Individuals with neurofibromatosis type 1 (NF1) are prone to developoptic pathway gliomas that can result in significant visual impairment. To explore the cellular basis for the reduced visual functionresulting from optic glioma formation, we used a genetically engineered mouse model of Nf1 optic glioma (Nf1+/−GFAPCKO mice). We performed multimodal functional and structural analyses both before and after the appearance of macroscopic tumors. At 6weeks of age, before obvious glioma formation, Nf1+/−GFAPCKO mice had decreased visual-evoked potential amplitudes and increased optic nerve axon calibers. By 3 months of age, Nf1+/−GFAPCKO mice exhibited pronounced optic nerve axonopathy and apoptosis ofneurons in the retinal ganglion cell layer. Magnetic resonance diffusion tensor imaging showed a progressive increase in radial diffusivity between 6 weeks and 6 months of age in the optic nerve proximal to the tumor indicating ongoing deterioration of axons. These data suggest that optic glioma formation results in early axonaldisorganization and damage, which culminates in retinal ganglion cell death. Collectively, this study shows that Nf1+/−GFAPCKO mice can provide a useful model for defining mechanisms of visual abnormalities in children with NF1 and lay the foundations for future interventional studies aimed at reducing visual loss.
Individuals with the inherited cancer predisposition syndrome, neurofibromatosis type 1 (NF1), develop tumors that involve both the central and peripheral nervous systems (1, 2). The most common central nervous system tumor is a glial neoplasm that arises along the optic pathway. Optic pathway gliomas (OPGs) are low-grade glial fibrillary acidic protein (GFAP)-immunoreactive tumors (3) that typically affect the prechiasmatic optic nerves and chiasm (4-6). Neurofibromatosis type 1-associated OPGs usually arise in children within the first decade of life, most often in preschool age children (7). Because of their location along the optic pathway, the typical presenting sign is visual impairment, and 25% to 40% of children with these tumors have decreased visual acuity at the time of initial OPG diagnosis (8).
Several barriers limit our ability to improve the clinical outcome for children with NF1-associated OPG. First, accurate methods for assessing visual function in very young children are required. Visual screening in the greatest at-risk population of children with NF1 (infants and toddlers) is often challenging using standard methods used for school-aged children (6). This has prompted clinicians to evaluate other modalities including visual-evoked potentials (VEPs) for identifying children with NF1-associated OPG visual deficits (9, 10). Moreover, prognostic markers of tumor growth are lacking. In this regard, the radiographic appearance of an OPG does not predict its individual clinical behavior, and there is no correlation between tumor size or contrast enhancement and visual function (11, 12). To provide such predictive information, diffusion-based magnetic resonance imaging (MRI) has been suggested as a potential method for assessing high-grade glioma growth (13-15) but has not been examined in low-grade brain tumors. Despite the ability to arrest tumor growth using chemotherapy in 60% to 80% of children with NF1-associated OPG, few patients exhibit improved visual acuity (16). The lack of visual improvement raises the possibility that the neuronal damage secondary to optic glioma formation is not reversible.
To gain insight into the cellular pathogenesis of NF1-associated OPG, we used a strain of genetically engineered mice (GEM) in which all cells are Nf1 haploinsufficient (Nf1+/−), except cells of glial origin, where neurofibromin expression was completely ablated by the expression of Cre recombinase (Nf1+/−GFAPCKO mice) (17). Nearly 100% of these GEM develop astrocytic neoplasms with low proliferative indices involving the prechiasmatic optic nerves and chiasm. Herein, we use this unique mouse model to assess optic pathway pathology using multiple modalities, including visual physiology, diffusion-based MRI, and transmission electron microscopy. We found that Nf1 mutant mice with optic gliomas have reduced VEP amplitudes and increased radial diffusivity (RD) in the optic nerve, enlarged optic nerve axons, and retinal ganglion cell (RGC) loss. These observations suggest that the Nf1 optic glioma GEM model may be an excellent experimental platform to understand the critical relationship between glioma formation and neuronal dysfunction relevant to visual loss in children with NF1-associated OPG.
Materials and Methods
Nf1+/−GFAPCKO mice (17) were generated by successive interbreeding of Nf1+/−, Nf1flox/flox (18) and GFAP-Cre mice (19). Age-matched C57BL/6, Nf1flox/flox, or Nf1flox/wt animals were analyzed as wild-type (WT) controls. Glial fibrillary acidic protein-Cre animals were also mated with rosa26R-enhanced yellow fluorescent protein (R26R-EYFP) mice (20) to detect Cre-mediated recombination in vivo. All mice were maintained on a C57BL/6 background.
VEP and Electroretinogram Measurements
Visual-evoked potentials (21) and full-field electroretinograms (ERGs) (22) were recorded on a UTAS-E 3000 Visual Electrodiagnostic System (LKC Technologies, Gaithersburg, MD). For all electrophysiologic measurements, mice were anesthetized by intraperitoneal injection of ketamine (80 mg/kg) and xylazine (15 mg/kg). Body temperature was maintained between 36°C and 37°C throughout the recordings with a heating pad and monitored using a rectal temperature probe. Corneal anesthesia was achieved with 1% proparicaine, and the pupils were dilated for recordings with 1% atropine. Corneas were kept moist with application of 1% carboxymethyl-cellulose and the placement of clear contact lenses (Metro Optics, Austin, TX), which stayed in place throughout the procedure.
For VEP measurements, stainless steel needle electrodes were placed as follows: recording electrode on the scalp over the visual cortex, reference electrode in the skin of the left ear, and ground electrode in the base of the tail. Brief white flashes at 0.2 log cd-seconds/m2 were delivered via a Ganzfeld sphere on a dark background. For each trial, 80 consecutive flashes were averaged at 1.9 Hz.
For ERG recordings, mice were dark adapted overnight and prepared for recordings under infrared illumination. A 2-mm-diameter stainless steel loop positioned gently on the cornea by a micromanipulator served as the recording electrode. The reference needle electrode was placed under the skin between the eyes, and the ground needle electrode was placed at the base of the tail. Stimuli were brief white flashes delivered via a Ganzfeld-integrating sphere, and signals were recorded with band-pass settings of 0.3 to 500 Hz. After a 10-minute stabilization period, an 11-step scotopic intensity series (−3.60 to 0.875 log cd-seconds/m2 stimulus) was recorded, which included rod-specific and scotopic bright flash responses. After a 10-minute light adaptation period on a steady white background (2.30 log cd/m2), a 5-step photopic intensity series was recorded (0.0-2.82 log cd-seconds/m2 stimulus). Scotopic and photopic B wave amplitudes and scotopic A wave amplitudes were recorded for all flash intensities.
Morphometry of Optic Nerve Axons
Mice were perfused with Ringer solution followed by 1.5% paraformaldehyde and 2.5% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.4). Both retrobulbar optic nerve segments from each mouse were postfixed in 1% buffered OsO4, dehydrated through an ethanol/acetone series, and embedded in Durcupan (Electron Microscopy Sciences, Hatfield, PA) for light and electron microscopy. Transverse semithin sections (0.5 μm) obtained at a 1-mm distance behind the eyeballs were stained with paraphenylenediamine (1% in isopropanol/methanol; 30 minutes) to optimize contrast of the myelinated fibers for automated axon counts. Mosaic images of paraphenylenediamine-stained nerve cross-sections were produced on an Olympus BX51WI Microscope at 100× equipped with a mechanical stage and Microbrightfield Neurolucida software and captured with an Optronics Microfire (1,600 × 1,200) charge-coupled device camera (0.075 μm/pixel). Postprocessing with an Adobe Photoshop action script sequence (levels, curves, contrast, sharpening, and posterization) provided high-contrast 2-bit renditions of the original images (Figure, Supplemental Digital Content 1, ). A MATLAB script was used to segment these images and to automate counts and size determinations for all the axons in each nerve. The measurements excluded the myelin sheath and used eccentricity and size criteria to eliminate glial and vascular profiles. Axon sizes (areas) were mapped into bins to assess size distributions of the axonal populations by area. Axonal profiles were also examined by transmission electron microscopy (JEOL-1200) in thin sections obtained from selected specimens.
Diffusion-Tensor Imaging of the Optic Nerve
Images were collected in an Oxford Instruments 4.7-T magnet equipped with 15-cm inner diameter, actively shielded gradient coils, and interfaced with a Varian INOVA console, as described previously (17). Diffusion-tensor imaging (DTI) data were acquired using a conventional spin-echo imaging sequence, modified by the addition of a Stejskal-Tanner diffusion-sensitizing gradient pair. Six images with different gradient directions were acquired with a b value of 785 seconds/mm2, together with a reference spin-echo (b = 0) image. Slice thickness was 0.5 mm, with a field of view of 1.5 cm × 1.5 cm2. The 3 primary diffusivities, λ1 > λ2 > λ3, were calculated by diagonalization of the diffusion tensor using software written in MATLAB. These primary parameters were combined into relative anisotropy (RA) and RD (the mean of λ2 and λ3) (23).
Immunofluorescence and Terminal Deoxynucleotidyl Transferase dUTP Nick End Labeling Staining
Mice were perfused transcardially with PBS and 4% paraformaldehyde in PBS. After overnight postfixation at 4°C, eyes and optic nerves were either transferred to 70% ethanol solution before paraffin embedding or into 30% sucrose for 36 hours before embedding into optimal cutting temperature medium and snap freezing. For neurofilament staining, 20-μm cryostat sections were incubated with solutions in the following order: 10% normal donkey serum, mouse monoclonal anti-Neurofilament 68 antibody (Sigma, St Louis, MO), and fluorescent anti-mouse secondary antibody in PBS. Finally, sections were stained with Hoechst (Molecular Probes, Eugene, OR) and mounted onto glass slides. Images were acquired with confocal microscopy (Olympus FluoView1000; Olympus, Tokyo, Japan). Terminal deoxynucleotidyl transferase nick end labeling (TUNEL) and immunofluorescence double labeling was used on 5-μm paraffin sections. First, NeuN (Chemicon, Temecula, CA), microtubule-associated protein-2 (MAP2) (PharMingen, San Diego, CA), glutamine synthetase (Millipore, Billerica, MA), or cyclin D3 (Cell Signaling, Beverly, MA) staining was performed using Alexa568 conjugated secondary antibodies (Molecular Probes), as described previously (24). Next, the same sections were TUNEL stained according to the manufacturer's instructions (Roche Diagnostics, Nutley, NJ). Finally, slides were coverslipped with 4′,6-diamidino-2-phenylindole·2HCl-containing mounting medium. The percentage of TUNEL+ cells in the ganglion cell layer was counted on 3 consecutive paraffin sections. For the GFAP-Cre, R26R-EYFP retina immunofluorescence staining was performed as described previously (24).
Each experiment was performed with at least 4 animals from 2 or more independent litters. Statistical significance (p < 0.05) was determined by unpaired t-test (with Welch correction) using GraphPad Prism 4.0 software (GraphPad, Inc, San Diego, CA).
Nf1+/−GFAPCKO Mice Exhibit Reduced VEPs
Visual acuity in children with NF1-associated OPG is usually assessed using standardized eye charts, including Teller acuity, Lea figure, HOTV matching, and Snellen testing modalities (6). Although these measurements provide reliable information about visual function, they are not applicable for visual testing in rodents. Instead, we chose to use electrophysiological assessments of neuronal function, as have been used in children with NF1-associated OPG (9, 25). Visual-evoked potentials and electroretinograms were recorded in WT and Nf1+/−GFAPCKO mice at 6 weeks and 3 months of age. These specific time points were chosen to reflect a period of early tumor evolution in which glial hyperplasia, neovascularization, and microglial infiltration are seen, but gross histological, pathological, and neuroimaging evidence of a glioma is lacking (6 weeks of age) and a time point when an obvious glioma is detected by gross examination (Fig. 1A), pathological features (e.g. nuclear atypia), and MRI (26). At both 6 weeks and 3 months, VEP recordings revealed significant decreases in amplitudes, with more modest increases in latency, in Nf1+/−GFAPCKO compared with WT mice (Figs. 1B, C). In contrast, there was no change in the amplitudes or latencies of either the photopic or scotopic full-field ERGs (Figs. 1D-F). These observations indicate that visual function is impaired in this Nf1 mouse OPG model at an early stage of tumor development.
Nf1+/−GFAPCKO Optic Nerves Have Increased Numbers of Large Diameter Axons
Next, we sought to identify structural correlates for the impaired visual function in Nf1+/−GFAPCKO mice using automated axon counting on paraphenylenediamine-stained semithin cross-sections (Figure, Supplemental Digital Content 1, ). There was no loss of axons at either 6 weeks or 3 months of age by automated axon counting (Fig. 2B). At both ages, however, there was an increase in the proportion of axons with larger (>0.25 μm2) calibers (Fig. 2A). There was no difference between the number of smaller (<0.15 μm2) and larger nerve fibers in WT mice (Fig. 2C). These results suggest that the impairment of visual function in Nf1+/−GFAPCKO mice is associated with ultrastructural changes in optic nerve axons as early as 6 weeks of age.
Nf1+/−GFAPCKO Mice Exhibit Progressive Increases in Optic Nerve Water Diffusivity by MRI
We next wished to determine whether we could use DTI to investigate axonal integrity and diffusivity in mice with optic gliomas. In contrast to the visual physiology or ultrastructural findings, there was no change in RA or RD in Nf1+/−GFAPCKO mice at 6 weeks of age compared with WT mice (Figs. 3A, B), but at 3 months of age, Nf1+/−GFAPCKO mice displayed significantly decreased RA and greater RD values compared with WT mice. Although the RA and RD values did not change significantly in WT mice between 3 and 6 months of age (data not shown), the RD progressively increased in Nf1+/−GFAPCKO mice at both 4 and 6 months of age (Fig. 3D), which likely reflects ongoing changes in water diffusion in these tumors resulting from disturbances in the integrity of the nerve and its axons.
Early Changes in Prechiasmatic Optic Nerve Fiber Alignment in Nf1+/−GFAPCKO Mice
To define the effect of tumor formation in the prechiasmatic optic nerve on axonal organization, we next performed neurofilament staining on longitudinal sections of the optic nerve. Interestingly, the alignment of optic nerve axons is disorganized in the region of the developing tumor (prechiasmatic optic nerve) of 6-week-old Nf1+/−GFAPCKO mice (Fig. 4A). This early appearance of improper alignment of fibers is also seen on longitudinal sections of paraphenylenediamine-stained optic nerves (Fig. 4B). Importantly, the retrobulbar portion of the optic nerve did not show similar alterations in the organization of axonal processes. In addition, sporadic irregularities in axonal morphology in Nf1+/−GFAPCKO, but not WT, mice were observed at 6 weeks (data not shown). These alterations were, however, more prominent in tumor-bearing mice and included multiple neurites in a single myelin sheath (Fig. 4C), lamellar bodies in degenerating axonal spaces (Fig. 4D), and voids with no axoplasm (Fig. 4E). These findings indicate that alterations of optic nerve fiber alignment occur early during optic glioma evolution and progress from initial changes in the region of the developing tumor to involve more distant sites along the optic nerve over time.
Nf1+/−GFAPCKO Mice Develop Progressive Loss of RGCs
Although GFAP is typically regarded as a glial protein, there is evidence from numerous laboratories that GFAP-Cre strains may result in Cre-mediated excision in neuronal populations (27, 28). To exclude the possibility that the VEP and axonal abnormalities observed in Nf1+/−GFAPCKO mice reflected Cre activity in RGCs, we used a Cre reporter line (20): Rosa-YFP; GFAP-Cre mice exhibit YFP expression exclusively in cells in which Cre is expressed. In these studies, YFP expression was observed in GFAP+ cells in the retina, including astrocytic cells of the optic nerve layer and Müller glial cells spanning through the entire retina. We found no evidence of Cre-mediated recombination in NeuN+ neurons (Fig. 5A).
To determine whether the VEP, MRI, and ultrastructural abnormalities observed in Nf1+/−GFAPCKO mice might reflect irreversible loss (apoptosis) of RGCs, we performed TUNEL staining. At 6 weeks of age, there were no apoptotic nuclei in any cell layer of the retina in either Nf1+/−GFAPCKO or WT mice (Fig. 5B). By 3 months of age, however, we found a ∼4-fold increase in the percent of TUNEL+ cells in the ganglion cell layer in Nf1+/−GFAPCKO mice compared with age-matched WT controls (Fig. 5B). This elevated level of apoptosis was not observed in mice lacking Nf1 expression in GFAP+ cells only (Nf1GFAPCKO mice without optic glioma) or in Nf1+/− mice at 3 months of age (data not shown).
To identify which cell types in the retinal ganglion layer had undergone apoptosis, we performed double-labeling immunofluorescence using glial and neuronal markers. In these studies, apoptosis was confined to neurons, as indicated by TUNEL/NeuN (Fig. 5C) or TUNEL/MAP2 double labeling (data not shown). Approximately 15% of neurons in the RGC layer were TUNEL-positive (Fig. 5C). In contrast, Müller glial cell bodies in the inner granular cell layer did not contain TUNEL-stained nuclei, as assessed by cyclin D3/TUNEL or glutamine synthetase/TUNEL double labeling (data not shown).
Although most TUNEL+ cells were neurons, we observed apoptotic cells in the inner nuclear layer, which most likely represent amacrine cells. Because amacrine cells are known to have reduced NeuN expression (29), these neurons would not be identified by NeuN/TUNEL double labeling. Based on the density of neurons in the ganglion cell layer (∼8200 cells/mm2) versus the inner nuclear layer (∼100,541 cells/mm2), containing ∼41% RGCs and ∼39% amacrine cells, respectively (30), we estimate that fewer than 2% of the amacrine cells are TUNEL+. In contrast, a significantly larger proportion of the ganglion cells are undergoing apoptosis.
Collectively, these observations indicate that RGC apoptosis results from optic glioma formation and is not a cell-autonomous effect of reduced or absent Nf1 expression in neurons or glia, respectively.
Genetically engineered mouse models of NF1 have been widely used to identify the key signaling pathways that regulate tumor growth and to perform preclinical therapeutic testing (31, 32). Recently, Nf1 GEM models have also been exploited to define the contribution of cells in the tumor microenvironment to tumor formation and growth. These studies have highlighted the critical relationships that exist between stromal cell types (e.g. microglia and mast cells) and nervous system tumor development and growth (33-35). In contrast, no studies to date have used Nf1 GEM to define the impact of tumor formation on normal nervous system function. The availability of robust and accurate mouse models of Nf1 optic glioma now positions us to directly address the cellular and molecular basis of visual loss resulting from tumor formation and growth.
In this report, we used multiple complementary technologies to define the neuronal abnormalities that develop as a result of optic glioma formation in Nf1+/−GFAPCKO mice. First, we showed that visual-evoked response amplitudes were diminished early during the course of optic glioma formation. The reduced amplitudes most likely reflect decreased numbers of conducting axons, rather than a demyelinating process, a hypothesis supported by an increase in the number of enlarged axons as well as neuronal disorganization observed in Nf1+/−GFAPCKO mice at 6 weeks of age. In contrast, no ERG abnormalities were observed, suggesting minimal impairment in photoreceptor or bipolar cell function in Nf1+/−GFAPCKO mice. Collectively, these observations suggest that VEPs may accurately detect visual dysfunction early in the course of disease and might be a predictor of the risk of tumor formation; however, the lack of a temporal correlation between visual physiology measurements and ultrastructural changes may limit the use of this modality for monitoring tumor growth and response to therapy. Future studies are planned using optometry as a method for assessing vision in these GEM (36).
Second, transmission electron microscopy and automatic cell counting were used to define the ultrastructural changes that result from OPG formation. The presence of increased axonal diameters in the optic nerves of Nf1+/−GFAPCKO mice suggests a degenerative process affecting axons. This progressive axonal degeneration is further supported by the finding that axonal disorganization is first observed in the region in which OPG develops in these mice (the prechiasmatic optic nerves and chiasm), followed only later by changes in the retro-orbital optic nerve segments and, finally, by death of neurons in the retina ganglion layer. This axonopathic process is most likely not cell autonomous because retinal ganglion layer cell death is not observed in Nf1+/− mice or mice-lacking Nf1 expression in GFAP+ cells. We favor the interpretation that neuronal cell death reflects the process of tumor formation that disrupts the normal stabilizing and metabotropic relationships between axons and supportive cells in the prechiasmatic optic nerve region.
Third, we sought to use diffusion-based MRI to examine changes in axonal integrity in the intact animal. Using DTI to detect alterations in water movement (37, 38), we found a time-dependent decrease in RA and increase in RD in Nf1+/−GFAPCKO mouse optic nerves. These changes are not obvious at 6 weeks of age, but they are detected by 2 to 3 months of age (26). The strong correlation between the ultrastructural abnormalities observed by electron microscopy and the diffusion changes detected by MRI suggests that DTI might be useful for assessing the effect of tumor progression on axonal and myelin integrity. Future studies will be required to determine whether diffusion-based imaging methods reflect or predict visual deterioration in children with symptomatic NF1-associated OPG.
Finally, our studies suggest the presence of bidirectional communication between neoplastic cells and nonneoplastic cells in the tumor surround. We have previously identified stromal cell types (microglia) and growth factors (CXCL12 and hyaluronidase) that influence optic glioma growth in Nf1+/−GFAPCKO mice, implying that the tumor microenvironment influences tumor growth (33, 39). The current study highlights the impact of tumor formation on the nonneoplastic cell types present in the region of the developing tumor. We show that only Nf1+/−GFAPCKO mice exhibit RGC apoptosis. This observation suggests that tumor development is required to initiate the cascade of degenerative events that culminate in permanent cell death and vision loss. Alternatively, it is possible that Nf1-deficient neoplastic glia elaborate proapoptotic signals that uniquely affect Nf1+/− neurons. In this regard, some studies have demonstrated that Müller glia and astrocytes can reduce excitotoxic RGC death (40, 41), whereas others have shown that astrocytes do not enhance RGC survival in vitro (42, 43). It is also conceivable that neurofibromin loss in astrocytes either leads to the generation of proapoptotic factors or results in impaired glia-mediated neuronal survival, either of which contributes to increased Nf1+/− RGC death. We are currently exploring these possibilities using astrocyte-RGC coculture experiments in vitro.
In summary, using multiple complementary modalities, we have shown that OPG formation in mice results in altered visual function. We observed decreased VEP responses and a shift toward larger caliber axons before macroscopic optic glioma formation, whereas after optic gliomas were obvious, we found progressively increasing radial diffusivities on MRI and ultrastructural damage in the optic nerve axons, culminating in RGC death. These findings highlight the use of GEM models of human cancer for recapitulating some of the functional abnormalities seen in children with similar tumors as well as providing important insights into their pathogenesis.
We thank Ryan Emnett and Emily Barr for expert technical assistance and Dr Bryan Smith for the development of the MATLAB program for axon analysis. We also thank Dr Frank Costantini (Columbia University) for generously providing the R26R-EYFP mice.