Neuropathology of genetically engineered mice: consensus report and recommendations from an international forum

Article metrics


The Mouse Models of Cancer Consortium of the NCI sponsored a meeting of neuropathologists and veterinary pathologists in New York City in November of 2000. A rapidly growing number of genetically engineered mice (GEM) predisposed to tumors of the nervous system have led to a concomitant need for neuropathological evaluation and validation of these models. A panel of 13 pathologists reviewed material representing most of the available published and unpublished GEM models of medulloblastoma, primitive neuroectodermal tumor, astrocytoma, oligodendroglioma, mixed glioma, and tumors of the peripheral nerve. The GEM tumors were found to have many similarities and some distinct differences with respect to human disease. After review of the biology and pathology for all models presented, participants were split into groups reflective of clinical expertise in human pathology, tumor biology, neuroimaging, or treatment/intervention. Recommendations were made detailing an extensive and complete neuropathological characterization of animals. Importance was placed on including information on strains, tumor clonality, and examination for genetic mutation or altered gene expression characteristics of the corresponding human malignancy. Specific proposals were made to incorporate GEM models in emerging neuroradiological modalities. Recommendations were also made for preclinical validation of these models in cancer therapeutics, and for incorporation of surrogate markers of tumor burden to facilitate preclinical evaluation of new therapies.


Tumors of the nervous system represent a critical group of malignancies because of their impact on quality of life, because of their overall poor prognosis, and because the outlook for patients with high grade tumors has not changed significantly over the past 20 years. Basic science research holds tremendous potential to impact on the improved treatment of these tumors. A major limitation to progress in this area has been the lack of available animal models that accurately reflect the biology, neuropathology, and clinical behaviors of these tumors. Mouse models for glioma and other tumors of the nervous system have traditionally focused on the use of human tumor-derived cell lines that are xenografted into the brains of immunosuppressed mice. As a group, this type of model fails to recapitulate the genetic heterogeneity and infiltrating neuropathology characteristic of the human tumors from which they were derived. Not surprisingly and perhaps as a result, xenograft models have generally been poorly predictive in preclinical trials of anti-tumor agents.

The identification of genes mutated in brain tumors is progressing at a tremendous rate. A growing number of genetic abnormalities have been characterized in association with the most commonly occurring human brain tumors. As new models emerge, it is critical that tumors arising in these animals be validated, and that their similarity with and distinction from the corresponding human tumor be clearly delineated.

The National Cancer Institute's Mouse Models for Human Cancer Consortium was initiated in 1999 as a collaborative program designed to derive and characterize genetically engineered murine (GEM) models that recapitulate the genesis and progression of human malignancies. As part of this effort, a panel of basic scientists, neuropathologists, neuro-oncologists, radiation therapists, neurosurgeons, and neuroradiologists convened on November 1–3, 2000 to review the basic biology and neuropathology of mouse models for tumors of brain and peripheral nerve, and to compare the neuropathology of these models with the corresponding human tumors. Following the presentation of models and neuropathology, break out groups assembled and gave recommendations in areas of mouse models, neuropathology, therapeutics, neuro-imaging, and tumor biology. The following report includes the considerations and recommendations of this group.



Medulloblastoma are believed to arise from precursors of the cerebellar granule cells, the most abundant neuronal cell type in the brain. Granule cell precursors undergo a dramatic proliferative expansion in the external germinal layer (EGL) of the cerebellum during early postnatal life and eventually differentiate and migrate to assume their position in deeper layers of the cerebellum (Goldowitz and Hamre, 1998; Hatten and Heintz, 1995). In mice, these cells differentiate and migrate to their mature positions in the internal granular layer by the third week of postnatal development. In humans, this process is completed by the ninth postnatal month (Sidman and Rakic, 1973). It is possible that pediatric brain tumors arise from defects in developmental signalling pathways that only operate in sub-populations of precursor cells in the developing nervous system. It remains uncertain whether there is a unique cell of origin specific to a particular brain area, or whether malignancies arise from an undifferentiated progenitor cell common to several areas of the nervous system.

Several laboratories have generated mouse models of medulloblastoma by engineering mutations or misexpressing the murine forms of genes mutated in human medulloblastoma (Figure 1). Mutations in several genes including the human homologue of the Drosophila segment polarity gene, patched (PTCH), have been reported in subsets of hereditary and sporadic medulloblastoma (Pietsch et al., 1997; Raffel et al., 1997; Wolter et al., 1997; Xie et al., 1997). Inactivation of one Ptc allele in mice results in a 14% incidence of medulloblastoma (Goodrich et al., 1997; Wetmore et al., 2000). These tumors express intermediate filament proteins characteristic of both neuronal and glial differentiation (Figure 1a). Single cells grown in vitro express both GFAP and neurofilament, suggesting that tumors arise from a stem cell capable of differentiating along neuronal or glial lineages. Surprisingly, the full-length, normal Ptc allele continues to be expressed in the tumors, which suggests that haploinsufficiency of Ptc is sufficient to promote oncogenesis in the central nervous system (Wetmore et al., 2000). This also implies that a second mutation in another gene is required for oncogenic transformation of these cells. Recent studies have determined that loss of p53 accelerates tumorigenesis in Ptc+/− mice. Greater than 95% of the Ptc+/− p53−/− mice develop medulloblastoma by 12 weeks of age compared to a tumor incidence of 14% medulloblastoma by 10 months of age in Ptc+/− mice (Wetmore et al., 2001).

Figure 1

Histopathology of GEM primitive neuroectodermal tumors. (a) Highly cellular cerebellar lesion in a mouse hetero zygous for Ptc (Wetmore et al., 2000). Tumor shows a ‘small blue cell’ pattern with numerous nuclei in the absence of discernible cytoplasm. (b) Infiltrative process of undifferentiated tumor cells involving cerebrum in a mouse engineered for the FGF1B promoter driving SV40 T antigen (Chiu et al., 2000). (c) Medulloblastoma in a mouse engineered for GFAP-Cre:RbLoxP/LoxP:p53LoxP/LoxP, that shows a highly cellular pattern similar to that in (a) (Marino et al., 2000)

Inactivation of both Rb and p53 in the GFAP-expressing cells of the cerebellum resulted in development of medulloblastoma while mice deficient in either Rb or p53 were not prone to this tumor (Marino et al., 2000). These tumors express genes prevalent during normal nervous system development and appear to arise from the cells of the external granule layer. These mouse brain tumors also express proteins similar to those expressed by human medulloblastoma and resemble the tumors histologically. While TP53 and Rb are only rarely mutated in medulloblastoma, it appears that deficiency in these pathways during development predisposes to tumor formation.

Expression of the SV40 large T antigen FGF1 gene under control of brain specific FGF1 promoter was used to perturb the programming of proliferation and differentiation in a subset of neural cells in regions of the hindbrain (Chiu et al., 2000). The resulting mice developed brain tumors that originated in the pontine gray, just rostral to the fourth ventricle. In mature animals, these tumors lacked neuronal and astrocytic markers typically seen in human medulloblastoma. However, they expressed high levels of proliferation markers PCNA (proliferating cell nuclear antigen) and vimentin. Therefore, these mice may provide a valuable model for studying the transformation of early precursor cells.

Astrocytoma and glioblastoma models

Several mouse modeling systems have generated tumors with histological characteristics of human astrocytic tumors. Unlike oligodendrogliomas that have round and regular nuclei, scant cytoplasm, and lack GFAP positive glial processes, astrocytomas display pleomorphic nuclei, robust eosinophilic cytoplasm, and glial processes that stain positive for GFAP. In human astrocytomas the clinical grade of the tumor is designated by the presence of other histologic findings such as mitotic activity, nuclear atypia, cellular density, presence of microvascular proliferation and regions of necrosis surrounded by dense tumor nuclei, referred to as pseudopalisading. Many mouse astrocytomas presented at this meeting had all of these characteristics (Figure 2). Most of the glioma models also showed sub-arachnoid spread of tumor, a feature found more rarely in human tumors.

Figure 2

Histopathology of GEM astrocytic tumors. (a) Astrocytoma with gemistocytic morphology. Tumor from a mouse engineered to overexpress v-src in GFAP-expressing cells (Weissenberger et al., 1997). Arrows indicate neoplastic astrocytes with abundant eosinophilic cytoplasm (‘gemistocytes’), a recognized subtype seen in human tumors. (b) Tumor from a mouse heterozygous for Nf1 and p53 (Reilly et al., 2000). An astrocytic tumor with less prominent gemistocytes (arrow) and an abnormal mitotic figure (arrowhead). (c) Tumor from a mouse engineered to overexpress ras in GFAP-expressing cells (Ding et al., 2001). A diffusely infiltrating process is present, as astrocytic tumor cells are present in a background of neuropil. (d) GFAP staining in a tumor from a mouse overexpressing PDGFB using a retroviral vector (Uhrbom et al., 1998). In this astrocytic tumor strong positivity for GFAP is seen (red-brown staining). Collections of astrocytic processes converging on vessels can be appreciated (arrows), which is common in human tumors

On a molecular level, mutations in human malignant astrocytic tumors activate signal transduction pathways downstream of tyrosine kinase growth factor receptors, such as Ras and Akt. In addition, malignant astrocytomas frequently mutate genes important for p53 and pRb dependent cell cycle arrest pathways. The most common mutation in this category is homozygous deletion of the INK4a-ARF locus that encodes p16INK4a and p14ARF genes. Mouse models for astrocytic tumors activate many of these signal transduction pathways, usually in combination with mutation(s) that affect the cell cycle arrest pathways. The technologies employed in generating such tumor models range from replication competent and tissue specific retroviral vectors carrying specific oncogenes, to GEM mice which over-express specific transgenes, or have engineered mutations in specific tumor suppressors.

Weissenberger et al. (1997) produced transgenic mice expression v-src from the GFAP promoter (limiting the expression to astrocytes) resulting in the formation of astrocytomas. V-src activates several signal transduction pathways that are activated in human gliomas. These GFAP/v-src gliomas were either low grade or anaplastic, but some cases had the histologic characteristics of glioblastomas.

A second transgenic model for astrocytoma used the GFAP promoter to express oncogenic H-RAS as a transgene (Ding et al., 2001). Different founder lines of these mice expressed various levels of oncogenic Ras. Lines with high levels of expression of H-Ras developed glioblastomas at a young age, and could not be propagated. A line with moderate expression of H-Ras showed a high incidence of low grade and anaplastic astrocytomas by 3 months of age, and could be propagated (Ding et al., 2001).

By combining deletion of Nfl and p53, Reilly et al. (2000) generated gliomas with astrocytic character. Nfl is a RasGAP protein that suppresses Ras activity, such that loss of Nfl results in elevated Ras activity. Mice heterozygous for Nfl developed astrogliosis (Nordlund et al., 1995; Weissenberger et al., 1997; Rizvi et al., 1999; Bajenaru et al., 2001; Zhu et al., 2001), while those with astrocyte specific inactivation of Nfl exhibit increased proliferation of astrocytes (Bajenaru et al., 2002). In contrast, mice heterozygous for both Nfl and p53 developed astrocytic tumors with characteristics of glioblastomas (Reilly et al., 2000).

Retroviral vector gene transfer of PDGF-B to somatic cells has been used to generate astrocytic gliomas (Uhrbom et al., 1998). In these experiments, replication-competent MMLV vector systems resulted in the formation of various CNS tumor morphologies. The most frequent histology seen was high grade gliomas with characteristics of glioblastoma (Uhrbom et al., 1998).

Somatic-cell gene transfer with tissue-specific ALV based RCAS retroviral vectors also show the formation of glioblastomas (Holland et al., 2000). These tumors arose after combined transfer of genes encoding activated Ras and Akt to nestin expressing CNS progenitors. In this system, neither Ras nor Akt alone were sufficient to generate glioblastomas (Holland et al., 2000).

Xiao et al. (2002) expressed a 121-amino acid N-terminal fragment of SV40 T antigen under control of the GFAP promoter. This protein dominantly inactivates pRb and related proteins, p108 and p130, but does not interfere with p53 function. All transgenic mice developed high grade astrocytoma at around 6 months of age. Transgenic mice heterozygous for Pten developed astrocytoma with shortened latency, indicating that Pten mutation contributes to disease in this model.

Oligodendroglioma and oligo-astrocytoma

Oligodendroglioma represents approximately 20% of glial tumors (Fortin et al., 1999). These tumors are unique among glioma because a subset of oligodendroglioma deleted for chromosomes 1p and 19q is relatively sensitive to available chemotherapy (Cairncross et al., 1998). Oligodendroglioma frequently over-express PDGF and EGFR (Di Rocco et al., 1998; Reifenberger et al., 1996; Robinson et al., 2001; Smith et al., 2000), and both of these events have been modeled in GEM mice.

Eric Holland and Bengt Westermark presented data demonstrating oligodendrogliomas arising in mice infected with retroviruses expressing PDGF-B (Figure 3). Westermark injected a retrovirus expressing PDGFB intercerebrally in neonatal mice deleted for ink4a/arf. The frequency of tumor development varied between 40 and 80%. Holland used the RCAS system (described above) to express PDGF-B in CNS precursors, or in astrocyte precursors (Dai et al., 2001). Approximately 60% of mice expressing PDGF-B in CNS precursors developed oligodendrogliomas, whereas 40% of mice expressing PDGF in astrocyte precursors developed oligodendroglioma or mixed oligo-astrocytoma. Tumors arose with shortened latency and higher grade in mice deleted for ink4a/arf.

Figure 3

Histopathology of GEM oligodendroglial tumors. (a) Infiltrating oligodendroglioma. Tumor cells with round nuclei and perinuclear haloes are evident in the upper aspect of the photomicrograph. The lower portion of this picture shows infiltration of tumor cells into gliotic brain tissue. (b) Oligodendroglioma. Using the RCAS system, mice were engineered to overexpress PDGF-B in precursor cells (Dai et al., 2001). Tumor cells with round-to-oval nuclei and perinuclear haloes are easily appreciated

Peripheral nerve sheath tumors

Patients with neurofibromatosis type 1 (NF1) are prone to benign peripheral nerve sheath tumors (PNSTs) known as neurofibromas. These tumors, which are likely derived from Schwann cells, can progress to become malignant PNSTs (MPNSTs). In contrast patients with neurofibromatosis type 2 (NF2) are predisposed to developing multiple Schwann cell tumors (schwannomas). The hallmark of NF2 is the development of bilateral schwannomas of the vestibular nerves. Mutations in the NF1 and NF2 tumor suppressor genes are responsible for these respective syndromes; thus germline heterozygous NF1 and NF2 mutation is responsible for the inheritance of cancer predisposition. Subsequent somatic inactivation of the retained allele results in tumor development.

Nfl+/− and Nf2+/− mice develop a spectrum of cancer types that overlaps with that of human NF patients; however these mice rarely develop peripheral nerve sheath tumors (Jacks et al., 1994; Mcclatchey et al., 1997, 1998; however, see Figure 4a). To test whether a mutation in the wild-type Nfl allele was rate-limiting in the formation of neurofibromas in Nfl+/− mice, Cichowski et al. (1999) generated mosaic animals composed of both wild-type and Nfl−/− cells. Importantly, these mice developed multiple peripheral lesions with pathology consistent with that of PNSTs. Although electron microscopy indicated the presence of multiple Schwann cells in these tumors, they were found to be generally S100-negative, suggesting that they may arise from a Schwann cell (neural crest) precursor.

Figure 4

Histopathology of peripheral nerve sheath tumors. (a) Spindle cell tumor with morphology reminiscent of neurofibroma (spindle cells in a loose background) in a mouse heterozygous for Nf1 and Nf2 (cis) (Mcclatchey et al., 1998). (b) Higher power of another spindle cell in a conditional Nf2 mutant mouse (P0Cre:Nf2loxP/LoxP) (Giovannini et al., 2000). The tumor cells show more cytoplasm than in (a) but still retain a delicate ‘loose’ character. (c) This peripheral nerve sheath tumor is derived from an Nf2 heterozygote, and shows loss of the retained Nf2 allele (Mcclatchey et al., 1998). This tumor shows increased cellularity compared to (a) and (b). Tumor cells have an increased nuclear/cytoplasmic ratio characteristic of malignant peripheral nerve sheath tumors

A similar strategy for generating mosaic animals composed of both wild-type and Nf2−/− cells was hampered by the requirement for Nf2 function at multiple stages of embryonic development (McClatchey, unpublished observations). Thus Giovannini et al. (2000) employed conditional targeting to specifically ablate Nf2 function in Schwann cells. These mice developed Schwann cell hyperplasia and occasional frank schwannomas (Figure 4b) that expressed Schwann cell markers including the S100 antigen. Thus loss of the wild-type Nf1 or Nf2 allele is rate-limiting for PNST and schwannoma developed, respectively.

Human MPNSTs frequently contain TP53 mutations, suggesting that TP53 mutation contributes to malignant conversion of PNSTs. Interestingly, in the mouse, the Nf1, Nf2 and p53 loci are all linked on the long arm of chromosome 11. Cooperativity between Nf1, Nf2 and p53 mutations was investigated by generating mice carrying compound Nf1 : p53, Nf2 : p53 and Nf1 : Nf2 heterozygous mutation both in a linked (cis) and unlinked (trans) configuration (Cichowski et al., 1999; Mcclatchey et al., 1998; Vogel et al., 1999). In all cases, the cis (linked) configuration significantly reduced tumor latency. Tumors in cis animals always lost both wild-type alleles, suggesting loss of the entire wild-type chromosome. Thus, the linkage of these three genes in murine but not in human genomes may contribute to the significant differences in tumor predisposition seen in the two species.

Importantly, these models also develop distinct tumor-types. For example, Nf2+/−; p53+/− cis mice develop multifocal osteosarcomas while many Nf1+/−; p53+/− cis mice develop MPNST (Cichowski et al., 1999; Vogel et al., 1999). Many of these MPNSTs express various neural crest markers, including S100 and exhibit histological features of MPNSTs. Interestingly, Nf1+/−; Nf2+/− cis mice also develop MPNST of a different histological type (McClatchey, unpublished, Figure 4); these tumors have a higher collagen content and have lower mitotic indices.

Important directions for the modeling and study of peripheral nerve sheath tumors in mice include the development of additional markers for validation of these models and an understanding of the basis for histopathological differences between them. Clearly an increased understanding of the molecular basis of tumor development in these models is warranted; of particular interest is the molecular basis for cooperativity between Nf1, Nf2 and p53 mutation in Schwann cell tumorigenesis. As the Schwann cell appears to be the cell type of origin of both schwannomas and neurofibromas, the basis for the distinct anatomical location of these tumors in these two disorders must be addressed. Furthermore, an important question remaining for both schwannomas and neurofibroma tumorigenesis is whether the tumor arises from a mature, myelinating Schwann cell, a progenitor or a resting non-myelinating Schwann cell. Finally, as for human NF patients, imaging of tumor development and therapeutic response in these models must be developed.

Choroid plexus tumors

The choroid plexus (CP), located in the ventricles of the mammalian brain, produces most of the CSF and maintains its composition through active transport. The protruding frond-like structures comprising the CP consist of a vascularized core surrounded by a single layer of highly specialized ciliated epithelium. The CP epithelial cells derived from neuroectodermal precursors and normally stop dividing within a few weeks after birth. CP tumors occur most frequently in children where they represent 2–3% of brain tumors. CP papillomas are characterized by features that recapitulate normal CP (grade I, WHO), whereas CP carcinomas show disappearance of regular papillary architecture, prominent mitoses, variable nuclear size and local invasiveness (grades III and IV, WHO). Atypical choroid plexus papillomas show limited malignant features, and specific histologic characteristics are not predictive of biological behavior. This problem is especially critical because benign papillomas may be curable with surgery, whereas carcinomas are associated with poor prognosis. Little is known about genetic mutations in these tumors. A significant fraction of CP tumors show SV40-like sequences by polymerase chain reaction analysis or immunohistochemistry for SV40 T antigen (Garcea, 2001). Further studies are warranted to determine whether viral infection represents a possible etiology of CP tumors.

Preclinical mouse models of CP tumors

SV11 transgenic mice carry the SV40 large T antigen gene under its own regulatory signals. SV40 T antigen binds to and inactivates the Rb family of proteins, pRb, p107 and p130 as well as p53. SV11 mice develop tumors with histological features of carcinomas with 100% penetrance (van Dyke et al., 1987).

TgT121 mice express an amino terminal fragment of SV40 T antigen T121, which inactivates the Rb proteins, but not p53, and is expressed widely in CP epithelium under control of the lymphotropic papovavirus. T121 induces rapid CP proliferation accompanied by p53-dependent apoptosis (Saenz Robles et al., 1994). Forty per cent of mice develop tumors that are histologically similar to human CP carcinomas, and have inactivated p53 (Symonds et al., 1994). Sixty per cent of mice develop slower growing tumors with poorly defined borders and retain active p53. The average survival time for TgT121 mice is 26 weeks, but the range of survival is broad. TgT121 p53+/− mice develop focally aggressive tumors with low apoptotic rates, leading to terminal illness by 12 weeks of age (Symonds et al., 1994). Tumors progress with the same morphology in 100% of the mice (Figure 5), and the transition to aggressive tumor growth coincides with spontaneous somatic inactivation of p53.

Figure 5

Histopathology of GEM choroid plexus tumors. (a) Low power shows tumor from a T121 p53+/− animal (Symonds et al., 1994) that is well-demarcated relative to brain parenchyma (arrows), a feature in choroid plexus tumors which distinguishes this tumor from the infiltrating gliomas. (b) Higher power shows papillary pattern of tumor cells, characteristic of choroid plexus papilloma

Discussion and recommendations

Neuropathology and nomenclature

Neuropathological examination remains a gold-standard for verification of genetically engineered murine brain tumors. The neuropathology working group recommended consultation with a neuropathologist or veterinary pathologist experienced in genetically engineered murine brain tumors for all investigators generating such tumors. For those investigators without experienced neuropathologists at their institutions, MMHCC-affiliated neuropathologists or veterinary pathologists could provide consultations. The MMHCC pathologists suggested that they meet annually to review the histopathology of new models, and reappraise issues relating to nomenclature and histological verification. Interactions among MMHCC pathologists and other interested pathologists would be augmented by a website forum for posting images of tumors.

The neuropathology working group recommended that nomenclature for these tumors should follow the 2000 World Health Organization (WHO) scheme for classification and grading of nervous system tumors (Kleihues et al., 2002). The tumor name, however, should be preceded by the qualifying phrase ‘genetically engineered murine’ (‘GEM’), as in ‘GEM glioblastoma’. Use of this qualifying phrase emphasizes that such tumors are not necessarily equivalent to their human counterparts. Nonetheless, the working group stressed that use of the 2000 WHO terms to describe murine tumors should only occur after extensive and complete characterization of the experimental tumors.

Complete characterization of GEM tumor models involves gross, light microscopic, immunohistochemical and ultrastructural examination of tumors, brain and spinal cord. Documentation of how tissue was processed is also essential. Importantly, gross and standard light microscopic examinations should be performed on all tumors, and should include evaluations of surrounding tissues and relevant normal tissues as well. The specific immunohistochemical stains necessary for complete evaluation depend on the tumor model. For primary neuroectodermal tumors, such as GEM gliomas and GEM medulloblastomas, immunohistochemistry must evaluate glial and neuronal differentiation with markers such as GFAP, S-100 protein, synaptophysin and NeuN. Other recommended markers for central nervous system tumors include nestin, neurofilament protein and vimentin. For peripheral nervous system tumors, immunohistochemistry should include evaluations for S-100 protein, neurofilament protein, and the 75 kilodalton low affinity nerve growth factor receptor LNGFR. Additional markers for peripheral nervous system tumors include collagen type IV, epithelial membrane antigen (EMA), and laminin. Nonetheless, because of concerns about differences in immunohistochemical reactivity between human and mouse tissues, and concerns about conclusions drawn from ‘negative’ immunohistochemical results, electron microscopy is also of importance to complete characterization of GEM tumors. Ultrastructural examination provides a means for detailed and objective assessment of cellular differentiation, including in poorly differentiated lesions. For these reasons, electron microscopy is strongly recommended for representative lesions from all GEM brain tumor models. These recommendations are outlined in Table 1.

Table 1 Complete characterization of GEM tumor models

Tumor biology

Malignant astrocytomas such as glioblastoma remain the best studied and best characterized tumors of the human central nervous system. As a result, there is a significant need for increased research aimed at identifying specific gene alterations in other major CNS tumor subtypes, especially oligodendrogliomas, medulloblastoma and ependymoma. In addition, the study of low grade lesions for all CNS tumor subtypes should be emphasized in order to facilitate the identification of early/initiating genetic, as well as epigenetic (i.e. methlyation), alterations.

A second area of concern regarding the current status of our understanding of CNS tumor biology involves the lack of knowledge regarding the cell of origin for most types of cancer. Research directed to this purpose is well-positioned to benefit from current efforts to develop expression fingerprints for several of the major CNS tumor subtypes (Pomeroy et al., 2002). Comparison of tumor expression profiles against expression profiles for various cell populations in developing as well as adult CNS could be used to devise testable ‘cell of origin’ hypotheses, and would additionally facilitate the identification of cell-type specific promoters, thereby encouraging the refinement of mouse brain tumor models.

As GEM models for cancer are developed, it will be of benefit for investigators to implement standardized analyses and reporting of model tumors in order to facilitate comparisons as well as the ability of other groups to reproduce original results. Examples of information that should be collected for new mouse models includes:

  • Detailed information on mouse strains – in addition to allowing others to address reproducibility, this information will help with the identification of strain-specific effects.

  • Tumor clonality: i.e., does the genetic modifier (transgene, knockout gene, or transduced gene) produce tumors from a single cell clone, or does the tumor result from the expansion of multiple, independent clones?

  • Examination of tissues for histology, for genetic alterations and for gene expression patterns that have been observed in the corresponding human tumor.

Efforts at model development should be based primarily, if not exclusively, on existing information about the genetic basis of the corresponding human tumor. The study of mouse brain tumors that develop from alterations having no corresponding change in a histopathologically equivalent human tumor may be valuable for uncovering novel signalling pathways of potential importance in the corresponding human neoplasm. However, in the absence of data showing similarities in signalling between human and murine tumors, such models may have limited value for developing effective therapies.

Neuroimaging and neuroradiology

A variety of non-invasive imaging techniques (primarily computed tomography (CT), magnetic resonance imaging (MRI) and spectroscopy (MRS), positron emission tomography (PET) and optical imaging) are available to image malignancies of the central nervous system in mice. Each of these can be used for the serial investigation of a single animal. The power offered by serial longitudinal imaging allows the development of preclinical therapeutic paradigms that measure response to therapy rather than survival, allowing substantially smaller cohorts of animals and shorter duration studies than typically required using survival paradigms. Using 3-dimensional imaging, high spatial resolution and intravenous contrast, quantitative estimation of tumor burden and volume can be made, and responsiveness to therapy subsequently measured.

What will neuroimaging be able to offer?

Emerging techniques in magnetic resonance and in PET promise physiological and ultimately biochemical sensitivity to be integrated into the non-invasive longitudinal characterizations of tissue state. In the shorter term, imaging techniques, such as diffusion weighted imaging (with its hypothesized sensitivity to tumor cellularity), fractional blood volume estimation and characterization of tumor microvascular permeability (properties of the vascular component of the tumor), as well as spectroscopic assessment of the relative concentrations of major metabolites (typically, N-acetyl-aspartate, creatine and choline-containing compounds) are already available and await testing in GEM tumor models. In the future, antibody-conjugated, targeted contrast agents, and contrast agents activated by specific binding may offer non-invasive assays of gene therapy delivery, expression, and efficacy.

Diffusion-weighted imaging

Diffusion-weighted MRI (DWI) exploits the random (Brownian) motion of water molecules, and the hypothesized differences in hindrances to such diffusion in the intracellular and extracellular spaces. DWI yields a signal intensity quantitatively related to the freedom of diffusion, and thus leads to inferences of the relative proportions of the intra- and extra-cellular spaces. DWI has shown value in the detection of cytotoxic edema associated with acute cerebral ischemia, and may be valuable in the characterization of tumor cellularity (and in particular cell-packing density).

Perfusion and flow: fractional blood volume and microvascular permeability

Although MRI offers a wide variety of sensitivities to intrinsic tissue contrasts, it can also be used with exogenous contrast media, in a fashion analogous to nuclear medicine techniques such as positron emission tomography (PET). In MRI, the contrast media (or tracers) used are not radioactive. Nevertheless, they do affect local signal intensity to a degree related quantitatively to contrast agent concentration. Applying macromolecular contrast to dynamic images acquired during bolus administration of such contrast media allows assessment of relative cerebral blood volume (rCBV), or vascularity, and even cerebral blood flow (rCBF). Extending dynamic imaging over a longer period allows assessment of contrast agent extravasation into tissue, quantitatively described by the microvascular permeability.

In order to fully test and develop the specific tools discussed, validation studies will be of crucial importance. Once the interpretation of these advanced techniques is established, their utility can be harnessed for development and understanding of new models as well as for treatment evaluation. With this in mind, the development of mouse models of human brain cancer can in fact proceed hand-in-hand with the expanding implementation of non-invasive imaging techniques.

Treatment and intervention

GEM mouse models hold great promise for preclinical testing of anti-tumor therapy, but the utility of these models, and whether they are more predicitive than xenograft models remains to be established. It is crucial that these models be tested in preclinical trials, using both empirically designed therapies, as well as therapies targeted to the initiating transgene or tumor suppressor that is altered in the respective models. It will be important, when possible, to validate each model based on its response to therapies known to be useful or known to have no effect in the corresponding human malignancy. In contrast to xenograft models, a major hurdle for many of these models is the ability to identify animals with tumors easily and inexpensively, and to follow tumor burden, and surrogate markers of therapeutic response, in the preclinical evaluation of new therapies.

Materials and methods

A set of 85 slides reflecting 19 xenograft models, and over 20 genetically engineered mice from 12 different laboratories were developed from tissue contributed by investigators from the USA and from Europe. The slide set was developed, distributed and studied in advance of the meeting. A conference call review recommended special stains for many models, and these were available at the meeting. All of the biologists who created the GEM models attended the meeting to present and discuss the biology of individual models. Following the meeting, individuals representing specific disciplines were asked to contribute summary statements for this document, and for the related MMHCC web site (URL: which provides images and details that supplement this manuscript.


  1. Bajenaru ML, Donahoe J, Corral T, Reilly KM, Brophy S, Pellicer A, Gutmann DH . 2001 Glia 33: 314–323

  2. Bajenaru ML, Zhu Y, Hedrick NM, Donahoe J, Parada LF, Gutmann DH . 2002 Mol. Cell. Biol. 22: 5100–5113

  3. Cairncross JG, Ueki K, Zlatescu MC, Lisle DK, Finkelstein DM, Hammond RR, Silver JS, Stark PC, Macdonald DR, Ino Y, Ramsay DA, Louis DN . 1998 J. Natl. Cancer Int. 90: 1473–1479

  4. Chiu IM, Touhalisky K, Liu Y, Yates A, Frostholm A . 2000 Oncogene 19: 6229–6239

  5. Cichowski K, Shih TS, Schmitt E, Santiago S, Reilly K, McLaughlin ME, Bronson RT, Jacks T . 1999 Science 286: 2162–2167

  6. Dai C, Celestino JC, Okada Y, Louis DN, Fuller GN, Holland EC . 2001 Genes Dev. 15: 1913–1925

  7. Di Rocci F, Carroll RS, Zhang J, Black PM . 1998 Neurosurgery 42: 341–346

  8. Ding H, Roncari L, Shannon P, Wu X, Lau N, Karaskova J, Gutmann DH, Squire JA, Nagy A, Guha A . 2001 Cancer Res. 61: 3826–3836

  9. Fortin D, Cairncross GJ, Hammond RR . 1999 Neurosurgery 45: 1279–1291 discussion 191

  10. Garcea RL . 2001 Dis. Markers 17: 149–151

  11. Giovannini M, Robanus-Maandag E, van der Valk M, Niwa-Kawakita M, Abramowski V, Goutebroze L, Woodruff JM, Berns A, Thomas G . 2000 Genes Dev. 14: 1617–1630

  12. Goldowitz D, Hamre K . 1998 Trends Neurosci. 21: 375–382

  13. Goodrich LV, Milenkovic L, Higgins KM, Scott MP . 1997 Science 277: 1109–1113

  14. Hatten ME, Heintz N . 1995 Annu. Rev. Neurosci. 18: 385–408

  15. Holland EC, Celestino J, Dai C, Schaefer L, Sawaya RE, Fuller GN . 2000 Nat. Genet. 25: 55–57

  16. Jacks T, Shih TS, Schmitt EM, Bronson RT, Bernards A, Weinberg RA . 1994 Nat. Genet. 7: 353–361

  17. Kleihues P, Louis DN, Scheithauer BW, Rorke LB, Reifenberger G, Burger PC, Cavenee WK . 2002 J. Neuropathol. Exp. Neurol. 61: 215–225 discussion 226–229

  18. Marino S, Vooijs M, van der Gulden H, Jonkers J, Berns A . 2000 Genes Dev. 14: 994–1004

  19. McClatchey AI, Saotome I, Ramesh V, Gusella JF, Jacks T . 1997 Genes Dev. 11: 1253–1265

  20. McClatchey AI, Saotome I, Mercer K, Crowley D, Gusella JF, Bronson RT, Jacks T . 1998 Genes Dev. 12: 1121–1133

  21. Nordlund ML, Rizvi TA, Brannan CI, Ratner N . 1995 J. Neuropathol. Exp. Neurol. 54: 588–600

  22. Pietsch T, Waha A, Koch A, Kraus J, Albrecht S, Tonn J, Sorensen N, Berthold F, Henk B, Schmandt N, Wolf HK, von Deimling A, Wainwright B, Chenevix-Trench G, Wiestler OD, Wicking C . 1997 Cancer Res. 57: 2085–2088

  23. Pomeroy SL, Tamayo P, Gaasenbeek M, Sturla LM, Angelo M, McLaughlin ME, Kim JY, Goumnerova LC, Black PM, Lau C, Allen JC, Zagzag D, Olson JM, Curran T, Wetmore C, Biegel JA, Poggio T, Mukherjee S, Rifkin R, Califano A, Stolovitzky G, Louis DN, Mesirov JP, Lander ES, Golub TR . 2002 Nature 415: 436–442

  24. Raffel C, Jenkins RB, Frederick L, Hebrink D, Alderete B, Fults DW, James CD . 1997 Cancer Res. 57: 842–845

  25. Reifenberger J, Reifenberger G, Ichimura K, Schmidt EE, Wechsler W, Collins VP . 1996 Am. J. Pathol. 149: 29–35

  26. Reilly KM, Loisel DA, Bronson RT, McLaughlin ME, Jacks T . 2000 Nat. Genet. 26: 109–113

  27. Rizvi TA, Akunuru S, de Courten-Myers G, Switzer III RC, Nordlund ML, Ratner N . 1999 Brain Res. 816: 111–123

  28. Robinson S, Cohen M, Prayson R, Ransohoff RM, Tabrizi N, Miller RH . 2001 Neurosurgery 48: 864–873 discussion 873–874

  29. Saenz Robles MT, Symonds H, Chen J, Van Dyke T . 1994 Mol. Cell. Biol. 14: 2686–2698

  30. Sidman RL, Rakic P . 1973 Brain Res. 62: 1–35

  31. Smith JS, Wang XY, Qian J, Hosek SM, Scheithauer BW, Jenkins RB, James CD . 2000 J. Neuropathol. Exp. Neurol. 59: 495–503

  32. Symonds H, Krall L, Remington L, Saenz-Robles M, Lowe S, Jacks T, Van Dyke T . 1994 Cell 78: 703–711

  33. Uhrbom L, Hesselager G, Nister M, Westermark B . 1998 Cancer Res. 58: 5275–5279

  34. Van Dyke TA, Finlay C, Miller D, Marks J, Lozano G, Levine AJ . 1987 J. Virol. 61: 2029–2032

  35. Vogel KS, Klesse LJ, Velasco-Miguel S, Meyers K, Rushing EJ, Parada LF . 1999 Science 286: 2176–2179

  36. Weissenberger J, Steinbach JP, Malin G, Spada S, Rulicke T, Aguzzi A . 1997 Oncogene 14: 2005–2013

  37. Wetmore C, Eberhart DE, Curran T . 2000 Cancer Res. 60: 2239–2246

  38. Wetmore C, Eberhart DE, Curran T . 2001 Cancer Res. 61: 513–516

  39. Wolter M, Reifenberger J, Sommer C, Ruzicka T, Reifenberger G . 1997 Cancer Res. 57: 2581–2585

  40. Xiao A, Wu H, Pandolfi PP, Louis DN, Van Dyke T . 2002 Cancer Cell 1: 157–168

  41. Xie J, Johnson RL, Zhang X, Bare JW, Waldman FM, Cogen PH, Menon AG, Warren RS, Chen LC, Scott MP, Epstein Jr EH . 1997 Cancer Res. 57: 2369–2372

  42. Zhu Y, Romero MI, Ghosh P, Ye Z, Charnay P, Rushing EJ, Marth JD, Parada LF . 2001 Genes Dev. 15: 859–876

Download references


WA Weiss and M Israel were the meeting organizers. WA Weiss edited the manuscript. K Aldape organized the pathology workshop, prepared and distributed slide sets, and prepared figures for the manuscript. Charles Cobbs, Eric Holland, C David James, David N Louis, Cheryl Marks, Andrea I McClatchey, Tim Roberts, Terry Van Dyke and Cynthia Wetmore contributed equally to writing subsections of this manuscript and are listed alphabetically. Ing-Ming Chiu, Marco Giovannini, Abhijit Guha, Robert J Higgins, Silvia Marino, Ivan Radovanovic and Karlyne Reilly contributed pathology specimens and are listed alphabetically following those who contributed to writing this manuscript. This meeting was sponsored by the Mouse Models of Human Cancer Consortium of the NCI. We would like to thank Susan Seweryniak, Angela Sammarco and Tricia Wallich for assistance in organizing this meeting, Brian Slater for assistance in digitizing images, and David Gutmann for critical review of the manuscript. We acknowledge our additional pathology panel members for also reviewing the slide set: Roderick Bronson, Peter C Burger, Robert Cardiff, V Peter Collins, Charles Eberhardt, Gregory N Fuller, Margaret McLaughlin, Bernd Scheithauer, Patrick Shannon and James Woodruff. We are grateful to the following investigators for providing information and materials for the slide sets: Adriano Aguzzi, Anton Berns, Henry Friedman, Yancy Gillespie, Tyler Jacks, Matthew Scott, Xiao-Yang Wang and Bengt Westermark. We also acknowledge these additional investigators for organizing specific sessions: Peter Burger, Howard Fine, Greg Fuller, Yancy Gillespie, Bob Jenkins, Andras Nagy, Michael Prados, James Provenzale, Brian Ross, Ed Shaw, James Woodruff and Al Yung.

Author information

Correspondence to William A Weiss.

Rights and permissions

Reprints and Permissions

About this article


  • genetically engineered mice
  • mouse brain tumor models
  • mouse nerve sheath tumor models
  • neuropathological evaluation

Further reading