Abstract
In situ hybridization coupled to immunohistochemistry for antigens of interest allows unequivocal identification of tumor cells from reactive stroma cells and normal adjacent structures in human glioblastoma multiforme grafts transplanted into nude mice. With this methodology, we have explored the development of glioblastoma multiforme solid grafts transplanted into nude mouse brains or flanks. The brain transplants closely resembled the human situation, particularly in relation to differentiation and growth patterns. The morphological features of peritumoral reactive gliosis were similar to those observed in humans. A mouse glial stroma within the main tumor masses was also demonstrated. Kinetic studies showed that the compartment of isolated tumor cells that infiltrated host brains and the reactive gliosis constituted two cycling cell populations. Despite VEGF protein expression by tumor cells and some reactive astrocytes, the abnormally permeable microvascular beds were not hyperplastic. The observation of a non-infiltrative pattern of growth when grafts were established in host flanks demonstrated that the organ-specific environment plays a determining role in the growth and invasive properties of glioblastoma. The phylogenetic distance between man and mouse and the recipient immunoincompetence should not impose serious limitations on the use of this model for studying malignant glioma biology and therapy in vivo.
I
With the availability of such non-selecting, non-perturbing, specific and topographically precise methods, we have explored the development of human glioblastoma multiforme (GBM) solid grafts transplanted into nude mouse brains after being passaged within mouse flanks. Particular attention has been paid to the following characteristics: (a) tumor differentiation and pattern of growth within the host brain; (b) the reactive gliosis at the tumor-host interface and in the tumor solid component; (c) the proliferative activity of the malignant astrocytoma cells, not only in the main tumor masses but also in the compartments of isolated tumor cells infiltrating the normal host brains, whose existence was easily demonstrated on ISH and immunohistochemically stained microscopic sections; and (d) the graft microvascular bed, because endothelial proliferation and rupture of the blood-brain barrier are two highly characteristic alterations of GBM in humans. Furthermore, we have studied the contrasting pattern of growth of GBMs established in host flank, demonstrating that, as for other malignant tumors, the organ-specific microenvironment plays a determining role in the growth and invasive properties of malignant astrocytoma.
Materials and Methods
Donor Tissue
The donor tissue consisted of fresh samples of typical glioblastomas multiforme (WHO classification 1993) from two patients (Patients V and D) taken from the pathological specimens obtained at surgery. Tumor tissue was dissected free of necrotic areas and blood clots, rinsed, and then cut into 1.5-mm-thick slices, which were placed in a sterile isotonic glucose solution until grafting.
Animals and Tumor Transplantation
Animals. Four- to 7-week-old, pathogen-free, congenitally athymic nude mice, on a Swiss nu/nu background, were obtained from IFFA-CREDO Breeding Laboratories (Lyon, France). Animal experiments were performed according to institutional and national guidelines. All the surgical procedures, as well as the magnetic resonance imaging (MRI) of brain xenografts, were carried out under general anesthesia with IP ketamine (0.1 mg/g body weight).
GBM Primary Flank Implantation into Nude Mice and Further Propagation. Tumor slices from both patients were inserted into the flanks of five mice (four fragments per mouse) between the rectus abdominis and the posterior lamina of its sheath, in close proximity to the epigastric vessels. The abdominal tumors were excised at 1-2 cm3 in size, morphologically examined, and transplanted into the abdominal wall of another group of mice or into mouse brains. In total, the two original tumors were passed from animals to animals for 3 years (five generations of mice, five mice per generation and per original surgical specimen). Finally, tumor grafts were implanted into the flanks of three more mice to study the early stage of the primary abdominal implant take.
Intracerebral Grafts. Forty animals (10 per generation of mice) that had received grafts into their abdominal wall were also transplanted intracerebrally, both transplantations being performed during the same surgical procedure. For transplantation into the mouse brain, a bifrontal coronal incision was made and the scalp, including the pericranium, was separated from the cranial vault and turned inferiorly over the face. Next, a right monofrontal osteoplastic bone flap was performed. After incision of the dura, one tumor fragment 2-3 mm3 in size was implanted a few millimeters behind the central sulcus to a depth of about 1.5 mm. The craniotomy flap was then replaced.
Magnetic Resonance Imaging
A 1.5 T clinical Signa version 5.5.1. (General Electric MS) was used for brain imaging. Seventeen apparently normal mice were studied at various times from Day 45 to Day 211 after tumor implantation. Five animals were examined twice, with an interval of 20 days between the examinations. The contrast material, gadopentate dimeglumine (Magnevist; Schering, Lys Les Tannoy, France) 0.5 ml (0.25 mmol) was injected IP 10 min before examination, performed with a 3GP antenna (GEMS). The console settings chosen to optimize signal-to-noise ratio and spatial resolution were as follows: image matrix 256 × 192; field of view 8 × 6 cm; slice thickness 3 mm. T1 spin-echo pulse sequence was used with a repetition time of 300 msec. The echo time was 12 msec, with two excitations. The duration of this sequence was 90 sec. Four coronal and sagittal 3-mm slices were obtained per brain, with 0.3-mm of interslice.
Study of Tumor Microvasculature Permeability
Abnormal permeability of the tumor microvasculature was looked for in three mice, sacrificed on Day 65 (one mouse) and Day 106 (two mice), by studying the volume of distribution of Evans blue after a single IV injection of 200 μl of a 2% (w/v) solution of the colored tracer in a 0.9%(m/v) NaCl solution injected 2 hr before sacrifice.
Tissue Collection and Morphological Studies
Tissue Collection. The three mice that received flank implants for studying the early stages of the primary abdominal graft take were sacrificed on Day 3, 10, or 15. Similarly, three mice from the second generation were sacrificed on Day 3, 10, or 15 to study early brain tumor take. Three of the 37 remaining mice with intracranial grafts died spontaneously and were autopsied. The other mice were sacrificed when they became cachectic or at various post-implantation times for morphological studies. From each of the two initial brain tumor samples and from each abdominal and intracerebral transplant, several slices were cut and fixed in either ethanol/acetic acid 3/1 (v/v) or 4% (m/v) buffered formaldehyde for 16 hr and embedded in paraffin. Slices of three intracerebral tumor transplants were also frozen in isopentane precooled in liquid nitrogen.
Histochemical Staining. Sections of all paraffin-embedded blocks were stained with hematoxylin-eosin (HE) and with the Bodian stain for demonstration of nerve fibers and neurofibrils.
Detection of Cell Genotypes by ISH. ISH was performed on ethanol-acetic acid-fixed, paraffin-embedded sections with fragmented total mouse or human genomic DNAs, digoxigenin-labeled by the random primer labeling method (Plénat and Duprez 1990). Detection of the digoxigenin label was achieved by incubating the sections in a solution of Fab fragments of alkaline phosphatase-labeled sheep antibodies (Boehringer; Meylan, France), diluted 1:200 in 0.1 M Tris-HCl, pH 7.5, 0.3% (v/v) Tween-20 for 1 hr at room temperature. The color development was carried out by incubating the slides for 30 min with substrate solution: 0.33 mg/ml 5-bromo-chloro-indolyl phosphate (BCIP), 0.33 mg/ml nitroblue tetrazolium (NBT), 0.77% (v/v) N,N-dimethylformamide in 0.1 M Tris-HCl, 0.1 M NaCl, 0.05 M MgCl2, pH 9.4 (McGadley 1970).
When double ISH with mouse and human genomic DNAs was carried out in successive steps on the same sections, the tissue sites of the first probe were visualized with the BCIP/NBT method, which produces a dark-blue reaction product. The second probe was also detected with the alkaline phosphatase-labeled anti-digoxigenin antibody, the alkaline phosphatase activity being developed with a naphthol AS-MX/Fast Red TR salt method.
Antibodies and Immunostaining Procedures
Antibodies. Formaldehyde-fixed tissue sections of the two original tumors and every transplant were studied with a species-specific monoclonal antibody to human vimentin (Dako, Trappes, France, diluted 1:400) and a cross-species-specific antiserum to glial fibrillary acidic protein (GFAP) (Dako; diluted 1:800) for analyzing the evolution of tumor cell differentiation with successive passages. A combined immunohistochemical-ISH method was carried out with antiserum to GFAP and the mouse DNA probe to study the morphological and topographical characteristics of gliosis reactive to tumor development in the mouse brain. Tumor growth fractions were measured on sections stained with the species-nonspecific monoclonal antibody MIB-1, directed to the cell cycle-associated antigen Ki-67 (Immunotech, Marseille, France; diluted 1:400), by recording the percentage of positively labeled tumor cells in five independent areas of the main tumor masses. Each area included 800-1000 cells at a X 200 magnification. Dual immunohistochemistry on the same section using successively antibodies to either human vimentin or GFAP and then to Ki-67 was also performed on selected sections to appraise the proliferative activity of the compartment of malignant astrocytes infiltrating the normal brain parenchyma. The presence of cycling astrocytes in the mouse reactive gliosis was looked for by double staining sections with antibodies to GFAP and Ki-67 and comparing the data thus obtained with those of ISH techniques with the mouse DNA probe performed on alternating sections. On microscopic preparations immunostained with the MIB-1 antibody, the endothelial cells were most often difficult to recognize with confidence. Therefore, dual immunohistochemical techniques with the MIB-1 antibody and a species-nonspecific antibody to mouse Type IV collagen (Immunotech; diluted 1:500), followed by light hematoxylin counterstaining, were also carried out to identify the fraction of cycling endothelial cells.
The development of tumor vasculature was investigated with a cross-species-specific antiserum to factor VIII-related antigen (Dako; diluted 1:50), as well as with species-specific antibodies to human Type IV collagen (a generous gift of S. Guerret-Stocker, Institut Merieux, France; diluted 1:500) or to mouse Type IV collagen (Immunotech; diluted 1:500). Dual immunohistochemical techniques on the same section were performed on selected slides with the antibodies raised against factor VIII-related antigen and murine Type IV collagen to look for possible blood channels without basement membrane. In addition, combined immunohistochemical-ISH studies with the antibody to mouse Type IV collagen and either the mouse or human DNA probe were carried out to determine the tumor endothelial cell genotype. Because preliminary studies had shown that the vascular basement membranes appeared continuous on sections immunostained for murine Type IV collagen, microvessels were counted on sections immunostained for this antigen. Angiogenesis was scored using the vasoproliferative component of the microscopic grading system scale used to quantify angiogenesis in a variety of tumors (Brem et al. 1972). The histological slides were first scanned at low magnification, and the areas of maximal vascular density were selected for grading. Microvessels were then counted in 10 high-power fields, each mesuring 0.414 mm2). Pericytes were looked for with a cross-species-specific antibody to α-smooth cell actin (Dako; diluted 1:1000) and VEGF protein synthesis with a cross-species-specific monoclonal antibody (Calbochiem, Meudon, France; diluted 1:40). The species of origin of the cells synthesizing this growth factor was determined on selected slides by a combined immunohistochemical-ISH approach. Finally, dual immunohistochemical techniques on the same section were carried out on selected slides with the antibodies raised against human vimentin and murine Type IV collagen to study the possible relationship between the migrating tumor cells and basal membranes in the brain.
The presence of fibrin and murine IgGs in the nonvascular tumor extracellular matrix, testifying to an abnormal permeability of the tumor microcirculatory bed, was looked for in the brains and the abdominal tumors of three mice sacrificed 65 (one mouse) or 106 (two mice) days after tumor implantation. These experiments were carried out with appropriate fluorescein-conjugated primary antibodies (Institut Pasteur, Paris, France; diluted 1:10 and 1:40 for fibrin and murine IgGs, respectively).
Staining Procedures. One-hr incubations with the antibodies to fibrin and murin IgGs were carried out on 6-μm-thick unfixed frozen sections that were subsequently counter-stained with a 6 mg/ml solution of ethidium bromide in a 0.1 M phosphate buffer, pH 7.4. All the remaining primary antibodies were applied for 16 hr at 4C. Detection of tissue-bound primary antibodies was performed using the biotinylated antibody/streptavidin-peroxidase detection system. Bound peroxidase was identified using either the diaminobenzidine-H2O2 or the 3-amino-9-ethylcarbazole method (Plénat et al. 1994). When double immunostaining was performed, streptavidin-alkaline phosphatase conjugate was often used in place of the streptavidin-peroxidase one. Alkaline phosphatase activity was visualized with either the naphthol AS-MX phosphate/Fast Red TR salt or the BCIP/NBT method for the first step, and a naphthol AS-MX phosphate/Fast Blue BB salt for the second step (Plénat et al. 1994). Some dual immunolabelings were performed, however, with streptavidin-peroxidase complex and the 3-amino-9-ethyl-carbazole method of revelation of peroxidase activity as the first step and streptavidin-alkaline phosphatase and the BCIP/NBT method for the second one.
Antibodies to vimentin, GFAP, and factor-VIII-related antigen were applied on sections previously pressure-cooked for 5 min in a 0.01 M sodium citrate buffer, pH 6.0. Furthermore, for Type IV collagen detection, Before being pressure cooked, slides were enzyme-digested by incubation in a 0.1% (w/v) solution of trypsin in 0.1% (w/v) CaCl2, pH 7.8, for 15 min at 37C. When the immunoenzymatic detection of vimentin, GFAP, and VEGF was combined with the detection of cell genotype by ISH on the same section, immunohistochemistry was performed before ISH, using the biotinylated antibody/streptavidin-alkaline phosphatase detection system and the naphthol AS-MX/Fast Red method of visualization of alkaline phosphatase activity. The DNA probes were visualized as described above. All immunohistochemistry was controlled by substituting non-immune serum for the primary antibodies.
Results
Tumor Take
All the mice survived the surgical procedures, and the animals' clinical status was in all cases normal after the recovery from anesthesia.
At least one of the four parieto-abdominal transplants in each animal grew, producing a palpable nodule. Initial growth of the primary abdominal implants was delayed for a mean time of 21 weeks before tumors became palpable. The delay was shorter after the first passage: 5.5 weeks for the second generation, 4.9 weeks for the third, and 2.2 weeks for the last two generations. Secondary regression of the abdominal tumors was never observed.
None of the mice with a brain graft developed focal neurological signs until the end of the observation period. Every intracranial transplantation produced a tumor. Seventeen of the 37 mice that were grafted intracranially were sacrificed at various intervals after surgery for morphological studies. Three other mice died spontaneously, and the remaining 17 were sacrificed because of cachexia. Death of these last 20 animals occurred as early as Day 45 and as late as Day 217, the mean survival time after transplantation being 106 days (99 for the mice grafted with the V tumor line and 125 for those from the D line).
Morphology of the Cerebral Tumor Transplants Harvested After Day 15 Post Grafting
Gross Morphology. In only 26 of the 37 animals that received intracerebral grafts was the tumor macroscopically evident, localized in the right hemisphere above the lateral ventricle. Of these 26 tumors, 22 measured more than 6 mm in largest diameter and four from 3 to 6 mm. In three tumors whose diameter exceeded 6 mm, the growth insinuated itself between the bony flap edges and the skull, and infiltrated the scalp soft tissue. In these three cases, the extracranial portion of the tumor was vascularized by branches of the external carotid artery. In all other cases, healing of the bony opening in the calvaria had occurred, and the tumors were vascularized by branches of the circulus arteriosus only. In five cases, which included the three tumors with extracranial extension, the tumor invaded the dura and the subdural space. There was no macroscopic tumor spread to the contralateral side, even though severe midline shift and compression of the ventricular system and brainstem were observed in the largest tumors. Except for the tumors with an extracranial extension, tumor necrosis was not macroscopically observed. In 11 animals, the tumors were microscopic, no brain alterations being visible at the macroscopic level.
Neuroimaging. The signal-to-noise ratio on T2-weighted imaging was poor, and the tumor grafts could never be detected. No signal abnormalities were seen on the precontrast T1-weighted images. Each time, the postgadolinium T1 sequences revealed typical intense contrast enhancement in the right hemisphere (Figure 1). The contrast-enhanced zones corresponded to the solid tumor component and did not represent the outer tumor border, because isolated tumor cells were identified on ISH-stained sections of the corresponding brains, far away from the main tumor masses. In one animal, contrast enhancement was observed as early as Day 45 post grafting. On two occasions, contrast enhancement that was not observed on Day 45 appeared clearly visible 20 days later. In three mice, tumor growth was well visible on successive MRI evaluations.
Histological Study. In HE-stained sections, features common to brain xenotransplants were representative of high-grade astrocytomas (Figure 2) and included (a) high cellular density with frequent nuclear crowding, (b) moderately differentiated cells, some multinucleated, with astrocytic features of abundant pink cytoplasm and short and ramified processes that disappeared into the background in a disorganized pattern, (c) frequent mitoses, and (d) pseudo-palisading of tumor cells around serpiginous areas of necrosis. This latter feature was observed in five cerebral transplants whose largest diameter was in excess of 6 mm.
All intracerebral transplants appeared to be composed of two components (Figures 3A and 3B): solid tumor tissue in which tumor cells were crowded, always containing remains of neurites in Bodian-stained sections, and tumor cells, most often isolated, sometimes grouped in small non-cohesive nests within an apparently normal brain parenchyma. Combined immunohistochemical-ISH assays showed that whatever the tumor and the site of engraftment, the cytoplasm of every tumor cell was strongly stained by the species-specific antibody to human vimentin. In addition, because all the cells in the transplants harvested more than 15 days after grafting that expressed the intermediate filament had a malignant glial phenotype, tumor invasion could be very precisely studied on vimentin-stained sections. The extent of normal brain invasion by the neoplastic astrocytes varied from tumor to tumor but was not correlated with postoperative survival. In every case, the main tumor mass extended through the cortex and came to the surface of the right hemisphere, penetrating the overlying sub-arachnoid space. In 20 tumors, astrocytoma cells extended further into the adjoining subdural space. From the main tumor masses, astrocytoma cells spread out laterally on the dorsal surfaces of the two cerebral hemispheres beneath the glia limitans, where they formed a carpet one or two cells thick, and into the spaces of Virchow-Robin as blood vessels entered the brain.
Coronal T1-weighted MRI showing intense contrast enhancement 30 min after IP injection of a gadolinium chelate contrasting agent. Neuroimaging was performed on Day 51 post grafting. The contrast-enhanced structures correspond to peripheral areas of the main tumor mass but do not represent the outer tumor border, because isolated tumor cells were identified in sections of the corresponding brain far away form the main tumor mass. Bar = 4 mm.
Serpentine area of necrosis surrounded by pleiomorphic neoplastic astrocytes (pseudo-palisading) in a xenograft removed on Day 119 post grafting. Bar = 45 μm.
Tumor cells, isolated or grouped in nests, were always observed within the semioval center of the right ventricle and in the cortex of the right frontal and parietal lobes. In 32 cases, a large number of astrocytoma cells were observed bilaterally along the corpus callosum-cortical Layer VI interface, but also deeper within the corpus callosum itself (Figure 3B). In 30 of the 35 tumors, neoplastic cells were found in the left internal capsule, a few tumor cells being detected within the deeper cortical layers of the left frontal and parietal lobes. In 10 tumors, malignant astrocytes were observed in the dorsal thalamus and the subependymal space, surrounding the lateral and third ventricles. In two tumors, the walls of these ventricles were focally destroyed and tumor cells proliferated within the corresponding ventricle cavities. In one tumor, astrocytoma cells had migrated to the ventral surface of the brain and formed a carpet many cells thick on the ventral chiasma and other medial structures.
Compared with the human surgical material, the mouse-grown tumors contained the same morphological cell types but appeared somewhat more homogeneous. As could be seen in HE-stained sections, tumor cell differentiation was stable with increasing passage. The stability of tumor cell differentiation was also substantiated by the study of intermediate filament expression by tumor cells. Whatever the tumor passage and the site of grafting, all the tumor cells expressed the intermediate filament vimentin, and 90% of them co-expressed GFAP, a proportion similar to that observed in the two original tumors. In sections immunostained for vimentin and then hybridized with the human DNA probe, the neoplastic astrocytes in the central core of the tumors, in the transition zone between areas of solid tumor and those showing isolated tumor cell infiltration, as well as those that lined up beneath the pia mater or grew in the subependymal regions, showed moderately abundant cytoplasm and multiple ramified cell processes of various lengths and were GFAP-positive. The vimentin-positive isolated tumor cells in otherwise intact brain parenchyma tended to be elongated, with scantier cytoplasm and unbranched processes. They most often appeared as naked nuclei on corresponding HE-stained sections and were GFAP-negative. Double immunostaining with the antibodies to human vimentin and to mouse Type IV collagen revealed that most of the migrating malignant astrocytes aligned themselves along the basement membranes of parenchymal blood vessels and the glia limitans externa, expansions of astrocytic foot processes being in intimate relationship with these basement membranes (Figure 3C).
Tumor Proliferative Activity. Proliferative activity was variable from tumor to tumor but was always prominent. The median distribution of the growth fractions was 26% for the intracranial transplants, with extreme values of 8% and 67%. Palisading cells around areas of necrosis always failed to express the cell proliferation-associated antigen. Dual immunohistochemistry on selected slides with the MIB-1 antibody and the antibody to human vimentin showed that in each animal the proliferative activity was of the same intensity in the main tumor mass and in the compartment of isolated cells that infiltrated the normal brain parenchyma (Figure 3D).
Reactive Gliosis in the Recipient Brain. An acute reactive astrogliosis always developed at the tumor-host interface. As observed in tissue sections immunostained for GFAP and then hybridized with the mouse DNA probe, reactive astrocytes in the brain parenchyma adjoining the tumors made monomorphous populations of uniformly spaced, stellate fibrillary astrocytes, with rare branched, attenuated processes of similar length radiating in all directions from small cell bodies, and a nuclear to cytoplasmic ratio less than that of the neoplastic cells. These reactive mouse astrocytes always appeared smaller than their human tumor counterparts.
Immunohistochemistry on successive sections with the antibodies to Ki-67 and to GFAP showed that reactive mouse astrocytes at the tumor borders constituted a cycling cell population (Figure 3E). Reactive astrocytes could also be detected within the main tumor masses (Figure 3F). Hybridization with the mouse DNA probe showed that an average of 5% of the cells in the main tumor masses were of murine origin, of which 30% appeared to be reactive mouse astrocytes in combined immunohistochemical-ISH studies. The remaining mouse cells were mainly the endothelial cells of the tumor microcirculatory bed (see below) and a few macrophages either dispersed or grouped around areas of necrosis. Neuron cell bodies were never observed.
Tumor Vascularization. The vascular density in the intracranial tumors harvested more than 15 days after grafting varied widely from tumor to tumor and within different areas of the same tumor. The median distribution of the number of vessels in 10 high-power fields in the 23 intracranial tumors in which this count could be performed was 61 (range 27-227). The mean vascular density in the intracranial tumors (62) was on average more than three times lower than that established for the normal mouse brain hemispheres (201) using the same methodology. Furthermore, the normally highly ordered arrangement of the brain microcirculatory bed was destroyed. The tumor vessels were tortuous, larger, and more irregular in cross-section compared to normal brain vessels.
(
The basement membranes of the tumor microcirculatory beds, as well as the glia limitans externae covering the brain cortical surface and in the subependymal spaces in the intracranial grafts, were uniformly stained by the murine-specific antibody to Type IV collagen but never by the human-specific antibody. The basement membranes appeared continuous, focally thickened, and often multilayered. In sections stained with the monoclonal antibody to factor VIII-related antigen, the tumor microvessels appeared to consist of a continuous single layer of thickened endothelial cells that never formed glomeruloid microvascular proliferation. The antibody to α-smooth cell actin showed exceptional isolated pericytes dispersed along the tumor microvascular network, whereas pericytes formed an extensive yet incomplete layer around the normal brain and abdominal wall vessels. In ISH-stained sections, all of the endothelial cells were of murine origin (Figure 3G). Because of the small size of the grafts, the number of endothelial cells that could be counted in each tumor section was always less than 100. Therefore, true MIB-1 indices could not be established. However, whatever the tumor and the site of grafting, the percentage of MIB-1 endothelium-positive cells within the tumor microvascular beds was always less than 1%.
In the three cases studied for disruption of the blood-brain barrier by analysis of the intracerebral distribution of Evans blue, dye extravasation into the tumor masses was marked but no dye leakage was observed elsewhere in the brain. In sections of each of these three tumors, fibrin and murine IgGs, proteins that are physiologically excluded from the brain extracellular space, were detected immunohistochemically within the nonvascular extracellular matrix in territories stained by Evans blue.
Immunohistochemical staining for VEGF protein was performed in sections from the two initial surgical biopsies and from each of the intracranial and abdominal transplants. The staining patterns of the main tumor masses were identical whatever the sections studied. Immunopositivity was, on average, observed in the cytoplasm of 30% of cells that made up the main tumor masses (Figure 3H), with a preferential localization to astrocytes organized in perinecrotic palisades. Cells of tumor vasculature were always negative. Dual immunohistochemical-ISH studies with the antibody to the VEGF protein and the human or the mouse DNA probe performed on selected intracranial transplants made it possible to demonstrate unequivocally that most of the VEGF-synthesizing astrocytes in the main tumor masses were of human origin. However, all the mouse reactive astrocytes at the tumor borders and a few normal-looking mouse astrocytes in subpial or subependymal locations also expressed the protein.
Morphology of the Abdominal Grafts Removed More Than 15 Days After Grafting
Abdominal tumors grew as discrete, lobulated, firm gray or yellowish masses (Figure 3I). Central intratumoral necrosis and petechial hemorrhage were observed in every transplant. Although at low magnification they appeared well circumscribed, the abdominal tumors were not encapsulated. Staining for human vimentin often revealed nests of tumor cells within the parietal striated muscles and the connective tissue layers of the superficial fascia, but not in the dermis or in the fascia deep to the muscle. However, serial sections revealed that the tumor nests that appeared isolated were actually connected to the main tumor masses. Spreading of isolated tumor cells from the main tumor masses into adjacent connective tissue was never observed. The tumors contained the same morphological cell types as those observed in the original surgical specimens and in the brain transplants.
Tumor differentiation was stable with increasing passages. Every tumor cell expressed vimentin, and on average 85% of them expressed GFAP. The median distribution of growth fractions was 40%, with extremes of 13% and 70%. The mean vascular density was 110 (range 75-257). The vessels in the abdominal graft transplants were always of larger diameter than those of the cerebral grafts. On sections immunostained with the antibody to mouse Type IV collagen, the basement membranes of the tumor microvessels appeared continuous and often multilayered. No staining was observed with the antibody to human Type IV collagen. On sections double-stained with antibodies to mouse Type IV collagen and to human vimentin, alignment of tumor cells along the tumor microvessels was never oberved. ISH showed that all endothelial cells were of murine origin. An increase in microvasculature permeability was shown by diffuse and intense bluing of the three tumors studied after IV injection of Evans blue, as well as by heavy and diffuse deposits of fibrin and murine IgG within the tumor extracellular matrix in frozen sections immunostained for these proteins. On average, VEGF immunopositivity was observed in the cytoplasm of 45% of the tumor cells, but cells of the tumor vasculature were always negative.
Morphology of the Grafts Removed Within the First 15 Days After Grafting
Coagulative necrosis was not observed in the three intracranial transplants removed within 15 days of grafting. Non-necrotic-looking tumor cells were dispersed in a spongy neuropil. In sections double-stained for human vimentin and mouse Type IV collagen, neoplastic astrocytes were always observed in close contact with the basement membranes of recipient vessels.
The abdominal transplant removed on Day 3 showed extensive areas of ischemic necrosis of both the tumor cells and the adjoining mouse connective tissue, with tissue congestion and with macrophage and neutrophil emigration. The inflammatory infiltrate was less dense in the two grafts removed on Days 10 and 15. At that time, granulation tissue had formed in the tissue surrounding the necrotic territories. In the center of each of these three grafts, vessels with a human basement membrane were visible on sections immunostained with the species-specific antibody to human Type IV collagen. On ISH-stained sections, most of the endothelial cells of the grafts' microcirculatory beds were human, whereas all the inflammatory cells and those of the granulation tissue were murine. However, in the graft removed on Day 15, some murine endothelial cells could be detected in vessels underlined by a human basement membrane.
Discussion
The desire to experiment on human glioblastomas within a convenient in vivo system eventually led to the use of immunodeficient animals, most commonly in athymic nude mice or rats, which can serve as tolerant hosts for human neoplasms (Rana et al. 1977). The resulting tumors are interspecific transplantation chimeras and therefore constitute experimental systems in which the relationship between the tumor cells and the host tissues can be easily analyzed. Species divergence in repetitive DNA sequences makes it possible to distinguish human GBM cells from normal or reactive cells of the mouse brains by ISH, in a system where it is possible to analyze the cell phenotypes and the extracellular matrix macromolecules by immunohistochemistry (Plénat and Duprez 1990).
Hapten-labeled oligonucleotide probes for the human specific repetitive sequences Alu and their mouse equivalents can be used for such studies and give excellent morphological results when each probe is used individually (De Armond et al. 1994). In our studies, however, single oligonucleotide probes nonradioactively labeled on their 3′ end with terminal deoxynucleotidyl transferase are not sensitive enough for dual hybridization or combined ISH/immunohistochemistry methods to be performed successfully on the same slide. It is now necessary for double ISH methods to be performed because, even in optimally stained sections, the nuclei of a few cells of one species may be unstained by the corresponding probe. We previously reported that hapten-labeled, sonicated total genomic DNA probes make it possible to perform dual or combined studies with excellent morphological results and essentially no cross-hybridization (Plénat and Duprez 1990; Plénat et al. 1992). Similar results should be achieved with pools of terminally labeled oligonucleotides whose sequences are complementary with Alu sequences or with probes generated by the polymerase chain reaction, using mouse or human DNA as a template and oligonucleotides, the sequences of which are complementary to the Alu sequence, as primers. For cell migration studies in transplantation chimeras, these ISH methods should take the place of other marker systems, such as (a) the prelabeling of cells with fluorescent vital dyes or lectins, because such tracers are necessarily progressively washed out due to halving of their amount on each cell division, allowing tumor cell migration studies for only 3-4 weeks (Bernstein et al. 1989; Laws et al. 1993), and (b) the transfer of non-human genetic material into the GBM cells, such as the lacZ gene that codes for a bacterial β-galactosidase for which the enzyme activity can be detected in histological sections (Lampson et al. 1993; Pedersen et al. 1995). Indeed, transfection might select subsets of human brain tumor cells or alter their invasiveness.
Because of their high proliferative rates, transplantable GBM models in which tumor cells are implanted as cell suspensions are often considered as failing to invade the contiguous brain, progressing by expansive rather than diffuse infiltrative/invasive growth (Pilkington et al. 1997). The success of recent gene therapy experiments using the 9L transplantable gliosarcoma may be attributed in part, to the lack of invasion in this model (Culver et al. 1992). However, some models, such as the rat C6 line, demonstrate invasion along vascular pathways and migration along white matter tracts (Nagano et al. 1993). In this work, the xenografted GBMs combined an expansile central mass with diffusely infiltrative margins, a mixed expansive-infiltrative pattern of growth that may be seen in low-grade tumors in humans but is characteristic of high-grade gliomas. The population of astrocytoma cells, isolated or grouped in small cell nests, infiltrates the normal brain along white matter tracts such as the corpus callosum, along blood vessels, beneath the pia, and between the ependymal and subependymal layers of the third and lateral ventricles. Furthermore, as described above, the phenotype of these migrating cells did resemble that of the infiltrating tumor cells in humans. The growth and invasive properties of human GBMs transplanted orthotopically very closely resemble those of GBMs in humans. Therefore, despite the phylogenetic distance between human and mouse, our results suggest that this should not impose any serious limitations on the use of nude mouse brain xenograft models of human GBM for studying brain invasion in vivo.
The combination of immunohistochemistry and ISH made it possible to demonstrate unequivocally the presence of a mouse glial nonvascular stroma within the main tumor mass. Such a stroma in the tumor tissue proper was impossible to discriminate both on HE-stained sections and on microscopic preparations immunostained for GFAP. In the tumor periphery and peritumorally, however, the human neoplastic astrocytes were easily distinguished from the mouse reactive astrocytes in that they lacked the regular arrangement and symmetric stellate processes of reactive cells, which is best visualized by the GFAP reaction. Our results therefore validate the morphologic criteria for identification of acute reactive peritumoral gliosis in humans established by Daumas-Duport et al. (1987). Moreover, the fact that the reactive astrocytes, both in the central core of the tumor and in the tumor periphery, constitute a cycling cell population and take part in the VEGF protein synthesis, although in a small way, is firmly established in this work. On the other hand, because of difficulties in distinguishing stromal reactive astrocytes from tumor cells, such properties are often impossible to ascertain in surgical specimens of human GBMs.
In this study, double immunostaining for human vimentin and MIB-1 antigen made it possible to demonstrate unequivocally that the compartment of isolated human tumor cells that permeate largely intact brain parenchyma constitutes a cycling population and possesses a growth potential similar to that of cells in the cellular cores. Because of the difficulty in discriminating between invading isolated cells and reactive astrocytes, especially in immunostained sections, the proliferative potential of this cell compartment in human GBMs is difficult to appreciate. If this potential is identical to that observed in our model, recurrence in humans may result from proliferation of individual tumor cells that escape the surgeon and infiltrate beyond the standard radiation treatment volume.
In contrast to brain transplants, the flank tumors studied here grew mainly by expansion. Migration of isolated cells away from the main tumor mass was never observed. The same applied to close contacts between expansions of astrocytic foot processes and the Type IV collagen component of the vascular basement membranes. The migrating capacities of the GBM cells are therefore not entirely determined by genetic variations of tumor cells but rather depend on interactions with host factors, including a particular environment specific to the site of engraftment. In the brain xenografts as in GBMs in humans, the basement membranes form a major route of tumor cell migration, a role that they do not play in the abdominal grafts. Additional differences between the brain and abdominal xenografts were the more rapid speed of growth of the abdominal tumors and the higher permeability of their microvascular tree.
Tumors other than GBMs proliferate in nude mice in a pattern similar to that in patients if the cells are grown orthotopically. Orthotopic rodent models have been developed for human gastric carcinoma (Yamashita 1989), colon carcinoma (Morikawa et al. 1988; Fu et al. 1991), renal carcinoma (Naito et al. 1987), lung carcinoma (McLemore et al. 1987; Howard et al. 1991), prostate carcinoma (Kozlowski et al. 1984), pancreas carcinoma (Vezeridis et al. 1989), and malignant melanomas (Juhasz et al. 1993). Studies in these models have demonstrated that the organ-specific microenvironment plays a determining role in tumor growth, stroma formation, angiogenesis, and invasion. Selection of the transplantation site is therefore of importance. When used for experimental therapy, the easily accomplished SC implantation of a tumor is a model that allows the question of cell sensitivity to be separated from the problem of drug delivery. However, it is clear that intracranial growth better simulates the clinical situation.
The mean vascular density in the brain xenografts removed more than 15 days after implantation was always low, even though tumor ischemic necrosis was never extensive. Vascular proliferation in glomeruloid tuff form was never observed. The DNA indices for the microvascular cell population were always estimated to be less than 1%, whereas Watanabe et al. (1995) showed that vessels in human gliomas have a mean MIB-1 index of 2.6%.
Angiogenesis is a complex process, which is regulated by multiple stimulatory and inhibitory factors that are able to modulate the migration and/or proliferation of microvascular cells (Ferrara and Davis-Smyth 1997). Recent data suggest that in human gliomas, VEGF protein is the principal mediator of tumor-induced angiogenesis (Berkman et al. 1993). VEGF binds to endothelial cells that express its receptors. In our model, about 30% of the tumor cells, some reactive astrocytes from the host, but no endothelial cells, were stained by the antibody to VEGF protein. The percentage of tumor cells secreting VEGF was similar to that observed in the original surgical biopsies in which capillary endothelial proliferation with a glomerulus-like structure and endothelial cell staining were observed. These data therefore suggest the absence of a paracrine angiogenesis loop that could be explained by the lack of appropriate receptors for the human VEGF protein on the endothelial lining of the tumor microcirculatory beds, all of which are of murine origin.
In most previously published orthotopic animal models of human GBM (Rana et al. 1977; Bradley et al. 1978; Shapiro et al. 1979; Bullard et al. 1981; Horten et al. 1981; Jones et al. 1981; Basler and Shapiro 1982; Marno et al. 1982; De Armond et al. 1994), the tumors were implanted not in solid pieces but by stereotaxic injection of cell suspensions. The cell suspension methods of grafting have the distinctive advantage of limited surgical trauma. However, it can be estimated from different published figures that direct implantation of cell suspensions results in an overall take rate of less than 70%. Furthermore, injected cells may flow back along the shaft of the needle and therefore into the arachnoidal space, or may be carried away into the normal brain parenchyma or the ventricles by the liquid of the cell suspensions. Recently, GBMs have been also transplanted as spheroids from precultured biopsy specimens, with a success rate of 87.7-100% (Engebraaten et al. 1999). As far as we know, only one group (Shapiro et al. 1979; Horten et al. 1981; Basler and Shapiro 1982) have reported GBM grafting as solid explants into brains of nude mice. The authors who transplanted fragments of human Grade II-IV astrocytomas directly from the patient into mouse brain observed an overall take rate of only 16%, the mean take rate of Grade III and IV astrocytomas being 24%. In the present work, we have shown that human GBM explantation into nude mouse brain is always successful when the tumors are grown within the mouse abdominal walls before being heterotransplanted into the brain.
In our experiments, the abdominal engraftment of human GBMs was always successful. Because the volume of the abdominal tumors after a few weeks is such that several scores of intracerebral transplants can be performed from a single abdominal tumor, a large number of mice can be grafted with tumor tissue from one original surgical specimen, making large homogeneous experimental lots of animals available for well-designed assays. The constant take of the abdominal primary transplants may be explained by the way we carefully seated the tumor fragments in a richly vascularized site, in close contact with the epigastric vessels. Furthermore, we showed that, during the first abdominal passage, the tumor grafts acquire a microvasculature from the host. This may allow fast subsequent graft revascularization by end-to-end anastomosis between the transplants and the host microvasculatory beds which, at the time of grafting, are made of both cells and extracellular matrix molecules of murine origin. An accelerated revascularization may largely explain the shortening of the lag phases with successive abdominal passages, the lag phase brevity between the tumor intracranial transplantation and tumor detectability by MRI, and the constant take of the brain transplants observed in this study. Another factor would be the multistep selection during the abdominal passages of cell populations growing more rapidly than those at earlier passages, with subsequent stabilization.
In addition to its constant take and unquestionable similarities in overall histological structures and patterns of invasion between the intracranial tumor transplants and spontaneous human malignant gliomas, the transplanted brain tumor model described above has the following additional advantages: (a) tumor development is rapid enough for testing of most therapeutic modalities within a reasonable time period, while mean survival times as long as 1 year are observed when GBMs are directly grafted into the brain; and (b) an increased blood-brain barrier permeability to macromolecules allows the issue of drug delivery to malignant brain tumors to be addressed. The superficiality of the sites of tumor implantation, contrasting with the large dimensions of the tumor areas in which the abnormal permeability could be demonstrated, is such that the breakdown of the structural blood-brain barrier may not result from damage to the brain parenchyma and its vascular system caused by the implantation procedures.
MRI evaluation of experimental GBMs in the mouse has not been previously reported. Our findings indicate that this technique is sensitive enough to allow longitudinal evaluation of tumor growth in the same animal. This might spare animals in therapeutic experiments, because the tumor take is assessable before the therapeutic regimen is tested. However, access to instrumentation and the time involved in regularly scanning a large number of animals may be a limiting factor for the use of this approach.
With the exception of their microvascular bed, which differs in several points from that of GBMs in humans, the malignant astrocytomas derived from the orthotopic transplantation of solid fragments of human GBMs previously passaged within the mouse flank constitute a model that closely resembles the human situation, particularly in relation to its growth characteristics and invasive properties. Differences in vascularization must be taken into consideration when this model is used to study therapies in which the tumor microvascular bed is the main target. For the remainding factors, the phylogenetic distance between human and mouse and the recipient immunoincompetence should not impose serious limitations on the use of nude mouse orthotopic models for studying malignant glioma biology or therapy in vivo.
Footnotes
Acknowledgements
Supported by grants from l'Association Française de Recherche contre le Cancer and by La Ligue Régionale de Lutte contre le Cancer.
We thank C. Daumas-Duport for helpful discussion, C. Bonnet, S. Pizzagalli, D. Thiébaut, and M.P. Pretagut for excellent technical assistance, and C. Maire for expert help in the preparation of this manuscript.