Target antigen-displaying extracellular vesicles boost CAR T cell efficacy in cell and mouse models of neuroblastoma

Abstract

Glypican-2 (GPC2) and the disialoganglioside GD2 are validated CAR T cell targets in neuroblastoma, but durable clinical responses remain limited. This modest CAR T cell efficacy is in part due to suboptimal T cell persistence, antigen downregulation, and a hostile tumor microenvironment which includes immune cell modulating extracellular vesicles (EVs). Neuroblastoma-derived EVs may contain CAR targets or other immunoregulatory elements that can modulate CAR T cell anti-tumor activity. Thus, we first profiled the surfaceome of neuroblastoma EVs and assessed their impact on both GPC2 and GD2 CAR T cell function. Neuroblastoma EVs displayed GPC2 and GD2, with minimal expression of programmed death-ligand 1 (PD-L1), and were detected in blood from tumor-bearing mice and patients. These EVs directly activated paired CAR T cells, suggesting a role for a peripheral source of CAR antigen. To exploit this therapeutically, we engineered non-tumor-derived GPC2+ synthetic EVs (SyntEVs) as CAR T cell enhancers and armored them with either albumin- or GD2-binding domains. In mice harboring human neuroblastoma cell line- or patient-derived xenografts, serial infusion of armored SyntEVs following GPC2 CAR T cells enhanced tumor control by boosting peripheral CAR T cell persistence. Moreover, GD2-targeting SyntEVs decorated low-antigen tumor cells with GPC2, circumventing antigen downregulation. This SyntEV platform offers a versatile system to address the therapeutic limitations of CAR T cells in solid tumors.

One Sentence Summary:

Vaccination with antigen-displaying SyntEVs enhances GPC2 CAR T cell persistence and anti-tumor efficacy in models of neuroblastoma.

INTRODUCTION

Chimeric antigen receptor (CAR) T cell therapy has demonstrated remarkable success in treating patients with hematological malignancies (1). Despite these advances, early clinical studies in patients with solid tumors have generally fallen short of achieving similar sustained clinical responses, even when the targeted antigens are differentially and abundantly expressed (2). Barriers to the effectiveness of CAR T cells in solid tumors include antigen heterogeneity (3), limited CAR T cell persistence, and immunoregulation induced by the tumor microenvironment (TME) (2, 4), in which intercellular messengers such as tumor-derived extracellular vesicles (TEVs) contribute to the complex crosstalk between tumor and immune cells. TEVs are nanosized lipid-bilayer vesicles secreted by cancer cells that carry bioactive molecules, including immunoregulatory proteins such as checkpoint inhibitors (5, 6), major histocompatibility complex (MHC) molecules, and tumor associated antigens (7), with the ability to diversely modulate interacting cells depending on their cargo composition (8). For example, EVs may exhibit immunogenic properties such as transferring antigen to dendritic cells (DCs) inducing cytotoxic T lymphocyte activation (9). Conversely, EVs carrying inhibitory checkpoint molecules, such as Fas ligand (10), programmed death-ligand 1 (PD-L1) (6), or immunosuppressive cytokines can potently inhibit immune cells (11). Moreover, the EV surfaceome can influence the clinical response to immunotherapeutics, including monoclonal antibodies (12), antibody-drug conjugates (13), and CAR T cells (14). For example, EVs that harbor therapeutic antibody target molecules can act as a decoy facilitating tumor cell evasion, such as what has been observed with PD-L1+ EVs sequestering anti-PD-L1 antibodies in melanoma and lung cancer (5, 6). In the context of CAR T cell therapy, studies in hematological malignancies suggest that CD19 CAR T cell dysfunction can be induced by tumor derived EVs due to concomitant presence of the target antigen CD19 and PD-L1 (7). However, how immunotherapeutic target antigens on EVs modulate CAR T cell functionality in solid tumors, especially those with limited PD-L1 expression like neuroblastoma (15), is not well known. Further, whether manipulation of EV cargo can confer therapeutically beneficial immunostimulatory properties and serve as CAR T cell adjuvants also remains unexplored.

High risk neuroblastoma is an unrelenting childhood cancer with long-term survival below 50% despite intensive multimodal chemoradiotherapy (16), with a critical need for innovative treatments. Our lab has identified glypican-2 (GPC2) as a differentially expressed immunotherapeutic target in high-risk neuroblastoma and other lethal cancers with no substantial expression on vital normal tissues (17). We and others have isolated high-affinity GPC2 binders and developed effective immunotherapies targeting this molecule, including potent and safe 2^nd generation GPC2 CAR T cells (18–22), one of which is currently being tested in a first-in-human phase 1 clinical trial (NCT05650749). Several clinical trials are also evaluating the safety and efficacy of GD2 CAR T cells in children with neuroblastoma and other GD2-positive tumors (23–25), work that recently provided the first convincing evidence that CAR T cells can be effective for this pediatric embryonal tumor (23). However, similar to other potent CARs and solid tumor histologies, downregulation of the targeted antigen under CAR pressure (3, 21), limited CAR T cell persistence, and a hostile immunosuppressive TME (26, 27) drive CAR T cell resistance. Collectively these data suggest that CAR vaccination strategies that deliver more accessible CAR ligand, for example by providing antigen+ EVs outside of the immunosuppressive TME, may be an effective strategy to reinvigorate CAR T cells in the setting of limited T cell persistence or antigen downregulation to ultimately achieve more durable clinical responses.

In this study, we deepen our understanding of the neuroblastoma EV surfaceome and the interaction between TEVs and CAR T cell therapies in this disease, along with providing proof of concept for utilizing antigen-displaying synthetic EVs (SyntEVs) as a therapeutic platform to enhance clinically available solid tumor CAR T cell therapies.

RESULTS

GPC2 and other immunotherapeutic targets are selectively exposed on neuroblastoma-derived extracellular vesicles.

First, to define the protein cargo of neuroblastoma extracellular vesicles (EVs), we isolated EVs from seven human neuroblastoma and two non-tumor control cell lines using ultracentrifugation (28). Transmission electron microscopy (TEM) of isolated EVs confirmed the presence of vesicle-shaped particles in the range of 50–300 nm in diameter (Fig. 1A), which was verified by Nanoparticle Tracking Analysis (NTA; Fig. 1B). EV secretion kinetics were evaluated across all cell lines, showing that EV abundance was significantly higher in neuroblastoma cells compared to non-tumor control cells (p<0.05; fig. S1A). Neuroblastoma EV-associated proteins were then characterized by immunoblotting, confirming expression of small-EV markers CD81 and Alix and absence of calnexin in all EV isolations [fig. S1B; (29)]. Neuroblastoma-associated antigens, including neural cell adhesion molecule 1 (NCAM1), L1 cell adhesion molecule (L1CAM), GPC2, and B7 homolog 3 (B7-H3), were exclusively present in EVs isolated from neuroblastoma cell lines, but not in EVs released by a non-tumor control cell line (Fig. 1C). Further, neuroblastoma-specific cell surface molecules such as GD2, GPC2, and NCAM1 were identified by flow cytometry exclusively on the surface of neuroblastoma EVs (Fig. 1, D and E), whereas CD81 abundance was comparable across all EV isolates. Neuroblastoma-derived EVs contained minimal amounts of PD-L1, with only EVs from the SK-N-AS cell line having any appreciable PD-L1 exposure (Fig. 1C and fig. S1C). However, SK-N-AS EV PD-L1 abundance was modest compared to those reported on TEVs from other cancers (5, 30) even after 48 hours of co-incubation with IFN-γ, conditions that are known to dramatically upregulate both parent cell and EV PD-L1 (5), as we observed in control melanoma WM164 cells (Fig. 1C and fig. S1C). Finally, to more globally define the neuroblastoma EV protein cargo composition, we performed mass spectrometry (MS) on EVs isolated from two distinct neuroblastoma cell lines, the adrenergic NB-EbC1 cell line and more mesenchymal SK-N-AS cells (31). Neuroblastoma cell heterogeneity and plasticity are best exemplified by the recent characterization of both adrenergic and mesenchymal cell states in neuroblastoma preclinical models and tumors (32, 33), which are epigenetically distinct but interchangeable and can exhibit substantially different surfaceomes (34–36). MS identified a total of 4,543 proteins, with 56.8% shared between NB-EbC1 and SK-N-AS EVs (Fig. 1F). High abundance of neuroblastoma cell surface molecules was detected in neuroblastoma-derived EVs, including proteins present in EVs from both cell lines, such as B7-H3, GPC2, cell adhesion molecule 1 (CADM1), vimentin (VIM), and GDNF family receptor alpha 2 (GFRA2) (Fig. 1G, top). These MS studies also confirmed low abundance of inhibitory checkpoint ligands on neuroblastoma EVs, such as PD-L1 (CD274) and poliovirus receptor (PVR; Fig. 1G, middle), with many others being undetectable, and the expected presence of sEV markers (CD63, CD81, and CD9; Fig. 1G, bottom). Finally, other proteins displayed differential EV abundance, largely reflecting the cell-state-specific surfaceome (31, 36, 37). Specifically, proteins associated with an adrenergic phenotype, such as solute carrier family 6 member 2 (SLC6A2), delta like non-canonical notch ligand 1 (DLK1), anaplastic lymphoma kinase (ALK), and L1CAM, were enriched in NB-EbC1 TEVs (Fig. 1H, top). Proteins linked to a more mesenchymal phenotype, such as CD44, platelet-derived growth factor receptor alpha (PDGFRA), EPH receptor A2 (EPHA2), Erb-b2 receptor tyrosine kinase 2 (ERBB2), and AXL receptor tyrosine kinase (AXL), were alternatively found in higher abundance in SK-N-AS TEVs (Fig. 1H, bottom). Taken together, neuroblastoma EVs are enriched in immunotherapeutic target molecules such as GPC2 and GD2, but display low PD-L1 and other inhibitory checkpoint ligands, which contrasts with TEVs from other tumor histologies where PD-L1 abundance is more predominant (5, 14).

Fig. 1. Neuroblastoma cells release EVs carrying GPC2, GD2, and other tumor-associated antigens.

(A, B) Representative characterization of EVs isolated from SK-N-AS cells by transmission electron microscopy (TEM) (A) and nanoparticle-tracking analysis (NTA; B). Scale bar in A, 200 nm.

(C) Western blot of EVs isolated from neuroblastoma, non-tumor RPE1, and melanoma WM164 cell lines. Neuroblastoma associated antigens (NCAM1, L1-CAM, GPC2, B7-H3), small EV (sEV) markers (Alix, CD81, CD9), and PD-L1 are shown. Equal amounts of total protein were loaded for each sample.

(D) Surface expression of neuroblastoma- and sEV-associated molecules on bead-bound EVs isolated from seven neuroblastoma (NB) and two non-tumor cell lines by flow cytometry. Filled colored plots represent staining with the antibody indicated on the x-axis and empty plots represent unstained samples.

(E) Heat map showing the surface exposure of tumor antigens on neuroblastoma-derived EVs as fold change over non-tumor RPE1 EVs.

(F) Venn diagram showing proteins identified in SK-N-AS and NB-EbC1 EVs by mass spectrometry (MS; n=3 biological replicates). Common and exclusive proteins identified from the SK-N-AS and NB-EbC1 EVs are shown as overlapping and non-overlapping areas, respectively, with the numbers of total proteins (and percentages) indicated.

(G) Heat map showing the MS normalized protein intensity for proteins of interest from SK-N-AS and Nb-EbC1 EVs (n=3 biological replicates). Neuroblastoma cell surface molecules (top), inhibitory checkpoint ligands (middle), and sEV markers (bottom) are shown.

(H) Heat map showing the MS normalized protein intensity for adrenergic (top) and mesenchymal (bottom) proteins of interest from SK-N-AS and Nb-EbC1 EVs.

Circulating GPC2+ and GD2+ EVs are present in neuroblastoma-bearing mice and patients.

Tumor-derived EVs can be detected in the peripheral blood of patients with cancer as circulating EVs (circEVs), several of which directly modulate immunotherapeutic efficacy (12, 30) or have demonstrated diagnostic utility (38, 39). Considering the recent focus of targeting GPC2 and GD2 with CAR T cells and other immunotherapies (18–23), we first investigated whether neuroblastoma-derived GPC2+ or GD2+ circEVs can be detected in the peripheral blood of neuroblastoma-bearing mice or humans. We established six human neuroblastoma patient-derived xenografts (PDXs) in severe combined immunodeficient (SCID) mice. When PDX tumors reached a volume of 2 cm³, EVs were isolated from the peripheral blood and assessed for EV particle count by NTA and for GPC2 and GD2 abundance with flow cytometry (fig. S1D). Peripheral blood from mice with human neuroblastoma PDXs contained a higher number of circEVs compared to non-tumor bearing mice, but with no difference in particle size between the groups (fig. S1, E and F). Moreover, GPC2 and GD2 were selectively detected on circEVs from mice harboring neuroblastoma PDXs (Fig. 2A, B) and the abundance of GPC2+ EVs isolated from each mouse correlated with the expression of GPC2 in the parent tumor (R²=0.3, p=0.005 Fig. 2A and fig. S1, G and H). Although GD2+ EVs were also selectively isolated from tumor-bearing mice, the amount of GD2+ EVs did not correlate as well with parent tumor GD2 (R²=0.13, p=0.072; Fig. 2B and fig. S1, G and H). No PD-L1 was detected on PDX-circEVs (fig. S1I). Finally, isolation of EVs from the peripheral blood of PDX-bearing mice with various tumor sizes revealed that the abundance of both GPC2+ and GD2+ circEVs significantly correlated with overall tumour burden (Fig. 2, C and D; GPC2, R²=0.92 for COG-N-561x; R²=0.82 for COG-N-452x; GD2, R²=0.86 for COG-N-561x, and R²=0.70 for COG-N-452x; all p<0.0001).

Fig. 2. Circulating GPC2+/GD2+ EVs are detectable in the peripheral blood of tumor-bearing mice and patients with neuroblastoma.

(A) Association between the percentages of circulating GPC2+ bead conjugated-EVs (magenta-colored dots; left y-axis) and tumor GPC2 expression (black squares; right y-axis) from neuroblastoma PDX-bearing mice.

(B) Association between the percentages of circulating GD2+ bead conjugated-EVs (purple-colored dots; left y-axis) and tumor GD2 expression (black squares; right y-axis) from neuroblastoma PDX-bearing mice.

(C, D) Correlation between the percentages of circulating GPC2+ (C) and GD2+ (D) bead conjugated-EVs and tumor size (COG-N-561x, n=12; COG-N-452x, n=13).

(E) Representative TEM image of neuroblastoma patient-derived circEVs. Scale bar, 200 nm.

(F) left: Comparison of the percentages of CD81+ bead conjugated-EVs between patients with neuroblastoma (red, n=15) and healthy children (blue, n=12). right: Representative CD81 flow cytometry histograms.

(G) left: Comparison of the amounts of circEVs between patients with neuroblastoma (red, n=15) and healthy children (blue, n=12). right: Representative particle size distribution by NTA between circEVs from a patient with neuroblastoma (red) and a healthy child (blue).

(H) left: Comparison of the percentages of GPC2+ bead conjugated-EVs between patients with neuroblastoma (red, n=15) and healthy children (blue, n=12). right: Representative GPC2 flow cytometry histograms.

(I) left: Comparison of the percentages of GD2+ bead conjugated-EVs between patients with neuroblastoma (red, n=15) and healthy children (blue, n=12). right: Representative GD2 flow cytometry histograms.

(J) Comparison of the percentages of PD-L1+ bead conjugated-EVs between patients with neuroblastoma (red, n=15) and healthy children (blue, n=12).

For each flow cytometry plot, filled colored plots represent staining with the antibody indicated on the x-axis and empty plots represent unstained samples.

Data are shown as mean ± SEM. Dots represent individual mice (A-D) or human patients (F-J). Statistical analysis was performed using simple linear regression to assess correlation between GPC2/GD2 expression on circEVs and tumor samples (A-D) or Mann-Whitney test to evaluate difference between two groups (F-J). **, p<0.01; ****, p<0.0001; ns, not significant.

To validate these data, we next determined if GPC2+ or GD2+ EVs could be selectively isolated from the peripheral blood of children with neuroblastoma. We isolated EVs from archived peripheral blood samples from 15 patients with neuroblastoma and 12 age-matched healthy controls. TEM analysis of purified human circEVs revealed the presence of vesicle-shaped particles consistent with EVs (Fig. 2E), and flow cytometry showed comparable CD81 positivity between the two cohorts (Fig. 2F). However, the relative concentration of circEVs by NTA was significantly higher in the peripheral blood of children with neuroblastoma compared to healthy donors (Fig. 2G), similar to what we observed in our PDX models and validating what has been observed in other oncology patient cohorts (38, 40). No difference in particle size between the two cohorts was observed (fig. S1J). However, EVs isolated from the peripheral blood of children with neuroblastoma again selectively showed high abundance of GPC2 and GD2 (P<0.0001; Fig. 2, H and I), whereas they did not have substantial PD-L1 (Fig. 2J and fig. S1K). Taken together, these studies show that GPC2 and GD2 are selectively found on circEVs from mice and children with neuroblastoma. In mice with neuroblastoma the abundance of GPC2+ and GD2+ circEVs correlates with both disease burden and at least partially with parent tumor cell GPC2/GD2 expression. Together, these studies suggest a potential role for GPC2+ and GD2+ circEVs as non-invasive diagnostic biomarkers in children with neuroblastoma.

Target antigen+ EVs bind and activate CAR T cells in an antigen dependent manner.

Our findings of clinically relevant tumor-associated immunotherapeutic targets (23) on neuroblastoma-derived EVs motivated us to investigate how these target+ EVs might modulate the functionality of paired CAR T cells. We first focused on GPC2 given the recent development of several safe and effective CAR T cells (18, 20, 21), one of which is being tested in an early phase trial (NCT05650749). To enable these studies, we first generated SK-N-AS-GPC2 and HEK293T-GPC2 isogenic cell lines (fig. S2A) as a source of GPC2+ EVs with different densities of target antigen (ultrahigh/high/low/negative). The expected variable EV surface GPC2 was validated by flow cytometry, which also revealed lack of the inhibitory checkpoint molecule PD-L1, whose presence on the EV surface has been previously linked with CAR T cell inhibitory effects (14) (Fig. 3A). Using human primary T cells, we generated GPC2-targeting CAR T cells with either CD28 or 4–1BB costimulatory domains (GPC2.28z and GPC2.BBz, respectively) and CD19 CAR T cells as a non-targeting control as described previously (fig. S2, B and C) (20).

Fig. 3. GPC2+ EVs bind and activate GPC2 CAR T cells in an antigen-dependent manner enhancing T cell activation and in vitro cytotoxicity.

(A) Representative flow cytometry histograms illustrating the abundance of GPC2, CD81 (sEV marker), and PD-L1 on EVs isolated from GPC2 isogenic cell lines. Filled colored plots represent staining with the antibody indicated on the x-axis and empty plots represent unstained samples.

(B and C) Representative immunofluorescent images showing EV - CAR T cell binding after incubation of PKH67-labeled GPC2^UltraHigh EVs (Green) with GPC2 CAR T cells (CD3 staining in red and DAPI in blue; left) but not CD19 CAR T cells (right). GPC2^Neg EVs showed no binding to CAR T cells (center). CD3 and DAPI staining used to label T cells. Scale bars represent 10 μm. (C) Flow cytometric gating strategy to detect EV - CAR T cell binding after incubation of PKH67-labeled GPC2+ EVs with GPC2 CAR T cells at a 10,000:1 EV:CAR+ T cell ratio. Bar chart shows the percentage of PKH67 positive GPC2 CAR+ T cells.

(D) The percentages of PKH67-labeled GPC2^{Neg/Low/High/UltraHigh} EV-bound GPC2 CAR T cells at different EV:CAR+ T cell ratios.

(E) CAR T cell activation evaluated by CD69 (top) or granzyme B (GrzB; bottom) positivity by flow cytometry after a 16-hour incubation with GPC2^{Neg/Low/High/UltraHigh} EVs.

(F, G) Concentrations of IFN-γ (F) and IL-2 (G) by ELISA after a 24-hour co-incubation of CAR T cells with GPC2+ EVs.

(H) EV-primed CAR T cell killing of luciferase-neuroblastoma cell lines at an E:T ratio of 1:1 after a 24-hour co-incubation.

Data are shown as mean ± SEM. Dots represent independent biological replicates. Statistical analysis was performed using a one-way ANOVA with Dunnett’s multiple comparisons test. *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001. E:T, effector to target cell ratio.

First, we studied the interaction of GPC2^{UltraHigh/High/Low/Neg} EVs with GPC2.28z CAR T cells by performing co-incubation studies using EVs fluorescently labeled with PKH67, followed by flow cytometry binding assays and fluorescence microscopy (Fig. 3, B and C). For each EV sample, NTA was first performed, and the same number of particles was used in T cell co-incubation assays across the different EV samples. Here, we observed that GPC2+ EVs selectively bind to the CAR+ subset of T cells and that the percentage of EV-bound CAR T cells varied in an EV antigen density-dependent manner, whereas no differences in EV binding were observed for the CAR-T cell subset or control CD19 (non-targeting) CAR T cells (fig. S2, D and E). We additionally noted a concentration-dependent effect of EVs on the interaction with GPC2 CAR T cells across all EV subsets (Fig. 3D).

We next evaluated the functional effects of GPC2+ EV binding to GPC2 CAR T cells in 24-hour co-incubation assays. In these studies, GPC2+ EVs activated GPC2.28z CAR T cells in an antigen dependent manner, as shown by increased expression of CD69 and Granzyme B on the CAR+ T cell subset (Fig. 3E) along with increased IL-2 and interferon-γ (IFN-y) secretion by GPC2+ EV-stimulated CAR T cells (Fig. 3, F and G). Comparable magnitudes of GPC2+ EV-induced GPC2 CAR T cell activation were observed with GPC2.BBz CAR T cells, suggesting that the GPC2+ EV/CAR T cell intersection is independent of the CAR costimulatory domain (fig. S2, F and G).

Next, the target cell cytotoxicity of EV-primed GPC2 CAR T cells was tested using 3 luciferase-tagged neuroblastoma cell lines with different amounts of cell surface GPC2. When co-incubated for 24 hours, GPC2+ EV-primed GPC2 CAR T cells were substantially more cytotoxic to target neuroblastoma cells compared to unstimulated CAR T cells or CAR T cells pre-incubated with GPC2^Neg EVs (Fig. 3H and fig. S2H). This enhanced CAR T cell cytotoxicity was again independent of CAR costimulatory domain (fig. S2I). Finally, as further proof that EV-induced GPC2 CAR T cell activation is not due to nonspecific EV uptake but rather results from EV antigen binding to the CAR on the cell surface, endocytosis inhibition using Dynasore (41) did not affect either EV binding measured by PKH67 staining or the CAR T cell activation induced by GPC2+ EVs (fig. S3, A to C).

To determine whether the stimulatory effects of EV antigen on CAR T cells are specific to GPC2 or represent a general characteristic of other CAR target molecules in neuroblastoma, we investigated the impact of GD2+ EVs on paired GD2 CAR T cells. We generated GD2-targeting CAR T cells (Fig. 4A) and validated their specific cytotoxicity towards GD2+ neuroblastoma cell lines (fig. S4, A and B). CAR T cell binding assays after GD2 CAR exposure to GD2^{UltraHigh/High/Low/Neg} EVs (Fig. 4B) isolated from GD2 isogenic cell lines (fig. S4C) showed, similarly to GPC2+ EVs, that GD2+ EVs bind GD2 CAR T cells proportionally to GD2 EV surface amount (Fig. 4C). Further, GD2+ EVs robustly activated GD2 CAR T cells as shown by the increased expression of CD69 and granzyme B after exposure to GD2+ EVs but not to control GD2^Neg EVs (Fig. 4D). The GD2+ EV: GD2 CAR T cell synapse also resulted in enhanced GD2 CAR T cell killing of the GD2^Low neuroblastoma cell line SK-N-AS, especially at lower E:T ratios (Fig. 4E), similar to what we observed for GPC2+ EV pre-incubated GPC2 CAR T cells. This interaction was again specific as no increased cytotoxicity was observed for CD19 CAR T cells, confirming the requirement for the specific interaction between EV-displayed antigen and the paired CAR (fig. S4D).

Figure 4. GD2+ EVs bind and activate GD2 targeting CAR T cells in an antigen dependent manner enhancing T cell activation and in vitro cytotoxicity.

(A) left: Schematic illustration of CAR design, created with BioRender.com. right: Representative transduction efficiency of GD2 and CD19 (non-targeting control) CAR T cells by flow cytometry.

(B) Representative flow cytometry histograms illustrating the abundance of GD2 and CD81 (sEV marker) on EVs isolated from GD2 isogenic cell lines. Filled colored plots represent staining with the antibody indicated on the x-axis and empty plots represent unstained samples.

(C) Flow cytometric gating strategy to detect EV - CAR T cell binding after incubation of PKH67-labeled GD2+ EVs with GD2 CAR T cells at a 10,000:1 EV:CAR+ T cell ratio. Bar chart shows the percentage of PKH67 positive GD2 CAR+ T cells.

(D) Flow cytometric gating strategy to detect CD69 (top) and granzyme B (GrzB; bottom) positivity on GD2 CAR T cells after a 16-hour co-incubation with GD2+ EVs. The bar charts show percentage of CD69 or GrzB positive GD2 CAR+ T cells.

(E) EV-primed CAR T cell lysis of luciferase-SK-N-AS cells at different E:T ratios after a 24-hour co-incubation.

Data are shown as mean ± SEM. Dots represent independent biological replicates. Statistical analysis was performed using a one-way ANOVA with Dunnett’s multiple comparisons test. *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001.

Tumor-derived antigen+ EVs stimulate tumor-infiltrating CAR T cells and are necessary for in vivo therapeutic efficacy.

Next, using GPC2 as a CAR target we explored whether target antigens on TEVs also directly modulate CAR T cells in vivo. We first generated xenograft models with decreased endogenous EV secretion by CRISPR/Cas9 knockout (KO) of RAB27A, an essential GTPase involved in EV release (42). RAB27A KO was efficiently achieved in neuroblastoma cell lines with two unique guide RNAs (Fig. 5A and fig. S5A). As expected, loss of RAB27A markedly suppressed TEV secretion both in vitro (Fig. 5B and fig. S5B) and in vivo after subcutaneous injection into NSG mice (Fig. 5C), without modulating the cell surface expression of GPC2 (Fig. S5C). However, despite only modest differences in cell proliferation in vitro compared to wild-type (WT) cells (Fig. S5D), RAB27A KO cells grew substantially slower in vivo compared to WT xenografts (Fig. S5E), preventing us from directly comparing CAR T cell response between the RAB27A KO and WT models and suggesting a potential role for EVs in neuroblastoma in vivo tumor growth. Alternatively, we used the SK-N-AS RAB27A KO xenograft - a model with depleted TEV secretion and low GPC2 expression - to compare GPC2 CAR T cell efficacy with and without exogenous intratumor administration of GPC2^High TEVs, isolated from SK-N-AS-GPC2 isogenic cells. (Fig. 5D). Here, we observed that injection of GPC2^High TEVs significantly enhanced the efficacy of co-infused GPC2 CAR T cells against GPC2^low SK-N-AS xenografts, as demonstrated by the difference in tumor growth (Fig. 5, E and F, and fig. S5F) and the modestly prolonged survival of tumor-bearing mice in the GPC2^High TEV cohort (Fig. 5G). GPC2^Low TEVs isolated from paired-WT SK-N-AS cells had minimal effect on CAR T efficacy, further confirming that the superior CAR T efficacy was specifically induced by the presence of GPC2 on co-infused TEVs (fig. S5G). Further, we analyzed satellite tumors 7 days after GPC2 CAR T cell administration to quantify CAR T cell infiltration, activation, and phenotype. Consistent with the observed enhanced anti-tumor response, the exogenous administration of GPC2^High TEVs resulted in significantly enhanced tumor infiltration of both total and activated (CD69+ CD25+) GPC2 CAR T cells (Fig. 5, H and I) with an increased effector memory (EM) phenotype (fig. S5H), but with no differences in expression of T cell exhaustion markers (PD-1, LAG3) compared to GPC2 CAR T cells alone (fig. S5I; gating strategy in fig. S6A).

Figure 5. Tumor-derived GPC2+ EVs stimulate tumor-infiltrating CAR T cells and are necessary for in vivo anti-tumor efficacy.

(A) RAB27A western blot of non-targeting control (NTC) and three RAB27A knock-out (KO) SK-N-AS clones. β-actin served as a loading control.

(B) EV counts from the supernatant of control NTC and three RAB27A KO SK-N-AS clones quantified by NTA.

(C) In vivo TEV secretion by RAB27A KO compared to NTC SK-N-AS xenografts by NTA. TEVs were isolated from tumors collected on day 20 after engraftment in SCID mice.

(D) Experimental design of in vivo study using the SK-N-AS RAB27A KO xenograft (GPC2^Low) to compare GPC2 CAR T cell efficacy with and without exogenous intratumor administration of GPC2^High TEVs from SK-N-AS-GPC2 cells.

(E) SK-N-AS RAB27A KO tumor growth curves measured serially with calipers after treatment with non-targeting control CD19 or GPC2 CAR T cells co-administered with I.T. injected GPC2^High TEVs or vehicle (n = 5–7 mice per group).

(F) Tumor volumes at day 11 after CAR T cell administration for treatment groups in E.

(G) Progression-free survival (PFS) for the different CAR treatment cohorts shown in E and F.

(H) Quantification of tumor-infiltrating human CD45+ (CAR T) cells by flow cytometry from satellite tumors treated according to D (n=3 tumors per group, day 10).

(I) Quantification of activated (CD69+ CD25+) CAR T cells gated on human CD45+ cells in H.

(J) Number of circEVs from SCID mice bearing SK-N-AS-GPC2 tumors treated daily with GW4869 (1.25 mg/kg I.P.) or vehicle. Treatment started when tumors reached a volume of approximately 0.2 cm³ and circEVs were collected from the peripheral blood after 10 days.

(K) In vivo study design testing the efficacy of a sub-therapeutic dose of GPC2 CAR T cells (1×10⁶ CAR+ cells) on SK-N-AS-GPC2 xenograft (GPC2^High) growth when given concurrently with GW4869 with or without I.T. rescue with GPC2^High TEVs from SK-N-AS-GPC2 cells.

(L) SK-N-AS-GPC2 tumor growth curves after treatment with non-targeting control CD19 or GPC2 CAR T cells co-administered with GW4869 or vehicle with or without I.T. injected GPC2^High TEVs (n = 5–6 mice per group).

(M) Tumor volumes at day 20 after CAR T cell administration for treatment groups in L.

(N) PFS for CAR treatment cohorts shown in L and M.

(O) Quantification of tumor-infiltrating human CD45+ (CAR T) cells by flow cytometry from satellite tumors (n=3 tumors per group) treated according to K. Tumor samples were collect at day 10 after CAR T cell injection.

(P) Quantification of activated (CD69+ CD25+) CAR T cells gated on human CD45+ cells in O.

Data are shown as mean ± SEM. Dots represent independent biological replicates. Statistical analysis was performed using Log-rank test for PFS curves or a one-way ANOVA with Tukey’s multiple comparisons test (B, F, H, I, M, O, and P) or Mann–Whitney test (C and J) to evaluate differences between groups. *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001. Illustrations in D and K created with BioRender.com. KO, knock-out; NTC, non-targeting control; I.T., intratumoral; I.P., intraperitoneal.

To further validate these findings in an orthogonal EV-depleted in vivo model, we next evaluated GPC2 CAR T cell efficacy in a GPC2^High neuroblastoma xenograft (isogenic SK-N-AS-GPC2 tumors), achieving exogenous EV inhibition with the neutral sphingomyelinase inhibitor GW4869 (43). We first demonstrated that daily treatment with GW4869 (1.25 mg/kg) induced a stable decrease in EV secretion (Fig. 5J), without modulating in vivo tumor growth kinetics or GPC2 cell surface density (fig. S6, B and C). Here, we treated isogenic SK-N-AS-GPC2 xenografts with a sub-therapeutic dose of GPC2 CAR T cells (1 million GPC2 CAR+ T cells; 10% of typical dose) concurrently with the EV-inhibiting drug GW4869 (Fig. 5K). Pharmacologic inhibition of GPC2+ EV tumor secretion significantly decreased CAR T cell efficacy which could be rescued by intratumoral injection of GPC2^High TEVs (Fig. 5, L to N). Analysis of satellite tumors 7 days after GPC2 CAR T cell infusion validated a direct effect of TEVs on tumor infiltrating CAR T cells as we also observed an increase in both total and activated (CD69+ CD25+) GPC2 CAR T cells with a more prevalent EM phenotype in tumors that had GPC2+ EVs present (either those that received vehicle or where EV inhibition was rescued by IT administration of GPC2^High TEVs; Fig. 5, O and P, and fig. S6D). Collectively, these studies demonstrate that the in vivo availability of antigen exposing TEVs facilitate CAR T cell tumor homing, activation, and anti-tumor efficacy.

Engineering target antigen displaying-SyntEVs with albumin or GD2 binding properties further enhances GPC2 CAR T cell in vivo persistence and anti-tumor efficacy.

We next looked to further capitalize on the importance of GPC2+ EVs to GPC2 CAR T cell activation and enhanced anti-tumor efficacy by testing whether intravenously administered GPC2+ EVs might serve to serially stimulate GPC2 CAR T cells either peripherally or in the TME utilizing different engineering strategies. For this therapeutic purpose, we chose genetically modified HEK293T cells as producer cells for non-tumor GPC2+ EV isolation (Fig. 6A), herein referred to as SyntEVs to distinguish them from previously described GPC2+ TEVs. We first used mass spectrometry to characterize these SyntEVs and confirmed high abundance of GPC2, low amounts of some inhibitory checkpoint ligands (PD-L1, PVR). Other immunomodulatory proteins were undetectable such as the programmed cell death ligand 2 (PD-L2), fibrinogen like 1 (FGL1), herpesvirus entry mediator (HVEM), v-set and immunoglobulin domain containing protein 3 (VSIG3), and galectin 9 proteins (fig. S7, A and B). To better understand the optimal localization for GPC2+ EVs to reside in order to most effectively prime GPC2 CAR T cells in vivo, we engineered our non-tumor GPC2+ SyntEVs to either bind serum albumin (ALB) to enhance their circulation half-life or the neuroblastoma-associated antigen GD2 to enhance their tumor infiltration (Fig. 6A). First, to achieve EV binding to albumin, we exploited two well characterized engineered albumin binding domains (ABDs) with affinity to both human and murine albumin (ABD035, ABD094) (44, 45), which were inserted into the extracellular loops of the tetraspanin CD9 (46). CD9 was either engineered with ABDs in both loops (×2), or a single ABD in the 2^nd loop (L2), for a total of four unique anti-ALB-CD9 combinations (Fig. S7C). Alternatively, GD2 binding was achieved by expression of a recombinant receptor including the anti-GD2 scFv, 14G2a, followed by PDGFR transmembrane and intracellular domains (Fig. 6A). A CD81-GFP fusion protein was also expressed in the EV producer HEK293T cells to achieve production of GFP+ EVs (Fig. 6A). The expression of the transgenes in the producer cells, as well as the binding properties of the corresponding isolated EVs, was verified by flow cytometry (Fig. S7, D and E). Among the four different iterations of anti-ALB-CD9, we selected the one with the highest binding to albumin as quantified by flow cytometry (CD9-ABD094-L2; Fig. S7D). The affinity of the isolated GPC2+ anti-ALB SyntEVs for both murine and human albumin, as well as the affinity of the GPC2+ anti-GD2 EVs for the 1A7 anti-idiotype antibody specific for the 14G2a binder, was also validated (Fig. 6B, top). Comparable amounts of GPC2 exposure and GFP positivity were observed among the different engineered EVs (Fig. 6B, bottom).

Figure 6. Engineering GPC2+ SyntEVs with albumin or GD2 binding properties as CAR T cell engagers.

(A) Schematic illustration showing the design of GPC2+ SyntEVs (GPC2+, GPC2+ anti-ALB, and GPC2+ anti-GD2) derived from HEK293T cells stably engineered with lentiviral transduction. Figure created with BioRender.com.

(B) top: Representative flow cytometry histograms showing GPC2+ anti-ALB SyntEV binding to murine (m) and human (h) serum albumin (purple) and GPC2+ anti-GD2 EV binding to the 1A7 anti-idiotype antibody specific for 14G2a (cyan). bottom: Flow cytometry histograms showing GPC2 cell surface expression and GFP signal in GPC2+ ± anti-ALB/GD2 SyntEVs and GPC2− GFP+ control EVs. Stained beads without EVs are shown as grey filled plots, whereas empty plots represent unstained beads.

(C) Representative flow cytometry plots for EV-CAR T cell binding after incubation of GFP+ GPC2− (control) or GPC2+ ± anti-ALB/GD2 SyntEVs with GPC2 CAR T cells at a 10,000:1 EV:CAR+ T cell ratio.

(D) Quantification of EV-bound GPC2 CAR T cells from C.

(E and F) CAR T cell activation evaluated by CD69 (E) or granzyme B (GrzB; F) by flow cytometry after a 24-hour co-incubation with GPC2+ ±anti-ALB/GD2 SyntEVs or vehicle.

(G) Representative flow cytometry plots for EV-neuroblastoma cell binding after incubation of SK-N-AS-GD2 (top) or SK-N-AS (bottom) cells with GFP+ GPC2+ ±anti-ALB/GD2 SyntEVs at a 10,000:1 EV:tumor cell ratio.

(H and I) Quantification of EV-bound SK-N-AS-GD2 (H) or SK-N-AS (I) cells from G.

(J) GPC2 abundance on SK-N-AS-GD2 or SK-N-AS cells from G-I.

(K) GPC2 CAR T cell lysis of luciferase-SK-N-AS-GD2 or luciferase-SK-N-AS cells after pre-incubation with GPC2+ ±anti-ALB/GD2 SyntEVs or vehicle for 1 hour.

Data are shown as mean ± SEM. Dots represent independent biological replicates. Statistical analysis was performed using a one-way ANOVA with Dunnett’s multiple comparisons test (D, E, and F) or a one-way ANOVA with Tukey’s multiple comparisons test (H, I, and K). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001. ABD, albumin-binding domain; Mod, moderate.

Next, we performed EV-GPC2 CAR T cell co-incubation studies to verify binding to GPC2 CAR T cells by flow cytometry leveraging the intrinsic fluorescence of the EVs provided by GFP. As expected, based on the similar GPC2 densities, all the GPC2+ engineered SyntEVs (GPC2+, GPC2+ anti-ALB, and GPC2+ anti-GD2) exhibited comparable GPC2 CAR T cell binding (63.4% ± 7.1), whereas only minimal binding was observed with the GPC2^Neg GFP+ control EVs, suggesting that the interaction is driven by direct CAR engagement with EV-displayed-GPC2 (Fig. 6, C and D). Accordingly, comparable GPC2 CAR T cell activation was induced by GPC2+ engineered SyntEVs, as shown by CD69 and granzyme B expression in the CAR+ T cell subset (Fig. 6, E and F). Conversely, no activation was observed in GPC2 CAR T cells after incubation with the GPC2^Neg GFP+ EVs (Fig. 6, E and F) or in CD19 CAR T cells following exposure to GPC2+ SyntEVs (fig. S7F). In order to further define the activation profile of GPC2 CAR T cells upon interaction with the engineered GPC2+ SyntEVs, we performed single-cell functional cytokine immune profiling. This analysis revealed increased expression of not only effector cytokines, such as IFN-γ, but also elevation of stimulatory (IL-2, GM-CSF) and chemoattractive (MIP-1β) cytokines compared to GPC2 CAR T cells exposed to GPC2^Neg EVs (fig. S7, G and H).

We also tested the ability of our GPC2+ GD2-binding SyntEVs to specifically bind GD2+ neuroblastoma cells and potentially deliver extra GPC2 CAR antigen to tumor cells. Using flow assays, we compared the percentages of EV-bound neuroblastoma cells following co-incubation with different GPC2+ SyntEVs. Binding, detected as GFP positivity, was observed exclusively in GD2+ cell lines exposed to GD2-binding SyntEVs and correlated with the GD2 abundance on the target cell surface (SK-N-AS-GD2 > SK-N-AS >>> HEK293T; Fig. 6, G to I and fig. S8, A to C). Furthermore, GD2-mediated EV-tumor cell coating increased GPC2 cell surface expression in SK-N-AS and SK-N-AS-GD2 (Fig. 6J), but not in control HEK293T (fig. S8D) cells. Finally, this higher GPC2 cell surface density induced by EV-binding also resulted in selectively and potently increasing susceptibility to GPC2 CAR T cell-mediated cytotoxicity (Fig. 6K and fig. S8E).

Next, to determine whether the GPC2+ ALB-binding SyntEVs can be better detected in the peripheral blood of mice and to validate tumor trafficking of the GPC2+ anti-GD2 SyntEVs, we intravenously injected NSG mice bearing SK-N-AS-GD2 xenografts with two doses of each type of engineered SyntEVs (GPC2+, GPC2+ anti-ALB, or GPC2+ anti-GD2; on Day 1 and 3) and collected peripheral blood and tumor 5 hours after the second EV administration. As predicted, GFP signal evaluated by western blot of peripheral blood-isolated circEVs and tumor lysates showed enhanced detectability of GPC2+ anti-ALB SyntEVs in the peripheral blood and more robust tumor trafficking for the GPC2+ anti-GD2 SyntEVs (Fig. 7A). Next, to understand if these engineered GPC2+ SyntEV variants may better enhance GPC2 CAR T cell efficacy in vivo, we serially intravenously infused them in combination with GPC2 CAR T cells in NSG mice bearing SK-N-AS-GD2 (GPC2^Low, GD2^High) cell line xenografts (Fig. 7B and fig. S9A). All GPC2+ SyntEVs variants resulted in significantly better tumor control than GPC2 CAR T cells alone (Fig. 7C and Fig. S9, B and C). However, the ALB and GD2-binding GPC2+ SyntEVs outperformed the unmodified GPC2+ SyntEVs, preventing overt tumor relapse and enabling survival until day 80 post-CAR T cell administration when the study was concluded due to most mice acquiring xenogeneic graft-versus-host disease (xGvHD; Fig. 7D), an expected outcome due to human T cell reactivity towards host murine tissues (47). All EV administrations were well tolerated as no signs of systemic toxicity or changes in body weight were observed, except the late weight loss associated with xGvHD onset (fig. S9D). The differences in anti-tumor response were associated with increased persistence of GPC2 CAR T cells in the peripheral blood of mice receiving GPC2+ SyntEVs, with the highest peripheral human T cell count detected in the anti-ALB SyntEV treatment group 20 days post-CAR T cell infusion (Fig. 7E). Tumor-infiltration of GPC2 CAR T cells in the presence or absence of co-administration of GPC2+ SyntEVs was also verified by human CD3 immunohistochemistry (IHC) in satellite tumors collected 10 days after CAR T cell injection (fig. S10, A to D).

Figure 7. Armored GPC2+ SyntEVs boost GPC2 CAR T in vivo persistence and anti-tumor efficacy.

(A) top: GFP western blot of SK-N-AS-GD2 tumors from mice treated with two I.V. doses of GPC2+ SyntEVs 5 hours after the last EV administration. bottom: GFP western blot of circEVs isolated from the peripheral blood of the same mice. Respective purified SyntEV controls are also shown to validate similar levels of GFP in each EV cohort.

(B) Experimental design of in vivo study testing GPC2 CAR T cells in combination with GPC2+ EVs in SK-N-AS-GD2 xenografts (GPC2^Low, GD2^High).

(C) SK-N-AS-GD2 tumor growth curves after treatment with CD19 or GPC2 CAR T cells in combination with naked or armored GPC2+ SyntEVs (n = 8–9 mice per group). The indicated p-values correspond to differences between groups at day 45.

(D) PFS for the different CAR treatment cohorts shown in B and C.

(E) Quantification of circulating hCD45+ hCD3+ (CAR T) cells by flow cytometry in peripheral blood samples from mice enrolled on study shown in B-D (n=4 mice per group). Samples were collected at day 7 (post-two EV doses) and day 20 (post-five EV doses) after CAR T cell infusion.

(F) Experimental design of in vivo study testing GPC2 CAR T cell vaccination with GPC2+ ± anti-ALB/GD2 SyntEVs in the neuroblastoma COG-N-561x PDX (GPC2^High, GD2^UltraHigh) mouse model.

(G) Quantification of circulating hCD45+ hCD3+ (CAR T) cells by flow cytometry from peripheral blood samples from COG-N-561x-bearing mice enrolled on the study shown in F (n=3–4 mice per group. Samples were collected at day 7 and 14 (pre-EV administration) and day 21 (post-two EV doses) after CAR T cell injection. Only one mouse remained in the control CD19 CAR group at day 21.

(H) PFS for COG-N-561x-bearing mice after treatment with CD19 or GPC2 CAR T cells in combination with naked or armored GPC2+ SyntEVs as shown in F.

(I) COG-N-561x tumor growth curves for the different CAR treatment cohorts shown in F-H. Statistical analysis was performed by comparing tumor volumes at day 37.

(J) GPC2 IHC on satellite COG-N-561x tumors 20 days post-treatment with CD19 or GPC2 CAR T cells with or without GPC2+ anti-GD2 SyntEVs according to F. A representative tumor (left) and GPC2 IHC staining intensity quantification for all tumors (n=3 per group; right) are shown. Scales bars represent 200 mm.

(K) Quantification of human T cells by flow cytometry in COG-N-561x tumors shown in J (20 days post-CAR T cell infusion). Flow cytometry plots showing CD3 staining (left) and T cell quantification as % of hCD3+ cells (right) are shown (n=3 per group).

Data are shown as mean ± SEM. Dots represent independent biological replicates. Statistical analysis was performed using Log-rank test for PFS curves or a one-way ANOVA with Tukey’s (C, E, G, and I) or Dunnett’s (J and K) multiple comparisons test to evaluate differences between groups. *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001. Dotted lines in E indicate CAR T cell count from unpaired peripheral blood samples, whereas solid lines in G represent CAR T cell count from paired peripheral blood samples. Illustrations in B and F created with BioRender.com. O.D., optical density.

To further validate the therapeutic relevance of antigen-displaying EVs on CAR T cell efficacy in neuroblastoma, GPC2+ anti-ALB or anti-GD2 EVs were next tested in the human neuroblastoma PDX model, COG-N-561x, which exhibits both high cell surface GPC2 and GD2. Previous studies from our laboratory showed that GPC2 CAR T cells can induce initial complete tumor regressions in this neuroblastoma PDX, but that tumor relapses occur approximately 2.5 weeks post-CAR T cell injection (19). Thus, this study was specifically designed to evaluate the ability of GPC2+ EVs to reinvigorate CAR T cells through EV-induced antigen stimulation after this initial tumor eradication. Following CAR T cell injection, mice were monitored for tumor volume and peripheral CAR T cell counts. On day 17 (week 2.5), when the T cell count in the peripheral blood dropped below 200 cells/μL, GPC2+ EV administration was initiated (Fig. 7F). Infusion of both anti-ALB and anti-GD2 GPC2+ EVs resulted in substantial peripheral CAR T cell expansion (Fig. 7G), prolonged animal survival (Fig. 7H), and enhanced tumor control compared to GPC2 CAR T cells alone (Fig. 7I and fig. S11A), without signs of toxicity (fig. S11B). Finally, six additional mice were also treated with CAR T cells when tumor volume reached a size of 0.3 ± 0.05 cm³, followed by administration of GPC2+ GD2-binding EVs or vehicle, according to the same in vivo study design. Immunohistochemistry and flow cytometry analysis of tumors collected on day 20, five hours after the second EV dose, confirmed that GPC2 was downregulated in tumors treated with GPC2 CAR T cells (Fig. 7J and fig. S11C). In contrast, mice vaccinated with GPC2+ GD2-binding EVs exhibited increased GPC2 tumor expression (Fig. 7J and fig. S11C) and a significant enhancement in human T cell recruitment (Fig. 7K), demonstrating the engineered EVs’ ability to traffic to the tumor, supplement the tumor with extra GPC2 antigen, and enhance GPC2 CAR T cell homing.

DISCUSSION

Antigen heterogeneity, limited tumor infiltration, and poor persistence of adoptively transferred T cells are major barriers limiting the durable efficacy of CAR T cell therapy in solid tumors (48), including the commonly lethal embryonal malignancy neuroblastoma. To overcome these challenges, it is essential to understand the molecular mechanisms regulating CAR T cell response, including the role of TEVs as intercellular messengers between tumor and immune cells. Here, we demonstrate that GPC2+ and GD2+ EVs can be specifically detected in the peripheral blood of mice bearing neuroblastoma PDXs, where the degree of GPC2 and GD2 positivity correlate both with tumor size and tumor antigen expression. GPC2 and GD2 were also selectively detected on circEVs isolated from plasma of patients with neuroblastoma, suggesting that they may also be useful diagnostically in the pediatric oncology clinic. Other glypicans have been previously identified on circEVs from patients with cancer, highlighted by GPC3 in hepatocellular carcinoma (39), and GPC1 in pancreatic cancer (38) where GPC1+ circEVs distinguished healthy subjects and patients with benign pancreatic disease from those with overt pancreatic cancer. Although our analyses are limited to a discovery cohort and will require further validation in a larger independent patient series, these proof-of-concept studies support further exploring circEVs as a potential non-invasive tumor biomarker for children with neuroblastoma and potentially other lethal GPC2+ malignancies. For example, much like circulating tumor DNA profiling might offer clinically relevant real time information on targetable genomic alterations for children with neuroblastoma (49), serial protein profiling of circEVs might provide data to determine eligibility for personalized cell surface targeted immunotherapies. Further, our results showing the functional effects of TEV-exposed target antigen on CAR T cells also suggest that the abundance of circEVs should potentially be taken into account when administrating immunotherapies. Correlative research studies monitoring EV kinetics pre- and post-therapy should be incorporated into ongoing CAR T cell trials, similar to currently active clinical trials monitoring the kinetics of circEVs during combination therapies using immune checkpoint inhibitors in patients with high-grade glioma (NCT03576612) or melanoma (NCT04581382). Considering that GPC2, GD2, and other tumor antigens (such as B7-H3 and L1CAM) exposed on the TEV surface are targets for clinically available CAR T cells (23), these data are immediately relevant to several ongoing clinical trials.

Prior studies in hematological malignancies have dissected the interaction between TEVs and CD19 targeting CAR T cells, showing that TEVs carrying both target antigen and the inhibitory checkpoint protein PD-L1 inhibit CD19 CAR T cells (50). Similarly, a recent study in pancreatic cancer showed inhibitory effects of TEVs on mesothelin-redirected CAR T cells (14). Zhong et al. elegantly demonstrated that although target antigen on TEVs is essential to induce the synapse of TEV and CAR, high abundance of PD-L1 on TEVs ultimately induces mesothelin CAR T inhibition. Conversely, our studies showed that target antigen exposed on neuroblastoma derived TEVs enhance the effectiveness of paired CAR T cells. This discrepancy with previous studies may be attributed to differences in the other protein cargo components of neuroblastoma-derived EVs, especially the limited expression of PD-L1 (51). Even after IFN-γ stimulation, resembling a condition occurring in the TME in the presence of an effective CAR T cell response (52), PD-L1 upregulation on neuroblastoma TEVs was minimal when compared to even the basal exposure on melanoma-derived EVs. Due to the specific cargo of neuroblastoma TEVs, we focused on defining the effects of target antigen exposure on EVs on CAR T cell functionality. However, TEV cargo composition appears to be cancer specific (53), and the effects of tumor antigens loaded on TEVs may vary across tumor types based on other co-exposed molecules, particularly immune inhibitory checkpoint or costimulatory molecules (5, 6).

In addition to being natural nanocarriers (53), EVs can also be engineered to therapeutically deliver functional molecules specifically to tumors, thus offering potential as next-generation targeted nanomedicines (54, 55). For example, genetically engineered EVs, called SMART-Exos, with surface display of anti-human CD3 and anti-human HER2 antibodies, have shown preclinical anti-tumor efficacy by redirecting cytotoxic T cells toward HER2-expressing breast cancer cells (56). In support of an immunotherapeutic application of EVs, CAR EVs derived from effector CAR T cells have been nominated as an ‘off-the-shelf therapeutic’ with lower risk of toxicities, such as cytokine release syndrome (CRS), compared to respective CAR T cell treatments (57). Clinically, vaccination strategies based on antigen-loaded DC-derived EVs are also currently being explored (NCT01159288). These studies collectively underscore the potential value of engineered EVs in the clinic. Here, we further extend the therapeutic potential of EVs by exploring antigen-displaying EVs as a CAR T cell vaccination strategy in preclinical solid tumor models. Despite finding minimal PD-L1 exposure on neuroblastoma TEVs, to avoid any potential immune-inhibitory effects commonly associated with TEVs (5, 14), we opted for using non-tumor EVs engineered to display the antigen of interest, GPC2, for these purposes (herein referred to as GPC2+ SyntEVs). We confirmed with mass spectrometry studies that these GPC2+ SyntEVs lack common inhibitory checkpoint molecules, such as PD-L1 and PVR, in their protein cargo.

The importance of long-term CAR T cell persistence for durable clinical efficacy has been clearly delineated for several cancers (58). However, CAR T cell levels typically decline rapidly after infusion in patients with solid tumors (59, 60), in part due to the impaired accessibility of adoptively transferred cells to target cells within solid lesions and the absence of additional co-stimulation signals upon target encounter in the hostile immunosuppressive TME. In contrast, CAR T cells targeting hematologic malignancies immediately encounter their targets on circulating cells upon entering the bloodstream, a factor that may contribute substantially to their sustained anti-tumor responses. To address these limitations, lipid nanoparticles delivering antigen-encoding RNAs (61) and amphiphile CAR ligands targeting lymph nodes through antigen-linking to albumin-binding phospholipid polymers (62) have been used to boost solid tumor CAR T cell responses. Both methods promote priming and expansion of CAR T cells outside the immunosuppressive TME (61, 62). Here, we explored an alternative approach to address the lack of effective extra-tumoral CAR stimulation in solid tumors, using GPC2+ SyntEVs to boost GPC2 CAR T cells. Although we saw enhanced in vivo GPC2 CAR T cell efficacy when given in combination with serial dosing of GPC2+ EVs, given that exogenous EVs introduced into the circulation are known to have a short half-life ranging from 30 min to 5 hours depending on the cell source (63), we further hypothesized that extending the circulation half-life of our GPC2+ SyntEVs by facilitating binding to serum albumin (46) might further enhance their ability to positively modulate GPC2 CAR T cells. In turn, our in vivo data using these “armored” GPC2+ SyntEVs indeed suggested that GPC2+ ALB-binding SyntEVs persisted longer in the circulation, compared to naked GPC2+ SyntEVs, and dramatically enhanced GPC2 CAR T cell peripheral proliferation. Although increasing CAR T cell persistence in the periphery might lead to on-target-off-tumor toxicities or cytokine release syndrome, these adverse effects were not observed in mice that received these mouse/human cross-reacting GPC2 CAR T cells along with serial IV injections of GPC2+ ALB-binding SyntEVs. However, since development of xGvHD due to expected human CAR T cell reactivity towards host murine tissues limits the time of evaluation in this model, the long-term risks of prolonged CAR T cell stimulation from GPC2+ SyntEVs requires further investigation.

We also tested a different armored-EV strategy aimed at addressing a second major limitation to CAR T cell solid tumor efficacy - antigen downregulation (3). CAR-induced tumor target downregulation has been observed with most effective CAR T cell therapies including GPC2 CARs (21), and CARs targeting CD19 (64), EGFRvIII (65), IL13Rα2 (66), among several others. Thus, we armored GPC2+ SyntEVs with tumor-targeting ability achieved through binding to a second tumor antigen, GD2, and demonstrated the ability of the anti-GD2 SyntEVs to decorate neuroblastoma cells with GPC2 proportional to GD2 abundance. Indeed, in vivo systemic administration of the GD2-binding SyntEVs resulted in increased GPC2 on tumor cells and enhanced CAR T cell tumor infiltration and anti-tumor efficacy. Our “EV vaccination study” using the COG-N-561x model further validated the feasibility of using GPC2+ armored SyntEVs to restimulate CAR T cells in vivo weeks after initial CAR T cell administration to maintain the number of circulating CAR T cells within a therapeutic window to enable durable tumor control.

Our study has several limitations. For example, although our use of human xenograft models in immunodeficient NSG mice offers the advantage of recapitulating the human expression of tumor antigens like GPC2 and GD2, it also poses some restraints to the understanding of the full therapeutic potential of antigen-displaying SyntEVs as CAR T cell enhancers as well as their broader effect on endogenous immune cells. For example, ALB-binding amph-ligands (62) or EVs (46) have demonstrated trafficking to lymph nodes where they transfer antigens to professional antigen-presenting cells. This process renders the antigen accessible for cognate CAR T cell stimulation in an optimal immune-activating environment (62). Thus, these studies using immunocompromised mouse models may potentially underestimate the therapeutic benefits of combining antigen presenting SyntEVs with CAR T cells. Future studies using syngeneic neuroblastoma immunocompetent models treated with murine CAR T cells and paired antigen+ EVs are needed to fully define the effects of EV antigen presentation in a competent adaptive immune setting and additionally evaluate SyntEV-induced effects on endogenous immune cell populations. Moreover, these studies would allow verification of the hypothesis that ALB-binding SyntEVs can decorate dendritic cells resulting in GPC2 cross-presentation in the lymph-nodes and engagement of endogenous anti-tumor T cells, similar to “amph-ligand” vaccines (62). Moreover, our single-cell functional cytokine immune profiling of GPC2 CAR T cells after exposure to GPC2-displaying SyntEVs revealed that EV stimulation induces not only secretion of effector and stimulatory cytokines, like IFN-γ and IL-2, but also chemoattractive chemokines like MIP-1α and MIP-1β which are known to induce recruitment and activation of monocytes/macrophages and NK cells (67). These data additionally suggest that SyntEV-mediated CAR T cell activation might also stimulate recruitment and bystander activation of endogenous anti-tumor immune cells, further amplifying their therapeutic benefits, an effect that is best studied in immunocompetent preclinical models. Finally, although we designed the studies here for stimulation of GPC2 CAR T cells in neuroblastoma, this SyntEV system can be easily adapted to not only different CAR target molecules but also to different secondary EV-targeted molecules alternative to GD2, that could be modified even during therapy. This additional engineering of EVs to display specific binding properties and target antigens may add complexity to the EV manufacturing process. However, HEK293T cells, which we utilize as SyntEV producer cells, are particularly amenable to stable genetic modifications and have been previously used to meet large-scale EV production for clinical use (68). Similarly to the uniCAR approach (69), this approach potentially allows the use of a single CAR that can be flexibly redirected towards multiple tumor targets during therapy driven by the specific binding properties of the co-infused antigen presenting EVs.

Our studies led to the following major findings, summarized in Fig. S12: (1) neuroblastoma tumors secrete GPC2+ and GD2+ TEVs that inform tumor antigen expression and tumor burden; (2) TEVs exposing immunotherapeutic target antigens, such as GPC2 or GD2, functionally bind to paired CAR T cells inducing T cell activation and enhanced specific cytotoxicity against neuroblastoma cells, including those with lower antigen expression; and (3) target antigen displaying-SyntEVs can be exploited therapeutically to boost CAR T cell persistence and anti-tumor efficacy in solid tumors by providing peripheral antigen stimulation or delivering additional antigen to the tumor to circumvent resistance from target downregulation. Taken together, this study offers a proof of concept of the versatility of EV engineering as a platform for developing CAR T cell engagers with desired properties to enable development of transformative combination immunotherapeutic approaches readily adaptable to different cancer types.

MATERIALS AND METHODS

Study design

The objectives of this study were to assess the functional effects of target antigen-displaying EVs on CAR T cells and to harness their stimulatory properties to develop CAR T cell adjuvants that can overcome the therapeutic limitations of treating solid tumors. GPC2 and GD2, validated immunotherapeutic targets in neuroblastoma being evaluated in CAR T cell phase I clinical trials (NCT05650749, NCT05990751), were used as model target molecules. All in vitro experiments were performed independently at least three times. In vivo experiments were performed with random assignment to treatment groups based on a predefined range of tumor size, which was used as an inclusion criterion for enrollment; animals with tumors outside this range were excluded from the study. Experiments were conducted in a blinded fashion, with investigators assessing tumor growth unaware of group allocation. Sample sizes for in vivo experiments were determined based on prior publications and laboratory experience, with a minimum of four mice per group. Tumor size was measured by a digital caliper. The sample and tumor size criteria used for enrollment in each experiment are indicated in the figures or corresponding figure legends. Animal experiments were conducted using protocols approved by the CHOP Institutional Animal Care and Use Committee (IACUC; Protocols #0643 and #1464) with adherence to the NIH guide for the Care and Use of Laboratory Animals accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) and ARRIVE guidelines. Human neuroblastoma PDXs (COG-N-561x, COG-N-421x, COG-N-424x, COG-N-619x, COG-N-452x, and COG-N-496x) were acquired from the Children’s Hospital of Philadelphia Neuroblastoma biobank and the Children’s Oncology Group (COG) Cell Culture and Xenograft Repository (www.cccells.org) and were characterized previously (70). Primary human T cells to generate CAR T cells were obtained from anonymous donors through the Human Immunology Core at the University of Pennsylvania. Donors provided informed consent through the University of Pennsylvania Immunology Core (RRID:SCR_022380). Twelve peripheral blood samples were collected with informed consent from children without cancer under the Children’s Hospital of Philadelphia Institutional Review Board protocol #15920. Fifteen peripheral blood samples were obtained with informed consent through postmortem blood draws from neuroblastoma patients who died of progressive disease 1 to 6 hours after death. These samples were used to illustrate the selective presence of GPC2+ or GD2+ EVs in the peripheral blood of patients with neuroblastoma.

Statistical analysis

All statistical analyses were performed using GraphPad software. For comparisons between two groups, either a Student’s t-test or a Mann-Whitney test was used. For comparisons among more than two groups with one independent variable, a one-way ANOVA followed by Tukey’s or Dunnett’s multiple comparisons test was applied, as indicated in the figure legends. Survival curves were generated using the Kaplan–Meier method and compared using the log-rank (Mantel–Cox) test. All tests were two-sided, and statistical significance was defined as a P < 0.05. No additional adjustments to alpha levels were made for multiple testing beyond the post hoc corrections specified. The exact statistical test and number of replicates (n) are reported in the respective figures or figure legends. All individual-level data for experiments with n<20 are available in Data File S1.

Supplementary Material

List of Supplementary Materials

Materials and Methods

Figures S1 to S12

Data File S1

MDAR reproducibility checklist

Data and materials availability

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD067632. Newly generated GPC2+ and GPC2− SyntEV-producer cell lines, as well as plasmid constructs used for their production, are available from the corresponding author upon completion of a Material Transfer Agreement with the Children’s Hospital of Philadelphia. There are no additional restrictions on the availability of these materials. All data associated with this study are present in the paper or supplementary materials.