Local Intracerebral Immunomodulation Using Interleukin-Expressing Mesenchymal Stem Cells in Glioblastoma

Glioblastoma belongs to the most aggressive brain tumor types and remains virtually incurable so far. Despite surgical resection followed by combined radio-chemotherapy the mean survival time is 15 months (1). Extensive diffuse invasion of the brain parenchyma, ineffective delivery of drugs across the blood–brain barrier and the genetic and epigenetic versatility of glioma cells are the major reasons for treatment failure and have complicated further therapeutic developments.

Another hallmark of malignant glioma is their immunosuppressive microenvironment, which by multiple soluble- and cell-mediated mechanisms impedes an effective tumor-specific immune response (2–4). Peripheral T cells are able to recognize tumor-specific antigens and infiltrate the tumor mass (5, 6). However, a meaningful execution of glioma cell lysis and effective perpetuation of a tumor-specific immune response is severely impaired in glioblastoma. Our group and others have shown that tumor-infiltrating lymphocytes in glioblastoma display distinct immune exhaustion profiles (7, 8), reflecting the glioma-induced immunosuppressive environment. Therefore, counteracting the local immunosuppression of malignant glioma and promoting a proinflammatory microenvironment is a promising strategy to overcome this immune exhaustion and to promote a sustained tumor-cell directed cytotoxic T-cell response. Recent immunotherapeutic treatment approaches in other cancer types have already demonstrated that a successfully activated immune system has the potential to result in long-term tumor remissions even in the metastatic setting (9).

Mesenchymal stromal-/stem cells (MSCs) are multipotent, fibroblast-like cells first described by Friedenstein and colleagues (10). They can be isolated from various tissues including bone marrow, cord blood, and fat and easily expanded in culture (11). MSCs have a low immunogenic profile, which allows the generation of cell products for allogeneic application (12). Furthermore, MSCs can easily be genetically modified using viral vectors to expand their efficacy by the introduction of therapeutic genes. As MSCs display an extensive tumor tropism, they can be used as drug delivery vehicles to achieve high local concentrations without systemic side effects (13). Previous in vivo studies have demonstrated that MSCs target infiltrating glioblastoma xenografts similarly as neural stem cells and are able to successfully deliver their therapeutic cargo (14, 15). In this study, tumor-targeting MSCs were genetically modified to express the combination of the potent proinflammatory cytokines IL7 and IL12 to specifically modulate the local tumor environment of invasive glioblastoma and to promote a naturally derived tumor-specific T-cell response.

IL7 is crucial for T-cell homeostasis and can be used to reconstitute the immune system (16, 17). Promising data with recombinant IL7 have been generated in early clinical analysis, showing an increase in effector versus regulatory T cells (18), as well as an increase in T-cell receptor repertoire (19). Those effects could be harnessed for anticancer immunotherapy but IL7 might not be potent enough on its own. IL12 can activate the adaptive (T cells) as well as the innate [natural killer (NK) cells] arm of the immune system, rendering it a very potent immunostimulatory agent with the potential to enhance the effects of IL7. One of its main effector molecules is IFNγ and it induces polarization of CD4⁺ T cells to the T_H1 type (20). IL12 has shown promising results in preclinical models but its success in the clinic has been hampered by short half-life and severe systemic toxicity, making novel, more localized approaches mandatory (20–22). Therefore, MSCs are ideal candidates for the local delivery of IL12 and IL7 in glioblastoma as they target invasive tumor cells and therefore provide an optimized distribution for effective local immunomodulation while avoiding systemic side effects.

In this study, we assessed the therapeutic effects of IL7- and IL12-expressing MSCs (apceth-301, MSC^IL7/12) in immunocompetent glioblastoma mouse models. Immunomodulatory effects were monitored by comprehensively profiling immune activation and exhaustion of tumor-infiltrating lymphocytes (TILs) and other immune populations. This study demonstrates the first preclinical results for an upcoming phase I/II study of apceth-301 (MSC^IL7/12) in recurrent glioblastoma and provides the basis for MSCs as a potential platform for the local immunomodulation in malignant brain tumors.

MSCs were isolated from bone marrow of healthy donors. Bone marrow was seeded into five-layer culture stacks (Corning) and cultivated in apceth's proprietary xeno-free Medium Bio-1. Once MSC colonies became visible, MSCs were harvested and cryopreserved (stock). Cryopreservation medium consisted of human serum albumin (Baxter), Hydroxyethyl starch (Fresenius Kabi) with 10% DMSO. Generation of transduces and native MSC batches occurred from these stocks. Cryopreserved MSC stocks were taken into culture in Bio-1 medium at a density of 1,000 to 5,000 cells/cm². After 3 to 5 days, MSCs were harvested for transduction with viral vectors. MSCs were mixed with vectors at a MOI between 2 and 15 and seeded in five-layer stacks. Three days after transduction, medium was removed and replaced with puromycin containing Bio-1 medium (3 μg/mL). Cells were selected for 3 to 5 days. After completion of selection, MSCs were further cultivated in standard Bio-1 Medium until cryopreservation. MSCs underwent three to five passages before being cryopreserved for the experiments. Informed written consent was obtained from each donor. The study was conducted in accordance with the Declaration of Helsinki.

Mice were anesthetized and sacrificed by cervical dislocation. Bone marrow was isolated from femur and tibia. Cells were seeded at 4 Mio. cells/cm² and expanded over several weeks. If the culture still contained macrophages, they were removed with CD11b magnetic beads. Upon expansion cells were frozen in Hydroxyethyl starch (Fresenius Kabi) with 10% DMSO. Generation of transduces and native MSC batches occurred from these stocks. For murine experiments, mMSC^IL7/12 were generated as following: A vial of mMSC was thawed and cells seeded in a cell stack with αMEM (Lonza) media containing 20% FBS (Merck) and 1% Glutamax (Gibco). Four days later, cells were transduced with a MOI of 2.5 to 10 and cells seeded on a five-layer cell stack. Cells were selected with 2 μg/mL puromycin until the selection control was dead. Cells were further expanded or directly frozen in 90% HAES/10 % DMSO. For GL261 experiments, IL7/12-producing MSCs were generated from C57BL/6 mice, therefore performing treatment with autologous MSC^IL7/12. For CT-2A experiments as well as for the fate mapping and the in vivo cytokines expression quantification, MSCs were generated from Balb/c mice, resulting in treatment with allogeneic MSC^IL7/12, which is similar to the planned clinical trial protocol.

Vectors were generated by transient transfection of 293T cells as described previously (23).

Tumor tissue was harvested 1 day after injection of MSC^IL7/12 (n = 7), MSC^native (n = 8), or PBS (n = 8) and processed for protein extraction in RIPA buffer. Protein quantification was assessed by BCA assay (Pierce) and 100 μg of total protein was used for the LEGENDplex assay (BioLegend, mouse customized panel for IL7, IL12p70, IL12p40, IFNγ, and TNFα), and run according to the manufacturer protocol and analyzed by flow cytometry.

After PBS perfusion, mice brains were extracted, fixed overnight in formalin, transferred to 20% sucrose, and then embedded in OCT. Ten-micrometer cryosections were cut on cryostat and fixed on superfrost+ slides to be stored at −80°C. After defrosting, slides were pretreated with an antigen retrieval citrate solution and stained with anti-GFP (Abcam, 1:400), followed by anti-rabbit alexa-488 or anti-CD3 (Abcam, 1:100) followed by anti-rabbit HRP and DAB. Immunofluorescent slides were then stained with DAPI and mounted in an aqueous solution. Pictures were taken under a Zeiss fluorescent microscope. For hematoxylin/eosin staining, tissues were embedded in paraffin and processed according to standard protocols.

All mice experiments were performed in accordance with the guidelines of animal welfare of the University Medical Center Hamburg-Eppendorf (Hamburg, Germany) and with explicit permission of the Institutional Review Board as well as the local authorities (Behörde für Soziales, Gesundheit und Verbraucherschutz Hamburg). GL261 and CT-2A (kindly provided by Dr. Darell D. Bigner, Brain Tumor Center, Duke University Medical Center, Durham, NC) syngeneic glioma cell lines were injected orthotopically into C57BL/6 J immunocompetent mice (Charles River Laboratories). GL261 cells were injected into the brain at a concentration of 0.7 × 10⁵ cells/μL in DMEM and 2 μL. Briefly, after anesthesia by injection of ketamine/xylazine, a small hole was drilled through the skull of the mouse after surgical incision of the skin and the cell solution was injected into the right striatum using a Hamilton syringe (30-gauge needle) at the following stereotactic coordinates from the bregma (0 mm, +2 mm, +3 mm depth). Mice were killed upon occurrence of neurologic symptoms or when significant weight loss occurred. MSCs were thawed, washed, and injected with the appropriate controls (MSCs with GFP or w/o) in a volume of 4 μL at the mentioned timepoints at a concentration of 1 × 10⁵ cells/μL in PBS. Cells and controls were injected stereotactically through the same burr hole directly into the tumor.

Tumor growth of intracerebral glioblastoma-bearing mice was evaluated on a 7T MR imaging system (ClinScan, Bruker). Mice were anesthetized with 1% isoflurane (Baxter) in oxygen (0.5 l/min). Respiratory rates were monitored using a small animal vital sign monitor (SA Instruments Inc.). Axial 2D T2-weighted turbo spin echo images were acquired to assess tumor location and size. Sequence parameters were: TE = 39 ms, TR = 2500 ms, BW = 250 Hz/pixel, turbo factor 7, matrix = 256 × 192, FOV = 20 × 15 mm², 19 slices, 0.4-mm slice thickness with a 0.1-mm gap. Volumetric quantification was performed using MRIcron software and manual segmentation of axial T2 images.

Mice were killed using CO₂ according to the animal welfare guidelines. Peripheral blood was drawn by perfusing the mice through injection of PBS into the left ventricle and incision of the right atrium. Blood was collected in EDTA-containing tubes. Peripheral blood was centrifuged, and erythrocyte lysis was performed (BioLegend) before resuspension and antibody staining. The tumor was removed from the brain by microsurgical dissection, minced, and digested for 30 minutes at 37°C in 0.01% DNAse and 10% Collagenase (Biochrom) diluted in Hanks. Afterwards, the cell suspension was filtered through 70-μm sieves, centrifuged, washed, Fc-blocked (Miltenyi Biotec), resuspended in FACS staining buffer (Thermo Fisher Scientific), and stained with the multicolor antibody staining (see list below). After staining, antibodies were washed off, cells were resuspended in FACS staining buffer and analysis was performed on a BD LSR Fortessa (Becton Dickinson). Data were analyzed with FACSDiva software (version 8.0.1; Becton Dickinson).

Genomic DNA was extracted from snap-frozen peripheral blood, after erythrocyte lysis, or freshly isolated tumor tissue. gDNA was used as a template for a multiplex PCR (primer pools summarized in Supplementary Table S1) amplifying the Tcrb gene locus, as described before (7, 24). Amplicons were tagged with Illumina adapters and indices in consecutive PCR reactions after size separation and purification using the NucleoSpinVR Gel and PCR Clean-up Kit (Macherey-Nagel). Concentration and purity of the amplicons were determined by Qubit (QIAGEN) and Agilent 2100 Bioanalyzer (Agilent Technologies). NGS was performed using an Illumina MiSeq sequencer [600 cycle single indexed, paired-end runs (V3 chemistry)](65, 66). Demultiplexing and FastQ data output was generated by the MiSeq reporter. Tcrb data analysis was performed using the MiXCR analysis tool (25). A clonotype was defined by the CDR3 amino acid sequence for further analysis and only sequences with a read count ≥2 were included in the analysis. Clonotype diversity was studied using Shannon-Wiener, inverse Simpson and clonality diversity indices (26, 27).

Analysis and graphical representation of was performed using R (The R Foundation for Statistical Computing), RStudio (RStudio, Inc.), Microsoft Excel 2016, GraphPad Prism 5.0 and Adobe Illustrator CC. Tcrb sequencing results with a minimal overlap of two samples per sequence were selected and frequencies were transformed into logicals (frequencies > 0 = logical 1). Tcrb- and flow cytometry data were used to create heat maps using the heatmap.2 function from the R package “gplots” (v3.0.1), statistical tests as indicated in the figures.

Native MSCs do not express a significant amount of IL7 and IL12. To generate cytokine-expressing MSCs, the human or murine IL7 and IL12 genes were cloned into a retroviral vector backbone (Fig. 1A). The expression of IL7 and IL12 is driven by the constitutive physiologic promoter EFS. P2A and T2A elements are used for efficient expression of multiple transgenes from the same promoter (28). Transduced human MSCs (MSC^IL7/12) retained the classical marker expression of CD73, CD90, and CD105 and absence of the negative markers CD34 and CD45 as defined by the ISCT consensus meeting (Supplementary Fig. S1A; ref. 29). Transduced murine MSCs also showed the expected marker expression, that is, were negative for CD11b and CD45 and positive for the murine MSC marker CD29, CD44, and Sca-1 (Supplementary Fig. S1B; ref. 30).

Both cell types, hMSC^IL7/12 and mMSC^IL7/12, were further characterized in vitro regarding their functionality to produce IL7 and IL12 (Supplementary Fig. S2A). Cytokine expression of virally modified hMSCs showed a mean level of human IL7 (hIL7) and human IL12 (hIL12) in the supernatant was 16 ng/10⁵ cells after 48 hours (Supplementary Fig. S1A), respectively. To perform in vivo testing using a syngeneic, immunocompetent tumor model, we also transduced murine MSCs with their murine cytokine counterpart (apceth-301^m). Mean level of murine IL7 (mIL7) in the supernatant was 14 ng/10⁵ cells and the mean level of murine IL12 (mIL12) was 57 ng/10⁵ cells (Supplementary Fig. S2B). The batch used for the in vivo experiments described below had a slightly higher expression (25 ng of mIL7 and 61 ng of mIL12). Given previous reports about the potential immunosuppressive function of MSCs (31, 32), we then tested the functionality of the cytokines IL12 and IL7 produced by the cells. MSC^IL7/12 or native MSCs were cocultured with anti-CD3/anti-CD28 stimulated PBMCs. As shown in Supplementary Fig. S2C, apceth-301 strongly increased the amount of IFNγ and TNFα secreted by T cells (Supplementary Fig. S2C).

Next, we performed in vivo testing of intratumorally injected MSCs (Fig. 1B). Fate mapping of GFP-labeled MSCs using immunofluorescent staining showed that MSCs, although continuously decreasing in number, could be detected at least up to five days after intracerebral injection (Fig. 1C). MSCs infiltrate and surround tumor cells, ensuring that cytokines are delivered directly in the tumor microenvironment. Additional IHC staining for CD3 (Fig. 1D) and automated quantification showed a significant increase in tumor-infiltrating T cells 5 and 10 days after MSC^IL7/12 injection compared with PBS or MSC^GFP controls (P = 0.0109–0.0001; Fig. 1E). IL7/12 production of MSCs compensated a potential immunosuppressive effect seen in MSC^GFP on day 5 after treatment injection and induced the accumulation of T cells (Fig. 1E). To confirm intratumoral IL7 and -12 cytokine production and the generation of a proinflammatory tumor microenvironment, we performed ex vivo analysis of intratumoral cytokines. The comparison between PBS and MSC^GFP controls showed that MSC^IL7/12 intratumorally produced high levels of IL7, IL12p40, and IL12p70, which resulted in strong intratumoral secretion of IFNγ (P < 0.001) and to a lesser extent also TNFα (P = 0.036–0.0006; Fig. 1F).

During treatment time course, no differences in weight between the groups was observed and additional histopathologic evaluation of the lung, liver, spleen, or kidney did not show any differences between the groups (Supplementary Fig. S2D and S2E). No systemic toxicity of MSC^IL7/12 treatment was observed.

Injection of genetically modified murine MSCs, which simultaneously expressed IL7 and IL12 (MSC^IL7/12), into established, orthotopically injected murine glioma (Fig. 2H) resulted in prolonged survival and long-term tumor cures compared with PBS and MSC^GFP control groups (Fig. 2A and D). Immunomodulatory MSC^IL7/12 were injected either 5 or 10 days after injection of 1.4 × 10⁵ GL261 or 0.5 × 10⁵ CT-2A syngeneic murine glioma cell lines. Both MSC^IL7/12 implantation time points resulted in significant survival benefits compared with the control groups (P < 0.05–0.001). Injection of MSC^IL7/12 on day 5 resulted in 30% survival in GL261 (P < 0.05–0.001). Surprisingly, injection of MSC^IL7/12 at day 10 also resulted in a substantial fraction of long-term survivors (GL261: day 10 = 50%; CT-2A: day 10 = 18%) in both models (P < 0.0001). Treatment efficacy of MSC^IL7/12 was assessed using MRI at day 20 for the GL261- and at day 12 for the more aggressive CT-2A glioma model (Fig. 2C and F). Volumetric quantification of T2 weighed images (T2w), as exemplarily shown in Fig. 2G, demonstrates significantly reduced tumor sizes in MSC^IL7/12-treated GL261 gliomas compared with the PBS control group [Fig. 2C: mean volume 38.56 mm³ (PBS) vs. 63.84 mm³ (MSC^GFP) vs. 13.91 mm³ (MSC^IL7/12 day 5) vs. 24.03 mm³ (MSC^IL7/12 day 10), (P = 0.0105–0.035)]. Even in the more aggressive CT-2A glioma model (median survival of CT-2A PBS controls 15.5 days vs. GL261 PBS controls 22 days), tumor volumes tended to be smaller when quantified at the early time point of day 12, although a statistically significant difference was not reached between all control and treatment groups (Fig. 2F). Analysis of T2w-MR images of GL261 glioma revealed distinct intratumoral T2 hypointensities of treated gliomas suspicious for tumor necrosis (Fig. 2G, arrow). Quantification of these T2 hypointensities in relation to the T2 hyperintense tumor masses revealed significant alterations in the MSC^IL7/12 treatment cohorts compared with the PBS and MSC^GFP control groups (Fig. 2I), indicating intratumoral alterations, which might serve as radiographic marker of treatment response in a clinical setting. Monitoring treatment effects by MR imaging (Fig. 2H) confirmed tumor-free long-term survival and demonstrated visible tumor regression (Fig. 2J).

We hypothesized that long-term survival of MSC^IL7/12 -treated animals might be due to the generation of a glioma-specific immunologic memory function in these animals. Therefore, long-term survivors (GL261: n = 7; CT-2A: n = 2) were rechallenged with the initially injected glioma cell line. All rechallenged animals rejected the implanted tumors and did not show any symptoms indicative of tumor growth throughout observation. Absence of tumors was confirmed by MRI at >50 days after rechallenge (Fig. 2J). In contrast to this, naïve C57BL/6 control animals rapidly died due to tumor growth (Fig. 2B and E; P < 0.0001) at a similar kinetic as observed for previous control groups.

To further analyze the underlying immunologic mechanisms, we next performed immunophenotyping of peripheral and tumor-infiltrating immune cells using ex vivo multicolor flow cytometry in the syngeneic GL261-C57BL/6 glioma model at two different timepoints. Mice were killed day 15 (PBS, MSC^wt, and MSC^IL7/12 d15) and day 30 (MSC^IL7/12 d30) after treatment (Fig. 3A). Correlative comparison of T2w-MRI hyperintense tumor volume with CD3⁺ T-cell infiltration showed that the local immunomodulation with MSC^IL7/12 interfered with the otherwise negative correlation of tumor size and T-cell infiltration as observed in the PBS and MSC^wt control groups (Fig. 3B). MSC^IL7/12 treatment significantly increased CD3⁺ T-cell infiltration into tumors (Fig. 3B; mean 23.9% PBS vs. 24.7% MSC^wt vs. 36.3% MSC^IL7/12 d15 vs. 35.0% MSC^IL7/12 d30, P = 0.02–0.048) and resulted in conversion of the CD4⁺/CD8⁺ T-cell ratio, with a significant predominance of tumor-infiltrating CD8⁺ cytotoxic T cells in the MSC^IL7/12 treatment group at day 15 (Fig. 3B, P = 0.0025).

A detailed analysis of the expression of markers for T-cell activation and potential exhaustion, including CD69, PD-L1, CTLA4, PD-1, KLRG1, and TIGIT, on peripheral and tumor-infiltrating T cells demonstrated profound alterations after intratumoral injection of IL7 and IL12-expressing MSCs in both, the peripheral blood and local tumor environment (Fig. 3C and D). Overall, TILs expressed the activation markers CD69 and PD-L1 at higher frequencies compared with peripheral T cells (33). Compared with T cells from control animals (PBS and MSC^wt), peripheral CD8⁺ and CD4⁺ T cells at both time points (day 15 and 30) showed higher proportions of CD69 and PD-L1–positive cells, while lower frequencies were observed for cells positive for CTLA4, PD-1, KLRG1, and TIGIT (Fig. 3C and D, left). In the tumor environment, expression of CTLA4 increased on both, CD4⁺ and CD8⁺ TILs, while PD-1, although the by far most prevalently expressed exhaustion marker, was the only marker without significant changes between the treatment groups (Fig. 3C and D, right). Interestingly, the frequency of cells expressing the killer cell lectin-like receptor G1 (KLRG1) differed between CD4⁺ and CD8⁺ TILs, decreasing in the CD4⁺ TIL population of the MSC^IL7/12 treatment group, while increasing in the CD8⁺ TILs.

The differentiation profile of T cells was evaluated using CD44 and CD62L expression (Supplementary Fig. S3A). Again, differences between the treatment groups were observed in the peripheral blood and tumor environment. Direct comparisons demonstrated that both CD4⁺ and CD8⁺ T cells in the peripheral blood displayed a significant increase in the frequency of central memory T cells (T_CM: CD44⁺ CD62L⁺) on day 30 in the MSC^IL7/12 group (Fig. 3E, top). MSC^IL7/12 treated tumors analyzed on day 15 only showed increased prevalence of peripheral naïve CD8⁺ T cells (T_NV: CD44⁻ CD62L⁺). Interestingly, in the tumor environment, CD8⁺ TILs in the MSC^IL7/12 treatment group analyzed on day 15 displayed a predominant memory (T_MEM: CD44⁺ CD62L⁻) and central memory (T_CM) phenotype (P < 0.01) compared with the controls, while the CD8⁺ TILs analyzed on day 30 were characterized by an effector T-cell differentiation (T_EFF: CD44⁻ CD62L⁻) and less by a T_CM phenotype (Fig. 3E, bottom).

In addition to the T-cell phenotype, we analyzed the composition of immune cells in peripheral blood and the tumor environment (Supplementary Fig. S4). Most prominent changes were alterations in frequencies of CD11b⁺ CD68⁺ cells, most likely describing macrophages (Fig. 3F). In the peripheral blood, MSC^IL7/12 treatment resulted in a decrease of CD11b⁺ CD68⁺ cells with a concomitant increase in CD11b⁻ Ly6C⁻ CD11c⁺ cells, presumably representing dendritic cells (DC). A similar reduction of CD11b⁺ CD68⁺ cells was detected in the MSC^IL7/12 treatment group on day 15 in the tumor environment (Fig. 3F, bottom). Even more pronounced changes in composition were found in the analysis of MSC^IL7/12 on day 30. Here, CD11b⁺ CD68⁺ cells increased significantly compared with all other groups (P = 0.021–0.002), while CD49b⁺ NK cells and CD11b⁺ Ly6C⁺ cells were reduced (Fig. 3F, bottom; P = 0.005–0.0001).

To further illustrate how the above-described individual changes, impact the peripheral phenotype and more specifically the local immune response in the tumor environment, we performed an unsupervised clustering analysis of all investigated immune markers, their combinations, and of the defined cell populations, totaling 40 parameters (Fig. 4A and B). The comparison of peripheral blood and tumor-infiltrating immune cells showed that TILs represent a highly specialized immune cell population, which is defined by strong expression of activation markers like CD69 and PD-L1, as well as the immune checkpoint molecule PD-1, and a central memory phenotype. On the other hand, KLRG1 positive CD4⁺ and CD8⁺ TILs, B- and NK cells are less prominent in the tumor environment in the presence of MSC^IL7/12 (Fig. 4B).

To demonstrate how the MSC^IL7/12 treatment modulates the immune system we subsequently performed a principal component analysis (PCA) of all markers separately for the peripheral blood (Fig. 4C–E) and tumor environment (Fig. 4F–H). Unsupervised clustering of all markers and samples in the peripheral blood displayed that most changes in the MSC^IL7/12 treatment group on day 15 were primarily driven by the high frequency of CD69⁺, PD-L1⁺ T cells, while the samples analyzed on day 30 clustered separately due to increased expression of KLRG1 and differences in maturation of the T-cell phenotype (Fig. 4D). Evaluation of the spleen size showed a decrease of weight in MSC^GFP-treated mice on day 15, implicating a potential immunosuppressive effect of MSCs alone (Supplementary Fig. S5A). Endpoint analysis showed increased spleen sizes of day 10 MSC^IL7/12-treated mice compared with PBS controls (P = 0.052; Supplementary Fig. S5B). In the tumor environment, the PCA revealed that both control groups (PBS and MSC^wt) clustered together, whereas samples of the MSC^IL7/12 treatment cohort analyzed on day 15 and 30 separated significantly in opposite directions (Fig. 4F). Subsequent unsupervised clustering again showed distinct immune phenotypes between the control and MSC^IL7/12 treatment groups (Fig. 4G). To give an accurate overview of the gene expression patterns in the different treatment groups [control (PBS and MSC^GFP) versus MSC^IL7/12 at day 5 vs. MSC^IL7/12 day 10] summarized heatmaps were added (Fig. 4E and H), which only included encircled samples from the PCA (Fig. 4C and F). After initial upregulation of activation and exhaustion markers, T cells of MSC^IL7/12-treated mice surviving until day 30, less frequently expressed markers of exhaustion like CTLA4 and PD-1, but rather displayed rather a T_EFF phenotype with expression of TIGIT on CD8⁺ TILs (Fig. 4H). These heatmap representations highlight that MSC^IL7/12-mediated immunomodulation resulted in distinct local and peripheral immune activation profiles compared with control groups. Furthermore, MSC^IL7/12 treatment promoted a tumor-specific immune response, which was characterized by increased activation and memory formation in the equilibrium phase and subsequent presumable tumor control associated with reduced expression of exhaustion markers on infiltrating effector T cells.

To evaluate the clonogenicity and diversity of the tumor-specific T-cell response, we performed next-generation sequencing of the T-cell receptor beta repertoire (Tcrb) from peripheral blood and TILs of GL261-bearing C57BL/6 mice treated with MSC^IL7/12 or controls 10 or 20 days after the initial tumor implantation (Fig. 5A). Peripheral blood and TILs were analyzed on day 15 (n = 6/group), as well as on day 30 for surviving mice of the MSC^IL7/12 treatment group. While no differences in the clonality, N50 frequency (number of clones needed to make up 50% of all frequencies) or number of clones was observed between the groups in the peripheral blood, TILs extracted five days after treatment with MSC^IL7/12 (day 15 after tumor cell injection) already exhibited a significantly increased number of clones compared with the other groups (mean ± SEM: 1,335 ± 76, P = 0.001–0.042; Fig. 5B, top right). TILs extracted 20 days after treatment with MSC^IL7/12 (day 30) on the other hand displayed a significantly increased clonality compared to all other groups (mean ± SEM: 0.392 ± 0.029, P = 0.001–0.003; Fig. 5B, middle right). While the N50 frequency at day 15 in the treated group is significantly increased indicative of a very broad and diverse T-cell response, the significant reduction of the N50 frequency at day 30 indicated a condensation of the T-cell response with a reduced diversity in this context (Fig. 5B, bottom right).

Next, we analyzed the overlap of the 50 most frequently present T-cell clones, as described by their unique amino acid sequences of the complementary determining region (CDR3aa) of their Tcrb. Overall, 62 T-cell clones (TCC) were identified to be present at least twice within tumors of all groups (Fig. 5C). Four clones of the most frequent TCCs were present in four or more tumors, with one clone being present in eight of 25 tumors (32%). The distribution of the 62 clones between the groups is displayed in Fig. 5D. Sixteen clones were only found within one group (Fig. 5D, top pie chart). Interestingly, 9 of 16 (56.3%) TCCs present only in one group were found in the MSC^IL7/12-treated tumors at day 30, indicating a potential common tumor-specific T-cell response after apceth-301^m treatment. Only one TCC found in eight tumors was distributed in all four groups (Fig. 5D, bottom).

To evaluate whether the expansion of tumor-specific T-cell response was reflected in the peripheral blood, we analyzed the overlap of TCC between the peripheral blood and tumor of each individual animal. The average number of clones (mean 3,406–3,789, size of pie chart) and the mean proportion of TCC in the peripheral blood that overlap with the tumors was constant throughout all groups (4.9%–6.8%, blue share; Fig. 5E, left). In comparison, as mentioned above, the average number of clones (Fig. 5B, top) differed significantly in the tumors (Fig. 5E, middle). The proportion of overlapping clones that were also present in the peripheral blood increased from 19.3% in the PBS group to a maximum of 33.9% of TCCs in the MSC^IL7/12 tumors (P = 0.102; Fig. 5E, right). This indicated that the tumor-specific T-cell response in the later disease course was detected in part in the peripheral blood.

Furthermore, all CDR3aa clones present in tumors isolated at least from two different animals were depicted in a grouped heatmap (Fig. 5F). The heatmap visualizes the initially present increase in TCC diversity in the MSC^IL7/12-treated group on day 15, and the following reduction of diversity with an increased expansion of a limited number of clones, as the potential drivers of the tumor-specific immune response.

Stem cell–based concepts of optimizing drug delivery in diffusely infiltrating gliomas are highly attractive and compelling preclinical data have been gathered over the last years (34–37). With the emerging successful immuno-oncological therapies in other cancer types translation of immunotherapeutic options to malignant gliomas is under intensive investigation. However, to overcome the substantially immunosuppressed tumor environment in gliomas, a powerful and localized switch to a proinflammatory tumor environment is going to be essential to safely foster the natural immune defensive mechanisms. Previous studies, for example, analyzed the effect MSC-expressing IL24 (38), a secreted variant of TRAIL (39), the use of IFNβ-expressing MSCs after tumor resection (40), and MSC-mediated suicide gene therapy (41). In this study, we assessed whether an MSC-based local immunomodulation is able to overcome the immunosuppressive glioblastoma microenvironment and to induce an antitumor immune response. Using IL7 and IL12-expressing MSCs induced a potent antitumor T-cell response with tumor remissions and resulted in prolonged overall survival and long-term immunity in murine glioblastoma models. These effects were the result of a significant intratumoral inversion of CD4⁺/CD8⁺ T-cell ratio with an intricate, predominantly CD8⁺ effector T-cell–mediated antitumor response.

The recent success of immunotherapeutic treatment approaches has boosted the field of immuno-oncology, demonstrating that the activated immune system can result in long-term tumor remissions even in the metastatic setting (9). However, checkpoint inhibition, as the most prevalently approved immunotherapeutic approach, did not result in increased overall survival in recurrent glioblastoma as preliminary results indicated (42). A possible reason why checkpoint inhibition did not lead to an increased survival might be that the systemic administration of anti–PD-1 antibodies did not sufficiently disinhibit tumor-specific T cells to mount an effective glioma-specific T-cell response or that the activated T cells are impeded once they migrate into the immunosuppressive tumor microenvironment. Local immunomodulation toward a proinflammatory environment as demonstrated in this study has therefore the potential to overcome these obstacles and directly disinhibit infiltrated tumor-specific T cells and may potentially be combined with other systemic immunotherapeutic approaches.

In this study, we explored the efficacy and mechanisms of IL7 and IL12-expressing MSCs in two orthotopic glioblastoma mouse models. Previous work has focused on soluble factors naturally produced by MSCs. MSCs were shown to inhibit proliferation and migration of various tumors, including NSCLC cells (43), Kaposi sarcoma (44), and, among others, leukemia cells (45). Furthermore, preclinical studies increasingly explored the use of genetically modified MSCs as vehicles, expressing for example IFNβ to treat glioma (14), IFNγ and IL10 to treat HCC (46), or expressing IL18 to inhibit proliferation and metastatic spread of breast cancer cells in vivo (47). An additional targeted approach to enhance tumor-cell recognition by T cells was mediated by MSCs genetically modified to express antigen-bispecific T-cell engagers, which induced direct cell–cell interaction of immune- and tumor cells (48). The expression of IL12 alone, as potential T-cell promoting factor was first investigated using umbilical cord MSCs and resulted in promising results in vivo (49). In our study, we further developed the ability of MSCs as immunomodulators in the local tumor environment by genetically modifying MSCs to additionally express IL7 as a crucial factor to enhance the IL12 effect and maintain the survival of naïve and memory T cells, as well as to antagonize the immunosuppressive network (50). Our results demonstrated, that the combination of both interleukins was able to induce long-term survival in two different glioblastoma models in vivo. Interestingly, MR imaging confirmed that the observed effect and long-term survival was not only the result of tumor growth inhibition, but more importantly, tumor remission, as established tumors disappeared during the disease course. Results of the rechallenging experiments in both models further supported that tumor remission was mediated by an adaptive immune response, which created a long-term tumor-specific memory.

One potential limitation of our study is that the available syngeneic glioma models do not perfectly reflect the extensive immunologic immunosuppression present in human glioblastoma. Although recent data show that immune exhaustion and a immunosuppressive environment can certainly be found in these models (8), the chemically induced GL261 and CT2A cell lines harbor significantly more mutations compared to human glioblastoma (ref), and therefore, tumor-specific immune responses can be generated more easily. Ultimately, our translational MSC^IL7/12-based immunotherapy approach will face greater challenges in the human immunosuppressive environment and need to prove its efficacy in clinical studies. However, as recurrent human glioblastoma after radio-/chemotherapy demonstrates a hypermutated phenotype in up to 15% of cases (51), specific patient populations might respond better to our immunotherapeutic approach, which has to be addressed in clinical study protocols.

Other studies have shown that native MSCs itself may also exert protumorigenic effects, as cells with MSC-like phenotype were found to infiltrate into the tumor stroma and to contribute to tumor progression (52). The percentage of glioma-associated MSCs even correlated negatively with overall survival in patients with high-grade glioma (53). Attracted by TGFβ, glioma-associated MSCs increased growth and maintained glioma stem-like cells, possibly also by direct cell–cell interaction (54–56). In a lung cancer model, MSCs were shown to increase the amount of produced IL10, thereby suppressing an antitumor immune response (57). However, in contrast, genetically modified MSCs to express IL7 and IL12 (apceth-301^m) did not only increased the overall T-cell infiltration of glioblastoma xenografts, but more importantly lead to a reversal of the CD4⁺/CD8⁺ ratio, clearly favoring a cytotoxic T-cell response in contrast to potentially infiltrating CD4⁺ regulatory T cells. The activated phenotype of CD8⁺ T cells was demonstrated by their strong expression of CD69 and PD-L1, especially in TILs. Although the tumor volume negatively correlated with the number of infiltrating CD3⁺ T cells in the control group, MSC^IL7/12-treated tumors did not follow this correlation, indicating a disruption of the immunosuppressive network by the local cytokine production.

The expression of molecules, presumably involved in T-cell exhaustion, like PD-1, KLRG1, TIGIT, or CTLA4, is still discussed controversially. It is not clear how well the expression of these markers reflects the functional impairment of TILs or if the upregulation is a physiologic response during tumor-specific T-cell activation. PD-1 is by far the most prevalently expressed marker of TILs in our study, reflecting the focus of recent studies and clinical assessment of checkpoint inhibitors (58, 59). However, other markers, like KLRG1 and TIGIT seem to reflect the local changes in the immune environment more sensitively. CTLA4, although expressed in only a minority of cells, which does not reduce its relevance in the immunologic synapse (60), increased during immune activation as expected (61). KLRG1 and TIGIT reacted oppositional. Although KLRG1 increased in MSC^IL7/12-treated tumors during the early time point and dropping in the long-term surviving mice, TIGIT was found to be decreased in MSC^IL7/12-treated mice during the early timepoint. Both markers appear to reflect changes in the immune environment in both, CD4⁺ and CD8⁺ subtypes. In one of our previous studies in which we investigated human glioblastoma, KLRG1 expression of CD8⁺ TILs was reduced significantly compared with peripheral blood (7), presumably indicating the functional impairment of TILs. The increased expression during MSC^IL7/12-treatment in the CD8⁺ compartment might therefore be a good predictor for an effective immune activation in the local immune environment and will be investigated as potential surrogate for clinical response in the upcoming studies.

So far, the dynamics of the immune phenotype and composition during tumor-specific immune activation in the local environment are not well understood. By including a follow-up analysis of long-term surviving mice on day 30 after glioma injection and comparing the T-cell phenotype with the early time point (day 15), our analysis showed how the immune populations and expression of activation markers and immune differentiation of TILs changed dynamically. Interestingly, the transient increase of effector memory CD8⁺ TILs shifted into a predominant effector CD8⁺ TIL phenotype in MSC^IL7/12-treated tumors. A similar trend with opposing dynamics between the early and late time point immunophenotyping was also observed in the CD11b⁺ CD68⁺ population, which represents mostly macrophages/microglia. The observed increase of macrophages/microglia in long-term surviving mice might reflect the recruitment of macrophages/microglia to the site of inflammation for removal of apoptotic/lysed tumor cells. In line with our observation, a similar increase of macrophage/microglia infiltration was found by others using the immune checkpoint inhibitor nivolumab in glioblastoma mouse models (62).

Although MSC^IL7/12 were injected into the tumor locally, the activation of the tumor-specific immune response was also reflected in the peripheral blood. Here, presumable markers of exhaustion were significantly reduced in both, CD4⁺ and CD8⁺ T cells in the peripheral blood. Future studies, focusing on the discovery of peripheral markers for intratumoral response will demonstrate whether the observed effect of peripheral immune activation is the result of cytokine release from the tumor or a direct consequence of the tumor-specific immune response. Interestingly, although the activation markers CD69 and PD-L1 of the CD8⁺ T-cell compartment reflected the activation pattern of CD8⁺ TILs, peripheral CD4⁺ T cells showed an almost inverse activation pattern compared with their TILs. One explanation might be that the local immunomodulation due to MSC^IL7/12 treatment and subsequent increase in T-cell infiltration was perpetuated and promoted additional peripheral T-cell activation. This was reflected in the shifts of immune marker activation and peripheral T-cell differentiation subsets, that is, increase in central memory T cells and replenishment of intermediate phenotype by recruitment of naïve T cells.

To investigate the T-cell specificity, their diversity and their dynamics in peripheral blood and TILs, we performed additional next-generation sequencing of the T-cell receptor beta (Tcrb). Our results showed that the increased infiltration of T cells due to MSC^IL7/12 treatment on day 15 resulted in an increased Tcrb repertoire diversity compared to control groups. This broad repertoire diversity of infiltrating T cells on day 15 then contracts to a highly oligoclonal TIL repertoire in which just 10 to 15 T-cell clones make up over 50% of all infiltrating lymphocytes. Comparing the CDR3 amino acid sequence, which identifies identical T-cell clones, of the most expanded TIL clones between individual mice then showed, that identical clones were found significantly more often within the MSC^IL7/12-treated group on day 30. This finding indicates, that the prolonged survival and initiation of tumor immunity during MSC^IL7/12 immunomodulation induced a common, oligoclonal T-cell response, which is driven by a selection of certain clones, which presumably targeted a limited repertoire of antigens. These observations are in line with previous studies in humans, which describe the contraction of the Tcrb repertoire diversity during tumor recurrence (2, 7), or that only a limited number of tumor antigens is available to induce a tumor-specific immune response (5, 6, 63, 64). Our data showed that local immunomodulation is able to generate this specialized, oligoclonal immune response by transiently increasing the amount of infiltrating T cells and promoting a broad T-cell repertoire as a basis for selection of the fittest tumor-specific clones. Furthermore, it is important to note for upcoming clinical studies, that a large fraction of the tumor-specific immune response in MSC^IL7/12-treated mice on day 30 was also found in the peripheral blood. This might enable monitoring immune activation and treatment success in upcoming clinical studies.

Taken together, our results demonstrated that intracranial immunomodulation using IL-expressing MSCs represents a safe approach to effectively activate the intrinsic antitumor response in vivo, resulting in CD8⁺ T-cell–mediated activation and potential long-term survival, even after tumors have been fully established. Our study provides detailed insights into the tumor-specific immune response and T-cell clonality during MSC-mediated modulation of the immunosuppressive glioma microenvironment. Future clinical trials will aim to confirm the efficacy of MSC^IL7/12 (apceth-301) treatment and further assess the dynamics of immune activation in humans. Immunomodulatory MSC as a localized therapeutic approach have the potential to serve as a platform technology for the targeted therapy of invasive glioma and to be combined with other systemic immunotherapeutic approaches by providing a localized proinflammatory microenvironment.

U. Geumann, F.G. Hermann, and C. Guenther are employees/paid consultants for apceth Biopharma. M. Binder reports receiving speakers bureau honoraria from Bristol-Myers Squibb, Celgene, and Gilead. N.O. Schmidt reports receiving commercial research grants from apceth Biopharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Mohme, C.L. Maire, U. Geumann, F.G. Hermann, N.O. Schmidt

Development of methodology: M. Mohme, C.L. Maire, N. Akyüz, M. Binder

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Mohme, C.L. Maire, S. Schliffke, L. Dührsen, K. Fita, M. Binder, C. Guenther, K. Lamszus

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Mohme, C.L. Maire, U. Geumann, S. Schliffke, L. Dührsen, K. Fita, M. Binder, C. Guenther, K. Lamszus, F.G. Hermann, N.O. Schmidt

Writing, review, and/or revision of the manuscript: M. Mohme, C.L. Maire, U. Geumann, L. Dührsen, M. Binder, M. Westphal, C. Guenther, K. Lamszus, F.G. Hermann, N.O. Schmidt

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Mohme, N. Akyüz, M. Westphal, K. Lamszus, N.O. Schmidt

Study supervision: M. Mohme, M. Binder, K. Lamszus, F.G. Hermann, N.O. Schmidt

Other (provided laboratory infrastructure): M. Westphal

We thank Svenja Zapf, Katharina Kolbe, Mareike Holz, and Barbara Gösch for excellent technical assistance. We thank the UKE FACS Core Facility and the Mouse core facility for their excellent support.

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.