Only in Congenial Soil: The Microenvironment in Brain Tumorigenesis

Abstract

Microenvironmental or stromal influences on tumor formation and growth have become an active area of research. The use of mouse models of human cancers to study the role of the microenvironment will yield unique insights into this aspect of tumor biology and should identify novel therapeutic targets for the treatment of human cancers. In the following, the author review the natural history of two pediatric brain tumors, optic pathway glioma in neurofibromatosis type 1 and medulloblastoma in Gorlin's Syndrome, whose patterns of growth suggest that microenvironmental factors are essential for tumor formation. Each of these brain tumors is faithfully modeled in genetically engineered mice and the use of these mouse models to investigate the role of the microenvironment should yield exciting new insights into this important field of study.

INTRODUCTION

In 1889, careful study of breast cancer metastases inspired Stephen Paget to propose that the pattern of secondary tumor growth (metastases) was determined by properties of the normal tissues to which breast cancer cells spread and not by properties of the tumor cells themselves (46). This revolutionary idea was not experimentally confirmed until the work of Hart and Fidler in 1980 (27), which together with groundbreaking studies by Bissell(56) established stromal or microenvironmental contributions to tumor biology as an important field of study. While considerable progress has been made in elucidating pathways that regulate the metastatic spread of cancers, the mechanisms by which stroma participates in the early stages of oncogenesis remain largely unknown. Genetically engineered mouse models of cancer predisposition syndromes provide a unique opportunity to identify microenvironmental factors that can promote or inhibit tumor formation. The relevance of these conditions and associated mouse models to the study of stromal contributions to oncogenesis is illustrated by the natural history of two pediatric brain tumors whose patterns of growth indicate that tumorigenesis depends upon microenvironmental factors. Each of these tumor types, optic pathway glioma (OPG) in neurofibromatosis type 1 (NF1) and medulloblastoma in Gorlin's syndrome, have faithful mouse models that present a singular experimental opportunity to define how microenvironmental factors participate in oncogenesis.

Consider what each soil will bear, and what each refuses (Virgil)

Just as the case of preferential breast cancer metastases to liver suggested to Paget that it was the liver and not the breast cancer cell that determined where secondary tumors grew, the predilection of gliomas to form along the optic pathway in patients with NF1 suggests that it is the optic pathway and not the tumor progenitor that dictates where and when these tumors grow.

NF1 is an autosomal dominant cancer predisposition syndrome and among the most common inherited genetic disorders (15). Neurofibromin is a tumor suppressor that functions as a negative regulator of the Ras oncogene 1, 4, 21, 42, 43, 48, 62. Cells that sustain loss of neurofibromin exhibit increased Ras signaling 5, 7, 10, 11, 13, 20, and as a consequence, individuals with NF1 are susceptible to a number of neoplasms, especially benign and malignant tumors of the peripheral and central nervous systems (57). The most common NF1‐associated tumor of the central nervous system (CNS) is the low‐grade astrocytoma 22, 23, 36, 40. These tumors are most commonly World Health Organization (WHO) grade I, glial fibrillary acidic protein (GFAP)‐positive tumors that exhibit complete loss of neurofibromin function. While it is presumed that loss of the second (functional) NF1 allele happens through a random process, the somewhat restricted occurrence of these tumors to the optic nerve and chiasm of young children indicates that a nonrandom process dictates where and when these tumors form. In fact, these observations suggest that signals derived from the microenvironment of the optic pathway must work in concert with complete loss of neurofibromin to promote OPG formation and growth (61).

Cooperative promotion of tumor formation involving genetic events (Nf1 loss) in tumor progenitors and microenvironmental‐derived signals is also suggested by studies of mouse models of NF1. Zhu and colleagues were the first to demonstrate that complete loss of neurofibromin was necessary but not sufficient for the formation of neurofibroma in a mouse model of NF1 (65). They found that neurofibromas would only form if in addition to complete loss of neurofibromin function in tumor progenitors derived from the Schwann cell lineage, there was heterozygous loss of neurofribomin in the tissue surrounding the evolving tumor (Nf1^flox/−, Krox20^cre mice). Similarly, Bajaneru et al found that while complete loss of neurofibromin in astroglial cells resulted in hyperproliferation of astrocytes (2), optic pathway tumors would only form in mice if neurofibromin was completely deleted within the astroglial lineage and the remainder of the mouse was otherwise heterozygous for neurofibromin loss (Nf1^flox/−, GFAP^cre mice) (Figure 1) (3). Further, when neurofibromin loss was phenocopied by constitutive activation of its downstream mediator K‐Ras, OPGs would only form in Nf1+/− mice (10). Thus, in humans with NF1 and in three different mouse models of NF1 there are compelling observations that identify the microenvironment as an active participant in tumorigenesis, specifying where and when tumors will form.

Figure 1.

Spontaneous glioma in Nf1^flox/−, GFAP^cre mice. Section through the optic chiasm of a 4‐month old Nf1^flox/−, GFAP^cre mouse demonstrating an abnormal focus of hypercellularity (boxed area). The tumor exhibits increased expression of CXCL12 and phosphorylated‐CXCR4 (pCXCR4). The phosphorylation of CXCR4 is dependent upon ligand activation of the receptor and these data suggest that stroma‐derived CXCL12 is active in NF1‐associated glioma biology. The mouse tumors do not exhibit many of the characteristic features of human pilocytic astrocytomas such as Rosenthal fibers, eosinophilic granular bodies or microcystic spaces. GFAP, glial fibrillary acidic protein.

Temporal specificity in the pattern of brain tumor formation is also seen in the incidence of medulloblastoma in Gorlin's syndrome. Gorlin's syndrome, originally described in 1960 as nevoid basal cell carcinoma syndrome involving basal cell carcinoma, jaw cysts and bi‐fid ribs (19), results from inherited mutation in the sonic hedgehog receptor Patched (Ptc) 24, 32. Ptc ordinarily maintains a tonic inhibition of the sonic hedgehog pathway which is relieved upon ligand binding (31). The mutation in Gorlin's syndrome inactivates a single Ptc allele and thereby hyperactivates the sonic hedgehog signaling pathway. Ptc, like neurofibromin, is a tumor suppressor and individuals with Gorlin's syndrome are susceptible to a variety of tumors, among which is medulloblastoma (30).

In contrast to OPG, medulloblastoma is a neural tumor. In sporadic forms it appears that medulloblastoma may arise from one of two progenitor populations in the developing cerebellum; one derived directly from the periventricular zone known as the rhombic lip, and the other from the committed granule neuron lineage 37, 63. Reflective of the role that sonic hedgehog plays in cerebellar granule neuron development, the latter subtype appears to be distinguished by evidence for increased sonic hedgehog pathway activation and a unique histology known as desmoplastic medulloblastoma 41, 49.

Desmoplastic medulloblastoma occurs both in Gorlin's syndrome and as a sporadic form. The peak incidence of Gorlin's syndrome‐associated medulloblastoma occurs during the first 4 years of life when granule precursor cells, stimulated by sonic hedgehog, expand to become the single most abundant neuronal population in the CNS (18). Interestingly, sporadic desmoplastic medulloblastoma more commonly arises during adolescence and adulthood (16). This apparent disparity between when normal sonic hedgehog‐induced granule lineage growth occurs and when the majority of desmoplastic medulloblastomas arise suggests that either an alternate progenitor exists within the older cerebellum, or that an additional factor derived from the microenvironment of the cerebellum might prevent tumor formation early in life or work together with sonic hedgehog pathway activation to drive tumor formation later in life.

The latter hypothesis is supported by the identification of preneoplastic cells from the cerebella of Ptc+/− mice 8, 17, 34, 45. These cells, which lack Ptc expression, are distinct from both normal granule precursor cells and frank medulloblastoma cells. They reside within remnants of the external granule cell layer of greater than half of all Ptc+/− mice at 4–6 weeks of age. However, despite the presence of ectopic foci of Ptc−/− cells, only 10%–20% of Ptc+/− mice will go on to develop medulloblastoma, and only after 3–6 months of age (Figure 2). This latency in tumor formation to well beyond infancy is similar to the natural history of desmoplastic medulloblastoma in humans. The fact that most of the Ptc−/− cellular rests do not undergo complete transformation suggests that loss of Ptc is necessary but not sufficient to drive oncogenesis.

Figure 2.

Spontaneous medulloblastoma in Ptc+/− mice. A large medulloblastoma is seen arising from the surface of a Ptc+/− mouse. Like the gliomas that arise in Nf1^flox/−, GFAP^cre mice, published studies demonstrate that these medulloblastomas exhibit co‐expression of CXCL12 and CXCR4 (59). While the mouse medulloblastomas exhibit clear evidence for increased activation of the sonic hedgehog pathway, they do not posses the nodular pattern of growth that is typical for desmoplastic medulloblastoma. GFAP, glial fibrillary acidic protein.

Several genes distinguish the preneoplastic cells from normal granule precursor neurons, including adhesion molecules that mediate interactions with the microenvironment (45). Among them is nidogen, a receptor for laminin. Laminin is recognized as a component of the extracellular matrix of stem cell niches within the CNS 33, 39 and an enhancer of sonic hedgehog‐induced granule precursor proliferation (50). These findings support the hypothesis that oncogenesis involves a preneoplastic state in which the relationship to the microenvironment is altered to enhance growth. The importance of the microenvironment to the growth of medulloblastoma is also indicated by studies in which the actions of stroma‐derived CXCL12 were specifically disrupted with the drug AMD 3100 (AnorMED Inc., Langley, Canada), resulting in an inhibition of the intracranial growth of Daoy, desmoplastic medulloblastoma xenografts (59).

Both NF1 and Gorlin's syndrome provide examples of autosomal dominant cancer predisposition syndromes where complete loss of function of the involved tumor suppressor increases proliferation within the tumor progenitor lineage. This expansion of the tumor progenitor pool is necessary but not sufficient for tumor formation. In addition, the study of both diseases in human and mouse models suggests that the location and/or timing of tumor formation may be dictated by factors derived from tissue surrounding the tumor progenitors.

The seed never explains the flower (Edith Hamilton)

The molecular bases for the unique pattern of tumor formation in NF1 and Gorlin's syndrome have not yet been defined. In each case, the consequence of loss of tumor suppressor function (neurofibromin or patched) is increased proliferation. In each case progression to tumor formation occurs with acquisition of other features of tumor biology such as enhanced survival signaling. Among the potential roles for stroma‐derived factors in tumorigenesis is the provision of pro‐survival, anti‐differentiation or migratory signals that can promote tumor‐like growth during the evolution of a fully transformed tumor cell (Figure 3).

Figure 3.

Oncogenesis is a cooperative event involving tumor progenitors and their microenvironment. The process of oncogenesis is depicted as a cooperative activity involving tumor progenitors (cell‐like cartoons) and their microenvironment (stroma). During the process of oncogenesis, tumor progenitors change from normal cells to fully transformed cells through the acquisition of functions that support dysregulated proliferation and survival, telomere maintenance, migration and tissue invasion, and angiogenisis. As pictured, this process involves the accumulation of genetic abnormalities within tumor cells that with time support increasing levels of cell‐autonomous growth. During the early stages of tumor development, when sufficient cell‐autonomy does not exist, and to some degree throughout the life of a tumor, the stroma provides factors that support tumor development and progression through the stimulation of these same processes.

Two recent studies by Warrington et al (61) and Daginakatte and Gutmann (9)have identified candidate stroma‐derived factors in the genesis of NF1‐associated OPG that could fulfill this kind of role. Warrington et al suggested that developmental regulation of CXCL12 expression in endothelial cells, neurons and/or microglia could contribute to the pattern of tumorigenesis in NF1 (61). In these studies, it was determined that high levels of CXCL12 are present along the optic pathway in young humans and mice, and that complete loss of neurofibromin function dysregulates signaling through the CXCL12 receptor, CXCR4. The combination of high levels of CXCR4 activation and dysregulation of CXCR4 signaling in Nf1−/− astroglial cells results in abnormally deep and sustained suppression of intracellular cAMP in response to CXCL12 and this promotes a unique and abnormal survival responses in Nf1−/− cells. These studies suggest a mechanism of oncogenesis in which dysregulation of a stromal signal (CXCL12) results in the acquisition of one of the hallmarks of a cancer cell, inhibition of apoptosis.

Similarly, Daginakatte and Gutmann identified microglia‐derived hyaluronidase as potential stimulator of NF1‐associated OPG growth (9). Optic pathway gliomas in patients with NF1 typically contain large numbers of infiltrating microglia, the tissue macrophages of the CNS. Similar to other inflammatory cells, microglia become activated in pathological states and secrete inflammatory mediators (26). Daginakatte and Gutmann determined that Nf1+/− microglia produce excessive quantities of MGEA5, a hyaluronidase that can stimulate tumor growth via activation of the MAP kinase pathway. Importantly, systemic inhibition of microglia activation with minocycline reduced proliferation within OPGs in NF1 mice. These data not only identify a potential stroma‐derived stimulator of OPG growth, but also demonstrate that heterozygous loss of neurofibromin has a significant effect on the quantity of the stromal signal. Interestingly, Warrington et al reported that Nf1+/− microglia express greater quantities of CXCL12 compared with Nf1+/+ microglia (61).

A number of secreted and extracellular matrix components have been reported to regulate normal granule cell development and been implicated in either the genesis or suppression of medulloblastoma (64). In addition to sonic hedgehog, other secreted factors include BDNF 6, 60, FGF (14), IGF 28, 47, CXCL12 35, 55, 66 and PACAP 38, 44. Relevant matrix components include heparan sulfate proteoglycans (58), laminin and vitronectin (50). The incidence of medulloblastoma in Ptc+/− mice was increased from approximately 25% to 66% after disruption of the PACAP gene (38). These findings support the hypothesis that disruption of this stroma‐derived factor has a significant impact on the genesis of medulloblastoma.

To be a successful farmer (neuro‐oncologist) one must first know the nature of the soil (Xenophon, Oeconomicus)

The goal of our own research into the role that stroma plays in oncogenesis, is to fit stromal function into the prevailing model for multistage oncogenesis. Specifically, rather than genetic events accounting for all the necessary steps in malignant transformation (25), stromal activation of proliferative, survival or migratory pathways could substitute, at least transiently for genetic events. These functions of stroma are readily compared with the role that microenvironmental cues play in the regulation of progenitor cell proliferation, survival and migration during normal development. Thus, if the genesis of gliomas and medulloblastomas in the NF1 and Gorlin's syndrome mice was viewed as a developmental event, the role of specific microenvironmental factors in the induction and the development of tumors could be studied in a manner analogous to the methodologies used to study normal development. Given the spatial and temporal restrictions on spontaneous tumor formation in both models, the assay for stromal effects would be changes in these patterns.

As the stroma consists of multiple cell types and stromal effects are likely to be mediated through both secreted factors and cell‐cell as well as cell‐matrix contacts, there remains much groundwork to be done in generating a list of candidate stroma‐derived factors that might regulate brain tumorigenesis. This kind of groundwork could be readily pursued in multiple ways. Among the potentially powerful applications of mouse models to this problem is illustrated by the work of Reilly. She was able to introduce Trp53 and Nf1 mutations in cis into several different strains of mice and demonstrate that levels of Nf1 expression was strain‐dependent as was susceptibility to tumor formation. Moreover, tumor formation appeared to be modified by genes on chromosomes 11 29, 53, 54. These data are reminiscent of observations with patients that indicate NF1 phenotype is determined by modifier genes. Easton et alfound high phenotypic correlation between monozygotic twins with NF1 but little phenotypic correlation between more distant relatives with NF1 (12). Both the human and mouse data indicate that disease is not determined by the specific NF1 mutation but rather by the specific NFI mutation and modifer genes. These modifier genes may be stromal factors themselves or genes that modify response pathways to stroma‐derived factors.

Xenograft systems may also provide valuable insights into stroma. Xenografting into different regions of the brain or into different age mice might reveal regional and age‐dependent differences in tumor take rates. Particularly exciting would be a forward genetic screen such as could be accomplished in zebrafish in which mutagenized animals served as recipients for standard xenografts and genes that affected tumor take rate, tumor growth and invasiveness were identified.

The generation of a list of candidate factors will be critical to efforts to do definitive experiments on stromal function. With candidates in hand, it becomes possible with either global manipulation or lineage‐specific manipulation to alter the stroma in tumor‐prone mice and ask whether this alters the pattern of spontaneous tumor formation. These manipulations might be most readily achieved with lineage, site and age‐specific alterations in gene expression through the use of stereotactic injection of viruses such as lentiviruses or RCAS/tv‐a system 51, 52. The combination of the tumor prone mice with these techniques will be particularly powerful for confirmatory experiments regarding the role that specific factors and specific lineages play in brain tumorigenesis.

In conclusion, there is little doubt that stroma participates in brain tumor formation and growth. In fact, some areas of the brain may even be permissive while others restrictive for the formation of individual tumor types. This kind of relationship between tumor and stroma is most evident in diseases like NF1‐associated glioma or medulloblastoma in Gorlin's syndrome. In these diseases, germline mutation in tumor suppressors predispose patients to the formation of brain tumors. In each case, while tumors form with loss of heterozygosity at the tumor suppressor locus, this is a necessary but not sufficient event for tumor formation. Instead, a nonrandom pattern of tumor formation suggests that a second, non‐tumor‐derived factor is essential for oncogenesis to occur. Identifying these factors and understanding how they contribute to brain tumorigenesis will require novel approaches to animal models.