Prognostic value of tumor microinvasion and metalloproteinases expression in intracranial pediatric ependymomas

Abstract

Ependymomas are common pediatric intracranial neoplasms that often appear well circumscribed on imaging yet, when treated with surgical resection alone, many recur. The current WHO histologic grading system does not accurately predict clinical behavior. The aim of this study was to identify histological and immunohistochemical features that might better correlate with clinical course in patients with ependymomas treated with gross total resection. In a retrospective study, we performed a histologic and immunohistochemical analysis of 41 pediatric ependymomas for microinvasion and for expression of metalloproteinases MMP2 and MMP14, ezrin and bcl-2 and correlated the findings with clinical outcome. We found gross total resection to have significant positive influence on overall survival and progression free survival. In 28 patients that underwent gross total resection, microinvasion was a predictor of poor overall survival (p= 0.003) and progression free survival (p=0.03).Gross totally resected tumors with high expression of MMP2 and MMP14 had significantly shorter overall survival. Our data shows that tumor microinvasion into adjacent brain and tumor expression of MMP2 and MMP14 are significant predictors of both overall survival and progression free survival in pediatric ependymomas and these are useful prognostic markers that may be helpful in stratifying patients for adjuvant treatments.

INTRODUCTION

Ependymomas are the third most common intracranial neoplasm in children. In adults, many ependymomas occur in the spinal cord and include myxopapillary ependymomas, i.e. Grade I tumors according to World Health Organization (WHO) classification (1). In contrast, pediatric ependymomas are predominantly intracranial and are WHO Grade II, or Grade III (anaplastic) tumors. The clinical management of pediatric ependymomas is challenging due to imperfect correlation between histologic grade and prognosis. Different histological or immunohistochemical criteria have been proposed as prognostic factors, however, none, including the widely-used WHO classification, correlate well with prognosis. The extent of surgical resection remains the most important variable in predicting outcome. The role of chemotherapy is yet undetermined.

On imaging, ependymomas appear well-demarcated, and display significant gadolinium enhancement. Despite their well-demarcated appearance on imaging, following full resection, the most common site of recurrence is within the original tumor bed suggesting that tumor cells infiltrate beyond the surgical margins. This is supported by the significant improvement in long-term survival for patients that receive focal radiation therapy to the primary tumor bed. The current pathological diagnostic criteria do not include the well demarcated versus infiltrative nature of these lesions; furthermore, there are no tested immunohistochemical markers for that purpose.

To test the hypothesis that microscopic invasive ependymoma cells can be identified by histological and immunohistochemical analysis and that the presence of invading cells is associated with poor prognosis, we performed a retrospective study on 41 pediatric ependymomas. Samples were evaluated for microinvasion, expression of matrix metalloproteinases MMP2 and MMP14 (also known as MT1-MMP) and correlated with clinical outcome. We chose MMP2 and MMP14 based on their role in matrix degradation, their presence in infiltrative CNS neoplasms (2, 3, 4) and their correlation with grade and invasion in malignant gliomas. In addition to metalloproteinases, we also evaluated expression of ezrin as it has been shown to play an important role in various human tumors (5, 6, 7, 8, 9) and is associated with increased migration potential in malignant gliomas (10). Its role in predicting prognosis in ependymomas (11) remains unclear. Finally, Bcl-2 has been associated with radioresistance (12) and plays a role in migration behavior of malignant gliomas (13, 14) has been evaluated for prognostic implications in ependymomas (15, 16) and was therefore included in this analysis.

MATERIALS AND METHODS

Patients

Histopathological analyses of tumors and medical chart reviews were performed on patients diagnosed with WHO Grade II or Grade III intracranial ependymomas, seen at the Dana- Farber Cancer Institute/ Children's Hospital Boston between 1984 and 2004 for which clinical data were available. Forty-one patients were identified. All tumors were primary with no known spread at the time of diagnosis. Clinical presentation, neuroradiologic imaging, extent of surgery, adjuvant therapy and follow-up were recorded from patients' medical records. Approval from the Institutional Review Board was obtained prior to the initiation of this study (DFCI protocol #98-179).

A gross total resection was defined as a complete removal of all tumor based on both the surgeons interpretation after completion of the procedure and a MRI scan that demonstrated no residual tumor.

A recurrence was defined as a new lesion that appeared after initial treatment (surgery, radiation and/or chemotherapy) while progression was defined as the enlargement of an existing lesion. If repeated resection was performed, histological analysis was used to confirm the presence of ependymoma. A new lesion in the brain consistent with tumor based on imaging and clinical symptoms or the presence of disease throughout the brain and spine on imaging were considered sufficient to make the diagnosis of recurrent/progressive disease when biopsy or resection was not considered clinically indicated.

Progression free survival was defined as the time from initial surgery to recurrence/progression as defined above. Overall survival was defined as the time from the initial surgery to death.

Histology

Tissue for all analyses was obtained from the initial surgery. No patients received radiation or chemotherapy therapy prior to the surgery. Formalin-fixed, paraffin-embedded sections were stained by the hematoxylin and eosin method (H&E) from all tumor blocks and were reviewed and graded by two neuropathologists independently (MS, UDeG) using WHO 2000 criteria (1). In cases of interobserver disagreement, the case was reviewed together and the final grade was established as a consensus of both observers. Both for grading and evaluation of invasion on H&E, the entire case was reviewed.

Immunohistochemistry

Representative paraffin blocks from each tumor were selected for immunohistochemical studies. Representative blocks were defined as those with the largest amount of viable and most anaplastic looking tumor and with tumor-brain margin, if available. Blocks containing CUSA content were excluded. Sections were deparaffinized and subjected to antigen retrieval using the heat-induced epitope retrieval method (HIER) by microwaving 30 minutes at 92.2°C or enzyme induced epitope retrieval (EIER) by 0.01% protease digesting 10 minutes in a 37°C water bath. Slides were incubated at room temperature for 40 minutes with the primary antibody and 30 minutes with the secondary antibody. Antibodies were detected by the Envision Horseradish Peroxidase Method using 3,3' diaminobenzidine (DAB) as a chromogen (DakoCytomation, Carpinteria, CA). The following antibodies were used: ezrin (3C 12), mouse monoclonal (Abcam Inc, Cambridge, MA); 1:100 dilution, HIER microwave with Dako citrate buffer (pH= 6.0); MMP2 (CA-4001 / CA719E3C), mouse monoclonal (Abcam Inc, Cambridge, MA); 1:100 dilution, no pretreatment is required; MMP14 antibody, rabbit polyclonal (Abcam Inc, Cambridge, MA); 1:100 dilution protease digestion 10 minutes; and bcl-2 (clone 124), 1:30, heat induced epitope retrieval, (DakoCytomation, Carpinteria, CA); HIER pressure cooker with Dako citrate buffer (pH=6.0). Neurofilament, mouse monoclonal (DakoCytomation, Carpinteria, CA) 1:80 dilution; microwaved in Tris Buffer (pH 10). Results were compared to the appropriate controls as recommended by the manufacturers.

For quantification of MMP2, MMP14 and bcl-2, the following semiquantitative 2-tier staining score scale was used: score 0 = negative or almost negative (<5% positive cells) vs score 1 = positive (>5% positive cells). For ezrin, the following 2-tier staining score scale was used: score 1 = all or almost all cells are strongly positive (90-100% positive cells) vs score 0 = less than 90% positive cells.

In addition to H&E, tumor invasion into adjacent brain was evaluated by neurofilament immunohistochemistry, highlighting entrapped axons. Invasion was evaluated (absent vs present) by two neuropathologists independently (MS, ASR) and called present if multiple axons entrapped in the tumor were observed and both observers agreed on invasion. An example of the neurofilament immunohistochemistry which demonstrates invasion is illustrated in Figure 1G.

Figure 1.

All specimens are from gross totally resected tumors. Grade II ependymomas, with well demarcated edges on H&E (A, B) highlighted by ezrin (D) which did not recur after surgery. Invasive edge of the tumor is apparent on H&E (C) highlighted by ezrin (F). Multiple tumor cells invading white matter are highlighted by ezrin (E). This tumor recurred in the early postoperative period. Figures G, H and I illustrate tumors with an aggressive clinical course in which invasion was not identified on H&E slides. However, immunohistochemistry for neurofilament highlighting multiple entrapped axons (G) and/or strong diffuse immunopositivity for MMP2 (H) or MMP14 (I) predicted an aggressive course. In contrast, tumors that are immunonegative or with rare MMP2 (K) or MMP14 (L) positive cells (arrows) show favorable outcome. Another weaker predictor for aggressive clinical behavior was bcl-2 immunopositivity (J).

Statistics

To determine whether invasion (as seen by microscopic examination) is a stronger prognostic factor than grade for overall survival and progression free survival, a univariate proportional hazards model including only invasion, and a model including only grade as a predictor, as well as a Cox multivariate proportional hazards model including both invasion and grade as predictors was used. We analyzed risk factors for the entire cohort and then separately for gross totally resected tumors only. Correlation between expression of metalloproteinases, ezrin, bcl-2 and survival based on immunohistochemical stains for tumor invasion were determined.

Kaplan-Meier curves for overall survival and progression free survival were evaluated for extent of resection, tumor grade, MMP2, MMP14, bcl-2, and ezrin, for all patients in the study. Kaplan-Meier curves for these last five variables are also presented stratified by degree of resection (gross total versus subtotal resection). For both overall survival and progression free survival, median times, 95% confidence intervals for the median, and logrank p-values are presented for degree of resection, grade, MMP2, MMP14, bcl-2, and ezrin.

In order to evaluate patients for survival (overall and progression free survival) based on all four variables of interest simultaneously (i.e. multivariate), we utilized a latent class model supervised by overall survival, as well as a latent class model supervised by progression-free survival. This supervised latent class model assumes that each patient belongs to an unobserved latent class with some estimated probability, and survival is used to inform the estimation of classes (17).

RESULTS

Of the 41 patients in our study, 11 patients had supratentorial tumors (27%) and 30 patients had infratentorial tumors (73%). Of the 11 supratentorial tumors, gross total resection was achieved in 9 cases (82%), while of the 30 infratentorial tumors; gross total resection was possible in 19 cases (63%).

Median age was 4.2 years (0.7-16.8yrs). Median follow-up time was 54 months and progression-free survival ranged from 0.5 month to 141 months (median, 26 months) for the entire cohort. In the subgroup of the 30 patients (73%) who recurred, progression-free survival ranged from 0.5 month to 141 months (median, 17 months) and in the subgroup of the 24 patients (59%) who died by the end of the study, overall survival ranged from 7 months to 141 months (median, 48 months). Of the 41 patients, 28 (68%) underwent gross total resection and 13 (32%) underwent subtotal resection. Of the 28 patients treated with gross total resection, 10 (36%) never relapsed, 5 (18%) relapsed but were salvaged after additional treatment, and 13 (46%) relapsed, progressed and died. Of the 13 patients who underwent subtotal resection, one patient did not relapse (8%), one relapsed and was salvaged after additional treatment (8%) and 11 (84%) relapsed, progressed and died. Thirty-seven patients (90%) received postoperative radiation therapy; two nonirradiated tumors underwent gross total resection (1 infratentorial and 1 supratentorial tumor) and two were subtotally resected (both infratentorial tumors). Twenty-six patients (63%) received adjuvant chemotherapy of different regimens. Fifteen patients with gross total resection received chemotherapy (8 as postoperative adjuvant therapy, 7 after recurrence), 13 did not. Eleven patients who underwent subtotal resection received chemotherapy (5 as postoperative adjuvant therapy, 6 patients after recurrence) and 2 did not.

Of the 28 patients who underwent gross total resection, 20 (71%) were WHO Grade 2 and 8 (29%) were WHO Grade 3. In the subtotally resected group, 11 (85%) tumors were WHO Grade 2 and 2 (15%) were WHO Grade 3. The clinical and immunohistochemical data are summarized in Tables 1 and 2.

TABLE 1.

SUMMARY OF CLINICAL DATA FOR ALL PATIENTS

n=41	Supratentorial tumors 11 (27%)	Infratentorial tumors 30 (73%)
WHO Grade 2	5 (45%)	26 (87%)
WHO Grade 3	6 (55%)	4 (13%)
Invasion identified	7 (64%)	7 (23%)
High ezrin expression	4 (36%)	11 (37%)
High bcl-2 expression	5 (45%)	23 (77%)
High MMP2 expression	4 (36%)	4 (13%)
High MMP14 expression	5 (45%)	22 (73%)
Median age at diagnosis 4.2 years (0.7-16.8)	5.5 years	3.1 years
Sex	5 males, 6 females	15 males, 15 females
Gross total resection	9 (82%) (4 Grade 2, 5 Grade 3)	19 (63%) (16 Grade 2, 3 Grade 3)
Subtotal resection	2 (18%) (1 Grade 2,1 Grade 3)	11 (37%) (10 Grade 2,1 Grade 3)
Radiation	10 (91%) (4 Grade 2, 6 Grade 3)	27 (90%) (23 Grade 2,4 Grade 3)
Chemotherapy	6 (55%) (3 Grade 2, 3 Grade 3)	20 (67%) (17 Grade 2, 3 Grade 3)
Recurrence/Progression	8 (73%)	22 (73%)
Progression free survival (months)	3-160 (median, 42)	0.5-141 (median, 22.5)
Dead	6 (55%)	18 (60%)
Overall survival (months)	27-160 (median, 84)	7-141 (median 54)

Table 2.

Patients with Gross Total Resection

	n=28
Clinical Outcome
No relapse, alive	10 (36%)
Relapsed, alive	5 (18%)
Relapsed, dead	13 (46%)
Invasion
Absent	20 (71%)
Present	8 (29%)
Outcome and Expression of Metalloproteinases
Dead and High MMP14	12/13 (92%)
Alive and Low MMP14	10/15 (67%)
Dead and High MMP2	6/13 (46%)
Alive and Low MMP2	14/15 (93%)

Microinvasion is associated with poor prognosis in WHO Grade II and Grade III ependymomas

Most of the tumors had well demarcated margins with brain (Figure 1A, 1B), which was also highlighted by ezrin (Figure 1D). The combination of H&E (Figure 1C) and neurofilament immunohistochemistry (Figure 1G) identified 14 (34%) tumors with microinvasion. Of the 14 patients with identified invasion, all patients relapsed and of the eight who were treated with gross total resection, 7 died of disease. Of the twenty patients who underwent gross total resection and showed no evidence of invasion, fourteen survived; ten never relapsed and four relapsed but remained stable after additional treatment. Six patients with no identified invasion relapsed and died. All six patients with subtotally resected tumors and identified invasion relapsed and died by the end of study. Two survivors from the patients who underwent a subtotal resection showed no evidence of invasion; one relapsed and survived and the other never relapsed.

Invasion was a significant predictor of poor overall survival (p=0.002) and progression free survival (p=0.03) for the entire cohort irrespective of the extent of resection (Figure 2 and Table 3) as well as separately for patients who underwent gross total resection (overall survival p=0.003; progression free survival p= 0.03, Figure 3 and Table 3).

Figure 2.

In the entire cohort of 41 patients, invasion is associated with decreased progression free survival (p=0.03) as well as overall survival (p=0.002). Both progression free survival and overall survival are significantly decreased with high expression (score 1) of MMP14 (p=0.04 for both variables). The same trend exists for MMP2 although is not significant (p=0.09 for progression free survival and p=0.08 for overall survivel).

TABLE 3.

ALL PATIENTS:

Overall Survival				Progression Free Survival
Variables	Median	95%CI	LRT p-value	Variables	Median	95%CI	LRT p-value
STR	54	48,NA	0.04*	STR	22	13,48	0.04*
GTR	NA	48,NA		GTR	129	22,NA
MMP14=0	NA	102,NA	0.04*	MMP14=0	42	22,NA	0.04*
MMP14=1	NA	48,NA		MMP14=1	22	15,48
No invasion	NA	78,NA	0.002*	No invasion	36	17,NA	0.03*
Invasion	48	42,NA		Invasion	19.5	16,NA

PATIENTS WITH GROSS TOTAL RESECTION:
Overall Survival				Progression Free Survival
Variables	Median	95%CI	LRT p-value	Variables	Median	95%CI	LRT p-value
MMP2=0	NA	NA,NA	0.02*	MMP2=0	52	17,NA	0.07
MMP2=1	48	42,NA		MMP2=1	22	7,NA
MMP14=0	NA	NA,NA	0.002*	MMP14=0	NA	17,NA	0.07
MMP14=1	48	36,NA		MMP14=1	29	13,NA
No invasion	NA	141,NA	0.003*	No invasion	141	17,NA	0.03*
Invasion	48	33,NA		Invasion	19.5	16,NA

LRT: log-rank test; CI: confidence interval; GTR: gross total resection; STR: subtotal resection; OAS: overall survival; PFS: progression free survival; NA: Quantity not estimable with the current data;

^*p-value<0.05

Figure 3.

In patients with gross totally resected tumors, progression free survival is decreased with invasion (p=0.03), and is decreased, although not significantly, with high expression of MMP2 and MMP14 (score 1) (p=0.07 for both comparisons). Overall survival is significantly decreased with invasion (p=0.003) and high expression of MMP2 and MMP14 (score 1), (p=0.02 and 0.002, respectively).

As per previous reports, analysis of our data showed that the extent of surgical resection was the only significant clinical variable influencing survival. Both progression-free survival and overall survival were significantly different between gross total resection and subtotal resection groups (p value 0.04, for both comparisons, Table 3). Tumor histological grade (WHO), location, age and gender were not significant prognostic indicators.

Ezrin as an immunohistochemical marker for microinvasive ependymomas

All tumors in our cohort expressed ezrin and 15 (37%) tumors showed strong, diffuse staining in almost all of the cells (score=1, Table 1). In all positive tumor cells, ezrin staining had a diffuse cytoplasmic pattern. Invasive cells showed the same staining pattern as the solid tumor mass (see Figures 1D, 1F). The intensity of staining did not differ among positive cells in individual tumor samples. In adjacent brain, faint staining was present in astrocytes while oligodendrocytes and neurons were negative. However, peritumoral reactive astrocytes expressed ezrin strongly (not shown). Ezrin expression of invading tumor cells was helpful in identifying single tumor cells in the white matter, which are difficult to identify on H&E (Figure 1E). Interestingly, Grade II ependymomas showed stronger expression of ezrin than Grade III tumors (data not shown).

Thirty-seven tumors (90%) expressed bcl-2, with 28 (68%) of the tumors having score=1 (Table 1); however expression of bcl-2 did not demonstrate a strong prognostic significance. The intensity of bcl-2 staining did not differ significantly among positive cells in individual tumor samples (Figure 1J).

Seven tumors with well-demarcated margins on H&E had atypical cells deep in the white matter that were identifiable as infiltrating tumor cells by ezrin and bcl-2. Invading tumor cells highlighted by ezrin were oval to slightly elongate without multiple processes (Figure 1E and 1F) which distinguished them from reactive astrocytes which had typical star-like morphology. MMP2 and MMP14 were not useful as markers for the invading tumor cells because the staining was generally too faint.

MMP14 and MMP2 expression in Grade II and Grade III ependymomas is associated with poor prognosis

Eighteen tumors (44%) expressed some MMP2 and 8 (20%) tumors had a score=1 (Figure 1H, Table 1). Thirty-one tumors (76%) expressed some MMP14 and 27 (66%) of the tumors had score=1 (Figure 1I, Table 1). The intensity of MMP2 and MMP14 staining did not differ significantly among positive cells in individual tumor samples.

Of the 28 patients who underwent gross total resection, 12/13 (92%) who died of disease had high MMP14 (score=1) and 10/15 (67%) who survived had low levels of MMP14 (score=0, Figure 1L). For MMP2, 6/13 (46%) patients who died had high expression, however, 14/15 (93%) patients who survived had low expression of MMP2 (score=0, Figure 1K). Data for patients who underwent gross total resection are summarized in Table 2. Results of the logrank test showed that patients with low expression of MMP14 (score=0) had significantly longer overall survival and progression free survival than those with high levels, irrespective of the extent of resection (p=0.04 for both comparisons, Figure 2 and Table 3). The prognostic value of MMP14 was confirmed by multivariate Cox Proportional Hazards model for overall survival (Hazard Ratio = 9.012, 95% CI = 1.04, 78.29) although not for progression free survival (Table 4).

TABLE 4.

Univariate proportional hazards model for Invasion and Grade

Overall Survival			Progression Free Survival
MODEL	Hazard Ratio	p-value	MODEL	Hazard Ratio	p-value
Invasion	5.18	0.0061*	Invasion	2.83	0.035*
Grade	0.615	0.460	Grade	0.332	0.083
Invasion +	4.97	0.0082*	Invasion +	2.469	0.130
Grade	0.785	0.720	Grade	0.379	0.068

Multivariate Cox Proportional Hazards model for MMP14
Overall Survival			Progression Free Survival
MMP14 score=1	9.012 [78.29, 1.04]	0.046*	MMP14 score=1	2.22 [0.62,7.91]	0.22

^*p-value<0.05

Among the gross total resection group, patients with low expression of MMP2 and MMP14 (score=0 for both markers), had significantly longer overall survival (p = 0.02 and 0.002, respectively). The same trend existed for progression free survival but was not significant (p=0.07 for both comparisons; Table 3 and Figure 3). All other variables were not associated with survival by univariate analysis.

To compare the prognostic value of invasion with grade, we evaluated tumors defined as invasive by H&E and/or neurofilament immunohistochemistry irrespective of their other immunohistochemical markers (not including cases in which invasive cells were seen only with ezrin immunohistochemistry). Proven invasion was a significant univariate predictor of overall survival (Hazard Ratio = 5.18, 95% CI = 1.59, 16.71) and progression free survival (Hazard Ratio = 2.83, 95% CI = 1.07, 7.45) while tumor grade was not. Adjusting for tumor grade, invasion remained a significant predictor of overall survival (adjusted Hazard Ratio = 4.97, 95% CI = 1.51, 16.34) (Table 4).

The two-class supervised latent class model separated patients well by survival, although estimates were highly variable in general. The model confirmed that both MMP2 and MMP14 were predictors of prognosis; with >5% cells positive for MMP2 and MMP14 (score=1) predicting poor overall and progression free survival in gross totally resected tumors, and overall survival independently of resection (Figure 4). More specifically, the probability of falling into the poor prognosis latent class given that these patients expressed either marker was nearly equal to 1. MMP14 was also a predictor of rapid progression independent of degree of resection, although the variability resulting from this model fit was larger. Although the latent class model results should not be over interpreted since the high variability and limited sample size, it confirms that MMP14 is the strongest predictor of poor overall and progression free survival in ependymoma patients with or without considering the extent of resection.

Figure 4.

The supervised latent class model shows that the expression of MMP14 and MMP2 are predictors of poor overall survival both for patients with gross totally resected tumors (A) as well as for the entire cohort (B). The model also designates high expression of ezrin as a marker of good overall survival in patients with gross totally resected tumors, and bcl-2 expression a marker of poor overall survival, although the prognostic value of these markers is less clear.

Results of the latent class model for ezrin were less clear. The presence of strong ezrin expression in >90% of the tumor cells (score=1) predicted good overall survival in patients who underwent gross total resection, but poor overall survival in the combined cohort (Figure 4). Results for progression free survival showed the opposite findings, however the results were not as stable as for overall survival, and should be interpreted with caution (data not shown). In all latent class analyses considered, bcl-2 was a predictor, although weak, of poor prognosis (Figure 4).

DISCUSSION

We tested the hypothesis that a subset of pediatric intracranial ependymomas invade surrounding brain tissue and, as a result, recur despite gross total resection. Although the tumor grade based on WHO criteria remains a major independent prognostic factor in many reports, our study did not show a significant difference in clinical outcomes between Grade 2 and 3 tumors. A recent study analyzed the value of histologic data and clinical variables in 96 patients enrolled in the Children's Oncology Group (18) and underlined the prognostic importance of tumor grade for event-free survival, although not for overall survival. This extensive review also pointed out intrinsic problems such as interobserver variability and the importance of establishing well-defined histological criteria for precise grading. Gross total resection of ependymoma represents the single most important prognostic clinical factor, although infratentorial location as well as younger age are associated with worse outcome in some (18, 19, 20) but not all (16) studies. These observations might be due to the fact that posterior fossa tumors are surgically challenging and gross total resection is more difficult to achieve than in supratentorial tumors, as was shown here. Posterior fossa ependymomas are more prevalent in younger children and, in addition to the surgical considerations mentioned above, long term effects of adjuvant treatment in children must be taken into account. In addition, a wide variety of adjuvant treatment protocols have been used for these tumors during the last 20 years making it difficult to reach firm conclusions from clinical data in relatively small series of cases.

Although patients with ependymomas are known to develop tumor recurrence after a variably long period of time, our data emphasize that a large proportion of these cases recur relatively early despite the gross total resection, often resulting in a fatal outcome. Therefore, additional prognostic markers are urgently needed to help guide treatment in the early postoperative period.

Our study shows that invasion, seen either on routine H&E or by demonstrating entrapped neuron processes by neurofilament immunohistochemistry, is a significant independent prognostic factor of adverse outcome. Evaluation of tumor both for invasion as well as for grading is significantly influenced by sampling and by the quality and amount of the tissue received at surgery. CUSA contents are difficult to interpret reliably for invasion and even well preserved fragments show good tumor-brain border only in a minority of the cases, more often in supratentorial than infratentorial tumors. In our study, although the absence of invasion did not exclude invasive behavior, its presence was strongly associated with both poor progression-free survival and overall survival. This shows the potential importance of additional markers for identifying tumors that might be locally invasive.

We found metalloproteinases MMP14 and MMP2 are good prognostic markers of pediatric ependymomas and can help to identify aggressive tumors that might require adjuvant treatment. Ezrin is especially helpful in identifying deeply infiltrating ependymoma cells not apparent on H&E although the exact predictive value of ezrin expression itself remains to be determined.

In evaluation of MMP2 and MMP14, only tumors that were negative or had few positive cells (<5%) belonged in the low-risk subgroup, while >5% of positive cells upgraded a tumor into the high-risk subgroup. This approach was chosen for practical purposes as evaluation of these markers into negative or almost negative (score = 0) versus positive (score = 1) categories may be easily translated to larger studies and to clinical practice, and avoids problems with counting of cells and establishing cut off points. However, the variability in staining, even in diffusely positive tumors with high percentage of positive cells, might be a confounding factor. Simultaneous use of multiple markers; one or both metalloproteinases with ezrin, addresses these issues as a useful approach to identifying aggressive infiltrating tumors. Furthermore, although we found both MMP14 and MMP2 useful, other members of the metalloproteinase family might show stronger correlation with invasion and even better prognostic value particularly for progression free survival, where our findings were not as strong as for overall survival.

Difficulties in determining histologic grade and the lack of consensus regarding other prognostic factors plague current diagnosis and treatment of ependymomas (21). The proliferation marker MIB-1 and p53 expression are used with some predictive success (15, 16, 22), although the threshold for these is not clearly established. One reason for the discrepancy between histological grade and prognosis in ependymomas may be tumor microinvasion, undetectable by neuroimaging or by gross examination at the time of surgery.

Gliomas migrate preferentially via white matter tracts (23, 24, 25) and metalloproteinases' role in migration of glioma cells along white matter track has been attributed to modifications in cell adhesion and metalloproteinase-induced destruction of the extracellular matrix (ECM) (26, 27).

MMP14 (also known as MT1-MMP) has three major roles that appear to account for much of its activity: cleaving adhesion membrane proteins, degrading ECM and activating soluble MMPs. MMP14 was shown to play a major role in the remodeling of non-permissive protein substrates seen in white matter tracts, thus making migration of glial cells possible (28). MMP14 activates other MMPs as well, particularly MMP2 (29). Several studies have shown that metalloproteinases MMP2 and MMP9 are not expressed in normal brain, and their expression is significantly elevated with increased grade of astrocytomas, and inversely correlated with survival (2, 3). Kunishio et al. (30) did not find a correlation between expression of metalloproteinases and survival, but showed that MMP2 expression was associated with a more invasive behavior in astrocytic tumors. In some gliomas MMP2 and MMP9 stained predominantly vascular structures suggesting that MMP2 and MMP9 might also play a role in both neoangiogenesis and invasion (4, 31, 32).

The expression of bcl-2 has been studied in several brain tumors (15, 16, 33, 34, 35), however only a few studies (15, 16) have examined its expression in ependymomas. The prognostic relevance of bcl-2 is not clear. Although one study (15) did not find prognostic value of bcl-2, only 4 high-grade cases were examined. In contrast, Zamecnik et al (16) evaluated a larger series of tumors (11 low grade, 20 high-grade tumors) and identified bcl-2 as a useful marker identifying particularly aggressive pediatric ependymomas. Bcl-2 has also been implicated in the regulation of intercellular interactions (36) and adhesion, specifically via R-ras integrin dependent regulation (37). Its role in migration and correlation with MMP activity was also suggested (13, 14). Bcl-2 might influence migration in at least two ways: either by down-regulation of TIMP2 or directly by up-regulation of MMP2, MMP9, MMP3 and MMP12 (13, 38). In fact, antibodies against αvβ3integrin down regulate bcl-2, leading to induction of apoptosis (39) and inhibition of invasion and migration (14).

Ezrin is a member of ezrin-radixin-moesin (ERM) family (40); physiologically accumulated in actin-rich parts of cells binding cytoskeleton to membranes, it plays an important role in cell-cell and cell-matrix interactions (41). Ezrin is expressed in multiple cell types within the central nervous system, namely ependyma, choroid plexus, and astrocytes while it is absent in oligodendrocytes (11). Wick et al (10) showed a functional connection between ezrin, bcl-2 and Transforming Growth Factor-β2 (TGF-β2) in glioma cell migration. Ezrin has been shown to play a role in the malignant phenotype of different tumors, although paradoxically, via opposing pathways. In colorectal carcinoma ezrin inhibition was shown to reduce cell adhesion, increase invasiveness and motility (41). In contrast, in transformed malignant fibroblasts up-regulated ezrin expression is associated with increased malignancy and inhibition of ezrin with decreased tumor invasiveness (42). Similarly, ezrin expression in soft tissue sarcomas (5), osteosarcomas (6, 7) and endometrioid carcinomas (8), is associated with poor prognosis while in serous ovarian carcinomas, with better prognosis (9). In ependymomas, higher expression of ezrin is noted in lower-grade tumors (WHO grade I and II) as compared to high-grade ependymomas (WHO grade III) (11). Our study confirmed these observations in that we also find higher expression of ezrin in low-grade ependymomas (grade II).

This study focused on the prognostic significance of microinvasion in ependymomas and evaluation of immunohistochemical prognostic factors that are associated with ependymal invasion. Our patients were stratified based on extent of resection, because recurrence in those with gross totally resected tumor is more likely related to tumor microinvasion. By contrast, recurrence in those with subtotally resected tumors is confounded by the extent of the residual tumor and proliferation rate. Another reason for the use of degree of resection rather than tumor grade results from the high intraobserver variability in grading of ependymomas as documented in many studies (16, 18) influencing histology, immunohistochemistry and clinical outcome. The importance of stratification of ependymoma patients based on extent of resection was previously shown (16).

Our study suggests that aggressive ependymomas infiltrate brain tissue beyond their grossly demarcated borders via white matter tracts, without significant destruction of extracellular matrix. Although the total number of patients in this study is small, our findings suggest that the infiltrative behavior in ependymomas can have a significant impact on the success of surgery and long term outcome of patients.