Personalized‐induced neural stem cell therapy: Generation, transplant, and safety in a large animal model

Abstract In this study, we take an important step toward clinical translation by generating the first canine‐induced neural stem cells (iNSCs). We explore key aspects of scale‐up, persistence, and safety of personalized iNSC therapy in autologous canine surgery models. iNSCs are a promising new approach to treat aggressive cancers of the brain, including the deadly glioblastoma. Created by direct transdifferentiation of fibroblasts, iNSCs are known to migrate through the brain, track down invasive cancer foci, and deliver anticancer payloads that significantly reduce tumor burden and extend survival of tumor‐bearing mice. Here, skin biopsies were collected from canines and converted into the first personalized canine iNSCs engineered to carry TNFα‐related apoptosis‐inducing ligand (TRAIL) and thymidine kinase (TK), as well as magnetic resonance imaging (MRI) contrast agents for in vivo tracking. Time‐lapse analysis showed canine iNSCs efficiently migrate to human tumor cells, and cell viability assays showed both TRAIL and TK monotherapy markedly reduced tumor growth. Using intraoperative navigation and two delivery methods to closely mimic human therapy, canines received autologous iNSCs either within postsurgical cavities in a biocompatible matrix or via a catheter placed in the lateral ventricle. Both strategies were well tolerated, and serial MRI showed hypointense regions at the implant sites that remained stable through 86 days postimplant. Serial fluid sample testing following iNSC delivery showed the bimodal personalized therapy was well tolerated, with no iNSC‐induced abnormal tissue pathology. Overall, this study lays an important foundation as this promising personalized cell therapy advances toward human patient testing.


| INTRODUCTION
The notoriously hard-to-treat brain tumor glioblastoma (GBM) consists of highly infiltrative cells. Standard treatment for patients with GBM consists of maximal tumor surgical resection followed by chemoradiotherapy. However, obtaining clean tumor margins is difficult, and residual tumor cells invade normal brain tissue, creating distant tumor foci. 1 Curtailing tumor recurrence with systemic therapies has seen limited success as the blood-brain barrier (BBB) tightly regulates the passage of molecules from systemic circulation to the brain parenchyma, which prevents many chemotherapies from reaching infiltrative GBM cells. 2 Treatment advances for GBM must combat tumor recurrence by targeting the infiltrative cells that remain after standard therapy.
Induced neural stem cells (iNSCs) generated by the direct transdifferentiation of fibroblasts from a patient's own skin have recently shown promise as drug carriers for GBM due to their innate tumor tropism and low immunogenicity. 3 These tumor-seeking cells are genetically engineered to express cytotoxic proteins as they migrate toward invasive cancer cells. Early testing of iNSC therapy in preclinical models has demonstrated the effectiveness of iNSCs generated from human donors against human xenografts of GBM in athymic nude mice. [3][4][5] Although the athymic mouse model recapitulates many aspects of human tumor resection surgery, it is understandably limited in its ability to test iNSC doses on par with what would be administered to a human patient as well as its ability to elucidate any potential adaptive immune reactions. Additionally, a small animal model cannot accurately mimic the size of the surgical cavity or fluid volumes that iNSCs may be delivered into during initial human patient testing. A large animal model is needed to bridge the gap between mouse models and a first-in-human Phase I clinical trial.
Canine models can be used to bridge this translational gap, as gliomas occur spontaneously in canines with phenotypes and genetic mutations similar to humans. 6,7 While this study focused on testing the safety and toxicity of iNSCs in healthy, nontumor-bearing canines, this was an important consideration when selecting an animal model that would be compatible with future efficacy studies. Despite the potential of this large-scale model, the generation of iNSCs from canine skin has not yet been reported. Additionally, the exploration of the delivery, dosing, and safety of autologous or allogeneic tumorhoming cell therapies has primarily focused on the murine model and is extremely limited in large animal models. [8][9][10] As such, we sought to demonstrate the feasibility of manufacturing, as well as the safety and toxicity of autologous therapeutic iNSCs using healthy canine patients. The iNSCs used in this study carry two therapeutic agents.
The first therapeutic agent is TNFα-related apoptosis-inducing ligand (TRAIL), a constitutively expressed protein that is continuously secreted into the extracellular space. TRAIL diffuses to nearby cells but initiates caspase-mediated apoptosis by engaging death receptors upregulated on cancer cells. 11 Importantly, TRAIL has been well tolerated and has shown negligible off-target toxicities in normal cells in both preclinical and clinical studies. [12][13][14][15] The second constitutively expressed protein is the enzyme thymidine kinase (TK). TK remains inactive until administration of its nontherapeutic prodrug substrate valganciclovir (VGCV). VGCV is first hydrolyzed into ganciclovir (GCV) in the liver and intestine, and subsequently, TK expressed by iNSCs phosphorylates circulating GCV into cytotoxic ganciclovir triphosphate (GCV-TP). Finally, GCV-TP inhibits DNA polymerase, consequently killing both the iNSC and nearby tumor cells via the bystander effect. 16,17 TK-VGCV therapy exhibits limited toxicity on normal brain cells due to their quiescent state. [18][19][20] Using our early-stage studies in murine models as a guide, we demonstrate the production, safety, and toxicity of autologous therapeutic iNSCs transdifferentiated from canine skin for the first time.
Healthy, purpose-bred canines were successfully administered two dose levels of autologous iNSCs using two clinically relevant delivery methods: intracerebroventricular (ICV) infusion and scaffold encapsulation. 21,22 We further demonstrate the safety and minimal toxicity signals in the canine model using magnetic resonance imaging (MRI), blood, urine, cerebrospinal fluid (CSF), neurological assessment, and histopathology. These promising results pave the way for future efficacy studies in a spontaneous canine glioma model as well as human clinical trials.

| Generation of canine iNSCs: Isolation, expansion, and conversion
Previously, we have shown the ability to convert mouse and human fibroblasts into tumor-homing iNSCs. 3 Using these studies as a guide, we first explored the feasibility of generating personalized canine iNSCs ( Figure 1a). First, full thickness (epidermis and dermis) skin biopsies were collected from donor canines. Once the animals were anesthetized, a skin biopsy was isolated from the base of the neck and placed into a tube containing collection media (Figure 1b,c). To isolate the fibroblasts, skin biopsies were manually diced into approximately 6 mm pieces ( Figure 1d) and placed in digestion media containing a low concentration of collagenase, thus enabling long dissociation conditions without negative impacts on cell viability. 23,24 Following dissociation, the tissue pieces were seeded in culture plates containing growth media, and fibroblasts were allowed to grow for 72 hr. Once fibroblasts had expanded, the cells were trypsinized, filtered to removed unwanted solid tissue, and the remaining fibroblasts were allowed to proliferate rapidly. As shown in Figure 1e, we observed robust outgrowth of fibroblasts from all biopsies. While fibroblasts initially proliferated reproducibly after each passage, cell senescence was observed as early as passage 12 for some fibroblast lines (data not shown). This timeline dictated the fibroblast transduction and transdifferentiation processes. As we have shown previously, we used the SOX2 transcription factor to and transdifferentiation media to convert fibroblasts into tumor-homing iNSCs. 3 To explore the potential of converting canine fibroblasts into canine iNSCs, we transduced the canine cells with lentiviral vectors encoding reverse tetracycline-controlled transactivator (rtTA) and SOX2, as well as the therapeutic and optical reporters mCherry-TK, and enhanced green fluorescent protein (eGFP) fused to TRAIL. In total, three fibroblast cell lines (biological replicates) were established and labeled iNSC-1, -2, and -3, respectively (Table 1).

| Investigating tumoritropic migration and tumor killing of canine iNSCs
After confirming transduction and transdifferentiation of canine fibroblasts to iNSCs, in vitro functional assays were conducted to test the migratory and tumor-killing capability of the new therapeutic cells.
The metastatic breast cancer cell line, MDA-MB-231-Br (231-Br), which was isolated from brain metastases, was selected as the model tumor for in vitro assays because of its ease of growth and sensitivity to TRAIL. Tumor-homing migration is one of the most critical aspects  In the absence of TRAIL and VGCV, we showed the iNSCs did not induce apoptosis and allowed the 231-Br cells to expand (Figure 1j).
Overall, these data suggest that canine iNSCs have the ability to seek out tumor cells and induce tumor apoptosis. Importantly, these results also showed similar migratory and killing potencies in all three canine iNSC lines, thus demonstrating the reproducibility of this technology.

| Developing methods for MRI tracking of cells
Tracking iNSCs in patients is highly beneficial, allowing insights into persistence and distribution. Although fluorescent-and bioluminescent-tracking is valuable in preclinical models, the use of these modalities in human patients is limited by light penetration and other challenges. To generate a method for tracking iNSCs that is compatible with canines and eventual human patients, we opted to label fibroblasts with superparamagnetic iron oxide nanoparticles (ferumoxytol) for MRI. Fibroblasts were used in lieu of iNSCs to expedite this experiment. Ferumoxytol is an FDA-approved iron replacement therapy and has recently been explored for its use as a cell-labeling agent. [25][26][27] Importantly, ferumoxytol has been shown to have no impact on cell viability or stemness, and the iron particles are not readily exocytosed, making it an ideal labeling agent for long-term tracking of iNSCs. 25,28 To investigate the potential of labeling fibroblasts for MRI tracking with this agent, the canine cells were transfected with ferumoxytol and assessed for particle uptake in vitro. As shown in Figure 2a, Prussian blue staining of cultured fibroblasts showed homogeneous uptake of iron oxide particles in the fibroblasts.
After confirming labeling in vitro, we next assessed imaging and tracking of the labeled cells in vivo. Unlabeled fibroblasts, ferumoxytollabeled fibroblasts, and free ferumoxytol were injected into the brain parenchyma of mice. Changes in volumes were then tracked using MRI over 3 days (Figure 2b). Analysis of images showed the cells and free ferumoxytol injection sites are clearly visible in all samples on the day of implant. The unlabeled and iron-labeled fibroblasts appear confined to a single location in the brain, whereas free ferumoxytol spreads from the injection site. By the third day postinfusion, unlabeled and iron-labeled fibroblasts remain constrained at their initial injection site, but the free ferumoxytol particles continue to spread medially, caudally, and invade the contralateral hemisphere. To confirm imaging results, mice were sacrificed following MR imaging on day three, brains were harvested, and tissue sections were probed for the presence of iNSCs ( Figure 2c). Strong eGFP fluorescence was observed at the site of injection, and Prussian blue staining confirmed the colocalization of iNSCs with iron oxide particles. We believe these studies suggest that we have the ability to isolate and transduce primary canine fibroblasts, transdifferentiate fibroblasts to iNSCs, kill tumor cells using both TRAIL and TK/VGCV, and track iNSCs in mice using MRI. Taken together, these in vitro and murine in vivo studies laid the foundation for the canine scale-up model. Building on this knowledge, we designed a canine study to evaluate the safety and toxicity of autologous canine iNSCs at low and high doses using two delivery mechanisms anticipated to be used clinically.

| Study design to test delivery and dosing
In the clinical setting, patient skin biopsies will be collected and converted into iNSCs, which will then be delivered back into patients following surgical resection as part of clinical standard of care for GBM patients. Using our expertise in canine neurosurgery, we created a surgical resection cavity and explored the delivery of iNSCs into the brain using two methods: (a) intracavity seeding within a biocompatible matrix, and (b) infusion into the CSF through a reservoir implanted in the lateral ventricle. Four healthy, male canines were randomized into one of two study arms: ICV injection or scaffold implantation.
One canine in each cohort was to receive 1 × 10 6 autologous iNSCs/ kg per dose, and the other canine was to receive 3 × 10 6 autologous iNSCs/kg per dose. Canines in the ICV cohort, referred to from here T A B L E 1 Abbreviations of cell types used

Cell type Abbreviation
Neural stem cells NSC Induced NSCs generated from skin via direct transdifferentiation iNSC iNSCs expressing both TRAIL and TK but not exposed to prodrug ganciclovir iNSC TRAIL+/TK− iNSCs expressing both TRAIL and exposed to prodrug ganciclovir At specified time points, canines were assessed for toxicity. If no toxicity or adverse safety signals were observed, the animal was to continue with the study; however, if a toxicity or safety signal was observed, the second canine in the cohort was redirected to the other study arm (Figure 3b,c).

| Manufacturing and scale-up of canine iNSCs
We next worked on expanding the canine iNSCs manufacturing process. Upon delivery, canines underwent a physical examination to ensure suitability for the study. CP01, CP02, and CP04 weighed 15-20 kg at the time of biopsy. CP03 weighed slightly less at 12 kg. After confirming animal health, full-thickness skin biopsies were surgically excised and autologous iNSCs were manufactured for each patient.
One of the biggest challenges in this study was scaling-up manufacturing procedures for a canine from a mouse model. As expected, interpatient variability was seen in fibroblast proliferation and transduction efficiency. This made it challenging to predict the iNSC yield after transduction and transdifferentiation. For CP01, we successfully generated the full 1 × 10 6 iNSCs/kg dose. CP01 received 15 × 10 6 , 14 × 10 6 , and 15 × 10 6 iNSCs at each dose, respectively.
CP02 did not receive the target dose of 3 × 10 6 iNSCs/kg any of its three doses; CP02 was administered 20 × 10 6 , 17 × 10 6 , and 12 × 10 6 cells for the first, second, and third doses, respectively. The goal of 1 × 10 6 iNSC/kg was reached for CP03 and the canine was implanted with a scaffold containing 12 × 10 6 iNSCs total. CP04 did not reach the dose goal of 3 × 10 6 iNSCs/kg but was implanted with the presence of iron in hemoglobin. However, it is expected that blood and its degradation products would resorb over time and a reduction in the hypointense signal would be observed. 29 As the hypointense region persisted nearly 60 days posttransplant, and without dramatic changes in volume, we believe this suggesting that the hypointense signal is indicative of iNSCs rather than hemoglobin.

| Assessing the safety of personalized iNSC therapy
Demonstrating iNSCs are safe is one of the most important parameters to test as iNSCs move toward human patients. Although human iNSCs have previously been shown to be safe, these studies were limited as they were performed in immune-depleted mice. The autologous transplant into fully immune-competent canines offers the potential to more accurately assess iNSC safety in a model mirroring personalized human iNSC therapy. To investigate the safety of iNSCs, canine health was assessed using complete blood count, blood chemistry, urine, and CSF ( Figure 6). As anticipated, elevations in white blood cells, segmented neutrophils, aspartate aminotransferase, and alanine aminotransferase were noted immediately following surgery. For CP02 and CP04, blood was drawn after administration of hydromorphone and midazolam but prior to propofol, isoflurane, and MRI.
There was some evidence of dilutional effects of the fluid therapy on blood cell counts. However, surprisingly neutropenia was observed on both MRI and non-MRI days, unlike in CP01 and CP03. There was also some evidence to suggest that VGCV can cause neutropenia.
However, we cannot rule-out iNSC-induced neutropenia. 30 Importantly, urine and CSF findings were unremarkable (data not shown).
We also assessed all major organs for any signs of abnormal pathology following necropsy ( Figure 7). As anticipated, there were some notable findings near the ICV reservoir implantation site and the scaffold implantation site. These findings included mild to moderate chronic inflammation, mild fibrosis, gliosis, and axon degeneration. Fibrous thickening and chronic inflammation were observed in the meninges in the region of the resection cavities of CP01 and CP02. No abnormalities were identified in the contralateral brain in all canines. However, histological changes in the brain and meninges near the resection cavities in the current study were attributed to postsurgical wound healing. Additionally, histology showed acute spinal cord hemorrhage in all canines, which was attributed to perimortem CSF collec-  Moreover, we also explored two delivery mechanisms to implant iNSCs in the brains of canines. We first investigated intrathecal delivery using a Rickham reservoir. ICV reservoirs are used in patients with GBM to deliver chemotherapeutics into the lateral ventricle. Here, chemotherapy follows the flow of CSF to cover the brain and spinal cord and reaches tumor foci via diffusion. While this system is FDA approved and bypasses the BBB, passive diffusion will prevent the chemotherapy from reaching deep tumor foci. 38 With the canine model, we were also able to establish the safety of iNSCs by conducting neurological assessments, screening fluids, and analyzing histology. Postural, gait, and vision abnormalities observed in canines mimicked possible adverse reactions during human brain surgery. Additionally, iNSCs did not elicit any concerning safety signals, regardless of ICV reservoir or FLOSEAL scaffold delivery, as determined by blood, urine, or CSF. The transient neutropenia and reproductive toxicity findings were unanticipated, but these could due to administration of the prodrug VGCV rather than the iNSCs.

| CONCLUSION
In conclusion, this research demonstrates the feasibility of manufacturing autologous iNSCs as well as their limited toxicity profile. We further show two clinically translatable methods for delivery.
Together, these data support the need for additional large animal efficacy studies and paves the way for future human clinical trials.

| Tissue harvest and digestion
Skin samples were harvested from canines at North Carolina State University; animals were euthanized for reasons unrelated to this F I G U R E 7 Canine histology. Hematoxylin and eosin staining contralateral brain, resection cavity, testicular tissue at necropsy, and testicular tissue collected presurgery. Scale bar, 100 μm study. Immediately following euthanasia, hair was removed using an electric trimmer. The harvest site was sterilized with betadine followed by 70% ethanol, repeated three times. Full thickness skin samples were acquired using a 6 mm diameter biopsy punch. Biopsies were placed in a 15 ml conical tube containing 10 ml 1X PBS and 2X anti-

| Lentiviral transduction and transdifferentiation
Fibroblasts were cotransduced with lentiviral rtTA, SOX2, TK-mCh, NIM + doxy media was replaced every other day for a total of 5 days.

| Immunohistochemistry
To

| Iron oxide labeling
On the fifth day of transdifferentiation, iNSCs were labeled with fer-

| Murine in vivo surgical procedure
All murine procedures were approved by the Animal Care and Use

| Canine iNSC production
The skin of the dorsal cervical region was prepared as described above and 1 × 3 cm full thickness, rectangular skin biopsies were obtained and placed into transport media on ice. iNSCs were manufactured as described above. For postoperative analgesia, canines were given carprofen (4.4 mg/kg SQ followed by 4.4 mg/kg PO q 24 hr for 3 days).

| Canine craniectomy and iNSC implantation
In preparation for craniectomy, canines were anesthetized as described above except that a continuous infusion of propofol (200 μg/kg/min) was utilized to decrease isoflurane requirements.
Canines were mechanically ventilated; the head was shaved and surgically prepared and placed in a head frame for neuronavigation (Brain-Sight, Rogue Research). The bite plate made prior to the first MRI was replaced and used to register the canine for neuronavigation. The iNSCs were delivered in a FLOSEAL scaffold in two canines and through an intraventricular catheter in the other two canines. To simulate a resection cavity remaining after tumor removal, a craniectomy was created using a high-speed drill followed by durotomy. A 2 × 2 cm region of cerebral cortex and underlying white matter was removed, centered on the marginal gyrus near the junction of the parietal and occipital lobes. In CP03 and CP04, the iNSC-FLOSEAL scaffold was injected into the resection cavity; in the canines with intraventricular catheters, the resection cavity was left unfilled. The dura was then manually apposed and covered with an artificial dural product (DuraGen Plus, Integra LifeSciences, Plainsboro, NJ). Neither the dura nor the craniectomy defect were primarily closed. In CP01 and CP02, an intraventricular catheter was placed into the right lateral ventricle after creation of a 6-mm burr hole. The catheter was attached to a Rickham reservoir that was seated into the burr hole to create a ventriculostomy system (Codman Holter Rickham reservoir, Integra LifeSciences). The temporalis muscle, subcutaneous tissues, and skin were closed using suture and the skin was closed using staples. A fentanyl patch (50-75 μg/hr) was placed immediately following surgery, and hydromorphone (0.1 mg/kg IV q 6 hr) was administered until the patch took effect approximately 12 hr after placement. For CP01 and CP02, ultrasound navigation was used to inject the second and third iNSC doses into the Rickham reservoir.

| Canine castration
In order to further investigate potential reproductive toxicity, CP02 and CP04 underwent unilateral castration for histopathological analysis and sperm evaluation. These procedures were performed using standard techniques and the contralateral testis was similarly evaluated at the time of autopsy. which was applied immediately following during surgery, was removed after 5 days. VGCV was administered 2 weeks following scaffold implantation and 2 weeks after each ICV dose at 450 mg PO q 5 days.

| Canine euthanasia and necropsy
At the study endpoint, all canines were euthanized via an intentional sodium pentobarbital overdose at 85 mg/kg. Necropsy was performed immediately after euthanasia. Tissue samples were fixed in 10% formalin overnight in preparation for histological analysis.

| Canine histology
Hematoxylin and eosin staining was performed on 5 μm sections of formalin-fixed, paraffin-embedded tissue samples. Histopathological analysis was performed by board-certified veterinary pathologists (L. B. B., D. A. T.).

| Statistical analysis
Data are expressed as mean ± SD.