Lentiviral Vector Induced Modeling of High-Grade Spinal Cord Glioma in Minipigs

Background: Prior studies have applied driver mutations targeting the RTK/RAS/PI3K and p53 pathways to induce the formation of high-grade gliomas in rodent models. In the present study, we report the production of a high-grade spinal cord glioma model in pigs using lentiviral gene transfer. Methods: Six Gottingen Minipigs received thoracolumbar (T14-L1) lateral white matter injections of a combination of lentiviral vectors, expressing platelet-derived growth factor beta (PDGF-B), constitutive HRAS, and shRNA-p53 respectively. All animals received injection of control vectors into the contralateral cord. Animals underwent baseline and endpoint magnetic resonance imaging (MRI) and were evaluated daily for clinical deficits. Hematoxylin and eosin (H&E) and immunohistochemical analysis was conducted. Data are presented using descriptive statistics including relative frequencies, mean, standard deviation, and range. Results: 100% of animals (n = 6/6) developed clinical motor deficits ipsilateral to the oncogenic lentiviral injections by a three-week endpoint. MRI scans at endpoint demonstrated contrast enhancing mass lesions at the site of oncogenic lentiviral injection and not at the site of control injections. Immunohistochemistry demonstrated positive staining for GFAP, Olig2, and a high Ki-67 proliferative index. Histopathologic features demonstrate consistent and reproducible growth of a high-grade glioma in all animals. Conclusions: Lentiviral gene transfer represents a feasible pathway to glioma modeling in higher order species. The present model is the first lentiviral vector induced pig model of high-grade spinal cord glioma and may potentially be used in preclinical therapeutic development programs.

the most extensively used models for immunotherapeutic studies is the GL261 line. Originally developed in 1939 through administration of methylcholanthrene pellets into the murine brain, it has served as a critical resource given its diffuse infiltration, histopathologic characteristics, and capacity to be employed in wildtype C57BL/6 mice. Unfortunately, since this is a syngeneic mouse model, transplantation of these cannot be scaled into large animals for use a more surgically translatable anatomic space 13 .
Vector-driven glioma models represent an immunocompetent option that could feasibly be scaled up to a large animal model, well reviewed elsewhere 3 . In particular, platelet-derived growth factor Beta (PDGF-B) driven glioma models have been demonstrated in-silico, in-vitro, and in in-vivo -using rodent models with retroviral expression of PDGF-B, HRAS-G12V, shRNA-P53, and other transgenes implicated in the disease pathogenesis [14][15][16][17][18][19] . In 2011, Lei et al. first reported that PDGF-B-driven models recapitulate the proneural glioblastoma subtype 20 . This was further investigated by  with robust characterization of the master regulator transcriptional network that drives glial progenitor transformation 21 . As recently as 2018, transcriptomic characterization of mouse models using lentiviral PDGF-B and shRNA-CDK2NA have also been shown to produce pro-neural subtype glioblastoma 22 . In mouse models, it is reported that additional genetic lesions may be required to model high-grade gliomagenesis with either constitutive RAS, or more importantly, knockdown of CDK2NA or p53 22,23 .
In optimizing a model system for high-grade glioma, the use of a large animal system would provide improved utility for translation of neurosurgical strategies, device development, drug delivery, and radiologic study [24][25][26] . From an anatomic standpoint, the rodent brain is lissencephalic creating a minimization of drug leakage and improvement of local drug delivery 27,28 . Furthermore, the murine brain and spinal cord have significant developmental differences in comparison to large animal systems [27][28][29] . In addition to these anatomic considerations, the aforementioned concerns with immunocompromised models using xenograft cell lines have generated a translational space that may explain some of the limited success with preclinical strategies 4,25,30 . Moreover, the use of viral vector-driven approaches in, targeting distinct genetic lesions implicated in the human disease, have been widely reported as reliable, highly penetrant, and with striking histopathologic and molecular validity 14,16,17,22 .
The genetic profile, immune system, and the size and anatomy of the porcine brain and spinal cord is recognized to better model the human 25,29,31 . For these reasons, device implantation, stereotactic radiosurgery, and intrathecal and intraparenchymal drug delivery have all been performed in pigs 7,27,[32][33][34][35][36] . Furthermore, while the Food and Drug Administration (FDA) has recognized proof-of-principle data on therapeutic efficacy in highly characterized rodent models, the use of large animals is considered critical for clinical relevance. Recognizing these considerations, groups have utilized the U87 xenografts in pigs, but required Cyclosporine-based immunosuppression and continued to face the known limitations of the U87 cell line 27,37 . Overall, existing animal models have limited translational utility due to a confluence of factors including immunogenicity, immunocompromised systems, variable recapitulation of histopathologic features, and limited anatomic translatability, highlighting the significance of generation and validation of an immunocompetent large-animal model.
Thus, the goal of the present study is to scale previous strategies from murine models into a large animal model. Here we describe the use of lentiviral gene transfer to generate an immunocompetent large animal model of high-grade glioma in the porcine spinal cord with subsequent histopathologic, radiologic, and behavioral analysis.

Materials and Methods
Vector design. It is well known that HGGs involve a mixed mutational profile with marked heterogeneity between patients. However, the RTK/RAS/PI3K and p14(ARF)/CDKN2A and TP53 pathways are involved in up to 88% and 50% of high-grade gliomas respectively 17 . Therefore, we decided to target these pathways, as has been performed in rodent models, to establish the feasibility of scaling this strategy into a pig model. Three individual lentiviral vectors were designed to target the RTK/RAS/PI3K and TP53 pathways. These vectors were designed separately to avoid the risk of decreased transduction efficiency from larger insert sizes, biosafety requirements, and to provide flexibility to adjust included vectors for future study 38 . Raw sequences of transfer plasmids are presented in Supplemental File 1 and schematics of transgenes in Supplemental Fig. 1.
All lentiviral vectors were VSV-G pseudotyped, third-generation, replication deficient systems titered at >10 9 infectious units (IU)/ml. Vector 1: PDGF-B-IRES-eGFP and Vector 2: HRASG12V-IRES-mPlum were designed using a pCDH transfer plasmid backbone with an Ef1α promoter and reporters following an internal ribosomal entry site (IRES) sequence. Vector 3 was designed using a pLKO1 backbone and expressed two sequences of shRNA targeting porcine p53 including sequences 787 and 944 under the H1 and U6 promoters, respectively as reported by Merkl et al. 39 . These shRNA sequences have 100% homology to two regions of porcine TP53 mRNA and have been used for efficient knockdown in porcine cell culture models 39 . Controls for vectors 1-3 only included reporters and shRNA scramble, referred to as CTRL. The oncogenic cocktail will be referred to as ONC. All vectors (ONC or CTRL) were thawed and combined in equal parts immediately prior to inoculation. Surgical approach. The development of an intraparenchymal mass forming lesion, as with a glioma in the spinal cord, would progress in human patients to yield an anatomically localizable neurologic deficit. As such, we selected the thoracolumbar spinal cord of the Gӧttingen minipigs to inject the oncogenic lentiviral vectors. This location of injection provides a distinct advantage, where tumor progression will incur clinically appreciable and quantifiable motor deficits affecting the hind-limb ipsilateral to the site of injection. The surgical procedure was performed as in prior studies and as depicted in Fig. 1A,B 40 . Briefly, the animals underwent a two-level laminectomy and dural exposure. The spinal derrick stereotactic platform was mounted onto the dorsum of the animal for targeting the lateral white matter (2.5 mm lateral to midline, at rostral or caudal positions for each injection, spaced by 2 vertebral levels) 40 . This spacing is necessary to allow for exclusive analysis of the injection sites, space www.nature.com/scientificreports www.nature.com/scientificreports/ system, and the size and anatomy of the porcine spinal cord is recognized to better model the human 25 . Gӧttingen minipigs >15 kg (n = 6, 3 males and 3 females) were used in the present study. All animals underwent daily modified Tarlov-motor scoring to provide an ordinal scale of behavioral motor deficits for quantitative analysis (scored 0-9, Supplemental Table 1) 42 . All scoring was performed by investigators blinded to the side of injection and study design. Other clinical deficits may manifest due to a mass forming lesion in the thoracolumbar spinal cord. As such, physical examinations were performed by laboratory staff in consultation with veterinary staff daily, with both baseline and post-operative assessments until endpoint. All surgical procedures, animal care, endpoints, and the respective experiments on pigs were approved by the Emory University Institutional Animal Care and Use Committee (IACUC). All experiments were conducted with adherence to approved protocols set forth by the IACUC in coordination with the division of animal resources (DAR) and veterinary staff.
Baseline and endpoint magnetic resonance imaging. The gold-standard for non-invasive tracking of tumor progression and recurrence of glioma in human patients is magnetic resonance imaging (MRI) 43 . To evaluate tumor growth, all animals underwent baseline and endpoint MRI scans. All scans were performed on a clinical Siemens 3T Trio MRI scanner with an integrated spine coil (Siemens Medical Solutions, Malvern, PA). Pre-contrast sequences included sagittal T1-weighted, T2-weighted, T1-fluid attenuated inversion recover (FLAIR) and T2-FLAIR sequences. Following intravenous gadolinium administration (0.1 mmol/kg, MultiHance-gadobenate dimeglumine, Bracco Diagnostics Inc.), post-contrast sequences were performed (sagittal and axial T1-FLAIR). Spatial saturation pulses were used for elimination of respiratory artifacts, and appropriate adjustments of field of view, flip angle, echo time, TR, and TI-interval were made. Radiologic features of interest included iso/hyper/hypo intensity, necrosis, contrast enhancement, rostro-caudal expansion, transverse cord involvement, tumor volume, spinal cord expansion, edema, and CSF obstruction. All scans were processed and representative images were acquired using ONIS-Dicom Viewer (DigitalCore, Co.Ltd, Tokyo, Japan).
Tissue processing, H&E, and immunohistochemistry. week endpoint, all animals underwent euthanasia through intravascular administration of pentobarbital sodium and intracardiac perfusion with heparinized 0.9% saline solution. Spinal cords were harvested, post-fixed in 4% paraformaldehyde overnight, and cut into blocks at 5 mm intervals for paraffin embedding. Serial transverse spinal cord sections were cut on a microtome at 8um thickness for hematoxylin and eosin (H&E) and immunohistochemistry. H&E staining was performed with deparaffinization by serial xylene incubations and ethanol gradients in a standard fashion (Hematoxylin Gill No. 3, Sigma Aldrich, Cat: GHS332; Eosin Y, Sigma Aldrich, Cat: HT110132). Qualitatively, H&E sections were evaluated by a board-certified clinical neuropathologist (S.N. and P.C.) under wide-field microscopy blinded to the study design. Immunohistochemical stains were performed with primary antibodies for GFAP (Dako, Z0034), Olig2 (abcam, ab109186), Ki67 (abcam, ab15580) with appropriate secondaries for subsequent diaminobenzidine (DAB) and hematoxylin counterstaining (Thermo Scientific, Autostainer 480S). Immunohistochemical stains for glial markers (Olig2, GFAP) were qualitatively evaluated. Ki-67 index was quantified with manual counting of at least 1000 cells, scored as positive or negative. Representative images were acquired at 2x, 10x, 20x, and 40x magnifications (DS-Qi1 high sensitivity Cooled CCD camera, Nikon E400 microscope, NIS-Elements imaging software, Nikon Instruments, Inc) with calibrated scale bars prepared in ImageJ 44 . Whole slides were scanned in a raster pattern at 40x (Leica Aperio AT2 Slide Scanner).
Data analysis and statistics. Descriptive statistics were used for all categorical variables, summarized as absolute and relative frequencies. Continuous and ordinal variables were summarized as appropriate using mean, standard deviation (SD), median and range. All clinical and behavioral analyses were performed by investigators blinded to the side of ONC or CTRL administration.

Animals develop motor deficits ipsilateral to ONC injections by post-operative week 3. Post-
operative behavioral analysis and physical examination was performed daily by blinded investigators until a 3-week endpoint. All animals developed clinically appreciable progressive motor deficits ipsilateral to the ONC injection, with 100% of animals (n = 6/6) developing hindlimb motor deficits by the 13 th post-operative day (Fig. 1C). 33% of animals (n = 2/6) progressed to develop bilateral hindlimb paralysis by the 3-week endpoint. At endpoint, Tarlov scores among animals ranged from 0-8, including either complete paralysis or impaired ambulation with the ability to walk greater than 1 minute respectively. Representative motor deficits are presented in Supplemental videos 1 and 2 taken at post-operative day 12 and 17 respectively in the same animal. Following tissue harvesting, appreciable mass-forming lesions were observable through the dura mater on gross pathology (Fig. 1C). No animals exhibited clinical signs of cachexia and the mean difference between baseline and endpoint body weight was 0.61 kg (SD: ± 0.94, Range: −0.5 to 2.1, Median: 0.55). A summary of hindlimb motor deficits, laterality, and clinical symptoms are outlined in Table 1. On gross examination, (Fig. 1D), a mass forming lesion was appreciated in 6/6 animals with appreciable expansion of the spinal cord. No tumoral or peritumoral cysts were appreciated.
Endpoint MRI demonstrates mass forming lesions consistent with high grade glioma. All 6 animals received baseline and endpoint MRI scans. Baseline imaging demonstrated normal spinal cord anatomy with unobstructed cerebrospinal fluid (CSF) pathways in the sub-arachnoid space that were visible on sagittal views. Representative images from two animals presented in Fig. 2A, T2 Histopathologically confirmed high-grade glioma. On H&E staining at low magnification, the tumors diffusely infiltrate the spinal cord, spreading rostral and caudal with respect to the injection site, and invading both grey and white matter. There is noticeable mass effect on the central canal and contralateral structures (Fig. 3A) in all animals. In addition, there is marked invasion along the white matter tracks with cellular regions showing hyperchromasia visible in sections distal to the site of injection (Fig. 3A). Histopathologic features identified at high magnification include: high cellularity with epithelioid, and fibrillary astrocytic morphology, microvascular proliferation, necrosis, thrombosed blood vessels, and invasion along the tumor border into surrounding normal parenchyma (Fig. 3B). Subsequent evaluation by board-certified neuropathologists (S.N. and P.C.) resulted in confirmation of the mass forming lesions as high-grade gliomas with astrocytic morphology, with consistent features in all animals. Ki-67 staining was highly positive across animals, (Fig. 3C) with a mean proliferative index of 37.1% (SD: ± 14.2). In addition, immunohistochemical staining was highly positivity for glial markers GFAP and Olig2 (Fig. 3C).

Discussion
The data of the present study illustrates the first lentiviral vector induced large animal model of high-grade glioma with histopathologic, radiologic, and behavioral characterization. We found that high-grade gliomas formed in a highly penetrant fashion following the injection of ONC lentiviral vectors and exhibited histopathologically confirmed features of high-grade gliomas. Importantly, this has not been performed in large animal models prior to this study and we assert that this paradigm using intraparenchymal gene transfer may be adapted for use in spinal cord, brainstem, cerebellum, and brain modeling of glioma.
Current preclinical studies for therapeutic approaches including, but not limited to, oncolytic viral therapy, immunotherapy, convection-enhanced delivery, fluorescein dyes and laser interstitial thermal therapy in high-grade glioma rely on large animal models with no ability to evaluate therapeutic efficacy in disease models 7 . By using orthotopic intraparenchymal gene transfer in higher-order species, given the advantages of an immunocompetent porcine model, investigators can reasonably adjust transgenes and/or promoters to dissect the disease process and develop translationally relevant systems in a variety of glioma grades and subtypes.
Utilizing gene transfer in pigs as a system for modeling tumors in the CNS offers several advantages. For one, this approach has been previously employed in porcine CNS gene transfer for other neurologic disease modeling and can be widely scaled for implementation and adaptation in wildtype pigs for use in preclinical therapeutic development programs 45,46 . Importantly, this would circumvent the need for generation and maintenance of numerous transgenic breeds, providing a pathway for rapid induction in widely available pigs. In addition, when using lentiviral gene transfer -third or second generation replication deficient lentiviruses permit large insert size, minimal immunogenicity, and a broad tissue tropism with a VSV-G pseudotype 38 . Furthermore, lentiviral genomic integration into the transduced cell permits the continued expression in daughter cells of the proliferative environment of an induced glioma 16 . As such, examination of the most common patient-population and respective patterns of mutations may allow for targeted modeling with ease of inclusion or exclusion of specific genetic lesions (e.g. loss of ATRX, IDH1 mutations, EGFRvIII, PTEN, p53, H3K27M) with multiple combined www.nature.com/scientificreports www.nature.com/scientificreports/ constructs. Consequently, this paradigm could provide a foundation for utilizing specific transgenes, promoters, and titer for a host of potential glioma grades, subtypes, and other CNS tumors in porcine models. This strategy can be reasonably implemented using non-viral systems (sleeping beauty transposon system) or other vector driven gene editing approaches (CRISPR) to develop an armamentarium of immunocompetent translational animal model systems for neoplasia of the central nervous system. By drawing from clinical insights, applying foundational research, and validating response to the clinical standard of care, we assert that such a modeling approach would provide a highly impactful space for both bench science and translational therapeutic initiatives.
Limitations and future directions. There are several areas that warrant investigation in future studies.
One area of interest is whether these tumors progress as the result of clonal expansion of virally infected cells, or through the recruitment and transformation of resident glial progenitors 15,16,22 . In addition, it is necessary to evaluate tumor biopsies for accumulated mutations that occur in the environment of p53 knockdown, PDGF-B expression and HRAS-G12V expression, and an overall picture of proliferative stress 47 . As such, the importance of sequencing tumor biopsies for evaluation of copy-number variation, single-nucleotide polymorphisms and mutational insertions or deletions cannot be overstated. The rise of molecular subtyping have provided putative categories for glioblastoma and high-grade glioma 48,49 , which have marked inter-tumoral and intra-tumoral www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ heterogeneity in patients 50,51 . To validate such models with respect to these subtypes and evaluate the degree of inter-and intra-tumoral heterogeneity, the use of bulk and single-cell RNA sequencing for transcriptomic investigation is warranted. These data can be readily compared to our existing understanding of the disease and open-source patient data sets such as The Cancer Genome Atlas (TCGA) as has been conducted in prior studies. Furthermore, it is critical from a translational standpoint to study the use of surgical resection, associated strategies and challenges, and the characterization of tumor recurrence in the present model. Lastly, the characterization of host immunologic responses in the swine in response to these lentiviral injections will need to be understood for testing and understanding immunotherapeutic interventions. Limitations and future directions notwithstanding, the present study describes the first histologically, radiologically, and behaviorally characterized immunocompetent porcine spinal cord glioma model. conclusion Lentiviral gene transfer of driver mutations represents a feasible pathway to modeling glioma in higher order species. Here we report the first lentiviral induced large animal model of high-grade spinal cord glioma. This can be potentially used in a variety of preclinical therapeutic development programs.