Endogenous CD28 drives the persistent activity of CAR T cells in myeloma and lymphoma models

Abstract

Chimeric antigen receptor (CAR) T cell therapy has reshaped the therapeutic landscape for multiple myeloma (MM), yet most patients treated with BCMA-targeted CAR T cells experience disease relapse. Consequently, we sought to determine if inhibition of CD28 survival signaling could increase MM sensitivity to CAR T cell therapy. Contrary to expectations, blockade of CD28 interaction with CD80/86 accelerated tumor regrowth in preclinical MM and lymphoma CAR T therapy models. Knockout studies revealed that endogenous CD28 on 4–1BB co-stimulated CAR T cells prolonged in vivo activity, reprogrammed mitochondrial metabolism to maintain redox balance, and stimulated proliferation and release of tumor-model specific inflammatory cytokines in the tumor microenvironment. Intriguingly, transient CD28 blockade decreased levels of certain TME cytokines without significantly affecting survival of CAR T cell treated mice. Collectively, these data provide direct evidence that endogenous CD28 signaling modulates CAR T cell responses in multiple myeloma and lymphoma models.

Introduction

Chimeric Antigen Receptor (CAR) T cells are a form of immunotherapy that has seen extraordinary success in treating hematologic malignancies(1). Expression of CARs in autologous T cells isolated from cancer patients redirects T cell specificity toward an antigen of interest and delivers activation signals upon antigen ligation(2,3). Activation signals in clinically relevant second-generation CAR T cells are mediated by CD3ζ as well as a co-stimulatory domain, most commonly CD28 or 4–1BB, although others including ICOS and OX40 have been investigated(4–7). Second generation CAR T cells targeting the B cell antigen CD19 are FDA approved for B cell leukemias and lymphomas(8–11) and have resulted in potent, decade long remissions for some of the initial patients who received this therapy(12). Within the last few years, FDA approval for CAR T cells has expanded to include those directed against B cell maturation antigen (BCMA) for the treatment of multiple myeloma (MM)(13–15). Despite robust initial response rates greatly outperforming standard-of-care in certain heavily pre-treated patient populations, clinical studies have shown that most CAR T cell treated myeloma and lymphoma patients will experience disease progression(13–15). Therefore, there is an urgent need to understand and overcome resistance mechanisms that hinder CAR T cell success(16).

A key mediator of MM cell survival is the canonical T cell co-stimulatory receptor, CD28, whose expression on MM cells correlates with myeloma progression(17). Importantly, CD28 ligands CD86 and CD80 are expressed on dendritic cells (DCs) and stromal cells within the bone marrow (BM) microenvironment(18–20) (Fig. 1A). MM cells themselves can also express CD86, which in MM has pro-survival activity that is dependent on its cytosolic domain(21,22). Further, ligation of CD28 on MM cells by CD86/CD80 transduces a PI3K/Akt pathway dependent, pro-survival signal leading to chemotherapy resistance(23–26). Consistent with the pro-survival roles of CD28 and CD86 on MM cells, low expression of CD86 and CD28 by MM cells is associated with better patient outcomes (22). More than half of mature B cell lymphomas (BCL) also express CD86 but, unlike MM, CD80 is expressed in nearly half of BCL patients and CD28 is never expressed(27). Antibody ligation of CD86, but not CD80, on BCL cells provides a survival and growth signal(28). Further complicating matters, activated T cells upregulate CD86 and CD80, and these molecules can ligate CD28 in cis or trans(29). Importantly for studies in pre-clinical mouse models, human and mouse CD28 are cross-reactive to human and mouse CD80 or CD86(30) (Fig. 1B) and cross-species interactions are functional(31).

Figure 1: Complex CD28 and B7 protein interactions in CAR T cell therapy models.

(A) General expression pattern of CD28 and B7 proteins (CD86 & CD80) on normal immune cells, CAR T cells, and B cell lineage tumors. CD28 is expressed on resting T cells, activated T cells, and CAR T cells, unless experimentally knocked out (CD28^KO). CD86 is not expressed on resting T cells but is upregulated with T cell activation and maintained on CAR T cells. Multiple myeloma (MM) cells can express CD28 and/or CD86 and are clinically targeted by CAR T cells specific for BCMA. B cell lymphoma (BCL) cells can express CD86 and/or CD80 and are clinically targeted by CAR T cells specific for CD19. B7 proteins are also expressed by endogenous antigen presenting cells (APCs), with reduced expression in immunocompromised mouse strains (NSG and Rag2^−/−).

(B) Table showing cross-reactivity of human CD28, CD86, CD80, and CTLA4-Ig with murine CD28, CD86, and CD80. + indicates cross-reactivity.

(C) Schematic of second-generation retroviral CAR construct used to generate human BCMA targeted human(45) or mouse(77) CAR T cells or murine CD19 targeted mouse CAR T cells(78).

(D) CD28-CD86/80 interactions likely to occur in CAR T cell therapy models used in this study. Left – Related to Figs. 2A – 2D, 5D,5E, & 7A, 7B. CD28 on human BCMA targeted, 4–1BB costimulated (hBCMABBζ) CAR T cells could be stimulated in trans by human CD86 on MM.1S myeloma cells or in cis by human CD86 expressed by hBCMABBζ CAR T cells. Additionally, murine APCs (mAPC) in NSG mice express low levels of murine B7 proteins capable of stimulating human CD28 in trans. CD28 acts as a survival signal on MM cells and could also be stimulated in cis or in trans by human or mouse B7 proteins (not shown in illustration). Abatacept (CTLA4-Ig) binds to B7 proteins with higher affinity than CD28, thereby blocking cis and trans CD28 stimulation. Middle – In the 5TGM1 mouse MM model used in Figs. 2F, 2G, 5A – 5C, & 6A – 6C, CD28 expressed by hBCMAmBBmζ CAR T cells can be stimulated in cis or in trans by reduced-level B7 proteins on mAPC in Rag2^−/− mice. CD28 on 5TGM1 cells can be stimulated in cis by B7 proteins expressed by mAPC or CAR T cells (not shown in illustration). Right – In the immune competent A20 BCL model used in Figs. 3A – 3C, 6D, 6E, & 7C, 7D, CD28 on CD19mBBmζ CAR T cells can be stimulated in cis or in trans by B7 proteins on A20 cells and mAPC. Figure 1A,D images made with BioRender.

Clinically, the interaction of CD28 with CD86/CD80 can be blocked by the human CTLA4-Ig fusion protein abatacept, which is FDA approved for the treatment of rheumatoid arthritis, psoriatic arthritis, polyarticular juvenile idiopathic arthritis, and acute graft versus host disease(32,33). Mouse studies further showed that human CTLA4-Ig can functionally block murine T cell activation(34). In myeloma, pre-clinical studies have shown that CTLA4-Ig, in combination with the chemotherapeutic melphalan, can significantly reduce tumor burden(24). Preliminary results from a phase II clinical trial of abatacept plus ixazomib and dexamethasone for treatment of patients with relapsed/refractory MM (NCT03457142) suggest that overall response can be increased by approximately 3-fold when abatacept is added(35).

In the current study, we hypothesized that systemic blockade of CD28 would similarly sensitize MM to CAR T cell therapy. We reasoned that unlike endogenous T cells that rely on CD28 co-stimulation to mount an antitumor response(36,37), second generation CAR T cells receive co-stimulation directly from the CAR and would therefore be relatively unaffected by blockade of endogenous CD28. FDA approved CAR T cell products for MM - idacabtagene vicleucel and ciltacabtagene autoleucel - and for BCL – tisagenlecleucel and lisocabtagene maraleucel - employ 4–1BB co-stimulatory domains, which have been shown to transduce a weaker signal than CD28 co-stimulatory domains(38,39). CARs used in this study similarly employed 4–1BB co-stimulatory domains (Fig. 1C). In CD19 targeted CAR T cells, the weaker 4–1BB co-stimulatory signal reduced CAR T cell exhaustion and enhanced in vivo persistence when compared to CD28 driven co-stimulation(40,41). However, recent evidence suggests that enhanced CD28 signaling in CTLA-4 knockout 4–1BB co-stimulated (BBζ) CAR T cells improves their antitumor efficacy(42). Moreover, endogenous tumor-reactive cytotoxic T cells rely on CD28 signaling to acquire effector properties in the tumor microenvironment(36,43,44), suggesting that perhaps the CAR 4–1BB co-stimulatory domain alone is insufficient to stimulate potent antitumor activity.

To test our hypothesis, we employed human and mouse orthotopic models of multiple myeloma (MM) and B cell lymphoma (BCL) to directly test whether endogenous CD28 affected tumor control by 4–1BB CAR T cells (Fig. 1D). Unexpectedly, data demonstrated that continuous blockade of CD28 interaction with CD86/CD80 using abatacept significantly impaired long-term CAR T cell tumor control in both MM and BCL preclinical models. Data further suggested that abatacept primarily affected endogenous CD28 signaling on CAR T cells, as inducible deletion of CD28 from 4–1BB co-stimulated CAR T cells also reduced sustained in vivo and in vitro antitumor activity. Mechanistically, we provide evidence that endogenous CD28 signaling increased in vivo expansion, stimulated oxidative phosphorylation, and maintained redox balance of 4–1BB co-stimulated CAR T cells. Endogenous CD28 on CAR T cells also stimulated production of tumor-model specific proinflammatory cytokines in the mouse MM and BCL tumor microenvironment (TME). In agreement, a higher frequency of grade 2 or greater cytokine release syndrome (CRS) was observed in a small cohort of CAR T cell-treated MM patients with CD86⁺ disease, when compared to their CD86^- counterparts, although this difference was not statistically significant. Transient inhibition of endogenous CD28 on 4–1BB co-stimulated, BCMA-targeted CAR T cells resulted in decreased accumulation of CD4⁺ CAR T cells and reduced inflammatory cytokine levels in the TME of myeloma-bearing mice without significantly impairing antitumor activity. In the more aggressive syngeneic A20 BCL model, transient CD28 blockade decreased levels of IL-6 in the TME without significant impairment of mouse survival following CD19 CAR T cell therapy. Collectively, these findings reveal that the persistent in vivo function of 4–1BB co-stimulated CAR T cells is enhanced by stimulation of endogenous CD28, which can be transiently blocked to reduce pro-inflammatory cytokine release while having limited effect on antitumor activity.

Results

Blocking CD28-CD86/CD80 interaction impairs hBCMABBζ CAR T cell anti-MM activity

Since inhibition of the CD28 survival signal in human or mouse multiple myeloma (MM) cells sensitizes them to in vitro or in vivo chemotherapy exposure(23,24), we sought to determine whether CD28 inhibition similarly sensitized MM cells to killing by CAR T cells. Human CAR T cells targeting BCMA and containing a 4–1BB/CD3ζ intracellular signaling domain (hBCMABBζ; Fig. 1C) along with a CD8α hinge and transmembrane domain were generated from healthy donor PBMCs transduced with a previously described retroviral vector(45) (Supplemental Fig. 1A). Co-culture of hBCMABBζ CAR T cells with human MM cell lines MM.1S or U266, which differ in their expression profiles of BCMA and CD86 (Supplemental Fig. 1B), or U266 overexpressing the extracellular domain of CD86 tethered to the membrane with truncated NGFR (tCD86) resulted in expected cytotoxicity across a range of effector to target ratios (Supplemental Fig. 1C). Addition of abatacept (CTLA4-Ig), a potent blocker of CD28-CD86 interaction, to co-cultures had no discernable effect on short-term in vitro killing by hBCMABBζ CAR T cells. In agreement, production of T cell cytokines in co-culture of CAR T cells with MM.1S cells was unaffected by abatacept, as determined by a standard T cell cytokine panel (Supplemental Fig. 1D).

We next evaluated effects of abatacept on in vivo control of CD28⁺CD86⁺ MM growth by hBCMABBζ CAR T cells. Luciferase-tagged human MM.1S cells were implanted intravenously ( i.v.) into NSG mice. Four weeks later, 3 × 10⁶ hBCMABBζ CAR T cells were infused with or without 3x/weekly injections of abatacept (200μg/mouse) continued until endpoint (Fig. 2A). Bioluminescence imaging (BLI) was used to normalize tumor burden across groups immediately prior to therapy and to monitor MM.1S burden following CAR T cell infusion. As expected, MM regression was observed in hBCMABBζ CAR T cell treated mice, with many mice lacking detectable tumor burden 2 to 3 weeks following infusion (Fig. 2B). Unexpectedly, abatacept treatment for the duration of the experiment shortened the time to progression following hBCMABBζ CAR T cell infusion (Supplemental Fig. 1E), resulting in significantly shortened survival of MM.1S bearing mice (Fig. 2C).

Figure 2: CD28 inhibition impairs BCMA-targeted CAR T cell anti-myeloma efficacy.

(A) Diagram of experimental setup. NSG mice were intravenously inoculated with 1 × 10⁶ MM.1S-luc myeloma cells on day −28, or day −35 for the high tumor burden model, and injected i.v. with 3 × 10⁶ hBCMABBζ CAR T cells on day 0. Mice received 200 μg abatacept i.p. 3x/week beginning the day before CAR T cell infusion and continuing through endpoint. Tumor burden was monitored by IVIS bioluminescent imaging (BLI) 2x/week. Image made with BioRender.

(B) Bioluminescent images of MM.1S bearing mice on specified days relative to CAR T cell infusion. The representative experiment shown was repeated three times (4 mice per condition per experiment; 3 experiments total).

(C) Kaplan – Meier survival analysis of hBCMABBζ CAR T or mock transduced T cells ± abatacept treated MM.1S-luc bearing mice (n = 8 – 12 mice per CAR T cell treated group & n = 4 – 6 mice per control group). Median survival of hBCMABBζ CAR T treated mice was >100 days post CAR T cell infusion vs. 55 days for hBCMABBζ CAR T + abatacept treated mice vs. 24 days for mock transduced T cell treated mice vs. 22.5 days for abatacept treated mice vs. 26 days for mock transduced T cell + abatacept treated mice. Groups were compared using Logrank Mantel-Cox tests; ****P < 0.0001, ns = P > 0.05.

(D) Kaplan – Meier survival analysis of hBCMABBζ CAR T or off-target control hCD19BBζ CAR T cells ± abatacept treated MM.1S-luc high tumor burden mice (n = 4 mice per group). Median survival of hBCMABBζ CAR T treated mice was 80 days vs. 45 days for hBCMABBζ CAR T + abatacept treated mice vs. 27.5 days for hCD19BBζ CAR T treated mice vs. 24 days for hCD19BBζ CAR T+ abatacept treated mice. Groups were compared using Logrank Mantel-Cox tests; ****P < 0.0001, **P =0.007, ns = P > 0.05.

(E) CD28 dependence of persistent in vitro anti-myeloma activity of human hBCMABBζ CAR T cells. Unmodified or CD28 CRISPR knockout human hBCMABBζ CAR T cells produced from two independent donors were co-cultured with fresh U266 or U266 cells expressing tCD86 (U266^tCD86) every two days for up to 12 repeated stimulations.

(F) Diagram of experimental setup. Rag2^−/− mice were inoculated intravenously with 2 × 10⁶ 5TGM1^hBCMA-luc cells on day −14 and treated with 3 × 10⁶ CD28^fl/fl or CD28^iKO CAR T cells on day 0. Tumor burden was monitored by IVIS bioluminescent imaging (BLI) 2x/week through endpoint. Image made with BioRender.

(G) Kaplan – Meier survival analysis of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cell or mock transduced T cell treated 5TGM1^hBCMA-luc bearing mice (n = 6 – 8 mice per CAR T cell treated group & n = 3 mice in the Mock group). Median survival of CD28^fl/fl hBCMAmBBmζ CAR T treated mice was 38 days post-CAR T cell infusion vs. 24 days for CD28^iKO hBCMAmBBmζ CAR T treated mice vs. 16 days for mice treated with mock transduced T cells. Groups were compared using Logrank Mantel-Cox tests; ***P =0.0008, **P = 0.0014, *P = 0.017.

Abatacept stimulation of CD86/CD80 on bone marrow resident DCs can contribute to an immunosuppressive MM TME by inducing IL-6 and activity of the tryptophan catabolizing enzyme, indoleamine 2,3-dioxygenase (IDO)(46). An immunosuppressive TME can predict response to CAR T cell therapy in MM patients(47,48), however, we found no indication that abatacept induced IL-6 or IDO activity in the bone marrow of MM.1S bearing CAR T cell treated mice (Supplemental Fig. 1F and 1G). Moreover, abatacept accelerated relapse (Supplemental Fig. 1H) and shortened survival (Fig. 2D) following hBCMABBζ CAR T cell infusion in the high tumor burden setting, as defined by an additional week of tumor growth (Fig.2A), where an immunosuppressive MM TME should be established prior to CAR T cell infusion. These data demonstrate that the primary effect of continuous blockade of CD28 via abatacept following CAR T cell therapy for MM is not sensitization. Rather, data suggest that CD28 signaling may be important for maintaining antitumor activity of CAR T cells.

Endogenous CD28 sustains 4–1BB co-stimulated CAR T cell antitumor activity

Despite the clear reduction of in vivo CAR T cell efficacy imparted by continuous abatacept treatment, data do not differentiate effects of abatacept on cells in the MM TME versus effects of blocking endogenous CD28 signaling in CAR T cells. To test whether endogenous CD28 supports persistent tumor control by 4–1BB co-stimulated CAR T cells, we generated human CD28 CRISPR knockout hBCMABBζ CAR T cells (Supplemental Fig. 2A) and performed repeated stimulation assays using U266 MM cells or U266 cells overexpressing a signaling deficient CD86-tNGFR chimeric protein (U266^tCD86). Fusion of the extracellular domains of CD86 to a signaling deficient NGFR transmembrane domain has been shown to uncouple the ligand function of CD86 from its survival signal(22). Killing of CD86-tNGFR expressing U266 cells was maintained over the course of several more rounds of hBCMABBζ CAR T cell co-culture restimulation than CD86-negative U266 cells. Importantly, the stimulatory effect of CD86-tNGFR on persistent cytotoxic activity was abrogated by deletion of CD28 from hBCMABBζ CAR T cells (Fig. 2E).

To further examine CAR T cell intrinsic effects of CD28, we generated a tamoxifen inducible CD28 knockout mouse model (CD28^iKO), in which CD28 is knocked out in all cells, by crossing CD28-floxed mice to mice expressing Cre ERT2 from the ROSA26 locus(49). Following hBCMAmBBmζ CAR transduction of CD28^iKO or littermate control mouse T cells lacking CreERT2 expression, CAR T cells were expanded in media containing 250 nM 4-hydroxytamoxifen (4-OHT) to induce CD28 deletion (Supplemental Fig. 2B). Importantly, CD28 surface expression was reduced to near background levels in CD28^iKO CAR T cells by 4-OHT exposure (Supplemental Fig. 2C), although CD28 staining was still statistically increased relative to isotype, while CD4:CD8 ratio and CAR expression was unaffected (Supplemental Fig. 2D, 2E). To interrogate functionality of murine CAR T cells, we engineered murine 5TGM1 MM cells(50) to express a chimeric human BCMA (hBCMA)-tNGFR target antigen (5TGM1^hBCMA; Supplemental Fig. 2F). Coupling the extracellular domains of hBCMA to a signaling deficient NGFR transmembrane domain allowed for uncoupling of the CAR target function of BCMA from its survival signal. In co-culture assays, CD28^iKO CAR T cells were as effective as littermate control CD28^fl/fl CAR T cells at killing 5TGM1^hBCMA myeloma cells (Supplemental Fig. 2G) and produced comparable amounts of cytokines (Supplemental Fig. 2H), indicating that endogenous CD28 does not directly impact CAR T cell cytotoxic function.

In contrast to in vitro findings, CD28^iKO hBCMAmBBmζ CAR T cells differed greatly from hBCMAmBBmζ CAR T cells generated from CreERT2-negative littermates in their ability to control in vivo myeloma growth (Fig. 2F). CD28^iKO hBCMAmBBmζ CAR T cells transiently controlled systemic growth of luciferase labeled 5TGM1^hBCMA myeloma in Rag2^−/− mice while CD28^fl/fl littermate control hBCMAmBBmζ CAR T cells demonstrated extended myeloma control (Supplemental Fig. 2I). As a result, the median survival of 5TGM1^hBCMA myeloma bearing mice treated with CD28^fl/fl hBCMAmBBmζ CAR T cells was 1.52 times (38 days vs. 24 days) that of those treated with CD28^iKO hBCMAmBBmζ CAR T cells (Fig. 2G).

To expand on these findings, we next asked whether 4–1BB co-stimulated CD19 targeting CAR T cells were similarly affected by CD28 blockade. BALB/c mice were inoculated i.v. with 1 × 10⁶ syngeneic CD86⁺CD80^- A20 BCL cells followed 5 days later with a single intraperitoneal (i.p.) injection of 100 mg/kg cyclophosphamide (CTX) to induce lymphodepletion(51). On the day following lymphodepletion, 3x/weekly i.p. injections of abatacept (200μg) were initiated, followed one day later with a single i.v. injection of 3 × 10⁶ CAR T cells targeting murine CD19 (m19mBBmζ; Fig. 3A). As in human MM models, continuous abatacept treatment of A20 BCL bearing Balb/cJ mice blunted the antitumor effect of CAR T cells (Fig. 3B), leading to shorter survival of mCD19mBBζ CAR T cell + abatacept treated mice compared to those treated with mCD19mBBζ CAR T cells alone (Fig. 3C). To further test effects of endogenous CD28 on CAR T cells in immunocompetent mice, CD28^iKO or CD28^fl/fl m19mBBmζ CAR T cells were injected into lymphodepleted C57BL/6 mice and their long-term ability to maintain B cell aplasia was examined (Fig. 3D). Flow cytometry assessment of peripheral B cells demonstrated that CD28^fl/fl m19mBBmζ CAR T cells were able to maintain B cell aplasia significantly longer than CD28^iKO m19mBBmζ CAR T cells (Fig. 3E), confirming the lack of persistent in vivo activity of CD28 knockout, 4–1BB co-stimulated CAR T cells. Taken together, these data are consistent with the notion that engagement of endogenous CD28 signaling on 4–1BB co-stimulated CAR T cells drives persistent antitumor activity.

Figure 3: CD28 inhibition limits CD19-targeted CAR T cell activity in immunocompetent mice.

(A) Diagram of experimental setup. Balb/cJ mice were intravenously inoculated with 1 × 10⁶ CD19⁺ A20-GFP/luc lymphoma cells on day −7 and injected i.v. with 3 × 10⁶ mCD19mBBmζ CAR T cells on day 0. Mice received 200 μg abatacept i.p. 3x/week beginning the day before CAR T cell infusion and continuing through endpoint. Tumor burden was monitored by IVIS bioluminescent imaging (BLI) 2x/week. Image made with BioRender.

(B) Bioluminescent images of A20-GFP/luc bearing mice on specified days relative to CAR T cell infusion. The representative experiment shown was repeated twice (5 mice per condition per experiment; 2 experiments total).

(C) Kaplan – Meier survival analysis of mCD19mBBmζ CAR T cell ± abatacept treated A20-GFP/luc bearing mice (n = 10 mice per group). Groups were compared using a Logrank Mantel-Cox tests; *P = 0.022.

(D) Diagram of experimental setup. C57BL/6J mice were lymphodepleted with cyclophosphamide (CTX, 200 mg/kg) on day −2 followed on day 0 by i.v. injection of 1 × 10⁶ murine CD19 targeted CAR T cells (m19mBBmζ) or mock transduced T cells from CD28^iKO mice or CD28^fl/fl mice. The frequency of CD19⁺ cells in the peripheral blood of mice was determined weekly for 8 weeks by flow cytometry. Image made with BioRender.

(E) Mean frequency ± standard error of the mean (SEM) of B cells (%CD19⁺ of live CD45⁺ cells) in the peripheral blood of C57BL/6J mice following injection of m19mBBmζ CAR T cells made from CD28^iKO mice (n = 9) or CD28^fl/fl mice (n = 8) or mock transduced T cells (n = 8 per group). Horizontal dashed lines depict the normal range of B cell frequency in C57BL/6J mice. Groups were compared using Two-Way ANOVA with Tukey correction for multiple comparisons; ***P < 0.001, ****P < 0.0001, ns = P > 0.05.

Endogenous CD28 regulates CAR T cell oxidative metabolism

CD28 controls metabolic reprogramming of activated T cells to enhance production of pro-inflammatory cytokines and antitumor immunity(52–55). Since CAR T cell efficacy is linked to the metabolic state of infused CAR T cells(56), we evaluated glycolytic and mitochondrial metabolism of unstimulated and 5TGM1^hBCMA stimulated CD28^fl/fl and CD28^iKO hBCMAmBBmζ CAR T cells using Seahorse assays. Results of Seahorse Glycolysis Stress Tests were equivalent between CD28^fl/fl and CD28^iKO CAR T cells (Fig. 4A, 4B), indicating that mBBmζ CAR signaling was sufficient to induce glycolytic metabolism. In contrast, CD28^iKO hBCMAmBBmζ CAR T cells displayed reduced mitochondrial respiration in Seahorse Mito Stress Tests (Fig. 4C,4D). Basal and uncoupled (Maximal) oxygen consumption rate (OCR) were decreased in unstimulated CD28^iKO CAR T cells while uncoupled OCR and spare respiratory capacity (SRC) were decreased in stimulated CD28^iKO CAR T cells. Reduced OCR in CD28^iKO CAR T cells is consistent with the established role of CD28 in priming mitochondria to support a robust recall response of memory CD8 T cells(54). Yet in contrast to memory CD8 T cells, reduced mitochondrial OCR in CD28^iKO CAR T cells did not result from diminished fatty acid oxidation nor from an inability of CD28^iKO CAR T cells to oxidize other major anaplerotic substrates glucose and glutamine (Supplemental Fig. 3A). Moreover, no difference in mitochondria content (Supplemental Fig. 3B), nor in the contribution of complex I or complex II to mitochondrial oxygen consumption (Supplemental Fig. 3C) was observed when comparing CD28^iKO and CD28^fl/fl hBCMAmBBmζ CAR T cells. The ratio of reduced (NADH) to oxidized (NAD+) nicotinamide adenine dinucleotide increased in a CD28-dependent manner when hBCMAmBBmζ CAR T cells were stimulated with hBCMA expressing 5TGM1 target cells (Fig. 4E). Additionally, mitochondrial ROS was elevated in both stimulated and unstimulated CD28^iKO CAR T cells (Fig. 4F). These data identify oxidative mitochondrial metabolism and redox balance as potential targets of endogenous CD28 in 4–1BB co-stimulated CAR T cells.

Figure 4: Perturbation of oxidative metabolism and redox homeostasis in CD28^iKO hBCMAmBBmζ CAR T cells.

(A) Representative Seahorse Glycolysis Stress Test performed following 24 hr. stimulation of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells with 5TGM1^hBCMA myeloma cells. Connected data points represent mean extracellular acidification rate (ECAR) ± standard deviation (SD) of four technical replicates at indicated time points during a representative experiment that was repeated at least 3 times. Dashes indicate the timing of glucose, oligomycin (Oligo), and 2-deoxyglucose (2-DG) injection.

(B) Quantified rates of glycolysis (left), glycolytic capacity (middle), and glycolytic reserve (left) in CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells ± 24-hr. stimulation with 5TGM1^hBCMA myeloma cells. Bars represent mean ± SD, dots represent independent experiments. Groups were compared using mixed-effects analysis with Šidák correction for multiple comparisons; ns = P > 0.05.

(C) Representative Seahorse Mito Stress Test performed 24 hr. after stimulation of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells with 5TGM1^hBCMA myeloma cells. Connected data points represent mean oxygen consumption rate (OCR) ± SD of four technical replicates at indicated time points during a representative experiment that was repeated at least 3 times. Dashes indicate the timing of oligomycin (Oligo), FCCP, and rotenone (Rot) + antimycin A (Ant. A) injection.

(D) Quantified basal OCR (left), uncoupled maximal respiration (middle), and spare respiratory capacity (SRC) expressed as percent of basal OCR (right) in CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells ± 24 hr. stimulation with 5TGM1^hBCMA myeloma cells. Bars represent mean ± SD, dots represent independent experiments. Groups were compared using mixed-effects analysis with Šidák correction for multiple comparisons; *P < 0.05, **P < 0.01.

(E) Ratio of NADH to NAD⁺ in CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells prior to 4-OHT mediated CD28 deletion or ± 24 hr stimulation with 5TGM1^hBCMA myeloma cells. Bars represent mean ± SD from 3 independent experiments. Groups were compared using mixed-effects analysis with Šidák correction for multiple comparisons; *P = 0.03, ns = P > 0.05.

(F) Representative histograms depicting MitoSOX Red staining (left), gated on MitoTracker Green-bright CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells ± 24 hr. stimulation with 5TGM1^hBCMA myeloma cells. Quantification of the frequency of cells highly stained with MitoSOX Red (right) within the MitoTracker Green-bright population. Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; **P < 0.01. The representative experiment shown was repeated three times.

Endogenous CD28 enhances CAR T cell expansion in the MM TME

Due to the known influence of mitochondrial respiration and redox balance on T cell proliferation(54,57,58), we next evaluated the ability of CD28^fl/fl and CD28^iKO CAR T cells to undergo proliferative expansion. Over the course of ex vivo CAR T cell manufacturing, CD28^fl/fl and CD28^iKO T cells expanded at similar rates (Supplemental Fig. 4A). Further, co-culture of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells with 5TGM1^hBCMA target cells similarly induced expression of the proliferation marker Ki67 (Supplemental Fig. 4B). However, when hBCMAmBBmζ CAR T cells in the MM TME or peripheral blood were enumerated 7 days after adoptive transfer into 5TGM1^hBCMA bearing mice (Fig. 5A), a marked decrease in CD4⁺ CD28^iKO CAR T cells was observed in both peripheral blood (Fig. 5B) and in the MM bone marrow TME (Fig. 5C). Importantly, CD28^iKO hBCMAmBBmζ CAR T cells maintained overall lower CD28 surface expression in the MM TME, compared to CD28-intact controls (Supplemental Fig. 5A). In a mouse model of human MM (Fig. 5D), abatacept treatment of MM.1S bearing mice reduced the frequency of human CD4⁺ hBCMABBζ CAR T cells in the TME (Fig. 5E), but had no effect on peripheral CAR T cells (Supplemental Fig. 5B). A similar reduction in CD4⁺ CAR T cells was observed in the MM TME of abatacept treated 5TGM1^hBCMA bearing mice (Fig. 5F). Interestingly, the frequency of CD8⁺ CAR T cells in the MM TME or peripheral blood of MM bearing mice was unaffected by CD28 knockout or abatacept treatment (Fig. 5 & Supplemental Fig. 5).

Figure 5: CD28 stimulates in vivo expansion of 4-BB co-stimulated CAR T cells.

(A) Diagram of experimental setup for murine myeloma CAR T cell model. 5TGM1^hBCMA bearing Rag2^−/− mice were treated with CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells and euthanized 7 days later for blood and hind limb bone marrow collection. Image made with BioRender.

(B) Quantification of CD4⁺ and CD8⁺ CAR T cells by flow cytometry in peripheral blood one week following adoptive transfer of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells into 5TGM1^hBCMA myeloma bearing mice. Bars represent mean ± standard deviation (SD), dots indicate individual mice (n = 3–6 mice per group). Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; ****P < 0.0001, ns = P > 0.05.

(C) Frequency or CD4⁺ and CD8⁺ CAR T cells assessed by flow cytometry in bone marrow (BM) one week following adoptive transfer of CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells into 5TGM1^hBCMA myeloma bearing mice. Bars represent mean ± SD, dots indicate individual mice (n = 3–6 mice per group). Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; ***P = 0.0001, ns = P > 0.05.

(D) Diagram of experimental setup for human xenograft myeloma CAR T cell model. MM.1S bearing NSG mice were treated with hBCMABBζ CAR T cells on day 0 ± 200μg abatacept on days −1, 1, 3, 5, 7 and then euthanized for blood and hind limb bone marrow collection.

(E) Quantification of CD4⁺ and CD8⁺ CAR T cells by flow cytometry in BM one week following adoptive transfer of hBCMABBζ CAR T cells into MM.1S myeloma bearing mice ± abatacept. Bars represent mean ± SD, dots indicate individual mice (n = 4–5 mice per group). Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; ****P < 0.0001, ns = P > 0.05.

(F) Frequency or CD4⁺ and CD8⁺ CAR T cells assessed by flow cytometry in BM one week following adoptive transfer of hBCMAmBBmζ CAR T cells into 5TGM1^hBCMA myeloma bearing mice ± abatacept. Bars represent mean ± SD, dots indicate individual mice (n = 3–6 mice per group). Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; ***P = 0.0007, ns = P > 0.05.

(G) IVIS bioluminescence images of T-lux luciferase expressing hBCMAmBBmζ CAR T cells ± abatacept (200μg, 3x/week) on day 7 or day 15 after infusion. The representative experiment shown was repeated twice.

(H) Photon flux by IVIS imaging within the hind limb region of interest (ROI) over a four-week period following T-lux hBCMAmBBmζ CAR T cell infusion ± abatacept into 5TGM1^hBCMA myeloma bearing mice. Dots represent individual mice (n = 3 mice per group), lines are best-fit sigmoidal curves, and curves were compared using mixed-effects modeling. The representative experiment shown was repeated twice.

The observed reduction in CD4⁺ CAR T cells in the MM TME upon deletion or blockade of endogenous CD28 led us to test the contribution of endogenous CD28 signaling to in vivo expansion of luciferase expressing hBCMAmBBmζ CAR T cells generated from T-lux mice, which were engineered to express firefly luciferase in T cells (59). Approximately one week after infusion, which aligns with the kinetics of tumor regression (Supplemental Fig. 2I), T-lux hBCMAmBBmζ CAR T cell luminescence within the hind limbs of 5TGM1^hBCMA myeloma bearing mice rapidly increased with a signal plateau observed approximately 1 week later (Fig. 5G, 5H). Abatacept treatment significantly blunted in vivo expansion of T-lux hBCMAmBBmζ CAR T cells in the MM TME (Fig. 5H), confirming that CD28 signaling supports in vivo expansion of 4–1BB co-stimulated CAR T cells.

Endogenous CD28 on CAR T cells stimulates inflammatory cytokine production in the TME

Because CAR T cell activation and expansion is associated with a spike in cytokine production(60), we next sought to test the effect of abatacept on CAR T cell stimulated cytokine production. To this end, myeloma bearing mice were treated with abatacept for one week following CAR T cell infusion. At this early timepoint, abatacept had no effect on the anti-MM activity of BCMA targeted human or mouse CAR T cells, as assessed by tumor burden (Supplemental Fig. 6A, 6B), and only a very minor effect on the levels of CAR T cell-induced human inflammatory cytokines in the MM TME (Supplemental Fig. 6C). These findings led us to conclude that CD28 blockade with abatacept was unlikely to affect in vivo CAR T cell cytokine secretion nor antitumor activity in the first week following infusion, consistent with in vitro findings.

Similar experiments conducted using a murine model of MM (Fig. 6A) yielded strikingly different results. Overall cytokine levels in the MM TME of hBCMAmBBmζ CAR T cell treated mice were reduced by abatacept or by knockout of CD28 (Fig. 6B). Importantly, abatacept did not further reduce cytokine levels in the MM TME of CD28^iKO hBCMAmBBmζ CAR T cell treated mice. Notable among inflammatory cytokines affected by both abatacept treatment and CD28 deletion were CXCL10, which is secreted by monocytes and stromal cells in response to IFN-γ(61), IL-1β, which is primarily expressed by myeloid cells(62), and IL-12, which is mainly secreted by monocytes, macrophages, neutrophils, and dendritic cells(63) (Fig. 6C). Cytokines primarily derived from T cells, including IFNγ and IL-2, were variably suppressed in the TME by CD28 knockout and unaffected by abatacept (Supplemental Fig. 6D). Despite marked effects within the bone marrow TME, cytokines detected in plasma of myeloma bearing mice were unaffected by CD28 knockout (Supplemental Fig. 6E).

Figure 6: CAR T cell CD28 stimulates inflammatory cytokine release in the TME.

(A) Diagram of experimental setup for murine myeloma CAR T cell model. 5TGM1^hBCMA bearing RAG2^−/− mice were treated with 5 × 10⁶ CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells on day 0 ± 200μg abatacept on days −1, 1, 3, 5, 7 and then euthanized for hind limb bone marrow collection. Image made with BioRender.

(B) Heatmap representation of mean log₂ transformed murine cytokine levels in the MM TME 7 days after treatment of 5TGM1^hBCMA bearing mice with CD28^f/f or CD28^iKO hBCMAmBBmζ CAR T cells ± abatacept. Bilateral hind limbs were harvested from 2 – 3 mice per group and BM was flushed into 15 μL PBS for CodePlex multiplexed cytokine analysis. Groups were compared using Two-Way ANOVA with Tukey correction for multiple comparisons; **** P < 0.0001, ***P = 0.0001, ns = P > 0.05.

(C) Bar graphs showing concentrations of myeloid-associated cytokines, murine CXCL10, IL-1β, and IL-12, in the TME 7 days after treatment of 5TGM1^hBCMA bearing mice with CD28^fl/fl or CD28^iKO hBCMAmBBmζ CAR T cells ± abatacept. Bars represent mean ± standard deviation (SD) of 2–3 biological replicates. Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; *P < 0.05, **P < 0.01, ***P < 0.001.

(D) Diagram of experimental setup for syngeneic murine lymphoma CAR T cell model. A20 bearing Balb/cJ mice were treated with mCD19mBBmζ CAR T cells on day 0 ± 200μg abatacept on days −1, 1, 3, 5, 7 and then euthanized for hepatic tumor collection. Image made with BioRender.

(E) Bar graphs showing concentrations of cytokines, murine IL-6, IFNγ, and IL-12, in the TME 7 days after treatment of A20 bearing mice with mCD19mBBmζ CAR T cells ± abatacept. Cytokine concentrations in tumor interstitial fluid were measured using Luminex Discovery Assays. Bars represent mean ± SD of 5 biological replicates. Groups were compared using One-Way ANOVA with Tukey correction for multiple comparisons; *P < 0.05, ns = P > 0.05.

Further examination of effects of CD28 on CAR T cell driven inflammatory cytokine secretion in the TME was performed in mCD19mBBmζ treated A20 BCL bearing mice (Fig. 6D). Critically, A20 BCL cells do not lose CD86 expression upon engraftment in the livers of BALB/c mice (Supplemental Fig. 6F). Within the A20 TME 7 days following CAR T cell infusion relatively minor CAR T cell driven increases in inflammatory cytokine release were observed. Among CAR T cell induced cytokines, only IL-6 was significantly decreased by abatacept treatment (Fig. 6E). IL-6 was not significantly altered in the TME of myeloma bearing mice (Supplemental Fig. 6D), reflecting one of many differences between the MM and BCL models used in these studies. Despite these differences, data demonstrate that inhibition of endogenous CD28 in 4–1BB co-stimulated CAR T cells can decrease release of certain proinflammatory cytokines in the TME.

Many cells within the TME, including inflammatory myeloid cells, B cells, and even MM and BCL cells express the B7 protein CD86(21,22). Our data show that CD86 can act as a ligand for CD28 on CAR T cells (Fig. 2E), and as such may stimulate inflammatory cytokine release in CAR T cell treated patients. To further investigate this possibility, we examined immune related adverse events (irAEs) and CD86 expression on CD138⁺ MM cells from 23 patients treated with CAR T cells at Roswell Park between 2021 and 2024 (Supplemental Table 1). Clinical flow cytometry(64–67) revealed that thirteen patients (57%) exhibited CD86 expression on greater than 10% of abnormal CD138⁺ cells in their bone marrow. The frequency of MM cell CD86 positivity ranged from 12% to 86%, with a median of 41% (Supplemental Fig. 7). Among the thirteen patients with CD86-positive MM, eleven (85%) experienced CAR T therapy associated cytokine release syndrome (CRS), with six patients (46%) experiencing grade 2 CRS. Of the ten CAR T cell treated patients with CD86-negative MM, seven (70%) experienced CRS, with only 2 patients (20%) experiencing grade 2 or higher CRS (Supplemental Table 1). Immune effector cell-associated neurotoxicity (ICAN) frequency was similar when comparing patients with CD86-positive (4 patients) and CD86-negative MM (4 patients), although the length of ICAN was greater among patients with CD86-positive than CD86-negative MM (mean = 26.7 days vs. 1.3 days), although this difference was not statistically significant. Of note, irAE severity in MM patients was not explained by differences in treatment. Nine of eleven (82%) CD86-positive and six of seven (86%) CD86-negative MM patients who developed CRS were treated with tocilizumab upon CRS onset, anakinra was used at a similar rate in both populations, and treatment with dexamethasone was more common among patients with CD86-positive than CD86-negative MM (Supplemental Table 1). These intriguing clinical data are consistent with increased proinflammatory cytokine levels observed in mouse models following treatment with CD28-expressing CAR T cells (Fig. 6), but require validation in larger cohorts of CAR T cell treated patients before conclusions regarding potential links between CD86 expression and CAR T cell toxicity can be made.

Transient CD28 blockade does not significantly affect CAR T cell antitumor activity.

Based on our observation that CD28 expressed on CAR T cells stimulates certain inflammatory cytokine responses to 4–1BB co-stimulated CAR T cells that were reduced by abatacept treatment, without affecting early antitumor CAR T cell activity, we predicted that transient abatacept exposure would not impair survival of CAR T cell treated tumor bearing mice. To test this, abatacept or vehicle control was administered every other day to MM.1S (CD86⁺) tumor bearing NSG mice from day −1 to day 7 post hBCMABBζ CAR T cell infusion followed by monitoring of MM growth and mouse survival (Fig. 7A). Transient abatacept exposure resulted in a mild reduction in tumor regression induced by hBCMABBζ CAR T cells (Supplemental Fig. 8) but did not significantly affect the long-term survival of hBCMABBζ CAR T cell treated myeloma bearing mice (Fig. 7B). Similar experiments performed using mCD19mBBmζ CAR T cells to treat A20 BCL bearing BALB/c mice (Fig. 7C) yielded similar results (Fig. 7D).

Figure 7: Transient CD28 blockade does not affect survival of CAR T cell treated mice.

(A) Diagram of experimental setup. MM.1S bearing NSG mice were treated with hBCMABBζ CAR T cells on day 0 ± 200μg abatacept on days −1, 1, 3, 5, 7 and tumor burden was monitored by bioluminescent imaging 2x/week until endpoint. Image made with BioRender.

(B) Kaplan – Meier survival analysis of hBCMABBζ CAR T ± transient (7 days) abatacept treated MM.1S-luc bearing mice (n = 9 – 10 mice per CAR T cell treated group). Mock transduced or untreated MM.1S-luc bearing mice (n = 5) were included as controls. Groups were compared using Logrank Mantel-Cox tests; ****p < 0.0001.

(C) Diagram of experimental setup. A20 bearing Balb/cJ mice were treated with mCD19mBBmζ CAR T cells on day 0 ± 200μg abatacept on days −1, 1, 3, 5, 7 and tumor burden was monitored by bioluminescent imaging 2x/week until endpoint. Image made with BioRender.

(D) Kaplan – Meier survival analysis of m19mBBmζ CAR T ± transient (7 days) abatacept treated A20-GFP/luc bearing mice (n = 6 – 7 mice per group). Groups were compared using a Logrank Mantel-Cox test.

Discussion

CAR T cell therapies targeting BCMA or CD19 have shown curative potential in patients with relapsed/refractory multiple myeloma (MM)(13–15) and B cell lymphoma (BCL)(8–11), respectively. However, achieving long-term remissions remains an ongoing challenge, with many patients experiencing disease relapse within one year of CAR T infusion. In this study, we set out to determine whether CD28 blockade using abatacept could sensitize MM cells to CAR T cell therapy in a manner analogous to standard chemotherapy(24,35). In contrast to expectations, we found that abatacept limited efficacy of 4–1BB co-stimulated CAR T cells in human xenograft MM and syngeneic BCL mouse models. Using a CD28 inducible knockout mouse to generate CD28-deficient (CD28^iKO) CAR T cells, we further revealed a role for the endogenous CD28 receptor in driving persistent in vivo antitumor activity of 4–1BB co-stimulated CAR T cells. CD28 deletion did not alter 4–1BB co-stimulated CAR T cell cytotoxic capabilities nor inflammatory cytokine production in vitro, but rather resulted in disrupted mitochondrial metabolism, as evidenced by decreased mitochondrial oxygen consumption coupled with increased reactive oxygen species, and a lower ratio of the oxidized form of nicotinamide adenine dinucleotide (NADH) to its reduced form (NAD⁺). These metabolic changes were accompanied by reduced in vivo expansion of CD28^iKO 4–1BB co-stimulated CAR T cells and CAR T cell induced inflammatory cytokine production in the TME. Blockade of CD28 interaction with its ligands through administration of abatacept throughout the entirety of experiments similarly reduced in vivo 4–1BB co-stimulated CAR T cell expansion and inflammatory cytokine production. Importantly, however, short-term blockade of endogenous CD28 using abatacept during the first week following CAR T cell infusion reduced inflammatory cytokine levels in the TME without significantly reducing long-term survival of CAR T cell treated tumor bearing mice. These data establish a role for endogenous CD28 in inducing cytokine release in the TME and sustaining the antitumor activity of 4–1BB co-stimulated CAR T cells.

Co-stimulation has long been known to be critical for antitumor effects of CAR T cells(68,69), with different CAR-encoded co-stimulatory domains having distinct effects on CAR T cell properties(4,5). Clinically available CARs contain either a CD28 or a 4–1BB co-stimulatory domain. CD28 co-stimulated CAR T cells exhibit rapid antitumor effector function but lack functional persistence associated with 4–1BB co-stimulated CAR T cells. Modulation of CAR-encoded CD28 signaling has resulted in improved functional persistence and reduced CAR T cell exhaustion in pre-clinical models(70,71). Studies have hinted at a role for endogenous CD28 in determining CAR T cell efficacy. However, evidence for endogenous CD28 modulation of CAR T cell function was either indirect, in the case of CTLA4 knockout(42), complicated by CAR-CD28 heterodimerization due to CD28 transmembrane domain incorporation in CARs(72,73), or by co-expression of IL-12 from a fourth-generation armored CAR construct(74). Data presented here provide direct evidence that endogenous CD28 signaling affects efficacy of second-generation, 4–1BB co-stimulated CAR T cells comparable to those used to treat myeloma and leukemia/lymphoma patients. These data raise important questions about how signaling from CAR co-stimulatory domains cross-talks with signaling from endogenous co-stimulatory, and/or co-inhibitory receptors. While we focus on 4–1BB co-stimulated CARs because of their clinical use in myeloma and leukemia/lymphoma, whether CD28 co-stimulated CAR T cells – axicabtagene ciloleucel and brexucabtagene autoleucel - similarly rely on endogenous CD28 signaling is an open question with clinical relevance for leukemia/lymphoma patients.

Robust CAR T cell activation and expansion can result in immune related adverse events (irAEs), including cytokine release syndrome (CRS), immune effector cell-associated neurotoxicity (ICAN), and neutropenia(60). In pivotal CAR T trials in MM(13–15), 76 – 84% of patients experienced CRS, 18 – 21% experienced ICAN, and nearly all patients experienced hematologic toxicity, although these were generally transient. Clinical data from a small cohort of CAR T cell treated myeloma patients suggest a difference in CRS severity among patients with >10% CD86 expressing MM cells in their bone marrow TME, with 46% of CD86-positive patients (6/13 patients) and 20% of CD86-negative patients (2/10 patients) experiencing grade 2 or higher CRS. However, while conclusions regarding toxicity cannot be drawn from such small patient cohorts, these observations may provide reasoning for further experimental testing. The rates of tocilizumab and anakinra treatment were similar between CD86-positive and CD86-negative MM patients who experienced CAR T irAEs and dexamethasone was more commonly used to treat irAEs in CD86-positive patients, suggesting CRS severity may not be explained by differences in patient treatment. While these clinical data are preliminary and require confirmation in a larger cohort of patients, CRS rates in this small cohort of CAR T cell treated MM patients, taken together with mouse TME cytokine data (Fig. 6), suggest that cytokine release commonly associated with CAR T cell therapy could potentially be exacerbated by CD86 interaction with endogenous CD28 on 4–1BB co-stimulated CAR T cells in the MM TME.

Interestingly, cytokines affected by CAR T cell therapy in a CD28-dependent manner were different when comparing MM and BCL models. In the 5TGM1 MM model, CAR T cell therapy induced, CD28-dependent cytokines were mainly myeloid associated (i.e. CXCL10, IL-1β, and IL-12), with only IFNγ known to be produced by T cells. Interestingly however, in the A20 BCL model, IFNγ and IL-12 were unaffected and IL-6 was reduced by abatacept treatment. These discrepancies may reflect inherent differences in tumor type, mouse strain, location and composition of associated TMEs, or aggressiveness of the tumor model relative to timing of when cytokine levels were measured. Our findings highlight the need for deeper understanding of the complexities of pre-clinical CAR T cell models, or at minimum use of several complementary pre-clinical models to control for inter-model variability, to improve predictive value for CAR T cell therapy in human cancer patients(75).

Overall, results presented here provide direct evidence that endogenous CD28 is important for driving persistent antitumor activity of 4–1BB co-stimulated CAR T cells. These results introduce many interesting and important biological questions about co-stimulation and signaling in CAR T cells and raise the possibility that blocking endogenous CD28 signaling could reduce inflammatory cytokine production associated with CAR T therapy. Future studies aimed at deciphering the crosstalk between endogenous receptor and CAR signaling have the potential to unlock novel clinical strategies capable of enhancing the efficacy and/or safety of CAR T cell therapy.

Materials and Methods

Detailed descriptions of reagents, cell lines, mouse strains, and software used in these studies can be found in the Key Resources Table.

Myeloma patient data:

Retrospective data from 23 multiple myeloma patients treated with CAR T cells at Roswell Park between 2021 and 2024 were collected in accordance with the Roswell Park IRB-approved protocol BDR183624 and with ethical standards. The Roswell Park IRB granted a waiver of patient informed consent because many patients were deceased at the time of study initiation. MM patient data are shown in Supplemental Table 1. Demographic and treatment data shown were manually abstracted from electronic patient charts. Flow cytometry data were collected using established and heavily validated methods(64–67) and were reanalyzed using WinList software to confirm and quantify antigen expression on myeloma cells.

Cell lines:

Parental 5TGM1 cells generously provided by G. David Roodman (Indiana University) were transduced with a lentiviral vector coding the extracellular domain of human BCMA (hBCMA) fused to the transmembrane domain of truncated NGFR and firefly luciferase (see Methods section Viral Vectors for details). hBCMA expressing 5TGM1 populations were enriched to >99% through fluorescence-activated cell sorting (FACS). Luciferase expressing clones of MM.1S and U266 were generated as previously described(76). Briefly, MM cell lines were transfected with pGL4-luc and single cell luciferase-expressing clones obtained by limiting dilution. U266-luc cells were further transduced with a lentiviral vector coding a chimeric protein consisting of the extracellular domain of human CD86 and the transmembrane domain of truncated NGFR (CD86-tNGFR; see Methods section Viral Vectors for details). Lentiviral transduction of MM cell lines was performed in Opti-MEM medium (Gibco) using lentiviral particles generated through cotransfection of lentiviral and packaging vectors into 293T cells using FuGENE 4K Transfection Reagent (Promega, catalog #E5911). GFP/luciferase expressing A20 cells were generated as previously described(51). Briefly, A20 cells were transduced to express GFP and luciferase and then FACS sorted to obtain a pure GFP-positive population. MM and BCL cell lines were maintained for a maximum of 3 months in RPMI 1640 (Gibco) supplemented with 10% heat-inactivated FBS (R&D Systems), 1% nonessential amino acids (Gibco), 1mM sodium pyruvate (Gibco), 10 mM HEPES (Gibco), 2 mM L-glutamine (Gibco) and 1% penicillin/streptomycin (Gibco). 293T cells (ATCC) and 293 Galv9 cells were maintained in DMEM, 10% FBS and 1% L-glutamine. All cell lines were routinely tested for mycoplasma using the Lonza MycoAlert Detection Kit.

Cell Line Authentication:

U266-luc and MM.1S-luc were authenticated by STR profiling, with 100% match, as follows. To acquire a DNA profile for each cell line, fifteen short tandem repeat (STR) loci and a gender-determining marker Amelogenin as well as positive and negative controls were amplified with the AmpFLSTR^® Identifiler^® Plus PCR Amplification Kit (ThermoFisher Scientific, catalog #A26182) on the Applied Biosystems Verti 96-well Thermal Cycler in 9600 Emulation Mode (Initial Denature: 95°C 11 min, 28 cycles of Denature: 94°C 20 sec and Anneal/Extend: 59°C 3 min, Final Extension: 60°C 10 min, and Hold: 12°C forever). The PCR products were run on the Applied Biosystems 3130xl Genetic Analyzer and analyzed with Applied Biosystems GeneMapper v4.0. Eight of the fifteen STRs and Amelogenin from the DNA profile for the cell line(s) are compared to ATCC STR database and the DSMZ combined Online STR Matching Analysis.

Viral vectors:

Retroviral CAR constructs have been previously described. The human BCMA targeted CAR (hBCMABBζ) was described by Smith et al(45), mouse hBCMA targeted CAR (hBCMAmBBmζ) described by Bezverbnaya et al.(77), and mouse CD19 targeted CAR (mCD19mBBmζ) described by Li et al. (78). Additional lentiviral vectors used to ectopically express hBCMA on 5TGM1 cells and CD86 on U266 cells were constructed using the LeGO-Luc2 vector, a gift from Boris Fehse (Addgene plasmid #154006). Briefly, gBlocks (IDT) coding the extracellular domain of human BCMA or CD86 fused to the transmembrane domain of NGFR were cloned into LeGO-Luc2 upstream of a P2A sequence and the gene coding for luciferase. All vectors were verified by whole plasmid sequencing performed by Plasmidsaurus.

Mouse CAR T cell production:

Mouse CAR T cell production was adapted from previous reports(79,80). Briefly, pan CD3⁺ murine T cells were isolated from single cell splenocyte suspensions of 6 – 12-week-old mice through a negative selection process (STEMCELL Technologies, catalog #19851). T cells were activated with αCD3/αCD28 Dynabeads as specified by manufacturer’s instructions in the presence of 100 IU/mL recombinant mouse (rm) IL-2 and 10 ng/mL rmIL-7 (BioLegend, catalog #575406 & #577804, respectively) and cultured in RPMI1640 supplemented with 10% FBS, 2 mM L-glutamine, 10mM HEPES, 0.5% 2-mercaptoethanol, and 1% penicillin/streptomycin. Retroviral transduction was achieved by spinoculation of 3 × 10⁶ mouse T cells on retronectin-coated plates (Takara Bio, catalog # T100B) with retroviral supernatant harvested from 293T packaging cells (2000×g, 60 min., 30 °C) at 24 and 48 hr. post activation. CAR T cells were maintained at 1 × 10⁶ cells/mL for 7–10 days in vitro in the presence of rmIL-2 and rmIL-7.

Human CAR T cell production:

Human CAR T cell production was adapted from previously published protocols(45). De-identified, healthy donor peripheral blood mononuclear cells (PBMCs) were obtained through the Roswell Park donor center under the approved protocol BDR 115919. PBMCs were isolated from whole blood through density gradient centrifugation using Cytiva Ficoll-Paque^™ PLUS Media (ThermoFisher Scientific, catalog #45–001-749). PBMCs were activated with T cell TransAct polymeric nanomatrix (Miltenyi Biotec, catalog #130–111-160) according to manufacturer’s specifications in the presence of 100 IU/mL recombinant human (rh) IL-2 and 10 ng/mL rhIL-7 (Peprotech, catalog #200–02-50UG & #200–07-10UG, respectively). Spinoculation with neat 293 Galv9 retroviral supernatant was performed at 48, 72 and 96 hr. post activation (3200 rpm, 60 min., 30 °C). Human CAR T cells were expanded for 14 days and subsequently cryopreserved in 90% FBS, 10% DMSO.

CRISPR-edited human CAR T cells:

Human T cells isolated from PBMCs were activated with Human T-Activator CD3/CD28 Dynabeads (ThermoFisher Scientific, catalog #11–132-D), after which CD28 knockout was completed using the P3 Primary Cell 4D-Nucleofector Kit (Lonza, catalog #V4XP-3024) following manufacturer’s instructions. Briefly, sgRNA (5’ TCGTCAGGACAAAGATGCTCAGG 3’), Alt-R^™ S.p. Cas9 (Integrated DNA Technologies, catalog #1081059), and Poly-L-glutamic acid (Millipore Sigma, catalog# P4886)) were combined at a 2:1:1 molar ratio and incubated at 37 °C for 15 minutes to form ribonucleoprotein (RNP) complex. T cells were resuspended in Nucleofector solution, combined with RNP at 1 million T cells per 50 pmol SpCas9, and transferred to cuvettes for electroporation using the 4D-Nucleofector system under the pulse code EO-115, generating CD28 knockout T cells (CD28^KO). As a control, T cells electroporated with only the sgRNA and Poly-L-glutamic acid, without Cas9, retain CD28 (CD28^WT). CD28^KO and CD28^WT T cells were spinoculated with retroviral supernatant to generate CD28^KO and CD28^WT hBCMABBζ CAR T cells, after which CAR T cells were expanded and cryopreserved.

Mice and in vivo CAR T models:

All animal studies were performed in accordance with the Roswell Park Comprehensive Cancer Center Institutional Animal Care and Use Committee guidelines under the approved protocol 1094M.

CD28^iKO:

CD28^iKO mice were generated by Ozgene (Australia). LoxP sites flanking exon 2 and 3 of the CD28 gene were introduced to allow for Cre-mediated deletion of the CD28 gene. Mice were generously provided by Kelvin Lee (Indiana University) and subsequently bred in-house under protocol 1425M. Splenocytes were isolated as previously described and CAR T cells were expanded in the presence of 250 nM 4-hydroxytamoxifen (Sigma-Aldrich, catalog #H6278) for 4 days to induce CD28 deletion.

T-lux:

Transgenic T-lux mice expressing firefly luciferase from a human CD2 minigene, as described in ref.(59), were generated by Casey Weaver at the University of Alabama at Birmingham were acquired by AJ Robert McGray (Roswell Park) under a material transfer agreement (MTA). Mice were utilized as splenocyte donors for CAR T cell manufacturing for in vivo imaging of CAR T cell trafficking and expansion.

MM Models:

NOD scid gamma (NSG) mice (NOD.CgPrkdc^scid Il2rg^tm1Wjl/SzJ) mice ages 6–12 weeks old were purchased from the Comparative Oncology Shared Resource in-house mouse colony at Roswell Park. RAG2^−/− (B6.Cg-Rag2^tm1.1Cgn/J, strain #008449) mice were purchased from the Jackson Laboratory and subsequently bred in our facility under the approved protocol 1425M. NSG mice were intravenously injected with 1 × 10⁶ MM.1S-Luc at week −4 and 3 × 10⁶ CAR T cells at week 0. RAG2^−/− mice were injected with 2 × 10⁶ 5TGM1^hBCMA-Luc at week −2 to and 3 × 10⁶ CAR T cells at week 0 to compensate for differences in tumor engraftment rate amongst the two models. Bioluminescence was measured 2x/week using an IVIS^® Spectrum In Vivo Imaging System housed in the Roswell Park Translational Imaging Shared Resource to assess tumor burden. Mice were injected with 150 mg Luciferin/kg of body weight and briefly anesthetized through isoflurane inhalation during image acquisition. Data was analyzed on the Living Image analysis software (PerkinElmer). In some settings, retroorbital blood collection was performed 1 week after CAR T cell infusion to examine CAR T cell frequency in circulation. Mice were monitored daily for signs of deteriorating condition or disease progression including decreased activity, hunched posture, ruffled coat, or hind limb paralysis and euthanized upon veterinary recommendation.

Syngeneic BCL Model:

6 week old female Balb/c mice purchased from The Jackson Laboratory (strain #000651) were intravenously inoculated with 1 × 10⁶ A20 cells expressing GFP and luciferase. Mice were lymphodepleted using cyclophosphamide (100mg/kg) 2 days before CAR T cell treatment and randomized into respective treatment groups. Beginning one day before CAR T cell infusion, mice received intraperitoneal injections of Abatacept (200μg/mouse; see acknowledgements for reagent information) 3 times a week until endpoint. For the survival study, mice were treated with 3 × 10⁶ CAR T cells, and tumor progression was monitored by bioluminescence imaging. Mice in the cytokine study received 10 × 10⁶ CAR T cells and 7 days after CAR T cell infusion, plasma was collected and mice were euthanized for tissue processing. Liver was excised and homogenized into single cell suspension by mechanical disruption in 500 μL of PBS containing cOmplete^™, EDTA-free Protease Inhibitor Cocktail (Roche, catalog #11836170001). Lysates were filtered and pelleted by centrifugation, and the resulting supernatant was collected and cryopreserved until analysis by multiplex Luminex assays (see Methods section Cytokine Analyses for details).

In vitro cytotoxicity assays:

In vitro CAR T cell killing assays were performed using firefly luciferase expressing target cells. Briefly, 2 × 10⁴ target cells were seeded in 96 well plates, varying numbers of CAR T cells were added to assess CAR T cell mediated killing within the linear range. Target cell viability following co-culture incubation was determined using the ONE-glo EX luciferase reporter assay (Promega, catalog # E8130).

Repeated antigen stimulation assays:

In vitro persistent CAR T cell activity was determined using repeated antigen stimulation assays, performed as in previous reports(81). 1.5 × 10⁶ CD86-negative or CD86-expressing U266 target cells were seeded in 6 well plates, after which CD28 wild type or CD28 CRISPR knockout human hBCMABBζ CAR T cells were cocultured with target cells at an E:T of 1:1. 48 hours after cocultures were established, an aliquot was collected to determine CAR T cell cytotoxicity using flow cytometry. CAR T cells were restimulated with additional target cells in fresh media at an E:T of 1:1, and cocultures were analyzed and established every 2 days until end of repeated stimulation.

Flow cytometry:

Data was acquired on either a LSR Fortessa (BD Biosciences) or Cytek Aurora full spectrum analyzer (Cytek Biosciences). Analysis was performed using FlowJo (Tree Star Inc.) or FCS Express software (De novo Software). Briefly, cell suspensions were harvested, washed, and stained with LIVE/DEAD^™ Fixable Blue Dead Cell Stain Kit (Invitrogen, catalog #L34962) in PBS followed by surface antibody staining in FACS buffer (1% BSA, 0.1% sodium azide in PBS). Antibodies were titrated for optimal staining for 20 min. at 4°C. Intracellular cytokine staining was conducted following fixation and permeabilization according to manufacturer’s instructions (BioLegend, catalog #420801 & #421002). Antibodies used in phenotypic analysis are included in the Key Resources Table and gating strategies can be found in Supplemental Gating Strategies.

Cytokine analyses:

Cytokine levels within the TME and blood of MM bearing mice was evaluated using Isoplexis’ CodePlex secretome chips (Isoplexis, catalog #CODEPLEX-2L10 or #CODEPLEX-2L01). Briefly, peripheral blood or bilateral, tumor-bearing hind limbs (femur and tibia) were harvested, and BM was collected into 30 μL PBS. Following centrifugation, soluble fractions were collected and stored at −80 °C in low-bind Eppendorf tubes until loaded onto a CodePlex secretome chip with CodePlex Secretome Mouse Inflammation Panel (Isoplexis, catalog #PANEL-2L10) or CodePlex Secretome Human Adaptive Immune Panel (Isoplexis, catalog #PANEL-2L01 )and analyzed in an IsoLight instrument housed in the Roswell Park Flow & Image Analysis Shared Resource. Cytokine levels in BCL bearing mice were measured in hepatic tumor interstitial fluid custom Luminex Discovery Assays (R&D, catalog #LXSAMSM-06). In vitro cytokine levels were measured using MILLIPLEX^® Th17 premixed panels (Millipore, catalog #HT17MG-14K-PX25 [human] or MT17MAG47K-PX25 [mouse]). A Luminex xMAP INTELLIFLEX system housed in the Roswell Park Flow & Image Analysis Shared Resource was used to acquire Luminex data, which was analyzed using the Belysa^® Immunoassay Curve Fitting Software (Millipore Sigma).

Seahorse:

The day prior to assay CAR T cells were stimulated overnight at an E:T ratio of 2:1. Cell‑Tak‑Coated XFe96 microplate (Agilent Technologies, catalog # 103794–100) was prepared to support testing of cells grown in suspension. Sensor cartridge was hydrated in XF Calibrant and incubated overnight at 37 °C in a non-CO₂ incubator. On the day of the experiment Seahorse XF DMEM Medium pH 7.4 (Agilent Technologies, catalog #103575–100) was supplemented with 10 mM glucose, 1 mM pyruvate and 2 mM glutamine for oxygen consumption rate (OCR) examination and 2mM glutamine for extracellular acidification rate (ECAR) examination and pre-warmed to 37 °C. Suspension cells were harvested, washed in 1X PBS, resuspended in the prepared assay media, and gently seeded on the Cell‑Tak‑Coated plate at 2 × 10⁵ cells/well. The seeded plate was incubated in a 37 °C non-CO₂ incubator for 1 hour prior to the assay. During the incubation, test compounds specific to the assay type were prepared and added to the ports of the hydrated cartridge. For Mito Stress Tests Oligomycin (2.0 μM), FCCP (1.0 μM) and Rotenone/Antimycin A (0.5 μM) were used. For Glycolysis Stress Tests Glucose (10 mM), Oligomycin (1.0 μM) and 2DG (50 mM) were used. For T Cell Metabolic Fitness Tests, the first injection included substrate pathway specific inhibitors Etomoxir (4.0 μM), long chain fatty acid oxidation, UK5099 (2.0 μM), glucose/pyruvate oxidation, or BPTES (3.0 μM) glutamine oxidation. Following inhibitor injections of Oligomycin (1.5 μM), BAM15 (2.5 μM), and Rotenone/Antimycin A (0.5 μM) were preformed for all wells. The loaded cartridge was moved to the XFe96 Analyzer housed in the Roswell Park Flow & Immune Analysis Shared Resource and initial calibration was performed. Following the 1-hour incubation the Cell-Tak plate was transferred to the Xfe96 Analyzer and the assay was initiated according to manufacturer’s recommendations. The final injection of each assay included Hoechst 33342 Nuclear Stain (ThermoFisher, catalog #H3570) to facilitate Normalization via fluorescent imaging and cell counting supported by the BioTek Cytation 5 Cell Imaging Multimode Reader (Agilent Technologies). Data was analyzed using the Wave 2.6.1 software and the Seahorse Analytics cloud-based resource.

NAD⁺/NADH Quantitation:

NADH:NAD⁺ ratios were determined using the NAD/NADH-Glo assay (Promega) according to manufacturer’s instructions. Briefly, 1 × 10⁵ stimulated CAR T cells were washed with 1x PBS prior to cell lysis. NAD⁺ and NADH levels were quantified independently using acid/base treatment. Luminescence values were read on a BioTek Synergy H1 plate reader (Agilent Technologies).

Statistical analyses:

All data shown, unless otherwise noted as representative data, were generated from at least two independent biological replicate experiments. Statistical analyses were performed using GraphPad Prism software. Data points represent independent biological replicates. Error bars represent standard deviation unless otherwise stated. Statistical significance between groups was determined by paired or unpaired Student’s t test, One-Way or Two-Way ANOVA, or mixed-effects modeling with appropriate corrections for multiple comparison testing applied. Survival analysis was performed using Logrank Mantel-Cox tests. In all experiments a p-value < 0.05 was considered significant.

Statement of Significance:

This study provides direct evidence that endogenous CD28 on 4–1BB co-stimulated CAR T cells promotes cytotoxic activity and production of inflammatory cytokines in the tumor microenvironment. These findings have important implications for ongoing efforts to improve CAR T therapy for the treatment of hematological malignancies.

Data Availability Statement

All data associated with this paper are included in the manuscript and supplementary materials. Requests for resources and reagents should be directed to and will be fulfilled by the corresponding author, Scott H. Olejniczak (scott.olejniczak@roswellpark.org)