Harnessing Human Stem Cells for the Treatment of Glioblastoma – Twenty Year Perspective – Review of Preclinical and Clinical Studies: 2000–2020

The potential of Neural Stem Cells (NSCs) to provide therapeutic benefit for a variety of neurologicaldisorders, including brain malignancies, has been long recognized and has inspired many scientists to design,test and successfully demonstrate that NSCs are efficient and effective therapeutic agents. Glioblastoma, thedeadliest form of primary brain tumor, despite extensive and sustained efforts to find better therapies, remainsa disease without cure, with a median survival after diagnosis of less than two years. Treatment resistance inglioblastoma is in large part attributed to limitations in the delivery and distribution of therapeutic agentsadministered either systemically or directly into the tumor due to the highly invasive nature of this cancer andits abnormal intratumoral vasculature. Stem Cells (SCs) have an innate tumor-tropic migratory behavior, canbe modified to deliver a variety of therapeutic agents and efficiently distribute their cargo into brain tumors,pursuing invading streams of tumor cells, deep into the brain parenchyma. Over the last twenty years,numerous preclinical trials have demonstrated the feasibility and efficacy of SCs as antiglioma agents, leadingto the development of trials to test these therapies in the clinic. In this review we present and analyze thesestudies and discuss mechanisms underlying their beneficial effect, highlighting experimental progress,limitations and the emergence of promising new therapeutic avenues. We hope to increase awareness of theadvantages of using SCs for the treatment of glioblastoma and inspire further studies that will lead toaccelerated implementation of effective therapies.


4/35
The CXCL12 chemokine and its cognate signaling receptor CXCR4, originally identified as a homeostatic chemokine/receptor complex that regulates hematopoietic stem cell (HSC) trafficking, activates one of the main signaling pathways that drives NPC migration in the developing CNS 38 , and also guides migration of GSCs 39 . CXCL12 is highly expressed in the SVZ 39 and in the bone marrow HSC niche where it promotes migration of primitive hematopoietic cells 40 . Production of CXCL12 is increased inside tumors in response to hypoxia and irradiation 41 . High levels of CXCL12 are also found in pseudopallisades 42 , a hallmark histological feature of WHO grade IV GBM, regions that also harbor GSCs 43 . GSCs and NSCs thus have shared migratory behavior, due at least in part to their responsiveness to CXCL12 chemotaxis.
Transplanted rodent NSCs exhibit an intrinsic migratory behavior, following developmental pathways 44,45 . Human NSCs, generated from the periventricular telencephalic region of a 15-week old human fetus and immortalized by retroviral transduction with v-myc, were injected into the lateral ventricles of newborn mice and their distribution into the brain was analyzed over time, for up to 5 weeks. As early as 24 hours after injection, the human NSCs had joined migrating endogenous mouse neuroblasts in the rostral migratory stream. At three weeks these cells were found in the olfactory bulb, differentiated into neurons 46 . Implanted human NSCs were also found to differentiate into oligodendrocytes, astrocytes or cerebellar granule cells depending on the site of transplantation, demonstrating their multipotency. Interestingly, it was observed that, while the immortalized human NSCs proliferated well in vitro, once transplanted into mouse brains, expression of v-myc was downregulated and the cells stopped dividing 46 . The underlying cause of this behavior is not known, but is assumed to be a consequence of normal developmental mechanisms that induce mitotic arrest during differentiation 46 . Similar observations were made with mouse immortalized NSCs 47,48 .
Considering the pressing need to find better strategies for intratumoral distribution of therapeutic agents, scientists tested immortalized NSCs in a mouse model of glioblastoma 49 . It was elegantly demonstrated that when murine NSCs (C17.2 cells that were derived from the cerebellum of 4 day old mice and immortalized with v-myc 50 ) were injected into the tumor or at a distance from the tumor (in the contralateral hemisphere, intraventricular, or intravenously), distributed widely throughout the tumor and followed invasive streams of glioma cells deep into the brain parenchyma, albeit with lower efficacy following systemic administration 49 . This tumor tropic migratory behavior was not altered when cells were transduced to express a therapeutic enzyme, proof of principle that NSCs have the potential to deliver therapeutic agents in glioblastoma.

ORIGINS OF STEM CELLS USED FOR THE TREATMENT OF BRAIN TUMORS
In addition to NSCs, Mesenchymal Stem Cells (MSCs) as well as induced NSCs (iNSCs), derived from pluripotent Stem Cells or transdifferentiated from somatic cells, have been shown to display tumor-tropic behavior and distribute extensively in intracranial gliomas. Many preclinical trials have tested these cells for 5/35 their ability to distribute throughout the tumor and deliver a variety of therapeutic agents: bioactive proteins, viruses, cytokines, antibodies, toxins or nanoparticles. The majority of these studies tested the use of human SCs and the rest, rodent SCs. Immortalized human NSCs were most often used, followed by MSCs derived either from bone marrow (BM), adipose tissue, umbilical cord or amniotic fluid, and one study used iNSCs transdifferentiated from fibroblasts. Rodent SCs were employed similarly (SupplementaryTable 1).
Interestingly, all but one study with human NSCs used cells that were obtained from fetal human brain and immortalized with the v-myc oncogene, for easy in vitro propagation and to prevent terminal differentiation when exposed to serum 51 .
One line, the HB1.F3 cell line was derived from the telencephalon of a human female 15-week gestation fetus, initially propagated on poly-lysine-coated tissue culture plates in DMEM supplemented with 5% horse serum and transduced with a replication incompetent retroviral vector encoding v-myc. Successfully transduced cells were plated at clonal density and several clones were selected, including HB1.F3 52 . The HB1.F3.CD cell line was derived from HB1.F3 cells that was transduced to express cytosine deaminase(CD) 49 , and subsequently extensively characterized and FDA approved for use in clinical trials 53  plates in serum free defined NSC media supplemented with EGF, FGF 54 . ReNCells were also immortalized with a retrovirus encoding v-myc (ReNCell VM) or c-myc (ReNCell CX) 54 and are currently commercially available. ReNCells differentiate upon growth factor withdrawal into early neurons expressing III-Tubulin, dopaminergic neurons expressing tyrosine hydroxylase, GFAP+ astrocytes and oligodendrocytes expressing Galactosylceramide 54 . Despite expression of the myc oncogene, these cell lines have not been shown to produce tumors in the brains of experimental animals or in clinical trials, and most of them do not persist in the brain. The differentiation potential of the HB1.F3.CD line has not recently been tested, nonetheless, it was demonstrated to be safe and effective in trials for glioblastoma.
The main function of NSCs used for the treatment of GBM is to migrate deep into the tumor and deliver therapeutic agents, most often as part of a suicide mission, their ability to differentiate into neurons, oligodendrocytes or astrocytes becoming less relevant. It was demonstrated that NSCs derived from different parts of the mouse and human CNS have different proliferation and differentiation potential and express transcription factors specific for their region of origin 32,55,56 . However, it was also reported that upon culture, expression of many transcription factors that indicate positional identity of the NSCs was downregulated or lost and these changes altered the differentiation potential of these cells 57,58 . It is likely that such changes affect functional aspects of NSCs used for GBM as well; rigorous mechanistic studies in this direction may lead to findings that will improve the efficiency of therapeutic NSCs.

6/35
MSCs are cells capable of differentiating into tissues that are derived from the embryonic mesoderm, such as adipose tissue, cartilage, bone and muscle 59 . MSCs can be found in a great variety of tissues.
Therapeutic MSCs have been derived from bone marrow (BM), adipose tissue, umbilical cord blood, and even placenta and have the advantage of being abundant, easy to isolate and propagate, have high tumor-tropic migratory potential and can be donor-matched. Some studies report that unmodified MSCs can promote tumor growth in other cancers 60 . So far, preclinical studies using MSCs for the treatment of glioblastoma have not reported tumor formation (Supplementary Table 1).
NSCs appeared to be better suited in their ability to distribute throughout brain tumors and deliver oncolytic viruses when compared to MSCs 61 ; also, HB1.F3.CD were more efficient in this process than ReNCells 62 . The underlying mechanisms for this therapeutic advantage have so far not been explored.
Regrettably, the authors 62 didn't specify whether the ReNCells used were midbrain or cortex derived (ReNCell-VM or ReNCell-CX).
For clinical use and to achieve FDA approval, each newly established SC line must be prepared following Good Manufacturing Practice protocols and requires extensive testing of safety and therapeutic properties according to standards of Good Clinical Practice. When rigorous procedural practices are not followed, as was the case in some for profit clinics that carry out unproven SC interventions for a variety of conditions as well as in the absence of regulatory oversight as encountered in some countries, administration of SCs can lead to major complications, including tumor formation 63,64 , embolism, vision loss, infectious events, autoimmune reactions, stroke, brain hemorrhage and even death 14 .
Induced NSCs (iNSCs), generated by transdifferentiation of somatic cells by transient expression of specific transcription factors, can be derived from the patient's own cells, evading a potential immune response, as may be encountered when using allogeneic NSCs. Both mouse and human iNSCs have been shown effective in preclinical studies for glioblastoma 65,66 . The primary safety concern for therapeutic NSCs is their potential to induce tumor formation, especially when implanted into the brain of patients with established tumors that generate a microenvironment permissive for tumor growth. It was demonstrated in a syngeneic mouse model that iNSCs were safe in this respect, and did not induce tumor formation, unlike induced pluripotent SCs (iPSCs) and embryonic SCs (ESCs) which generated aggressive, deadly tumors 67

. Induced
NSCs have not yet been tested in the clinic. Considering the rapid progression of glioblastoma, concerns arise whether such an approach would allow for a timely and in depth quality and safety analyses to validate and expand autologous iNSCs to the necessary amounts needed for therapeutic use, as well as the costs associated with such an endeavor. Nonetheless, transdifferentiation of somatic cells may be a powerful tool for drug delivery. If factors that promote optimal NSC migration and intratumoral distribution are uncovered and engineered in iNSCs, with thorough characterization and testing, the therapeutic NSC toolbox could be considerably expanded. Detailed strategies for generation of iNSCs have been described in a recently published review 68 .  The first preclinical studies using SCs for the treatment of malignant glioma emerged twenty years ago,   testing four different therapeutic strategies: enzyme/prodrug 49 , oncolytic virus 69 , cytokine therapy 70 and delivery of pro-apoptotic molecules 71 . Today, these mechanisms are still being tested, optimized, fine-tuned and combined with other treatment modalities to overcome limitations and improve efficacy. Pilot/feasibility and phase I clinical trials using NSCs loaded with oncolytic viruses or delivering enzymes for intratumoral prodrug conversion have recently been completed. Exciting new therapeutic avenues have been opened by the tremendous progress in bioengineering technologies, generating nanoparticles and nanorods that can be conjugated to glioma-tropic SCs for drug delivery and targeted photo-thermal ablation therapy. We describe some of these studies in the following sections, highlighting experimental progress, limitations and efforts to translate these strategies into the clinic.

STEM CELLS IN ENZYME/PRODRUG STRATEGIES
Enzyme/Prodrug strategies for cancer treatment have long been pursued in an effort to artificially create selective and local cytotoxicity for tumor cells, while leaving normal cells unharmed 72 . The most widely used combinations are the Herpes Simplex Virus-Thymidine Kinase (HSV-TK) with Gancyclovir (GCV) and the bacterial Cytosine Deaminase (CD) with 5-Fluorocytosine (5-FC). The HSV-TK enzyme converts GCV into Ganciclovir monophosphate that is further phosphorylated to Gancyclovir triphosphate, a toxic antimetabolite that undergoes erroneous incorporation into DNA, leading to the death of dividing cells ( Figure 1A). Cells lacking HSV-TK can still be targeted for apoptosis through a phenomenon called the "bystander effect", which entails transport of the active drug through GAP junctions from neighboring cells, enhancing the cytotoxic response. Similarly, the bacterial enzyme CD is able to convert the nontoxic prodrug 5-FC into the powerful cytotoxic compound 5-fluorouracil (5-FU), that primarily inhibits the production of thymidine, which is required for DNA replication, thus killing dividing cells exposed to it ( Figure 1B). Enzyme/Prodrug gene therapy strategies using viral vectors have been extensively explored for the treatment of glioblastoma for over 40 years, but still face many challenges, mainly related to the distribution of the viruses into the tumor 73 .The high migratory capacity of NSCs to distribute throughout the tumor and tumor satellites may serve to improve this therapeutic strategy. TS converts dUMP to dTMP. Inhibiting TS results in nucleotide imbalance, excess dUTP and lack of dTMP, leading to DNA damage. FUTP (5-fluorouridine 5'-triphosphate) is extensively incorporated into nuclear and cytoplasmic RNA, leading to impaired RNA synthesis, stability, processing and methylation. FdUTP (5-fluoro-2'-deoxyuridine 5'triphosphate), when incorporated into DNA, inhibits DNA elongation and induces DNA fragmentation. 5FU is deactivated and converted to FUH2 (fluoro-5,6-dihydrouracil) through the catalytic action of dihydropyrimidine dehydrogenase (DPD), the initial and rate-limiting step in the catabolism of 5-FU. FUH2 can be further degraded to FUPA (fluoro-ureidopropionate) and subsequently to FBAL (fluoro--alanine). (C) Carboxyl esterase (CE) converts the water soluble compound CPT-11 into the more potent, lipophilic metabolite SN38. During the DNA synthesis phase of the cell cycle Topoisomerase 1 (TOP1) attaches to the 3' end of the cleaved DNA and forms a reversible DNA-TOP1 cleavage complex (TOP1cc). SN-38 binds to TOP1 and stabilizes this complex, halting DNA synthesis and leading to the accumulation of single strand DNA breaks that trigger apoptosis.

9/35
A proof of principle study, using an in vitro co-culture model, demonstrated that NSCs expressing HSV-TK can induce the death of glioma cells when combined with GCV, and this cytotoxic effect was dependent on cell to cell contact and on the presence of the GAP junction protein Connexin 43 74 . The bystander effect of HSV-tk delivered by NSCs was also demonstrated in a mouse model in which C6 glioma cells were implanted with NSCs that were transduced with HSV-tk (NSC-tk) 75 . Long-term survival was observed in all animals treated with NSC-tk, whereas all control animals succumbed before 30 days. It was also demonstrated that human BM derived MSCs transduced to express HSV-TK and then injected into mouse brains at the time of glioma cell implantation, induced long term survival in all animals 76 . In a rat glioma model, the same MSC-TK cells improved survival of tumor bearing animals with 40% of animals becoming tumor-free 76 . These studies demonstrate that delivery of HSV-TK by either NSCs or MSCs provide a significant survival benefit in animal models of glioblastoma. One caveat associated with such studies comes from the simultaneous administration of tumor cells and therapeutic SCs, scenario that cannot be applied in the clinic. Other animal studies have administered SCs at later time following tumor cell implantation, and demonstrated therapeutic effect, albeit with lower efficacy (Supplementary Table 1

).
A pioneering preclinical study that illustrated the migration of mouse and human NSCs towards intracranial gliomas 49 , tested the ability of immortalized mouse NSCs (C17.2) 50 that were transduced with cytosine deaminase (CD-NSC) to induce glioma cell death in the presence of 5-FC. Results demonstrated the feasibility of this approach and tumor reduction upon intratumoral administration of CD-NSCs; proliferating NSCs were also killed by the treatment 49 . Human NSC cells HB1.F3 were also transduced to express CD, generating HB1.F3-CD cells that were tested for their migration potential and tumor killing efficacy in mice injected with Daoy medulloblastoma cells 77 . Survival was not analyzed in this study, however HB1.F3-CD cells injected into established tumors, when combined with 5-FC, were able to significantly reduce the tumor volume (by 74%) after three weeks of treatment, demonstrating a strong bystander effect from the CD producing NSCs 77 .
Adipose tissue derived MSCs (AT-MSC) were transduced with yeast derived Cytosine Deaminase (CDy) and uracil phosphoribosyltransferase (UPRT) generating CDy-AT-MSCs, that migrated towards gliomas when implanted at a distance 78  When administered to mice with intracranial tumors and combined with 5-FC, tumor volume was significantly decreased in animals receiving the highest dose of NSCs (1x10^5) 53 . This study highlights the fact that therapeutically efficacy is intimately dependent on the number of NSCs available in the tumor. The pilot/feasibility clinical study designed to use these cells in patients with recurrent high grade glioma (rHGG) (NCT01172964), planned a classic 3+3 dose escalation regimen (1x10^7 cells to 5x10^7 and 5-FC from 75 to 150 mg/kg/day, administered for 7 days, starting 4 days after surgery) with injection of NSCs into the walls of the resection cavity. Results from 15 patients found steady state levels of 5-FU in the brain, much higher than in the blood, indicating local production of 5-FU by the NSCs 82 . Overall, administration of HB1.F3.CD was safe, resulted in no detection of NSCs outside the brain and elicited no humoral immune responses. The median overall survival (OS) at the highest dose level was 15.4 months compared to patients on doses 1-2, with OS of 2.9 months 83 . Distribution of NSCs was analyzed in postmortem brains by v-myc PCR. Several selected sites from brain sections from a male patient, who had died at some time after surgery, tested positive for v-myc by PCR, indicating presence of HB1.F3.CD cells. In situ hybridization with an XY probe (the NSCs are female) on adjacent sections confirmed the presence of the injected NSCs and showed that they were not proliferating. It was estimated that the NSCs traveled about 11 cm from the site of injection 83 . These results are indeed exciting, proof of safety and of principle, with added optimism for efficacy and identification of live, possibly still therapeutic cells, more than 2 months after administration. The follow up phase I study (NCT02015819) started in 2014, designed to include 18 patients with rHGG, has been modified to include an intraventricular Rickham catheter, placed at the time of surgery, to be used for the administration of subsequent doses of NSCs every two weeks, followed each time by a 7-day course of oral 5-FC. Leucovorin was also added to the protocol to enhance the cytotoxic effects of 5-FU. This study is still ongoing.

11/35
Another enzyme/prodrug combination used for cancer treatment is Carboxyl Esterase (CE) and CPT-11 (Irinotecan), a water soluble semisynthetic derivative of Camptothecin. CPT-11 has antitumor activity on its own, however its lipophilic metabolite SN-38 is a more potent cytotoxic agent. Camptothecin and its derivatives induce DNA damage during the S-phase of the cell cycle by inhibiting Topoisomerase I (TOP-1), forming a stable ternary DNA-TOP1 cleavage complex (TOP1cc) and inducing apoptosis. CPT-11 is metabolized to SN-38 by CE, an enzyme naturally present in the liver, albeit at low concentrations ( Figure 1C). High expression of CE in the tumor results in efficient conversion of CPT-11 into SN-38 and high cytotoxic effect, mitigating systemic toxicity. Human Adipose Tissue derived MSCs transduced to express the rabbit CE (hAT-MSC.rCE) were tested in a rat brainstem glioblastoma model, in a protocol that entailed two injections of hAT-MSC.rCE at 12/35 HB1.F3.CD.hCE1m6 cells were as efficient as HB1.F3.CD.rCE in converting CPT-11 and had similar migratory behavior and cytotoxic activity against glioma cells. In vitro degranulation assays with the two enzymes using human peripheral blood mononuclear cells, indicated that hCE1m6 had lower immunogenic potential than the rabbit counterpart 88 .
The distribution of HB1.F3.CD.hCE1m6 cells in intracranial glioma tumors was also analyzed following administration into the lateral ventricles, demonstrating that these cells migrated efficiently towards tumors, even in the case of multifocal tumors, located in different hemispheres 89 . NSCs tended to locate around the tumor margin but were also found within the tumor core; some cells aggregated in the ventricles, and this was thought to be due to the speed of the injection or to a cell suspension that was too concentrated (up to 4x10^5 cells in 2 ul). It has become evident that optimizing delivery of therapeutic NSCs for maximal tumor coverage necessitates very careful attention to all procedural details. An important step was made with the large scale quantitative analysis by Barish and collaborators 90 who analyzed the distribution of HB1.F3.CD.hCE1m6 cells, injected either intracranially (i.c.) or intravenously (i.v) into Es1e/SCID mice with orthotopic gliomas. Tumor volumes were calculated from analysis of serial brain sections, number and location of migrating NSCs were scored and the area of the tumor covered was calculated based on the CE molecular weight and diffusion parameters for a tissue like the brain. The careful analysis established, as observed before, that higher doses of NSCs yielded better tumor coverage, but up to a point, when the percentage of actual NSCs present in the tumors declined both in the i.c. and i.v. administration routes. The authors theorize that the rate limiting factors could be tumor dependent or related to the administration technique; high density of cells may lead to aggregation or decreased survival that would limit intratumoral migration and distribution. It was also found that larger tumors attracted more cells, due likely to chemotactic tumor-derived cytokines. Systemic administration required about 10 times more NSCs than i.c injections to reach the same degree of tumor coverage; other parameters of distribution were similar between the two administration routes.
Following in depth analysis of safety, efficacy and pharmacokinetics of the HB1.F3.CD.hCE1m6 cells in combination with i.v. administration of Irinotecan in a mouse glioma model, these cells, identified now as: hCE1m6-NSCs, were approved as an investigational new drug to be tested in a phase I clinical trial for recurrent HGG therapy (NCT02192359) 91 . The study plans to administer hCE1m6-NSCs twice, at two week intervals, and to repeat the cycle every month in the absence of disease progression or toxicities. The goal is to determine the recommended phase II dose of hCE1m6-NSCs, monitor for adverse events, measure SN-38 in the brain and blood, test for immune responses elicited by the allogeneic NSCs and assess potential clinical benefit.

STEM CELLS CARRYING PRO-APOPTOTIC MOLECULES
Dysregulated apoptotic pathways are a hallmark of malignancy. The TP53 tumor suppressor gene, that initiates intrinsic apoptotic pathways in response to cellular stressors, including genomic aberrations and DNA damage is often inactivated in cancer 92 and very frequently in glioblastoma 93 . Malignant cells evade apoptotic 13/35 signals and therapeutically induced DNA damage and continue to proliferate. Activation of the extrinsic apoptotic pathway via ligands, like the TNF related apoptosis-inducing ligand (TRAIL) or FasL, that bind to cell surface death receptors (DRs) can induce activation of the caspase enzymatic cascade leading to apoptosis independent of TP53. TRAIL has therefore been extensively explored as a promising anticancer agent 94,95 , as it specifically targets tumor cells, leaving normal cells unharmed 96 . Soluble TRAIL is very rapidly cleared from systemic circulation, and strategies for extending its stability and delivery by linking it to nanoparticles (NP) or SC carriers have been intensely studied 97 .
If cells express the TRAIL ligand on their surface they can induce apoptosis in neighboring cells. It was demonstrated that antigen specific CD4 T cells expressing TRAIL can kill glioma cells in vitro 98 . Primary mouse NSCs derived from the forebrain of embryonic mice modified to express human TRAIL were able to specifically induce apoptosis in human glioblastoma cells, in mouse orthotopic xenografts, while sparing the NSC population 71 . Also, mouse NPCs (C17.2), transduced to express a secreted form of TRAIL (S-TRAIL) were able to decrease viability of human glioma cell lines through caspase mediated apoptosis 99 . This effect was synergistically increased if NPC-S-TRAIL was combined with miR-21 knock-down. It had been previously reported that miR-21 is highly expressed in human glioblastoma, and its knockdown led to apoptosis of glioblastoma cell lines 100  While many glioma cells are sensitive to TRAIL mediated apoptosis, some, especially GSCs are not.
Resistance to TRAIL has been linked to several mechanisms, including upregulation of anti-apoptotic proteins of the Bcl-2 family, inactivating mutations or epigenetic silencing of Caspase 8, or upregulation of FLIP, a molecule that blocks the formation of the death inducing signaling complex (DISC) 94 . In some cells, resistance to TRAIL-induced apoptosis can be overcome by simultaneous administration of the proteasome inhibitor Bortezomib, which results in enhanced Caspase 8 activation that depends on accumulation of the cyclin dependent kinase inhibitor p21 CIP1 and inhibition of cyclin-dependent kinase activity (cdk1/2) 102   TRAIL was also delivered into the brain of tumor bearing rats via iron oxide magnetic nanoparticles (NPs), and it was shown that NP-TRAIL were effective in inducing apoptosis of glioma cells and significantly extended the MS of tumor bearing rats when compared to rats treated with free soluble TRAIL 105 . NP-TRAIL were also able to induce apoptosis and increase radiation sensitivity and efficacy of the proteasome inhibitor Intratumoral administration of h-iNSC TE -TK when combined with GCV also increased MS of glioma bearing mice. The group further went on to demonstrate that post-surgical treatment with h-iNSC TE -TK cells, encapsulated in ECM hydrogel that was placed into the tumor cavity, followed by GCV administration, a paradigm that more closely resembles the clinic, also resulted in an increased MS of animals 66 .
An interesting development of cell mediated therapy for glioblastoma is represented by the use of modified tumor cells to deliver therapeutic agents into the tumor. This is based from observations of the "selfseeding" behavior of cancer cells, where metastatic cells migrate back toward the primary tumor or cells from the primary tumor migrate toward established metastases 107 . Glioma cells, especially GSCs, share many characteristics with NSCs, including migration and invasion towards CXCL12 and TGFb 108,109 . Genetically modified glioma cells have therefore been tested for their ability to deliver pro-apoptotic molecules for the treatment of glioblastoma and other metastatic cancers 110 . Two strategies for this approach were designed: allogeneic "off the shelf" glioblastoma cells, intrinsically resistant to TRAIL mediated apoptosis and expressing a secreted form of TRAIL (sTRAIL), or autologous glioma cells with CRISPR/Cas9 mediated genetic deletion of the two TRAIL signaling receptors (TRAIL-R1(DR4) and TRAIL-R2(DR5)) and lentiviral expression of sTRAIL.
The cells were also modified to express HSV-TK, to render them sensitive to GCV mediated elimination, yet insensitive to the actions of TRAIL. Mice with established intracranial tumors injected with these cells showed increased median survival, efficiency being further enhanced with GCV administration 110 .
Overall, these studies demonstrate that SC represent efficient vehicles for delivery of TRAIL into the tumors and that the therapeutic effect is dependent on the tumor sensitivity to TRAIL mediated apoptosis. This 16/35 strategy may be useful in combinatorial approaches with agents that mitigate resistance to TRAIL and other synergistic cytotoxic drugs.

STEM CELLS AND ONCOLYTIC VIROTHERAPY
Oncolytic viral therapy uses viral vectors that are either able to selectively replicate in tumor cells and induce tumor cell lysis that is propagated outwards from the site of virus administration, while sparing normal brain cells, or employs viruses that are replication-deficient and are used for the delivery of therapeutic genes 111,112 . In addition to the direct tumor lytic effect, oncolytic viruses strongly stimulate the innate immune  . These antigens activate NK cells that induce direct tumor cell killing through production of granzyme and perforin, and also induce apoptosis through the release of TRAIL, TNF and IFN. DAMPs and TAA activate Antigen Presenting Cells that travel to the draining lymph nodes where they cross present antigens to naïve T lymphocytes and activate them. Activated T helper and Cytotoxic T lymphocytes migrate into the tumor and release cytokines, perforin and granzyme that amplify the cytotoxic effect.
Taken together, these clinical studies are indeed exciting, and provide hope that oncolytic virotherapy may provide therapeutic benefit. Nonetheless, limitations exist, particularly in relation to the limited spatial distribution of the virus following injection, especially in large tumors and those that invade at a distance and necessitate multiple rounds of injections at different coordinates. As a result, many oncolytic viruses (OVs) designed to specifically replicate in glioma cells that were shown to be safe in early clinical trials did not 18/35 advance to phase III studies 123 . Tumor tropic SCs may be able to improve viral distribution and enhance therapeutic efficacy, especially in tumors that are difficult to reach surgically.
Numerous preclinical studies have demonstrated an advantage of using SCs as vehicles for the delivery of OVs in intracranial gliomas. Several strategies have been employed to allow specific replication of Conditionally Replicating Adenoviruses (CRAds) in glioma cells: deleting a viral region responsible for inactivating p53, as in the ONYX-015 virus 124 , or deleting the E1A viral region that binds to Retinoblastoma (Rb), allowing for replication in cells with defects in the Rb pathway, frequently altered in glioblastoma, as in the Delta24 virus 125 . Later it was established that the ONYX-015 was also active in cells with intact p53, and the selectivity for tumor cells was dependent on a tumor specific viral RNA export mechanism 126  allowed for great specificity of the oncolytic effect. CRAd-S-pk7 was engineered with survivin promoter dependent replication and a poly-lysine modification (pk7) that binds to heparin sulfate proteoglycans expressed on the surface of tumor cells, facilitating viral entry. This virus was able to infect both NSCs and human glioma cells with high efficiency, allow NSCs to migrate into orthotopic implanted tumors in mice and reduce the size of glioma flank tumors by about 50%, when compared to administration of loose viral particles 136 . The same CRAd-S-pk7 vector was used to optimize parameters of viral loading of NSCs to achieve relevant therapeutic efficacy that can be explored clinically 137  Previously, it was shown that HB1.F3.CD cells infected with CRAd-S-pk7 were effective in improving survival of mice with GBM43 xenografts when radiation was administered after intracranial delivery of NSCs, and less so when XRT was administered prior to NSCs 141 . These studies illustrate that the timing of NSC administration relative to radiation therapy and the route of administration are critical factors to consider when combining OV-NSCs with XRT.
To improve the delivery of SCs using the intranasal (IN) route, Spencer and collaborators tested two methods of reducing the clearance of NSC from the nasal cavity and improving transmigration of NSCs through the nasal epithelium into the brain: administration of a biodegradable fibrin and thrombin based glue into the nasal cavity immediately after the administration of NSCs loaded with OV, or administration of methimazole, a drug used to treat hyperthyroidism, that also causes a thinning of the nasal epithelium 142  In preparation for clinical use, the HB1.F3.CD NSC cell line, loaded with CRAd-S-pk7 (now named: NSC-CRAd-S-pK7) has been extensively tested in laboratory animals for tumor coverage and distribution 143,144 , safety, viral delivery, persistence in immunocompetent semi-permissive hosts 145 and efficacy when combined with standard of care 141 . It was demonstrated that administration of NSC-CRAd-S-pK7 is safe, results in extensive coverage of the tumor area by migrating NSCs and is more effective when administered prior to radiation and temozolomide therapy. Evidence was presented in support of a mechanism by which pretreatment with CRAd-S-pk7 impaired the DNA damage response and increased radio-sensitivity of glioma cells 141 .
The human HB1.F3.CD loaded with the CRAd-Survivin-pK7 oncolytic virus has been approved by the FDA as a clinical grade agent. The first in human phase I clinical trial for primary grade III and IV gliomas using NSCs as delivery agents for oncolytic viruses (NCT03072134, 2017-2020) was recently completed 146 . The study enrolled 13 patients with resectable or unresectable tumors, with the primary goal of establishing safety and toxicity of the NSC-CRAd-S-pK7 therapy when combined with SOC, including a dose escalation regimen with three doses of NSCs: 50, 100 and 150 million cells, corresponding to 6.25x10^10, 1.25x10^11 and 1.875x10^11 viral particles respectively, which were injected at the site of tumor biopsy or resection, followed by standard radiotherapy and Temozolomide treatment. The study also aimed to follow tumor progression by MRI and measure changes in cytokines and T cell subgroups 146 . Results from this study, not yet revealed, are expected with heightened anticipation.

THERAPEUTIC CYTOKINES, ANTIBODIES AND TOXINS DELIVERED BY STEM CELLS
Developing effective immunotherapy strategies for glioblastoma has been the goal of numerous preclinical and clinical studies over many decades, with limited success so far. New enthusiasm followed successes in the treatment of other solid tumors with checkpoint inhibitors and sustained efforts to replicate such results in glioblastoma are ongoing 147 . The promise of long-term efficacy by eliciting durable antitumor immunological memory is a highly desired outcome, especially in a cancer that inevitably recurs.
Immunotherapy for gliomas needs to address the severe tumor-induced immune suppression, in addition to being mindful of the minimal tolerance for inflammation in the brain 148 . In an effort to boost the antitumor immune response and aid antiglioma vaccine therapies, cytokine therapy has been pursued in preclinical and clinical trials for glioblastoma. The most promising results were shown with delivery of IL-2, IL-4, IL-12, and GMCSF 149,150 . In addition to antigen presenting cells (APCs) that bring the added advantage of expressing costimulatory proteins able to directly activate immune cells, glioma tropic stem cells represent good vehicles for cytokine delivery.
IL-4 strongly activates the antitumor immune system in a variety of cancers 151,152 . Expression of the IL-4 receptor has been found extensively on glioblastoma cells, and not on normal neurons and astrocytes, representing a good target for specific therapeutic intervention 153 . The therapeutic potential of mouse and rat NPCs, genetically modified to produce IL-4, has been tested in immunocompetent preclinical models of glioma 70  IL-12, a cytokine with important roles in modulating crosstalk between the native and adaptive immune system, induces proliferation of T lymphocytes and NK cells and stimulates production of other cytokines, especially of IFN. These activities result in a strong antitumor response in numerous cancers 154,155 . It was shown that intracranial administration of mouse NSCs transduced to express IL-12 improved survival in immunocompetent C57BL/6 mice 156 . Importantly, animals that survived the first administration of tumors were able to reject tumors implanted into the contralateral hemisphere three months later, demonstrating development of antitumor immunological memory. Increased infiltration of CD4 and CD8 T cells into the tumor was also observed. 156 . Similarly, in a rat glioma model, intratumoral administration of human NSCs, that were isolated from the hippocampi from 3-5 months old human embryos and transduced to express IL-12, significantly improved median survival (MS) of tumor bearing rats from 17 to 87 when NPC were administered at the same time with tumor cells, or to 73 days when NPCs were administered 5 days later 157 . Increased infiltration with T cells was also observed. In this study, the large survival benefit was likely due to the high 22/35 number of administered hNSCs: 10 and 30-fold higher than the number of tumor cells injected 157 . MSCs derived from either mouse BM 158 or human umbilical cord 159 modified to express IL-12 were also found to provide therapeutic benefit in preclinical glioma models.
Evidence that cytokine delivery by SCs induces immunogenic cell death (ICD) of glioma cells in preclinical models is provided by an elegant study with mouse BM derived NSC-like cells, transduced to express IL-23 (BM-NSC-IL23), a cytokine that induces IL-12 and IFN production 160 . Sixty percent of mice treated with BM-NSC-IL23 showed long term survival (>60 days) in immune competent B6 mice, while the MS of control animals was about 30-35 days. Antibody depletion of either CD8+ cytotoxic cells (CTLs), CD4+ or NK cells reduced the survival benefit of tumor bearing mice, demonstrating that CTLs were the primary cells responsible for the observed effect with minor contributions from CD4+ and NK cells 160 . When surviving animals were re-challenged with glioma cell implantations, they rejected the tumors, becoming tumor free, indicating establishment of long term antitumor immunity; this process was associated with increased levels of IFN. Studies using MSCs that delivered IL-18, IFN or both cytokines also showed a therapeutic effect and establishment of long-term anti-tumor immunity in glioma bearing rats, an effect associated with increased intratumoral production of IL-2 and IFN 161,162 . Combined delivery of IL-7 by MSCs with peripheral immunization, with IFN transduced irradiated tumor cells also resulted in decreased tumor size in a rat model of glioma 163 . MSCs transduced to express IFN had a direct cytotoxic effect on mouse glioma cells and human primary GBM cells and induced survival benefit in tumor bearing mice, an effect that was more pronounced when MSC-IFN was placed in the post-resection cavity and encapsulated in a synthetic ECM 164 . The melanoma differentiation associated gene-7 (mda-7/IL-24) cytokine, expressed in terminally differentiated cells and less so in malignant cells, also induces a direct, specific cytotoxic effect in a variety of tumor cells and increases sensitivity to radiotherapy 165 . It has been modified for increased stability and expression of luciferase, for in vivo luminescence tracking, generating the multifunctional protein SML7 166 . Primary mouse NSCs, transduced to deliver SML7 induced a reduction in intracranial glioma growth; co-administration of sTRAIL by NSCs, further improved this beneficial effect. 166 .
BMP4, a critical developmental growth factor in embryogenesis, was reported to inhibit the tumorigenic potential of brain tumor initiating cells (BTIC) and increase survival of glioma bearing animals when administered into the tumor as BMP4 coated acrylic beads 167 . Treatment with human adipose derived MSCs (hAMSCs), transduced with a retroviral vector to express BMP4 and administered systemically into the left cardiac ventricle of mice with intracranial gliomas (U87 cells), also elicited great survival benefit, with over 75% of tumor bearing animals becoming tumor free 168 . To increase safety of the therapeutic MSCs, avoiding possible insertional mutagenesis or other viral antigen induced inflammatory-responses, hAMSCs were transfected to express BMP4 using polymer nanoparticles (NP-BMP4.hAMSCs) 169 . Conditioned medium from these cells decreased the viability of patient derived human brain tumor initiating cells (BTIC) in vitro.
Intranasal administration of NP-BMP4.hAMCs in rats bearing intracranial tumors with BTIC significantly improved MS from about 14 days to about 17 days. It was also demonstrated using fluorescence imaging that 23/35 both intravenous and intranasal administration of these MSCs resulted in extensive distribution into the tumor 169 . The difference in the extent to which the two BMP4.hAMSCs preparations increased survival in the mouse vs. the rat model is large: 75% of mice were tumor free vs. a ~3-day increase in MS in rats. However, the studies cannot be properly compared due to other variations in the two experimental paradigms. The difference could be explained by the different distribution of MSCs following intracardiac vs. intranasal delivery, but is most likely due to the different tumor cells used, highlighting the fact that impressive results when using the U87 glioblastoma cell line may not reflect efficacy on primary patient tumor BTICs.
Another strategy for cancer immunotherapy, kindled by advances in toxinology and immunology, emerged in the late 1970s, when hybrid molecules were constructed from potent toxins conjugated to antitumor antibodies to specifically target tumor cells, leaving normal cells unharmed 170 . Debinski and collaborators demonstrated that the receptor for IL-13 is amply expressed on numerous glioblastoma cell lines, and that a chimeric protein made of the human IL-13 and a modified form of Pseudomonas exotoxin (PE): PE38QQR had a powerful cytotoxic effect on glioblastoma cells that was neutralized by IL-13 171 . It was later discovered that IL-13 binds with high affinity to IL13R2, which is present almost exclusively on tumor cells, acts as a decoy receptor and has a higher affinity for IL-13 than IL13R1, a receptor found on most normal cells 172 . IL13-PE38QQR was embraced with great enthusiasm and developed as an investigational drug. A phase III multicenter randomized controlled study in first recurrence glioblastoma patients comparing BCNU wafers with IL13-PE38QQR administered by convection enhanced delivery through a catheter placed into the resection cavity, found no significant difference in OS between the two groups 173  editing following treatment with CDX110, with 82% of treated patients losing EGFRvIII expression at recurrence 177 . The large randomized multicenter ACT IV study with Rindopepimut for the treatment of patients with newly diagnosed EGFRvIII positive glioblastoma was discontinued as it did not meet significant OS improvements 178 . Failure of this trial illustrates the difficulty in designing specific therapies for glioblastoma, even when unique tumor associated antigens are identified and carefully targeted; disease progression leads to the development of a continuously moving target, increasingly difficult to manage.

STEM CELL TRAFFICKING OF THERAPEUTIC NANOPARTICLES
Chemotherapeutic drugs are highly toxic to normal tissues. Intratumoral administration of therapeutic agents has significant advantages over systemic administration, allowing for increased therapeutic index and concentration of the active drug in the tumor tissue, while reducing off-target effects. However, effective distribution and intratumoral retention of the drug following injection, or local placement of hydrogels or other drug-laden scaffolds is hindered by limited diffusion capacity and adverse pressure gradients that rapidly clear the drug from the tumor. Combining NSC delivery of chemotherapeutics with nanotechnology allows for more efficient distribution of drugs throughout the tumor, including to invading margins and tumor satellites, as well as decreased clearance and release of targeted agents where they are most effective.
Nanoparticles (NPs) are synthetically generated vesicles with sizes as small as 10 nm and as large as 1000 nm. They are prepared from organic (lipids, polymers, dendrimers) or inorganic (metallic, ceramic) materials, can be solid or colloidal and have applications in numerous biomedical fields including theranostic applications of drug delivery, gene therapy, immunomodulation, imaging of treatment and disease progression, targeted radio-and photodynamic therapy, modulating release of bioactive molecules, minimizing toxicity and enhanced biocompatibility. Many advances have been made over the last 15 years in the field of nanotechnology and increased translation into cancer treatment is justifiably anticipated 179 . An excellent review of the use of NPs in brain cancer has recently been published 180 . In this section we will highlight preclinical studies that have employed NPs conjugated to stem cells to improve treatment of glioblastoma.
Systemic administration of NPs for cancer therapies still faces many challenges related to efficient delivery, distribution and retention in brain tumors, especially due to NP sequestration by the reticulo-edothelial system, renal clearance, transport through the BBB, overcoming the elevated intratumoral interstitial fluid pressure as well as phagocytosis by resident microglia and tumor associated macrophages. SC mediated delivery of therapeutic NPs is able to overcome these roadblocks. A proof of concept study by Roger and collaborators 181 demonstrated that poly-lactic acid NPs (PLA-NPs) and lipid nanocapsules (LNCs) can be uploaded onto human BM derived MSCs and intracranially delivered into mice with established U87 orthotopic brain tumors.
NPs conjugated to NSCs have been used in an intracranial mouse model of glioblastoma, demonstrating that distribution and retention of NPs within the tumor was enhanced when NPs were surfacebound to NSCs and administered either adjacent to the tumor, in the contralateral hemisphere or 25/35 intravenously 182 . These NPs were large (~800nm, with high drug loading capacity), manufactured from polystyrene and conjugated to biotinylated NSCs (HB1.F3 cells) via streptavidin, resulting in substantial NP cargo onto the NSCs, approximately 175 NPs/NSC 182 . In a preclinical model of ovarian carcinoma, using the same human NSC line (HB1.F3 ), treatment of tumor-bearing animals with NSCs loaded with Silica Platinum Nanoparticles SiNP[Pt] was able to induce a higher level of [Pt] in tumors compared to treatment with NPs or drug alone 183 . Silica NPs loaded with Doxorubicin (DOX), named a silica nanorattle (SN), was loaded onto human BM derived MSCs using antibody (CD73, CD90) mediated conjugation, as a delayed release "time bomb" for antiglioma therapy, and tested for use in a mouse flank tumor model with U251 glioma cells 184 . MSC administration of DOX showed high level, broad distribution and prolonged retention compared to injection of free DOX or NP-DOX and induced increased apoptosis in U251 glioma cells.
Biotin/avidin conjugation of NSCs (HB1.F3) to Docetaxel (DTX) loaded NPs was employed in a preclinical mouse model of triple negative breast cancer 185 . In this study the NPs, manufactured using a pH responsive biotinylated polymer (PEG-PDPAEMA) were conjugated to biotinylated NSCs via an avidin linker.
Acidic pH (6.3-6.9) is found in the extracellular space of many solid tumors, including glioblastoma 186-189 , making it an ideal signal for the unloading of NPs cargo. This study demonstrated that DTX loaded NPs conjugated to NSCs induced a significant decrease in tumor proliferation and vascularization after 7 days, when compared to free NP-DTX, which were rapidly cleared from the tumor 185 .
Doxorubicin loaded pH-sensitive mesoporous silica nanoparticles (DOX-NPs) have been tested in a preclinical intracranial glioma model on their own or when loaded onto human NSCs (HB1.F3.CD) and injected into the tumor or in the contralateral hemisphere (5 mm away) 190 . It was demonstrated that these NPs, when loaded onto NSCs, show a delayed release of the drug by about 20 hrs., presumably until the DOX-NPs are exposed to the acidic lysosomal compartment inside NSCs. During this time, the tumor-homing capacity of HB1.F3.CD was not compromised and they were able to reach the tumor to widely distribute and release DOX that induces the death of both tumor cells and NSCs. When compared to the intratumoral administration of loose DOX-NPs, NSCs loaded with DOX-NPs were able to significantly increase the MS of tumor-bearing mice 190 .
In vitro experiments have also demonstrated that NSCs can be loaded with magnetic spinning disks (SD), 2 microns in diameter, made of several layers of cobalt, iron and boron separated by platinum and covered in gold 191 . In the absence of a magnetic field (MF), the SD are innocuous to the cells. Application of a rotating MF causes the disks to spin and inflict mechanical damage inside the cells. In co-culture experiments of NSCs loaded with SD and glioma cells (U87), sequential application of the magnetic field allowed for release of SDs by NSCs, which were taken up by glioma cells that were killed when the rotating MF was applied again 191 .
An innovative use of NSCs as vehicles for glioma therapeutic agents was illustrated by Rachel Mooney and collaborators with the use of gold nanorods (AuNR) for plasmonic photo-thermal ablation of flank tumors in mice 192 . It was demonstrated that human NSCs (HB1.F3) induced a wider, more homogeneous distribution of 26/35 AuNR in flank tumors compared to administration of free AuNR and reduced tumor recurrence post photoablative treatment. We anticipate seeing this strategy tested in intracranial glioblastoma preclinical models. Inspired by successes in the treatment of other solid tumors with immune checkpoint inhibitors, and the hope for long-term efficacy by inducing antitumor immunological memory, extensive and sustained efforts to develop effective immunotherapy strategies for glioblastoma are much more challenging than for other cancers 147 . Oncolytic viruses have the proven capacity to induce immunogenic cell death (ICD) 113 and can also be regarded as a form of immunotherapy. Following successful early clinical trials, many oncolytic viruses received an expedited drug review process by the FDA to be advanced to large scale randomized trials 115 .
Study after study demonstrated that delivery of oncolytic viruses by NSCs results in enhanced tumor coverage, distribution and efficacy when compared to delivery of free virus 131,132,137 , advocating for efforts to fast track many more trials using this strategy into the clinic. The first-in-human clinical trial using NSCs as delivery agents for oncolytic viruses has been recently completed and results are eagerly awaited. It is anticipated that more trials with combination therapies using SC delivery of oncolytic viruses and/or immunostimulatory agents will soon be conducted. Combinatorial strategies using multimodal therapeutic SCs are likely to be more successful than current therapies. Successful targeting of glioblastoma will require combination therapy with multiple agents amenable to adaptation as the tumor progresses. Developing an expanded depository of clinically approved SCs, delivering a variety of antiglioma agents with different mechanisms of action that would be available for quick use, should bring significant improvements to currently available antiglioma therapies.
Exciting progress comes from the field of nanotechnology, with enhanced biomaterials being developed for drug and nucleic acid delivery. Conjugation of NSCs to nanoparticles (NPs) has proven to be feasible and more effective in delivering chemotherapeutics in preclinical models of glioma than direct administration of NPs. The acidic intratumoral environment brings an added bonus for pH responsive NPs, facilitating local 27/35 unloading of their cargo. Great hope is provided by the ability of NPs to deliver nucleic acids into gliomas. This strategy, if optimized for use with SC mediated delivery, would overcome many current therapeutic limitations in gliomas. Targeting glioma specific genes with siRNAs or with CRISPR/Cas9 vectors represents a powerful therapeutic approach, as many of the drivers of malignant glioma and of the glioma stem cell phenotype are developmental transcription factors 193 , generally considered "undruggable". SCs conjugated via a pH-sensitive bond to NPs loaded with nucleic acids may provide a viable avenue to overcome current limitations. An emerging field with exciting potential therapeutic applications in glioblastoma is the ability of NSCs and MSCs to deliver proteins and nucleic acids via exosomes 194 .
In this review we aimed to increase awareness of the great potential and advantages brought by the use of Stem Cells as vehicles for efficient and effective delivery of targeted anti-glioma therapies. At the same time, we emphasized the importance of critical experimental parameters, especially regarding the number of effectively distributed intratumoral SCs needed to ensure therapeutic efficacy. There is a great need to develop more clinically approved SCs for glioblastoma therapy and good practice manufacturing protocols that would make sufficient cells available for therapeutic use. This entails dedicated and diligent work to demonstrate and document the safety of these cellular therapeutic agents and a push for regulatory approval. Treatment failure in glioblastoma is in large part due to limitations in the delivery of therapeutic agents into invasive intracranial tumors. Stem Cells represent a viable solution for this problem. Why wait?