PROX1 is a prospero-related transcription factor that plays a critical role in the development of various organs including the mammalian lymphatic and central nervous systems; it controls cell proliferation and differentiation through different transcription pathwaysand has both oncogenic and tumor-suppressive functions. We investigated PROX1 expression patterns in 56 human astrocytic gliomas of different grades using immunohistochemistry. An average of 79% of cells in World Health Organization Grade IV (glioblastoma, n = 15) and 57% of cells in World Health Organization Grade III (anaplastic astrocytoma, n = 13) were strongly PROX1 positive; low-grade diffuse astrocytomas (Grade II, n = 13) had 21% of cells that were strongly positive; Grade I tumors (n = 15) had 1.5%; and non-neoplastic brain tissue (n = 15) had 3.7% of cells that were PROX1 positive. Double immunolabeling showed that PROX1+ cells in glioblastomas frequently coexpressed early neuronal proteins MAP2 and βIII-tubulin but not the mature neuronal marker protein NeuN. Analyses of coexpression with proliferation markers suggest that PROX1+ cells have a marginally lower rate of proliferation than other tumor cells but are still mitotically active. We conclude that PROX1 may constitute a useful tool for the diagnosis and grading ofastrocytic gliomas and for distinguishing Grade III and Grade IV tumors from Grade I and Grade II tumors.
Gliomas are the most common brain tumors in adults and are thought to arise from astrocytes, oligodendrocytes, or their precursors (1). The hypothesis that brain tumors originate from a tissue stem cell or a progenitor cell has recently attracted interest (2-4). Astrocytic gliomas are classified by the World Health Organization (WHO) into 4 grades that are correlated with clinical outcome: Grade I pilocytic astrocytoma, Grade II astrocytoma (low-grade diffuse), Grade III anaplastic astrocytoma, and Grade IV glioblastoma (GBM) (5). Glioblastomas diffusely infiltrate the brain parenchyma and are poorly differentiated tumors; they show cellular polymorphism, high proliferative activity, necrosis, and neovascularization (6). Glioblastomas are the most aggressive type of brain tumor and are associated with a poor prognosis. Current glioma classification and grading are based on histopathologic methods and are thus subjective to some extent (6, 7). The development of new molecular markers for diffuse astrocytic gliomas could improve diagnosis and clinical management of patients with astrocytomas.
PROX1 is a homeobox gene cloned in humans by homology with the Drosophila gene Prospero and mouse Prox1 (8, 9). PROX1 is an early specific marker for the developing liver and pancreas in the mammalian foregut endoderm (10, 11). It is also one of the most specific and reliable markers for lymphatic endothelial cells (12). Overexpression of PROX1 in vascular endothelial cells promotes differentiation into a lymphatic endothelial phenotype and induction of cell proliferation markers (13). In the developing eye lens, the loss of PROX1 function led to downregulation of the cell cycle inhibitors p27 and p57 and deregulation of E-cadherin (14).
Decreased expression of PROX1 has been detected in some tumors, suggesting that it has a possible role as a tumor suppressor gene. In support of this notion, overexpression of PROX1 decreased hematopoietic cell growth, and its expression was lost in several hematological malignancies caused by DNA intron methylation in PROX1 (15). A hypermethylated CpG island within the first intron of PROX1 leading to a decrease in PROX1 expression was identified in a microarray-based study of sporadic breast cancer (16); it was also found to be decreased because of epigenetic silencing in carcinomas of the biliary system (17). In contrast, a recent study shows that PROX1 promotes colorectal cancer progression and transition to a more malignant phenotype (18). The functions of PROX1 in tumors as well as in development are, therefore, cell type dependent.
PROX1 regulates gene expression during cell differentiation in the mammalian central nervous system (CNS). It is expressed in a defined transitory stage during interneuron neurogenesis in the mouse retina (14). PROX1 expression also coincides with early stages in the differentiation of neuroepithelial stem cells to neuronal and glial cells (19, 20). Indeed, PROX1 protein is present in the cells of the subventricular zone, and at early postnatal stages, it is mainly detected in neuron nuclei in the thalamus, cerebellum, and hippocampus (19, 21, 22).
We previously observed differential expression of PROX1 mRNA in cultured human GBM cells (data not shown) (23). This suggested to us that PROX1 may have a role in brain tumors. Here, we describe the expression patterns of PROX1 in astrocytic brain tumors of different grades using immunohistochemistry (IHC).
Materials and Methods
Cases of Grade I (n = 15), Grade II (n = 13), Grade III (n = 13), and Grade IV (n = 15) astrocytic tumors were collected at the Karolinska University Hospital in Solna between 2004 and 2007. Fifteen cases of various non-neoplastic intracerebral conditions were also analyzed. Formalin-fixed paraffin-embedded sections were anonymized for this study; the use of these specimens was approved by the regional ethical committee in Stockholm (2005/542-31/1). Pathologist A.O. reviewed all hematoxylin and eosin-stained slides to confirm diagnoses and grading.
The following antibodies were used: rabbit anti-PROX1 (ab11941; Abcam, Cambridge, UK) diluted 1:100; goat anti-PROX1 (R&D Systems, Minneapolis, MN), 1:100; mouse anti-proliferating cell nuclear antigen ([PCNA] clone PC10, sc-56; Santa Cruz Biotechnology, Santa Cruz, CA), 1:500; mouse anti-glial fibrillary acidic protein ([GFAP] clone GF12-24; Chemicon/Millipore, Billerica, MA) 1:1000; mouse anti-neuronal βIII-tubulin (clone TuJ1; BabCO, Richmond, CA), 1:500; mouse anti-Ki67 (clone MIB-1; Dako, Glostrup, Denmark), 1:100; mouse anti-NeuN (clone A60; Chemicon), 1:100; rabbit anti-SOX2 (Chemicon) diluted 1:500; and mouse anti-microtubule-associated protein 2 ([MAP2] clone HM-2; Sigma Aldrich, St Louis, MO), 1:200. The secondary antibodies used were: polyclonal biotinylated swine anti-rabbit IgG (Dako E0353), 1:200, and polyclonal biotinylated rabbit anti-goat (Dako E0466), 1:500. Texas Red or fluorescein isothiocyanate-conjugated anti-rabbit, Texas Red anti-goat, and Texas Red or fluorescein isothiocyanate-conjugated anti-mouse antibodies were from Vector Laboratories (Burlingame, CA).
Immunostaining was performed using the Ventana Discovery Automated Stainer (Ventana Medical Systems, Tucson, AZ), following the vendor's suggestions. Deparaffinization was done in the Ventana machine. The heat-induced epitope retrieval incubation was in Tris-borate EDTA buffer pH 8.0. Primary antibodies were diluted in 1% bovine serum albumin, 0.1% Tween-20 in PBS. A streptavidin-biotin horseradish peroxidase-based diaminobenzidine kit (Ventana Medical Systems) was used for detection of antibody staining. Secondary antibodies were diluted in antibody diluent (Ventana Medical Systems). Sections were counterstained with hematoxylin. After staining, the slides were rehydrated in graded ethanols, cleared in xylene, and mounted in Pertex. Omission of primary antibodies served as negative controls.
Evaluation of Immunostaining
Each section was examined under 100× to 400× magnifications using an Olympus light microscope (UPMTVC, Japan) and a coupled Leica camera (DFC320). Immunoreactivity was evaluated for the quantity and intensity of staining. Numbers of PROX1+ cells were determined in high cell density areas, and at least 600 cells were counted in each section. Positive staining intensities were scored as either weak positive or strong positive. With the exception of obvious inflammatory cells, all neoplastic and non-neoplastic cells were counted in areas with the highest cell densities in the tumors. To rule out the possibility that infiltrating non-neoplastic cells introduced a counting bias that might result in a difference between Grade II and Grade III tumors, we examined the same areas from consecutive tumor sections for SOX2 expression (n = 5 for each grade) (Figure, Supplemental Digital Content 1, ). SOX2 is equally expressed in Grade II and Grade III tumors but expressed in less than 10% of cells in Grade I pilocytic astrocytomas or in normal brain (24); therefore, we used it as a marker of neoplastic cells. All cases were stained with the goat anti-human PROX1 antibody and scored. As a further control, 3 randomly picked cases of each of the tumor grades were stained with both rabbit anti-human PROX1 and goat anti-human PROX1, and the percentages of PROX1-expressing cells in the sections were calculated for both and the relative difference was calculated (Figure, Supplemental Digital Content 2, ; Table, Supplemental Digital Content 3, ). Continuous variables were expressed as mean ± SD, and a difference between means was considered significant if p < 0.05 using the t-test. Correlations were calculated using a bivariate correlation analysis using Pearson correlation coefficient followed by significance test.
Double Immunofluorescence on Paraffin Sections
Sections were incubated for 2 hours at 60°C. For deparaffinization, sections were left overnight in xylene, followed by rehydration in graded solutions of ethanol 100% (2 × 5 minutes), 95% (2 × 2 minutes), 80% (2 minutes), and dH2O (5 minutes). Antigen retrieval was performed by 10 minutes of microwave exposure in 10 mmol/L citrate buffer (pH 6.0) followed by 20 minutes at a sub-boiling temperature. Slides were then allowed to cool for at least 30 minutes and were washed 3 times in dH2O and once in PBS. Nonspecific antibody binding was blocked by incubating the sections for 1 hour at room temperature in 5% milk in PBS. Primary antibodies were diluted in blocking solution and incubated on the sections overnight at 4°C. Sections were rinsed in PBS 3 × 5 minutes followed by incubation with fluorescein isothiocyanate or Texas Red-conjugated secondary antibodies for 1 hour at room temperature. Sections were mounted using Vectashield mounting media with 4′,6-diamidino-2-phenylindole ([DAPI] Vector Laboratories). We used a Zeiss Axioplan II microscope controlled by Axiovision 3.1 software and equipped with Plan-Apochromat 63_/1.4 and Plan-Neofluar 100_/1.30 objectives. Images were assembled in Adobe Photoshop CS3.
PROX1 Expression in Astrocytic Gliomas of Different Grades
We used commercially available goat and rabbit polyclonal anti-PROX1 antibodies, raised against the N-terminus and C-terminus of PROX1, respectively. The antibodies were tested by immunofluorescence staining in SW480 colon carcinoma cells, known to be positive for PROX1, and also in U2OS osteosarcoma cells as a negative control (18) (Data, Supplemental Digital Content 4, ). These antibodies gave identical strong nuclear PROX1 staining in SW480 cells, whereas no reactivity was seen in U2OS cells (Supplemental Digital Content 2, parts A-C, ). Staining subsequent to the use of siRNA against PROX1 and overexpression of PROX1 cDNA in negative cell lines further confirmed the specificity of the antibodies, in agreement with published findings (18) (Supplemental Digital Content 2, parts A-C, ); thus, both antibodies specifically recognize PROX1. After antigen retrieval, both antibodies produced strong and distinctive nuclear staining in tissue sections of gliomas, but not in cells from “normal” adult brain (Fig. 1A; Supplemental Digital Content 2, part D, ). PROX1 expression was confined to the cell nuclei and was particularly evident in areas of tumor cell crowding.
Using the goat anti-human PROX1 antibody, we found that PROX1 exhibited a strikingly positive and statistically significant correlation with glioma tumor grade (Fig. 1A; Tables 1 and 2). An average of 79% of cells in the GBM and 57% of cells in WHO Grade III (anaplastic astrocytoma) were strongly positive. Low-grade diffuse astrocytomas showed significantly less expression of PROX1; 21% of cells were clearly PROX1+ in the Grade II tumors; a fraction of cells with ambiguous staining were scored as weakly PROX1+ (Table 1). There was a significant correlation between the increased percentages of strongly PROX1+ cells and higher glioma grade (r = 0.78, p < 0.01). For strongly positive cells in Grade III and Grade IV compared with Grade II cases, the difference was still significantly different (p < 0.001; Table 2). Thus, PROX1 expression might be a useful diagnostic tool to distinguish high-grade tumors (III, IV) from Grade II tumors. Grade I astrocytomas showed significantly lower expression of PROX1, that is, on average, 1.5% of cells were strongly positive. PROX1 expression in diffuse astrocytic glioma Grade II compared with that in Grade I tumors was significant (p < 0.001; Table 2). This is consistent with the fact that pilocytic astrocytomas (Grade I) constitute an entirely different entity, associated with specific genomic aberrations (25, 26). Nearly identical results were obtained using the other PROX1 antibody (Supplemental Digital Content 2, part D, ; Table, Supplemental Digital Content 3, ). Also in support of these findings, our evaluation of recently published global SAGE data from 12 cases of GBM (27) showed that gliomas, on average, express a 4-fold increase of the PROX1 mRNA compared with normal brain tissue.
PROX1 Is Not Expressed in Glioma-Associated Microvascular Structures
More numerous PROX1-expressing cells were seen in the peritumoral normal-appearing tissue adjacent to the high-grade tumor areas compared with non-neoplastic control samples (Fig. 1B; Table 1). The PROX1+ cells seemed to be randomly scattered in the peritumoral tissue with no preference for perivascular areas. Tumor-induced angiogenesis is a hallmark of high-grade malignant glioma. Glioma vessels are characterized by irregular shape, glomeruloid formations, and bleeding; and they may also express lymphatic vessel markers (28). PROX1 was, however, absent in vessels and was confined to the tumor cells in all diffuse astrocytic tumors (Fig. 1C).
PROX1 Expression in “Reactive” Non-Neoplastic Conditions
PROX1 expression was very low in reactive non-neoplastic brain tissue, with an average of 3.7% of cells with strong staining (Fig. 2). This is much lower than the average of PROX1+ cells in the astrocytic tumors (Fig. 3; Table 1). In 2 of the non-neoplastic 15 cases, there was a somewhat higher percentage of strong positive cells (16% and 12%); another case showed 8% PROX1+ cells. The remaining 9 cases had 0% to 5% strong positive cells. Taken together, this suggests that strong PROX1 expression is significantly more common in all grades of astrocytic glioma tumors than in non-neoplastic lesions (Tables 1 and 2); moreover, in reactive lesions, the staining was mostly weak (Fig. 2).
Expression Pattern of PROX1 in Relation to Proliferation Markers
Proliferating cell nuclear antigen has been widely used as a marker for proliferation in brain tumors because it shows maximum expression in S-phase cells (29); increased PCNA reactivity may also occur as part of the cellular DNA damage response (30). Moreover, PROX1 interacts physically and functionally with PCNA through a conserved PIP-box motif in the Prospero domain (31). Therefore, we assessed a possible relationship between PCNA and PROX1 in the gliomas using double immunofluorescence protocols adapted for paraffin-embedded sections. The PCNA+ cells were frequently scattered throughout the sections from tumors of all grades (Fig. 4A and data not shown). In high-grade tumors, most cells that exhibited strong PCNA signals were only weakly, or not at all, PROX1+ (Fig. 4A). We calculated PCNA labeling index by counting at least 200 cells per field in 5 different fields of a Grade IV tumor and found that, on average, 35% of all cells in this sample expressed PCNA, whereas 20% of all cells scored as strongly PROX1+ (6.3% PCNA/PROX1 double-positive cells of all cells in the section). Thus, the frequency of PCNA+ cells among PROX1+ cells was 32%, which is marginally lower than the percentage of PROX1-negative cells that were PCNA+ (36%). In this assay, the number of PROX1+ cells is presumably slightly underestimated because of lower sensitivity when using double immunofluorescence on formalin-fixed paraffin-embedded material.
We next assessed Ki67 as a proliferation marker, which produced a more distinct positive signal with less or no background (Fig. 4B). Calculating the apparently more reliable Ki67 labeling index (13%) in relation to the number of PROX1 high expressing cells (20%) again revealed that PROX1+ cells are more likely to be Ki67 negative than the average cell population. Only 2% of all cells were Ki67+/PROX1+ (Fig. 4C). The frequency of Ki67+ cells among PROX1+ cells was 10%, which was marginally lower than the percentage of Ki67+ cells in the PROX1-negative cell population (13%). This suggests that PROX1+ cells have a marginally lower rate of proliferation compared with the average tumor cell but are still mitotically active.
PROX1-Expressing Cells Are Positive for Markers of the Neuronal Cell Lineage
To phenotype PROX1+ cells, immunofluorescence with double labeling was performed with neural differentiation markers. The βIII-tubulin is an immature neuronal marker expressed in neurons of the peripheral and central nervous systems (32, 33) and in brain tumors (34). Microtubule-associated protein 2 is an early mature neuronal marker found specifically in the perikarya and dendrites of postmitotic neurons and frequently expressed in brain tumors (35,36). NeuN is a protein specifically localized in the nucleus of terminally differentiated neuronal cells and is usually not expressed in GBM (37). We found focal and abundant expression of βIII-tubulin and MAP2 in Grade IV tumors (Fig. 5A) and more widespread and strong immunoreactivity for GFAP. Interestingly, the tumor cells did not express NeuN, indicating that the MAP2- and βIII-tubulin-positive cells were blocked in early stages of neuronal differentiation and had failed to undergo terminal differentiation (Fig. 5A). A few randomly scattered NeuN-positive cells present in the normal adjacent tissue served as an internal positive control. Indeed, by IHC, NeuN was not expressed in areas of high cell density in any Grade IV tumors but could easily be seen in adjacent normal areas from the same sections (Figure, Supplemental Digital Content 5, ). Glial fibrillary acidic protein showed lower expression in regions positive for PROX1 (Figs. 5A-C), whereas MAP2 and βIII-tubulin showed a high overlap with PROX1+ regions within the double-stained sections using low magnification (Figs. 5A-C). Further analysis counting another double-stained Grade IV sample revealed that PROX1+ cells frequently coexpressed βIII-tubulin (62%) and MAP2 (79%) and to a lesser extent GFAP (32%) (Figs. 5B, C). Grade IV tumors also exhibited strong SOX2 immunoreactivity in most tumor cells (>50%), indicating that a fraction of PROX1+ cells must also coexpress SOX2 in high-grade tumors (Figure, Supplemental Digital Content 1, ). In line with this, a subset of PROX1+ cells did not express neuronal markers MAP2 and βIII-tubulin, indicating that these cells could represent an intermediate stage of differentiation between an early neural precursor and a postmitotic neuronal cell, in agreement with earlier studies on PROX1 in normal brain (19). SOX2 is equally expressed in Grade II and Grade III tumors but expressed in less than 10% of cells in Grade I pilocytic astrocytomas or in normal brain (24). Therefore, we compared the ratios of SOX2- and PROX1 (strong)-expressing cells in the same tumor areas using consecutive sections (n = 5 per grade). A high frequency of SOX2+ cells remained in Grade II astrocytomas, whereas the frequency of strongly PROX1+ cells was much lower in Grade II (Fig. 5D).
New molecular tools for improving diagnosis and grading of astrocytomas and a better understanding of their molecular pathogenesis are needed. Here, we focused on the homeobox gene PROX1 that encodes a transcription factor that plays a major role in the development of the lymphatic system (38) and is also involved in the differentiation of cells in the CNS (39). Induction of MASH-1 and PROX1 could be critical molecular events that control early development of the CNS (20).
We found that PROX1 is expressed in diffuse astrocytic gliomas and that its expression increases with tumor grade. Low PROX1 expression was found in Grade I astrocytomas, and lower levels were seen in non-neoplastic conditions. In GBMs, PROX1 was intensely expressed in densely cellular highly proliferative areas of the tumor, but close inspection showed that tumor cells and not the vasculature were PROX1+.
Proliferation measurements are used as an adjunct tool for grading gliomas (40), and we found that individual PROX1+ cells can express the proliferation markers PCNA and Ki67. For both proliferation markers, however, the labeling index of PROX1+ cells was marginally lower than the mean labeling index of all cells within the section, indicating that PROX1+ cells are dividing at a slightly slower rate than the rest of the tumor cells but are still mitotically active. The PROX1+ cells did not show increased proliferation, indicating that PROX1 is not simply a proliferation marker but represents a separate quality likely related to a progenitor stage. Thus, PROX1 IHC might be a useful tool for grading astrocytic gliomas. To determine whether PROX1 IHC can assist in difficult cases of grading uncertainties and/or be of prognostic value in the clinically heterogeneous group of patients with astrocytoma Grade II (41), however, it will be necessary to study a large number of patients with long-term follow-up and known clinical outcome.
In clinical situations, it is also important to distinguish neoplastic from non-neoplastic CNS tissue; discriminating between astrocytoma Grade II and inflammatory/reactive lesions and between radiation necrosis and tumor recurrence can be difficult. Similarly, identifying a high-grade glioma when a biopsy is from the periphery of a glioma and in which tumor cells cannot be distinguished from normal resident cells is often challenging. Our findings of scattered strongly PROX1+ cells in the normal-appearing tissue adjacent to high-grade gliomas (Fig. 1B) and of higher numbers of strongly PROX1+ cells in Grade II astrocytoma than in non-neoplastic lesions suggest the potential use of PROX1 IHC in these situations. Non-neoplastic lesions do, however, have PROX1+ cells, especially many weakly positive cells (Table 1). Nevertheless, PROX1 IHC might be used for discriminating between neoplastic and reactive lesions, with the caveats that only strongly positive cells are considered and that other types of markers, such as p53 or WT1 (42, 43), are included in the evaluation. Future evaluations of the clinical use of PROX1 IHC in a larger patient series with detailed clinical and radiological data are warranted.
Our findings raise questions regarding the identity of the PROX1+ brain tumor cells. The PROX1 fly homologue Prospero is an important regulator of Drosophila neural cell differentiation needed by ganglion mother cells to activate neuronal differentiation genes and to repress neuroblast-specific and cell cycle regulatory genes (44-46). The size of PROX1 is significantly different from that of Prospero, and it is unclear to what extent Prospero functions are conserved in the mammalian PROX1 protein. PROX1 is, however, transiently expressed at an early stage of mammalian neuronal specification, that is, it is involved in promoting cell cycle exit but is not capable of executing a full differentiation program (14, 19, 20). Thus, PROX1 likely is a progenitor cell marker, and it is possible that PROX1+ cells in the tumors are progenitor cells. An increase in the number of multipotent progenitors in high-grade tumors as part of an undifferentiated phenotype would explain the increased PROX1 expression we observed. This is supported by our finding that PROX1+ cells can express both neuronal markers (MAP2, βIII-tubulin) and also, but less frequently, GFAP. This interpretation is also consistent with the fact that most of the cells in Grade IV tumors seem to express SOX2, a transcription factor normally present in neuroepithelial stem cells and in amplifying precursor cells, and that can be coexpressed with neuronal markers in GBM (24). The equal percentage of SOX2+ cells in Grade II and Grade III gliomas (Fig. 5D) is a strong indication that we counted an equally representative fraction of tumor cells in both grades. Why then is there a difference in PROX1 expression between Grade II and Grade III? Because Grade II tumors have a lower mitotic index and are sometimes well differentiated; the number of PROX1+ cells (presumed neuronal precursor cells) would actually be expected to be lower in these tumors. The PROX1+ cells are more frequently detected in high-grade tumors with anticipated accumulation of tumor cells with progenitor or stem cell-like properties.
Brain tumor stem cells can produce morphologically aberrant cells expressing both glial and neuronal antigens (47), and similarities between amplifying neural progenitors and GBM cells may have implications for understanding the pathogenesis of GBM. PROX1 emerges as an additional marker protein distinct from commonly used proliferation markers that may help in the identification of different cell populations within GBM. Our study also indicates that strong PROX1 expression may be most useful as a diagnostic tool for grading diffuse astrocytic gliomas. When the finding of a significantly higher PROX1 expression in Grades III and IV than in Grade II astrocytic tumors has been validated, this simple IHC application may be adopted for clinical use.
The immunohistochemistry was performed at the core facility Mouse Tissue Analysis at Karolinska Institutet/Karolinska University Hospital in Solna.