Gain of 1q Is a Potential Univariate Negative Prognostic Marker for Survival in Medulloblastoma

Medulloblastomas are the most common malignant brain tumors in children (1). With recent advances in treatment options involving surgery, radiotherapy, chemotherapy as well as computed tomography and magnetic resonance neuroimaging techniques, long-term survival rates of 60% to 80% can be achieved (2). Unfortunately, cognitive deficits and other sequelae of therapy are common among survivors (3). A major challenge, therefore, is to differentiate between high- and low-risk patients to tailor therapy to the degree of biological aggressiveness of the tumor. Currently, patients are divided into average and high-risk groups, with the high-risk group being characterized by three clinical parameters: age group (<3 years), residual tumor after resection >1.5 cm (3), and evidence of metastasis at presentation (3, 4). Recent reports, however, indicate that these clinical variables are an inadequate method of defining disease risk (5). To further develop parameters for clinical outcome prediction, molecular genetic markers such as expression of trkC (6), c-myc (7), erbB (8), PDGFRB (9, 10) chromosomal abnormality iso-17q (11), a gain of 8q (12), or anaplastic histology (13) have also been suggested to be potential aids in predicting prognosis. Recently, gene expression profiling was suggested to be a better outcome predictor (14), where an eight gene predictor model was used to segregate survivors from those experiencing treatment failure. When tested in an independent cohort of patients, this model was apparently the only univariate prognostic indicator of survival that reached statistical significance (2), although disease dissemination at diagnosis showed a nonsignificant trend with decreased survival despite a favorable gene expression profile.

To develop potentially more predictive biomarkers for medulloblastoma risk stratification, we analyzed data obtained by array-based comparative genomic hybridization (aCGH) to systematically explore the relationship between copy number abnormalities (CNA) in the karyotype and overall survival in a cohort of 49 medulloblastoma samples. As a result of this analysis, we have shown a statistical association between the gain of 1q and overall survival.

Samples. A total of 49 medulloblastoma samples were obtained via Institutional Review Board (IRB)–approved protocols from patients diagnosed at the Children's Hospital (Boston, MA) and the Johns Hopkins Hospital (Baltimore, MD). Patients more than 3 years of age underwent surgical resection followed by a combined regimen of chemo- and radiation therapy. The median duration of follow-up was 39 months. During the time of the last follow-up, 16/49 (32%) patients had died of their disease. All of these tumors were analyzed for aCGH using the RPCI 6K BAC array as previously described (15, 16).

Array comparative genomic hybridization. Briefly, the BAC array contains ∼6,000 RPCI-11 library BAC clones that provide an average resolution across the genome of 420 kb. BACs were printed in triplicate on amino-silanated glass slides (Schott Nexterion type A) using a MicroGrid ll TAS arrayer (Apogent Discoveries). Genomic pooled normal control DNA and tumor DNA were fluorescently labeled by random priming and hybridized as described previously (15). Hybridizations of normal and tumor DNA were done as sex mismatches to provide an internal hybridization control for chromosome X and Y copy number differences. The hybridized slides were scanned using an Axon 4200a scanner to generate 10-μm resolution images for both Cy3 and Cy5 channels, and image analysis was done using ImaGene (version 4.1) software (BioDiscovery, Inc.).

Circular binary segmentation analysis. For each reference point on the array, background corrections were done for both test and control channels. The log₂ ratios of the background-corrected test/control were normalized using a loess correction. For each BAC, a median of the log₂ ratios of all its replicates was computed, and BACs with less than two successful replicates were excluded.

The corrected log₂ ratio for each BAC was then used to segment the array using Circular Binary Segmentation (CBS) using DNAcopy software (17). For each segment, the median absolute deviations (MAD) of the corrected log₂ ratios were computed (18). Segments with a median-corrected log₂ ratio greater than 1× the median of the MAD (MMAD) were considered gained, and segments with a median-corrected log₂ ratio less than −1× MMAD were considered lost (18). All BACs were assigned a call of gained, lost, or normal, based on the call of the segment they are in and the karyotypes of the tumors generated accordingly.

Statistical analysis. Survival rates were computed using the methods described by Kaplan and Meier (19) The log-rank test was used to evaluate the difference in survival rates between genetic groups (20). Because of the relatively small number of samples in this cohort, “Gain P values” (Table 1) were calculated by segregating all of the samples involving a single copy gain along the chromosome arm, with those tumors combined showing either a normal copy number or a single copy loss. Similarly, “Deletion P values” were obtained by comparing tumors showing losses involving chromosome arms with samples combined with either a normal copy number or a single copy gain.

Where individual log-rank tests revealed a significant correlation with survival, additional confirmational tests were done. Due to the small number of events, the log-rank test was incorporated into a permutation test to validate the asymptotic significance level under the null hypothesis. The values were negligibly different, and therefore, the asymptotic χ² significance level is presented.

aCGH analysis of medulloblastomas. To establish whether any CNA in the 49 aCGH profiles of tumors correlated with overall survival of the patient cohort, chromosomal abnormalities involving a segment identified by CBS analysis were scored as either “+” to indicate a single copy gain of the arm or “−” to signify a single copy loss of the chromosomal arm. More complicated rearrangement events involving multiple gains and losses within a chromosomal arm, although rare, are designated by “+//−”. A summary of these events is described in Table 1. Grouping specific aberrations that involve the same chromosome arms, but with different extent of the CNA, may potentially lead to over- or underestimation of prognostic significance due to the lack of large sample sizes with identical aberrations. It was not possible, therefore, to segregate various subcategories of aberrations for significance testing.

Correlation studies. Initially, a log-rank test was done for each cytogenetic event along each chromosomal arm to determine whether any specific abnormality was correlated with survival. The results of this analysis are given in Table 2. In some cases, because specific chromosome arms were infrequently or never involved in CNAs, they were not included in the analysis. As a result of this analysis, loss involving 4q, 7p, and 18q and gain of 1q were found to be significantly associated with survival. Losses of 7p (P = 0.00107 by the log-rank test), 4q (P = 0.02503), and 18q (P = 0.02503), however, occurred at low frequency (<5%) in this series of tumors, which made it difficult to establish whether either of these events could be a potential biomarker for survival. A gain of 1q, on the other hand, was a relatively more frequent event (9/49) and showed a highly significant correlation with overall survival (P = 0.00004 by the log-rank test). To investigate this correlation further, we did a permutated log-rank test to simulate the null P value distribution, which again showed significance (P < 0.0001), correlating survival and gain of 1q status in this cohort of medulloblastoma patients.

The Kaplan-Meier survival curve showed a distinctive pattern between patients with a 1q gain and patients without the 1q gain (Fig. 1A). At the median follow-up time from this cohort (39 months), only ∼20% of patients with 1q gains were alive, whereas ∼80% of patients without 1q gain were alive. Because of the availability of additional clinical data for many of the patients in the current series, we compared the prognostic significance with two other traditional risk stratification parameters—age at diagnosis and presence of metastasis at presentation—with that of the presence of a 1q gain. Using age at diagnosis of greater/less than 3 years (Fig. 1B), the log-rank analysis did not identify a statistically significant association with survival (P = 0.7256). The same was true using a Cox proportional hazards model. In the same way (Fig. 1C), evidence of metastasis at the time of surgery did not yield a significant correlation with survival (P = 0.2295, log-rank test), although some previous reports have indicated that evidence of dissemination is the most sensitive prognostic factor for risk stratification (21). Moreover, multivariate analysis between survival and age at diagnosis, with the presence of metastasis, did not reveal a significant association either (Fig. 1D). Thus, patients that were >3 years old at diagnosis, with no metastasis (most favorable), showed essentially the same survival as those that are <3 years old with metastatic disease (least favorable) as seen in Fig. 1D. Overall, therefore, gain of 1q seems to be the strongest indicator of overall survival from the analysis of this cohort of medulloblastoma patients.

Previous studies have suggested independently that both a gain of 8q and the presence of iso-17q (see Discussion) are potentially related to survival. To examine these possibilities in the current series, we did Kaplan-Meier analysis for these two events. As shown in Fig. 2, using the log-rank test for 8q gain, the P value was 0.188, and for the iso17q, the P value was 0.6106. Thus, although the frequencies of these CNAs in the 49 samples reported in this study were comparable with the frequencies seen in other studies, the associations were not significant. On rare occasions, 17q gain can occur independently of the loss involving 17p, but Kaplan-Meier analysis failed to reach statistical significance for the correlation between survival and overall 17q gain (P = 0.30). There are several different mechanisms by which gains of chromosome material are achieved in cancer cells. Whole chromosome gains may be a consequence of non-disjunction, whereas partial gains are usually due to deletions and translocations involving the chromosomes in addition to non-disjunction. When we excluded gain of a whole copy of chromosome 1 from the analysis, the same association with survival was seen in patients with gains of 1q (data not shown). Interestingly, all nine tumors with 1q gains were of the classic or anaplastic morphologic subtype, and none of the 10 tumors with desmoplasia or nodules had this molecular change. However, because only four of the nine cases with 1q gain were of the more aggressive anaplastic subtype, the 1q gain abnormality does not seem to be a surrogate marker for anaplasia.

Minimal region of 1q gain. The resolution provided by aCGH allowed the high-resolution definition of the breakpoints giving rise to each of the 1q gains seen within this cohort. The results are summarized in Fig. 3. Although only one case of subregional 1q loss was recorded in this series, this case allowed a tentative assignment of the minimal consistent region of gain among these tumors to region Chr1:144250173-193504146. Clearly, whereas it is potentially important in defining candidate genes from 1q related to poor survival, it must be seen whether subsequent studies can support this suggestion. Because 1q gain apparently provides a highly significant correlation with overall survival, it is implicit that either a gene along the length of 1q or several genes on 1q acting in concert is related to poor survival as a result of an increased dosage effect that contributes to more aggressive cancer cell growth. There are currently 1,369 annotated genes on the long arm of chromosome 1. Although defined by only a single case, the 1q gain in tumor MD3, which defines the minimum region of overlap in 1q12-23 involving the Chr1:144250173-193504146 region, still contains 749 genes.

Currently, clinical parameters such as (a) evidence of metastasis at presentation, (b) age at diagnosis (<3 years), and (c) extent of residual tumor post-resection have been the most traditional risk stratification indicators for medulloblastoma patients (3, 4). We now present evidence that, using both single and multivariate analysis, gain of 1q seems to be a significantly more accurate predictor of overall survival in this disease than at least two of the traditional risk stratification parameters. The correlation between a specific biomarker and survival, however, depends on a number of critical variables, only some of which can be adequately controlled for between studies. For example, one of the current criteria that is used to predict outcome in medulloblastomas is the amount of residual tumor after surgery. Quantification of the levels of residual disease, however, may often be a subjective assessment. Such a metric, therefore, might be expected to show more variability between neurosurgical teams. Because our study drew samples from different clinical centers, we were not confident in the overall analysis to include this parameter in our comparison with 1q gain, in contrast to some of the other more objective phenotypes.

The association of 1q gain with survival was particularly evident in the current series of tumors, but has curiously not been reported before in similar studies, although the overall incidence of 1q gain in these different studies is approximately the same (11, 12), suggesting that there is no significant bias in the genetic abnormalities seen in the different tumor cohorts (see also below). One possible explanation for this discrepancy is that the techniques used to identify genomic changes between studies may not have allowed unequivocal identification of 1q gain or, more likely, that the cohort sizes in these cases are too small to reveal the association. Furthermore, in the study by De Bortoli et al. (12), there was in fact an association approaching significance for 1q gain (log-rank P = 0.088) and progression-free survival in that cohort. Interestingly, of the 9 medulloblastoma samples with 1q gain in this analysis, 6 showed no evidence of metastasis at diagnosis, which reaffirms that the gain of 1q is an independent prognostic marker rather than being coincident in patients with metastatic disease. The mechanism through which the extra copy of 1q confers biological aggressiveness of the tumor, however, is unclear. Although gene dosage may have severe consequences during fetal cerebellum development, the lack of concordance between the gain of 1q and age at diagnosis argues against this hypothesis. An alternative scenario is that the cytogenetic event that results in the gain of 1q may result in the biological aggressiveness of this tumor. However, without further investigation, it is not possible to identify the specific event, if any, that results in the gain of 1q.

One of the most common cytogenetic events in medulloblastomas involves the loss of genetic material from 17p with a concomitant gain of 17q (11, 12, 22). These two cytogenetic events can also occur independently, albeit less frequently (22–24). The prevalence of this cytogenetic event has prompted attempts to establish a relationship between this particular CNA and survival. Although some studies have suggested a poorer outcome from tumors carrying this abnormality (11, 23, 25, 26), others have not (24, 27). The number of patients in many of these studies, however, was still relatively small, and the technique used to identify chromosome 17 abnormalities may have been suboptimal. In our series, the aCGH profiles unequivocally identify 17p⁻/17q⁺ (16, 22, 28). However, despite 22% of tumors in this series showing involvement of chromosome 17, no statistically significant correlation between these abnormalities and overall survival was detected. Interestingly, none of the five patients with the iso(17q) described by Pan et al. (11) reached the 5-year survival mark, although the numbers in this study were small compared with our series where 8/11 tumors with this specific cytogenetic event remain alive at the time of the last follow-up (35–98 months). A gain of 8q has also been suggested to represent a prognostic biomarker in medulloblastoma (12). We did not, however, observe a significant association between a gain of 8q with survival at an α level of 0.05 in the series of tumors reported here. We cannot at this time, however, conclude that 8q gain is not associated with survival because this cohort may have been underpowered for detecting this particular association. Interestingly, the study by De Bortoli et al. (12) revealed 7/71 cases with 8q gain, which is a similar frequency to that seen in our study, again supporting the idea that there is no particular bias in terms of the genetic abnormalities seen in these different series. There was, however, a dramatic difference in survival between the two reports.

One of the promising opportunities in the post–human genome project era is that greater information will lead to improved prognostic and diagnostic biomarkers. As technologies such as gene expression arrays and aCGH have achieved higher resolution, the extensive data generated by these platforms can be assessed for utility in the clinical management of patients. In this cohort of medulloblastomas, we have provided evidence that 1q gain significantly correlates with overall survival. It is implicit that a gene(s) along the length of 1q is related to poor survival as a result of an increased dosage effect that contributes to more aggressive cancer cell growth. Several genes on 1q have been reported to be overexpressed in different cancer types, although in most cases, these represent high-level fold increases over control samples. The MDM4 gene in 1q31, for example, is highly amplified and overexpressed in rare cases of gliomas (29, 30), and the KIF14 gene in 1q31-32 has been shown to be overexpressed in retinoblastomas (31). To our knowledge, neither of these genes have been implicated in medulloblastoma. Clearly, more work is needed to elucidate the candidate gene(s) within this minimal region of overlap. However, fluorescence in situ hybridization analysis of biopsy-processed tissues for 1q gain may lead to a potentially more sensitive predictor for survival that can augment traditional clinical parameters used for risk stratification of medulloblastoma patients.

We are grateful to the staff of the Roswell Park Cancer Institute Microarray Core Facility for performing aCGH analysis.