Clinical association analysis of ependymomas and pilocytic astrocytomas reveals elevated FGFR3 and FGFR1 expression in aggressive ependymomas

This study was approved by the Ethical Committee of Tampere University Hospital and the National Authority for Medico-legal Affairs in Finland. The study cohort included 108 ependymal tumors from 88 patients, 97 pilocytic astrocytomas from 97 patients (Table 1).

Ependymoma patients underwent neurosurgical operation with the intention of gross radical tumor resection between 1984 and 2009 at Tampere University Hospital, between 1979 and 1998 at Kuopio University Hospital, and between 1986 and 1999 at Turku University Hospital, Finland. The clinical data detail about radicality of tumor resection is imperfect, but radical resection has always been performed when possible for the patient. Grade I tumors included 17 myxopapillary ependymomas and 1 subependymoma. Grade II tumors included 68 ependymomas, while Grade III tumors included 22 anaplastic ependymomas, as classified according to WHO criteria [ 23].

Pilocytic astrocytoma patients underwent tumor surgery at the Tampere University Hospital between 1985 and 1999, at the Kuopio University Hospital between 1980 and 1992, at the Turku University Hospital between 1981 and 1992, and at the Helsinki University Hospital between 1986 and 1993.

Tumor samples were fixed in formaldehyde (buffered with 4% phosphate) and embedded in paraffin. The samples were processed into paraffin blocks and sections were stained with hematoxylin and eosin (H&E). Histopathological typing and grading, evaluation, and identification of histologically representative tumor regions on each slide were performed by an experienced neuropathologist. Tissue microarray (TMA) blocks were constructed using representative sample regions and a custom-built instrument (Beecher Instruments, Silver Spring, MD, USA). The diameter of the tissue core on the microarray block was 0.6 or 1 mm, depending on the TMA type. Five-micrometer-thick sections were cut from representative array paraffin blocks.

Paraffin was removed with hexane. After rehydration in ethanol, the pre-processing stage was performed using Target Retrieval Solution citrate buffer (Dako). The samples were stained using rabbit monoclonal FGFR1 antibody (#9740, Cell Signaling Technology, 1:100 dilution) and mouse monoclonal FGFR3 antibody (sc-13,121, Santa Cruz Biotechnology, 1:600 dilution). ‘Envision + System-horseradish peroxidase and diaminobenzidine (DAB)’ kit (Dako) was used for FGFR3. The nuclei were stained with hematoxylin. A mouse monoclonal antibody MIB-1 (Ki-67 antigen, dilution 1:40, Immunotech, S.A. Marseille, France) was used to analyze cell proliferation. The tissue sections were counterstained with methyl green. The percentage of tissue MIB-1-positive nuclei was quantitatively evaluated using a computer-assisted image analysis system (CAS-200 TM Software, Becton Dickinson & Co., USA) and ImmunoRatio analysis. Only neoplastic cells were included in the analysis (necrotic and hemorrhagic areas were omitted).

The intensity of FGFR3 and FGFR1 immunopositivity was scored by two observers (HH and KG) on a scale from 0 to 3 as follows: 0 (no staining), 1 (weak immunostaining), 2 (moderate immunostaining), or 3 (strong immunostaining).

All data were analyzed using R packages or IBM SPSS statistics 21.0 software (SPSS Inc., Chicago, IL, USA) for Windows. Tests for pairwise association between discrete variables were performed using Fisher’s exact test for count data. For tables larger than 2 × 2, the p-values of Fisher’s exact tests were calculated using Monte Carlo simulation with 2.5*10^7 replicates. p-values were not corrected for multiple testing. Log-rank test was used for the analysis of prognostic factors. In cox regression analysis, cox model was built using a stepwise forward likehood-ratio testing.

All the tissue samples were formalin fixed and paraffin embedded (FFPE). A turXTRAC FFPE DNA kit (Covaris) or AllPrep DNA/RNA Mini Kit (Qiagen) was used for DNA isolation. We used 1 μg of extracted DNA for targeted sequencing using the Sureselect XT Target enrichment system together with custom-designed RNA probes (Additional file 1: Table S1). The sequencing library was prepared according to the kit instructions (200 ng of DNA samples) with a shorter DNA-shearing protocol (220 s) and sequenced with MiSeq (Illumina). Tumors Epe002 and Epe003 were derived from the first and the third tumor surgery (after second recurrence) of one patient. In addition, the tumors Epe004 (1st tumor surgery) and Epe005 (2nd tumor surgery) were derived from a separate ependymoma patient.

The resulting data were aligned against the GRCh37 human reference genome using Bowtie 2.2.4 [ 24]. Mutations were identified in tumor samples by searching for sites with an alternate allele fraction of at least 10%, and at least 5 reads with the mutation. Additionally, the allele fraction was required to be 20 times higher than the background error rate (i.e., the average allele fraction across control blood samples from healthy patients). Protein-level consequences of variants were predicted using ANNOVAR software tool [ 25]. Mutations with a known or suspected pathological function were identified manually. To discover chromosomal rearrangements for fusion detection, unaligned reads from each sample were split into two 30 bp anchors (one from both ends) that were aligned to the hg38 genome using Bowtie-1.1.2. Discordant anchor pairs were grouped by position, and groups with 8 or more supporting reads were flagged as rearrangement candidates and manually curated using IGV and BLAT.

Log ratios of amplicon read counts were used for DNA copy number calling. Differences in average coverage between samples were corrected on the basis of control amplicons in chromosomes 5, 8, 11, and 18 (14–21 amplicons per chromosome), positioned in regions with the lowest rate of reported copy number alterations. Blood-derived DNA from healthy individuals was used as a negative control for the copy number analysis.

Immunohistochemistry was used to investigate FGFR3 expression levels in 108 ependymal tumor samples applied to TMAs. The TMA cohort (Table 1), representing different grades of ependymomas and disease subtypes, has been partly reported previously [ 26]. FGFR3 immunoreactivity was detected in 27 (37%) of the cases; 11 (15%) showed weak immunostaining, 11 (15%) showed moderate immunostaining and 5 (7%) were strongly immunopositive. Increased staining was also observed in pseudorosette structures (Additional file 1: Figure S1b). Recurrent tumors showed typically similar staining levels as the primary tumor. With respect to the association analysis (Additional file 1: Figure S2), FGFR3 staining was significantly associated with a higher tumor grade ( p < 0.01, Fisher’s exact test, Fig. 1b, Table 2). None of the grade I cases showed detectable FGFR3 expression. Moderate-to-strong FGFR3 immunostaining was predominantly detected in cerebral tumors as compared to other locations ( p < 0.001, Fisher’s exact test, Fig. 1c, Table 2). Elevated FGFR3 immunopositivity in high-grade cerebral tumors suggests that FGFR3 immunostaining may be typical for pediatric ependymomas. Indeed, patients with age < 20 years at tumor onset had a higher frequency of FGFR3 immunopositive staining ( p < 0.05, Fisher’s exact test, Fig. 1d). Cases with moderate-to-strong FGFR3 immunostaining tend to show a high proliferation rate (Fig. 1e), although this association was not statistically significant ( p = 0.07, Fisher’s exact test). Importantly, moderate-to-strong FGFR3 immunostaining was significantly associated with shorter overall patient survival ( p < 0.05, log-rank test, Fig. 1f) and shorter time to tumor recurrence ( p < 0.01, log-rank test, Fig. 1g). The association with disease-free survival remained significant after adjustment for tumor location, grade, and proliferation ( p = 0.003, RR = 1.82, 95% CI 1.23–2.68 for FGFR3, other variables not significant in the final equation, N = 77, stepwise Cox regression), but only tumor location ( p = 0.022, RR = 2.47, 95% CI 1.42–5.34, N = 77, stepwise Cox regression) was a significant prognostic predictor for disease-specific survival in multifactorial analysis. It is relevant to note the patient numbers ( N = 77) are rather low for multifactorial analysis using four different variables. Still, the obtained results suggest that FGFR3 immunopositivity is associated with more aggressive ependymomas.

As pediatric and adult ependymomas differ in many respects and the age association might influence the observed associations, we analyzed the pediatric and adult sample cohorts independently. Patients that were at least 16 years old were considered as adults according to general practice in Finnish pediatric clinics. There were 35 pediatric and 73 adult samples in our cohort. Moderate-to-strong FGFR3 staining was slightly more common in pediatric than adult samples (34.3% vs 13.7%, p = 0.055, Fisher’s exact test, Table 2). In pediatric patients, moderate FGFR3 immunostaining was observed in cerebellar (31%, n = 16) and cerebral (29%, n = 14) tumors and strong FGFR3 staining only in cerebral tumors (21%, n = 14), whereas all the spinal cases ( n = 5) were negative for FGFR3 ( p = 0.065, Fisher’s exact test). FGFR3 staining was not associated with tumor grade or proliferation index in pediatric ependymomas. In adults, FGFR3 associations were largely very similar as in the whole sample cohort: stronger FGFR3 staining was associated with tumor grade ( p < 0.01, n = 73, Fisher’s exact test), tumor location ( p < 0.001, n = 71, Fisher’s exact test) and there was a close-to-significant association with proliferation index ( p = 0.095, n = 66, Fisher’s exact test). Prognostic associations were mostly nonsignificant in separate survival analyses in pediatric ( n = 14) and adult ( n = 30) sample cohorts, but this was likely due to low sample count in the analysis, as the trend remain the similar. Of note, when FGFR3 staining was divided into four groups, it was associated with worse disease-specific ( p < 0.01, log-rank test) and disease-free ( p < 0.001, log-rank test) survival in pediatric patients.

The interpretation of the FGFR1 immunostaining data was not as straightforward as FGFR3 staining, partly because macrophages, neurons, and necrotic areas showed immunopositive staining. Therefore, FGFR1 immunohistochemical scoring was based on the presence of FGFR1-positive malignant cell clusters or larger tumor areas (i.e. diffuse staining), and scoring of individual cells was omitted in the analysis. Sporadic moderate-to-strong FGFR1 immunopositivity was also detected and characterized by high outlier expression in individual malignant cells. These observations support those from previous reports [ 27]. FGFR1 staining was detected in the cytoplasm and membrane compartments, while granular staining was also observed in a subpopulation of positively-stained samples. Interestingly, moderate-to-strong FGFR1 immunostaining was also observed in ependymal rosettes (Additional file 1: Figure S3).

Diffuse FGFR1 immunoreactivity was detected in 42 (58%) of ependymal tumors. Twenty-four cases (33%) showed weak immunostaining, 15 (21%) cases showed moderate immunoreactivity, and 3 (4%) cases showed strong immunopositivity (Fig. 2a). Consistent with FGFR3 expression, FGFR1 immunostaining was significantly associated with a higher tumor grade ( p < 0.05, Fisher’s exact test, Fig. 2b, Table 2) and cerebral location ( p < 0.01, Fisher’s exact test, Fig. 2c, Table 2). Diffuse FGFR1 staining was not significantly associated with overall or recurrence-free survival but cases with high FGFR1 expression had a tendency toward decreased survival rates in this cohort (Additional file 1: Figure S4). When ependymomas were divided into pediatric ( n = 34) and adult ( n = 72) patients, no associations were observed for FGFR1 in the pediatric cohort. However, FGFR1 staining was similarly associated with tumor location ( p < 0.001, n = 70, Fisher’s exact test) and higher tumor grade ( p < 0.01, n = 72, Fisher’s exact test) in the adult cohort as in the whole sample cohort. Furthermore, a weak association was observed between stronger FGFR1 staining and higher tumor proliferation index ( p = 0.061, n = 68, Fisher’s exact test) among adult patients.

Among ependymomas, marked (moderate-to-strong) immunostaining for FGFR1, FGFR3, or both proteins occurred more frequently in cerebral than in non-cerebral tumors (76, 32, and 19% in cerebral, cerebellar, and spinal tumors, respectively, p < 0.001, Fisher’s exact test, Fig. 2d). Increased FGFR1 and/or FGFR3 expression was therefore a common characteristic of cerebral tumors. Strikingly, tumor tissues expressing marked (moderate-to-strong) levels of both FGFR1 and FGFR3 were associated with significantly worse patient survival than tissues obtained from other cases, in terms of both overall mortality ( p < 0.05, log-rank test, Fig. 2e) and recurrence-free survival ( p < 0.05, log-rank test, Fig. 2f). Furthermore, the combined variable for FGFR1 and FGFR3 (both are negative-to weak, either staining is moderate-to-strong or both are moderate-to-strong) was the only significant predictor for the disease-specific survival ( p = 0.014, RR = 1.91, 95% CI 1.14–3.20, N = 77, stepwise Cox regression) and disease-free survival ( p = 0.007, RR = 1.75, 95% CI 1.17–2.62, N = 77, stepwise Cox regression), when it was combined together with tumor location, grade, and proliferation index as explanatory factors in the multifactorial analysis. It is good to remember that the patient numbers ( N = 77) are rather low for multifactorial analysis using four different variables when interpreting these results. Still, the obtained results support the aggressive nature of tumors with moderate-to-strong staining of both FGFR1 and FGFR3. Our results are also concordant with previous notions (e.g. [ 28]) that supratentorial and infratentorial ependymomas are largely different and appear to represent distinct tumor entities.

In the pilocytic astrocytoma cohort, 60 (82%) samples were negative for FGFR3 expression, while only 21 cases (22%) failed to show any FGFR1 expression (Fig. 3c-d). Among samples with FGFR3 immunoreactivity, 7 samples (9%) showed weak immunostaining, 5 samples (6%) showed moderate immunostaining, and 2 samples (3%) were strongly immunopositive. Immunopositive FGFR3 staining was detected in both microcystic and pilocytic areas. Among samples with positive FGFR1 staining, 59 samples (61%) showed weak immunopositivity, 16 samples (16%) samples showed moderate immunopositivity, and 1 sample (1%) was strongly immunopositive. Moderate-to-strong FGFR1 immunostaining was detected predominantly in microcystic areas. Clinical association analysis (Additional file 1: Figure S5) did not reveal any significant associations between FGFR1 staining and other clinical factors. Interestingly, moderate-to-strong FGFR3 protein levels were associated with increased patient age (≥16 years, p < 0.01, Fisher’s exact test, Fig. 3e). All but one of the six primary cases showing moderate-to-strong FGFR3 immunostaining were from patients who were at least 15 years old. FGFR3 immunostaining was not associated with tumor location or aneuploidy.

Ten tumors showing moderate-to-strong FGFR1 or FGFR3 immunostaining were selected for targeted sequencing analysis. All analyzed ependymomas were supratentorial. In addition to FGFR3 and FGFR1, the sequencing panel incorporated genes with reported alterations in gliomas, including IDH1, IDH2, TP53, ATRX, CIC, CDKN2A, RB1, RELA, and BRAF (Additional file 1: Table S1). We did not detect FGFR coding mutations or fusions in any of the samples (Fig. 4, Additional file 2: Table S2, Additional file 1: Figure S6). FGFR3 fusions were detected with high sensitivity from large diffuse glioma cohort using the same sequencing panel and methodology [ 34], suggesting that the lack of detectable FGFR fusions was not due to methodological limitations. The tumors selected for analysis contained many known alterations, including a C11orf95-RELA fusion and CDKN2A alterations in ependymoma tumors (Epe001, Epe002 and Epe003). RELA fusions and loss of CDKN2A have been routinely observed in aggressive ependymomas [ 17, 29, 30]. A TERT promoter mutation was observed in tumors Epe004 and Epe005 obtained from the same ependymoma patient. In addition, one pilocytic astrocytoma tumor harbored the KIAA1549-BRAF fusion, which is the most frequent MAPK pathway alteration in this tumor type [ 7]. It is interesting that majority of sequenced PA samples did not carry any BRAF or FGFR1 alterations, but limited sample size does not allow full generalization of this result. A total of 4 cases in our cohort did not carry any alterations in targeted genes. This may be due, in part, to the fact that all genomic regions were not covered during targeted sequencing. In addition, pilocytic astrocytomas are known to harbor very few alterations [ 7].