Improved Antitumor Efficacy of Chimeric Antigen Receptor T Cells that Secrete Single-Domain Antibody Fragments

Chimeric antigen receptor (CAR) T-cell therapy relies on T cells that have been redirected with a receptor designed to recognize an antigen of choice. CAR T cells have been used to successfully treat hematologic cancers (1, 2). In the case of tumors of B-cell origin, CD19-targeted CAR T cells lead to remission in patients refractory to several other lines of therapy (3). CAR T cells have shown less success in the treatment of solid tumors. Obstacles include a paucity of tumor-specific targets and a high rate of antigen escape (4, 5). A dense extracellular matrix, characteristic of many solid tumors, and the presence of inhibitory checkpoint signals both blunt the immune response (6–8). Suppressive cell types, such as T-regulatory cells or myeloid-derived suppressor cells, can further inhibit or cause exhaustion not only of CAR T cells but also of endogenous T cells that have infiltrated the tumor (6). This combination of factors prevents CAR T cells from recognizing and attacking the tumor. To overcome some of these challenges, “armored” CAR T cells that secrete cytokines such as IL12, IL15, and IL18 can modify the tumor microenvironment, make it more hospitable for T-cell activity, and thus promote an antitumor response (9–11). Other means of manipulating the tumor microenvironment, for example by having CAR T cells release proteins that affect cell–cell interactions, therefore deserve consideration as well.

Variable heavy domain of heavy chain antibodies (VHH), also referred to as nanobodies, are single-domain antibody fragments derived from the variable region of camelid heavy-chain-only antibodies (12). They are small, standalone proteins of approximately 15 kDa that retain binding affinities comparable with full sized mAbs (13). VHHs are stable, soluble, and can be expressed in excellent yields without the need for extensive optimization (14). For use as therapeutic agents, their small size and sequence similarity to human immunoglobulin V regions render them less immunogenic than murine mAbs, and humanization of VHHs is possible (15). CAR T cells can be engineered so that they express and secrete VHHs with immunomodulatory properties to enhance their antitumor effect. Current immunotherapy approaches aim to improve adaptive immune recognition and killing of tumor cells, but the importance of engaging the innate immune system for enhanced antigen presentation and epitope spreading is increasingly recognized (16, 17). For example, the CD47 protein delivers a “don't eat me signal” to phagocytes. Consequently, a blockade of this signal synergizes with the efficacy of mAb therapy for a variety of cancers in preclinical models (18–21). Here, we show that engagement of the innate immune system through blockade of CD47 by anti-CD47 VHH-secreting CAR T cells improves their antitumor effect.

We further demonstrate the modularity of VHH and VHH fusion–secreting CAR T cells by developing CAR T cells that secrete VHHs and VHH fusion proteins specific for multiple checkpoints. Such CAR T cells show less exhaustion and improved persistence when compared with nonsecreting CAR T cells. Because CAR T cells traffic to and persist at sites where their antigen is present (in our case, the tumor microenvironment), this strategy can limit immune-related adverse effects associated with checkpoint blockade (22). Furthermore, we show that, owing to their modest size, multiple VHH and VHH-fusions can be expressed from the same vector in the same cell, underscoring the flexibility of this approach.

Using CAR T cells as a vehicle for local delivery of VHH and VHH fusion proteins should limit systemic exposure to immune modulators such as anti-CD47 mAbs that have off-tumor toxicity. Anti-CD47 blocks engagement of the CD47 ligand, SIRP1a, while enhancing FcγR-mediated phagocytosis (23). The ubiquitous expression of CD47, especially on red blood cells (RBC), complicates systemic administration of anti-CD47. The anemia that results from such treatment is sufficiently severe to have led to interruption of several clinical trials because of safety concerns (24, 25). Even so, CD47 blockade remains a promising treatment if the toxicity concerns can be dealt with (26). We demonstrate that an anti-CD47 VHH, fused to an Fc portion and secreted by CAR T cells, provides a stimulus for macrophage engulfment. CAR T cells that secrete anti-CD47 Fc fusions show superior antitumor efficacy compared with CAR T-cell therapy without such supplementation. Toxicity associated with systemic exposure to anti-CD47 Fc is thus avoided.

Vectors used for the generation of CAR T cells were comprised of a fluorescent protein tracer linked to a P2A sequence followed by a CD8 signal sequence and the VHH domain linked to a CD8 hinge and transmembrane domain. The VHH sequences used are as follows: A4 – QVQLVESGGGLVEPGGSLRLSCAASGIIFKINDMGWYRQAPGKRREWVAASTGGDEAIYRDSVKDRFTISRDAKNSVFLQMNSLKPEDTAVYYCTAVISTDRDGTEWRRYWGQGTQVTVSS; A12 – QVQLVESGGGLVQAGGSLRLSCTASGSTFSRNAMAWFRQAPGKEREFVSGISRTGTNSYDADSVKGRFTISKDNAKNTVTLQMNSLKPEDTAIYYCALSQTASVATTERLYPYWGQGTQVTVSS; H11– QVQLQESGGGLAQPGGSLRLSCAASGSTISSVAVGWYRQTPGNQREWVATSSTSSTTATYADSVKGRFTISRDNAKNTIYLQMNSLKPEDTAVYYCKTGLTNWGQGTQVTVSS; and B2 – QVQLVETGGGLVQAGGSLRLSCAASGSTFSHNAGGWYRQAPEKQRELVAGISSDGNINYADSVKDRFTISRDNASNTMYLQMNNLKPEDTAVYVCNIRGSYGNTYYSRWGQGTQVTVSS. These were connected to a CD28 costimulatory signal and a CD3ζ activation signal as described previously (27, 28). For VHH-secreting CARs, a P2A, IRES, or separate cytomegalovirus (CMV) construct followed the CD3ζ activation domain, and these were linked to an IgK signal sequence to drive secretion of the VHH, which was tagged at the C-terminus with an HA tag. The CAR constructs are contained in a murine stem cell virus–based γ retroviral vector, the XZ vector, a gift from the Lodish laboratory (29). To generate virus, HEK 293T cells were transfected using Fugene 6 (Promega) with the XZ CAR vector and a pCL-Eco packaging vector. Cloning was carried out in DH5a E. coli cells.

To generate CAR T cells, pan T cells were isolated from mouse splenocytes using a magnetic bead-based–negative selection strategy according to the manufacturer's instructions (Thermo Fisher Dynabeads Untouched Mouse T Cell Kit). T cells were activated overnight with anti-mouse CD3 and CD28 (BD Biosciences) in complete RPMI [RPMI supplemented with 10% heat-inactivated FBS (IFS), nonessential amino acids, 1 mmol/L sodium pyruvate, 2 mmol/L l-glutamine, 100 U/mL Penicillin/streptomycin, and 50 μmol/L 2-mercaptoethanol] and mIL2 (recombinantly, house-made) media and then transduced using the retrovirus and polybrene. T cells were spinfected twice in 6-well plates at 2,000 g with the virus and polybrene mixture for 90 minutes. Fresh media was added after each spinfection, as described previously (27, 30). After two spinfections, cells were expanded for 2 days in complete RPMI IL2 and used for in vitro and in vivo assays.

Anti-HA immunoprecipitations (IP) were performed on culture supernatants of transduced cells using the Pierce HA-Tag IP/Co-IP Kit (Thermo Fisher Scientific). HA tags were located at the C-terminus of the secreted VHHs. Supernatant (850 μL) was incubated with anti-HA agarose beads overnight, and eluates were run on a 12% SDS PAGE gel and transferred to a PDVF membrane (Bio-Rad TurboBlot). Membranes were blocked in 5% milk in TBST and incubated with anti-HA-HRP mAb overnight. Membranes were developed with western ECL substrate (PerkinElmer) and imaged on a Bio-Rad ChemiDoc imager. For immunoblots on cell lysates, cells were lysed with RIPA buffer (25 mmol/L Tris, 150 mmol/L NaCl, 1% NP-40, 0.5% sodium deoyxcholate, and 0.1% SDS) and protease inhibitor (complete mini protease inhibitor, Sigma). DNA was removed through benzonase digestion, and lysates were boiled with Laemmli sample buffer and 2-mercaptoethanol.

Cytotoxicity of CAR T cells was measured using various coculture assays. The cytotoxicity of the PD-L1–targeted CAR T cells was shown by coculture with B16F10 melanoma cells, which express PD-L1. B16 cells were plated in complete RPMI (Corning) and recombinant murine Il2, and A12 or A12 A4 CAR T cells were added to the B16F10 cells for a 24-hour incubation. T cells were then washed out of the plate using PBS, and the number of remaining viable B16F10 cells were measured using a Cell Titer Glo Assay (Promega). Supernatants of each coculture were collected for IFNγ ELISA measurements (BD Biosciences).

B16F10 cells obtained from the ATCC and IDEXX cleared for Mycoplasma (2017) were used in the animal tumor models and in vitro studies. Cells were not reauthenticated within the past year. Cells were maintained in complete RPMI, RPMI1640 supplemented with 10% heat-IFS, nonessential amino acids, 1 mmol/L sodium pyruvate, 2 mmol/L l-glutamine, 100 U/mL Penicillin/streptomycin, and 50 μmol/L 2-mercaptoethanol. All cell lines were kept in culture for up to 2 months. B16F10 PD-L1^–/– lines were generated using CRISPR-Cas9, and single clones were selected and sequenced to verify gene disruption. MC38 was obtained from Kerafast (2017) and cleared for Mycoplasma (2017), and cells were grown in Dulbecco medium supplemented with 10% heat-IFS, nonessential amino acids, 1 mmol/L sodium pyruvate, and 100 U/mL penicillin/streptomycin.

Recombinant VHHs were produced by periplasmic expression in WK6 cells. Recombinant VHH-Fc fusions were produced in Expi293 cells, which were cultured for 4 days after polyethylenimine PEI transfection. Proteins were purified from supernatants or periplasmic fractions on Ni-NTA beads (Qiagen) and eluted with 500 mmol/L imidazole. Eluted proteins were further purified with size exclusion chromatography on a SuperDex 75 column run in PBS. Endotoxin removal was performed by extensive washing of column-bound materials with endotoxin-free PBS and 0.1% TritonX114.

All animal procedures performed were in accordance with institutional guidelines and approved by the Institutional Animal Care and Use Committee of Boston Children's Hospital (IACUC protocol 16-12-3328). To test the efficacy of the A4-secreting PD-L1–targeted CAR T cells, PD-L1–deficient mice were inoculated with 5 × 10⁵ cells of B16F10 wild-type (WT) cells. Two days later, mice were treated with A12 A4-secreting CAR T cells, A12 CAR T cells with daily systemic A4, no CAR T cells with daily systemic A4, or left untreated. A total of three CAR T-cell injections were given intravenously in combination with a twice weekly intraperitoneal injection of the anti-TRP1 antibody TA99, a gift from the D. Wittrup laboratory. Tumor measurements were taken by caliper every 1–2 days. Animals were sacrificed following CAC protocols when the tumor area was >125 mm², and mice with ulcerated tumors were removed from the studies. For the MC38 model, 3 × 10⁵ cells were inoculated into PD-L1–deficient mice, and 5 days later, mice were treated with A12-A4Fc–secreting CAR T cells, A12 CAR T cells with a single dose of systemic A4Fc, A12 CAR T cells only, no CAR T cells with a single dose of systemic A4Fc, or left untreated. A total of three intravenous CAR T-cell injections were given, and tumor measurements were taken by caliper every 1–2 days. To test the effect on persistence of the A12-secreting B2-targeted CAR T cells, WT mice were inoculated with 1 × 10⁵ cells of B16F10 WT cells. Four days later, mice were treated intravenously with A12-secreting B2 CAR T cells, B2 CAR T cells, or left untreated. To test the effect on persistence of the H11Fc-secreting B2-targeted CAR T cells, WT mice were inoculated with 1 × 10⁵ cells of B16F10 WT cells. Four days after, mice were treated with H11Fc-secreting B2 CAR T cells, B2 CAR T cells with three systemic doses of H11Fc, only three systemic doses of H11Fc, or left untreated. In persistence experiments, spleens, lymph nodes, and tumors were removed to check for the presence of CAR T cells. Mice from each treatment group were cohoused to minimize variability due to external factors. Experiments were repeated at least twice.

IFNγ ELISPOT assays were performed by coculturing with B16F10 or B16F10 PD-L1–knockout (KO) cells using a mouse IFNγ ELISPOT Kit (BD Biosciences). B16F10 and B16F10 PD-L1–KO tumor cells were cultured with IFNγ overnight. Splenocytes (1 × 10⁶) from each treated mouse were then cocultured with 25,000 irradiated tumor cells (100 Gy) per well. Plates were developed according to the manufacturer's protocol and imaged with CTL Immunospot S6 analyzer.

Statistical analyses were performed using GraphPad Prism. The log rank (Mantel Cox) test was used to analyze survival curves. Flow cytometry data were analyzed using two-tailed, unpaired Student t tests, considering multiple comparison corrections. Statistical significance was determined using the methods described in the Results section or in the figure legends using the Bonferroni correction for comparison between multiple groups.

CD47 VHH-secreting anti–PD-L1 (A12) CAR T cells were generated using the constructs depicted in Fig. 1A (27). We chose to develop CAR T cells that secrete an anti-CD47 VHH without the Fc portion to limit the toxicities associated with Fc engagement, as CD47 is expressed across a wide range of normal cells. We generated viral vectors that encode fluorescently labeled A12 (anti–PD-L1) CARs linked to a self-cleaving P2A peptide or IRES sequence and a secreted anti-CD47 VHH (A4; refs. 18, 27). We also tested constructs with monocistronic expression, in which one promoter controls expression of the CAR and a second promoter controls expression of a VHH (Fig. 1B). Success of transduction was gauged by expression of the fluorescent reporter molecule (Fig. 1C). To verify that the transduced CARs express anti-CD47 VHH, a flow cytometry–based assay was used. A population of both transduced and control CAR T cells was stained with a fluorescently labeled anti-CD47 mAb that competes for binding with the A4 anti-CD47 VHH. Populations of non-A4–secreting CAR T cells were also stained by the anti-CD47 mAb, because all T cells express CD47. Populations that contain the A4-secreting CAR T cells showed a reduction of anti-CD47 staining in both transduced and untransduced cells. The A4 anti-CD47 VHH is thus secreted and can bind to any cell present, regardless of whether that cell itself was transduced (Fig. 1D; Supplementary Fig. S1). Expression and secretion of A4 was further confirmed by performing an anti-HA immunoblot on CAR T-cell lysates or on an anti-HA immunoprecipitate of the culture supernatant of transductants (Fig. 1E). The 1B7 pCMV A4 construct is a VHH-based CAR T cell, in which the CAR VHH recognizes a Toxoplasma gondii (irrelevant) antigen. The 1B7 CAR T cell was likewise engineered to secrete the A4 VHH to serve as a control. The A12 (anti–PD-L1) CAR in which a separate CMV promoter drives expression of A4 showed the most A4 secretion. Because of limitations in transduction capabilities, possibly due to the large construct size of the CMV and IRES vectors, we performed all further experiments with the A12 P2A A4 construct. To confirm that killing by A12 CAR T cells was not negatively affected by secretion of the A4 VHH, we performed a killing assay on PD-L1⁺ B16F10 cells. Cell killing was measured with a CellTiter Glo assay, and T-cell activation was measured by production of IFNγ (Fig. 1F). Killing of B16F10 cells was comparable for the A12 CAR T cells and A12 P2 AA4 CAR T cells. Enforced secretion of A4 therefore neither impairs CAR T-cell activity nor production of IFNγ.

Having confirmed the generation of A4-secreting A12 (anti–PD-L1) CAR T cells, we examined whether secretion of A4 by CAR T cells would improve their antitumor activity (Fig. 2A). PD-L1 KO mice were used as a source of T cells to generate A12 CAR T cells, as we have observed better persistence of these cells due to a reduction in chronic CAR signaling in cis, caused by basal PD-L1 expression on WT T cells (27). WT CAR T cells generated in the presence of the anti–PD-L1 VHH behave similarly to CAR T cells generated in a PD-L1–KO background. We have previously demonstrated that chronic signaling from WT CAR T cells leads to upregulation of exhaustion markers (27). B16F10 and MC38 tumors express CD47, as determined by flow cytometry (Supplementary Fig. S2). C57BL/6 PD-L1–KO mice were inoculated with B16F10 tumors on day 0, and received the treatments depicted in Fig. 2A. All groups also received the melanoma-specific anti-TRP1, TA99, to enhance antitumor activity. This allows macrophages and other phagocytic cells to exert an effect on tumor growth through FcγR engagement, which is enhanced by A4-VHH–mediated blockade of the CD47 “don't eat me” signal (Fig. 2A). Tumor sizes were measured by caliper every 1–2 days. Treatment with the A12 (anti–PD-L1) CAR T cells in conjunction with systemically delivered A4 VHH improved survival when compared with mice treated with A4 VHH only, or in comparison with mice that received no treatment (Fig. 2B). However, the A4-secreting A12 CAR T cells provided a significant survival benefit when compared with all other conditions. Secretion of (anti-CD47) A4 by CAR T cells further inhibits, but does not fully control, tumor growth in this aggressive, syngeneic tumor model (Fig. 2B). The antitumor effects of A12 P2A A4 CAR T cells were also evaluated in WT mice. WT C57BL/6 mice were treated with A12 CAR T cells or A4-secreting A12 CAR T cells, or left untreated, and tumor sizes were measured (Fig. 2C). TA99 treatment was given to all groups. In these WT, immunocompetent models, even at lower doses, A12 P2A A4 CAR T cells also show improved antitumor activity compared with A12 CAR T cells (Fig. 2D).

A nanobody, B2 that recognizes a splice variant, EIIIB, of fibronectin detects the tumor neovasculature and tumor stroma and was used to generate CAR T cells (27, 28). In similar fashion, we developed EIIIB-targeted CAR T cells that secrete A4 VHH (B2P2AA4 CARs) to explore the A4 secretion strategy in a different model. Flow cytometry shows that these B2-A4 CAR T cells secrete A4 VHH (Fig. 2E). We tested an in vivo tumor model in WT mice that compared no treatment with treatment with either B2 CAR T cells or B2P2AA4 CAR T cells (Fig. 2F). Mice were inoculated with B16F10 tumors on day 0, treated with the various CAR T cells on days 4, 9, and 18, and survival was measured (Fig. 2G). Secretion of the A4 VHH by B2 CAR T cells improves survival when compared with treatment with B2 CAR T cells only in this immunocompetent mouse model (Fig. 2H). Although the secretion of anti-CD47 VHH by both B2 and A12 CAR T cells improved outcomes in an aggressive, syngeneic tumor model, tumors eventually relapsed.

To determine whether secretion of (anti-CD47) A4 inhibits expansion and persistence of the CAR T cells, we monitored immune cell populations in the course of treatment (Fig. 3A). Mice were given daily doses of A4 systemically, received A12 CAR T cells in conjunction with daily doses of A4, or received A4-secreting A12 CAR T cells (Fig. 3B). On day 21 after tumor inoculation, lymphoid organs and tumors were harvested and flow cytometry was performed on cells recovered from lymphoid organs. The numbers of CD8⁺ and CD4⁺ CAR T cells in spleen and draining lymph nodes were comparable for mice treated with A12 CAR T cells or with the A12 P2A A4 CAR T cells (Fig. 3C). Secretion of the (anti-CD47) A4 VHH therefore did not affect persistence of the CAR T cells. CD11b⁺ and CD11c⁺ cells showed a slight, although not statistically significant, increase in mice that received the A12 P2A A4 CAR T cells, consistent with greater involvement of the innate immune system (Fig. 3D). For each group of mice, tumors were excised on day 21. One of the tumors extracted from the A12 P2A A4 treatment group showed a loss of melanin, suggesting that epitope spreading had occurred (Fig. 3E). To further examine this possibility, we performed an ELISPOT assay and cocultured splenocytes from each mouse with either B16F10 WT cells or B16F10 PD-L1–KO cells (Fig. 3F): the latter lack PD-L1, the target molecule recognized by the A12 CAR. If epitopes other than PD-L1 were recognized, splenocytes should also respond to the B16F10 PD-L1–KO cells. The ELISPOT results for splenocytes obtained from mice that received the A4-secreting A12 CARs, as well as from mice that received the A12 CARs together with systemic administration of A4 VHH showed reactivity against both the B16F10 and B16F10 PD-L1–KO lines (Fig. 3F and G). This shows that one or more antigens other than PD-L1 are recognized by T cells in the CAR T-cell recipients, consistent with epitope spreading.

An advantage of using VHHs as the secreted output of CAR T cells lies in their stability and solubility. With these properties, VHH-secreting CARs can be made modular. We have generated other VHH-secreting CAR T cells to improve their persistence through checkpoint blockade. We have produced EIIIB-specific CAR T cells that secrete an anti–PD-L1 VHH (A12) with a design similar to that of the A4-secreting CAR T cells (Fig. 4A). To confirm secretion of the A12 VHH, we performed a competition assay similar to that described for the A4-secreting CAR T cells (Supplementary Fig. S3A). T cells transduced with the A12-secreting B2 CAR are not stained by the PD-L1 mAb, showing that secreted A12 blocks recognition by anti–PD-L1. Immunoblotting confirmed release of the A12 VHH (Fig. 4B).

Transduced cell populations, which are a mixture of transduced T cells and unmodified bystander T cells, were stained for the exhaustion markers PD-1, CTLA-4, TIM3, and LAG3. Populations that contain the (anti–PD-L1) A12-secreting CAR T cells show less staining for these markers compared with CAR T cells that do not secrete the A12 VHH (Fig. 4C). To determine whether these in vitro results correlate with improved persistence in vivo, C57BL/6 mice were inoculated with B16F10 tumors and treated with either B2 CAR T cells, A12-secreting B2 CAR T cells, or left untreated (Fig. 4D). The spleens, draining lymph nodes, and tumors of each mouse were harvested after the tumor reached >125 mm², and CAR T cells were enumerated. The A12-secreting B2 CAR T cells showed improved persistence and expansion compared with the B2 CAR T cells, but survival was not significantly impacted (Fig. 4E; Supplementary Fig. S3B). In similar fashion, we generated B2 CAR T cells that secrete an anti–CTLA-4 VHH-Fc fusion (B2-H11Fc; Fig. 4F). Flow cytometry–based competition assays confirmed that the B2-H11Fc CAR T cells indeed secrete H11Fc (Supplementary Fig. S3C), as verified also by immunoblotting (Fig. 4G). Staining of B2-H11Fc CAR T cells for exhaustion markers shows less PD-L1 and PD-1 than on B2 CAR T cells (Fig. 4H). Analysis of the lymphoid organs and tumors of treated tumor-bearing mice showed that the B2-H11Fc–secreting CAR T cells show better persistence of both CD4⁺ and CD8⁺ CAR T cells (Fig. 4I). Administration of the B2-H11Fc–secreting CAR T cells did not significantly improve survival, compared with treatment with a combination of B2 CAR T cells and systemically delivered H11Fc, but was significantly better than no treatment (Supplementary Fig. S3D). Although, the data establish that these CAR T cells secrete anti–CTLA-4 (H11Fc), the amount released may not have been sufficient to provide an additional survival benefit.

We further demonstrated the modularity of this approach by generating CAR T cells that secrete multiple VHHs or VHH fusions. As a proof of concept, we generated A12 CAR T cells that secrete both an anti–CTLA-4 VHH-Fc fusion and an anti-CD47 VHH, encoded by a single vector (Fig. 4J), as validated biochemically and by cytofluorimetry. Anti-CD47 flow cytometry assays show that A4 VHH is indeed secreted (Fig. 4K). An anti-HA IP on the supernatant of cultures confirmed that H11-Fc is secreted as well (Fig. 4L). We did not evaluate the antitumor activity of such double-secreting CAR T cells.

Because the A12 (anti–PD-L1) CAR T cells localize to and can be retained in the tumor environment due to the presence of PD-L1, the cognate antigen recognized by A12, they can produce and serve as a delivery vehicle for biologics. An advantage of such localized delivery is the possible avoidance of side effects caused by therapeutics applied systemically. Secretion of an anti-CD47 VHH by CAR T cells was beneficial in the B16F10 animal model, but CD47 blockade is more effective if there is a signal to drive phagocytosis (18). In the B16F10 model, inclusion of the TA99 antibody, which targets TRP1, provides this signal. However, in the MC38 model and for several other tumor models, such enhancing antibodies are not currently available. Therefore, we generated PD-L1–targeted CAR T cells that secrete an A4-Fc fusion to block CD47 while at the same time providing an activating IgG2a Fc portion to enhance phagocytosis of anti-CD47–decorated targets (Fig. 5A). Anti-human CD47 mAbs have shown promise in preclinical xenograft models, but in syngeneic settings, the widespread expression of CD47 on RBCs in particular has caused complications and prevented further exploration (24). By delivering anti-CD47 VHH-Fc fusions locally to the tumor microenvironment, we might avoid such side effects. Anti–PD-L1 CARs that secrete A4Fc fusions were generated using a single vector containing a fluorescent reporter protein, a CAR linked to a self-cleaving peptide sequence, and a secreted VHH-IgG2a Fc fusion with an HA tag (Fig. 5A). This construct was transduced into T cells and adequately expressed (Fig. 5B). To determine whether the transduced cells secrete (anti-CD47) A4Fc, we performed flow cytometry–based assays. Staining with a fluorescently labeled anti-CD47 that competes for binding with A4 shows that the A4Fc-secreting CAR T cells block binding of the fluorescent anti-CD47 (Fig. 5C). The extent of CD47 blockade is similar to that of cells incubated in the presence of excess soluble A4 VHH. The population of A4Fc-secreting CARs binds both an anti-IgG2a and an anti-HA. The A4-Fc fusion is thus secreted and can bind to CD47 on the T-cell surface (Fig. 5C). Because these cells comprise both transduced and untransduced cells, secreted A4Fc can bind to bystander cells in amounts sufficient to block all available CD47 molecules. This may increase tumor killing efficacy. We further confirmed secretion of A4Fc by immunoblot (Fig. 5D) and detected the presence of a polypeptide of approximately 50 kDa, consistent with the predicted size of the A4-Fc fusion.

To determine whether secretion of A4Fc affects killing by A12 CAR T cells, we performed a coculture assay with B16F10 cells. Killing of B16F10 and production of IFNγ by A12-A4Fc and A12 CAR T cells was comparable (Fig. 5E). To address possible toxicity of anti-CD47 treatment, immunocompetent mice received either systemic (anti-CD47) A4Fc (Supplementary Fig. S3), (anti–PD-L1) A12 CAR T cells, (anti–PD-L1) A12 CAR T cells together with systemic (anti-CD47) A4Fc, (anti-CD47) A4Fc-secreting (anti–PD-L1) A12 CAR T cells, or no treatment at all. We measured body weight to gauge the toxicity of each treatment (Fig. 5F). Systemic delivery of A4Fc or A12 CAR T cells together with systemic A4Fc led to rapid weight loss (Fig. 5G). However, treatment with the A4Fc-secreting A12 CAR T cell showed no significant weight loss, comparable with untreated mice. Mice from each group were bled 3 days after treatment, and a complete blood count was done. Both groups that received systemic A4Fc injections showed a drop in RBC numbers and hemoglobin levels, but the group with CAR T-cell–mediated delivery of A4Fc did not (Fig. 5H and I). A large fraction, around 80%, of circulating RBCs were decorated with systemically delivered A4Fc, whereas circulating RBCs from mice treated with the A4Fc-secreting CAR T cells showed no such binding (Fig. 5J). To determine whether A4Fc-secreting A12 CAR T cells provide an antitumor benefit, we examined the MC38 tumor model (Fig. 5K). We inoculated mice with MC38 tumors and treated the recipients with either A12 CAR T cells, A12-A4Fc CAR T cells, or left mice untreated. We measured tumor sizes every 1–2 days with calipers. A concern was the possibility that binding of the A4-Fc to the CAR T cells themselves could lead to early elimination and thus compromise their therapeutic efficacy. However, upon analysis of the of lymphoid organs and tumors at the endpoint of the MC38 experiments, the A12-A4Fc CAR T cells persisted, although to a slightly lesser, not statistically significant, degree than the A12 CAR T cells (Supplementary Fig. S4). Phagocytosis likely requires a balance of signals, and blockade of CD47 alone may not be sufficient to ensure elimination (31, 32). We saw an improvement in survival of the MC38-inoculated mice upon treatment with the A12-A4Fc CAR T cells in comparison with no treatment or treatment with A12 CARs (Fig. 5L and M). Within the tumors in the A12-A4Fc–treated group, the number of CD11b⁺CD11c⁺ cells showed an increase, one of the indicators of the presence of M1-like macrophages (Fig. 5N; refs. 33, 34). We also saw slightly more CD11c⁺ cells, as well as significantly more CD69⁺ T cells and CD45⁺ cells in the groups treated with A12-A4Fc CAR T cells (Fig. 5N). These observations suggest that local secretion of the A4Fc generates a more inflammatory environment within the tumor, improving the activation of immune cells.

CAR T-cell therapies have shown success as single-agent therapeutics in hematologic cancers, but an immunosuppressive environment poses a challenge for their use in the treatment of solid tumors. Antibody-based approaches, alone or in combination with other treatment modalities can establish a favorable antitumor environment. However, each additional component may also cause side effects and complicate production and implementation. Because CAR T cells are living therapies that can be programmed to execute different functions, we developed CAR T cells that can secrete immune-modulating VHHs and VHH fusions to improve their therapeutic efficacy in immunocompetent animals.

VHHs are single-domain antibodies that can perform a wide range of functions depending on their targets. Because of their small size and stability, they can be produced at high yields and are readily secreted in functional form. Because VHHs are derived from immunoglobulin heavy chain–only antibodies, they can traverse the secretory pathway without the need for additional partner proteins. Where simultaneous production of multiple single-chain variable fragments (scFv) in a single cell can compromise their assembly and secretion through domain swaps between the V_H and V_L portions of the different scFvs, VHHs have no such drawback (35). This facilitates the generation of CAR T cells that make use of more than a single VHH. CAR T cells serve as ideal vehicles for VHH delivery through local secretion, so that VHHs target and accumulate in the tumor microenvironment. This strategy not only limits potential adverse effects due to off-tumor–targeting caused by systemic exposure, but also produces a higher local concentration of VHH, which should be helpful where complete blockade of a target in the tumor microenvironment is required.

Blockade of CD47 can increase phagocytosis of antibody-decorated targets by macrophages and other myeloid cells (21). However, the ubiquitous expression of CD47 on self-cells causes off-tumor side effects that are dose-limiting for clinical application of anti-CD47 treatment. Dose-escalation studies have shown that treatment-induced anemia is of particular concern and prevents attaining the requisite dose of anti-CD47 needed for increased tumor killing in some cases (36). We show it is possible to confine the presence of the agent mostly to the tumor environment and thus prevent potential systemic toxicities by CAR T-cell–mediated local delivery of anti-CD47 VHHs or anti-CD47 VHH-Fc fusions. These studies are performed in immunocompetent hosts, which should more accurately represent systemic toxicities, if any, than the frequently used immunocompromised models. Because phagocytosis by macrophages depends on a balance between stimulatory and inhibitory signals, delivery of the anti-CD47 VHH without an Fc portion requires an additional cue for phagocytosis (31, 32). In the case of the B16F10 melanoma model, the TRP-1–specific TA99 mAb can provide that phagocytic stimulus (37). However, in less well-characterized models, or in models where such antibodies are not readily available, it would be beneficial to have a molecule that can simultaneously provide a stimulatory signal and block inhibitory interactions. We therefore generated CAR T cells that secrete anti-CD47 VHH-Fc fusions. Treatment with A4Fc-secreting A12 CAR T cells shows an accumulation of CD11c⁺ macrophages within the tumor, suggesting an accumulation of proinflammatory M1-type macrophages (33, 34). An increase in CD69⁺ CD45⁺ cells also indicates enhanced leukocyte activation with inflammatory activity. Secretion of an anti-CD47 VHH or VHH-Fc fusion by CAR T cells thus inhibits tumor growth, while at the same time avoiding the toxicities associated with systemic anti-CD47 treatment. However, we believe that engineering CAR T cells to secrete the anti-CD47 VHH, paired with administration of a separate Fc engagement molecule is preferable. This solution avoids the potential risk of increased clearance of VHH-Fc–secreting T cells, as CD47 is present on the T-cell surface.

Different tumor indications have diverse immune characteristics, with variable immune infiltration, immunosuppressive capabilities, and infiltrating cell types. An ideal CAR T-cell therapy should be designed to tackle particular aspects of each tumor, with modifications that should be easy to engineer. To demonstrate the flexibility of VHH-secreting CAR T cells, we generated several additional VHH-secreting CAR T cells based on a standard vector design. In each case, we find that VHHs are secreted effectively. For both the PD-L1 VHH- and CTLA-4 VHH-Fc fusion–secreting CAR T cells, cells appear less exhausted in the course of their generation. In vivo, this translates to improved persistence and survival of the CAR T cells. Tumor-bearing mice treated with the A12-secreting B2 CAR T cells and the H11Fc-secreting B2 CAR T cells show more CAR T cells in the spleen, draining lymph nodes, and tumors when compared with treatment with B2 CAR T cells. A single retroviral vector can encode for the secretion of multiple VHHs and VHH-fusions from the same CAR T cell.

In conclusion, we have shown that secretion of VHHs by CAR T cells can improve their antitumor activity in immunocompetent hosts with syngeneic tumor models. Treatment with an anti-CD47 VHH-secreting CAR T cell enhanced survival over treatment with systemically delivered anti-CD47 VHH along with a CAR T cell. Localized secretion of VHHs and VHH-Fc fusions by CAR T cells results in an improved safety profile when compared with systemic administration of these antibody derivatives and can contribute to improved persistence of CAR T cells. Personalized, multi-modal treatments tailored toward the characteristics of each tumor indication thus become possible.

M. Dougan is a consultant for Tillots Pharma and Partner Therapeutics, is a scientific advisory board member for Neoleukin Therapeutics, and reports receiving a commercial research grant from Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y.J. Xie, M. Dougan, J.R. Ingram, H.L. Ploegh

Development of methodology: Y.J. Xie, T. Fang, H.L. Ploegh

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y.J. Xie, N. Pishesha

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.J. Xie, M. Dougan

Writing, review, and/or revision of the manuscript: Y.J. Xie, M. Dougan, H.L. Ploegh

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.J. Xie, N. Momin

Study supervision: Y.J. Xie

Other (made essential contributions to the generation and characterization of the nanobodies and the conceptualization of the VHH-based CAR T cells that were used in this study): J.R. Ingram

This research was supported by the Lustgarten Foundation Grant 80939 (to H.L. Ploegh) and the Melanoma Research Alliance (to H.L. Ploegh). Y.J. Xie was supported by a National Science Foundation Graduate Research Fellowship. M. Dougan was supported by the NIH Mentored Clinical Scientist Development Award 1K08DK114563-01 and the American Gastroenterological Association Research Scholars Award. N. Pishesha was supported by the Harvard Junior Fellowship. We thank Andrew Woodham for technical assistance with the ELISPOT assay. We thank Laura Kummer for her technical assistance in amplifying viral DNA. We thank Richard Hynes for helpful discussion in work regarding the B2 CAR T cell. We are grateful to Steve Kolifrath for his assistance in animal care. We thank Nicholas McCaul for his assistance in revisional experiments. We thank the flow cytometry core at Boston Children's Hospital (BCH) Karp. We are grateful to the Lodish laboratory for providing the XZ vector. We thank Dane Wittrup for providing the TA99.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.