyo, and FUJIFILM. The other authors have no financial conflicts of interest.