Molecular Stratification of Diagnostically Challenging High-Grade Gliomas Composed of Small Cells: The Utility of Fluorescence In situ Hybridization

Diffuse gliomas are therapeutically vexing in general, although some of these tumors are chemosensitive, and rare cures are observed (1). There is considerable morphologic overlap between various glioma entities, and high-grade gliomas with different cytological origin and biological behavior can represent cytologically similar microscopic appearance. As a consequence, their distinct histopathologic interpretation, as well as a further planning of optimal treatment regimen, is a controversial issue. Thus, the clear distinction of malignant oligodendrogliomas from small-cell glioblastomas remains somewhat ill defined (2, 3, 4, 5, 6). The presence of exuberant microvascular proliferation and necroses with pseudopalisading in a classical oligodendroglioma should not occasion a change of diagnosis on a glioblastoma. Conversely, in the settings of a typical small-cell glioblastoma, it is possible to find regions of uniform cells with round nuclei and perinuclear haloes that raise the possibility of an oligodendroglial component (7, 8). Nevertheless, the clinical outcomes for both of these tumor entities are extremely different (1). The vast majority of glioblastomas is resistant to most available therapeutic modalities, whereas anaplastic oligodendrogliomas are chemosensitive tumors, although a part of these lesions are therapeutically resistant too.

To develop more objective diagnostic approaches, recent investigators focused on molecular genetic analysis of glial neoplasms. Some authors suggested that expression of oligodendrocyte lineage genes and corresponding proteins (9, 10, 11) or myelin transcripts (12) could be of help in the differential diagnoses of anaplastic oligodendrogliomas versus small-cell glioblastomas. Nevertheless, these markers are still not widely included in routine settings of neuropathological methods. In addition, cytological identification of glial neoplasms in itself is not sufficient for its prognostic evaluation because specific molecular signatures were found to be more closely associated with tumor chemosensitivity. Thus, the allelic status of chromosomal arms 1p and 19q has emerged as a powerful means to predict treatment response in high-grade oligodendrogliomas, whereas a similar analysis could also help to establish a cytologic diagnosis of oligodendroglioma (13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24). Moreover, loss of 1p and 19q may be a marker of better prognosis and treatment response, not only in pure oligodendrogliomas but in some malignant astrocytic gliomas as well (23, 25). In contrast, small-cell glioblastomas are largely amplified by the epidermal growth factor receptor (EGFR) oncogene (4, 26), but an amplification of this gene in oligodendroglial tumors is uncommon (14, 16, 19, 24). In addition, deletions of CDKN2A/p16 gene and the loss of chromosome 10 with frequent inactivation of the PTEN gene are molecular hallmarks of glioblastoma (4), although these patterns may also be observed in ∼20–30% of malignant oligodendrogliomas that are related with their chemoresistance and adverse outcomes (14, 16, 19, 21, 24, 27, 28).

Recently, Fuller et al. (29) have applied the fluorescence in situ hybridization (FISH) method with the above-mentioned molecular markers for genetic analysis of morphologically ambiguous glial neoplasms with various grades of malignancy. The authors established that distinct and prognostically relevant molecular subsets of these tumors might be identified with this methodology, although they were unable to associate individual genetic patterns with specific morphologic features. Given the continued difficulties of glioma histopathologic classification, our present study was focused on the diagnostically challenging and morphologically uniform high-grade gliomas that were composed of small cells. FISH was used for their molecular stratification with the markers noted above.

A total of 114 adult patients who were treated in the Burdenko Neurosurgical Institute from January 1, 1993, to January 1, 2002, and who had newly diagnosed, supratentorial high-grade gliomas with “nonclassical” morphology (i.e., gliomas, for which the histologic diagnosis glioblastoma versus anaplastic oligodendroglioma was controversial) were included in this study. These cases were selected by the chief neuropathologist (A. K.) after re-evaluation of the 1509 diffuse glioma specimens with a histologic diagnosis of either glioblastoma (1142 cases) or anaplastic oligodendroglioma (367 cases). All of the examined tumors showed a uniform histologic appearance and were composed predominantly from small cells with high nuclear/cytoplasmic ratio and scanty cytoplasm (for a detailed description, see Results).

In all cases, immunohistochemistry with the antibody to glial fibrillary acidic protein (GFAP; clone 6F2, prediluted, DakoCytomation, Carpinteria, CA) was done. For differential diagnoses with primitive neuroectodermal tumors, the antibodies to synaptophysin (clone SY-38, 1:200, DakoCytomation), neurofilament proteins (clone 2F11, prediluted, DakoCytomation) and microtubule-associated protein (MAP LC3, polyclonal, 1:200, Santa Cruz Biotechnology, Santa Cruz, CA) were also applied.

The end point of the follow-up analysis was January 1, 2004, and the dates of death or last contact were considered the end of the study. Data on overall survival are shown as both the median survival times and the 5-year survival rates. The Institutional Review Board obtained approval to link laboratory data to clinical data.

A two-color FISH assay was done on 4-μm-tick sections. The following four commercial sets of locus-specific and fluorochrome-labeled probes were applied (all produced by Vysis, Inc., Applied Biosystems, Downers Grove, IL). Centromere (CEP)7/EGFR dual color probe set (Cat. No 32-191053), CEP9/p16 dual color probe set (Cat. No 32-190078), CEP10/PTEN dual color probe set (Cat. No 32-231010), and mixed 1p36/1q25 and 19p13/19q13 dual color probe sets (Cat. No 32-231004). After protease pretreatment of the slides and its fixation, the probes (10 μL per slide) were applied to the sections. Simultaneous probe/specimen denaturation at 76°C for 8 minutes with subsequent overnight incubation at 37°C was done in the HYBrite hybridization chamber (Vysis Inc.). Posthybridization process included subsequent washing in 50% formamide/2× SSC (4 times for 5 minutes at 42°C) and 0.5× SSC with NP40 L(three times for 5 minutes at 60°C). Nuclei were counterstained with DAPI (1000 ng DAPI/mL in antifade mounting solution, Vysis Inc).

The sections were studied with an Axioplan 2 fluorescent microscope (Karl Zeiss, Göttingen, Germany) that was equipped with a set of the appropriate filters (Vysis Inc.). Signal detection was done as described previously (17, 29, 30, 31). Briefly, samples showing sufficient FISH efficiency (>90% nuclei with signals) were evaluated. Two independent referees (A. K. and R. S.) were unaware of clinical data. Signals were scored in at least 300 nonoverlapping, intact nuclei. Nonneoplastic temporal lobe specimens (n = 8) were used as a control for each probe pair. Specimens were considered amplified for EGFR when more than 10% of tumor cells exhibited either a EGFR/CEP7 ratio of >2 or innumerable tight clusters of signals of the locus probe. Hemizygous deletions of the p16 locus were defined as >50% nuclei containing a single locus probe signal and two CEP9 signals (mean +3 SDs in nonneoplastic controls). Homozygous deletions of p16 were identified by the simultaneous lack of both of the locus signals and by the presence of CEP9 signals in more than 30% of cells. Deletions of 1p, 19q, and PTEN were defined as >50% of tumor nuclei containing one signal (mean +3 SDs in nonneoplastic controls). Most often there were absolute deletions (i.e., one signal of locus-targeted probe and two signals of reference probe), but sometimes, relative deletions (two signals of locus-targeted probe and four signals of reference probe) were seen. No homozygous PTEN deletions were found. Polysomies (chromosomal gains) were defined as >10% of nuclei containing three or more CEP signals, because no such findings were found in the control brain tissue. Monosomy for each chromosome was defined by the presence of one CEP signal per cell in >50% (mean +3 SDs in nonneoplastic controls). A balanced profile-compatible pattern showed two of each locus probe and two CEP signals for CEP7/EGFR, CEP9/p16, and CEP10/PTEN probe pairs and two of each locus probe for 1p/1q and 19p/19q probe pairs in more than 51% of cells and no chromosomal gains.

Images were captured by with a high-resolution CCD microscopy camera AxioCam MRm REV2 (Karl Zeiss). The resulting images were reconstructed with green (FITC), red (Cy3) and blue (DAPI) pseudocolors with AxioVision 4 multichannel fluorescence basic workstation (Karl Zeiss) according to the manufacturer’s instructions.

According to prior experience (17, 29), homozygous and hemizygous p16 deletions, as well as absolute and relative 1p and 19q deletions, were considered biologically equivalent.

For categorical data, the χ² test was used. Survival analyses from the date of operation were estimated with Kaplan–Meier method. The comparisons among various patient subgroups were done by the Log-Rank test. Multivariate analysis for survival was done with the Cox Proportional Hazard models. In addition, the classification and regression tree models for censored data (CART) as modified by LeBlanc and Crowley (32) were used. A significant correlation between two parameters was taken at the 95% confidence interval. P values < 0.05 were considered significant.

The 114 patients included 67 males and 47 females. The age of patients ranged from 19 to 72 years with a median age of 47 years, and 61 patients were younger than 50 years. According to preoperative contrast computed tomography and/or magnetic resonance imaging data, tumor occupied the frontal lobe in 35 (31%) of the patients. In all of the cases, an intense and heterogeneous ring-like contrast enhancement was identified. All of our patients had undergone open surgery with either gross total (78 cases) or partial (36 cases) tumor resection that was confirmed by postoperative neuroradiologic studies. Postoperative Karnofsky performance scores ranged from 80 to 90. All of the patients received postoperative radiotherapy (external beam irradiation) with a total dose 56 to 62 cGy (mean 58 Gy) after chemotherapy. As a first-line treatment, all 114 patients received polychemotherapy with procarbazine, lomustine, and vincristine (PCV regimen), and, thereafter, 32 patients with rapid tumor progression received a second-line monochemotherapy with temozolomide. At the date of end of the follow-up, 92 patients (80%) had died within the period of 4 to 88 months after surgery; the median survival time was 14 months. Survival time for the remaining 22 survivors varied from 28 to 117 months with the median survival time of 67 months. The median overall survival for the entire cohort of patients was 17 months, and the five-year survival rate was 18%.

Histologically, there were highly cellular and cytologically monotonous gliomas showing brisk mitotic activity, prominent microvascular proliferation, and necroses, with the presence of pseudopalisades in 63 tumors and without them in the remaining 51 samples. All 114 tumors were ambiguous as to cell type throughout a whole specimen. The cytological composition of these tumors included simultaneously both areas of haphazardly arranged small spindled cells that resembled a glioblastoma and clusters of rounded small cells with relatively regular and uniform arrangement that reminded us of an oligodendroglioma. The tumor cells were rather monomorphic, with high nuclear-to-cytoplasmic ratio and small-to-medium size hyperchromatic nuclei that were carrot-shaped, slightly elongated, or rounded.

There were no textbook oligodendroglioma patterns (uniform round nuclei with conspicuous nucleoli and perinuclear haloes, even cell distribution, branching vasculature with chicken-wire pattern) nor features of ependymal differentiation (specific cytological patterns, true ependymal rosettes, and perivascular pseudorosettes).

Expression of glial fibrillary acidic protein (GFAP) was found in all of the cases examined, and at least 30% of tumor cells showed an intense immunolabeling of their cytoplasm and processes. None of the neuronal markers exhibited expression in any of the samples tested.

Representative results of FISH with corresponding microscopic tumor appearance are presented in Fig. 1. Of the 570 total paired hybridizations, all yielded interpretable results. Notably, amplification of EGFR was always accompanied by polysomy of chromosome 7, and, similarly, chromosomal loci 1p and 19q were always codeleted. The most frequently encountered alteration was the deletion of p16 locus, which was found in 71 cases (62%) and was accompanied by monosomy 9 in 8 cases. Other detectable molecular abnormalities, in descending order of frequency, included polysomy of chromosome 7 [67 samples (59%)], deletion of PTEN locus [61 samples (54%)] accompanied by monosomy 10 in 36 cases, EGFR amplification [46 samples (40%)], combined loss of 1p and 19q [14 samples (12%)], and solitary deletion of 19q [12 samples (11%)]. The small minority of cases showed polysomies 9 [8 samples 7%), 19 [6 samples (5%)], and 1 [4 samples (4%)]. In addition, 13 tumors (11%) revealed the lack of any discernable molecular aberration and showed two of each probe and two CEP signals per nuclei. Hence, these tumors were considered having balanced and diploid cytogenetic profile for chromosomes 1, 7, 9, 10, and 19.

The most frequently encountered multiple alterations were found for combined p16 deletion and polysomy 7 (54%), combined PTEN deletion and polysomy 7 (45%), p16 and PTEN codeletion (43%), combined EGFR amplification and polysomy 7 (40%), combined EGFR amplification and p16 deletion (36%) and, finally, combined EGFR amplification and PTEN deletion (32%). Thirty-three tumors (29%) displayed the presence of four above-mentioned molecular alterations simultaneously.

Codeletion of 1p and 19q was unaccompanied by additional molecular alterations in all of the samples examined except one, in which there were additional deletions of p16 and PTEN. Only 7 of 12 tumors with solitary 19q deletion exhibited no other molecular alterations, and in the remaining 5 samples, this cytogenetic feature was accompanied by deletions of p16 and/or PTEN.

After acquisition of FISH data, the cases examined were reviewed to search for potential associations between clinical variables, tumor histology, and molecular features (Table 1). The patients whose tumors showed EGFR amplifications, polysomy 7, and deletions of p16 and PTEN were older, whereas the age of the patients whose tumors exhibited either 1p/19q codeletion or balanced cytogenetic profile was usually less than 50 years. Tumors harbored codeletion of 1p and 19q, more frequently centered in the frontal lobe. There were no associations between molecular findings and the presence of pseudopalisades.

By univariate analysis (Table 2), the patient survival time was shorter for EGFR amplification, p16 and PTEN deletions, and polysomy 7. In contrast, the patients whose tumors displayed either 1p/19q codeletion or the lack of any alteration, exhibited significantly prolonged survival times. By multivariate analysis (Table 3), only molecular features were reached at an independent level. Thus, amplification of EGFR and deletion of PTEN represented prognostically unfavorable variables, whereas 1p/19q codeletion and balanced molecular profile were strongly associated with prolonged survival.

Figures 2 and 3 display, respectively, the diagram and the Kaplan–Maier curves generated by the CART modeling process. We identified four final groups of patients with distinctly different survival times (see Figs. 1, 2, and 3, and Table 4; all differences are statistically significant, P < 0.05), as follows: (a) 13 patients whose tumors exhibited none of the discernable molecular alterations (balanced and diploid profile); (b) 20 patients whose tumors harbored either 1p/19q codeletion or isolated deletion of 19q unaccompanied by other molecular abnormalities; (c) 35 patients whose tumors revealed p16 and/or PTEN deletions unaccompanied by EGFR amplification; and (d) 46 patients whose tumors harbored EGFR amplification.

It has become more customary to search in any diffuse glioma specimen for areas with rounded nuclei that could represent oligodendroglioma, because the therapeutic and prognostic features of this glial tumor entity differ markedly from those for astrocytomas. Therefore, concepts have changed over time, and definitional shifts have lead to a widening of the accepted morphologic spectrum of the textbook oligodendroglial phenotype (33). As a consequence, the diverse range of approaches to the diagnosis of glial neoplasms showing ambiguous histology is not uncommon and reflects the diagnostic diversity of the World neuropathologic community in general. Coons et al. (34) found that diagnostic concordance among four neuropathologists reviewing diffuse glioma specimens reached 69%. Fuller et al. (29) disclosed some disagreement in the diagnostic philosophy for the six independent referees invited for the evaluation of 109 mixed diffuse gliomas with ambiguous histologic features. In this study, some of the reviewers preferred to diagnose these tumors as pure astrocytomas, another group diagnosed mixed gliomas more often, and, finally, a third cohort rendered the greatest number of descriptive histologic diagnoses (29).

The recent studies demonstrated that considerable molecular stratification could be achieved within the group of gliomas that have overlapping morphologic features (35, 36). A few reports have shown that FISH is a rapid and sensitive method of genetically stratifying morphologically ambiguous gliomas and provides prognostically useful information (15, 17, 20, 23, 29, 30). In the present study we applied FISH for molecular profiling of diagnostically challenging high-grade gliomas, keeping in the mind its applicability to paraffin-embedded specimens and the presence of well-established criteria for the interpretation of reproducible results. For this purpose, we analyzed a set of 114 high-grade small-cell gliomas, for which we experienced diagnostic difficulties in the interpretation of their distinct cytological origins. These tumors consisted of the small minority (less than 10%) of all malignant gliomas diagnosed at our department as either glioblastoma or anaplastic oligodendroglioma. Frank histologic malignancy of these neoplasms is in line with their biologically aggressive character, because ∼80% of the patients had died by end of the follow-up period, regardless of combined treatment.

Nevertheless, our data showed that further stratification of morphologically uniform high-grade gliomas composed from small cells could be achieved more precisely on the base of molecular analysis, which allowed their subdivision in the four clinically relevant subsets. In line with previous studies (4, 36), such genetic aberrations as EGFR amplification and deletions of p16 and PTEN chromosomal loci associated closely with an adverse outcome. Because of an obvious histologic malignant disease and the biological aggressiveness of the tumors examined, the high frequency of these unfavorable abnormalities in the present series came as no surprise. In contrast, 1p/19q codeletion was found in only 12% of cases, thus, supporting either the notion that such genetic aberration are not absolutely correlative with textbook oligodendrogliomas or, conversely, the fact that only a handful of lesions having real oligodendroglial origin fell into the present series. Another small tumor subset harbored loss of 19q, which has been found to be associated with astrocytoma malignant progression (4, 22) and long-term survival for a group of glioblastoma patients (37).

Most of the tumors with 1p/19q codeletion involved the frontal lobe, which corroborated a site-specificity of this molecular signature (20, 29, 38). Recently, these chromosomal alterations, detected by loss of heterozygosity (LOH) or FISH analyses, have been proposed to be a powerful prognostic indicator for favorable response to chemotherapy and better survival in anaplastic oligodendrogliomas (13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24). However, a few reports revealed that 1p/19q codeletion showed less impressive prognostic significance in the absence of classic oliogodendroglial features. Thus, Schmidt et al. (39) revealed that glioblastoma patients with combined 1p/19q alteration showed better survival, but TP53 mutation was found to be a more powerful and independent predictor of favorable clinical course. J. He et al. (7) performed molecular analysis of 25 glioblastomas with “oligodendroglial” features. The authors found LOH of 1p and 19q in seven tumors, but they were unable to show any association between these molecular alterations and patient survival. Finally, Fuller et al. (29) revealed no significant association between the codeletion of 1p and 19q and clinical outcomes for patients with morphologically ambiguous diffuse gliomas.

In our cohort of diagnostically challenging small-cell gliomas, 1p/19q codeletion was independently associated with better survival on univariate and multivariate analyses. Nevertheless, the patients whose tumors exhibited a lack of detectable genetic alterations revealed more favorable outcomes, and their 5-year survival rate reached 80%. It gives evidence that the panel of molecular markers applied by us is insufficient to cover an entire spectrum of the prognostically relevant genetic aberrations that have to be determined further. Perhaps, microarray-based screening of these frankly anaplastic-appearing tumors will be necessary to identify the relevant candidate genes and chromosomal regions underlying favorable biological behavior (35).

In summary of all that is noted above, morphologically uniform high-grade gliomas composed of small cells vary greatly in terms of their molecular features and clinical outcomes. Notably, the vast majority of patients whose lesions harbored “favorable” genetic profiles survived more than 36 months, although the real cytological origin and distinct nosologic position of these tumors still remain unclear. However, they merit to be separated from those that display obviously “unfavorable” genetic alterations, because the ability to distinguish molecular profiles of diagnostically challenging gliomas enables appropriate therapies to be tailored to specific tumor subsets (1). Because all patients included in the present study received uniform postoperative treatment, it is difficult to consider whether the long survival of the patients whose tumors harbored favorable molecular signatures was a result of their high sensitivity to treatment or whether it might have been caused by other biological events that are still unknown.

In conclusion, despite the new insight in glioma molecular pathology, standard light microscopy assessment remains the gold standard in final tumor diagnosis. In most cases, the histologic end point has primacy in the designation of high-grade glioma being of either astrocytic or oligodendroglial origin. In such cases, molecular analysis should be recommended as an ancillary method that allows evaluating a tumor response to further treatment. Nevertheless, for a handful of morphologically ambiguous small-cell gliomas, the diagnostic role of molecular analysis put forward on the first line because an identification of distinct cellular origin for such tumors has no prognostic significance. In our opinion, the descriptive diagnosis of such a tumor as “high-grade glioma composed from small cells” followed by the notification of its molecular signature provides clinically useful information, which could potentially improve the quality of treatment.