Abstract
Purpose: The high incidence of recurrence and unpredictable clinical outcome for pediatric ependymoma reflect the imprecision of current therapeutic staging and need for novel risk stratification markers. We therefore evaluated 1q25 gain across three age- and treatment-defined European clinical trial cohorts of pediatric intracranial ependymoma.
Experimental Design: Frequency of 1q gain was assessed across 48 ependymomas (42 primary, 6 recurrent) using Affymetrix 500K single-nucleotide polymorphism arrays. Gain of 1q25 was then evaluated by interphase FISH across 189 tumors treated on the Children's Cancer Leukaemia Group/International Society for Pediatric Oncology (SIOP) CNS9204 (n = 60) and BBSFOP (n = 65) adjuvant chemotherapy trials, or with primary postoperative radiotherapy (SIOP CNS9904/RT, n = 64). Results were correlated with clinical, histologic, and survival data.
Results: Gain of 1q was the most frequent imbalance in primary (7/42, 17%) and recurrent ependymomas (2/6, 33%). Gain of 1q25 was an independent predictor of tumor progression across the pooled trial cohort [HR = 2.55; 95% confidence interval (CI): 1.56–4.16; P = 0.0002] and both CNS9204 (HR = 4.03; 95% CI: 1.88–8.63) and BBSFOP (HR = 3.10; 95% CI: 1.22–7.86) groups. The only clinical variable associated with adverse outcome was incomplete tumor resection. Integrating tumor resectability with 1q25 status enabled stratification of cases into disease progression risk groups for all three trial cohorts.
Conclusions: This is the first study to validate a prognostic genomic marker for childhood ependymoma across independent trial groups. 1q25 gain predicts disease progression and can contribute to patient risk stratification. We advocate the prospective evaluation of 1q25 gain as an adverse marker in future international clinical trials. Clin Cancer Res; 18(7); 2001–11. ©2012 AACR.
This article is featured in Highlights of This Issue, p. 1819
Because current clinicopathologic classification criteria for pediatric intracranial ependymoma are inconsistent, the introduction of novel prognostic markers for therapeutic stratification is an important requirement of future clinical trials. In this study of age- and treatment-defined trial cohorts, 1q25 gain was identified as an independent and reproducible marker of intracranial ependymoma progression in all patients, particularly in the younger children treated according to European primary postoperative chemotherapy protocols. Furthermore, incorporating degree of surgical resection with tumor 1q25 status enabled patient stratification according to disease progression risk groups across all 3 trial cohorts, irrespective of patient age or adjuvant therapy administered. We therefore advocate the prospective evaluation of 1q25 gain as an adverse risk marker in future international trials.
Introduction
Improvements in the risk stratification and treatment of several cancers have been achieved in the postgenomic era through an appreciation of tumor-specific molecular abnormalities. Although our understanding of ependymoma biology has advanced in recent years with respect to tumor initiation and heterogeneity (1, 2), the development of novel prognostic classifications and targeted therapies is still required to enhance patient outcome for this tumor group, particularly in children.
Ependymomas represent the third most common pediatric tumor of the central nervous system (3). Although able to arise at any age, the majority occurs in children aged below 5 years (4). Significant differences are now apparent in the clinical and biologic characteristics of childhood versus adult ependymomas (5). Presently, prognostication for pediatric ependymoma is based solely on clinical parameters. Of these, the extent of primary tumor resection remains the most consistently reported correlate of outcome (4). European treatment schedules have hitherto been stratified by age. Trials of adjuvant chemotherapy for young children were initiated because of concerns of radiation-induced neuropsychologic and cognitive damage to the immature central nervous system (6–8), whereas postoperative radiotherapy has been reserved for older children (8). Despite these measures, the prognosis for pediatric intracranial ependymomas remains poor when compared with other childhood malignancies, with local tumor recurrence a frequently reported event, even after complete tumor excision (5). After 5 years, progression-free survival (PFS) rates range from 23% to 74% (3, 9, 10), whereas mortality is reported in up to 40% of affected children (3).
The need to incorporate novel biomarkers into future prognostic stratifications for childhood intracranial ependymoma is therefore apparent. However, although several candidates have been proposed, markers showing reproducible results in sizeable groups of young ependymoma patients are lacking (5). Indeed, several purported biologic prognostic markers in ependymoma have been shown to lose this capacity when assessed across clinical trial cohorts (11), highlighting the importance of analyzing standardized therapeutic groups.
Copy number gain of chromosome 1q has been reported as a frequent genetic aberration in both primary and recurrent childhood intracranial ependymomas (5). Retrospective analyses of cohorts comprising children and adults have identified gain of either the entire long arm or the 1q25 amplicon as adverse prognostic markers in intracranial ependymoma (12–14), although little evidence exists for such a role exclusive to a pediatric setting (15).
In this study, we established the frequency of 1q gain in pediatric ependymoma, identifying 1q21–25 among the most common subregions of gain. We then evaluated 1q25 gain as a robust prognostic marker in pediatric intracranial ependymoma by carrying out interphase FISH (iFISH) across 189 primary tumors, incorporating 3 European clinical trial cohorts. To our knowledge, this is the first study to assess the reproducibility of a genomic marker in both comparable (CNS9204 and BBSFOP) and contrasting (CNS9904) therapeutic trial groups of pediatric intracranial ependymoma patients.
Methods
Patients and clinical specimens
Forty-eight snap-frozen ependymomas (42 primary, 6 first recurrent) from 42 patients were obtained from Children's Cancer and Leukaemia Group (CCLG) registered centers in the United Kingdom for analysis using Affymetrix 500K single-nucleotide polymorphism (SNP) arrays. Constitutional blood samples from 38 of 42 (90%) patients contributing tumors were analyzed as controls. From the tumor cohort, a subset of 18 formalin-fixed paraffin-embedded (FFPE) intracranial ependymomas were used to validate microarray 1q gain results by iFISH (see below). Fifteen of these samples were also included in the clinical trial cohort analysis.
A total of 172 FFPE primary intracranial ependymomas from trial patients were analyzed by iFISH on tissue microarrays (TMA). Patients were enrolled in either the CNS9204 (ref. 6; n = 60), BBSFOP (ref. 7; n = 65) or CNS9904 clinical trials (n = 47) and were diagnosed between 1989 and 2007. An overview of each trial is provided (Supplementary Methods). To supplement the CNS9904 cohort, 17 primary tumors (9 supratentorial, 8 posterior fossa) from therapeutically matched, nontrial patients were also examined. These children were aged between 5 and 14 years and had intracranial ependymomas treated only with cranial irradiation (54 Gy) following primary surgery.
Patient clinical information was obtained from respective trial centers. For all cases, central pathologic review was done according to WHO criteria (DWE, FA, PV; ref. 16). Cases with differing pathologic diagnoses at review were resolved by consensus opinion following discussion between responsible neuropathologists. The degree of surgical resection was evaluated by central review of postoperative imaging according to International Society for Pediatric Oncology (SIOP) guidelines (17). The study obtained CCLG, Société Française d'Oncologie Pédiatrique (SFOP), SIOP, and Multiple Centre Research Ethics Committee (MREC) approval. Consent for tumor tissue use was taken in accordance with national tumor banking procedures (uk:05/MRE/04/70).
Nucleic acid isolation
DNA was extracted from 10 mg of frozen tumor tissue and peripheral blood mononuclear cells as described previously (18). Before tissue extraction, hematoxylin/eosin stained smears from each specimen underwent pathologic review to confirm tumor presence and viability.
500K SNP array analysis
SNP microarray profiles for tumor and constitutional DNA were generated using the Affymetrix GeneChip Human Mapping 500K assay, with data analysis and visualization done as described previously (Supplementary Methods; refs. 1, 18, 19, 20). Chromosomal arms and cytobands were defined as gained or lost if more than 80% of encompassed probes showed copy number gain or loss, as defined previously (18). The microarray data generated during this study has been deposited in GEO with an accession number GSE32101.
Ependymoma TMA construction
TMAs were constructed from blocks of FFPE tumor material. Viable and representative tumor areas were identified by a neuropathologist using hematoxylin/eosin stained sections from each block before TMA incorporation (JL, DE, FA, and PV). Typically for each tumor sample, 3 to 4 × 0.6 mm cores of 4 μm thickness were included, incorporating the different representative areas defined.
Interphase fluorescence in situ hybridization
Dual color iFISH was carried out as described previously (21), using a commercial 1q25 (spectrum green) and 1p36 (spectrum orange) probe (Vysis). A commercial probe was chosen in view of the need for a prognostic biologic marker to be robust and widely available for multicentre application. The evaluation criteria and scoring system adopted was based on that used by several preceding analyses (Supplementary Methods; refs. 13, 14, 22).
Statistics
Statistical analysis was carried out in SPSS (version 17.0, SPSS) and in SAS, Version 9.1.2 (SAS Institute Inc.). A detailed definition of analyses used is provided (Supplementary methods; refs. 23).
Results
500K SNP array analysis
Clinical characteristics of the SNP array ependymoma cohort are summarized in Table 1, with results from survival analysis shown in Table 2 (Comprehensive clinicopathologic data and chromosome arm imbalance results for each tumor sample are provided in Supplementary Table S1). The median age of the primary tumor cohort was 6.8 years (range: 1–20.9 years) with a male:female ratio of 1.2:1. Children with posterior fossa ependymomas were significantly younger than those with spinal tumors (ANOVA with Tukey HSD test; P = 0.009, eta 0.2), whereas the age difference between patients with posterior fossa and supratentorial tumors was not significant. The median follow-up period for all 42 patients was 9.6 years (range: 0.5–21 years). Disease progression occurred in half of the cohort with a median time to progression of 1.5 years (range: 0.3–8.8 years), whereas 12 patients died with a median survival time of 3.0 years (range: 1–9.6 years). Incomplete resection was the only clinicopathologic variable to confer an adverse prognosis, associated independently with a worse PFS [HR = 3.19; 95% confidence interval (CI) = 1.26–8.08; P = 0.01].
Patient data . | . | 500K SNP array cohort (42 patients) . |
---|---|---|
Age | ||
<5 y | 18 (43) | |
>5 y | 24 (57) | |
Gender | ||
Male | 23 (55) | |
Female | 19 (45) | |
Five-year PFS | 38 ± 9% | |
Five-year OS | 78 ± 8% | |
Survival status | ||
Alive | 30 (71) | |
Dead | 12 (29) | |
Tumor data | Primary tumors (n = 42) | First recurrent tumors (n = 6) |
Location | ||
PF | 24 (57) | 3 (50) |
ST | 12 (29) | 3 (50) |
Spinal | 6 (14) | — |
WHO grade | ||
III | 16 (38) | 2 (33) |
II | 23 (55) | 4 (67) |
I | 3 (7) | — |
Surgical resection | ||
Complete | 21 (50) | 1 (17) |
Incomplete | 21 (50) | 3 (50) |
Unknown | 2 (33) | |
Adjuvant therapy | ||
RT | 12 (28) | 2 (33) |
CT | 15 (36) | 1 (17) |
Both | 10 (24) | 3 (50) |
Nil | 5 (12) | — |
1q gain | ||
No | 35 (83) | 4 (67) |
Yes | 7 (17) | 2 (33) |
Patient data . | . | 500K SNP array cohort (42 patients) . |
---|---|---|
Age | ||
<5 y | 18 (43) | |
>5 y | 24 (57) | |
Gender | ||
Male | 23 (55) | |
Female | 19 (45) | |
Five-year PFS | 38 ± 9% | |
Five-year OS | 78 ± 8% | |
Survival status | ||
Alive | 30 (71) | |
Dead | 12 (29) | |
Tumor data | Primary tumors (n = 42) | First recurrent tumors (n = 6) |
Location | ||
PF | 24 (57) | 3 (50) |
ST | 12 (29) | 3 (50) |
Spinal | 6 (14) | — |
WHO grade | ||
III | 16 (38) | 2 (33) |
II | 23 (55) | 4 (67) |
I | 3 (7) | — |
Surgical resection | ||
Complete | 21 (50) | 1 (17) |
Incomplete | 21 (50) | 3 (50) |
Unknown | 2 (33) | |
Adjuvant therapy | ||
RT | 12 (28) | 2 (33) |
CT | 15 (36) | 1 (17) |
Both | 10 (24) | 3 (50) |
Nil | 5 (12) | — |
1q gain | ||
No | 35 (83) | 4 (67) |
Yes | 7 (17) | 2 (33) |
NOTE: The values in parenthesis are given in percentage.
Abbreviations: PF, posterior fossa/infratentorial; ST, supratentorial; RT, radiotherapy; C, chemotherapy.
Factor (numbers) . | Progression-free survival . | Overall survival . | ||||||
---|---|---|---|---|---|---|---|---|
. | Univariate . | Multivariable . | Univariate . | Multivariable . | ||||
. | HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . |
Patient age | ||||||||
>5 y (n = 18) | 1 | 1 | ||||||
<5 y (n = 24) | 1.33 (0.54–3.23) | 0.54 | 1.29 (0.38–4.35) | 0.68 | ||||
Gender | ||||||||
Female (n = 19) | 1 | 1 | ||||||
Male (n = 23) | 1.07 (0.45–2.56) | 0.88 | 0.77 (0.25–2.40) | 0.65 | ||||
Location | ||||||||
ST/SP (n = 18) | 1 | 1 | ||||||
PF (n = 24) | 0.83 (0.31–2.22) | 0.71 | 0.93 (0.28–3.09) | 0.90 | ||||
WHO grade | ||||||||
I/II (n = 26) | 1 | 1 | ||||||
III (n = 16) | 1.19 (0.49–2.86) | 0.69 | 1.17 (0.79–3.69) | 0.78 | ||||
Surgery | ||||||||
CR (n = 21) | 1 | 1 | 1 | 1 | ||||
IR (n = 21) | 3.19 (1.26–8.08) | 0.01 | 3.19 (1.26–8.08) | 0.01 | 2.00 (0.63–6.31) | 0.24 | 1.74 (0.54–5.65) | 0.36 |
Adjuvant Radiotherapy | ||||||||
No (n = 20) | 1 | 1 | 1 | |||||
Yes (n = 22) | 0.66 (0.27–1.61) | 0.36 | 0.43 (0.13–1.42) | 0.17 | 0.24 (0.06–1.03) | 0.06 | ||
1q gain result | ||||||||
No gain (n = 35) | 1 | 1 | 1 | |||||
Gain (n = 7) | 1.73 (0.66–4.51) | 0.26 | 2.60 (0.73–9.26) | 0.14 | 4.62 (0.99–21.20) | 0.05 |
Factor (numbers) . | Progression-free survival . | Overall survival . | ||||||
---|---|---|---|---|---|---|---|---|
. | Univariate . | Multivariable . | Univariate . | Multivariable . | ||||
. | HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . |
Patient age | ||||||||
>5 y (n = 18) | 1 | 1 | ||||||
<5 y (n = 24) | 1.33 (0.54–3.23) | 0.54 | 1.29 (0.38–4.35) | 0.68 | ||||
Gender | ||||||||
Female (n = 19) | 1 | 1 | ||||||
Male (n = 23) | 1.07 (0.45–2.56) | 0.88 | 0.77 (0.25–2.40) | 0.65 | ||||
Location | ||||||||
ST/SP (n = 18) | 1 | 1 | ||||||
PF (n = 24) | 0.83 (0.31–2.22) | 0.71 | 0.93 (0.28–3.09) | 0.90 | ||||
WHO grade | ||||||||
I/II (n = 26) | 1 | 1 | ||||||
III (n = 16) | 1.19 (0.49–2.86) | 0.69 | 1.17 (0.79–3.69) | 0.78 | ||||
Surgery | ||||||||
CR (n = 21) | 1 | 1 | 1 | 1 | ||||
IR (n = 21) | 3.19 (1.26–8.08) | 0.01 | 3.19 (1.26–8.08) | 0.01 | 2.00 (0.63–6.31) | 0.24 | 1.74 (0.54–5.65) | 0.36 |
Adjuvant Radiotherapy | ||||||||
No (n = 20) | 1 | 1 | 1 | |||||
Yes (n = 22) | 0.66 (0.27–1.61) | 0.36 | 0.43 (0.13–1.42) | 0.17 | 0.24 (0.06–1.03) | 0.06 | ||
1q gain result | ||||||||
No gain (n = 35) | 1 | 1 | 1 | |||||
Gain (n = 7) | 1.73 (0.66–4.51) | 0.26 | 2.60 (0.73–9.26) | 0.14 | 4.62 (0.99–21.20) | 0.05 |
NOTE: Probability (P) values for univariate and multivariable survival analysis obtained by the Cox proportional hazard model (see Supplementary Methods).
Abbreviations: PF, posterior fossa; ST, supratentorial; SP, spinal; IR, incomplete resection; CR, complete resection.
In keeping with previous comparative genomic hybridization (CGH) studies of ependymoma (13–15, 24), the SNP array analysis categorized primary tumors according to their broad genomic imbalance profile. Seven tumors (17%) showed 4 or more chromosomal aberrations, 11 tumors (26%) revealed 1 to 3 imbalances, whereas 24 tumors (57%) showed no whole chromosome or arm imbalance. Within this latter group, 15 ependymomas (36%) had a high-resolution balanced profile (≥95% of all SNP probes showing a diploid copy number) and were associated with children aged below 3 years (Fisher's exact test; P = 0.04). Even when accounting for different tumor location by restricting the analysis to posterior fossa ependymomas, the number of chromosome arm imbalances between patients aged below and above 5 years remained significantly different (Wilcoxon rank sum test; P = 0.0001).
Gain of chromosome 1q was the most frequent aberration in both the primary and recurrent ependymomas [7/42 (17%) and 2/6 (33%) respectively], identified in 7 patients. Higher resolution cytoband analysis revealed 1q21–25, 1q32, and 1q42–44 to be amongst the most frequently gained subregions on this arm (11/48, 23%). Whole gains of chromosomes 9 and 18 were also relatively common, seen in 6 of 42 (14%) primary tumors. The most frequent loss was of chromosome 22q, present in 3 of 42 primary tumors (7%) and 1 recurrent case.
Distinct patterns of genomic imbalance between primary ependymomas from different CNS locations were evident (Supplementary Fig. S1). Gain of chromosome 1q was associated with posterior fossa ependymomas (Fisher exact test; P = 0.03). Relatively few broad chromosomal changes were seen with supratentorial tumors. In contrast, spinal ependymomas were characterized by numerous arm and whole chromosomal aberrations when compared with intracranial tumors, particularly gain of chromosomes 17, 20p, 16, 12q, 20q, 21q, 9, and 18.
In view of its frequency across primary and recurrent ependymomas, the impact of entire chromosome 1q gain on patient survival was assessed, revealing a trend toward predicting worse overall survival (OS) in multivariable analysis (Table 2; HR = 4.62; 95% CI = 0.99–21.20; P = 0.05). FISH was used to validate the SNP array copy number results for 1q25 copy number gain across 18 primary intracranial ependymomas (Supplementary Fig. S2; Spearman's rank = 0.79; P < 0.0001).
1q25 FISH trial cohort analysis
Clinicopathologic data and 1q25 FISH results for the 3 therapeutic trial cohorts are summarized in Table 3.
Variable . | CNS9204 . | BBSFOP . | CNS9904 + RT only . |
---|---|---|---|
Patient number | 60 | 65 | 64 |
Median age (range), y | 2.0 (0.3–3.1) | 1.9 (0.6–5.1) | 7.8 (3.0–16.7)a |
Sex | |||
Female | 21 (35%) | 30 (46%) | 26 (41%) |
Male | 39 (65%) | 35 (54%) | 38 (59%) |
Male:Female ratio | 1.9:1 | 1.2:1 | 1.5:1 |
Status | |||
Alive–no disease progression | 20 (33%) | 13 (20%) | 29 (45%) |
Alive–disease progression | 11 (18%) | 17 (26%) | 11 (17%) |
Death from disease | 29 (48%) | 34 (52%) | 23 (36%) |
Death unrelated to disease | 1 (2%) | 1 (2%) | |
Survival | |||
5-year PFS ± SE (%) | 38 ± 6 | 26 ± 6b | 47 ± 6 |
5-year OS ± SE (%) | 63 ± 6 | 58 ± 6 | 71 ± 6 |
Primary tumor location | |||
ST | 7 (12%) | 12 (18%) | 28 (44%)a |
PF | 53 (88%) | 53 (82%) | 36 (56%) |
Primary tumor WHO grade | |||
II | 33 (55%) | 11 (17%) | 30 (47%) |
III | 27 (45%) | 54 (83%)a | 34 (53%) |
Primary tumor surgical resection | |||
Complete | 28 (47%) | 43 (66%)c | 35 (55%) |
Incomplete | 32 (53%) | 22 (34%) | 29 (45%) |
Metastatic disease at presentation | |||
Yes | 4 (7%) | 2 (3%) | 2 (3%) |
No | 58 (93%) | 63 (97%) | 62 (97%) |
Scorable tumors (n = 147) | 52 (87%) | 41 (63%) | 54 (84%) |
1q25 result | |||
No gain | 41 (79%) | 35 (85%) | 41 (76%) |
Gain | 11 (21%) | 6 (15%) | 13 (24%) |
Variable . | CNS9204 . | BBSFOP . | CNS9904 + RT only . |
---|---|---|---|
Patient number | 60 | 65 | 64 |
Median age (range), y | 2.0 (0.3–3.1) | 1.9 (0.6–5.1) | 7.8 (3.0–16.7)a |
Sex | |||
Female | 21 (35%) | 30 (46%) | 26 (41%) |
Male | 39 (65%) | 35 (54%) | 38 (59%) |
Male:Female ratio | 1.9:1 | 1.2:1 | 1.5:1 |
Status | |||
Alive–no disease progression | 20 (33%) | 13 (20%) | 29 (45%) |
Alive–disease progression | 11 (18%) | 17 (26%) | 11 (17%) |
Death from disease | 29 (48%) | 34 (52%) | 23 (36%) |
Death unrelated to disease | 1 (2%) | 1 (2%) | |
Survival | |||
5-year PFS ± SE (%) | 38 ± 6 | 26 ± 6b | 47 ± 6 |
5-year OS ± SE (%) | 63 ± 6 | 58 ± 6 | 71 ± 6 |
Primary tumor location | |||
ST | 7 (12%) | 12 (18%) | 28 (44%)a |
PF | 53 (88%) | 53 (82%) | 36 (56%) |
Primary tumor WHO grade | |||
II | 33 (55%) | 11 (17%) | 30 (47%) |
III | 27 (45%) | 54 (83%)a | 34 (53%) |
Primary tumor surgical resection | |||
Complete | 28 (47%) | 43 (66%)c | 35 (55%) |
Incomplete | 32 (53%) | 22 (34%) | 29 (45%) |
Metastatic disease at presentation | |||
Yes | 4 (7%) | 2 (3%) | 2 (3%) |
No | 58 (93%) | 63 (97%) | 62 (97%) |
Scorable tumors (n = 147) | 52 (87%) | 41 (63%) | 54 (84%) |
1q25 result | |||
No gain | 41 (79%) | 35 (85%) | 41 (76%) |
Gain | 11 (21%) | 6 (15%) | 13 (24%) |
NOTE: The difference in median age between cohorts was assessed by the independent samples t-test. The difference in PFS between cohorts was assessed by the log rank test. Differences in the proportion of other clinical factors between cohorts, such as patient sex, tumor location, WHO grade and resection status, were assessed by the Fisher's exact test.
Abbreviations: PF, posterior fossa/infratentorial; ST, supratentorial.
aP < 0.0001 for comparison against remaining cohorts.
bP < 0.01 for comparison against remaining cohorts.
cP < 0.05 for comparison against remaining cohorts.
The median age of the CNS9204 cohort was 2.0 years (range: 0.3–3.1 years). The median follow-up period for all 60 patients was 8.9 years (range: 0.6–16.1 years). Disease progression occurred in 40 patients with a median time to progression of 1.6 years (range: 0.3–9.5 years). Death occurred in 29 patients with a median survival time of 3.3 years (range: 0.6–8.9 years). Four primary cases (7%) were metastatic at presentation.
The BBSFOP cohort was comparable with the CNS9204 group with respect to patient age (median 1.9 years, range: 0.6–5.1 years), gender, tumor location, and primary adjuvant therapy administered. However, WHO grade III ependymomas (Fisher exact test; P < 0.0001) and complete tumor resection (P = 0.032) were more prevalent in the French cohort, which had a poorer PFS. The median follow-up period across the entire BBSFOP cohort was 7.7 years (range: 0.3–16.9 years). Fifty-one patients had suffered disease progression, with a median time to progression of 1.3 years (range: 0.3–9.3 years). Thirty-four patients had died of disease [median survival time 3.4 years (range: 0.3–10.1 years)]. Two primary cases (3%) were metastatic at diagnosis. No associations between clinical variables were evident within either trial group.
The nature of the adjuvant therapy administered to the CNS9904/RT cohort resulted in an older patient age when compared with the chemotherapeutic trial groups [independent t test; median age 7.8 (range: 3.0–16.7); P < 0.0001]. A higher proportion of supratentorial ependymomas were also observed in this cohort (Fisher exact test; P < 0.0001), these being predominantly of anaplastic histology (P = 0.032). This reflected a correlation of posterior fossa tumors with young children and supratentorial tumors with older patients identified across the entire FISH cohort [independent t test; mean age posterior fossa = 3.5 (SD 3.2) years vs. supratentorial = 6.7 (SD 4.6) years; P < 0.0001]. The median follow-up period for all 64 patients was 10.6 years (range: 0.6–11.8 years). Tumor progression occurred in 34 patients with a median time to progression of 1.4 years (range: 0.2–7.9 years) and 23 patients had died [median survival time of 3.4 years (range 0.6–5.8 years)]. Two CNS9904/RT cases (3%) were metastatic at presentation.
FISH was successful in 147 of 189 tumors (78%; Table 3). Concordance was achieved between independent scorers (Kappa 0.94; P < 0.0001). Unsuccessful cases were the result of core loss, autofluorescence or insufficient signal generation. Gain of 1q25 was evident in 30 of 147 tumors (20%). This proportion of gain did not differ significantly between the 3 therapeutic trial groups (range: 15%–24%; Table 3). No association was identified between 1q25 copy number imbalance and variables including patient age, gender, tumor histology, resection status, or intracranial location. Of the 8 metastatic primary tumors, 1q25 gain was observed in 3 cases (2 CNS9204, 1 CNS9904) and was not present in 1 case (CNS9204), whereas 4 samples were unscorable.
The prognostic impact of 1q25 gain and putative clinical factors on PFS and OS was initially evaluated across the pooled cohort of 147 eligible cases from the CNS9204, BBSFOP, and CNS9904/RT groups. Univariate analysis by log rank (Fig. 1A; P = 0.008) and Cox proportional hazards model stratified on therapeutic cohorts (Table 4; P = 0.002) identified 1q25 gain as a marker of adverse PFS. This was confirmed on multivariable analysis (Table 4; HR = 2.55; 95% CI = 1.56–4.16; P = 0.0002). The only clinical variable to be independently associated with poor outcome was incomplete surgical resection, both for PFS (Table 4; HR = 2.60; 95% CI = 1.64–4.11; P < 0.0001) and OS (Supplementary Table S2; HR = 1.75; 95% CI = 1.05–2.93; P = 0.03).
. | Univariate . | Multivariable . | ||
---|---|---|---|---|
Factor . | HR (95% CI) . | P . | HR (95% CI) . | P . |
Location | ||||
ST (n = 37) | 1 | 1 | ||
PF (n = 110) | 1.35 (0.80–2.29) | 0.25 | 1.21 (0.71–2.76) | 0.48 |
WHO grade | ||||
II (n = 63) | 1 | |||
III (n = 84) | 0.91 (0.58–1.41) | 0.67 | ||
Sex | ||||
Female (n = 59) | 1 | |||
Male (n = 88) | 0.91 (0.60–1.39) | 0.67 | ||
Surgery | ||||
CR (n = 79) | 1 | 1 | ||
IR (n = 68) | 2.3 (1.47–3.59) | 0.0003 | 2.6 (1.64–4.11) | <0.0001 |
1q25 result | ||||
No gain (n = 117) | 1 | 1 | ||
Gain (n = 30) | 2.16 (1.34–3.48) | 0.002 | 2.55 (1.56–4.16) | 0.0002 |
. | Univariate . | Multivariable . | ||
---|---|---|---|---|
Factor . | HR (95% CI) . | P . | HR (95% CI) . | P . |
Location | ||||
ST (n = 37) | 1 | 1 | ||
PF (n = 110) | 1.35 (0.80–2.29) | 0.25 | 1.21 (0.71–2.76) | 0.48 |
WHO grade | ||||
II (n = 63) | 1 | |||
III (n = 84) | 0.91 (0.58–1.41) | 0.67 | ||
Sex | ||||
Female (n = 59) | 1 | |||
Male (n = 88) | 0.91 (0.60–1.39) | 0.67 | ||
Surgery | ||||
CR (n = 79) | 1 | 1 | ||
IR (n = 68) | 2.3 (1.47–3.59) | 0.0003 | 2.6 (1.64–4.11) | <0.0001 |
1q25 result | ||||
No gain (n = 117) | 1 | 1 | ||
Gain (n = 30) | 2.16 (1.34–3.48) | 0.002 | 2.55 (1.56–4.16) | 0.0002 |
NOTE: The final model includes surgery and 1q25 result. Italic values are results for other variables added one by one to the final model.
Abbreviations: PF, posterior fossa; ST, supratentorial; IR, incomplete resection; CR, complete resection.
After adjusting for surgical resection (Table 5), 1q25 gain was identified as an independent predictor of adverse PFS for the CNS9204 (HR = 4.03; 95% CI = 1.88–8.63; P = 0.0003) and BBSFOP patients (HR = 3.10; 95% CI = 1.22–7.86; P = 0.02), but not the CNS9904/RT group (HR = 1.39; 95% CI = 0.61–3.20; P = 0.43). However, 1q25 gain was not a chemotherapy cohort-specific PFS marker (prognostic heterogeneity test of 1q25 for chemotherapy vs. radiotherapy cohorts; P = 0.13). Gain of 1q25 did not translate into a significantly worse OS across the cohorts (Supplementary Table S2).
. | Multivariable (n = 147) stratified on cohort . | . | Multivariable (n = 147) stratified on adjuvant therapy . | ||
---|---|---|---|---|---|
Factor . | HR (95 % CI) . | P . | Factor . | HR (95 % CI) . | P . |
Surgery | Surgery | ||||
CR (n = 79) | 1 | CR (n = 79) | 1 | ||
IR (n = 68) | 2.64 (1.67–4.18) | <0.0001 | IR (n = 68) | 1.99 (1.31–3.03) | 0.001 |
Global effect of 1q25 result | 0.0003 | Global effect of 1q25 result | 0.001 | ||
Heterogeneity between adjuvant therapy modalities | 0.13 | ||||
1q25 result in CNS9204 | 1q25 result in chemotherapy patients | ||||
No gain (n = 41) | 1 | No gain (n = 76) | 1 | ||
Gain (n = 11) | 4.03 (1.88–8.63) | 0.0003 | Gain (n = 17) | 2.9 (1.64–5.12) | 0.0003 |
1q25 result in BBSFOP | |||||
No gain (n = 35) | 1 | 1q25 result in radiotherapy patients | |||
Gain (n = 6) | 3.1 (1.22–7.86) | 0.02 | No gain (n = 41) | 1 | |
1q25 result in CNS9904 + RT only | Gain (n = 13) | 1.33 (0.58–3.05) | 0.50 | ||
No gain (n = 41) | 1 | ||||
Gain (n = 13) | 1.39 (0.61–3.20) | 0.43 |
. | Multivariable (n = 147) stratified on cohort . | . | Multivariable (n = 147) stratified on adjuvant therapy . | ||
---|---|---|---|---|---|
Factor . | HR (95 % CI) . | P . | Factor . | HR (95 % CI) . | P . |
Surgery | Surgery | ||||
CR (n = 79) | 1 | CR (n = 79) | 1 | ||
IR (n = 68) | 2.64 (1.67–4.18) | <0.0001 | IR (n = 68) | 1.99 (1.31–3.03) | 0.001 |
Global effect of 1q25 result | 0.0003 | Global effect of 1q25 result | 0.001 | ||
Heterogeneity between adjuvant therapy modalities | 0.13 | ||||
1q25 result in CNS9204 | 1q25 result in chemotherapy patients | ||||
No gain (n = 41) | 1 | No gain (n = 76) | 1 | ||
Gain (n = 11) | 4.03 (1.88–8.63) | 0.0003 | Gain (n = 17) | 2.9 (1.64–5.12) | 0.0003 |
1q25 result in BBSFOP | |||||
No gain (n = 35) | 1 | 1q25 result in radiotherapy patients | |||
Gain (n = 6) | 3.1 (1.22–7.86) | 0.02 | No gain (n = 41) | 1 | |
1q25 result in CNS9904 + RT only | Gain (n = 13) | 1.33 (0.58–3.05) | 0.50 | ||
No gain (n = 41) | 1 | ||||
Gain (n = 13) | 1.39 (0.61–3.20) | 0.43 |
Abbreviations: IR, incomplete resection; CR, complete resection.
In view of the above findings, the degree of surgical resection and tumor 1q25 status were integrated to enable stratification of the 147 ependymomas into 3 distinct risk groups for disease progression (Fig. 1B–E). High-risk disease was defined by ependymomas that were incompletely resected and showed 1q25 gain (2 risk factors). Intermediate-risk tumors were those which were either incompletely resected or showed 1q25 gain (1 risk factor), whereas standard risk disease encompassed completely resected ependymomas which did not exhibit 1q25 gain (no risk factors). This risk classification system showed significant differences in PFS, across both the pooled cohort (Fig. 1B; P < 0.0001), and independent trial subgroups (Fig. 1C–E; P = 0.009, 0.0002, and 0.01 for the CNS9204, BBSFOP, and CNS9904/RT groups, respectively) by the log rank test.
Discussion
The incorporation of biologic prognostic markers to enhance current risk stratification, guide therapy, and improve long-term outlook for pediatric intracranial ependymoma is an aim of future clinical trials. Several candidates have been proposed, yet recent review has shown that few have been analyzed in sufficiently large cohorts to allow informed evaluation of their prognostic efficacy in childhood (5). Independent validation of such purported pediatric ependymoma markers has often yielded contradictory results, such as those for ERBB2/ERBB4, Ki-67, and Nucleolin expression (25–27), although Tenascin-C expression has reported reproducible prognostic value across standardized cohorts (11). In addition, to date, no genomic imbalance has been assessed in pediatric ependymoma patients treated within the context of a prospective clinical trial (5).
We initially confirmed 1q gain as the most frequent chromosome arm imbalance in 36 primary childhood intracranial ependymomas and a patient-matched subset of 6 recurrent tumors by SNP array analysis, finding gain correlated with a trend toward worse OS. High-resolution analysis identified 1q21–25 among the most common subregions of gain. We explored these findings by evaluating iFISH results for 1q25 copy number increase across 147 primary pediatric intracranial ependymomas, spanning 3 European clinical trial cohorts. This identified 1q25 gain as an independent marker of tumor progression for ependymomas in 2 independent trial cohorts of young children treated with primary postoperative chemotherapy (CNS9204 and BBSFOP), but not from older children administered focal radiotherapy after surgery (CNS9904/RT). Nevertheless, incorporating tumor resectability with 1q25 status enabled patient stratification according to disease progression risk groups across all 3 trial cohorts, irrespective of patient age or adjuvant therapy administered.
We found no association between 1q25 gain, as determined by FISH, and a specific intracranial tumor location. Although this contrasted with the association of 1q gain with posterior fossa ependymomas identified from our smaller SNP array study, it was consistent with a sizeable CGH meta-analysis of 175 pediatric intracranial ependymomas that revealed gain involving 1q to be a frequent aberration in both posterior fossa and supratentorial tumors (5).
The only clinical factor adversely influencing outcome across both the array and pooled trial cohorts of this study was incomplete tumor resection, although this was not applicable to the CNS 9204 cohort when assessed independently (PFS, P = 0.36, OS, P = 0.44), in keeping with results for the entire cohort previously published (6). Indeed, this reflects current literature, in which resection status has been reported as the most consistent adverse prognostic marker in pediatric ependymoma (4, 5, 7, 28, 29), albeit not universally (5, 6, 30–32). Histologic anaplasia did not confer a worse patient outcome, supporting a recent multiprofessional pathologic review of the CNS9204, BBSFOP, and CNS9904 trials which found the current WHO classification system subjective and lacking prognostic accuracy (33).
Results from our FISH analysis of prospective trial cohorts lends some support to the findings of 2 sizeable retrospective FISH studies of intracranial ependymomas which have previously reported 1q25 gain as a marker of reduced PFS, but also OS on mixed age cohorts (13, 14). The threshold used to define gain for a tumor in this study (15% of nuclei showing signal gain) was higher than that of the retrospective series (10% of nuclei), as the latter did not yield a satisfactory measure of agreement between independent scorers. Nevertheless, the proportion of primary ependymomas showing 1q25 gain across the 3 therapeutic cohorts (15%–24%) was comparable with the 20% to 25% reported in these preceding studies. Although FISH proved an efficient means of screening our cohorts for copy number imbalance, unsuccessful cases were observed and mostly attributed to the use of tissue fixative agents incorporating acetic acid, a practice being rectified by contributing institutions.
In contrast to the retrospective analyses, 1q25 gain was not significantly associated with worse OS across the pooled trial patients of this study, including the CNS9204 or BBSFOP cases. Demographic and therapeutic differences between retrospective and prospective studies could account for this disparity, such as the inclusion of adults in the previous analyses, or a potential beneficial effect of introducing cranial radiotherapy as standard salvage therapy post relapse in the chemotherapeutic trials (6, 7). In addition, the median patient follow-up times for the 2 retrospective studies (7.3 and 5.2 years) were shorter than that of the pooled trial cohort (9.1 years, chemotherapy cases only 8.4 years). It is therefore possible that with continued observation of the retrospective cohorts, a less significant effect on overall outcome may have been observed, particularly, as late adverse outcome events are not uncommon for this tumor type (34). Nevertheless, as more than 70% of young children who experience ependymoma recurrence despite adjuvant chemotherapy will not survive longer term (9), a biologic marker that predicts progression in these patients remains an important discovery upon which therapy can be stratified. For high-risk patients with resistant disease to conventional chemotherapy, postoperative conformal radiotherapy may be a feasible and effective alternative adjuvant therapy on the basis of results from the SJCRH RT1 trial (29). Alternatively, novel chemotherapeutic and biologic agents could be considered, including tyrosine kinase inhibitors and antiangiogenic therapy (3).
Unlike the chemotherapeutic trial results, 1q25 gain in the CNS9904/RT group was less predictive of a worse patient PFS or OS. Although this could reflect a different biologic milieu for ependymomas from older children (5), it may suggest that primary radiotherapy is an effective counteractive adjuvant measure despite the adverse effects of 1q25 gain. Postoperative focal radiotherapy was standardized across the entire CNS9904/RT cohort, as opposed to chemotherapy in the other trial groups. This also differed from that of the retrospective studies, in which radiotherapy was administered to certain patients, but not uniformly (13, 14). However, this explanation must be considered cautiously as our iFISH study could not conclude that the reduction in PFS associated with 1q25 gain was specific to the chemotherapeutic trial regimens when compared with radiotherapy (P = 0.13), hindered by the smaller sizes of the individual treatment groups. Moreover, incorporating the degree of surgical resection with tumor 1q25 status enabled a significant 3-tier stratification for disease progression risk in the CNS9904/RT cohort (Fig. 1E), suggesting a prognostic role, albeit possibly not independently, for 1q25 gain in this patient group.
Survival data from the SNP profiling cohort did not help establish or refute the prognostic value for 1q gain in patients treated with radiotherapy, despite comparable median follow-up times with the pooled trial cohort analysis. Although 1q gain was independently associated with a trend toward inferior OS (Table 2), the 3 of 7 patients with ependymomas exhibiting 1q gain who remained alive had all been treated with postoperative cranial irradiation. The differences in patient outcome observed between 1q gain from the array study and the 1q25 FISH data are again plausibly explained by the considerably smaller size of the heterogeneous SNP array cohort, together with the variable treatments given to these children compared with the administration of standardized adjuvant therapy for the trial patients. Moreover, the difference in results could suggest that regions on 1q are more sensitive and robust markers of progression in childhood ependymoma than gain of the whole arm itself, a hypothesis supported by other genomic work on ependymoma (14, 34).
The integrated clinical and biologic risk stratification for ependymoma progression reported in this study reflects the current aspiration to develop novel prognostic models for this tumor. Indeed, recent work has reported a molecular staging system for ependymoma that defined 3 cytogenetic categories, including a high-risk group (group 3) characterized by 1q gain and/or homozygous CDKN2A deletion (13). In our analysis, such a classification would also have accounted for every high-risk patient, together with 25% of intermediate-risk patients (1q gain and complete tumor resection). However, incorporating 1q25 gain and tumor resectability seemed a more robust method of stratification across the 3 therapeutic trial groups (Fig. 1) compared with 1q25 status/group 3 categorization alone (Supplementary Fig. S3). This is supported by the previous study, in which addition of the genomic classification to established clinical variables in the previous work also improved risk prediction (13).
A transcriptional and genomic profiling study of posterior fossa ependymomas, delineating 2 distinct molecular subgroups of tumors with contrasting prognosis (groups A and B; ref. 31), offers further support to the present work. Although 1q gain was a feature of the less favorable group A tumors, survival for these patients was influenced more by the degree of surgical resection than 1q gain in isolation (34). Despite this, all 8 posterior fossa tumors designated high risk from our analysis most likely correspond to the group A category.
In summary, 1q25 gain as determined by FISH appears an independent marker of tumor progression in European primary chemotherapeutic cohorts of pediatric intracranial ependymoma (CNS9204 and BBSFOP). Although up to 42% of young children with ependymoma can remain free of disease with prolonged chemotherapeutic regimes (6, 7), 1q25 gain can be used to delineate a high-risk group of children, accounting for approximately 20% of patients who will experience recurrence or local progression despite this therapeutic strategy. However, the relatively small size of the individual therapeutic trial groups analyzed in this study precluded a decision on whether the prognostic role of 1q25 gain was specific to this patient age and treatment group, and thereby not applicable to older children treated with postoperative radiotherapy. Such a conclusion cannot be made until prospective trials are undertaken evaluating such patient cohorts in larger numbers. Nevertheless, this study showed that tumor 1q25 status, in conjunction with the degree of surgical resection, enabled a 3-tier patient stratification system of distinct disease progression risk groups across the therapeutic trial sets, irrespective of patient age or adjuvant therapy received.
We therefore advocate the prospective evaluation of 1q25 gain as a prognostic marker in forthcoming large international clinical trials of pediatric intracranial ependymoma, both independently and integrated with tumor resectability. Upon successful validation, 1q25 gain could be incorporated into future clinical trial design to improve risk stratification for children diagnosed with this tumor.
Disclosure of Potential Conflicts of Interest
P. Varlet is a consultant and is on the advisory board of Roche study. The other authors disclosed no potential conflicts of interest.
Acknowledgments
This work was a combined CCLG, Société Française d'Oncologie Pédiatrique (SFOP), and International Society of Pediatric Oncology (SIOP) biologic study. The authors thank Lisa Storer and Sarah-Leigh Nicholson for sample collection, Keith Robson and Tom Jacques for their involvement in the pathologic review process, and Lee Ridley for TMA construction. The sponsors had no role in study design, data collection, interpretation and analysis, report preparation, or submission. R.G. Grundy and J. Grill had access to all study data and final responsibility to submit for publication.
Grant Support
The study was supported by the James Tudor and Joseph Foote Foundations, the Institut National du Cancer (INCa)–Canceropole 01 Ile de France, and the charity l'Etoile de Martin. CCLG is supported by Cancer Research-UK.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.