Genetic alterations have been isolated from medulloblastomas within signaling pathways that normally regulate proliferation and differentiation of granule neuron progenitors (GNPs) during cerebellar development [13]. Medulloblastomas are induced in the SmoA1 mouse model through the constitutive activation of the sonic hedgehog (Shh) [GenBank: NM_009170] pathway within GNPs, signaling normally required for GNP proliferation. p27Kip1 loss in mice can increase tumor formation [4], and p27Kip1 haploinsufficiency has been further supported by studies using the Patched1 [GenBank: NM_008957] heterozygous (Ptc1+/−) and SmoA1 mouse medulloblastoma models [5, 14], where Shh-induced medulloblastoma incidence was accelerated by loss of one or both p27Kip1 alleles.
To determine whether p27Kip1 loss affected medulloblastoma initiation or progression, we generated mice hemizygous for the SmoA1 transgene that were heterozygous (p27wt/-) or nullizygous for p27Kip1 (p27−/−) and examined tumor incidence at two months of age, a time point preceding symptomatic medulloblastoma onset. Tumor incidence was 71.4% in SmoA1, p27wt/wt mice (n = 35); 73.1% in SmoA1, p27wt/- mice (n = 52); and 60% in SmoA1, p27−/− mice (n = 20). The differences were not significant (p = 1.0 and p = 0.55, respectively). Tumors from heterozygous SmoA1, p27wt/- mice did, however, display significantly more invasive phenotypes (Figure 1). 60.5% of SmoA1, p27wt/- mice had invasive tumors and tumors with effacement compared to 36.0% of SmoA1, p27wt/wt mice (Figure 2A, p = 0.04). Similarly, 66.7% of SmoA1, p27−/− mice displayed invasive and effaced pathology (Figure 2A, p = 0.08). The proliferative index, measured by bromodeoxyuridine (BrdU) incorporation, was higher in tumors with invasive or effaced pathology and greater in advanced tumors from SmoA1 mice lacking a single or both p27Kip1 alleles than in tumors from SmoA1, p27wt/wt mice (Figure 2B, p = 0.05 and p = 0.0001, respectively). Increased tumor progression was associated with decreased p27Kip1 and increased cyclinD1 [GenBank: NM_009829] protein levels in tumors from both study groups (Figure 2C-D).
An important observation can be drawn from comparing tumor incidence at two months to the survival at later time point results reported previously. Early tumor formation was relatively unchanged by the loss of a single or of both p27Kip1 alleles in our study. In contrast, SmoA1 mice with wildtype p27Kip1 lived on average twice as long as mice with a single copy of p27Kip1
[5], and mice retaining wildtype p27Kip1 in the heterozygous (Ptc1+/−) background survived significantly longer than counterparts lacking p27Kip1
, which succumbed due to an increased tumor incidence [14]. In combination with the aggressive pathology observed in early tumors from SmoA1 mice lacking p27Kip1, the higher tumor incidence in Ptc1+/− mice lacking p27Kip1 further confirms that p27Kip1 is haploinsufficient in Shh-mediated medulloblastomas and that its loss of function contributes to medulloblastoma progression.
We next assessed whether P27 levels could distinguish children likely to fail therapy from those likely to be long-term survivors. To properly evaluate P27 as a potential marker of therapeutic response, we established quantitative criteria for P27 staining in a training set of patient samples then applied the same criteria to a larger test set. Human medulloblastoma tissue microarrays containing tissues from 87 patient samples were analyzed for P27 expression [10]. 80% of samples evaluated had 20% or fewer P27 positive cells, with 16% of samples having less than 1% P27 positive cells. In contrast, intensity scoring of control sections from normal human cerebella revealed an average of 74% P27-positive granule cells (Figure 3A). Additional P27 staining on individual human medulloblastoma sections from five patients revealed significant loss of P27 protein (Figure 3B). Two tumor samples with regions lacking P27 contained additional regions resembling the remaining normal cerebellar architecture that stained positive for P27 (Figure 3C). Overall survival analysis demonstrated that patients with very low P27 expression had a poor outcome. Significant correlations were found when patients were stratified between high and low P27 levels, with patients with higher P27 levels (>1% P27 positive cells) having a more favorable outcome (n = 79, p = 0.027, data not shown). This correlation persisted when patients were additionally stratified by factors that define the standard-risk subgroup. P27 expression had prognostic value in patients greater than three years old with total surgical resection (n = 30, p = 0.001) or a lack of metastasis (n = 44, p = 0.007). While the number of patients with clinical data for all three standard-risk parameters was too small for Kaplan-Meier analysis, multivariate analysis with overall survival as the dependent variable demonstrated that P27 expression levels had prognostic value, independent of the extent of tumor resection or metastasis.
Together, these data suggested that P27 absence could further stratify standard-risk patients between survivors and those that fail therapy and do not survive. Because incorrect patient stratification could result in death for children placed improperly into a lower risk group, or unnecessary toxicity in children placed improperly into a higher risk group, we sought to confirm these findings by examining tissues from a second, larger study population. Tissues from 141 patient samples from children treated on the SIOP PNET03 clinical trial in Europe were stained for P27 and analyzed as above [11, 12]. Only two patients were identified that met the <1% cutoff criteria, indicating that the threshold established in the training set was not suitable for the test data set. To determine if this threshold was too stringent, we relaxed the criteria to <10% to capture more patients with low P27 expression. No significant differences were found in overall survival (P = 0.67) or in progression free survival (P = 0.55) between patients with little to no (less than 10%, F/N) P27 cells and those with greater than 10% P27 positive cells (M) (Figure 3D). We therefore conclude that the first test set was a false positive signal, likely due to small numbers and retrospective establishment of the cutoff criteria. This discrepancy underscores the importance of evaluating potential prognostic indicators in large patient populations and using both training and test data sets when developing candidate prognostic indicators for clinical use. To our knowledge, this is the first quantitative analysis of P27 expression in a large cohort of human medulloblastoma samples.