Small cell astrocytoma: An aggressive variant that is clinicopathologically and genetically distinct from anaplastic oligodendroglioma


  • Arie Perry M.D.,

    Corresponding author
    1. Division of Neuropathology, Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, Missouri
    • Division of Neuropathology, Washington University School of Medicine, Box 8118, 660 S. Euclid Avenue, St. Louis, MO 63110-1093
    Search for more papers by this author
    • Fax: (314) 362-4096

  • Kenneth D. Aldape M.D.,

    1. Division of Neuropathology, Department of Pathology, The University of Texas M. D. Anderson Cancer Center, Houston, Texas
    Search for more papers by this author
  • David H. George M.D.,

    1. Division of Neuropathology, Foothills Medical Center, Calgary, Alberta, Canada
    Search for more papers by this author
  • Peter C. Burger M.D.

    1. Division of Neuropathology, Department of Pathology, Johns Hopkins University School of Medicine, Baltimore, Maryland
    Search for more papers by this author



Small cell glioblastoma (GBM) is a variant with monomorphous, deceptively bland nuclei that often is misdiagnosed as anaplastic oligodendroglioma.


To elucidate its clinicopathologic and genetic features, the authors studied 71 adult patients (median age, 57 years), including 22 patients who were identified from a set of 229 GBMs (10%) that had been characterized previously by epidermal growth factor receptor (EGFR)/EGFR-vIII variant immunohistochemistry. Tumors also were analyzed by fluorescence in situ hybridization for 1p, 19q, 10q, and EGFR copy numbers.


Radiologically, 37% of tumors that were not selected for grade showed minimal to no enhancement. Similarly, 33% of tumors had no endothelial hyperplasia or necrosis histologically, qualifying only as anaplastic astrocytoma (Grade III) using World Health Organization criteria. Nevertheless, such tumors progressed rapidly, with mortality rates that were indistinguishable from their Grade IV counterparts. The median survival for 37 patients who were followed until death was 11 months. Oligodendroglioma-like histology included chicken-wire vasculature (86%), haloes (73%), perineuronal satellitosis (58%), and microcalcifications (45%), although mucin-filled microcystic spaces were lacking. No small cell astrocytomas had 1p/19q codeletions, whereas EGFR amplification and 10q deletions were present in 69% and 97% of small cell astrocytomas, respectively. The tumors expressed EGFR and EGFR-vIII more commonly than nonsmall cell GBMs (83% vs. 35% [P < 0.001]; 50% vs. 21% [P < 0.001] respectively).


Small cell astrocytoma is an aggressive histologic variant that behaved like primary GBM, even in the absence of endothelial hyperplasia and necrosis. Despite considerable morphologic overlap with anaplastic oligodendroglioma, clinicopathologic and genetic features were distinct. Fifty percent of small cell astrocytomas expressed the constitutively activated vIII mutant form of EGFR, and molecular testing for 10q deletion improved the diagnostic sensitivity over EGFR alone. Cancer 2004. © 2004 American Cancer Society.

Unlike the typical multiforme or pleomorphic appearance of conventional glioblastoma (GBM), the small cell variant is characterized by a monomorphic proliferation of small, oval nuclei with only mild hyperchromasia, regular nuclear contours, and a brisk mitotic index.1 Due to this nuclear uniformity and several other overlapping features, such tumors often are misdiagnosed as high-grade oligodendroglial neoplasms. Nevertheless, the distinction is critical, given the improved prognosis and therapeutic responsiveness of oligodendrogliomas, particularly those with chromosome 1p and 19q codeletion.2–9 In contrast, glioblastomas are frustratingly chemoresistant and follow a highly aggressive course, with an average survival of roughly 1 year. It was shown recently that epidermal growth factor receptor (EGFR) amplification is particularly common in the small cell variant,1 but the spectrum of clinicopathologic and genetic features has not been explored fully. Our interest in further characterizing this variant was sparked additionally by 1) frequent consults diagnosed as a high-grade oligodendroglial tumor and sent for 1p and 19q genetic testing and 2) examples that qualified cytologically as small cell GBM but lacked endothelial hyperplasia and necrosis.


Patient/Tumor Selection and Histologic Review

Tumor specimens were retrieved using two strategies. The first cohort consisted of in-house specimens from Washington University (St. Louis, MO), Johns Hopkins University (Baltimore, MD), and the Foothills Medical Center (Calgary, Alberta, Canada), as well as consultation specimens that were sent to two of the authors (A.P. and P.C.B.). Specimens that were diagnosed as having a small cell astrocytic component, regardless of the grade, were retrieved and reviewed for diagnostic accuracy. The small cell phenotype, as outlined recently,1 was defined on routine hematoxylin and eosin-stained sections by a predominant population of small, monomorphic, uniformly oval nuclei with minimal hyperchromasia, minimal discernable cytoplasm, and a surprisingly high mitotic index, given the otherwise bland nuclear cytology. The diagnosis was supported further in a subset of tumors by the presence of long, thin, glial fibrillary acidic protein (GFAP)-positive cytoplasmic processes. Parenchymal infiltration was similar to that encountered in other diffuse gliomas and could be extensive in some specimens, occasionally qualifying for the diagnosis of gliomatosis cerebri radiologically. Others also have included within this nomenclature glioblastomas with primitive-appearing, markedly hyperchromatic nuclei resembling those of medulloblastoma or primitive neuroectodermal tumor (PNET). However, for the sake of diagnostic uniformity, such PNET-like tumors were not included in the current study. Similarly, tumors that did not consist predominantly or exclusively of the small cell variant (> 80% of tumor), tumors with limited sampling (e.g., stereotactic biopsy specimens), and tumors from pediatric patients (age < 18 years) were excluded from further study. A second set of gliomas was selected from a group of 229 newly diagnosed glioblastomas from patients who were seen at The University of Texas M. D. Anderson Cancer Center (Houston, TX) from 1996 to 2003, on whom clinical follow-up data and adequate archival tissues were available. All available slides were reviewed by one author (K.D.A.) for small cell features. Slides that qualified were sent to a second author (A.P.), who determined whether the small cell features were predominant (> 80%) or focal, and the latter slides were excluded from further study. Specimens from this second set were not included in the analyses of radiographic features and tumor grade, because there was a selection bias; i.e., by definition, they all were glioblastomas (World Health Organization [WHO] Grade IV). In addition, all available slides from both cohorts were assessed for the presence or absence of five oligodendroglioma-like features: a chicken wire-like capillary network, clear perinuclear haloes, perineuronal satellitosis, microcalcifications, and mucin-rich microcystic spaces. Clinical and radiographic data, including clinical follow-up, were obtained by chart review at the various medical centers, in accordance with local Institutional Review Board approvals. Most radiology data were obtained from available reports at each of the various medical centers, particularly in specimens from the consult patients. However, the magnetic resonance images with and without gadolinium also were available in a subset of patients. Follow-up in some of the consult patients was provided by referring pathologists and treating clinicians.

Fluorescence In Situ Hybridization

Fluorescence in situ hybridization (FISH) was performed on 5-μm-thick sections from a representative paraffin block (when available), as reported previously.8, 9 In some of the consult specimens, either the genetic studies already had been performed, or unstained sections were available for further studies rather than a paraffin block. Paired fluorescein isothiocyanate (FITC)/rhodamine-labeled DNA probes included 1p32 (RPCI-11 human BAC 260I23; Washington University Human Genome Sequencing Center, St. Louis, MO)/1q42 (RPCI-11 184E11), 19p13 (RPCI-11 575H1)/19q13 (RPCI-11 426G3), CEP7 (Vysis Inc., Downers Grove, IL)/EGFR (RPCI-11 148P17), and PTEN (10q23)/DMBT1 (10q25–q26) (both donated by Dr. Robert Jenkins, Mayo Clinic, Rochester, MN). Deparaffinization of the sections was carried out with two 10-minute immersions in Citrisolv (Fisher Scientific, Pittsburgh, PA), followed by three 3-minute immersions in isopropanol. Next, the slides were rinsed in running water for 5 minutes, followed by distilled water for 3 minutes. Target retrieval was achieved by immersing the slides in citrate buffer, pH 6.0, within a plastic Coplin jar, which was steamed for 20 minutes, then slowly cooled to room temperature. The slides were then rinsed in running water for 5 minutes, followed by distilled water for 3 minutes. This was followed by 0.4% pepsin (P-7012; Sigma-Aldrich, St. Louis, MO) digestion for 15 minutes at 37 °C, and then a rinse in 2 × standard saline citrate (SSC) on a rotator for 5 minutes. Slides were then air dried. Paired probes were diluted from stock with tDenHyb hybridization buffer (Insitus Biotechnologies, Albuquerque, NM) to a concentration of 1:25 and dispensed at 10–20 μL per slide, depending on the surface area of tissue to be covered for any given section. Slides were coverslipped, and denaturation was achieved by placing the slides on the metal surface of a light-shielded slide moat that was preheated to 90 °C for 13 minutes. The slides were removed and kept in darkness until the slide moat reached a temperature of 37° C. Slides were then replaced into the slide moat, which was then used as a 37 °C humidified chamber for the overnight incubation period, during which hybridization occurred. The next day, slides were removed from the 37 °C humidified chamber. Coverslips were removed, and the slides were washed in 50% formamide/1 × SSC solution and placed on a rotator for 5 minutes. This was followed by 2 washes in SSC for 2 minutes each. Slides were removed and allowed to air dry. Depending on the surface area of the section to be covered, 10–20 μL of 4,6-diamidino-2-phenylindole (DAPI) in Fluorgard (Insitus) were applied to each of the slides, which were then coverslipped.

Green and red fluorescent signals were enumerated under an Olympus BX60 fluorescence microscope with appropriate filters (Olympus, Melville, NY). For each hybridization, a minimum of 100 nonoverlapping nuclei were assessed for numbers of green and red signals. An interpretation of deletion was made when > 50% of the nuclei harbored only 1 red or 1 green signal. This was based on the frequencies of nonneoplastic nuclei that contained 1 signal for the same probes in nonneoplastic control (seizure-resection) specimens (median ± 3 standard deviations). Polysomies or gains were set arbitrarily at 10% nuclei with > 2 signals, because such cells are seen rarely in nonneoplastic brain specimens. In addition, specimens were considered amplified for EGFR when they demonstrated nuclei containing innumerable red signals and an EGFR:CEP7 ratio > 2. FISH images were captured using a black-and-white, high-resolution Cohu CCD camera (Cohu Inc., San Diego, CA); a Z-stack motor; and a CytoVision basic workstation (Applied Imaging, Santa Clara, CA) with sequential DAPI (1 level), FITC (10 levels), and rhodamine (10 levels) filter settings. The resulting images were reconstituted with blue, green, and red pseudocolors using CytoVision software.


Sections were stained using a previously prescribed protocol.10 Briefly, sections were deparaffinized, rehydrated, and treated with 3% hydrogen peroxide for 30 minutes. For EGFR staining, sections were treated with 0.025% trypsin for 30 minutes. For EGFR-vIII staining, sections were placed in 50 mM sodium citrate buffer, pH 6.0, in a pressure cooker and microwaved. EGFR antibody (clone 528; Oncogene Science) was used at a 1:200 dilution in phosphate buffered saline/10% goat serum. EGFR-vIII antibody (rabbit polyclonal; Zymed) was used at 1:2000 dilution in the same buffer. After an incubation overnight in the refrigerator followed by washing, appropriate secondary antibodies (Envision kit; Dako, Carpinteria, CA) were added for 30 minutes. Staining was visualized using diaminobenzidine.


Patient/Tumor Cohort, Neuroimaging Studies, and Clinical Follow-Up

Combining the 2 groups described above (see Materials and Methods), the entire study cohort consisted of 71 surgical specimens from 71 adult patients, ages 19–90 years (median, 57 years). There were 43 men and 28 women (1.5:1.0), with 42% of all patients derived from outside consultations. The primary tumor location was cerebral in all patients, except for one patient who presented with a primary tumor in the cerebellum. Primary sites of involvement at the time of presentation included temporal lobe (46%), frontal lobe (35%), parietal lobe (20%), occipital lobe (13%), and thalamus (7%), with some of the tumors involving > 1 site. Presenting symptoms often were nonlocalizing and most commonly included new-onset seizures, mental status changes, and headaches. Of the 49 patients who were not selected specifically from the subset of previously studied GBMs, radiographic data were available on 35 patients. Thirteen patients (37%) presented with nonenhancing or minimally enhancing masses (Fig. 1). The radiologic differential diagnosis in such patients often included cerebral infarct, encephalitis, and low-grade glioma. In fact, the findings occasionally were so subtle that the initial studies were reported as normal. However, on follow-up imaging studies 2–3 months later, the tumors typically progressed to clearly identifiable, rapidly growing masses, often with ring enhancement (Fig. 1). The remaining patients presented initially with either heterogeneous enhancement or ring enhancement, and 5 tumors (14%) were considered multifocal. Radiographic data also were available in 20 of the 22 patients with glioblastoma from The University of Texas M. D. Anderson Cancer Center. Only one tumor in that group was characterized as ‘predominantly nonenhancing’. The remaining tumors showed either ring enhancement or a heterogeneous pattern of tumoral enhancement. When combining these patients from the other cohort, only 7% of the 43 patients with WHO Grade IV tumors with available radiology reports had either nonenhancing or minimally enhancing tumors (Table 1).

Figure 1.

Representative (A,C,E,G) T2-weighted and (B,D,F,H) gadolinium-enhanced T1-weighted magnetic resonance images obtained from patients who presented with nonenhancing or minimally enhancing small cell astrocytomas. The images in (E–H) are from a single patient who had (E,F) minimal enhancement at presentation and (G,H) progression of disease to a ring-enhancing mass 3 months later.

Table 1. World Health Organization Grade III versus Grade IV Small Cell Astrocytomas
Patient characteristicNo. of patients (%)
Grade IIIGrade IV
  1. EGFR: epidermal growth factor receptor.

Age (yrs)  
No or minimal enhancement11/12 (92) 3/43 (7)
Survival (mos)  
10q deletion10/12 (83)52/52 (100)
Polysomy 710/12 (83)47/53 (89)
EGFR amplification 7/12 (58)38/53 (72)

Thirty-seven patients (52%) were followed until death, with a median survival of 11 months (range, 2–22 months). The remaining 34 patients who were known to be alive had limited clinical follow-up (range, 0–20 months; median, 2 months), because many patients were diagnosed only recently. The length of survival for patients who were diagnosed with anaplastic astrocytoma (WHO Grade III) did not differ significantly from the length of survival for patients who were diagnosed with glioblastoma (WHO Grade IV) (Table 1). The only statistically significant difference between these 2 groups was that 92% of patients with Grade III small cell astrocytomas had no or minimal contrast enhancement compared with 7% of patients with tumors in the Grade IV category (P < 0.001; Fisher exact test). This finding is not surprising, because contrast enhancement correlates well with endothelial hyperplasia, the presence of which defines the tumor as Grade IV. Two patients had evidence of metastasis, one to the lung and the other to the lumbosacral region (i.e., drop metastases).

Histopathologic Features

By definition, tumors were composed either predominantly or entirely of small cell astrocytic cells with monomorphous oval nuclei, mild nuclear hyperchromasia, bland chromatin, occasional small nucleoli, minimal discernable cytoplasm, and frequent mitotic figures (Fig. 2A). In addition to the high proliferation index, 33 of 49 tumors (excluding the 22 tumors that were selected from the set of glioblastomas) had endothelial hyperplasia and/or pseudopalisading necrosis (usually both). However, the remaining 16 tumors (33%) had neither and, thus, only qualified as Grade III anaplastic astrocytoma using current WHO criteria. The majority of such patients presented with nonenhancing masses (Table 1). In the second cohort of 229 glioblastomas that were reviewed at The University of Texas M. D. Anderson Cancer Center, 22 tumors were predominantly small cell in appearance, suggesting that this variant represents roughly 10% of such tumors. An additional 25 tumors (11%) had focal small cell features.

Figure 2.

Histologic features of small cell astrocytomas, many of which overlap with oligodendrogliomas, include (A) nuclear uniformity and microcalcifications; (B) cytologically bland, oval nuclei with numerous mitotic figures; (C) chicken wire–like capillaries; (D) clear perinuclear haloes; and (E) perineuronal satellitosis.

In addition to the nuclear regularity, other oligodendroglioma-like features were common in the 71 specimens studied, including chicken-wire vasculature (86%), clear haloes (73%), perineuronal satellitosis (58%), and microcalcifications (45%) (Fig. 2). In most tumors, the clear perinuclear haloes were not as prominent as those seen in classic oligodendrogliomas, instead representing a focal finding. In addition, mucin-rich microcystic spaces, a common finding in high-grade oligodendrogliomas, were not found in any of our small cell astrocytomas. Although it was not a specific focus of our study, additional immunostains for GFAP were performed as part of the routine work-up in 16 specimens, revealing thin, elongated, immunopositive, cytoplasmic processes in 15 specimens (94%) (Fig. 3A). MIB-1 (Ki-67) labeling indices generally were high, ranging from 17% to 61% (median, 24%) in the 9 specimens in which it was quantitated (Fig. 3B). Stains for p53 protein were performed in only 8 specimens, with 5 specimens that were immunonegative and 3 specimens that showed minor populations of positive cells.

Figure 3.

Typical immunohistochemical and fluorescence in situ hybridization findings include (A) thin, glial fibrillary acidic protein (GFAP)-positive cytoplasmic processes; (B) a high MIB-1 (Ki-67) labeling index; (C) epidermal growth factor receptor (EGFR) immunoreactivity; (D) EGFR-vIII immunoreactivity; (E) EGFR gene amplification, with innumerable red signals (CEP7 in green, EGFR in red); and (F) chromosome 10q deletion, with only one green and one red signal, in most nuclei (PTEN in green, DMBT1 in red).

Genetic Features

Representative genetic findings are illustrated in Fig. 3C–3F. Of the 71 specimens, 66 had tissue blocks or unstained sections available for FISH analysis. One specimen was noninformative for all markers tested due to a lack of discernable signals. Of the remaining 65 specimens, EGFR amplification was evident in 45 specimens (69%), most commonly in association with gains of chromosome 7 centromere signals or polysomy 7 (Fig. 3E). In the vast majority of specimens, the amplification was widespread, typically involving nearly all tumor cells. However, 2 specimens had focal EGFR amplification, with only 5–10% of tumor cells identified as positive. Evidence for chromosome 10q deletion was even more common and was encountered in 97% of specimens (Fig. 3F). None of the specimens had evidence of 1p deletion, and only 1 specimen had a demonstrable 19q deletion. Polysomies of one or both chromosomes were seen occasionally. Of the 28 specimens (43%) with polysomy 19, 11 specimens also showed gains of chromosome 1, consistent with an overall state of aneuploidy or polyploidy. The remaining 17 specimens (26%) likely represented a specific gain of chromosome 19. There were no examples of chromosome 1 gain in the absence of polysomy 19. Despite these occasional gains, the overall impression from reviewing all of the FISH enumeration counts was that the majority of small cells had a near-diploid complement (i.e., two signals for reference probes in most cases). This differs from conventional GBMs, in which aneuploidy/polyploidy is extremely common.

Tissue was available for EGFR and EGFR-vIII immunostaining from 46 tumors and 50 tumors, respectively. EGFR staining was at least focally positive in 38 of 46 tumors (83%), and EGFR-vIII staining was positive in 25 of 50 tumors (50%) (Fig. 3C, 3D). In previous studies of unselected GBMs using identical staining protocols, we found EGFR positivity in 39 of 110 tumors (35%)10 and EGFR-vIII positivity in 22 of 105 tumors (21%).11 Therefore, comparisons of the current cohort with previous unselected series demonstrate higher overall expression levels of both EGFR and EGFR-vIII in small cell GBM (P < 0.001 for each comparison; chi-square analyses). The overall concordance between EGFR gene amplification status by FISH and EGFR immunohistochemistry was 83%. However, if chromosome 7 gains also are accepted as a potential explanation for increased EGFR expression, then the concordance reaches 94%.


In a recent article, the pathologic definition of small cell GBM was clarified, and its common association with EGFR amplification was emphasized.1 Nevertheless, it is a variant that remains under-recognized. In fact, our review of 229 previously studied GBMs suggests that it is relatively common, accounting for as many as 10%, with another 11% showing focal, small cell features. Although the name suggests a small, blue-cell tumor similar to PNET, the chromatin pattern is surprisingly bland and only mildly hyperchromatic; so that, in fact, the nuclear cytology more often is reminiscent of oligodendroglioma. Moreover, it even may be difficult to recognize that regions with widely spaced tumor cells are neoplastic. Clearly, glioblastomas with PNET-like foci also occur, but their cytologic features are quite different (primitive-appearing, hyperchromatic, rounded to carrot-shaped nuclei, often with molding); therefore, they were not included in the this study.

The current study showed that the mimicry of oligodendroglioma is exacerbated further by other histologic features, such as a branching, chicken wire-like capillary networks, clear perinuclear haloes, perineuronal satellitosis, and microcalcifications. Thus, it is not surprising that these tumors commonly are misdiagnosed as anaplastic oligodendroglioma, anaplastic oligoastrocytoma, or glioblastoma with oligodendroglial features. However, building on prior studies, a distinct clinicopathologic and genetic pattern has emerged, and a number of features serve to distinguish small cell astrocytomas from oligodendrogliomas (Table 2). For example, although both ring-enhancement and pseudopalisading necrosis may be encountered in high-grade oligodendrogliomas, they are relatively uncommon and always should prompt the consideration of small cell astrocytoma as an alternate possibility. Because mucin-rich spaces (‘microcysts’) never were encountered in our patients, such a finding would favor a high-grade oligodendroglioma. In contrast, small cell astrocytomas have oval rather than rounded nuclei; a surprisingly brisk proliferative index given the paucity of nuclear atypia; and often have thin, GFAP-positive cytoplasmic processes. In our experience, foci of anaplastic oligodendrogliomas more frequently have larger epithelioid cells with vesicular nuclei, prominent nucleoli, and a greater degree of pleomorphism compared with foci of small cell astrocytomas. Despite these histologic distinctions, the differential may be extremely difficult in selected patients, and additional studies probably should be performed in patients with suspected, high-grade, oligodendroglial tumors in the presence of an unusually rapid clinical progression, ring enhancement, a dissociation between bland cytology and brisk proliferation, and/or pseudopalisading necrosis (Table 2). Our data show that, in addition to the total lack of 1p/19q deletions and the EGFR amplification in roughly 70% of tumors, EGFR-vIII immunoreactivity and 10q deletion determined by FISH can be extremely helpful for confirming the diagnosis. The latter is particularly sensitive, because it was nearly universal in our cohort. Caution is warranted in using simple EGFR immunohistochemistry; because, in contrast to gene amplification, immunopositivity reportedly is common in oligodendrogliomas of all grades, although this varies somewhat depending on the antibody used.12–15 In contrast, the vIII mutation is associated specifically with glioblastomas. It should be noted that EGFR amplification and 10q deletions also have been reported in anaplastic oligodendrogliomas and oligoastrocytomas, nearly always with a lack of 1p/19q deletions and with poor survival.16–18 In contrast, we have not found these patterns in our own histologically classic oligodendrogliomas, including those with pseudopalisading necrosis.9 Therefore, the possibility must be considered that some of the previously reported, genetically and clinically aggressive variants of anaplastic oligodendroglioma, in fact, are small cell astrocytomas.

Table 2. Features that Favor Small Cell Astrocytoma Diagnoses or Anaplastic Oligodendroglioma Diagnoses
FeatureSmall cell astrocytomaAnaplastic oligodendroglioma
  1. GFAP: glial fibrillary acidic protein; EGFR: epidermal growth factor receptor.

RadiologicRing enhancement; lack of enhancement despite high-grade histologyPatchy or nodular enhancement
HistologicOval rather than round nuclei; bland cytology despite markedly elevated mitotic/proliferative index; pseudopalisading necrosisUniformly round nuclei; epithelioid cells with clear-to-amphophilic cytoplasm, vesicular nuclei, and nucleoli; mucin-rich microcystic spaces; minigemistocytes
ImmunohistochemicalGFAP-positive cytoplasmic processes; EGFR-vIII immunoreactivityGFAP-negative or GFAP-positive minigemistocytes and gliofibrillary oligodendrocytes
GeneticEGFR gene amplification; chromosome 10q deletion1p/19q codeletions

Another interesting finding in our series was the surprisingly common radiologic presentation as a subtle, nonenhancing mass, prompting consideration of nonneoplastic conditions, such as cerebral infarct and encephalitis. Because the radiology studies from our patients were performed at many different institutions, it is possible that some of this represents variabilities in volumes of contrast material administered and/or the interval between administration and imaging. Nevertheless, this seems unlikely, because one-third of our patients similarly showed no evidence of endothelial hyperplasia or necrosis, thus qualifying only as anaplastic astrocytoma (Grade III) using strict WHO grading criteria. Sampling error similarly seems unlikely, because the vast majority of our material consisted of large biopsy or resection specimens. Finally, follow-up neuroimaging studies (typically at 2–3 months after the initial diagnosis) showed that the majority of such patients quickly developed the full-blown ring-enhancement and rapid clinical progression typical of GBM. Grading is problematic in such patients, and our data suggest that such neoplasms simply may represent the earliest phase of primary GBM, perhaps becoming symptomatic earlier than most, due to the high proliferative index and rapid growth. In other words, they are glioblastomas molecularly, but not yet histologically. Therefore, in our pathology reports, we typically add a comment stating that the small cell variant, particularly when accompanied by EGFR amplification and/or 10q deletion, generally behaves like a GBM, even though it may not be possible to designate the tumor as Grade IV purely on histologic grounds. This raises a philosophic dilemma regarding whether all small cell astrocytomas should be diagnosed as Grade IV (i.e., glioblastoma) by definition, based primarily on the length of patient survival, as suggested by our own data, compared with reserving the Grade IV designation solely for tumors that have with microvascular proliferation or necrosis, as practiced currently for all other forms of diffuse astrocytoma. This nomenclature issue is beyond the scope of the current study and likely will require a consensus decision at the next WHO classification meeting.

Finally, the remarkably common expression of EGFR amplification and EGFR-vIII expression in our patients also has potential interest from a therapeutic perspective, because that there are now a number of clinical trials specifically targeting this receptor, including the constitutively activated vIII mutant. If these therapeutic modalities are effective, then it may become even more critical to recognize the small cell astrocytoma variant and to apply ancillary genetic studies. In summary, we conclude that 1) small cell astrocytoma is an aggressive histologic variant that behaves like primary GBM, even in the absence of endothelial hyperplasia and necrosis; 2) despite considerable morphologic overlap with anaplastic oligodendroglioma, the clinicopathologic and genetic features of these two gliomas are distinct; 3) 50% of small cell astrocytomas express the constitutively activated EGFR-vIII mutant; and 4) molecular testing for 10q deletion improves the diagnostic sensitivity over EGFR alone.


The authors thank all of their colleagues who submitted cases for review and provided clinical follow-up data. The authors also thank Ruma Banerjee for her expert assistance with the fluorescence in situ hybridization assays performed in the current study.