Self-driving armored CAR-T cells overcome a suppressive milieu and eradicate CD19⁺ Raji lymphoma in preclinical models

Abstract

Chimeric antigen receptor (CAR) T cells typically use a strong constitutive promoter to ensure maximal long-term CAR expression. However, recent evidence suggests that restricting the timing and magnitude of CAR expression is functionally beneficial, whereas constitutive CAR activation may lead to exhaustion and loss of function. We created a self-driving CD19-targeting CAR, which regulates its own function based on the presence of a CD19 antigen engaged by the CAR itself, by placing self-driving CAR19 constructs under transcriptional control of synthetic activator protein 1 (AP1)-nuclear factor κB (NF-κB) or signal transducer and activator of transcription (STAT)5 promoters. CD19 antigen-regulated expression was observed for self-driving AP1-NFκB-CAR19, with CAR19 upregulation within 18 h after exposure to target CD19, and corresponded to the level of tumor burden. Self-driving CAR-T cells showed enhanced tumor-dependent activation, expansion, and low exhaustion in vitro as compared to constitutively expressed EF1α and murine stem cell virus (MSCV) CARs and mediated tumor regression and survival in Raji-bearing NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice. Long-term CAR function correlated with upregulated CAR expression within 24 h of exposure to tumor antigen. The self-driving AP1-NFκB-CAR19 circuit was also used to inducibly express dominant-negative transforming growth factor β receptor II (TGFBRIIdn), which effectively countered the negative effects of TGF-β on CAR-T activation. Thus, a self-driving CAR approach may offer a new modality to express CAR and auxiliary proteins by enhancing CAR-T functional activity and limiting exhaustion.

Graphical abstract

Self-driving chimeric antigen receptor (CAR) transcriptional circuits are developed, where expression is dependent on tumor antigen levels. Self-driving CAR circuits have high CAR effector functions and are useful in co-regulating the expression of auxiliary proteins in CAR-T cells.

Introduction

Chimeric antigen receptor (CAR)-modified T cells are an important new tool in the armamentarium of cancer immunotherapies. Remarkable success has been achieved with CAR-T cell therapies directed against CD19⁺ B cell leukemias and lymphomas, offering promise for the treatment of chemo-refractory or relapsed malignancies.1, 2, 3 However, in a subset of patients, CAR-T cell therapy falls short of desired outcomes. Major challenges to the field include insufficient CAR-T cell persistence and side-effects, including cytokine release syndrome and neurotoxicity.^4,5

Traditionally, CAR expression cassettes incorporate a strong constitutively active promoter to ensure robust long-term CAR expression. However, constitutively high expression of CARs differs from the native T cell receptors (TCRs), which are subject to stringent control of function via a variety of cellular mechanisms (CD45, checkpoint molecules, reduction of CD3 surface expression post-activation).⁶ CARs, by contrast, are much less susceptible to this intrinsic feedback, in part, due to being transcriptionally regulated by strong constitutive promoters. Constitutive CAR expression and activation may result in premature CAR-T exhaustion and may, in part, be responsible for CAR-associated adverse events, such as cytokine release syndrome or neurotoxicity.

Recent observations suggest that limiting the magnitude or the duration of CAR expression may augment its anti-tumor function. When CAR expression in T cells was governed by the endogenous constant region of a TCR α chain (TRAC) promoter, which limited CAR expression levels to that analogous to the native TCR, superior tumor rejection in vivo was seen as compared to constitutively expressed CARs.⁷ Pharmacologic control of CAR-T function by administration of the tyrosine kinase inhibitor dasatinib, which led to a transient downregulation of CAR expression in the course of anti-tumor responses, achieved superior tumor clearance in vivo as compared to the constitutively expressed CAR and mitigated cytokine release syndrome.^8,9 Finally, tonic, ligand-independent signaling observed in some CAR constructs, such as anti-GD2 CAR, especially those expressed at a high level, may lead to uncontrolled T cell proliferation, cytokine production, and eventual exhaustion and a lack of memory T cell formation.10, 11, 12

Multiple approaches have been explored to control the expression of chimeric receptors in CAR-T cells extrinsically.¹³ Transient CAR expression has been achieved via electroporation of plasmids or mRNAs encoding CARs into T cells.¹⁴ Although safer, such approaches are unlikely to persist and fully eliminate disease. Moreover, molecular switches controlled by tetracycline,¹⁵ small-molecular dimerization domains,¹⁶ self-dimerization domains,¹⁷ epitope tags,¹⁸ or the logic-gated synNotch system have been developed.¹⁹ Although promising, these approaches often require inclusion of non-human sequences in the composition of the molecular switch, posing the risk of inducing anti-CAR immunity. In addition, these approaches do not account for the intrinsic dynamics of the tumor microenvironment, such as the number of tumor cells present at a given time, the levels of homeostatic cytokines available to support T cell function (e.g., interleukin [IL]-2), or the presence of immunosuppressive factors (e.g., transforming growth factor β [TGF-β]).

To address this problem, we constructed a self-driving CAR system whereby the expression of CAR19 (anti-CD19 CAR) molecules on the T cell surface is operatively linked to the presence of CD19 antigen on target cells in the immediate vicinity of the self-driving CAR-T cell at a given time and tested them under conditions of varied cytokine and cognate antigen (CD19) levels.

T cell activation and subsequent signal transduction cascades rely on several key mediators, including the activator protein 1 (AP1), nuclear factor κB (NF-κB), and signal transducer and activator of transcription (STAT) family of proteins. AP1 and NF-κB transcription factors are activated downstream of TCR and CD28 engagement and upregulate the expression of genes involved in T cell proliferation and anti-tumor function, including IL-2.^20,21 The Jak-STAT pathway plays a central role in the transmitting homeostatic and pro-inflammatory signals in T cells, in part, via the common γ chain cell surface receptors for cytokines, such as IL-2, IL-15, IL-7, and IL-9.²² Because CAR-T cells are believed to utilize similar downstream signaling pathways as the canonical TCR, we conditioned the self-driving CAR expression proximally on the presence of activated AP1 and NF-κB, or distally on the presence of STAT5. We hypothesized that the use of self-driving promoters will mediate low basal levels of CAR19 expression in the absence of cognate antigen and that CAR expression levels will be rapidly upregulated upon CAR recognition of its cognate antigen. Moreover, we hypothesized that by operatively linking the regulation of CAR expression on T cells to the site-specific tumor burden and the duration of tumor antigen presence, it may be possible to improve CAR-T effector function and persistence.

Another challenge to successful CAR-T therapy is posed by immunosuppressive tumor microenvironments, which may limit CAR-T cell effector function. TGF-β, a soluble mediator, inhibits T cell effector function directly via its cognate receptor on T cells and indirectly by inducing T cell PD1 expression.²³ TGF-β has been implicated in tumor resistance to treatment, both in liquid and in solid malignancies.²⁴ TGF-β receptor II dominant-negative (dn) form (TGFBRIIdn), containing the extracellular moiety capable of binding TGF-β and dimerization with TGF-βRI, but lacking the intracellular components necessary to transmit an anti-inflammatory signal,²⁵ functions as a dn receptor.²⁶ Although expression of TGFBRIIdn was shown to alleviate TGF-β inhibition of T effector function,²⁷ constitutive expression on T cells increases the risk of inappropriate T cell activation and autoimmunity.²⁵ This variant of TGFBRIIdn was previously used successfully in prostate-specific membrane antigen (PSMA)-targeted CARs to allow them to resist the immunosuppressive effects of TGF-β.²⁸ As a proof of principle, whereby CAR-T cells can restrict the expression of potentially dangerous “armor” genes such as TGFBRIIdn to conditions where tumor cells are present, we designed self-driving bicistronic constructs for combined inducible expression of TGFBRIIdn armor and CAR19 conditioned on the presence of the CD19 tumor antigen.

We demonstrate that the use of a self-driving AP1-NF-κB promoter leads to low basal CAR19 expression, which is rapidly upregulated (within 18 h) following engagement with the CD19 antigen, a property unique among the self-driving and constitutive promoters tested. The proximal signaling of the AP1-NF-κB promoter in driving CAR19 T cell expression yielded equivalent tumor lysis efficiency as compared to the constitutively expressing elongation factor 1α (EF1α)-CAR19 T cells in vitro and in vivo, which retain moderate CAR19 expression following coculture with CD19⁺ tumor cells. Additionally, AP1-NFκB-CAR19 T cells displayed enhanced efficiency as compared to the constitutively expressing murine stem cell virus (MSCV)-CAR19 or proximally inducible STAT5-CAR19. Notably, the AP1-NFκB-CAR19 system increased CAR-T persistence and expansion across multiple rounds of antigen stimulation in vitro. We further demonstrate that the self-driving AP1-NFκB-CAR19 system can be used to control the CD19 antigen-dependent expression of an auxiliary armor protein, TGFBRIIdn, which provides protection from the anti-inflammatory effects of TGF-β.

Results

Design of self-driving CAR constructs

To confer self-driving CAR expression to T cells, we generated CAR19 constructs under the control of one of the two inducible promoters, STAT5 or AP1-NF-κB (Figure 1A). Constitutively expressed CAR19 constructs under the control of either the human EF1α promoter or MSCV promoter were constructed for comparison with the STAT5- and AP1-NF-κB-regulated CARs (Figure 1A). The STAT5-CAR19 and AP1-NFκB-CAR19 constructs were designed by placing CAR19 expression under the control of a tandem (5×) STAT5 response element (STAT5-CAR19) or tandem (6×) AP1 and (5×) NF-κB response element (AP1-NFκB-CAR19) followed by a minimal promoter sequence (Figure 1A).

Figure 1.

Rapid induction and re-induction of CAR19 expression by the AP1-NF-κB promoter in the presence of CD19 antigen

(A) Graphical representation of CAR19 vectors used in this study in the context of a self-inactivating lentiviral backbone. The sequence of the transcription factor response elements (REs) is shown, as is the minimal promoter sequence. (B) Quantification of CAR19 expression by flow cytometry following 18 h treatment of CAR-T cells with cytokines (IL-2, TNF-α, IL-6) at the given concentrations. CAR-T cells were cultured in the absence of exogenous cytokines prior to treatment. n = 3 independent donors, day (D)9 post-activation. Mean ± SEM is shown. (C) Graphical representation of CAR-T culture/coculture protocol for the experimental data represented in (D) and (E). n = 3 independent donors. IL-2-starved cultures were cultured in the absence of cytokines throughout the culture/coculture period, and IL-2-replete cultures were cultured in 30 IU/mL of IL-2 from day 2 to day 6 post-activation. (D) Flow cytometric plots representing CAR-T CAR19 expression prior to coculture with CD19⁺ Raji cells (day 6) or 18 h following coculture (E:T 1:3, day 7 post-activation). One representative donor shown of n = 3 independent donors. (E) Quantification of CAR-T CAR19 expression by flow cytometry (% T cells CAR19⁺) at various time points pre- and post-coculture with CD19⁺ Raji-GFP cells (shown with arrowheads, repeated cocultures 1−2, E:T 1:3). UTD, untransduced. n = 3 independent donors. Mean ± SEM is shown.

Cytokine- and CAR-antigen-mediated induction of self-driving CAR constructs

In order to test the responsiveness of the self-driving promoter systems to induction by cytokines, which may be found in a tumor milieu containing activated T cells, CAR-T cells were treated with the T cell activating cytokines IL-2, tumor necrosis factor (TNF)-α, or IL-6. T cells were transduced with lentiviral vectors (LVs) expressing CAR19 constructs and cultured through day 6 (D6) post-activation in the absence of cytokine supplementation. CAR19 regulated by either the constitutive EF1α promoter or by self-driving promoters STAT5 or AP1-NF-κB or untransduced (UTD) control was tested (Figure 1). The constitutive EF1α-CAR19 expression did not change due to IL-2, TNF-α, or IL-6 supplementation (Figure 1B). By contrast, STAT5-CAR19 expression was greatly enhanced by supplementation with cytokines IL-2 and IL-6 but not by TNF-α (Figure 1B). The AP1-NFκB-CAR19 expression was moderately increased by each of the three cytokines (Figure 1B).

In separate experiments, the ability of the STAT5 and the AP1-NF-κB promoters to drive CAR19 expression was assessed by CD19 antigen stimulation, as compared to the two constitutively expressing promoters EF1α and MSCV. Long-term CAR expression was assessed over two or three consecutive cocultures with Raji target cells, for a total period of 20−30 days (Figure 1C; Figures S1A and S1C). Primary human T cells were cultured in the presence or absence of 30 IU/mL IL-2 (IL-2 replete or IL-2 starved, respectively) to investigate the impact of IL-2 priming on the inducible CAR expression. In the IL-2-replete group, 30 IU/mL IL-2 was supplemented in the cultures from days 2 to 6 post-activation to prime the expression of the STAT5-CAR19 and AP1-NFκB-CAR19, whereas IL-2-starved cells remained IL-2 free for the duration of the experiment. CD19⁺ Raji non-Hodgkin’s lymphoma tumor cells stably expressing GFP were added to CAR cultures and incubated overnight. The AP1-NFκB-CAR19 T cell expression was strongly induced within 18 h post-stimulation as compared to the pre-incubation period (Figures 1D and 1E; Figures S1B and S1D). Induction of CAR surface expression in the IL-2-replete AP1-NFκB-CAR coculture was greater than for IL-2-starved AP1-NF-κB-CAR coculture. In contrast to AP1-NFκB-CAR, the surface expression of both the EF1α-CAR19 and MSCV-CAR19 was dramatically decreased, especially for MSCV-CAR19 (Figures 1D and 1E; Figure S1B). On the other hand, the STAT5-CAR19 expression decreased after 18 h co-incubation with the Raji target for IL-2-replete conditions (Figures 1D and 1E; Figure S1). Under IL-2-starved conditions, STAT5-CAR19 expression was very low prior to co-incubation with Raji targets and remained so after co-incubation until late in cocultures.

Although AP1-NFκB-CAR19 demonstrated rapid upregulation of surface expression after each Raji target cell addition and equally rapid downregulation of CAR expression at the end of each target elimination cycle, regardless of IL-2 starvation (Figure 1E; Figure S1D), STAT5-CAR19 induction was poor in the absence of IL-2 priming and was delayed as compared to AP1-NFκB-CAR19 (Figure 1E; Figure S1D) (IL-2 starved). By contrast, surface expression of both the EF1α-CAR19 and MSCV-CAR19 declined rapidly after the initial Raji addition and exhibited delayed recovery as compared to AP1-NFκB-CAR. Overall, expression of the constitutively expressed CARs (MSCV-CAR19, EF1α-CAR19) was more sustained and less strictly dependent on target cell addition (CD19 stimulation), whereas the AP1-NFκB-CAR expression was strictly linked to antigen-driven stimulation. The STAT5-controlled self-driving CAR performed poorly under IL-2-starved conditions, only reaching >70% CAR19 expression at the end of the first coculture period. Even when the STAT5-controlled CAR was primed under the IL-2-replete condition, there was no significant expression in the third round of antigenic stimulation, likely indicating a drop in available cytokine, below levels that could stimulate CAR expression (Figure S1D).

Efficacy of self-driving CAR constructs

We then proceeded to assess the CAR-mediated cytotoxicity, T cell expansion, activation, and exhaustion of self-driving CARs as compared to the constitutive CAR19. CAR19-constitutive EF1α or MSCV promoters were compared to the inducible promoters STAT5 or AP1-NF-κB under either IL-2-replete or IL-2-starved conditions (Figure 1C; Figure S1A). CAR19-dependent cytotoxicity was assessed by coculture of CAR-T cells with CD19⁺ Raji cells stably expressing GFP for 20 days. During this time period, Raji cells were added to the cultures on day 6 (coculture 1) and again on day 13 (coculture 2) (Figure S1A). In both cocultures, a low effector-to-target (E:T) ratio of 1:3 CAR-T:Raji cells was used to maximize the challenge to CAR-T cultures.

T cell lytic function, expansion, CAR surface expression, T cell activation, and exhaustion were examined by flow cytometry (Figure 2; Figure S2). Both groups of self-driving (AP1-NF-κB, STAT5) and constitutive CAR19 T cells (MSCV, EF1α) significantly suppressed Raji expansion during coculture 1 (Figures 2A−2D; Figures S2A−S2D). (The details and significance values of all statistical tests in this manuscript are presented in Table S1.) However, in coculture 2, the self-driving AP1-NFκB-CAR19 and constitutive EF1α-CAR19 were more cytolytic than self-driving STAT5-CAR19 and constitutive MSCV-CAR19 as compared to the negative controls, Raji alone, and UTD. The cytolytic activity of the self-driving AP1-NFκB-CAR19 was robust, as even under IL-2-starved conditions, the AP1-NFκB-CAR19 achieved the greatest target cell lysis as compared to other CARs by coculture 2 (Figure 2C; Figure S2C).

Figure 2.

Equivalent tumor cell cytotoxicity and superior CAR-T expansion by AP1-NF-κB-driven CAR19 expression relative to constitutive CAR19 expression

Data from this figure also represent the experimental data shown in Figures 1C−1E. (A−D) Quantification of CAR19-dependent cytotoxicity of CD19⁺ Raji-GFP cells by flow cytometric enumeration of GFP⁺ cells in repeated CAR-T:Raji-GFP cocultures at the end of the coculture period (repeated cocultures 1−2). Coculture 1 (A and B) was initiated on day 6 post-activation (E:T 1:3), and coculture 2 (C and D) was initiated on day 13 post-activation (E:T 1:3). Geometric mean + geometric SD is shown. (E and F) Quantification of CAR19-dependent CAR-T expansion by flow cytometric enumeration of CD3⁺ cell expansion at the end of repeated coculture 2 of CAR-T:Raji-GFP cocultures (day 20 relative to day 0 post-activation). Geometric mean + geometric SD is shown. (G) Quantification of T cell activation marker expression (CD69) by flow cytometry at day 1 post-coculture 1 with CD19⁺ Raji-GFP cells (day 7 post-activation). Mean + SEM is shown. (H) Quantification of T cell intracellular cytokine production (IL-2) in CD4⁺ T cells by flow cytometry at day 1 post-coculture 1 with CD19⁺ Raji-GFP cells (day 7 post-activation). Mean + SEM is shown. (A−H) n = 3 independent donors. n.s., non-significant; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Of note, the AP1-NF-κB-driven CAR-T cells displayed a number of characteristics related to improved CAR-T function aside from CD19-dependent cytotoxicity. Expansion of AP1-NFκB-CAR19 T cells during the experimental period was the greatest among the four CAR-T groups, and it was significantly higher than both MSCV and STAT5 CARs, regardless of IL-2 supplementation history (Figures 2E and 2F; Figures S2E and S2F). Further, the ability of self-driving CAR-T cells to be activated by target cells was assessed by measuring the expression of the activation marker CD69. 1 day after target addition, CD69 expression was most highly upregulated on AP1-NFκB-CAR19 T cells in the first coculture, followed by STAT5-CAR19, and then by EF1α- and MSCV-CAR19, regardless of IL-2 supplementation history (Figure 2G). Moreover, at the end of both coculture periods, the AP1-NFκB-CAR19 group had the lowest number of cells co-expressing one, two, or three of the exhaustion markers TIM3, LAG3, or PD1 (Figures S2G−S2J). By contrast, STAT5 CAR19 T cells displayed an increased exhaustion phenotype by the end of the coculture period, defined as triple positivity for TIM3, LAG3, and PD1 (Figures S2G and S2H). We also assessed T cell memory phenotypes at various points pre- and post-coculture by CD45RO and CD62L expression. No significant differences were observed before coculture (data not shown); however, under IL-2-starved conditions, both AP1-NFκB-CAR19 and EF1α-CAR19 T cells demonstrated a significantly higher proportion of effector memory T cells (T_EM; CD45RO⁺, CD62L⁻) relative to STAT5-CAR19 and STAT5-CAR19 T cells (Figures S2K and S2L).

The long-term function of CAR19-expressing cells was concordant with CAR expression in the first 24 h following exposure to antigen. Although MSCV-CAR19 (both IL-2 starved and IL-2 replete) and STAT5-CAR19 (IL-2 replete) demonstrate high levels of CAR19 expression before exposure to CD19 tumor antigen, both promoters demonstrate a substantial decrease in CAR19 expression following exposure to tumor antigen by both percentage CAR19⁺ cells and the level of CAR19 expression (Figures 1D and 1E; Figure S1B). By contrast, the constitutive EF1α promoter retained high-level CAR19 expression on a subset of CAR-T cells (Figures 1D and 1E; Figure S1B), although the overall percentage of CAR19⁺ cells decreased, and the AP1-NF-κB promoter demonstrated a sharp increase in percentage of CAR19⁺ cells, as well as CAR19 expression.

Importantly, the AP1-NF-κB and, to a lesser extent, the EF1α promoter displayed superior long-term CD19⁺ tumor cell cytotoxicity across multiple cocultures relative to the STAT5 and MSCV promoters (Figures 2A−2D; Figures S2A−S2D), as well as long-term CAR19-dependent T cell expansion (Figures 2E and 2F). This long-term functionality did not depend on high-level CAR19 expression prior to coculture but appeared to result from sustained and/or robust expression in the first 24 h following coculture, as AP1-NFκB-CAR19 had low-level CAR19 expression prior to coculture but rapidly increased CAR19 expression following exposure to tumor antigen (Figures 1D and 1E; Figures S1B and S1D). The superior long-term functionality of AP1-NFκB-CAR19 CAR-T cells also correlated with CAR19-dependent IL-2 production from CD4⁺ T helper cells following repeated coculture with Raji cells (Figure 2H).

Self-driving CAR induction by varying levels of CAR antigen

Previous experiments demonstrated that the AP1-NF-κB promoter was efficient at inducing high levels of CAR expression downstream of CAR signaling. However, this induction was performed using a single CD19⁺ Raji-GFP cell line and represents one fixed surface density of the CD19 antigen. We therefore tested the induction of the AP1-NFκB-CAR19 construct in response to tumor cell lines expressing various amounts of CD19 antigen. Stable Raji target cell clones with CD19^High (1.59 × 10⁵ CD19/cell), CD19^Med (5.25 × 10⁴ CD19/cell, native), or CD19^Low (4.65 × 10³ CD19/cell; Figure 3A) target cell density were cocultured with AP1-NFκB-CAR19 T cells transduced at a multiplicity of infection of 5. The EF1α-CAR19 was used as a comparative control due to its robust and sustained expression relative to MSCV-CAR19 (Figures 1D and 1E; Figures S1B and S1D). The STAT5-CAR19 was not used in these experiments due to its poor induction by the CAR19 antigen. As some clinical protocols require longer CAR-T production times than the 6−8 days used in prior experiments, we extended the analysis of self-driving AP1-NFκB-CAR19 induction to separate cocultures at both 8 days and 15 days after T cell activation (Figure 3B). Primary T cell cultures were also performed in the presence of either 30 IU/mL or 200 IU/mL IL-2 to assure T cell persistence during the longer expansion phase and as existing clinical protocols use 200 IU/mL IL-2 for CAR-T manufacture.²⁹ The results of the experiments were comparable between both IL-2 doses, but only the results from the 200-IU/mL dose is shown in keeping with clinical manufacture.

Figure 3.

AP1-NF-κB promoter induces CAR19 expression at low CD19 antigen density and during prolonged production cultures

(A) Antigen density of target cell lines (Raji-CD19^Low, Raji-CD19^Med, Raji-CD19^High) expressed as CD19 molecules per cell. (B) Graphical representation of CAR-T culture/coculture protocol for the experimental data represented in (C−K). Primary T cell cultures used in coculture 1 were cultured in 200 IU/mL of IL-2 from day 2 to day 8 post-activation, and primary T cell cultures used in coculture 2 were cultured in 200 IU/mL of IL-2 from day 2 to day 15 post-activation. Cocultures were performed with CAR-T cells and Raji-CD19^Low, Raji-CD19^Med, Raji-CD19^High (E:T 1:1) at day 8 post-activation (coculture 1) or day 15 post-activation (coculture 2). (C and D) Quantification of CAR-T surface marker expression by flow cytometry at various time points during primary T cell culture; (C) % T cells CAR19⁺ and (D) % T cells CD69⁺. Mean ± SEM is shown. (E) Flow cytometric plots representing CAR-T CAR19 expression 18 h post-coculture with Raji cell lines or cultured without target cells (CAR-T only). Coculture 1 plots are at day 9 post-activation, and coculture 2 plots are at day 16 post-activation. One representative donor shown of n = 3 independent donors. (F and G) Quantification of T cell CAR19 expression by flow cytometry. Mean + SEM is shown. (F) 18 h post-coculture 1 (day 9 post-activation). (G) 18 h post-coculture 2 (day 16 post-activation). (H and I) Quantification of T cell CD69 expression by flow cytometry (H) 18 h post-coculture 1 (day 9 post-activation). (I) 18 h post-coculture 2 (day 16 post-activation). (J and K) Quantification of CAR19-dependent cytotoxicity of CD19⁺ Raji cells by flow cytometric enumeration of CD3⁻ cells in CAR-T:Raji cocultures (E:T 1:1). Geometric mean + geometric SD is shown. (J) 5 days post-coculture 1 (day 13 post-activation). (K) 5 days post-coculture 2 (day 20 post-activation). n = 3 independent donors. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Of note, the low level of CAR19 expression from the AP1-NF-κB promoter was maintained throughout the primary T cell culture (i.e., before target cell addition; Figure 3C) and did not differ in activation marker expression from EF1α-CAR19 or UTD T cells (Figure 3D). When AP1-NFκB-CAR19 T cells were cocultured with Raji target cells with varying CD19 surface antigen densities, AP1-NFκB-CAR19 expression was strongly induced in Raji^Low and Raji^Med groups, as compared to CAR T-only control, and was equal to or greater than the EF1α-CAR19 expression under matched conditions (Figures 3E and 3F). The percentage of AP1-NFκB-CAR19 expression remained similar even at a 10-fold reduction in CD19 target cell antigen density (Raji^Low) relative to the standard antigen density (Raji^Med; Figure 3E). The sustained induction of CAR19 in coculture with Raji target cells persisted even in CAR T cells after the longer CAR-T expansion phase of 15 days (coculture 2), albeit at a lower efficiency (Figure 3G). Surprisingly, co-incubation of CAR19 cells with Raji^High targets resulted in equally repressed CAR surface expression in both EF1α-CAR19 and AP1-NFκB-CAR19, suggesting a high level of CD19 antigen-mediated internalization (Figures 3E and 3G). By contrast, the expression of the CD69 activation marker did exhibit a significant CD19 density-dependent upregulation, both during the 1^st and the 2^nd coculture (Figures 3H and 3I). CD69 induction was also higher in AP1-NFκB-CAR19 relative to EF1α-CAR19 T cells, as it was previously (Figure 2G). Additionally, both EF1α-CAR19 and AP1-NFκB-CAR19 T cells demonstrated efficient CAR-dependent cytotoxicity at all target antigen densities tested and varying CAR-T production lengths (Figures 3J and 3K). These results indicate that AP1-NFκB-CAR19 surface expression is more highly upregulated than EF1α-CAR19 under medium or low CD19 densities and is equal to EF1α-CAR19 under high CD19 density and that the AP1-NFκB-CAR19 activation is CD19 target density dependent, and its cytolytic function is robust even at the low CD19 target density.

To assure that the self-driving AP1-NF-κB promoter does not induce CAR expression in the absence of the cognate antigen, we used suspension human embryonic kidney 293T target cell lines (Figure S3A) stably expressing CD19 (293T-CD19, CD19⁺) or CD22 (293T-CD22, CD19⁻, negative control). These cell lines were cocultured with AP1-NFκB-CAR19- T cells or UTD controls at day 6 post-activation (Figure S3B). The 293T-CD19⁻ cells strongly induced AP1-NFκB-CAR19 and CD69 expression, whereas the 293T-CD22 cells did not (Figures S3C and S3D). In parallel, AP1-NFκB-CAR19 T cells lysed the 293T-CD19 cells but not the 293T-CD22 cells (Figures S3E and S3F). To further assure that the constructs EF1α-CAR19, STAT5-CAR19, and AP1-NFκB-CAR19 demonstrate no cytotoxic activity against CD19⁻ cells, short-term cytotoxicity assays (Figures S3G and S3H) were also performed in additional CD19⁺ Reh cell lines and CD19⁻ K562 cell lines stably expressing firefly luciferase. Importantly, none of the constructs demonstrated significant lysis of the CD19⁻ K562 cell line (p > 0.05, E:T 10:1) relative to the UTD control. In this short-term, 18-h coculture period, EF1α-CAR19 T cells displayed superior lysis of CD19⁺ T cells relative to AP1-NFκB-CAR19 and STAT5-CAR19 T cells (Figure S3G), likely due to the higher expression of CAR19 under the control of the EF1α promoter (CAR19 median fluorescence intensity of CAR19⁺ T cells: EF1α-CAR19: 69.8, STAT5-CAR19: 18.36, and AP1-NFκB-CAR19: 6.04; Figure S3I) prior to coculture with CD19⁺ target tumor cells (Figure S3I).

Efficacy of self-driving CAR constructs in vivo

After verifying that the AP1-NFκB-CAR19 self-driving construct displayed potent anti-tumor activity comparable to constitutively expressed EF1α-CAR19, we proceeded to determine the ability of the self-driving CAR constructs to clear tumors in vivo. The constitutive EF1α-CAR19 was used as a comparison in the in vivo studies due to its robust performance in long-term coculture assays, as compared to MSCV-CAR19 (Figures 2C and 2D; Figures S2C and S2D). Additionally, to highlight differences between the treatment groups and to approximate clinical CAR-T production conditions, a limiting pro-viral DNA integration threshold (viral copy number [VCN] < 5 per cell) was used for CAR-T production for in vivo studies (Figure S4A). The CAR-T cells from the same donor as those used in the in vivo study and CAR-T from an additional two donors were subjected to three rounds of coculture, and CAR efficacy was assessed over three repeated cocultures (Figure S4B). As in previous experiments, AP1-NFκB-CAR19 and EF1α-CAR19 T cells exhibited higher efficacy, T cell expansion, and IL-2 production and reduced exhaustion marker expression relative to STAT5-CAR19 T cells (Figures S4C−S4I).

In vivo, the ability of CAR-T cells to regress CD19⁺ tumors was assessed in a NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mouse model of disseminated B cell lymphoma. STAT5-CAR19, AP1-NFκB-CAR19, and EF1α-CAR19 CAR-T cells were used in this study. To provide a challenging model for CAR-T effects on tumor burden, mice were implanted with a low dose of 2 × 10⁶ total T cells per mouse, and all T cell groups had low CAR integration at ~2 copies/cell (Figure S4A). The estimated effective CAR-T cell dose was based on maximal CAR19 expression levels from all constructs under both basal and antigen-stimulated conditions (Figure S4F, left panel) from 0 h to 48 h after CD19 antigen induction: STAT5-CAR19: 66.3% CAR19⁺, AP1-NFκB-CAR19: 66.3% CAR19⁺, and EF1α-CAR19: 81.7% CAR19⁺. This suggests an overall in vivo dose of ≤1.6 × 10⁶ CAR-positive T cells, as compared to the standard dose of 5 × 10⁶ CAR-positive T cells per mouse in this model.^30,31 Tumor growth and persistence were assessed by bioluminescence imaging, and blood was collected for flow cytometric analysis throughout the study.

Each of the CAR19 treatment groups displayed a significant, approximately 5-log reduction in tumor burden relative to the UTD T cell control and tumor-only groups (Figure 4A). Despite the very low dose of CAR-T cells administered and the very low proviral copy number per cell in CAR-T groups (Figure S4A) by study termination date, six out of eight mice in the EF1α-CAR19 and AP1-NFκB-CAR19 groups and three out of eight mice in STAT5-CAR19 completely rejected tumors (Figure 4B; Figures S5A−S5C). Furthermore, no significant difference was observed in tumor-burden reduction among STAT5-CAR19, AP1-NFκB-CAR19, and EF1α-CAR19 CAR-T treatment groups at any time point (Figure 4A). Additionally, all mice survived in each of the CAR19 treatment groups relative to both UTD and tumor-only controls (Figure 4C), indicating that all CAR-T groups were equally protective at the level of tumor survival.

Figure 4.

Equivalent tumor cytotoxicity in vivo by constitutive and self-driving CAR19

(A) Quantification of Raji tumor burden by in vivo luminescence (bioluminescence) analysis of luciferase expression by engrafted Raji cells at various time points post-tumor engraftment. Geometric mean ± geometric SD is shown. (B) Graphic representation of Raji tumor burden by in vivo luminescence analysis of luciferase expression by engrafted Raji cells at various time points post-tumor engraftment. (C) Survival curves of tumor-implanted, CAR-T-treated NSG mice at various time points post-tumor engraftment. (D) Flow cytometric-based quantification of human T cell levels in the blood of CAR-T-engrafted mice at various time points post-tumor engraftment. Mean ± SEM is shown. (E) Flow cytometric-based quantification of T cell exhaustion (PD1 expression levels) on human T cells present in the blood of CAR-T-engrafted mice at various time points post-tumor engraftment. Mean + SEM is shown. (A−E) 0.5 × 10⁶ Raji-luciferase cells were implanted on day 0 into 8 NSG mice per group, followed by 2.0 × 10⁶ CAR-T cells or PBS (“Tumor only”) on day 7 by intravenous injection. Bioluminescence measurements and blood collections were performed on days indicated. Human T cells were defined as lymphocytes/singlets/live/human CD45⁺. n = 8 mice per condition. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Despite receiving the same initial total T cell dose, the expansion of human T cells in the blood of engrafted mice was significantly greater in the two self-driving CAR-T groups, STAT5-CAR19 and AP1-NFκB-CAR19, as compared to the constitutive EF1α-CAR19 (Figure 4D). This may reflect the heightened T cell activation of the self-driving STAT5-CAR19 and AP1-NFκB-CAR19 T cells in vivo as compared to the constitutive EF1α-CAR19 T cells, in accordance with in vitro results (Figures 2G and 3G). Also, in line with in vitro studies (Figure S4H; day 16 post-activation), the engrafted STAT5-CAR19 T cells demonstrated significantly higher exhaustion marker (PD1) expression at later time points post-engraftment (p < 0.01, day 14; Figure 4E). In summary, the self-driving AP1-NFκB-CAR19, which was similar to the constitutive EF1α-CAR, mediated potent tumor regression and survival.

Use of self-driving CAR19 to drive expression of auxiliary protein: TGFBRIIdn

Given the high functionality of the AP1-NFκB-CAR19 in vitro and in vivo, we explored the feasibility of including auxiliary proteins to be transcriptionally regulated by the inducible promoter of the self-driving CAR in order to extend CAR-T activity under immunosuppressive conditions. To illustrate this potential, we utilized a TGFBRIIdn lacking the intracellular signaling domains to allow the CAR-T cells to evade the immunosuppressive cytokine TGF-β, which is often found in the tumor microenvironment. The TGFBRIIdn protein was co-expressed with the CAR19, via a ribosomal skipping site (P2A) bracketed by furin protease cleavage sites (FP2AF). Constitutively expressed EF1α promoter-driven CAR19 was constructed as a control and tested in parallel with the AP1-NF-κB-regulated CAR TGFBRIIdn. The architecture of each of these armored CAR constructs with TGFBRIIdn coexpression is depicted in Figure 5A.

Figure 5.

Protection of CAR-T cells from anti-inflammatory signaling by the use of a dominant-negative TGF-β receptor (TGFBRIIdn)

(A) Graphical representation of CAR19-TGFBRIIdn vectors used in this study in the context of a self-inactivating lentiviral backbone. FP2AF represents a furin cleavage site-P2A ribosome skip site-furin cleavage site. (B) Graphical representation of CAR-T culture/coculture protocol for the experimental data represented in (C)−(G). All T cells were cultured in 30 IU/mL of IL-2 from day 0 to day 8 post-activation. Coculture 1 was initiated on day 8 post-activation, and coculture 2 was initiated on day 14 post-activation. Cocultures were performed in the in the presence or absence of 10 ng/mL TGF-β. (C) Flow cytometric plots representing T cell TGFBRII expression (native TGFBRII and TGFBRIIdn transgene expression) and quantification of TGFBRII expression prior to coculture with CD19⁺ Raji-GFP (day 8) or following coculture (day 9) in the absence of TGF-β. FMO, fluorescence minus-one control. (D) Quantification of CAR19-dependent cytotoxicity of CD19⁺ Raji-GFP cells, by flow cytometric enumeration of GFP⁺ cells in repeated CAR-T:Raji-GFP cocultures (coculture 1). Geometric mean + geometric SD is shown. (E) Quantification of CAR19-dependent cytotoxicity of CD19⁺ Raji-GFP cells by flow cytometric enumeration of GFP⁺ cells in repeated CAR-T:Raji-GFP cocultures (coculture 2). Geometric mean + geometric SD is shown. (F and G) Quantification of CAR19-dependent CAR-T expansion by flow cytometric enumeration of CD3⁺ cell expansion at the end of repeated coculture 2 of CAR-T:Raji-GFP cocultures (day 20 relative to day 0 post-activation) in the absence (F) or presence (G) of 10 ng/mL TGF-β. Geometric mean + geometric SD is shown. (A−G) n = 3 independent donors. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

T cells were transduced with CAR19 constructs with auxiliary components in the presence of IL-2 and cultivated up to study day 8. The experimental groups included constitutively expressed EF1α-CAR-19 or EF1α-CAR19-TGFBRIIdn and self-driving AP1-NFκB-CAR19 or AP1-NFκB-CAR19-TGFBRIIdn or UTD control (Figures 5A and 5B). For coculture, T cells were co-incubated with target Raji-GFP cells either in the absence of cytokines or in the presence of 10 ng/mL TGF-β in order to simulate a tumor-suppressive microenvironment (Figure 5B). CAR-T cells were co-incubated with target cells for 12 days across two repeated cocultures (cocultures 1 and 2), and changes in CAR, TGFBRII, activation, and exhaustion markers were monitored periodically by flow cytometry (Figure 5C; Figures S6A−S6E).

The expression of CAR19 (Figure 5C; Figures S6A and S6B) was similar between the EF1α-CAR19/EF1α-CAR19-TGFBRIIdn (83.5% and 87.3% CAR19⁺, respectively, prior to coculture) and AP1-NFκB-CAR19/AP1-NFκB-CAR19-TGFBRIIdn groups (86.3% and 83.0% CAR19⁺, respectively, 18 h post-coculture), indicating a lack of effects of the TGFBRIIdn transgene on CAR19 expression (Figure S6A). Furthermore, the CAR19 expression was inducible, following CD19⁺ Raji stimulation in cocultures with AP1-NFκB-CAR19 or AP1-NFκB-CAR19-TGFBRIIdn. Although the flow cytometric analysis could not distinguish between native TGFBRII and the TGFBRIIdn transgene, native TGFβRII expression was low for both EF1α-CAR19⁻ and AP1-NFκB-CAR19 T cells (<8% TGFβRII⁺ T cells under all conditions). In the absence of TGF-β supplementation, TGFBRII expression prior to coculture with Raji target cells was low in AP1-NFκB-CAR19-TGFBRIIdn T cells, and it was significantly upregulated after exposure to Raji targets overnight (two-way ANOVA, p < 0.0001; Figure 5C; Figure S6B). By contrast, EF1α-CAR19-TGFBRIIdn T cells did not change TGFBRII expression following stimulation with CD19⁺ Raji cells (p > 0.05) and displayed weaker TGFBRIIdn expression after antigen stimulation as compared to the AP1-NF-κB promoter (Figure 5C; Figure S6B). Under TGF-β supplementation, a strong downregulation of TGFβRII/TGFBRIIdn was observed (Figure S5B), as postulated previously.^32,33 We demonstrate that the AP1-NF-κB promoter can be used to co-induce expression of proteins other than CARs (e.g., TGFBRIIdn) in the presence of antigen signaling through a cognate CAR.

The expression of the TGFBRIIdn transgene enhanced CAR19-dependent cytotoxicity in the presence of the immunosuppressive cytokine TGF-β, as assessed by repeated cocultures (E:T 1:3) of CAR-T cells with CD19⁺ Raji-GFP cells in the presence or absence of TGF-β (10 ng/mL). In the absence of TGF-β, all CAR19 T cell constructs strongly suppressed Raji expansion as compared to negative controls UTD and Raji only. By contrast, in the presence of TGF-β, the AP1-NFκB-CAR19 and EF1α-CAR19 vectors displayed poor cytotoxicity of Raji cells (Figures 5D and 5E; Figures S7A and S7B). Strikingly, in the presence of TGF-β, the EF1α-CAR19-TGFBRIIdn demonstrated a fully restored CAR-dependent cytotoxicity. Similar restoration of CAR-dependent cytotoxicity was also observed for the AP1-NFκB-CAR19-TGFBRIIdn group, despite the initially lower TGFBRII expression in these CARs (Figures 5D and 5E; Figure S6B).

TGF-β signaling also dramatically affected CAR-T expansion, reducing the EF1α-CAR19 T cells expansion from 1,230-fold (relative to day 0 post-activation) in the absence of TGF-β to 269-fold in the presence of TGF-β and reducing the expansion of AP1-NFκB-CAR19 T cells from 4,265-fold in the absence of TGF-β to 407-fold in the presence of TGF-β (Figures 5F and 5G; Figures S7C and S7D). These anti-proliferative effects of TGF-β were reversed when the TGFBRIIdn transgene was co-expressed under either the EF1α or AP1-NF-κB promoters. Moreover, CAR19-dependent T cell expansion was impaired in EF1α-CAR19 T cells (1,230-fold) relative to AP1-NFκB-CAR19 (4,265-fold; Figure 4F), as observed previously (Figures 2E and 2F; Figure S4G). CAR-T cell expansion was also improved in EF1α-CAR19-TGFBRIIdn relative to EF1α-CAR19 T cells even in the absence of TGF-β supplementation and may reflect protection of CAR-T cells from paracrine production of TGF-β, which is known to occur at low levels in activated T cells.³⁴ Therefore, tumor antigen-inducible expression of TGFBRIIdn under the AP1-NF-κB promoter provided equivalent protection from the anti-inflammatory effects of TGF-β to TGFBRIIdn expression under the control of the constitutive EF1α promoter.

In the presence of the TGF-β cytokine, the expression of the TGFBRIIdn was also associated with reduced expression of the exhaustion markers. When stimulated with CD19⁺ tumor cells, the percentage of CAR-T cells co-expressing the three canonical exhaustion markers, TIM3, LAG3, and PD1, was significantly increased in the presence of TGF-β (Figure S6E). However, the upregulation of the exhaustion markers was reversed in CAR-T cells expressing the armored EF1α-CAR19-TGFBRIIdn and AP1-NFκB-CAR19-TGFBRIIdn groups. As previously reported,²³ TGF-β treatment of CAR-T cells stimulated with CD19⁺ Raji cells leads to upregulation of the exhaustion marker PD1 and downregulation of the activation marker CD25 at late time points in both cocultures 1 and 2 (Figures S6C and S6D). In contrast, expression of the TGFBRIIdn armor prevented the upregulation of the PD1 and downregulation of CD25 (Figures S6C and S6D).

Overall, we demonstrate successful utilization of CARs with auxiliary components (e.g., TGFBRIIdn) in a single-vector system controlled by the self-driving AP1-NF-κB promoter. The self-driving promoter provided equivalent or superior functionality as compared to the highly effective constitutive EF1α promoter, both at the level of the CAR and the auxiliary protein (TGFBRIIdn).

Discussion

In this study, we demonstrate an application of the AP1-NF-κB-CAR signaling-induced promoter to self-regulate CAR19 expression on the basis of engagement with the tumor antigen CD19. The self-driving CAR signaling-proximal promoter AP1-NF-κB provided the best long-term CAR19 anti-tumor function in vitro, as well as the ability to return to basal, low-level expression following tumor clearance, as compared to the CAR signaling-distal STAT5 or the constitutive promoters. The induction of CAR19 under the control AP1-NF-κB promoter was rapid and prolonged over multiple rounds of stimulation, whereas CAR19 expressed under a constitutive promoter (EF1α, MSCV) was either downregulated or internalized within the first 2−3 days after antigen exposure and only recovered afterward. The AP1-NFκB-CAR19 was significantly upregulated following exposure to the Raji cells expressing CD19 at low and medium antigen-surface densities, even after an extended 15-day CAR manufacturing period. Surprisingly, at high CD19 densities, both the constitutive EF1α-CAR19 and the inducible AP1-NFκB-CAR19 surface expression were reduced; however, the cytolytic and the activation responses of the AP1-NFκB-CAR19 remained equal to or greater than those of EF1α-CAR19 at all conditions tested. The AP1-NFκB-CAR19 also demonstrated greater expansion and lower exhaustion even in the absence of IL-2 pre-supplementation, in comparison to EF1α-CAR19. Finally, the control of CAR19 by this self-driving promoter led to equivalent tumor lysis both in vitro and in vivo compared to the constitutively expressed EF1α-CAR19, demonstrating AP1-NFκB-CAR19 equivalence to EF1α-CAR19.

In contrast to the signaling-proximal AP1-NFκB-CAR19, the signaling-distal STAT5-CAR19 was nearly completely downregulated initially upon CD19⁺ Raji addition and was only re-induced by 3−5 days after stimulation. The STAT5 promoter itself was functional in responding to cytokine-mediated activation, by inducing CAR19 expression within 18 h after treatment with STAT5-inducing cytokines, IL-2 or IL-6,³⁵ but not TNF-α. A similar, near-complete downregulation of surface CAR19 within 18 h post-stimulation, the target antigen was observed with the constitutive promoter MSCV. Of note, both the STAT5- and MSCV-driven CAR19 had substantial defects in long-term cytotoxicity and T cell expansion/persistence, suggesting that rapid induction and sustained surface CAR expression are necessary for successful long-term CAR function.

The lack of long-term CAR efficacy and T cell persistence in STAT5-CAR19 and MSCV-CAR19 T cells may be attributed to a defect in production of IL-2 downstream of CAR19 ligation, which stood in contrast to both AP1-NFκB-CAR19 and EF1α-CAR19 T cells. Indeed, IL-2, an important T cell homeostatic cytokine produced by CD4⁺ helper T cells, is only produced transiently following TCR and/or CAR stimulation.³⁶ Therefore, upregulated CAR surface expression immediately following antigen stimulation, such as afforded by the self-driving AP1-NFκB-CAR19, would likely boost CAR-T function in the long term by increasing production of homeostatic cytokines like IL-2. Similarly, the STAT5-CAR19 T cells completely cleared tumors only in three out of eight animals in the Raji NSG xenograft model in vivo, whereas AP1-NFκB-CAR19 and EF1α-CAR19 each cleared tumors in six out of eight treated animals. In line with these results, a recent study using CAR19 mutated to block CAR ubiquitination demonstrated increased surface expression following antigen engagement, which led to superior in vitro and in vivo functionality in terms of tumor cytotoxicity and T cell expansion, and increased 4-1BB signaling,³⁷ which is known to increase IL-2 production.³⁸ Therefore, sustained or increased surface expression of CAR following antigen engagement appears to be important for long-term CAR function.

The fact that the self-driving AP1-NFκB-CAR19 effectively mediated tumor clearance and survival in vivo under low CAR T dose, similar to the constitutive EF1α-CAR19, and was superior to the constitutive CAR in in vivo expansion enabled us to carry out our goal of building a multi-functional “transcriptional circuit.” In this capacity, the self-driving AP1-NF-κB promoter served to restrict both CAR and auxiliary protein expression to conditions where a tumor antigen (CD19) was present. Although no difference was observed in tumor clearance in Raji NSG xenograft mouse models between self-driving AP1-NFκB-CAR19 and constitutive EF1α-CAR19 T cells, it is possible that specialized in vivo experimental conditions, such as a further reduction in CAR T dose, or repeat tumor challenge models may help pinpoint additional advantages of the self-driving CAR per se. For example, the study demonstrating an advantage of ubiquitination-resistant CAR19 used a dose of only 5 × 10⁵ CAR-T cells to demonstrate its efficacy,³⁷ relative to 1.6 × 10⁶ CAR-T cells in this study. However, one of our primary purposes in this study was to demonstrate that the self-driving AP1-NFκB-CAR19 promoter can be used as part of a transcriptional circuit to restrict both CAR and auxiliary protein expression to conditions where a tumor antigen (CD19) was present. Therefore, we elected to determine whether the self-driving AP1-NFκB-CAR19 and constitutive EF1α-CAR19 promoters were equivalent under challenging but relatively standard conditions for in vivo tumor clearance studies. Future studies will determine whether the in vitro advantages of the self-driving promoter translate into increased tumor control in vivo under more restrictive conditions.

As a proof of concept that the self-driving promoter may simultaneously control the expression of multiple payloads, we co-expressed the CD19 CAR with an auxiliary TGFBRIIdn protein under the control of the self-driving AP1-NF-κB promoter. Others have previously used inducible promoters, such as NFAT, to drive IL-12 or IL-18 expression in the context of both tumor-infiltrating lymphocytes³⁹ and CAR-T cells.^40,41 However, these studies utilized dual-promoter bicistronic constructs, whereas the first constitutive promoter drives the expression of a TCR or a CAR, and the engagement of these receptors in turn activated the expression of a cytokine from the inducible NFAT promoter. Unfortunately, under this design configuration, toxicity associated with “leaky” expression of a transgene, e.g., IL-12, in a clinical setting prevented further implementation of this system.³⁹ Also, the use of dual-promoter bicistronic constructs may lead to promoter silencing⁴² and requires high vector payload capacity.

In contrast to these approaches, we elected to use the single inducible AP1-NF-κB promoter to regulate both the expression of the CAR and the auxiliary protein in a CAR activation-dependent manner. Although constitutive, long-term expression of TGFBRIIdn can lead to autoimmunity,²⁵ its expression is less likely to be systemically toxic at low levels²⁸ and may benefit from a system restricting high-level expression to encounter with tumor antigen. We demonstrate that the self-driving circuit of the AP1-NF-κB promoter combined with a prototypical anti-CD19 CAR can be used to limit high expression of both a CAR and auxiliary proteins such as TGFBRIIdn armor to conditions where the tumor antigen is present.

Additionally, payload capacity is critical for lentiviral vector manufacturing.⁴³ The AP1-NF-κB promoter (214 bp) is substantially smaller than the constitutive EF1α promoter (1,196 bp) and would favor incorporation of large CAR auxiliary components, such as pro-homeostatic cytokines (e.g., IL-12, IL-18, IL-15), dn or “switched” immunosuppressive receptors (e.g., TGFBRIIdn, PD1 dn, PD1-CD28), or chemokine receptors (e.g., CXCR2).^28,40,44, 45, 46, 47, 48

In the context of our anti-CD19 CAR, the use of the AP1-NF-κB circuit relative to EF1α constitutive expression provided equivalent tumor lysis, superior CAR-T expansion, and reduced exhaustion. Moreover, the inclusion of TGFBRIIdn armor in this circuit also afforded protection from TGF-β-induced immunosuppression. As a simple, compact promoter-based system, the AP1-NF-κB-CAR transcriptional circuit demonstrates promise for next-generation CAR-T therapies.

Materials and methods

Creation of CAR expression vectors

CAR19 was generated using the single-chain variable fragment (scFv)-targeting domain derived from FMC-63 mouse hybridoma (FMC-63: amino acid [aa] 1−267; GenBank: HM852952.1) in orientation variable light chain (VL) to variable heavy chain (VH) and connected by the (GGGGS)3 flexible intrachain linker. The targeting domain was linked in frame to a human CD8 hinge (aa 138−179, Ref sequence: NP_001759.3), human TNF receptor superfamily 19 transmembrane domain (TNFRSF19; aa 167−196, UniProt sequence: Q9NS68), human 4-1BB co-stimulatory domain (CD137, aa 214−255, UniProt sequence: Q07011), and human CD3ζ signaling domain (CD247, aa 52−163, Ref sequence: NP_000725.1). Leader sequence was derived from the human granulocyte macrophage colony-stimulating factor receptor alpha subunit (aa 1−22, GenBank: EAW98673.1). CAR sequence was codon optimized and cloned into a third-generation lentiviral plasmid backbone under the regulation of surface antigen-regulated inducible promoters STAT5 or AP1-NF-κB or constitutive EF1α or MSCV promoters as a control. The inducible promoters were comprised of five tandem STAT5 response elements or an alternation of six AP1 response elements and five NF-κB response elements, respectively, placed upstream of a minimal TATA box promoter derived from the adenovirus E1b promoter (minP).⁴⁹ Exact sequences of the promoters are shown in Supplemental information.

In order to generate the TGFBRIIdn, the extracellular and transmembrane portion of human TGF-β (TGF-βRII: aa 1−191, UniProt: P37173) was included downstream of the CAR19 sequence, separated by a cleaved ribosome skip site (FP2AF) consisting of the following: a consensus furin cleavage site (aa: RAKR) fused to a ribosome skip site (P2A) derived from the porcine teschovirus-1 polyprotein (aa 976−997, GenBank: CAB40546.1, mutated residue P977S) fused to a consensus furin cleavage sequence (aa: RAKR).

Cell lines used to demonstrate CAR activity

The Burkitt lymphoma cell line Raji, the acute lymphocytic leukemia cell line REH, the chronic myelogenous leukemia line K562, and culture reagents were purchased from American Tissue Culture Collection (ATCC; Manassas, VA, USA), unless otherwise noted. Cells were cultured in RPMI-1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS; HyClone, Logan, UT, USA) and 2 mM GlutaMAX (Thermo Fisher Scientific, Grand Island, NY, USA). Human embryonic kidney line 293T was purchased from ATCC and propagated in CD FortiCHO medium (Gibco/Thermo Fisher Scientific, Grand Island, NY, USA).

Single-cell clones of luciferase-expressing cell lines were generated by stably transducing wild-type tumor lines with lentiviral vector encoding firefly luciferase (Lentigen Technology, Gaithersburg, MD, USA), followed by cloning and selection of luciferase-positive clones. Next, luciferase-transduced tumor lines were transduced with lentiviral vector encoding GFP, followed by fluorescence-activated cell sorting (FACS)-based separation of GFP⁺ cells. For CD19/CD22-expressing 293T cell lines, wild-type 293T cell lines were stably transduced with bicistronic lentiviral vector encoding CD19 and the puromycin-resistance gene via F2A peptide. After 2 weeks of selection with 1 μg/mL puromycin, transduced 293T cell pools were banked, and surface target molecule expression was confirmed by flow cytometry using CD19 antibody (clone: LT19) (Miltenyi Biotec, Bergisch Gladbach, Germany).

To generate Raji cell lines with different CD19 density, a CD19 knockout cell line was first generated from the parental Raji cells using the LentiCRISPR version (v.)2 plasmids carrying a human single guide RNA (sgRNA; AAGCGGGGACTCCCGAGACC; GenScript, Piscataway, NJ, USA) by electroporation on the Amaxa device (Lonza, Basel, Switzerland). CD19-knockout (KO) Raji cells were isolated by CD19 microbead depletion, followed by single-cell cloning and verified by CD19 antibody LT19-phycoerythrin (PE; Miltenyi Biotec, Bergisch Gladbach, Germany). The CD19-KO Raji cell line was stably transduced with a lentiviral vector encoding CD19 fused with puromycin by a 2A peptide. After single-cell cloning by limiting dilution, all available clones were screened with the CD19 antibody for target molecule expression by flow cytometry to generate Raji-CD19^Low and Raji-CD19^High cell lines. The antibody-binding capacity (ABC) assay to determine CD19 molecules per cell was performed as per the manufacturer’s protocol using the BD PE Fluorescence Quantitation Kit (BD Biosciences, San Jose, CA, USA) and BD PE-labeled anti-human CD19 (clone SJ25C1) on Raji-CD19^Low and Raji-CD19^High cell lines, as well as a parental Raji-GFP-luciferase clone (Raji^Med).

Primary human T cells used to demonstrate CAR activity

Whole blood was collected from healthy volunteers at Oklahoma Blood Institute (OBI) with donors’ written consent. Processed buffy coats were purchased from OBI (Oklahoma City, OK, USA) and leukapheresis material from Hemacare (Los Angeles, CA, USA; leukopak), also with donors’ written consent. CD4-positive and CD8-positive human T cells were purified from buffy coats or leukopaks via positive selection using a 1:1 mixture of CD4⁻ and CD8⁻ MicroBeads (Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer’s protocol.

Primary T cell transduction

Human primary CD4⁺ and CD8⁺ T cells from normal donors were cultivated in TexMACS medium at a density of 1 × 10⁶ cells/mL, activated with CD3/CD28 MACS GMP T Cell TransAct reagent on day 0 (all reagents from Miltenyi Biotec), and transduced on days 1/2 with lentiviral vector encoding CAR constructs overnight and media exchanged on days 2/3. Supplementation with cytokines (IL-2, TNF-α, IL-6, TGF-β; Miltenyi Biotec, Bergisch Gladbach, Germany) was performed as described. Cultures were propagated until harvest on days 5−15 for co-incubation analysis. In some cases, CAR-T cells were cryopreserved at harvest in 10% DMSO (Amresco), 70% FBS (HyClone, Logan, UT, USA), and 20% TexMACS in a controlled-rate freezer (Mr. Frosty; Nalgene) and stored at −160°C until reculture.

Immune effector assays: Short-term killing

To determine cell-mediated cytotoxicity (killing assay), 5,000 tumor target cells stably transduced with firefly luciferase were combined with CAR-T cells at various E:T ratios and incubated overnight. SteadyGlo reagent (Promega, Madison WI, USA) was added to each well, and the resulting luminescence quantified as counts per second (sample CPS). Target-only wells (max CPS) and target-only wells plus 1% Tween 20 (min CPS) were used to determine assay range. Percent specific lysis was calculated as: (1− [sample CPS − min CPS]/[max CPS − min CPS]). Three technical replicates were performed for each condition, and each experiment was repeated using CAR-T cells generated from at least three independent donors.

Immune effector assays: Long-term killing and cytokine

Generally, for long-term coculture killing assays, CAR-T effector and GFP⁺ CD19⁺ Raji-GFP target cells were mixed at a ratio of 1 CAR-T:3 GFP⁺ target cells (2.5 × 10⁵ CAR-T: 7.5 × 10⁵ Raji-GFP in a total volume of 2 mL TexMACS), and cells were propagated in TexMACS media without IL-2 supplementation. In some cases, TexMACS was supplemented with 10 ng/mL TGF-β (Miltenyi Biotec). At various time points post-coculture, additional TexMACS media were added to the culture as cells expanded.

In the case of coculture with Raji cell lines expressing varying levels of CD19, CAR-T effector and Raji target cells were mixed at a ratio of 1 CAR-T:1 target cells (5 × 10⁴ CAR-T: 5 × 10⁴ Raji in a total volume of 200 μL TexMACS).

In the case of coculture of 293T cell lines transduced to express CD19 or CD22, the 293T cell lines were labeled immediately prior to coculture with carboxyfluorescein diacetate succinimidyl ester (CFSE; Invitrogen) to distinguish targets and effectors by flow cytometry. CAR-T effector and 293T target cells were mixed at a ratio of 1 CAR-T:1.2 target cells (2.5 × 10⁵ CAR-T: 3 × 10⁵ Raji in a total volume of 200 μL TexMACS).

The extent of target cell population killing and CAR-T population survival and expansion was determined by flow cytometry. Briefly, for cell staining, CAR-T cells and GFP+ targets were harvested from cocultures, and staining was performed in cold AutoMACS buffer supplemented with 0.5% bovine serum albumin (Miltenyi Biotec). Cell populations were gated based on forward and side scatter and viability (7AADnegative; BD Biosciences, San Jose, CA, USA). Percentages of surviving cells in cocultures were determined based on GFP positivity for Raji targets and CD3-VioBlue positivity (Miltenyi Biotec, Bergisch Gladbach, Germany) for CAR-T effectors. In addition, CAR-T expression in live CD3-positive cells was determined by staining with the CD19 Fc peptide (R&D Biosystems, Minneapolis, MN, USA), followed by anti-Fc (Fab′)2-FL-AF647 reagent (Jackson ImmunoResearch, West Grove, PA, USA). Absolute cell counts were obtained by addition of 4 × 10³ CountBright beads (Invitrogen) to each sample prior to staining.

Additional flow cytometry panels included a T cell activation panel: CD8 VioGreen, CD25 PE, and CD69 APC-Vio770; a T cell exhaustion panel: PD1 PE-Vio770, LAG3 APC, TIM3 BV510 (BioLegend); and a cytokine production panel: IL-2 PE-Vio770, interferon (IFN)-γ-APC, and TNF-α-APC-Vio770. All flow cytometry staining was performed in cold AutoMACS buffer supplemented with 0.5% bovine serum albumin with the exception of the cytokine panel, where staining was performed using the Inside Stain Permeabilization Kit, as per the manufacturer’s protocol. All reagents were from Miltenyi Biotec (Bergisch Gladbach, Germany) unless otherwise noted.

For re-coculture long-term killing assays (cocultures 2−3), the time chosen to restimulate CAR-T cells was typically when a standard control CAR (e.g., EF1α-CAR19) under standard conditions (no cytokine supplementation, wild-type CD19⁺ target cells) had killed 95%–99% of the target cells. Briefly, the absolute counts of remaining CAR-T cells and GFP⁺ targets were assessed by flow cytometry, and these counts were used in performing restimulation of cells. Typically, a total number of 2.5 × 10⁵ CAR-T (CD3⁺) cells were added to a culture well, and the total number of GFP-expressing target cells was adjusted to 7.5 × 10⁵ to obtain an E:T ratio of 1:3. Following this coculture, cells were propagated as per coculture 1, and cytotoxicity was assessed by flow cytometric analysis.

VCN analysis

VCN per cell was determined by a quantitative PCR duplex analysis on genomic DNA from human primary T cells transduced with lentiviral vectors as described elsewhere.⁵⁰

In vivo analysis of CAR-T function

This study was carried out in accordance with the recommendations of the National Institutes of Health and Covance Animal Care and Use Committee (formerly Molecular Imaging [MI] Bioresearch, Ann Arbor, MI, USA). The protocol was approved by the MI Bioresearch Animal Care and Use Committee. The function of CD19-targeting CAR-T cells was assessed in vivo. 6- to 8-week-old NSG mice, 8 per group, were injected intravenously (i.v.) with 0.5 × 10⁶ Raji-luciferase cells on day 0. Tumor burden was determined by in vivo imaging system (IVIS) bioluminescent imaging on day 6, mice were then randomized to groups with equal mean tumor burden, and a total of 2.0 × 10⁶ CAR-T cells/mouse was administered on study day 7. Tumor regression was determined by bioluminescent imaging on days 8, 12, 15, 21, 28, and 35 using a Xenogen IVIS-200 instrument (PerkinElmer, Shelton, CT, USA). Images were analyzed using Living Image, v.4.1, software (PerkinElmer), and the total bioluminescent signal flux for each mouse (ventral + dorsal) was expressed as average radiance (photons per second per square centimeter per steradian). Survival was recorded and analyzed at the end of the study. To determine the presence of CAR-T and tumor cells, peripheral blood was collected from all animals on study days 8, 14, 22, and 29. The absolute numbers of blood CAR-T cells and the expression of activation and exhaustion of T cell surface markers CD69 and PD1 were determined by flow cytometry.

Statistical analysis

All statistical analyses were performed with Prism 7 software (GraphPad, San Diego, CA, USA). Technical replicates, representing repeated measurements or treatments of the same donor-derived population of cells, were averaged prior to analysis. Biological replicates represented separate donor-derived cellular populations or separate mice. Bioluminescence, cytotoxicity of target cells, and expansion of T cells, as log-normal parametric data, were log transformed prior to analysis using parametric tests. Mouse experiments were designed to detect a 1.5-fold difference in mean of parametric data at a type I error rate (α) of 0.05 and a statistical power (β) of 0.8 (n = 8 per condition, 5 treatment groups, assumed coefficient of variation = 0.375). In vitro experiments were generally designed to detect a 2-fold difference in mean of parametric data at a type I error rate of 0.05 and a statistical power of 0.8 (n = 3 donor-derived cells per condition, 5 treatment groups, assumed coefficient of variation = 0.375). Statistical significance was determined by one- or two-way ANOVA, followed by Tukey’s post hoc test for multiple comparison correction. Survival was evaluated by Kaplan-Meier test. p values were reported as the following: ns (non-significant), p > 0.05, ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001. Exact p values are provided for all statistical comparisons performed in Table S1. Error bars represent standard deviation or standard error of the mean.