Intratumoral Delivery of Chimeric Antigen Receptor T Cells Targeting CD133 Effectively Treats Brain Metastases

This article is featured in Highlights of This Issue, p. 469

Brain metastases (BM) account for 90% of all brain malignancies yet remain largely incurable. Clinically, BM are mainly treated palliatively, with survival past the 1-year mark being rare (1). Chemotherapy is rarely utilized for BM treatment due to its limitation to effectively cross the blood–brain barrier (BBB), which has also historically led patients with BM to be excluded from clinical trials that use small-molecule agents and more recently, immunotherapy with immune checkpoint inhibitors (2). Recent efforts into the identification of the molecular drivers of primary cancers has led to the attempted repurposing of primary tumor targeted agents for BM, but these have limited BBB permeability and efficacy against BM (3), highlighting the need to identify BM-specific drivers for therapeutic development.

It has been shown that only approximately 0.01% of metastasizing primary tumor cells are capable of initiating and sustaining the growth of a distinct, secondary tumor (4). This cell population is theorized to have inherent stem-like properties that allow it to drive malignant tumor growth, contribute to drug resistance, and drive relapse (5–8). In the early 2000s, the human neural stem cell marker glycoprotein CD133 was identified as a marker for cancer stem cells (CSC) that are enriched in brain tumors (9, 10). These CD133⁺ cells, termed brain tumor-initiating cells, have been shown to successfully propagate brain tumor growth both in vitro and in vivo, unlike their CD133⁻ non-CSC counterparts, providing additional support for the CSC hypothesis in brain cancer. More recently, a select subpopulation of CSCs capable of initiating BM has been reported, termed brain metastasis initiating cells (BMIC), providing further support for the presence of metastatic CSCs (11). Whereas the cells that comprise the bulk of a neoplasm are generally variable in their proliferative, differentiation, and self-renewal capabilities, BMICs hold the greatest potential to evade conventional therapies, migrate away from their primary tumors and home to the brain to form secondary lesions; therefore, targeting BMICs could lead to the attenuation of BM. In fact, CD133⁺ tumorigenic CSC fractions isolated from primary cancers have both been shown to be virtually unaffected by standard-of-care cancer therapies (12, 13). Similar limited effectiveness against CSCs by conventional chemotherapy and radiotherapy has been proven in a variety of models (6, 12, 14–16).

High expression of CD133 correlates with disease progression, recurrence, chemoresistance and radioresistance, and overall poor prognosis (17). We recently reported that adult glioblastoma responds well to targeted immunotherapeutic chimeric antigen receptor (CAR) T-cell against CD133 (CART133; ref. 18). Here, we apply a CART133 modality for the treatment of CD133-expressing BM using a clinically relevant model of brain metastases (Fig. 1A); immunocompromised NSG mice were engrafted with patient-derived BM tumor cells and the T-cell therapy was delivered via intratumoral route leading to an effective preclinical response after a single dose.

BM cell lines from primary lung (BT530), breast (MBT382), colon (BT750), and melanoma (BT673) cancers were derived from primary patient samples with written consent from the patients and approved by the Hamilton Health Sciences McMaster Health Sciences Research Ethics Board (REB #16-078), in compliance with Canada's Tri-Council Policy Statement on the Ethical Conduct for Research Involving Humans and International Ethical guidelines for Biomedical Research Involving Human Subjects. Cells were grown in the presence of MycoZap Mycoplasma Elimination Reagent (Lonza #LT07-818) for 2 weeks following harvest. Cell lines were minimally cultured (<10 passages) and maintained in NeuroCult Complete media consisting of NeuroCult NS-A Basal Medium (Stemcell Technology #05750) and supplemented with 50 mL of NeuroCult Supplement, 20 ng/mL EGF, 10 ng/mL FGF, 0.1% heparin, and 1% antibiotic-antimycotic solution. All cell lines were maintained at 37°C with a humidified atmosphere of 5% CO₂ and authenticated periodically by short tandem repeat analysis.

Levels of CD133 (Miltenyi Biotec, catalog no. 130-113-184) on patient-derived BM lines were analyzed using flow cytometry. The gating strategy to determine expression of CD133 was set up using an isotype control (Miltenyi Biotec, catalog no. 130-113-446). Forward scatter (FSC)/side scatter gate was used to determine the select cell population of which singlets were selected using FSC-W versus FSC-H. Viability dye 7-aminoactinomycin D (1:100 dilution, Invitrogen catalog no. 00-6993-50) was used to stain for live cells of which the expression of CD133 was analyzed. For the limiting dilution studies, cells were sorted on the basis of CD133 positivity into 96-well plates.

Limiting dilution analysis of BT530 at day 7 was performed using extreme limiting dilution analysis that implements a generalized linear model approach to the limiting dilution assay (19). Multiple group analysis was performed using a 95% confidence interval. The stem cell frequency for each group with the 95% confidence interval was calculated and tests for inequality in frequency between multiple groups were performed.

The lentivirus (LV) plasmid encoding the binder and CAR used here largely reflect our previous publication (18) with slight modifications. We utilized a modified single-chain variable fragment substituting a single amino acid substitution to avoid a potential glycosylation site on the CAR. We also altered the CAR's structural and signaling components to: a hinge comprising a flexible GS linker and human CD8a, the transmembrane and endodomain from human CD28, and the same CD3z endodomain sequence. Finally, instead of relying on a secondary promoter-driven open reading frame to infer CAR expression, these sequences were removed and CAR expression was measured directly via GFP expression.

LV for T-cell transduction was produced using standard best practices. Briefly, 80% confluent adherent HEK293T cells were transfected using fourth-generation LV production plasmids (all Addgene: Gag/Pol #12252, REV #12253, VSVg #12259) and the CAR-expressing transfer plasmid at a 2:1:1:4 ratio using polyethyleneimine. At 24-, 48-, and 72-hour marks following transfection, media was harvested, stored at 4°C, and fresh media (DMEM containing 2 mmol/L glutamine and 1 mmol/L sodium pyruvate with 10% FBS, 10 mmol/L HEPES, and 1.0 mmol/L sodium butyrate) was added. After the 72-hour harvest, all media was pooled, centrifuged at 1,000 × g for 10 minutes to remove debris, sterile-filtered using 0.2 μm SteriFlip (Millipore-Sigma), and concentrated using ultra-centrifugation (4 hours at 20,000 × g and 4°C). Finally, concentrated LV pellets were resuspended in OptiMEM, aliquoted into single-use vials, and stored at −80°C. LV formulations were functionally titrated by serially transducing HEK293T cells at 10 ratios and measuring CAR expression 4 days later; one or more ratios at 10%–30% transduction was then used to calculate titers for each LV batch.

Peripheral blood mononuclear cells (PBMC) from consenting healthy donors were obtained from whole blood using SepMate (StemCell Technologies) according to the McMaster Health Sciences Research Ethics Board. Next, T cells were isolated from frozen PBMCs using CD4 and CD8 magnetic beads (Miltenyi Biotec). These cells were then cultured in PRIME-XV T Cell Expansion XSFM media (Irvine Scientific, FujuFilm) with 100 U/mL hIL2 and stimulated with TransAct (Miltenyi Biotec) according to the following ratio: 1.2 million cells in 990 μL media with 10.0 μL TransAct. A total of 24 hours later, cells were transduced with CD133-CAR LV at multiplicity of infection =3 by simply adding appropriate volumes of concentrated LV formulations, incubated overnight, and then given 500 μL more complete media. The next day, cells were transferred to T175 flasks and maintained at 1–2 million/mL by adding complete media every other day according to a predetermined schedule. CAR-T cells were harvested on day 9, counted, viability checked, and transduction was analyzed by flow cytometry detection of GFP. Transduction efficiency ranged between 45% and 55% and experiments were conducted after normalizing for specific CAR expression.

Cytotoxicity, activation, proliferation, and cytokine release assays were performed as described previously (18, 20). For cytoxicity assays, luciferase-expressing BM cells were plated at a concentration of 10,000 cells per well in triplicates with D-firefly luciferin potassium salt (100 mg/mL) to ensure equal distribution of target cells (measured with a luminometer, Omega) before adding T effector cells at 1:1, 0.5:1, 0.25:1, and 0:1 effector:target ratios. Cells were incubated at 37°C with a humidified atmosphere of 5% CO₂ for 10–24 hours depending on the target cell doubling time. For proliferation, activation, and cytokine release assays, effector cells were incubated with BM cells at a 1:1 ratio in triplicate. After 24 hours, supernatants were collected and analyzed for human TNFα (DuoSet ELISA kit, catalog no. DY210-05) and Human IFNγ (DuoSet ELISA kit, catalog no. DY285B-05) while cells were collected and either analyzed for activation markers CD25 (Miltenyi Biotec, catalog no. 130-113-283) and CD69 (Miltenyi Biotec, catalog no. 555533) by flow cytometry or sorted (CD3⁺ T cells) into 96-well plates for 72 hours of incubation and subsequent proliferation analysis using the Presto Blue assay.

A lentiviral vector expressing Firefly Luciferase (Addgene, RRID:Addgene_118017) was used for this study. Replication-incompetent LV was produced by cotransfection of the Firefly Luciferase vector and packing vectors pMD2.G and psPAX2 in HEK293T cells at approximately 80% confluency using Lipofectamine 3000 reagent (Thermo Fisher Scientific) as per manufacturer's instructions. Viral supernatant was harvested every 24 hours for a total of 3 days and concentrated by PEGit (System Biosciences) as per manufacturer's instructions. The viral pellet was resuspended in 1.0 mL of DMEM, aliquoted, and stored at −80°C. BMIC lines were transduced with lentiviral vectors and treated with puromycin after 48 hours of transduction as a selection marker to develop stable cell lines.

All experimental procedures involving animal work has been reviewed and approved by McMaster University Animal Research Ethics Board (AREB). Non-obese diabetic-severe combined immunodeficient IL2rγnull (NSG) mice were used for all experiments. An even number of male and female mice were used for these studies. Mice were anesthetized by gas anesthesia using isoflurane (4% induction, 2.5% maintenance) before procedure. Cells were engineered to express firefly luciferase and were injected intracranially as described previously (18, 21–25). Briefly, each mouse was injected with 1 million luciferase-tagged BM cells suspended in 5 μL PBS with a Hamilton syringe (Hamilton, catalog no. 7635-01) into right frontal lobes of 6–8 weeks old NSG mice. Following successful tumor engraftment 4 days after tumor cell injection, as determined by in vivo bioluminescent imaging, 1 million untransduced CAR-T cells or CART133 cells were injected into the tumor injection site at a single dose. To evaluate tumor volume, a CART133-treated mouse was sacrificed when a control mouse reached humane endpoint. To evaluate survival advantage, a cohort of CART133-treated mice was kept until they either reached humane endpoint or until the experiment reached completion (90 days after injection) for cured mice. Culled mice were overdosed with intraperitoneal injection of tribromoethanol (Avertin) and perfused with 10% formalin, and brain was collected for IHC.

Bioluminescent imaging was performed weekly using an IVIS Spectrum In Vivo Imaging System (PerkinElmer) as per the manufacturer's instructions. Quantification of bioluminescent signals was performed using the analysis software Living Image (Xenogen). Mice were injected intraperitoneally with 10 μL/g of 15 mg/mL solution of d-Luciferin firefly solution (PerkinElmer) in PBS (Invitrogen) before being imaged, and anesthetized (4% induction, 2.5% maintenance isoflurane). Mice were then placed onto a warmed stage inside the instrument and imaged for a maximum of 5 minutes depending on the tumor size.

Brains were sliced at 2 mm thickness using brain-slicing matrix for paraffin embedding and hematoxylin and eosin (H&E) staining or COXIV staining for human cells, as described previously (18). Images were captured using an Aperio Slide Scanner (Leica Biosystems) and analyzed using ImageScope v11.1.2.760 software (Aperio). In a neurotoxicity study where an experimental neuropathologist was blinded to the treatment of mice, brain slices were stained with Luxol fast blue with H&E counterstain (LFB+H&E) and analyzed under a Nikon 50i Eclipse light microscope.

Replicates from at minimum three samples are used for all applicable experiments for mean comparisons. Data collected from respective in vitro experiments are represented using GraphPad Prism 8 software. Student t tests and two-way ANOVA analyses are conducted using the same software, with a P value < 0.05 deemed as statistically significant. For in vivo studies, medium survival differences were measured using Kaplan–Meier survival curves and significance determined by the log-rank test.

All animal experiments were performed in accordance with the Canadian Council on Animal Care under animal utilization protocol (19-01-01) approved by the AREB. Human tissues were isolated using protocols approved by the Human Integrated Research Ethics Board.

The data generated in this study are available within the article and its Supplementary Data, and the raw data are available upon request from the corresponding author.

We evaluated 35 in-house patient-derived BM cell lines for their surface CD133 antigen expression (Supplementary Fig. S1a; Supplementary Table S1). Our analysis showed variable expression of CD133 both within and among different primary tumor cohorts. To identify the clinical significance of CD133 expression on patient prognosis, we sorted the patients into CD133^low and CD133^high cohorts based on the median expression of 3.41% and found that patients in the CD133^high cohorts had a significantly lower survival time after BM surgical resection (Fig. 1B and C). In many solid cancers, CD133 expression has been correlated to CSC enrichment and decreased survival (26). In addition, CSCs and CD133⁺ cell populations within tumors have been shown to be resistant to chemotherapies and radiotherapies (12, 13). On the basis of these data, we reasoned that the clinical utility of CD133 as a diagnostic marker in patients with BM should be explored.

We selected BT530 (lung-BM), BT750 (colon-BM), and MBT382 (breast-BM) as appropriate models to pursue for this study based on their variable CD133 positivity (Fig. 1D; Supplementary Fig. S2). FACS of CD133⁺ and CD133⁻ populations and assessing them for sphere-forming capability (a surrogate measure of self-renewal) showed that CD133⁺ fractions have a significantly higher ability to form spheres compared with CD133⁻ fractions (Fig. 1E). Patient-derived xenografts confirm the presence of CD133 in vivo (Fig. 1F). We used BT673 (melanoma-BM), a cell line that does not express surface CD133, as a negative control moving forward (Supplementary Fig. S1B and S1C).

We repeated sphere formation for BT530 (lung-BM) in an enhanced limiting dilution analysis, which allows for the determination of the minimal frequency of repopulating tumor sphere cells within the cell population (19). The number of cells required to generate at least one tumor sphere per well was calculated as 1.61 in the CD133^high-expressing cells and 25.02 in CD133^low-expressing cells (Supplementary Table S2; Supplementary Fig. S3), indicating that the frequency at which one tumor sphere cell will proliferate to form a new tumor sphere varies according to CD133 expression, with high CD133 expression exhibiting a statistically significant increased self-renewal capacity. These data show that the CD133 expressing subpopulation of cells have a higher capacity to self-renew and provides further rationale to explore targeting CD133 in the context of BM.

A variant of our previously described CD133 binder (18) was used to generate second-generation GFP+ CART133 cells for this study (Fig. 2A). Untransduced T cells from the same healthy donor (herein referred to as Untd-T cells) were used as controls. GFP was used to confirm the efficiency of transduction and surface expression of the CAR targeting CD133 to be approximately 45% (Fig. 2B).

To evaluate the antigen-specific cytotoxicity of CART133 cells against CD133-expressing BM lines, we performed a luciferase-based cytotoxicity assay (20) to assess the specific lysis of CD133-expressing BM cells by CART133 as a direct measure of the specific killing of CD133⁺ cells. Our results show a dose-dependent response of CD133-expressing BM cell lysis in the presence of CART133 cells, whereas Untd-T cells did not show cytotoxic activity toward BM cells (Fig. 2C). Notably, the CD133-lacking cell line, BT673 (melanoma-BM), was unaffected by both CART133 and Untd-T cell controls. These results confirm that CART133 cells show specific activity against CD133-expressing BM lines.

To further assess the functionality of CART133 cells against our CD133-expressing BM lines, we cocultured CART133 or Untd T-cells with either CD133-expressing or CD133-lacking BM lines at a 1:1 ratio for 24 hours (20). CART133 cells cocultured with CD133-expressing BM lines (BT530, lung-BM; BT750, colon-BM; MBT382, breast-BM), but not the CD133-lacking BM line (BT673, melanoma-BM), were highly proliferative compared with Untd-T cells (Fig. 3A) and showed elevated expression levels of the activation markers CD25 and CD69 by flow cytometry compared with control Untd T-cells (Fig. 3B). Finally, the release of cytokines TNFα and IFNγ was significantly upregulated in the supernatants of the coculture, as determined by an ELISA (Fig. 3C and D). Together, the data obtained from coculture experiments suggest an antigen-specific CART133 response upon exposure to CD133-expressing BM cells, supporting the feasibility of CART therapeutic development for CD133-expressing BM.

We next assessed the in vivo activity of CART133 cells in tumor-bearing mice. We decided to pursue the BT530 lung-BM and BT750 colon-BM models in this preclinical study due to the relative ease of achieving the cell numbers required for in vivo studies. Briefly, CD133-expressing BM cells stably expressing firefly luciferase were intracranially xenografted into immunocompromised mice (Fig. 4A,⁵). After tumor formation was confirmed by in vivo imaging 4 days after tumor cell injection, Untd T-cells or CART133 cells (n = 6 per cohort for each cell line) were delivered in a single dose intratumorally. This type of local or regional route of administration circumvents concerns of “on-target off-tumor” toxicity to reduce systematic targeting of antigens present on normal tissues and has demonstrated efficacy and safety following local delivery for central nervous system tumors both in preclinically mouse models (18, 27) and human trials (28–30). This type of delivery method also addresses the problem of suboptimal CAR-T cell trafficking since the CAR-T cells are directly administered into the tumor space.

Mice treated with CART133 were showing significant tumor reduction after only single dose in both models used (Fig. 4B and C), and a significant survival advantage determined by Kaplan–Meier analysis (Fig. 4C and D). A single dose of CART133 cells exhibited a striking efficacy and potency against BM tumor cells despite the variability of CD133 protein expression in these cell lines (see Fig.1). In the BT750 (colon-BM) model, complete eradication of the tumor cells was observed after a single dose and the mice did not exhibit any relapse at the experimental endpoint. The significant difference in tumor formation can also be visualized by H&E staining, where mice treated with CART133 were not yet showing significant tumor burden when mice in the control cohort reached humane endpoint (Fig. 5A and B; Supplementary Fig. S4A, S4B, S5A, and S5B). Moreover, the mice injected with BT750 (colon-BM) cells that exhibited a “cure” by bioluminescent imaging were sacrificed when the experiment ended, and their brains stained for H&E (Supplementary Fig. S4C and S5C) the human marker COXIV by IHC showed no trace of human cells left in the brain after 3 months following a single dose of CART133 (Fig. 5C). Importantly, as CD133 is a marker for neural stem cells, we investigated any potential damage to healthy neural stem cells following CART133 therapy and did not detect any signs of toxicity (Supplementary Table S3).

With autopsy studies suggesting that up to 40% of patients with cancer develop BM over the course of their disease, there is an urgent need for more effective therapeutic strategies for patients who present with established BM. The glycoprotein CD133 has been linked to tumor relapse of several primary cancers that contribute to BM; however, there are no reports of anti-CD133 therapies for this disease. Our group has previously shown that CART133 cells present a therapeutic strategy to target the CD133⁺ BTIC population that drives glioblastoma recurrence and therapy resistance (18). In this follow-up study, we expand the utility of a CART133 immunotherapy for a metastatic brain tumor that is currently mainly treated palliatively.

CD133⁺ cells represent a stem-like population of cells within the bulk of a tumor that are functionally relevant in driving the formation and/or relapse of a tumor, and according to the CSC hypothesis, sit at the apex of the tumor cell hierarchy (9, 10, 18). Recent CAR-T therapy trials for malignant brain tumors have shown promising results, yet in all of these studies, recurrence was associated with outgrowth of antigen-negative tumor cells (28, 30, 31). Because of the well-known fact that heterogeneous cell lineages within solid tumors and cell lines may respond variably to therapies and often compensate for signaling pathway-specific therapy by upregulating a different signaling pathway, a multitargeted approach to cancer therapy is warranted because it is likely not feasible for one drug to target all the signaling pathways and interaction networks present in BM, or malignant tumor growth in general. Nonetheless, we have shown here a striking eradication of brain tumor cells in vivo after a single dose of CART133 therapy, with complete remission in one of two models, suggesting that targeting the important stem-like cell population that is ultimately responsible for tumor formation and maintenance.

In the BT530 (lung-BM) model, CART133-treated mice that were sacrificed as time-match culls for control mice succumbing to brain tumor burden did not have detectable brain lesions by H&E staining. However, the mice in this model eventually relapsed and succumbed to their brain tumor burden, suggesting that CAR-T therapies should be administered adaptively because tumors may need a variable number of doses before remission can be achieved. In the BT750 (colon-BM) model, CART133-treated mice reached complete remission after a single dose of CART133, despite the cell line's low CD133 antigen expression as determined by FACs (see Fig. 1D). One plausible explanation for this is that the CD133 expression in this cell line appears to be elevated in vivo (see Fig. 1F): this could be explained by the significantly lower ability of CD133⁻ cells to form spheres compared with their CD133⁺ counterparts (see Fig. 1E), which is an in vitro surrogate property for self-renewal. Furthermore, the therapy did not show any signs of toxicity in mice (see Supplementary Table S3), confirming the data shown by Vora and colleagues who assessed the safety profile of the CART133 in a humanized model of hematopoiesis to show that intracranial and intravenous CART133 treatment is safe (18). In this study, we also show that the CART133 cells are very specific toward cells that are positive for CD133 because they do not show any activity in cells that do not express CD133 (see Figs. 2 and 3). These data add to the growing literature that eradication of the CD133⁺ subpopulation of cells within a tumor, regardless of its expression level, can halt tumor growth because this is the cell population that sits at the apex of the tumor hierarchy and is ultimately responsible for tumor formation (12, 32, 33).

Brain permeability can be a major hurdle when it comes to developing targeted therapies for brain tumors, and concerns of “on-target off-tumor” toxicity are derived from the potential systematic targeting of antigens present on normal tissues. We circumvented these concerns by administering the T cells directly into the site of the brain tumor rather than systemically, because the placement of an intracerebroventricular shunt after surgery is standard practice in the clinic and often used to administer therapies thereafter to bypass the BBB (34). However, it is well known that BM are often not limited to a single lesion, and future directions should explore systemic injection of CART133 cell therapy be better to target the lesions that are in other parts of the brain.

In summary, the data shown here suggest that CD133 is an important diagnostic marker for BM and should be exploited to slow the growth or eradicate BM independent of antigen expression levels. By targeting the critical cell population within the tumor that is responsible for its outgrowth and therapy resistance, CART133 therapy can offer an effective way to transform a palliative standard-of-care into one that is focused on extending patient survival.

C. Venugopal reports a patent for CD133 binding agents and uses thereof issued. S.K. Singh reports grants from Century Therapeutics Inc. during the conduct of the study; in addition, S.K. Singh has a patent for CD133 therapeutic targeting licensed. No disclosures were reported by the other authors.

A.M. Kieliszek: Conceptualization, data curation, formal analysis, validation, visualization, methodology, writing–original draft, writing–review and editing. D. Mobilio: Data curation, writing–review and editing. D. Upreti: Conceptualization, resources, writing–review and editing. D. Bloemberg: Resources, data curation, writing–review and editing. L. Escudero: Formal analysis, methodology, writing–review and editing. J.M. Kwiecien: Data curation, formal analysis, writing–review and editing. Z. Alizada: Project administration. K. Zhai: Conceptualization, writing–review and editing. P. Ang: Data curation. S.C. Chafe: Conceptualization, writing–review and editing. P. Vora: Conceptualization, supervision, writing–review and editing. C. Venugopal: Conceptualization, writing–review and editing. S.K. Singh: Supervision, writing–original draft, writing–review and editing.

We thank Minomi Subapanditha for performing the flow cytometry analyses shown in this study. We thank Blessing Bassey-Archibong, Emily Ford, and Mohini Singh for contributing to the sample CD133 expression analysis presented in Supplementary Fig. S1.

S.K. Singh holds a Senior Canada Research Chair in Human Cancer Stem Cell Biology. This work was funded by Canadian Cancer Society Research Institute (CCSRI), donations from Boris Family and McMaster University Department of Surgery, and the Ontario Institute Cancer Research. A.M. Kieliszek was supported by the Ontario Graduate Fellowship and the MITACS fellowship. D. Mobilio was supported by the Brain Tumor Foundation of Canada Summer Scholarship (Taite Boomer Foundation).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.