A transcriptomic signature mediated by HOXA9 promotes human glioblastoma initiation, aggressiveness and resistance to temozolomide.

Glioblastoma is the most malignant brain tumor, exhibiting remarkable resistance to treatment. Here we investigated the oncogenic potential of HOXA9 in gliomagenesis, the molecular and cellular mechanisms by which HOXA9 renders glioblastoma more aggressive, and how HOXA9 affects response to chemotherapy and survival. The prognostic value of HOXA9 in glioblastoma patients was validated in two large datasets from TCGA and Rembrandt, where high HOXA9 levels were associated with shorter survival. Transcriptomic analyses identified novel HOXA9-target genes with key roles in cancer-related processes, including cell proliferation, DNA repair, and stem cell maintenance. Functional studies with HOXA9-overexpressing and HOXA9-silenced glioblastoma cell models revealed that HOXA9 promotes cell viability, stemness and invasion, and inhibits apoptosis. Additionally, HOXA9 promoted the malignant transformation of human immortalized astrocytes in an orthotopic in vivo model, and caused tumor-associated death. HOXA9 also mediated resistance to temozolomide treatment in vitro and in vivo via upregulation of BCL2. Importantly, the pharmacological inhibition of BCL2 with the BH3 mimetic ABT-737 reverted temozolomide resistance in HOXA9-positive cells. These data establish HOXA9 as a driver of glioma initiation, aggressiveness and resistance to therapy. In the future, the combination of BH3 mimetics with temozolomide should be further explored as an alternative treatment for glioblastoma.


REpository for Molecular BRAin Neoplasia DaTa (REMBRANDT) analyses
The REMBRANDT platform (https://caintegrator. nci.nih.gov/rembrandt/) [2] was used to evaluate HOXA9 expression and its prognostic value in an independent set of 181 GBM patients. HOXA9 expression was categorized as high for fold-change values > 4 compared to normal samples (corresponding to 95 of 181 patients).

Microarray data validation by RT-PCR
Upon identification of HOXA9-mediated differentially expressed genes in each cell line by microarray data analysis (see statistical methods for details), confirmatory reverse-transcriptase PCR (RT-PCR or qPCR) analyses were performed in a subset of target genes to validate the microarray data. Specifically, genes among the most differentiallyexpressed were selected and tested by RT-PCR or qPCR in each cell line (RAC2, CXCL1, NDRG1, and TOX2 for hTERT/E6/E7 cells; ICAM2, BAMBI, ANGPT2, and PDGFRB for U87MG cells; TOX2, NDRG1, RAC2 and NPR3 for GBML18 cells; C10orf10, PDGFRβ, DKK1 and SOX2 for U251 cells). See Supplementary Table 3 for information about PCR parameters and primer sequences.

Neurosphere formation assay
Cell lines were plated at low density (500 cells/ mL) in NeuroCult NS-A Proliferation Kit (Human) supplemented with 20 ng/mL EGF, 20 ng/mL b-FGF, 20 ng/mL heparin, 2 mM glutamine, and 1% B27 (all from Life Technologies). Cells were incubated for up to 8-15 days. Pictures were taken and quantification was performed by ImageJ software (version 1.44).

Immunohistochemistry and immunofluorescence
Tissues sections were deparaffinized and rehydrated by xylene and ethanol series. Immunohistochemical staining was performed using the LabVision kit (UltraVision Large Volume Detection System Anti-polyvalent, HRP) according to the manufacturer's instructions. For PECAM1 staining, the immunohistochemical protocol was performed using a biotinylated horse anti-goat antibody. For all stainings DAB substrate (DAKO) was used as chromogen, followed by counterstaining with hematoxylin. For immunofluorescence, cells were fixed with Paraformaldehyde (PFA) 4% for 15 minutes and incubated with the primary antibody against Nestin. Nuclei were stained with DAPI (VECTASHIELD ® Mounting Medium with DAPI, Vector Laboratories).
All antibodies used and their respective concentrations are listed in Supplementary Table 2.
Cell viability under basal conditions and after exposure to temozolomide (TMZ, Sigma-Aldrich, dissolved in DMSO) was determined by Trypan blue (Gibco ® ) exclusion assay and MTS (Promega) tests. For the Trypan blue assay, cells were plated at an initial density of 1.5×10 4 cells/well in 6-well plates, in triplicates, and incubated at 37 o C (5% CO 2 ). After 24 hours, hTERT/E6/ E7, U87MG, U251 and GBML18 with differential levels of HOXA9, were treated with TMZ (600 μM, 600 μM, 10 μM and 50 μM, respectively) or vehicle (1% DMSO) in DMEM medium supplemented with 10% FBS for 9 days. Culture medium containing TMZ or vehicle was renewed every 3 days. At each timepoint (1-9 days), total cells were trypsinized and the suspension mixed with trypan blue (1:1 ratio). Viable cells were counted in triplicates under the microscope using hemocytometers. For MTS assays, cells were plated at an initial density of 4x10 3 cells/well in 24-well plates, in triplicates. After 24 hours, cells were treated with DMEM supplemented with 10% FBS and drugs (TMZ and/or ABT-737, Selleck Chemicals) or vehicle (1% DMSO), and incubated for 2, 4, and 6 days (medium with drugs or vehicle was renewed every 3 days). At each time point, cells were exposed to medium containing MTS in a 5:1 ratio for 2 h in a humidified atmosphere at 37°C and 5% CO 2 , and the optical density was determined at 490 nm.

Cell death
Cell lines hTERT/E6/E7, U87MG, U251 and GBML18, with differential levels of HOXA9, were plated at an initial density of 7.5 ×10 4 cells for hTERT/E6/E7, and for the remaining cell lines 1.0 ×10 5 cells, per T25 flask in 3 mL of DMEM supplemented with 10% FBS. After 24 hours, cells were treated with TMZ (600 μM, 600 μM, 10 μM and 50 μM, respectively) or vehicle (1% DMSO). Cell death was evaluated after 5 days of treatment by annexin V-fitc or annexin V-alexa fluor 647, according to the manufacturer's instructions (BD Biosciences), followed by flow cytometry analyses. A total of at least 10,000 events were acquired by flow cytometer. The results were analyzed by FlowJo software (version 7.6).

Cell invasion
Invasion was measured using the BD BioCoat™ Tumor Invasion System (BD Biosciences) as indicated by the manufacturer. Briefly, cell lines were seeded in triplicate into the apical side of Matrigel coated chambers at an initial density of 5×10 4 cells/mL in 500 µL DMEM with 5% FBS, with TMZ or vehicle (1% DMSO). The lower chambers were filled with 500 µL of DMEM containing 10% FBS and 20 ng/mL EGF as a chemoattractant. After 48 h of incubation, cells that invaded through Matrigel to the membrane were stained with DAPI (1:1000, Vector Laboratories) and scanned using an inverted fluorescent microscope. To quantify the number of invading cells, 10 pictures at 20x-magnification were acquired per well, and the total cell numbers were obtained using ImageJ software (version 1.44).

In vivo GBM xenografts
For subcutaneous models, a total of 2x10 6 cells (hTERT/E6/E7-MSCV, hTERT/E6/E7-HOXA9, U87MG-MSCV and U87MG-HOXA9) were injected into the right flank of nude mice (8-weeks old athymic nude Foxn1 nu male mice, from Harlan Laboratories). Tumor size and body weight were measured every 3 days. Tumor volume was calculated by assessing the two largest sides (v = (3.14xL1xL1xL2)/6). After euthanasia, tumors were divided for molecular analyses (frozen samples) or for immunohistochemical analyses, for which samples were fixed by immersion in formalin and subsequently embedded in paraffin.
For intracranial orthotopic models, a total of 2×10 5 cells (hTERT/E6/E7-MSCV, hTERT/E6/E7-HOXA9, U87MG-MSCV and U87MG-HOXA9) were stereotactically injected into the brain striatum (1.8 mm right, 0.4 mm back, and 2.5 mm deep from the bregma) of 8-weeks old athymic nude Foxn1 nu male mice. Mice injected with U87MG cells were randomly assigned to control or treatment groups (8 mice/group) 2 weeks after injection. TMZ treatment was administered by oral gavage with a daily dosage of 50 mg/kg diluted in saline solution, over 2 cycles of 21 days (5 days on and 2 days off). Animals' body weight was evaluated 3 times per week, and general behavior and symptomatology daily. Humane endpoints for sacrifice were established as severe weight loss (>20%), neurological dysfunction, seizures, or moribund condition. All brains were collected for histological and molecular analyses.

Statistical analyses
The effect of HOXA9 in the overall survival of GBM patients from TCGA was represented by Kaplan-Meier survival curves, and the differences evaluated by univariate (Log-rank test) and multivariate survival analysis (Cox proportional hazard model, adjusted for the potential confounding effect of other putative prognostic factors, including patient age, gender, KPS, MGMT methylation status and treatment with chemotherapy). These analyses were made with SPSS 19.0 software (SPSS, Inc.).
For gene expression microarray data processing and analysis, the limma package [5] of the Bioconductor software platform (http://www.bioconductor.org) was used. Data was first pre-processed performing background correction (using the normexp method), normalized between arrays using quantile normalization and log 2 -transformed. Probes representing control spots were filtered, keeping only the ones with status equal to "Gene". Replicated probes (with the same ProbeName) were averaged. Genes consistently differentially expressed across quadriplicates for each paired cell line (hTERT/E6/ E7-MSCV vs. hTERT/E6/E7-HOXA9; U87MG-MSCV vs. U87MG-HOXA9, U251-shControl vs. U251-shHOXA9, and GBML18-shControl vs. GBML18-shHOXA9) were identified by fitting linear models for each gene, followed by the calculation of the relevant statistics using the Empirical Bayes method as implemented by the limma Bioconductor package [6]. Genes were ranked according to their adjusted p-values, where false positive rates were assessed to address the issues related to multiple testing.
For in vitro tests, a repeated measures ANOVA was used to assess differences in Trypan blue viability curves. For MTS viability assay, cell death, cell invasion, cell migration, neurospheres formation, and levels of protein and gene expression, t-tests were used to assess statistical differences.
Overall survival of orthotopic GBM xenografted mice was compared between groups (HOXA9-positive versus HOXA9-negative, with or without TMZ treatment) by the log-rank test, and plotted as Kaplan-Meier curves.
Spearman correlation was used to evaluate the association between the expression levels of HOXA9 in the subcutaneous tumor models and the final tumor volume.
All statistical tests were two-sided, and significance was considered when p < 0.05.