Supplementary Figure S1. Grade-specificity aberrant expression of HOXA genes in gliomas. (A) Representative RT-PCR analyses of HOXA gene expression in human astrocytomas. Exemplified glioma samples include four WHO grade 4 glioblastomas (GBM 4, GBM 5, GBM 3, GBM 6), three WHO grade 2 astrocytomas (A-1, A-2, A3), and one WHO grade 3 anaplastic astrocytoma (AA-1). Fetal brain white (FBW) matter was used as a non-tumoral control; HEK 293T cells (293T) were used as positive control. The criterion for assessment of positive versus negative gene expression was based on the presence or absence of specific PCR bands at the expected molecular weights, respectively. In some instances, non-specific PCR bands weree also detected at incorrect molecular weights on repeated experiments (e.g. HOXA5 and HOXA11 in FBW); such bands were thus not consideredd to correspond to positive expression. HOXA1 and HOXA9 genes present two specific PCR bands, representing two alternative splicing variants. GBM samples show strikingly higher activation of HOXA genes as compared to lower grade gliomas or normal brain tissue. (B) HOXA9 expression analysis by RT-PCRR in primary-recurrent astrocytoma (WHO grade 2). The time interval reflects the time from initial surgery for a low-grade astrocytoma to the time of surgery for tumor recurrence, each diagnosed histologically as glioblastoma (WHO grade 4). Note that the tumor pairs from three patients who each initially presented with low-grade low-
grade primary tumors demonstrate absence or negligible expression of HOXA9, while all the three high-grade recurrences show significant expression of HOXA9. (C) HOXA9 expression analysis by RT-PCR in seven recurrent low-grade infiltrative gliomas (lanes 1-7), a positive control (lane 8) and a negative control (Water). Each tumor was initially diagnosed as a WHO grade 2 astrocytoma and remained a WHO grade 2 at recurrence. Prior to resection of the recurrent tumor, five of the patients were treated with external beam radiation therapy and three were treated with chemotherapy. Expression of HOXA9 was not detected in any of the WHO grade 2 astrocytoma recurrences, suggesting that HOXA9 activation is not therapy-related.
Supplementary Figure S2. HOXB, HOXC, and HOXD gene clusters demonstrate chromosomal domains of transcriptional activation in the UCSF tumor set. (A to C) Gene expression heat maps centered on HOXB, HOXC, and HOXD clusters and extending 1 Mb in each direction for the UCSF tumor set (37 GBMs) demonstrates HOX gene domains of activation resembling those detected for the HOXA cluster. Genes contained in boundaries flanking the activation domains are relatively silent. For each heat map, the gray side
bar indicates the estimated boundaries of the activation domain. The colored horizontal bar across the top of the heat map indicates which tumors express HOXA9 (blue, high expression; yellow, low or no expression). Red and green colors reflect high and low/no expression of each gene, respectively.
Supplementary Figure S3. HOXB, HOXC, and HOXD gene clusters demonstrate chromosomal domains of transcriptional activation in the MDA tumor set. (A to C) Gene expression heat maps centered on HOXB, HOXC, and HOXD clusters for the MDA tumor set (63 GBMs), analyzed and represented as in Figure S2. White color corresponds to samples for which the expression status could not be assigned with high (>95%) confidence.
Supplementary Figure S4. Aberrant transcriptional activation across HOX clusters in glioblastoma does not recapitulate the co-linear expression pattern of HOX genes during embryonic development. (A and B) Gene expression heat maps were generated for the two sets of glioblastomas, 37 from UCSF (A) and 63 from MDA (B). Clustering tumors using Euclidian distance suggests two primary tumor groups displaying HOX-activated or HOX inactive expression patterns. The colored horizontal bar across the top of the heat map indicates which tumors express HOXA9 (blue, high expression; yellow, low or no expression). Red and green colors reflect high and low/no expression of each gene, respectively. White color corresponds to samples for which the expression status could not be assigned with high (>95%) confidence. Note that the highly ordered HOX gene expression patterns observed during development are not recapitulated by the tumors with aberrant HOX gene expression.
Supplementary Figure S5. The PI3K pathway regulates HOXA9 expression in glioblastoma cells. (A) Quantitative real-time PCR (qpcr) assessment of HOXA9 mrna expression levels in A172 cells following treatment with the PI3K inhibitor, LY294002, for time points ranging from 1 to 24 hours. Also included are measurements of HOXA9 levels after treatment of A172 cells with LY294002 for 24 hours and followed by incubation with media lacking LY294002 for 8, 24 and 32 hours. Expression levels were normalized to hgus, a housekeeping gene with relatively constant levels of expression. A significant decrease in HOXA9 mrna levels occurs as early as
4 hours following treatment with LY294002. HOXA9 levels return to pre-treatment levels 32 hours after LY294002 removal. The results are representative of 3 independent experiments. * P<0.05, 2-sided Student s t-test (P<0.001 at 4h; P<0.001 at 8h; P=0.002 at 24h; P=0.008 at 24+8h; P<0.001 at 24+24h). (B) A172 cells were treated with LY294002 for 24 hours and HOXA9 protein levels assessed by immunoblot analysis. PI3K inhibition resulted in a marked decrease of HOXA9 protein levels in A172 cells, which is consistent with our findings of PI3K-mediated regulation of HOXA9 transcription. α-tubulin was used as an internal loading control. The results are representative of 3 independent experiments for each blot. (C) Treatment of A172 cells with rapamycin for 24 hours resulted in successful inhibition of mtor activity as assessed by decreased levels of phosphorylated p70/p85 S6Kinase, a downstream target of mtor-mediated phosphorylation. The 2 bands indicate both p70 and p85 isoforms of the same kinase. HOXA9 protein levels were not affected by mtor inhibition, supporting our hypothesis that mtor is not a major mediator of PI3K-regulation of HOXA gene expression. Total levels of p70/p85 S6Kinase and α-tubulin were used as internal loading controls. The results are representative of 3 independent experiments for each blot. (D and E) Quantitative real-time PCR (qpcr) assessment of HOXA9 expression levels in A172 cells and two sublines of a primary GBM grown in neurosphere conditions (NSP_7030/SF6969-3 and NSP_7081/GS2-3), deleted at the PTEN locus (data not shown), following treatment with the PI3K inhibitor, LY294002 (D), or the mtor inhibitor, rapamycin (E), for 24 hours. Expression levels of HOXA9 were normalized to human GUS (hgus), a housekeeping gene with relatively constant levels of expression. The relative gene expression levels of control-treated cells were normalized to 100% to facilitate the comparisons. A significant decrease in HOXA9 mrna levels occurred following treatment with LY294002 in both A172 cells (*P<0.001) and GBM-derived neurosphere cell lines (NSP_7030, *P=0.021; NSP_7081, *P=0.049). In contrast, rapamycin treatment did not show a consistent effect on HOXA9 transcriptional levels across the tested cell lines (NSP_7030, *P=0.011). The results are representative of 3 independent experiments.
Supplementary Figure S6. HOXA9 expression status improves MGMT-based prediction of progression-free survival of glioblastoma patients from the UCSF tumor set. (A) Kaplan-Meier progression-free survival curve for 43 GBM patients from UCSF (including cases used for expression array analysis and those used for initial assessment of HOXA expression) whose tumors demonstrated (green line) or lacked (red line) MGMT promoter methylation. Patients whose tumors had methylation of the MGMT promoter had a trend toward longer progression-free survival (P=0.060). (B) Kaplan-Meier progression-free survival curve for the same set as in (A), based on HOXA9 expression and MGMT promoter methylation status. Patients whose tumors expressed HOXA9 and/or lacked methylation of the MGMT promoter (red line) had a significantly shorter progression-free survival (P=0.001) than patients whose tumors did not express HOXA9 and had a methylated MGMT promoter (green line). (C) Kaplan-Meier progression-free survival curve for the 24 patients whose tumors demonstrated MGMT promoter methylation, stratified based on whether or not the tumors expressed HOXA9.
Expression of HOXA9 was associated with significantly shorter progression-free survival (P<0.001), even considering only the patients with a methylated MGMT promoter.
Supplementary Figure S7. Examples of methylation-specific PCR (MSP) analysis of the MGMT promoter for glioblastoma tumor tissue from three patients (lanes 1 to 3) and a fetal brain specimen with DNA either untreated (F. Brain) or in vitro methylated with SssI (SssI F. Brain). Note the presence of bands in both the unmethylated (U) and methylated (M) lanes for glioblastoma samples #1 and #3, reflecting MGMT promoter methylation. The lack of a band in the lane corresponding to methylation-specific primers for glioblastoma sample #2 reflects a lack of MGMT promoter methylation. PCR reactions in the absencee of DNA ( H 2 O) were performed as negative controls for both the unmethylation and methylation reactions.