The impact of initial tumor microenvironment on imaging phenotype.

Models of human cancer, to be useful, must replicate human disease with high fidelity. Our focus in this study is rat xenograft brain tumors as a model of human embedded cerebral tumors. A distinguishing signature of such tumors in humans, that of contrast-enhancement on imaging, is often not present when the human cells grow in rodents, despite the xenografts having nearly identical DNA signatures to the original tumor specimen. Although contrast enhancement was uniformly evident in all the human tumors from which the xenografts' cells were derived, we show that long-term contrast enhancement in the model tumors may be determined conditionally by the tumor microenvironment at the time of cell implantation. We demonstrate this phenomenon in one of two patient-derived orthotopic xenograft (PDOX) models using cancer stem-like cell (CSC)-enriched neurospheres from human tumor resection specimens, transplanted to groups of immune-compromised rats in the presence or absence of a collagen/fibrin scaffolding matrix, Matrigel. The rats were imaged by dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and their brains were examined by histopathology. Targeted proteomics of the PDOX tumor specimens grown from CSC implanted with and without Matrigel showed that while the levels of the majority of proteins and post-translational modifications were comparable between contrast-enhancing and non-enhancing tumors, phosphorylation of Fox038 showed a differential expression. The results suggest key proteins determine contrast enhancement and suggest a path toward the development of better animal models of human glioma. Future work is needed to elucidate fully the molecular determinants of contrast-enhancement.

Introduction

Glioblastoma (GBM), the most common of high grade gliomas (HGGs), presents a deadly puzzle for medicine. Glioblastoma is treated with maximal surgical removal, followed by radiation therapy (RT) and DNA-alkylating agent temozolomide (TMZ) [1]. Despite this aggressive multi-modality treatment, most patients show treatment refractory relapse within months, with a dismal 2 year survival rate of 15.2% [2]. Even with the best care, i.e. surgery, radiotherapy and temozolomide, GBM is incurable and quickly fatal, with a mean time from diagnosis to death of about 15 months [3, 4]. This dismal therapeutic landscape is made even darker by a dearth of truly representative animal models of GBM [5-7]. Huszthy et al. [6] noted, “… there are a vast number of reports showing the therapeutic efficacy of novel treatment modalities in various animal models, yet when these are further evaluated in the clinic, they fail in phase II/III clinical trials.” In the intervening years, little has changed: Aldape et al., in a 2019 review [5], described seven challenges for the scientific community, one of which is to develop preclinical models of human cancers. In studying this malignant disease, animal models are a necessary first step in testing putative therapies; in order to be useful, they should recapitulate the salient pathophysiology of the disease.

Neurosphere cultures derived from tumor biopsy and grown in serum-free media have been shown to preserve the genomic features of the original tumor, and form patient-derived orthotopic xenograft (PDOX) tumors in immunocompromised mice [6-15]. Because they so closely echo the genotypes of their parent tumors [8, 16, 17], PDOXs can be considered to be the most representative and important tool to study treatments intended to interfere with the complex genetic alterations involved in in vivo tumorigenicity and progression of GBM [14, 15, 18, 19]. However, these best current models of GBM have an unrecognized flaw. We [20] and others [6] have observed that PDOX’s mimicry of the human disease is impaired, in that instances of magnetic resonance imaging (MRI)-visible blood-brain barrier (BBB) breakdown are rare. In fact, PDOXs are typically infiltrative [15] and generally do not exhibit MRI contrast enhancement, despite the clinical MRI displaying florid central enhancement in the main tumor mass from which these models were derived.

Two mutually exclusive cellular behaviors have been observed in glioma tumors - slow moving proliferative cells (‘grow’), or mainly infiltrative migratory cells (‘go’) with lower proliferation rate [21]. The enhancing mass of the tumor (the ‘grow’ population) is characterized by an incompetent vasculature, central hypoxia, pseudopalisading necrosis, increased interstitial fluid pressure, and peri-tumoral flux [22, 23]. This physiology is substantially different from the physiology of the ‘go’ population, where proliferation is suppressed and resistance to apoptosis and drug therapy is increased [21]. Thus, PDOXs without BBB breakdown model the infiltrative cell population, but not the enhancing proliferative tumor mass, and therefore represent an incomplete model of the human pathology from which they were derived. This partial fidelity to the original pathology has potential to compromise the assessment of trial therapies, and of combination therapies, regardless of dependency on BBB breakdown for delivery. Interestingly, glioma stem cells can adopt invasive or proliferative phenotypes [24], accumulating in perivascular niches and hypoxic regions [25].

The lack of phenotypic recapitulation of human tumors is suggested to lie in the interaction between the implanted tumor cells and host stroma [26]. We hypothesized that manipulation of the tumor stromal composition by addition of Matrigel, a basement membrane preparation, would produce a consistent BBB breakdown in murine PDOX models of GBM. To test this hypothesis, reconstituted basement membrane derived from Engelbreth-Holm-Swarm (EHS) mouse sarcoma Matrigel [27, 28], with high and low growth factor content, was used to manipulate the brain microenvironment during patient-derived cancer stem-like cell (CSC) implantation and the radiological presentations of the resultant tumors were examined by DCE-MRI.

Materials and methods

Athymic, immune-compromised female RNU/RNU rats were acquired from Charles River, Frederick, MD. Matrigel Matrix-Growth factor reduced and Matrigel Basement Membrane Matrix-High concentration were purchased from Corning and BD Biosciences, respectively. All other reagents and labware used were of analytical grade.

Patient-derived neurosphere cell lines

Following an approved Institutional Review Board protocol, resected brain tumor specimens were collected and processed at Hermelin Brain Tumor Center, Henry Ford Hospital, with written consent from patients. Tumors were pathologically graded according to the WHO criteria. GBM specimens were dissociated and cultured in serum-free neurosphere medium to enrich CSC [9]. HF2303 and HF3016 CSC lines were derived from newly diagnosed tumors from two glioblastoma patients in which MRI studies showed typical ring-enhancing lesions. These neurosphere lines were tumorigenic, and molecular profiling show that genomic abnormalities present in the parental tumors were preserved in these models [17, 29, 30].

Orthotopic rat xenografts

For intracranial implantation, neurospheres were dissociated to single cells in Dulbecco’s phosphate buffered saline (PBS) following published methods [9, 11]. A 10 μl volume containing 3 × 105 cells was used for implantation in each rat. The cells were suspended in 10 μl sterile PBS (M0), 5 μl PBS+5 μl Matrigel with reduced growth factors (M−) or 5 μl PBS+5 μl Matrigel with high growth factors (M+). Thirty-two female, athymic, immune-compromised rats were randomly distributed between the 3 subgroups, M0, M− and M+. Following published methods [31, 32], one of the two neurosphere cell lines was implanted into the right hemisphere in these rats under a protocol approved by the Institutional Animal Care and Use Committee. Briefly, animals were anesthetized with isoflurane (4% for induction, 1.5% for maintenance, balance N2O:O2 = 2:1). Under aseptic conditions, the scalp was swabbed with Betadine and alcohol, the eyes coated with Lacri-lube (Allergan Inc., Madison, New Jersey) and the head immobilized in a rat stereotactic device (Kopf, Cayunga, California). A 1 cm incision was made 2 mm right of the midline and the skull exposed and a burr hole drilled 3.5 mm to the right of bregma without penetrating the dura mater. A 10 μl Hamilton syringe with a 26 gauge needle containing the tumor neurosphere cells was lowered to a depth of 3.0 mm and then raised back to a depth of 2.5 mm. Cells were then injected at a rate of 0.5 μl/10 s until the entire volume was injected. The needle was gently withdrawn, the burr hole sealed with sterile bone wax and the scalp sutured with sterile silk sutures. The rats were observed until complete recovery and then placed back in their home cages.

Imaging

All studies used a Varian/Magnex (Santa Clara, California), 7 Tesla, 20 cm bore magnet with a Bruker console running Paravision 6.0 software. Gradient maximum strengths and rise times were 250 mT/m and 120 μs. All MRI image sets were acquired with a 32 × 32 mm2 FOV. Transmit and receive coils included a Bruker Quadrature Birdcage (transmit) and 4-channel phased-array surface coil receiver (Rapid MR International, Columbus OH). T1-weighted images (T1WIs) were acquired pre- and post-contrast (Magnevist, 0.25 mmol/kg) with the following parameters: matrix: 256 × 192, 27 slices, 0.4 mm thickness, 0.1 mm gap, number of echoes (NE) = 1, number of averages (NA) = 4, TE/TR = 16/800 ms.

A complete set of MRI (T1 and T2) and DCE-MRI parameters (plasma distribution volume, vp; blood-to-brain forward volume transfer constant, Ktrans; interstitial volume fraction, ve; VD, extracellular distribution volume) for the MR contrast agent (CA) Magnevist (gadopentetate dimeglumine; Bayer Healthcare, Wayne, New Jersey) were acquired using published methods [31, 33]. A unique feature in the DCE-MRI analysis was the use of a model selection paradigm, which attempted to best balance variance and bias in the estimation of model parameters, and thus stabilize estimates of cerebral physiology produced by DCE-MRI [34].

Histopathology and reverse phase protein array (RPPA) assays

Rats were euthanized by isoflurane overdose, followed by decapitation. Whole brains were quickly removed and placed in rat brain matrix. With the implantation needle entry as the center, 2 mm thick, coronal slices were cut anterior and posterior to the center to generate frozen and formaldehyde fixed, paraffin embedded (FFPE) tissues. Slices for frozen sections were placed on cork circles and were covered in OCT freezing medium prior to flash freezing in 2-methylbutane cooled to —40 °C with dry ice. Slices for FFPE were placed in mesh cassettes and immersion fixed in formaldehyde for later paraffin embedding. Five μm thick sections from the FFPE slices were deparaffinized and processed through a graded alcohol series for hematoxylin and eosin (H&E) staining. Additional 5 μm thick sections underwent a 15 min incubation in pH 6.0 citric acid buffer for antigen retrieval, followed by incubation in Rodent Block M (Biocare Medical, Pacheco, California) for 20 min. They were then incubated with human marker major histocompatibility complex I (MHC; ab52922, Abcam, Cambridge, UK), 1:1000 for 60 min. After primary antibody incubation, slides were incubated for 30 min in Rabbit on Rodent Polymer (Biocare Medical, Pacheco, California) and for 4 min in Betazoid DAB (Biocare Medical, Pacheco, California). Images were captured using a Nikon E800M microscope and DXM1200C digital camera with Image Pro-Plus software (Media Cybernetics, Rockville, MD). Stained sections were examined for tumor location from both H&E and MHC stained specimens.

RPPA

The flash frozen rat brain tissue blocks were stored in a —80 °C freezer until processed for RPPA. One 5 μm section was cut from each slice and stained with H&E to localize the tumor boundaries to aid in dissecting out the tumor tissue. With reference to this staining, tumor tissue was macrodissected, weighed and placed in 5 ml tube on ice. Freshly prepared lysis buffer was added to the tube at a ratio of 40 mg tumor/ml buffer. The tumor tissue was homogenized in the tube on ice for 8 s and then the tubes were centrifuged at 4 °C, 14,000 RPM for 10 min. The supernatants were transferred to another set of microcentrifuge tubes and their protein concentrations were measured and the buffer volume was adjusted to 1.5 μg protein/ml. The lysates were then mixed in 3:1 ratio with sample buffer with 4X sodium dodecyl sulfate, but without bromophenol blue, boiled for 5 min and stored at —80 °C until shipped to MD Anderson Cancer Center for analysis. Brain tissue samples from two rats with no tumor implantation were included as controls.

Data analysis

Parameter maps for T2, T1 pre- and post-contrast were made using published procedures [32, 35]. The analysis of DCE-MRI data followed established procedures [34], with a model selection paradigm centered around the multi-parameter Patlak Plot, and producing, at most, non-biased estimates of the forward vascular transfer constant, Ktrans, the extravascular, extracellular volume fraction ve, and the plasma volume, vp in tissue that had sufficiently permeable vasculature. In general, estimates of Ktrans and ve could only be constructed from those tumors exhibiting true contrast extravasation and could not be constructed in tissues that exhibited post-contrast T1 hyperintensities due to hypercellularity and/or vasodilation.

In addition to the parametric analysis of the DCE-MRI data, the T1 and T2 image sets were examined and graded. A board-certified radiologist blinded to the conditions of the study reviewed all cases and graded each for radiological evidence of BBB breakdown on a scale of 0 to 4, with 0 being no evidence, and 4 being florid enhancement. The radiological images were visually compared to their corresponding histopathology to match tumor location and extent. They were also compared to clinical T2 and post-contrast T1 images after patient de-identification.

Ordinal logistic regression (OLR) modeling and Wilcoxon rank sum tests were used to analyze radiologic BBB breakdown scores. The RPPA data were examined for significant changes in levels of any of the 384 analyzed proteins between tumors from control rats (n = 3) and those implanted with Matrigel (n = 5) using multiple t-tests. Statistical significance was determined using the Holm-Sidak method, with alpha = 0.05.

Results

In 32 animals there were 5 extra-axial masses and 30 intra-axial cerebral tumors. All 5 extra-axial masses enhanced floridly. Two animals with only extra-axial masses were removed from further consideration. The average post-implantation days for all rats in a given neurosphere group at which the tumors were studied were, 84.4 ± 5.8 days for HF2303 PDOX model (mean post-implantation termination time: M0=86.4 days; M−=77.8 days; M+=89 days) and 57.2 ± 12.6 days for the HF3016 PDOX model (mean post-implantation termination time: M0=42.7 days; M−=64.3 days; M+=64.8 days).

Visually, the two cell lines demonstrated quite different responses to Matrigel, with no subjective difference between M+ and M− in either cell line. Typical control HF3016 cell lines (M0) showed little or no T1WI post-Gd enhancement (Fig. 1), and typical M+ and M− cell lines showed florid enhancement (Fig. 1). Although HF2303 PDOXs were visible both pre- and post-contrast (Fig. 2), subtraction images showed little post-Gd enhancement under any condition (M0, M+, or M−). Fig. 3 summarizes the radiologist’s scores via a dot plot; the trends described are clearly reflected therein.

Fig. 1.

A collage of MRI and histology images from three HF3016 studies, one each from M0, M− and M+ groups along with their associated patient MRI images in the bottom panel. The embedded tumor is indicated by a solid, white arrow. Note the agreement between MRI and histology in tumor location, extent and central necrosis when present, in all cases. All three experimental groups showed T2 hyperintensity, but only M− and M+ groups showed T1-weighed contrast enhancement and could be used to produce maps of model selection. The H&E and MHCs-stained sections showed the contrast-enhancing tumor part, as well as the MRI-invisible, infiltrative tumor along white matter. The patient tumor showed both T2 hyperintensity and T1 contrast enhancements (bottom panel). The large T2 contrast enhancement in M0 is the enlarged, contralateral ventricle.

Fig. 2.

A collage of MRI and histology images from three HF2303 studies, one each from M0, M− and M+ groups along with their associated patient MRI images in the bottom panel. The embedded tumor, when visible, is indicated by a solid, white arrow. Note the agreement between MRI and histology in tumor location and extent. All three showed T2 hyperintensity, but none showed T1-weighed contrast enhancement. The H&E and MHCs-stained sections showed embedded as well as infiltrative tumor tissue. Note the small tumor segment on the cortical surface that showed T1 enhancement in all three experimental groups. Similar to the HF3016, HF2303 human tumor also showed both T2 hyperintensity and T1 contrast enhancements (bottom panel).

Fig. 3.

Dot plot of blinded scores from a neuroradiologist’s assessment of T1-weighted contrast enhancement in the two tumor types. More rats in the HF3016 cohort were scored toward 4, florid enhancement.

Statistical analysis supported these observations. The OLR model was fit with factors treatment (three levels: M0, M+ and M−) and cell line (two levels HF2303 and HF3016). The test for interaction between cell line and M+ factor was not significant (p = 0.451). There was also no evidence of a growth factor effect (p = 0.725 for cell line HF2303 and p = 1.0 for cell line HF3016). The Matrigel treated groups with and without growth factor were then combined. After this, the test for interaction between cell line and Matrigel was still not significant (p = 0.195). However, when the Matrigel effect was tested using a Wilcoxon Rank Sum test separately within each cell line, the test was significant for HF3016 (p = 0.004), but not for HF2303 (p = 0.328).

A comparison of MRI parameter maps with histologies showed spatial agreement between the two main tumor masses in both (Figs. 1 and 2). However, the histologies showed infiltrating tumor tissue along white matter tracts that was not observed on MRI. Such infiltrating tumor regions were more prominently visible on MHC-stained than on H&E-stained images (Fig. 1). The M− and M+ cohorts in the HF3016 group could be analyzed for DCE-MRI studies with well-defined regions of vascular breakdown, and amenable for Patlak plot-model selection maps. These quantitative DCE analysis provided information about virtual tumor segmentation in terms of its vascular permeability and contrast agent extravasation for Magnevist (Table 1). A preliminary analysis suggested comparable vascular volumes (vp), but a trend of increased interstitial spaces in M+ rats (elevated ve and VD values). However, the differences were not statistically significant (t-test).

Table 1

Tumor transvascular kinetics values for Magnevist in HF3016 M− and M+ tumors.

HF3016 M−	vp	Ktrans	ve	VD
Mean	0.0094	0.0386	0.1815	0.1651
SD	0.0062	0.0152	0.0479	0.0137
HF3016 M+	vp	Ktrans	ve	VD
Mean	0.0126[1]	0.0482[2]	0.3503[3]	0.2710[4]
SD	0.0050	0.0312	0.2175	0.1155

p values:

0.49.

0.56.

0.12.

0.08.

Apart from the spatial correlation between tumor cells and MRI-visible tumor segments in contrast enhancing PDOXs, the histopathology from the PDOXs exhibited microscopic infiltration that recapitulated human glioma (Fig. 4). Tumor cells, singly and in small clusters that were unaccompanied by contrast enhancement on MRI, were observed to be infiltrating into the surrounding brain tissue. These tumors also showed regions of psueudopalisading necrosis within the main tumor mass.

Fig. 4.

A1, MHC-stained coronal brain section (main tumor mass is indicated by a black solid arrow; also showed in Fig. 1); A2, box in A1 enlarged to show infiltrating glioma cells along the white matter, but invisible on MRI (some tumor cells are indicated by black arrows). B1 to B4 depict H&E-stained coronal brain sections from the two cell lines as described below to show microscopic, histopathologic features. B1, HF2303 M−; B2, HF3016 M0; B3, HF3016 M−; B4, box in B3 enlarged. Note the nearly uniform distribution of glioma cells in both B1 and B2 in comparison to regions with necrosis and irregular blood vessels in B3. B4 shows a region with pseudopalisading necrosis typical of GBM.

An interesting feature noted was that extra-axial tumors close to the brain surface, presumably contiguous with pial vasculature, typically showed T1 contrast enhancement independent of its Matrigel status. An example is shown in Fig. 5 wherein two distinct tumor segments are visible, but only the extra-axial tumor segment shows T1 contrast enhancement whereas its intra-axial, embedded segment does not. This phenomenon is also visible in T1 post-pre panels in both figures 1 and 2 wherein extra-axial tumor tissue on cortical surface shows contrast enhancement.

Fig. 5.

As in Figures 1 and 2, a typical set of images from an animal implanted with cell line HF2303, with co-injection of M+. There is florid T1 contrast enhancement of the extra-axial (black star), but not of the intra-axial tumor (solid white arrow).

To compare the molecular characteristics of the terminal HF3016 xenograft tumors grown without Matrigel (M0 group, n = 3) or with Matrigel (M+, M−; n = 5) we employed a targeted proteomics approach. The levels of 384 proteins and post-translational modifications, representing key pathways in cancer, were measured in tumor lysates via Reverse Phase Protein Arrays (RPPA). Lysates from two HF3016 preimplant CSC cultures were included (Supplemental Table 1). Using multiple t-test with statistical significance determined using the Holm-Sidak method, with alpha = 0.05, we determined that only 3 analytes were significantly different between the M0 and M(+,−) groups (Fig. 6). The phosphorylation of the transcription factor FoxO3 at S318/S321 was significantly decreased in the M(+,−) tumors. The Akt directly phosphorylates these conserved serine residues, leading to FoxO3A nuclear export, thus inhibiting its transcriptional regulation activityOn the other hand, the levels of phospho-S6 (S240/S244), a canonical indicator of mammalian target of rapamycin (mTOR) activity, and of the UV excision repair protein Rad23 homolog A (Rad23A) were significantly increased in the Matrigel group. Prior to implant, HF3016 presented low levels of phospho-S6, phospho-FoxO3A and Rad23A relative to PDOXs.

Fig. 6.

Long term effect of Matrigel at implant on the proteomic composition of rat HF3016 models. Normalized protein and phosphoprotein levels in the CSCs (n=2), PDOX tumors with no-matrigel control (n = 3), or with Matrigel (n = 5), measured by reverse-phase protein arrays (RPPA) and compared among groups using Holm-Sidak test. p-values are shown; N.S., non-significant.

Discussion

Using two patient-derived neurospheres, this study describes a simple method that shows promise for inducing an intra-axial PDOX brain tumor that presents radiological features of the parent tumor. Such tumors showed a contrast MRI-visible, space occupying tumor mass with MRI-invisible, infiltrating tumor regions, similar to human GBM. However, the method needs to be tested with other neurosphere cultures and matrix scaffolding preparations to fine-tune the combination to suit given human GBM neurospheres. As a practical matter for preclinical studies, a reliable BBB breakdown in PDOX models represents a significant advance in the assessment of trial therapies because it produces an experimental condition closer to the human pathology of interest. However, as demonstrated herein, the effect of stromal enhancement is cell-line dependent, and the biological correlates that accompany BBB breakdown after stromal enhancement still need to be described, an interesting question in itself. Despite these limitations, note that tumor visualization via T1 contrast-enhancement is the clinical standard that determines the extent of tumor resection in both newly diagnosed and recurrent GBM [36, 37]. An orthotopic animal model that recapitulates this critical feature while also retaining the molecular fingerprints of the patient tumor will be highly valuable in testing new therapies.

Using targeted proteomics to evaluate any lasting effect of Matrigel on molecular composition of the tumors, we showed that the terminal tumors from cells implanted with Matrigel were indistinguishable from those implanted without Matrigel in relation to the levels of the majority of analyzed key proteins and post-translational modifications measured. In the tumors from the Matrigel group we observed a decrease in phosphorylation of FoxO3 at the S318 and S321 residues, suggesting increased nuclear accumulation. FoxO3 is a transcription factor involved in homeostasis regulation in neural stem cells [38]. Interestingly, an increase in phospho-S6, a marker of mTOR activity, was also observed in the Matrigel group.

In neural stem cells, FoxO3 is the major member of the FoxO family. Phosphorylation of FoxO3 via phosphoinositide-3-kinase (PI3K)-AKT results in its binding of protein 14–3–3, and its subsequent exclusion from the nucleus to the cytoplasm, i.e., its inactivation. Once FoxO3 is removed from its nuclear site, its suppression of cell cycling is disabled, which leads to in-situ proliferation [39]. FoxO’s are lineage-restricted redundant tumor suppressors [40]. According to Paik et al., FoxO3 also plays a role in angiogenesis, although not exclusively. Complete knockout of FoxO’s in mice resulted in a hemangiomic phenotype, mainly due to FoxO1 knockout, but also related to FoxO3 knockout in endothelial cell populations [40]. Its relevance to the PDOX phenomena needs further studies because the vasculature recruited by the tumor cells is probably murine vasculature with a full complement of FoxO’s.

We note that there are some limitations to these data with regards to known inconsistencies in Matrigel preparations. This is also evident in the present data where the two Matrigel preparations gave rise to slightly different tumors. Use of M− resulted in tumors with slightly lower Ktrans, and ve and VD suggesting more tumor cell density [35] than in M+ tumors. The stated compositions of the two Matrigel matrix preparations were similar, i.e., laminin as a major component, collagen IV, heparan sulfate proteoglycans, entactin/nidogen, and several growth factors. However, the proportions of these moieties and the growth factors could be different between the two leading to slightly different tumor subtypes in vivo. A tightly controlled Matrigel composition can overcome this limitation which can then be tested with other GBM neurosphere cell lines to generate PDOX models. Also, although no BBB breakdown was observed in rats after Matrigel injection (Supplemental data), we concede that more subtle parenchymal effects than a disturbance of the BBB may have occurred in the vicinity of the injection.

While no animal model is completely representative of human disease in all its complexity, the establishment of tumors in immune-deficient animals by xenografting human tumor material, mostly in the form of cell lines and biopsies, has been invaluable in the search for mechanisms that determine tumor growth and progression [11, 12, 14, 15, 26]. Xenograft models include human biopsy spheroid xenograft, human glioma cell monolayers grown in serum-containing media and human glioma-derived neurospheres grown in serum-free neurobasal media. Xenograft tumors from established 2D serum-cultured glioma cells lines are fast growing, reproducible, and generate a BBB breakdown, but the tumors deviate in genotypic and phenotypic characteristics (typically reduced invasiveness) from the original patient tumor [6, 26]. Further, developing models from low grade gliomas has proven difficult with this approach. In contrast, neurospheres grown in serum-free media more closely reiterate the molecular makeup of the original patient tumor, with molecular profiles being more stable over time [6, 8, 9, 11, 26]. This approach is also suitable for low grade gliomas. Neurospheres grown as PDOXs are now considered as the most representative and important tool to study in vivo tumorigenicity and progression of GBM [14, 15, 18, 41].

As we and others have noted [5, 6], multiple experimental therapies effective in animal models have failed in clinical trials [42, 43]. The success of treating model brain tumors in rodents and the failure in human GBM in a clinical setting is attributed to such factors as the tumor models: (1) do not reflect the biological properties of the patient tumors, and (2) they do not reflect the cellular and molecular heterogeneity of human tumors. The inability of current preclinical systems to mimic the biology of human GBM in situ is increasingly cited as a key cause of low success rates for translation of drug discovery efforts to the clinic [44]. Therefore, there is an urgent need for more reliable in vivo models in order to facilitate the elucidation of pathophysiology and for testing new drugs and therapies. In this work, we present compelling evidence that a relatively minor manipulation of the extracellular matrix (ECM) during tumor implantation can result in PDOX tumors that are radiologically similar to the human tumors they were derived from.

Matrigel is used to provide a stiff matrix for tumor cells’ development into three dimensional and organoid tumor models [45-47]. Matrigel has been extensively used for subcutaneous and to a lesser extent orthotopic in vivo tumor cell implantation to presumably enhance tumor take rate and growth [48-50]. Although the specific mechanisms leading to tumor growth enhancement by Matrigel are not well understood, it likely involves stimulation of CSC survival and proliferation, as observed by neural progenitor cells implanted in rodent spinal cord and brain [51, 52]. Some protocols recommend mixing tumor cells with low growth factors-Matrigel before initial heterotopic implantation and later heterotopic in vivo passaging [53]. Interestingly, this protocol does not recommend the use of Matrigel in intra-cranial implantations [53]. However, other studies have used co-implantation of tumor cells and Matrigel in both heterotopic [53, 54] and orthotopic GBM preparations [55, 56]. Das et al. used a 1:1 mixture of GBM cells and Matrigel for intra-cranial implantations [56], but did not specify the Matrigel type. The orthotopic tumors derived from them and implanted in mice brains seemed to be continuous with pial distribution and showed hyperintensity. The imaging sequences not being specified and the observed contiguity with pial vasculature, however, make it difficult to determine the extent of T1 contrast enhancement in these tumors [56]. Tsoli et al. generated in vivo models of diffuse intrinsic pontine glioma with implantations with and without Matrigel, but reported no specific effects of Matrigel on resultant tumors [55].

One study has reported evidence of nodular tumor growth and a modest correlation in post-contrast enhancement to the human tumor in one of the seven mouse PDOX models studied [57]. The authors concluded that the MRI features of patient-derived xenograft glioma models were different compared with those of the original patient tumors. Nonetheless, generation of PDOX models of GBM consistently recapitulating radiologic features of patient tumors is still not widely reported. It is important to note that some previous attempts have been made to develop rat PDOX models that exhibit both the characteristic GBM phenotypes, i.e., high neovascularization and diffuse infiltration [58-60]. This is deemed necessary since a typical GBM exhibits a solid, angiogenesis-dependent tumor with mass effect as well as infiltrative, vascular co-opting segments and, thus, essential for an experimental tumor to replicate these features [61]. Sakariassen et al. reported developing angiogenesis-dependent tumors from infiltrative tumors after several successive in vivo passages in immune-suppressed rats [59]. First generation tumors in this series showed diffuse lesions that occupied most of the implanted hemisphere and just T2 hyperintensity, atypical of patient tumors. After serial in vivo passaging for 4–5 generations, the resultant tumors became more vascular and circumscribed with emerging necrotic regions along with T1 contrast enhancement. However, as noted before, the procedures required that GBM spheroids from rat tumors be passaged in vivo into the next generation of nude rats. Similar results have been reported by two other groups. Thorsen et al. used spheroids generated from tumor biopsy specimens for initial orthotopic implantation; they reported forming a more typical GBM-like orthotopic tumor with extensive angiogenesis and necrosis from an initially highly infiltrative, non-angiogenic tumor after three serial passages in immune-suppressed rats [60]. It took about six in vivo passages in another study to arrive at a similar model [58]. In comparison to these studies, the approach described in the present work will likely result in a representative tumor phenotype from a GBM neurosphere cell line in the first generation of implanted rats, saving time and resources, and minimizing possible clonal selection alterations over serial in vivo passaging. Further work with other available human neurosphere lines that are being generated to test precision medicine approach will elucidate the wider applicability of this technique. With the short overall survival for GBM, minimizing the time needed to test a putative therapy in the animal model that most closely resembles the characteristics of human tumor and reflects its response to therapy is critical for overcoming the significant challenges for effective treatment for GBMs.