Tumoricidal stem cell therapy enables killing in novel hybrid models of heterogeneous glioblastoma

Abstract

Background

Tumor-homing tumoricidal neural stem cell (tNSC) therapy is a promising new strategy that recently entered human patient testing for glioblastoma (GBM). Developing strategies for tNSC therapy to overcome intratumoral heterogeneity, variable cancer cell invasiveness, and differential drug response of GBM will be essential for efficacious treatment response in the clinical setting. The aim of this study was to create novel hybrid tumor models and investigate the impact of GBM heterogeneity on tNSC therapies.

Methods

We used organotypic brain slice explants and distinct human GBM cell types to generate heterogeneous models ex vivo and in vivo. We then tested the efficacy of mono- and combination therapy with primary NSCs and fibroblast-derived human induced neural stem cells (iNSCs) engineered with tumor necrosis factor–related apoptosis-inducing ligand (TRAIL) or enzyme-prodrug therapy.

Results

Optical imaging, molecular assays, and immunohistochemistry revealed that the hybrid models recapitulated key aspects of patient GBM, including heterogeneity in TRAIL sensitivity, proliferation, migration patterns, hypoxia, blood vessel structure, cancer stem cell populations, and immune infiltration. To explore the impact of heterogeneity on tNSC therapy, testing in multiple in vivo models showed that tNSC-TRAIL therapy potently inhibited tumor growth and significantly increased survival across all paradigms. Patterns of tumor recurrence varied with therapeutic (tNSC-TRAIL and/or tNSC–thymidine kinase), dose, and route of administration.

Conclusions

These studies report new hybrid models that accurately capture key aspects of GBM heterogeneity which markedly impact treatment response while demonstrating the ability of tNSC mono- and combination therapy to overcome certain aspects of heterogeneity for robust tumor kill.

Key Points.

1. We generated novel hybrid GBM models to test the impact of intratumoral heterogeneity.

2. We treated novel heterogeneous GBM models with engineered induced neural stem cells.

Importance of the Study.

Understanding the impact of GBM heterogeneity on tNSC therapy is vital for effective treatment in human patients, yet the lack of relevant preclinical models has left their impact largely unexplored. We report new hybrid models capable of capturing multiple aspects of molecular and spatial GBM heterogeneity. In initial tests of these models, our results show that tNSC-TRAIL therapy still imparts marked tumor kill despite the presence of significant heterogeneity, while tNSC‒thymidine kinase therapy requires further optimization to achieve the same response. These new models will continue to be useful for testing multiple types of targeted GBM therapy, while our results can begin to serve as a guide for selecting the most efficacious tNSC therapy in human patient testing.

Glioblastoma (GBM) is the most common primary adult malignancy of the brain and one of the most aggressive and deadly cancers, carrying a 3-year survival rate of only 6%.¹ The aggressiveness of high-grade gliomas such as GBM derives from their high rate of invasion into distant brain regions, as well as from high degrees of intratumoral heterogeneity in morphology and drug resistance.^2–4 To date, murine xenografts using a single human GBM cell line are the mainstay of preclinical models, yet they fail to recapitulate numerous key aspects of tumor biology that make GBM difficult to treat in the clinical setting. Improving preclinical tumor models to reflect the invasiveness and heterogeneity of GBM should allow researchers to more effectively refine therapies in order to improve the outcomes of clinical trials.

Many clinical trials in oncology now focus on cell-based therapies. An emerging cell-based therapy for GBM utilizes tumoricidal neural stem cells (tNSCs), which track chemokine gradients to migrate toward solid GBM foci and cancer cells invading into the nondiseased brain.^5–8 Recently, clinical trials exploring tNSC monotherapies have been launched on the success of promising preclinical data. In particular, numerous studies have utilized tNSCs carrying the pro-apoptotic agent tumor necrosis factor alpha (TNFα)–related apoptosis-inducing ligand (TRAIL) or enzyme/prodrug approaches such as thymidine kinase (TK)/ganciclovir (GCV). Despite promising results, each of these monotherapies face limitations due to the invasiveness and intratumoral heterogeneity present in patient GBM. Heterogeneity in the expression of death receptor 4/5 and subsequent downstream signaling allows GBM to escape NSC-TRAIL therapy,^9–12 and the limited range of TK/GCV efficacy—which occurs via either gap junction transfer or bystander effect—in turn limits its effectiveness in invasive GBM regions.^13,14 Current preclinical orthotopic GBM tumor models derived from single cell line xenografts are thus unsuitable for testing these therapies. Achieving comprehensive and accurate preclinical optimization and validation requires the development of preclinical models which better simulate such clinically relevant hurdles.

We thus sought to create a new orthotopic human GBM modeling approach that better recapitulates the key clinical challenges arising from the intratumoral heterogeneity of GBM tumors. Ex vivo and in vivo testing of distinct GBM cell types from established lines and patient biopsies showed that these new hybrid models incorporate aspects of drug resistance, variable invasiveness, and tumor morphology that could not be achieved by any single cell type. Using these new models, we evaluated the strengths and limitations of NSC-TRAIL and NSC-TK therapy. We found that NSC-TRAIL significantly increased survival across all treatment paradigms despite large numbers of TRAIL-resistant and highly invasive GBM cells present in the tumor. In contrast, the efficacy of NSC-TK displayed greater dependence on the tumor growth profile and method of NSC delivery. Taken together, the hybrid modeling approach afforded key insights into differences between the two main strategies in tNSC therapy that had not been apparent from conventional preclinical models, and may prove useful in future studies to reveal further insights into GBM drug resistance and tumor recurrence. Testing the ability of tNSC mono- and combination therapy to overcome key aspects of heterogeneity and achieve robust tumor kill should serve as an important guide to designing and optimizing next-generation tNSC therapies and new clinical trials.

Materials and Methods

Details of this section can be found in the Supplementary Material.

Ethics Statement

All work performed on female athymic nude mice (therapy studies) or Sprague-Dawley rats (brain slice preparation) was approved by the Institutional Animal Care and Use Committee at the University of North Carolina–Chapel Hill or Duke University. All patient MRIs presented here have been fully de-identified.

Cell Lines and Lentiviral Vectors

U87, LN229, and C17 cells were purchased from American Type Culture Collection. GBM-8, G-EF, and G-FBS cells were gifts from H. Wakimoto (Massachusetts General Hospital). Human fibroblasts were provided by W. Kauffman (UNC School of Medicine). Cells were grown in vitro as previously described.^6,15 Lentiviral vectors encoding human telomerase reverse transcriptase (hTERT) and sex determining region Y–box 2 (SOX2) were purchased from Addgene. Induced NSC (iNSC) generation was completed as previously described.⁶

In Vitro Migration

Twenty-four hours after seeding, a line of cells was scraped off the plate. An EVOS FL Auto imaging system (ThermoFisher) was used to take a 40 h time lapse of cell migration. The Chemotaxis Tool plugin in FIJI quantified the migrated distance for cells along the scraped border (n = 10).

Brain Slice

Brain slice explants were prepared from postnatal day 10 Sprague-Dawley rat pups of either sex using previously described protocols.^16,17 Foci of concentrated cells were added and grown on the slice surface and imaged using fluorescence or bioluminescence imaging (BLI).

Hybrid Tumor Implantation

Tumor implantation into female athymic nude mice was carried out as previously described.⁶ Stereotactically into brain parenchyma implanted were 350k U87-LSSO-nLuc, LN229-mCh-FLuc, and GBM-8-GFP-FLuc (1:3:10) in 3 μL at 1 μL/min and coordinates (x, y, z = 0, 2.7, 3.5) as measured from distal needle tip at bregma.

Preparation of Brains at Study Endpoint

Cardiac perfusion was performed with phosphate buffered saline (PBS) and 10% formalin. Brains were dissected, soaked in 10% formalin overnight, and cut across the tumor region. One half remained in formalin for an additional 48 h before paraffin embedding; the other half was moved to 30% sucrose in PBS overnight before embedding in optimal cutting temperature compound (OCT).

Fluorescence Imaging

Tumor-bearing brains in OCT were sectioned at 6 μm thickness. OCT was dissolved in PBS and Hoechst stain was applied. Sections were washed and coverslips mounted using Prolong Gold mounting medium. Tumor fluorescence was imaged using EVOS FL Auto or an Olympus IX73.

Hematoxylin/Eosin and Immunohistochemical Staining

Paraffin-embedded brains were stained with hematoxylin and eosin by the Translational Pathology Lab core facility at UNC, which also performed immunofluorescence/immunohistochemical (IHC) stainings.

Live Animal Bioluminescent Imaging

An IVIS Kinetic was used for in vivo BLI. XenoLight D-Luciferin was injected i.p. into mice at 3 mg/mouse in 200 μL PBS.

Hybrid Co-Culture Implant

Tumor cells were co-implanted alongside different amounts of therapeutically engineered mouse NSCs (C17-TRAIL and C17-TK). Mice given TK were given 2 mg GCV i.p. in 200 μL PBS daily from day 4 to day 15 after tumor implant. Tumor growth was quantified by BLI. For untreated, low-dose TRAIL, high-dose TRAIL, TK, and TRAIL + TK groups, n = 13, 4, 4, 4, and 5 mice, respectively.

Hybrid Established Tumor

Mice were treated in 4 groups 8 days after tumor implant: untreated (n = 4), 250k iNSC-TRAIL (n = 4), 250k iNSC-TK (n = 4), and 250k iNSC-TRAIL + 250k iNSC-TK (n = 5). Mice given TK were given 2 mg/mouse GCV i.p. in 200 μL PBS on days 11, 12, 13, 15, 16, and 17 after tumor implant.

Patient-Matched Established Tumor

Tumors were implanted at 600k cells/mouse (1:5 G-EF:G-FBS). Mice were treated in 4 groups 8 days after tumor implant: untreated (n = 4), TRAIL (n = 4), TK (n = 3), and TRAIL + TK (n = 7).

Statistical Analysis

Data were analyzed by Student’s t-test, one-way ANOVA/Bonferroni, or log-rank test when appropriate. Data are expressed as mean ± SEM. Significance between groups is denoted by *P < 0.05, **P < 0.01, ***P < 0.001.

Results

Ex Vivo Cell Line Validation

In engineering our initial hybrid tumor model, we first sought to identify human GBM cell lines that could recapitulate individual characteristics of patient GBM, such as a solid core, infiltrative margins, and varied response to targeted cytotoxic agents such as TRAIL. Literature led us toward evaluating 3 human GBM cell lines that collectively could satisfy these properties: U87,¹⁸ LN229,¹⁹ and GBM-8.²⁰ Initial tumor cell migration studies performed on polystyrene tissue culture plates showed that U87 cells migrate more quickly than LN229 or GBM-8 (P < 0.001; Fig. 1A, B), contradicting established in vivo trends demonstrating nonmigratory tendencies of U87 cells and extremely invasive patterns of GBM-8 cells. This highlighted the need for a more predictive experimental setting in which to quantify tumor growth and migration.

Fig. 1.

Brain slice studies. (A) In vitro migration on polystyrene dish. (B) Quantification of cell velocity, n = 10. (C) Tumor foci on corpus callosum of brain slice. (D) Qualitative images of tumor foci growth and migration on brain slice. Quantification from day zero to 6 showing (E) change in diameter, n = 6; (F) cell growth, n = 6. (G) Foci height on slices. (H) Intravital images of foci. (I) Quantification of foci height; n = 3. (J) Interactions between cell lines on brain slices, t = 3 h and 6 days (left and middle) and in vivo (right). (K) Plot profile quantification of fluorescent signal from cell distribution in (J). ***P < 0.001, **P < 0.01.

We thus turned to living organotypic brain tissue slice explants (brain slices) to analyze the functional characteristics of individual cell lines in a setting that better simulates the in vivo environment. Small foci of concentrated tumor cells were added topically to brain slices along the corpus callosum (Fig. 1C). Similar to U87 growth in vivo, U87 tumor foci did not migrate outwardly across the brain slice tissue (Fig. 1D, E) even as they grew in cell number (Fig. 1F) and increased in height in the z-direction (Fig. 1G‒I). In contrast, GBM-8 tumor foci recapitulated the aggressive invasiveness of GBM-8 tumors in vivo, with cells migrating radially outward, most quickly along the corpus callosum of the brain slice, increasing in diameter over time (P < 0.001 vs U87) and maintaining a flat profile along the slice (P < 0.01 vs U87).

We also used brain slices to model the cell-cell interactions seen in vivo (Fig. 1J). U87 + GBM-8 cells segregated on the slices, while LN229 + GBM-8 cells maintained a homogeneous mixture. These same growth patterns were observed in vivo. Quantification of the relative fluorescence intensity of each cell type along the corpus callosum on the brain slice and in vivo confirmed this correlation (Fig. 1K).

We next used brain slices to analyze the relative sensitivity of U87, LN229, and GBM-8 to TRAIL, one of the most common antitumor agents delivered by NSCs,²¹ and to standard-of-care chemotherapy (temozolomide). We measured TRAIL resistance by mixing each tumor line with different ratios of NSC-TRAIL before addition onto the slice (Supplementary Fig. 1A, B). U87, LN229, and GBM-8 cells showed significant differences in TRAIL response at t = 3 days. When NSC-TRAIL and tumor cells were mixed 1:1, the growth of LN229, U87, and GBM-8 cells was inhibited by 5%, 97%, and 99.7%, respectively. The 3 cell types also showed heterogeneity in sensitivity to temozolomide 3 days after the drug was added to media under a transwell membrane with the brain slice + tumor foci lying atop the membrane (Supplementary Fig. 1C).

These results highlight the usefulness of brain slices in characterizing different tumor cell types and in assaying responsiveness to both cell therapy and chemotherapy. Importantly, this ex vivo tissue-based approach revealed differences in the migration patterns and drug sensitivities among U87, LN229, and GBM-8 cells, indicating that a comingled mixture of these 3 cell lines could be used to create a hybrid tumor recapitulating the solid core, infiltrative margins and varied drug response in clinical GBM.

Generating the Hybrid Tumor Model

Brain slice and in vivo (Supplementary Fig. 2) studies of U87, LN229, and GBM-8 revealed that none of these lines generates a tumor with the heterogeneity in growth and drug resistance observed in the clinic.²² De-identified patient GBM MRIs (Supplementary Fig. 3) show representative tumors with solid components and invasive regions, some of which extend across the corpus callosum into the contralateral hemisphere. We simulated this pathology by mixing fluorescently labeled U87-LSSmOrange (yellow), LN229-mCh (red), and GBM-8-GFP (green) cells and stereotactically implanting the cells into the brain parenchyma of nude mice. We sacrificed mice at 7, 10, 13, and 18 days post-implantation, sectioned brains axially or coronally, and measured fluorescence post mortem to analyze the growth and interactions among the cell types (Fig. 2). Seven days after implant, the 3 tumor types had segregated into 3 distinct regions in the growing tumors. The U87 cells composed the majority of the tumor center, while GBM-8 cells formed an invasive rim. Certain LN229 cells integrated within the U87-heavy center, while others migrated alongside the GBM-8 cells. As the tumors grew, multiple solid foci of U87 and LN229 cells formed. Initial invasion of LN229 and GBM-8 cells occurred inferiorly along the ipsilateral hemisphere, followed by increased GBM-8 invasion along the corpus callosum to the contralateral hemisphere. Between day 7 and 18 after tumor implant, the number of cells in each section migrating above the contralateral ventricle rose from just one to several hundred, with significant increases at each time point (days 7‒10, P < 0.001; days 10‒13, P < 0.01; day 13‒18, P < 0.01). Interestingly, subpopulations of GBM-8 showed far less invasion, with some GBM-8 cells near the tumor center remaining densely packed and nonmigratory.

Fig. 2.

Heterogeneous tumor morphology. Top row: Axial view showing growth of four different hybrid tumors from day 7 to 18 after implant. Second row: enlarged image of tumor area from axial section. Third row: Coronal view showing growth of 4 additional tumors; one at each time point. Bottom row: enlarged images of corpus callosum above ipsilateral and contralateral ventricles. Green GBM-8 cells in each contralateral view were quantified, n = 3: day 7 to 10, ***P < 0.001; day 10 to 13, **P < 0.01; day 13 to 18, **P < 0.01.

We further characterized this model by measuring changes in blood vessel structure (cluster of differentiation [CD]31), cancer stem cell populations (CD133), and microglial infiltration (CD11b) over time. Seven days after implant, the tumors contain a large amount of poorly defined vasculature (Supplementary Fig. 4), a small amount of CD133 expression that co-localizes with GBM-8 cells (Supplementary Fig. 5), and high CD11b expression (Supplementary Fig. 6), suggesting high microglial infiltration. As the tumors grow, the CD133+ population proliferates and invades but does not grow within the tumor core. During this time, microglial infiltration into the tumor center decreases while it increases along the tumor periphery, and vessel density decreases while vessel size increases.

Additional staining demonstrated the ability of the hybrid model to develop modified features distinct from tumors derived from U87, LN229, or GBM-8 cells alone. U87 and LN229 formed much of the tumor core in the hybrid model, and Ki-67 (proliferation) staining was lower in this region than in more infiltrative zones (Supplementary Fig. 7). This differed from the robust Ki-67+ staining we detected in tumors developed from U87 or LN229 alone (Supplementary Fig. 2). Furthermore, while no single tumor model expressed high levels of hypoxia-inducible factor 1 alpha (HIF1-α), pockets of cells expressing HIF1-α were detected in the heterogeneous model, with highest expression just outside the less proliferative areas. The high cell density in these regions and the hypoxia designated by this high HIF1-α expression are indicative of high-grade gliomas.^23,24

The above ex vivo and in vivo characterization suggests that this hybrid model displays heterogeneity in TRAIL and chemotherapeutic resistance, proliferation, invasion, blood vessel structure, cancer stem cell populations, immune infiltration, and hypoxic regions that cannot be captured by models using each human GBM cell line alone. The improved clinical accuracy of this model makes it ideal to investigate the impact of these parameters on NSC therapy.

Evaluation of Different tNSC Therapy Approaches in the Hybrid Tumor Model

Using the hybrid tumor approach described above, we investigated the strengths and limitations of NSC-TRAIL and NSC-TK therapies using mouse NSCs as well as novel human skin–derived iNSCs as the cell-based drug carriers. In each experiment, only one dose of tNSCs was implanted to most effectively discern differences in tumor regrowth in response to treatment.

We first tested NSC-TRAIL and NSC-TK therapies against hybrid tumors using an optimized mouse NSC delivery strategy. Mouse NSCs were used to increase NSC persistence in this initial optimized model. NSCs and tumor cells were mixed and co-implanted into the brain (Fig. 3A) to eliminate the initial cell/drug delivery hurdle and more easily determine patterns of drug resistance and tumor recurrence. Tumors implanted with a low dose of NSC-TRAIL (1 tNSC: 4 tumor cells) displayed a significant inhibition in growth rate (P < 0.01; Fig. 3B, C). Nevertheless, post mortem analysis showed that both TRAIL-resistant (LN229; red) and TRAIL-sensitive (U87 and GBM-8; green) tumor cell types eventually recurred (Fig. 3D). When the dose of NSC-TRAIL was increased 4-fold, optical imaging did not show a marked increase in overall GBM killing. Post mortem fluorescence analysis showed the higher NSC-TRAIL dose decreased four-fold, but did not eradicate, the population of TRAIL-sensitive cells relative to TRAIL-resistant cells in the recurrent tumor (Fig. 3E). These results show that NSC-TRAIL therapy is able to induce dose-dependent killing, although TRAIL-resistant cells are eventually able to evade therapy and drive tumor recurrence even when the dose of TRAIL is increased.

Fig. 3.

Treatment of heterogeneous tumor co-implanted with NSCs. (A) Schematic showing co-implantation of heterogeneous tumor and NSCs. (B) In vivo tumor growth curve, n = 13 untreated mice, and n = 4 for all treated groups. (C) Kaplan–Meier plot quantifying mice with tumors below a tumor growth threshold of 1e8 via BLI; **P < 0.01. (D) TRAIL-sensitive (green; U87 and GBM-8) and resistant (red; LN229) cells are both present 39 days after NSC-TRAIL implant, n = 6. (E) TRAIL-sensitive cells decrease in abundance relative to TRAIL-resistant cells when NSC-TRAIL dose is increased, n = 5. (F) Minimum residual disease eventually regrows long after NSC-TK/GCV treatment ends.

The DNA intercalator ganciclovir triphosphate (GCV-TP), which acts nonselectively on dividing cells by competing with deoxyguanosine triphosphate and subsequently inhibiting DNA polymerases,²⁵ has not shown the resistance issues associated with TRAIL when delivered via tNSCs. NSCs expressing TK, which catalyzes the prodrug GCV into GCV-TP, have shown preclinical efficacy against GBM after systemic injection of GCV.^14,26 NSC-TK cells were therefore co-implanted with hybrid tumors at a ratio of 1 NSC-TK to 4 tumor cells to gauge NSC-TK antitumor effect. Tumors initially grew before GCV initiation on day 4, with daily doses of GCV given for 12 days. Tumor volumes decreased long after GCV treatment ended, reaching a trough on day 25 after implantation at 20% of initial (day 0) size. The majority of these tumors did not show significant growth (above a BLI of 1E8) until day 72, significantly longer than tumors treated with NSC-TRAIL (P < 0.01; Fig. 3B, C). On day 72, mice were sacrificed and the recurrent tumor was imaged, showing regrowth of the residual tumor (Fig. 3F). These results suggest that GCV-TP catalyzed by NSC-TK was more effective than NSC-TRAIL therapy in this idealized therapeutic scenario, emphasizing the impact of TRAIL-resistant GBM cells on the therapeutic success of tNSCs.

To evaluate the clinical relevance of the hybrid modeling approach in the setting of personalized medicine, we next used novel human skin–derived iNSCs as the drug carriers. In the clinical setting, the ideal tNSC carrier would be autologous and generated from patient cells (eg, fibroblasts) to avoid rejection. Capitalizing on next-generation advancements in cellular reprogramming, we have previously converted human skin fibroblasts into tumor-homing iNSC therapies as a novel approach to personalized cell therapy.^5,6

To even more closely mimic the clinical scenario, hybrid GBM tumors were implanted into mice and allowed to establish and invade for 8 days before injection of therapeutic iNSCs into the established tumors (Fig. 4A). We first focused on iNSC-TK, as it was highly effective in the mouse NSC studies described above. Induced NSC-TK cells were directly injected into the established tumors and allowed to migrate for 3 days before daily doses of GCV were initiated. GCV treatment slowed the rapid progression of the hybrid tumors; however, unlike the dramatic tumor kill observed above, the growth delay by iNSC-TK therapy in this established tumor model was minimal and transient (Fig. 4B), and failed to statistically increase survival compared with untreated mice (19 ± 9 vs 13 ± 3 days) (Fig. 4C). In contrast, iNSC-TRAIL therapy resulted in a large, sustained decrease in tumor growth that resulted in tumors that were 16-fold smaller than untreated controls at 13 days post-iNSC injection (P < 0.001). This led to a significant increase in survival, with iNSC-TRAIL treated mice surviving an average of 33 ± 8 days (Fig. 4B, C).

Fig. 4.

Treatment of established heterogeneous tumor. (A) Induced NSC preparation, treatment, and brain processing schematic. (B) Established tumor growth curve; n = 4 for all groups. (C) Mouse survival curve, **P < 0.01. Low and high-magnification images at treatment endpoints of: (D) tumor growth (fluorescence); (E) Ki-67 immunofluorescence; (F) HIF1-α IHC (3,3′-diaminobenzidine).

At treatment endpoints for untreated mice and mice treated with iNSC-TK, we observed large solid tumors with only a small amount of invasive cells reaching the contralateral hemisphere (Fig. 4D). IHC staining showed that several less proliferative regions were present in solid tumor areas (Fig. 4E). These less proliferative regions contained a lower density of blood vessels (Supplementary Fig. 8), some of which were very large. Densely packed cells surrounding the nonproliferative areas stained highly positive for HIF1-α (Fig. 4F).

The recurrent tumor in mice treated with iNSC-TRAIL displayed a substantially modified morphology. The recurrent tumor appeared more invasive, with substantially more tumor cell burden present in the corpus callosum and contralateral hemisphere (Fig. 4D). This resembles the increase in invasive cell populations after solid tumor treatment in human GBM patients.⁴ The solid tumor remnant, primarily composed of TRAIL-resistant LN229 cells, still displayed abnormal vasculature, but the vascular density within the diffuse tumor burden remained more similar to areas of healthy brain. Additionally, the pockets of HIF1-α+ cells present in the large solid tumors were not present in those recurrent tumors previously treated with iNSC-TRAIL. These results show that successfully managing solid tumor growth with a single dose of iNSC-TRAIL increased survival and induced several molecular and morphological changes in the tumor; however, diffuse tumor was still able to recur. Future studies using this model could help elucidate whether recurrent tumor cells were able to escape treatment due to insufficient persistence of iNSCs, migration away from treated areas of the brain, acquired resistance, or other means.

Patient-Matched Hybrid Tumor Model

The hybrid tumor modeling approach is not limited to just one type of cell mixture. To more broadly test our therapies, we built an additional model using invasive and non-invasive subclones from the same patient GBM biopsy. Previously, the GBM biopsy material was separated and cultured in different media that selected for 2 distinct cell types, which were separately characterized in vivo.^15,27–29 We obtained limited-passage primary cells that had either been selected in stem cell media containing epidermal growth factor and fibroblast growth factor (G-EF cells) or in Dulbecco’s modified Eagle’s medium with fetal bovine serum (G-FBS cells) from the Wakimoto lab. Our initial assays on brain slices showed that G-EF cells (red) quickly grew and migrated across the corpus callosum, while the G-FBS cells (green) grew slowly and densely (Fig. 5A–C). Co-culture assays on brain slices with NSC-TRAIL showed that both G-EF and G-FBS had similarly high sensitivities to TRAIL. A 1:1 ratio of NSC-TRAIL:G-EF or G-FBS cells killed 98% or 99% of cells, respectively (Fig. 5D). Treatment with temozolomide on brain slices showed heterogeneity in sensitivity, as G-FBS was much more resistant than G-EF (Supplementary Fig. 9).

Fig. 5.

Patient-matched tumor studies. (A) Images and (B) quantification of patient-matched tumor foci migration on brain slice. (C) Growth of cells in tumor foci from 3 to 72 h. (D) Quantification of TRAIL sensitivities of G-EF (left) and G-FBS (right) cells on brain slices seeded with different ratios of NSC-TRAIL. (E) Interactions of patient-matched cells on brain slice, mixed at 1:1 G-EF:G-FBS (top) or 1:5 G-EF:G-FBS (bottom) with (F) quantification of changes in G-EF migration compared with G-EF seeded alone. (G) 1:5 G-EF:G-FBS tumor growth in vivo, t = day 10. (H) G-EF tumor growth in vivo, t = day 10. (I) Patient-matched workflow schematic. (J) Tumor treatment growth curve; n = 4 for untreated and iNSC-TRAIL groups and n = 3 for iNSC-TK group. (K) In vivo survival curve. (L) Fluorescent images of treated tumors in vivo at time of euthanasia. Percent of tumor growth in ipsilateral hemisphere quantified at bottom left of each image. *P < 0.05, **P < 0.01, ***P < 0.001.

Brain slices were then used to determine a G-EF:G-FBS ratio that would yield a heterogeneous tumor with both solid and invasive regions. G-EF:G-FBS ratios of 1:0, 1:1, and 1:5 were seeded on brain slices and analyzed after 6 days. The outward migration rate of the G-EF cells was significantly decreased compared with G-EF cells alone when mixed at the 1:5 ratio, but not the 1:1 ratio (Fig. 5E, F). When implanted in vivo, the 1:5 ratio (600k cells/mouse) yielded a tumor with a core of G-FBS cells and a dense leading edge of G-EF cells 10 days after implant (Fig. 5G). When implanted alone, G-EF cells grow diffusely and do not form the solid tumor components observed in patient GBM (Fig. 5H).

In an orthotopic treatment study, iNSC-TRAIL or iNSC-TK were injected into patient-matched tumors 8 days after implant, both resulting in a significant delay in tumor growth and increase in survival time, although iNSC-TRAIL was significantly more effective than iNSC-TK (Fig. 5J, K). IHC staining of the untreated tumor at treatment endpoint showed active proliferation throughout the brain, but blood vessel structure and hypoxia were less significantly modulated than in the model containing U87, LN229, and GBM-8 (Supplementary Fig. 10). Untreated tumors were invasive with a densely packed core and 80% of the tumor signal in the ipsilateral hemisphere, causing significant midline displacement of the brain (Fig. 5L). In contrast, tumors treated with iNSC-TRAIL contained only 56% of the tumor in the ipsilateral hemisphere (P < 0.001 compared with untreated), showing a more evenly distributed recurrent tumor without a dense primary mass.

Combination tNSC Therapy

In the clinical setting, chemotherapy regimens commonly consist of combination strategies to overcome the heterogeneous response of cancer cells. We next explored whether targeted tumor cell killing via TRAIL could be combined with the general killing mechanisms of TK/GCV to enhance treatment response. We thus used hybrid tumors to test NSC combination therapy in the 3 experimental paradigms used in the studies described above.

Co-implantation of tNSC and tumor

Combination therapy allowed initial NSC-TRAIL treatment to limit the growth of tumors before the onset of GCV. Two days after GCV dosing began, tumors had grown nearly 6-fold in mice treated with NSC-TK only, but in mice co-implanted with both NSC-TRAIL and NSC-TK, tumors had already decreased in size (Fig. 6A, inset). Over the next weeks, the efficacy of NSC-TK monotherapy decreased the separation between these 2 groups, although by day 72 after implant, tumors receiving NSC-TK monotherapy had grown significantly larger than tumors receiving combination tNSC therapy (P < 0.01).

Fig. 6.

Combination therapy. (A–F) Tumor growth curves and survival plots with combination therapy groups added; n = 5, 5, and 7 mice in combination therapy groups, respectively. Inset graph in (A) compares 1:4 NSC-TK:tumor (green) with 1:1:4 NSC-TK:NSC-TRAIL:tumor (purple). (G) Table comparing significance of all treatments compared with untreated control. (H) Comparison of days in complete remission (tumor signal below background) for all patient-matched treatment groups. *P < 0.05, **P < 0.01, ***P < 0.001.

Established hybrid tumors

Interestingly, iNSC combination therapy in the initial established hybrid tumor paradigm was not more effective than iNSC-TRAIL treatment alone (Fig. 6C, D). In contrast, in the patient-matched model, combination therapy led to longer tumor growth inhibition (Fig. 6E) and an enhanced and more consistent remission period, with all mice in this group spending at least 23 days in remission (P < 0.05 vs iNSC-TRAIL monotherapy; Fig. 6H). The long remission provided by combination therapy delayed the onset of tumor recurrence, but did not translate to a statistically significant increase in mouse survival (P = 0.054) compared with iNSC-TRAIL treatment alone (67 ± 4 vs 63 ± 3 days) (Fig. 6F).

Discussion and Conclusions

Preclinical GBM models need to closely mimic the challenges of treating GBM in patients in order to support rational and targeted drug development. Improvement of preclinical models should thus focus on clinically relevant aspects of GBM growth, treatment response, and recurrence, such as intratumoral heterogeneity in cellular migration/invasion patterns, drug sensitivities, and tumor cell morphologies/phenotypes. In this context, we report novel hybrid GBM tumor models that recapitulate several key aspects of patient GBM, including heterogeneity in TRAIL and chemotherapeutic sensitivity, proliferation, migration patterns, hypoxia, blood vessel structure, cancer stem cell populations, and immune infiltration.

We first built a model by repurposing the widely used U87 tumor line to build a dense, hypoxic tumor core interspersed with LN229, with the patient-derived GBM-8 line added to create highly invasive populations. Recognizing the limitations of high-passage established cell lines such as U87, we generated a second patient-matched model that was the first to combine 2 distinct types of tumor cells from the same human patient. We found that each model had unique strengths, and began to demonstrate the power of hybrid models to serve as a guide for future developments that could incorporate transcriptionally defined cells of differing GBM subtypes to better recapitulate the genetic diversity in GBM.

In addition, we show the utility of tumor cell implantation onto live organotypic brain tissue slice explants as a “staging ground” for in vivo studies. While a brain slice cannot recapitulate all aspects of the in vivo situation, we show that the brain slice environment sufficiently resembles brain stroma to better predict the growth and invasiveness of disparate GBM tumor cell types compared with conventional in vitro culturing systems. This enables the brain slice to be used as an initial assay for identifying relevant cell types and cell interactions for building in vivo models and for rapidly surveying a broad range of possible experimental protocols and parameters before refinement in much more labor-intensive and expensive in vivo models.

In an initial application of this novel hybrid tumor approach, we have found unexpected differences among 3 treatment approaches, speaking to varying mechanisms of TRAIL- and TK/GCV-mediated tumor killing. NSC-TRAIL quickly kills both adjacent and more distant tumor cells via TRAIL secretion, but TRAIL-resistant cell populations can escape treatment. Escape from NSC-TK treatment occurs in a different manner. Catalyzed GCV-TP effectively kills most adjacent dividing cells, but treatment fails when tNSCs are remote from tumor cells. While the contrasting strengths and weaknesses of NSC-TRAIL and NSC-TK should be amenable to effective combination therapy, our findings suggest that simply introducing both cell types simultaneously in the same tumor location will not necessarily lead to significantly improved clinical benefit.

Analysis of these data suggests several future approaches that could increase the efficacy of combination therapy. First, delaying the onset of GCV dosing could help iNSC-TK cells migrate farther from the implantation site before GCV-induced cytotoxicity of iNSCs begins, as well as decrease the number of TRAIL-sensitive cells redundantly killed by GCV-TP. Second, implanting iNSC-TRAIL cells in the tumor periphery and iNSC-TK cells nearer the tumor center could further decrease redundant overlap of treatments and decrease TRAIL-related toxicity toward iNSC-TK cells. Third, intrathecal administration of iNSCs may help quickly spread cells into distant brain regions. Furthermore, these models will be effective in analyzing more clinically relevant treatment/regrowth after partial tumor resection, as well as assessing treatment improvement as iNSC persistence and migration ability continue to increase.

In conclusion, we have developed a novel hybrid GBM tumor modeling approach that recapitulates key aspects of growth and drug response, and that can be derived from tumor cell lines as well as from patient biopsies. In an initial application of this new approach, we have identified unexpected strengths and limitations of two major avenues of tNSC therapy that form a basis for future drug optimization, including critical preclinical studies for combination tNSC therapies. As therapies are improved, this work lays the foundation for developing increasingly sophisticated hybrid GBMs that will continue to challenge therapeutic strategies and provide insights toward clinical success.

Funding

A.B.S. is supported by a TL1 fellowship from the National Center for Advancing Translational Sciences, NIH, grant TL1TR002491. This work is supported by National Institutes of Health grant NIHR01NS099368 and by the Eshelman Institute for Innovation.

Conflict of interest statement. Shawn Hingtgen is the founder of Falcon Therapeutics.

Authorship statement. Study design: Andrew Satterlee, Don Lo, and Shawn Hingtgen. Initial preparation of brain slice tissue: Denise Dunn. In vitro, ex vivo, and in vivo experiments and downstream analysis: Andrew Satterlee. Patient MRI acquisition: Simon Khagi. Secure Funding: Don Lo and Shawn Hingtgen.