Tumor cells were implanted into the cerebellum of SCID-beige hosts. Animals were sacrificed when they developed symptoms, and tumor tissue was stained with H&E. (A) Primary tumor from a Ptc^+/- mouse. (B) Secondary tumor from a SCID-beige host that received 5 × 10⁵ Ptc^+/- tumor cells. (C) Tumor-free survival of animals receiving indicated numbers of Ptc^+/- tumor cells. Animals implanted with 2-5 × 10⁵ cells developed tumors within 6-8 weeks (median latency = 49 days). Animals implanted with 0.5 × 10⁵ or 1 × 10⁵ cells developed tumors with a lower incidence (42% and 17% respectively) and a longer latency (58 days and 159 days respectively). Scale bars = 100 μm.

(A-B) Tumor cells from Ptc^+/- or Math1-GFP/Ptc^+/- mice were stained with anti-CD133 (black histograms) or isotype-matched control antibodies (gray histograms) and analyzed by FACS. (A) Representative histograms from CD133-low (top) and CD133-high (bottom) tumors. (B) Scatter-plot showing percentage of CD133+ cells in 24 tumors. Horizontal line indicates median percentage of CD133+ cells (2.5%). (C-D) CD133+ cells were sorted from neonatal cerebellum (C) or from Ptc^+/- tumors (D) and cultured at clonal density in serum-free medium with 25 ng/ml EGF and bFGF for 10 days. Cells were photographed under bright field (top, scale bar = 200 μm) or stained using the Live-Dead Assay Kit, which labels live cells green and dead cells red (bottom, scale bar = 100 μm). Passageable neurospheres were consistently obtained from neonatal cerebellum but not from Ptc^+/- tumors. (E) Survival of SCID-beige mice implanted with 3 × 10⁵ unsorted, CD133+ or CD133− tumor cells. Tumors never developed in mice implanted with CD133+ cells (cerebellum shown in F), but developed in all mice implanted with CD133− cells (cerebellum with tumor shown in G). Scale bars for F and G = 200 μm.

(A) FACS analysis of tumors from Math1-GFP/Ptc^+/- mouse shows that the majority of tumor cells are Math1-GFP+. (B) Survival of SCID-beige mice implanted with 3 × 10⁵ GFP+ or GFP− cells. Tumors developed in 100% of animals implanted with GFP+ cells (median latency: 48 days). (C-F) Tumors from Math1-GFP/Ptc^+/- mice (1° tumors in C, E) and from SCID-beige mice implanted with Math1-GFP/Ptc^+/- tumor cells (2° tumors in D, F) were cryosectioned and stained with DAPI (blue) to label nuclei and with anti-GFP antibodies to label Math1-expressing cells (green, C-D). Additional tumors were paraffin embedded, sectioned and stained with hematoxylin (blue) to label nuclei and with anti-Ki67 to label proliferating cells (brown, panels E-F). Scale bars = 200 μm.

(A) Tumor cells from Ptc^+/- or Math1-GFP-Ptc^+/- mice were stained with anti-CD15 antibodies and sorted into CD15+ and CD15− fractions. 3 × 10⁵ cells from each fraction were implanted into SCID-beige mice, and animals were monitored for tumor formation. Animals transplanted with CD15+ cells all developed tumors, whereas those transplanted with CD15− cells did not. (B) Expression of CD15 in neonatal cerebellum. Cerebellar cells from P7 Math1-GFP mice were stained with anti-CD15 antibodies and analyzed by FACS. Among cells that expressed Math1-GFP (right quadrants), a subset (15% of all cells, 18.5% of Math1-GFP+ cells) expressed CD15 (red dots). (D-F) Expression of CD15 in Ptc^+/- tumor cells. (C) Scatter-plot showing the percentage of CD15+ cells in a panel of 34 Ptc^+/- tumors (black dots) and 13 Math1-GFP/Ptc^+/- tumors (green dots). Horizontal lines indicate median percentage of CD15+ cells: 47% for Ptc^+/- and 4.3% for Math1-GFP/Ptc^+/- tumors. (D) FACS analysis of CD15 expression in a representative Math1-GFP/Ptc^+/- tumor. Among tumor cells expressing Math1-GFP, 13.4% (11% of all tumor cells) expressed CD15 (red dots). (E-F) Tumor-bearing Math1-GFP/Ptc^+/- mice (E) and Ptc^+/- mice (F) were injected intracranially with anti-CD15-secreting hybridomas to label CD15+ cells in vivo (see Supplementary Methods). Cerebella were sectioned and stained with secondary antibodies to detect cells that had bound anti-CD15 antibodies (red) and with anti-GFP antibodies to label Math1-expressing cells (green) or with DAPI (blue) to label all nuclei. Note the predominance of CD15 staining on the edge of the tumor (E) and at the boundary between tumor and normal tissue (F). Scale bars = 100 μm.

Cells from 6 Ptc^+/- tumors were sorted into CD15+ and CD15− samples, and RNA from these samples was labeled and hybridized to Affymetrix expression microarrays. (A) Principle components analysis. Each sample is represented by a spot (red = CD15+, blue = CD15−) whose position in the grid reflects its overall expression profile. The distance between spots is proportional to the difference in gene expression. Note that CD15+ samples are clustered together (red ellipse) and occupy a region of the grid distinct from that occupied by CD15− samples (blue ellipse). (B) Unsupervised hierarchical clustering highlights groups of genes whose expression is significantly increased or decreased in CD15+ vs. CD15− cells. Each column represents a distinct sample (CD15− or CD15+) and each row represents an individual gene. The normalized relative level of gene expression is denoted by color (blue = low, gray = intermediate, red = high), as indicated in the gradient at the bottom. (C) Validation of differential gene expression by RT-PCR. CD15+ and CD15− samples from five independent Ptc^+/- tumors (distinct from those used for microarray analysis) were analyzed by quantitative RT-PCR using primers for the indicated genes. Gene expression was normalized based on Actin levels.

(A) CD15+ and CD15− tumor cells were cultured for 48 hours in serum-free medium, and then pulsed with tritiated thymidine (³H-Td) for 14 hours before assaying thymidine incorporation. Data represent means of triplicate samples ± SEM. (B-C) RNA from CD15+ and CD15− tumor cells was analyzed by real-time RT-PCR using primers specific for the Hh target genes cyclin D1 and Gli1. Expression levels are normalized to actin. In A-C, percentages of CD15+ cells are shown below bars for each tumor. CD15+ cells consistently exhibited higher proliferation and Hh target gene expression than CD15− cells from the same tumor. (D) Math1+CD15− cells have lost the remaining copy of Ptc. RNA from Math1+/CD15+ and Math1+/CD15− tumor cells, and from WT GNPs (as a positive control) were analyzed by RT-PCR using primers specific for Ptc exons 7-9 (found in both the mutant and WT allele) and Ptc exons 2-3 (found only in the WT allele). Transcripts containing exons 7-9 are found in all cell populations, but only GNPs express transcripts containing the WT allele of Ptc. These data suggest that both CD15+ and CD15− cells are tumor cells that have lost the WT allele.

(A) Variable immunoreactivity for CD15 in a classic medulloblastoma. (B) CD15 immunoreactivity in the matrix of a nodule in a medulloblastoma with extensive nodularity. (C) Strong immunoreactivity for CD15 in clusters of tumor cells in a large cell medulloblastoma (Scale bars = 70 μm for all panels). (D) Increased probability of CD15+ signature correlates with poor survival in human medulloblastoma. Black tick marks represent surviving patients at time of last follow-up. Log rank for trend p = 0.02.