Cells with intense EGFR staining and a high nuclear to cytoplasmic ratio are specific for infiltrative glioma: a useful marker in neuropathological practice

Abstract

Background

The differential diagnosis between infiltrative glioma (IG) and benign or curable glial lesions, such as gliosis, pilocytic astrocytoma, dysembryoplastic neuroepithelial tumor, ganglioglioma, or demyelinating disease, may be challenging for the pathologist because specific markers are lacking. Recently, we described a strong EGFR immunolabelling pattern in cells with a high nuclear to cytoplasmic ratio that enables the discrimination of low-grade IG from gliosis. The aim of this study was to extend our observation to high-grade glioma to assess whether EGFR expression pattern is of value in the discrimination of all IG from noninfiltrative glial lesions (NIG), including gliosis, benign tumors, and demyelinating disease.

Methods

One hundred one IG and 58 NIG were compared for immunohistochemical expression of EGFR with use of an antibody that recognizes an epitope in the extracellular domain of both EGFRwt and EGFRvIII. Highly EGFR-positive cells with a high nuclear to cytoplasmic ratio were isolated and further characterized.

Results

Cells with intense EGFR staining and a high nuclear to cytoplasmic ratio were significantly associated with the diagnosis of IG (P < .0001). The sensitivity and specificity of this staining pattern for the diagnosis of IG were 95% and 100%, respectively. EGFR expression was independent of IDH1 mutations and EGFR amplification. Finally, we showed that these particular cells displayed the phenotype and properties of glial progenitors and coexpressed CXCR4, a marker of invasiveness.

Conclusions

We demonstrate that cells with intense EGFR staining and a high nuclear to cytoplasmic ratio are specific criteria for the diagnosis of IG, irrespective of grade, histological subtype, and progression pathway, and their identification represents a tool to discriminate IG from benign or curable glial lesions.

Infiltrative gliomas (IG), including astrocytomas, oligodendrogliomas, and glioblastomas, are the most common primary brain tumors in humans.¹ IG display a poor prognosis, especially glioblastomas, which represent the most aggressive subtype.¹ The tumors of this group are characterized by a pattern of infiltrative growth into the surrounding normal brain,² making complete surgical excision very difficult. Recurrence or progression after surgery almost always occurs despite adjuvant treatment with radiotherapy and/or chemotherapy.¹

Morphological patterns of IG are often heterogeneous, and the diagnosis may be challenging for pathologists.² Glial tumor cells may also be difficult to distinguish from reactive cells on the basis of morphology alone. Reactive astrocytes may become very large and may even display pleomorphic nuclei resembling neoplastic tumor cells.³ The differential diagnosis between IG and gliosis can thus be difficult, especially in small biopsy samples.⁴ IG can also be difficult to differentiate from benign and/or curable noninfiltrative glial tumors, including pilocytic astrocytomas, gangliogliomas, and dysembryoplastic neuroepithelial tumors,^1,2 and from demyelinating disease.⁵ It is therefore very important for pathologists to develop histological markers to identify infiltrating glioma cells. To date, no ideal marker has been identified for this purpose. In practice, the most useful markers are MIB1/Ki67^6,7 and p53,^8,9 but these lack sensitivity and specificity.¹⁰ More recently, the mutated R132H form of isocitrate dehydrogenase 1 (IDH1), which can be specifically detected by immunohistochemistry,^11,12 has been shown to be a good marker of grade II and grade III gliomas and of secondary glioblastomas.^13–15 However IDH1 (R132H) is rare in primary glioblastomas¹⁶ and is not useful for the diagnosis of this subtype, which represents the most common IG. A marker specific for infiltrating cells, which are characteristic of IG regardless of grade or histological subtype, would be of value.

Recently, we showed that the EGFR immunolabelling pattern can discriminate low-grade glioma from gliosis.¹⁷ The criterion that characterized low-grade gliomas was strong EGFR immunostaining in cells with a high nuclear to cytoplasmic ratio. EGFR overexpression is frequent in gliomas. In 40% of glioblastomas, EGFR overexpression is secondary to gene amplification and is often associated with expression of EGFRvIII, a constitutively activated mutated variant.¹⁸ Conversely, EGFR amplification is rare in low-grade gliomas,^18–21 although protein overexpression has been detected with a frequency ranging from 11.5% to 100% of cases in the literature.^{19,20,22–24} The mechanism of this overexpression in low-grade IG remains unknown. Several ligands, including EGF and TGFα, may activate EGFR. The activation of EGFR is involved in several processes, including cell survival, differentiation, proliferation, and migration.²⁵ Because EGFR has been shown to influence cell migration during the development of the central nervous system^26–28 and in gliomas,^29–31 we postulated that EGFR could be a marker of migrating cells, specific for IG. The aim of the present study was to assess whether elevated EGFR expression in cells with a high nuclear to cytoplasmic ratio, as we previously observed in low-grade glioma,¹⁷ may be a valuable criterion to discriminate infiltrative gliomas of any grade or histological subtype from noninfiltrative glial lesions. We also sought to further characterize these strongly EGFR-positive cells.

Materials and Methods

Tissue Collection

This retrospective study comprised a total of 159 human glioma tissue samples and nonneoplastic cerebral tissue samples selected from the database of the Departments of Pathology of Nice and Montpellier (Supplementary Materials).

Immunohistochemistry

EGFR immunohistochemistry was performed on paraffin-embedded tumor sections with the use of an anti-mouse monoclonal antibody (clone 2-18C9, Dako EGFR pharmDX Kit K1494; Carpinteria) as previously described.¹⁷ Clone 2-18C9 recognizes an epitope in the extracellular domain and has been found to recognize both EGFRwt and EGFRvIII forms. The evaluation of staining intensity was performed using the same controls as previously described¹⁷ (Supplementary Materials and Methods). IDH1 mutational status was determined using immunohistochemistry with an antibody specific for the R132H mutant of IDH1 (clone H09, Abnova, 1/100). Deparaffinization, rehydration, and antigen retrieval were performed using the pretreatment module PTlink (Dako). Double-immunolabelling EGFR/Mib1 on paraffin-embedded tumor sections was performed as previously described.¹⁷

Measurement of Nuclear to Cytoplasmic Ratio

In 7 cases of IG (3 glioblastoma, 2 oligodendroglioma grade II, 1 astrocytoma grade II, and 1 oligodendroglioma grade III), paraffin-embedded sections immunolabelled with EGFR were scanned using the Slide Scanner Leica SCN400. For each case, nuclear and cytoplasmic areas of strongly EGFR-positive cells were measured using the software Leica SlidePath Gateway. In each case, these measurements were made both on the population of small undifferentiated cells displaying morphological criteria previously described¹⁷ and on a population of more differentiated cells (astrocytic or oligodendroglial).

Immunofluorescence

Sorted cells were seeded on polylysine-coated glass slides and subjected to immunostaining using anti-EGFR at 1/100 (ab24293, mouse monoclonal, Abcam), anti-Oct4 at 1/50 (H-134, sc-9081, rabbit polyclonal, Santa-Cruz), anti-Sox1 at 1/50 (AB15766, rabbit polyclonal, Millipore), anti-Sox2 at 1/50 (sc-20088, rabbit polyclonal, Santa-Cruz), and anti-A2B5 at 1/100 (cl A2B5-105, mouse monoclonal, Millipore).

For double labelling, sections were incubated with mouse anti-EGFR antibody (clone 2-18C9, Dako EGFR pharmDX Kit K1494) or rabbit anti-EGFR (ab2430, Abcam) and either a goat polyclonal anti-Olig2 (R&D Systems, 1/50), a rabbit polyclonal anti-GFAP (Dako, 1/500), or a mouse monoclonal anti-Nestin (MAB353, Chemicon) antibody.

Slides were counterstained with 4′,6′-diamidino-2-phenylindole (DAPI). Microscopic analysis was performed using a Nikon eclipse Ti microscope (Nikon, Champigny sur Marne, France) (Supplementary Materials).

Fluorescent In Situ Hybridization (FISH)

EGFR gene copy number per cell was investigated by FISH performed on 5-µm sections cut from formalin-fixed, paraffin-embedded gliomas (Supplementary Materials).

Fluorescence-Activated Cell Sorting (FACS) and Flow Cytometry

After enzymatic and mechanical dissociation, cells from fresh GBM (n = 5) were stained with rabbit antihuman EGFR (ab2430, Abcam), followed by FITC-anti-rabbit IgG (BD Biosciences) and with PE-Cy5 mouse anti-human CXCR4/CD184 (BD Biosciences). Small, highly positive EGFR-expressing cells were sorted on a FacsAriaI cytometer (BD) equipped with 3 lasers (488 nm, 405 nm, 633 nm) (Supplementary Materials and Methods).

Cell Culture

Sorted cells were grown in NS34+ medium containing EGF and bFGF, specific for neural stem cells (DMEM-F12 1/1, glutamine 10 mM, Hepes 10 mM, Sodium bicarbonate 0.025%, N2, G5 and B27).

Statistical Analysis

The association of the criterion of strongly stained cells showing a high nuclear to cytoplasmic ratio with the diagnosis of IG was studied using the χ² test. The analysis was adjusted according to patient age using the Mantel-Haenszel test. Lastly, the diagnostic efficacy of this criterion to identify IG was evaluated by calculating its sensitivity, specificity, and positive and negative predictive value, both for the whole series and for adults and children separately. The staining intensity between infiltrative and solid areas of IG was compared using χ² test. Mean values of nuclear to cytoplasmic ratio were compared using the t test. All the tests were considered to be statistically significant at a 5% type I error rate (P < .05). Statistical analyses were performed using SPSS, version 11.0 (SPSS; Chicago, IL).

Results

Strongly EGFR-Positive Cells with a High Nuclear to Cytoplasmic Ratio Are Specific for Infiltrative Gliomas

In a previous study,¹⁷ we identified EGFR staining criteria specific for tumor cells, which allowed the discrimination of low-grade IG from gliosis—namely, strong EGFR staining in cells that displayed a high nuclear to cytoplasmic ratio. To assess whether these criteria could faithfully discriminate IG, regardless of grade or histological subtype, from noninfiltrative glial lesions, we compared EGFR immunostaining between 101 IG and 58 noninfiltrative lesions (Table 1). EGFR immunolabelling was detected in 99 of 101 cases of IG. The intensity of staining, often heterogeneous within the same tumor (Fig. 1), was predominantly weak in 15 cases (14.9%), moderate in 14 cases (13.9%), and strong in 70 cases (69.2%). The proportion of positive cells varied from 0% to 100% (Table 2). Nearly all tumors (96/101) contained at least some strongly EGFR-positive cells with a high nuclear to cytoplasmic ratio, including the tumors with a predominantly weak level of EGFR expression (Fig. 1). These cells were poorly differentiated, consistently characterized by scant cytoplasm, with or without a small unipolar process and without conspicuous astrocytic or oligodendroglial differentiation (Fig. 1). The nuclear to cytoplasmic ratio of these cells had a mean value of 0.58 ± 0.1 (n = 300). This was significantly higher (P > .0001) than the mean value of 0.19 ± 0.06 observed in more differentiated (astrocytic or oligodendroglial) cells, which were strongly EGFR positive (n = 300) and which were identified in the same tumors (Supplemental Fig. S1).

Table 1.

Patients and lesions

	Non infiltrative lesions	Infiltrative gliomas
Children	12 PA
	6 DNET	2 oligo II
	3 gangliogliomas	4 astro II
	1 gliosis	3 astro III
Adults	1 PA	20 oligo II
	2 DNET	13 oligo III
	1 ganglioglioma	15 astro II
	24 gliosis	4 astro III
	5 PML	1 oligo-astro II
	2 ADEM	2 oligo-astro III
	1 multiple sclerosis	37 GBM

Abbreviations: PA, pilocytic astrocytoma; DNET, dysembryoplastic neuroepithelial tumor; PML, progressive multifocal leukoencephalopathy; ADEM, acute disseminated encephalomyelitis; oligo, oligodendroglioma; astro, astrocytoma; oligo-astro, oligo-astrocytoma; GBM, glioblastoma; II, WHO grade II; III, WHO grade III.

Fig. 1.

EGFR immunostaining in infiltrative glioma (×400). Diffuse and strong EGFR expression in a glioblastoma (A). Strongly EGFR-stained cells with a high nucleus to cytoplasm ratio and an apolar or unipolar shape in the invasive edge of a glioblastoma (B), in an anaplastic astrocytoma (C), in an anaplastic oligo-astrocytoma (D), and in a low-grade astrocytoma (E). Low-grade oligodendroglioma showing few cells with strong EGFR staining and scant cytoplasm (arrow) intermingled with positive and negative typical tumor cells with a fried-egg appearance (F).

Table 2.

Proportion of EGFR-positive cells in infiltrative gliomas

	Proportion of EGFR positive cells
	0	≤25%	26%– 50%	51%– 75%	>75%
Oligo II (n = 22)		7	6	2	7
Oligo III (n = 13)		2	4	1	6
Astro II (n = 19)	1	7	7	1	3
Astro III (n = 7)	1	1	1	2	2
OA II (n = 1)					1
OA III (n = 2)					2
GBM (n = 37)		11	7	2	17

Abbreviations: Oligo, oligodendroglioma; Astro, astrocytoma; OA, oligo-astrocytoma; GBM, glioblastoma; II, WHO grade II; III, WHO grade III.

The proportion of the poorly differentiated strongly EGFR-positive cells was extremely variable across the series, with a mean of 37.8% (range, 0.3%–100%). Although EGFR staining was detectable in some cases of non-infiltrative lesions (31/58), the proportion of stained cells was low (≤25%) in most cases (Table 3), the staining intensity was weak, and the positive cells displayed a stellate, spindle, or gemistocytic morphology, suggestive of differentiated astrocytes (Fig. 2). Conversely to IG, no cells displaying strong EGFR staining with a high nuclear to cytoplasmic ratio were observed in noninfiltrative lesions.

Table 3.

Proportion of EGFR-positive cells in noninfiltrative lesions

	Proportion of EGFR positive cells
	0	≤25%	26 to 50%	>50%
Gliosis (n = 25)	2	13	5	5
PA (n = 13)	10	2		1
DNET (n = 8)	6	1	1
Ganglioglioma (n = 4)	4
PML (n = 5)	4	1
ADEM (n = 2)	2
Multiple sclerosis (n = 1)	1

Abbreviations: PA, pilocytic astrocytoma; DNET, dysembryoplastic neuroepithelial tumor; PML, progressive multifocal leukoencephalopathy; ADEM, acute disseminated encephalomyelitis.

Fig. 2.

EGFR immunostaining in noninfiltrative lesions. Absence of EGFR expression (A) in a pilocytic astrocytoma and (B) in a dysembryoplastic neuroepithelial tumor (×400). Oligodendroglioma-like pattern entirely immunonegative for EGFR (C) in a pilocytic astrocytoma (×200) and (D) in a dysembryoplastic neuroepithelial tumor (×400). (E) Weak EGFR expression in stellate astrocytes in a case of gliosis (×400). (F) EGFR-negative atypical cells in a pilocytic astrocytoma (×400). (G and H) Weak EGFR expression in large bizarre astrocytes in a case of progressive multifocal leukoencephalopathy (×400).

The specificity and sensitivity of the pattern of strong EGFR immunostaining in the context of a high nuclear to cytoplasmic ratio for the diagnosis of IG were 100% and 95%, respectively. Epidemiological data indicate that the diagnosis of IG as opposed to noninfiltrative lesions is correlated with patient age (P < .0001); therefore, we verified that patient age was not a confounding factor in our study. After adjustment for age, the association between this pattern of strong EGFR immunostaining and diagnosis remained significant (P < .0001). In adults, the specificity and sensitivity of this pattern of strong EGFR immunostaining for IG were 100% and 98.9%, respectively (Table 4). The positive predictive value and negative predictive value were 100% and 97.3%, respectively. In children, the specificity and sensitivity were 100% and 55.6%, respectively, with positive predictive value and negative predictive value of 100% and 84.6%, respectively.

Table 4.

EGFR status according to diagnosis and age of patients

		Infiltrative gliomas	Non infiltrative lesions
Adults	EGFR+	91 (98,9%)	0
	EGFR-	1 (1,1%)	36 (100%)
Children	EGFR+	5 (55,6%)	0
	EGFR-	4 (44,4%)	22 (100%)

EGFR status was defined by the presence (+) or not (−) of strongly EGFR-positive cells with a high nucleus-cytoplasm ratio.

Taken together, our observations, strengthened by a statistical analysis, show that the criterion of strong EGFR staining in cells with a high nuclear to cytoplasmic ratio is significantly and highly restricted to the diagnosis of IG (P < .0001), regardless of patient age.

EGFR Overexpression in Infiltrative Glioma Is Independent of EGFR Amplification and of IDH1 Mutation Pathway

EGFR amplification is classically associated with primary glioblastomas. Because strongly positive EGFR-expressing cells with a high nuclear to cytoplasmic ratio were constant in nearly all IG regardless of grade or histological subtype, we hypothesized that EGFR amplification was probably not required for the overexpression of EGFR in these cells. To test this hypothesis, we chose to analyze EGFR copy number by FISH in a series of 33 IG, including numerous cases (n = 32) of low-grade and/or oligodendroglial glioma, which are tumors unlikely to have EGFR gene amplification. EGFR amplification and EGFR gain (ratio EGFR/centromere >1.5) were detected in only 2 cases (1 glioblastoma and 1 high-grade oligoastrocytoma, respectively) of 33 cases (Fig. 3), whereas cells with intense EGFR staining and a high nuclear to cytoplasmic ratio were present in 31 cases (Supplementary Table). This result demonstrated that, in IG, EGFR amplification was not required for strong EGFR expression in cells with a high nuclear to cytoplasmic ratio.

Fig. 3.

Correlation between EGFR expression, EGFR copy number, and chromosome 7 number in infiltrative gliomas. This experience was performed using EGFR immunohistochemistry (×400) and FISH analysis of EGFR gene (red) and centromere of chromosome 7 (green) (×1000). (A) Glioblastoma showing numerous strongly EGFR-stained cells and EGFR amplification. (B) High grade oligo-astrocytoma showing strongly EGFR-stained cells and gain of EGFR (ratio EGFR/centromere >1.5). In both cases, the picture was taken in an infiltrative area where they were more easily individualized. (C) Low-grade astrocytoma showing strongly EGFR-stained cells and normal FISH analysis. (D) Low grade oligodendroglioma showing strong EGFR-stained cells and polysomy 7. (E) Low grade oligo-astrocytoma showing strongly EGFR-stained cells and normal FISH analysis. (F) High-grade oligo-astrocytoma showing strongly EGFR-stained cells and normal FISH analysis. (G) Anaplastic astrocytoma showing polysomy 7 and no EGFR-stained cells.

IDH1 mutations, particularly IDH1 R132H, are involved in the progression of grade II/III gliomas and secondary glioblastomas, whereas they are rare in primary GBMs. To assess whether the presence of strongly EGFR-positive cells with a high nuclear to cytoplasmic ratio was dependent on the IDH1 mutation, we compared EGFR and IDH1 R132H expression by immunohistochemistry. As expected, IDH1 R132H was essentially found in grade II and grade III IG (45/64) and in only 1 glioblastoma (1/37). Cells displaying intense EGFR staining and a high nuclear to cytoplasmic ratio were observed in both IDH1 R132H–positive (46 EGFR+/46 IDH1 R132H+ cases) and IDH1 R132H–negative IG (50 EGFR+/55 IDH1 R132H- cases). Intensely stained EGFR-expressing cells were detected in all IDH1 R132H–negative GBMs, which could be considered as primary GBMs. The presence of these cells was therefore not correlated with IDH1 mutation and, therefore, enabled the discrimination of GBM, regardless of the progression pathway, from noninfiltrative glial lesions.

Strongly EGFR-Stained Cells with High Nuclear to Cytoplasmic Ratio Predominated in Infiltrative Areas of High-Grade Gliomas and Expressed the Invasion Marker CXCR4

High-grade gliomas are classically composed of a solid component of tumor cells and neovascularisation. At the periphery of this solid component, there is an invasive edge of isolated tumor cells that infiltrate the parenchymal tissue. We compared EGFR immunostaining between the solid and infiltrative areas in 30 high-grade gliomas from our series, for which both components were assessable (Fig. 4A). In all cases, the infiltrative area showed a predominant strong intensity of EGFR staining (30/30) in cells displaying high nuclear to cytoplasmic ratio, whereas comparable staining was present in the solid component of only 15 (50%) cases (P < .0001). These results indicate that, unlike the solid areas, the infiltrative areas were enriched in these particular cells displaying the typical morphology of high nuclear to cytoplasmic ratio and a strong EGFR staining. Of interest, we also demonstrated by flow cytometry and immunohistochemistry that EGFR was coexpressed with Ki67/Mib1 and the well-known invasiveness marker CXCR4 (Fig. 4B and C and Fig. 5C), suggesting that these particular cells represent proliferating or infiltrating cells.

Fig. 4.

(A) Comparison of EGFR expression between solid and infiltrative areas in high-grade glioma. In this glioblastoma, the solid area was negative for EGFR, whereas the infiltrative area was rich in small strongly stained cells (immunohistochemistry ×100). (B) Analysis of EGFR and CXCR4 expression in high-grade glioma by flow cytometry. Small EGFR overexpressing cells showed a high expression of CXCR4. (C) Double immunolabelling of EGFR (brown) and ki67/Mib1 (red) in the invasive edge of a glioblastoma (×400): some of the strongly positive EGFR cells coexpressed the proliferative marker ki67/Mib1 (arrow).

Fig. 5.

(A) Characterization of strongly EGFR-positive cells in infiltrative glioma tissue sections. Double-staining was performed on paraffin-embedded tissue section of infiltrative glioma using anti-EGFR antibody (green), and (a) anti-GFAP antibody (red) or (b) anti-olig2 antibody (red), or (c) anti-nestin antibody (red) (×400). Strongly EGFR-stained cells with a high nucleus to cytoplasm ratio lacked GFAP expression (a) and coexpressed olig2 (b) and nestin (c). (B and C) Characterization of EGFR-overexpressing cells sorted from fresh glioblastoma. Immunofluorescence performed immediately after FACS sorting of small EGFR-overexpressing cells (B) confirmed that cells overexpressed EGFR and showed that these cells expressed the neural stem cell markers oct4, sox1, sox2, and A2B5 and (C) expressed invasion marker CXCR4 and CD44 (×1000). (D) Culture of small EGFR overexpressing cells in a medium specific for neural stem cells (×1000). Cells slowly proliferated and formed small neurospheres (left) or remained nonadhesive quiescent isolated cells (right).

Strongly EGFR-Stained Cells Display Properties of Neoplastic Glial Precursor Cells

The peculiar morphology of the strongly EGFR-stained cells (scant cytoplasm and often a unique small process, without astrocytic or oligodendrocytic differentiation) and their presence in nearly all infiltrative gliomas, regardless of the histological subtype and grade, support the hypothesis that these cells could be glial precursor tumor cells.

First, we showed that small strongly stained EGFR-expressing cells in tumor tissue sections were negative for the astrocytic differentiation marker GFAP but were positive for the glial marker olig2, regardless of the histological subtype (Fig. 5A). Expression of olig2 confirmed the glial nature of the cells. Almost all of the strongly EGFR stained cells (88.23%) coexpressed nestin, a marker of neural stem cells (Fig. 5A).

Second, we sorted by FACS from fresh gliomas, small cells strongly prositive for EGFR. The sorted cells were then immediately fixed with methanol and characterized by immunofluorescence. Cells expressed EGFR, as expected, and several stem cell markers, including sox 1, sox 2, oct3/4, A2B5, and CD44 (Fig 5B and C).

Neurosphere formation is the hallmark of neural stem/progenitor cells in culture. To explore stem cell–like properties of the cells, fresh sorted cells were grown in a medium specific for neural stem cells. The cells slowly proliferated and formed small neurospheres (Fig. 5D). In addition, these cells were capable of self-renewal, because a single cell was able to lead to a novel neurosphere up to 7 successive times. Within the same population, certain cells were able to remain isolated, nonadhesive, and quiescent and survive for several weeks without proliferating (Fig. 5D). Taken together, these results suggest that the strongly EGFR-stained cells with a high nuclear to cytoplasmic ratio display glioma-initiating cell (GiC) properties.

Discussion

In our study, we identified a pattern of EGFR expression specific for infiltrative gliomas, defined by strongly stained cells with a high nuclear to cytoplasmic ratio. These strongly stained cells often exhibited a scant cytoplasm with or without a small unipolar process. In a previous study,¹⁷ we showed that this pattern of staining allows the discrimination of low-grade glioma from gliosis. Here, we demonstrated that this criterion is sufficient to discriminate IG, regardless of histological subtype and grade, from a large range of differential diagnoses including gliosis, benign gliomas (WHO grade I), and demyelinating diseases. This EGFR expression pattern displays 100% specificity for infiltrative gliomas. Furthermore, the sensitivity of this EGFR pattern for IG was very high in adults (98.9%), indicating that its absence could virtually exclude the diagnosis of infiltrative glioma. The sensitivity for IG was lower in the pediatric population (55.6%). The difference in sensitivity between adults and children could be attributable to the small number of pediatric infiltrative gliomas in our series. This result needs to be confirmed in larger series of cases. Nevertheless, pediatric infiltrative gliomas are known to be genetically different from adult gliomas, and in particular, EGFR overexpression is less common in childhood gliomas than in adult gliomas.^32,33

The differential diagnosis between infiltrative glioma and benign noninfiltrative glial lesion, such as WHO grade I glioma, gliosis, or demyelinating disease, can be challenging for the pathologist.⁴ Frequently, magnetic resonance imaging data are suggestive of the diagnosis and have to be taken into account. Nevertheless, despite the development of novel functional imaging techniques (magnetic resonance spectroscopy, perfusion, or diffusion imaging), which improve the accuracy of diagnosis of brain lesions, some cases of neoplastic and nonneoplastic lesions are still misinterpreted radiologically.^34,35 Therefore, neuropathological examination is still considered to be critical for diagnosis. Elimination of other differential diagnoses is crucial because of prognostic and therapeutic implications.¹ In neuropathological practice, the most useful immunohistochemical markers used are MIB1/Ki67 and p53 and, more recently, the mutated form of IDH1, IDH1 R132H. Although a high MIB1/Ki67 proliferation index and a high p53 immunolabelling index are arguments in favor of infiltrative glioma,^6–10 a low proliferation index and low or null p53 immunolabelling index are insufficient to exclude the diagnosis. Positive IDH1 R132H immunolabelling is very useful because of a high specificity for grade II and grade III infiltrative gliomas.^13–15 Nevertheless, IDH1 R132H is usually negative in the most frequent subtype of infiltrative glioma, the primary GBM.^14,16 The main limitation of all of these markers is the dependence of their staining after either grade for MIB1/Ki67, histological subtype for p53 (overexpressed in astrocytic gliomas, compared with oligodendroglial gliomas^3,8), or the progression pathway for IDH1.¹⁴ In our study, we showed that the presence of cells with a high nuclear to cytoplasmic ratio and with strong EGFR staining is specific for IG regardless of grade, histological subtype, or progression pathway, and thus, it represents a very useful and reliable tool for the diagnosis of IG.

EGFR overexpression is common in IG.² It is classically reported that EGFR expression increases with the grade of glioma.^21,36 In our study, nearly all IG showed EGFR-positive cells. In some of the cases, the proportion of positive cells was very low. Such cases would probably be considered as negative in other studies using a higher threshold for positivity or using a less sensitive technique of detection. Discrepancies among studies investigating EGFR expression in glioma may also relate to the antibody used. Some antibodies recognize the extracellular domain of the receptor, whereas others recognize the intracellular domain and this affects the results,^37,38 especially in glioma where different isoforms of EGFR may be expressed.³⁹ In our study, we used an antibody directed against the extracellular domain of EGFR. Guillaudeau et al.³⁹ showed that antibodies recognizing the extracellular domain of EGFR were more sensitive for the detection of EGFR expression. This could explain why we found EGFR expression in nearly all IG. The property of tumor cells to overexpress EGFR has already been used for their visualization in living tumor tissue of high and low grade.⁴⁰ As described in the literature, the presence of EGFR-overexpressing cells in our study was not correlated with EGFR amplification.^36,41,42 Data concerning EGFR expression in noninfiltrative glioma are far fewer but corroborate our results. Studies described a much lower EGFR expression in pilocytic than in infiltrative glioma.^43–45 To our knowledge, EGFR overexpression has never been reported in ganglioglioma or in DNET, thus indicating that the EGFR pathway is not involved in the development of these tumors.

The major difference between infiltrative glioma and benign glioma, gliosis, or demyelinating diseases is the presence of infiltrating cells in infiltrative glioma. We showed that the invasive edge was enriched with strongly positive EGFR-expressing cells with a high nuclear to cytoplasmic ratio, compared with the solid tumor components and that these cells coexpressed the invasion marker CXCR4,^46,47 suggesting that they may be infiltrating cells. The migratory properties of these cells have not been assessed because of difficulties in maintaining sufficient numbers of cells in culture. The reduced capacity for growth of EGFR-positive cell subpopulations after FACS has already been described.⁴⁸ However, our results are in accordance with the known role of EGFR in cell migration^{26–28,30,31} and with the crosstalk between EGFR and CXCR4 pathways as described in the literature.^49–53

Strongly stained EGFR-positive cells had a peculiar morphology, characterized by scant cytoplasm and often a unique small process, without astrocytic or oligodendrocytic differentiation. Our results showed that they expressed stemness/progenitor markers but lacked expression of markers of astrocytic differentiation, such as GFAP. In addition, in culture medium appropriate for neural stem/progenitor cells, they displayed the ability to self-renew. All of these features are characteristic of neural stem/progenitor cells and indicate that these particular cells resemble glial progenitors. Cancer neural stem cells/progenitors (CSC) have been identified in glioblastoma since 2002⁵⁴ and are likely to be determinants of tumor aggression and resistance to treatment.⁵⁵ Although CSC are largely described and studied in glioblastomas, they have also been identified in grade II and grade III gliomas.^56–58 CSC in glioblastomas may express EGFR, and it has been shown that EGFR expression was associated with a more tumorigenic and invasive behavior, compared with EGFR-negative CSC.⁴⁸ EGFR-knockout in CSCs led to morphologic changes in vitro, typical of differentiated cells. Several ligands may interact with EGFR, including TGFα, which is a member of the EGF family. Several tumors have been shown to coexpress TGFα and EGFR, indicating the existence of autocrine activation loops driving tumor growth.⁵⁹ Of interest, TGFα is able to down-regulate GFAP expression and up-regulate nestin expression in glial tumor cells, and this regulation is correlated with increased motility and a less stellate morphology, suggesting that the TGFα-EGFR loop may be required to induce de-differentiation in tumor cells to promote motility.⁶⁰ Therefore, EGFR expression in tumor cells seems to be involved in the acquisition of glial precursor characteristics and migration properties.

In conclusion, cells displaying strong EGFR staining with a high nuclear to cytoplasmic ratio are specific to IG, regardless of grade and histological subtype. EGFR immunolabelling using an antibody which recognizes an epitope in the extracellular domain of both EGFRwt and EGFRvIII forms could be a very useful tool in neuropathological practice to distinguish these tumors from benign or curable glial lesions. The cells that we have described and characterized show stem-like properties and may be important in influencing the aggressive features typical of IG.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (http://neuro-oncology.oxfordjournals.org/).

Funding

This work has been supported by grants from the Association pour la Recherche sur le Cancer (subvention 3161), Association Sauvons Laura, Agence Nationale pour la Recherche (ANR Jeunes Chercheurs, Jeunes Chercheuses, « GLIOMIRSTEM project »).