Glioblastoma stem cells reprogram chromatin in vivo to generate selective therapeutic dependencies on DPY30 and phosphodiesterases

To explore the epigenetic drivers of in vivo tumor growth of GSCs, we interrogated our RNA interference (RNAi)–based screen of epigenetic regulators, which was performed simultaneously in patient-derived matched IC tumors in mice and cultured GSCs (1). Among the in vivo–specific dependencies was DPY30, an integral core subunit of SET/MLL family of methyltransferases regulating H3K4me3, particularly its trimethylation (Fig. 1A). Four independent, nonoverlapping short hairpin–mediated RNAs (shRNAs) targeting DPY30 were depleted in the final population of GSCs grown orthotopically (Fig. 1B), whereas the same shRNAs did not dropout in cultured GSCs (Fig. 1C), indicating context-specific dependency on DPY30. We ranked the IC-specific hits by preferential expression in The Cancer Genome Atlas (TCGA) glioblastoma data (https://tcga-data.nci.nih.gov/docs/publications/tcga/-JC) relative to normal brain, with DPY30 as one of the top overexpressed genes in glioblastoma (Fig. 1D). To interrogate the role of DPY30 in GSC biology, we targeted DPY30 expression in GSCs with two nonoverlapping shRNAs (Fig. 1E and fig. S1A). As predicted by the screening results, in vitro viability upon targeting DPY30 was unaffected in three patient-derived GSCs (IN528, 3565, and 1919) in vitro (Fig. 1F). GSC frequency and self-renewal as assayed by extreme limiting dilution assays (ELDAs) were not affected after the depletion of DPY30 in two patient-derived GSCs in culture (fig. S1B). DPY30 knockdown showed no effect on sphere formation in two GSCs relative to a nontargeting control shRNA (fig. S1C). These results confirmed that DPY30 is dispensable for GSC growth in vitro.

The gold standard for cancer stem cells is in vivo tumor initiation. Therefore, we knocked down DPY30 expression in GSCs and then implanted these cells in the brains of immunocompromised NSG [nonobese diabetic (NOD) severe combined immunodeficient (scid) gamma, NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ] mice. Knockdown of DPY30 prolonged tumor latency, as measured by time to onset of neurological signs, and reduced tumor volumes compared to mice bearing GSCs transduced with a nontargeting control shRNA in three patient-derived GSCs (Fig. 1, G to L), confirming that DPY30 knockdown impaired in vivo tumor formation of GSCs. To gain further mechanistic insights of DPY30’s effects on tumor growth and development in vivo, we used an inducible shRNA system to knockdown DPY30 expression in implanted tumors and observed that DPY30 depletion was sufficient to inhibit the growth of preestablished tumors (Fig. 1, M and N). These results suggested that DPY30 was required for the in vivo growth of GSC tumors. IC tumors consist of both GSCs and differentiated glioma cells (DGCs), and DGCs contribute to in vivo tumor growth (19). Because targeting DPY30 had no effect on enriched cultured GSCs, we asked whether effects of DPY30 on tumor growth were mediated via DGCs. We measured DPY30 expression after in vitro serum differentiation of GSCs and observed that differentiation of 3565 and IN528 GSCs did not alter DPY30 mRNA expression (fig. S1D). Furthermore, similar to observations in GSCs, DPY30 depletion had no effect on the viability of DGCs in vitro (fig. S1, E and F). These results established that DPY30 functions were not dependent on the differentiation state of glioma cells. To further support our observation that DPY30 was required for tumor initiation and stemness in vivo, we performed an in vivo limiting dilution assay to quantify the effect of DPY30 on GSC frequency in IC tumors. DPY30 knockdown led to a substantial decrease in stem cell frequency (table S1) and tumor initiation capacity of GSCs in vivo (fig. S2). Collectively, these observations established DPY30 as a selective in vivo–specific dependency in glioblastoma.

To address the molecular effects of DPY30, we analyzed an additional glioblastoma dataset independent from TCGA, the Chinese Glioma Genome Atlas (CGGA) (20), which revealed that DPY30 mRNA expression was up-regulated in patient tumors compared to normal brain tissue (Fig. 2A). DPY30 mRNA expression was also elevated in recurrent glioblastoma samples compared to primary tumors (Fig. 2B). To predict the function of DPY30 in glioblastoma, we interrogated TCGA glioblastoma gene expression data and identified programs correlated with DPY30 expression. DPY30-correlated genes were positively enriched for processes of ribosomes and translation, RNA metabolism, electron transport chain and cellular respiration, hypoxia, regulation of angiogenesis, and peptide synthesis, whereas DPY30 expression negatively correlated with histone modification, transcriptional regulation, and cellular catabolism (Fig. 2C). DPY30 expression strongly correlated with genes involved in angiogenesis (fig. S3, A and B) and hypoxia (Fig. 3, C and D) in glioblastoma samples from the TCGA glioblastoma and low-grade glioma RNA sequencing (RNA-seq) dataset. DPY30-correlated genes were highly enriched for structural constitution of ribosomes, eukaryotic translation, and response to hypoxia (Fig. 2D). Furthermore, DPY30 expression was highly enriched in mesenchymal and classical glioblastomas (Fig. 2E). These data suggested that DPY30 may be important for allowing adaptation of GSCs to a stressful TME through regulation of metabolic and ribosomal processes.

DPY30 regulates H3K4 methylation (21, 22), and H3K4me3 predominantly marks poised and active promoters leading to induction of gene expression (23, 24). To interrogate global regulation of selective gene expression in vivo, we performed H3K4me3 ChIP-seq and RNA-seq on five patient-derived, matched GSCs in CC or IC xenografts in mice (Fig. 3A). We compared H3K4me3 profiles between xenografts and cultured cells, revealing differential H3K4me3 epigenetic landscapes (Fig. 3B). More than 5000 H3K4me3 peaks were uniquely present in vivo (Fig. 3C), among which >2000 peaks were located around the promoter and transcriptional start sites (TSS) (Fig. 3D), indicating a shift in H3K4me3 occupancy in vivo (fig. S4, A to D). Pathway enrichment of gene sets marked by IC-specific H3K4me3 peaks were enriched for programs of transcription factor binding, transcriptional regulation, hypoxia, differentiation, and phosphodiesterase (PDE) activity (Fig. 3E). In contrast, genes with CC-specific H3K4me3 peaks were selectively enriched for cell cycle, neuronal development and differentiation, and cytokine signaling, among others (fig. S4E). Furthermore, extensive differences were evident in gene expression profiles between GSCs grown concurrently for about 3 weeks in IC xenografts and in CC, as determined by RNA-seq (Fig. 3F). Genes highly expressed in IC xenografts were enriched for programs related to neurogenesis (such as nerve growth factor–stimulated transcription, neuronal, and embryonic morphogenesis) and microenvironmental features (such as blood vessel development) (fig. S4F). Gene set enrichment analysis (GSEA) showed that genes selectively up-regulated in IC tumors were highly enriched with a primary glioblastoma signature derived from RNA-seq studies of glioblastoma surgical resection specimens compared to GSCs grown in CC (Fig. 3G) (25). Leveraging clinical TCGA datasets, genes up-regulated in the IC cohort were also up-regulated in glioblastoma tumors compared to normal brain tissue (Fig. 3H). Genes marked by IC-specific H3K4me3 peaks tended to be expressed at higher levels in IC models compared to genes marked by CC-specific H3K4me3 peaks (Fig. 3I), including nuclear factor κ light-chain enhancer of activated B cells 1Z (NFκB1Z), phosphodiesterase 4B (PDE4B), retinol-binding protein 1 (RBP1), and FosB proto-oncogene (FOSB), were highly expressed in GSCs compared to neural stem cells (Fig. 3J). Collectively, the IC TME drives an epigenetic state that manifests in selective gene expression profiles mimicking human glioblastoma signatures.

To dissect TME-specific mechanisms regulated by DPY30 under IC and CC conditions, we interrogated targets downstream of DPY30 using RNA-seq in paired IC and cultured samples from two patient-derived GSCs upon DPY30 knockdown. Depletion of DPY30 induced widespread gene expression changes in IC GSCs (Fig. 4A), with minimal effects in cultured GSCs (Fig. 4B). Because DPY30 regulates H3K4 methylation, we overlapped IC DPY30 knockdown RNA-seq data with H3K4me3 ChIP-seq data to identify the direct targets of DPY30 in glioblastoma. Of 122 genes down-regulated upon DPY30 depletion, 66 were marked by IC-specific H3K4me3 modifications (Fig. 4C). This restricted gene set was enriched for pathways associated with angiogenesis, synaptic transmission, vascular development, and hypoxia (Fig. 4D). The effects of DPY30 knockdown on decreasing mRNA expression of certain targets [including PDE4B, NOP2/Sun RNA methyltransferase family member 7 (NSUN7), and angiogenesis-associated genes, including cholinergic receptor nicotinic alpha 7 subunit (CHRNA7), acetyl–coenzyme A acyltransferase 2 (ACAA2), angiopoietin-like 4 (ANGPTL4), and prostaglandin I2 synthase (PTGIS)] were validated by quantitative polymerase chain reaction (qPCR) in tumors derived from two patient-derived GSCs (Fig. 4, E and F). Although we did not observe a statistically significant decrease in PDE4B expression in IC-specific RNA-seq data, we still validated PDE4B by qPCR as PDE4B expression and the H3K4me3 signal on its promoter were highly elevated in IC tumors compared to cultured GSCs (Fig. 3, B and F). Overlapping the RNA-seq data from IC DPY30 knockdown with IC-specific gene expression data confirmed that genes overexpressed in IC xenografts were commonly down-regulated upon DPY30 depletion (fig. S5A). To establish the importance of DPY30 in glioblastoma tumors, we overlapped the DPY30 knockdown gene expression data with genes overexpressed in gliomas in TCGA and observed that genes down-regulated upon DPY30 depletion were predominantly up-regulated in glioblastoma surgical specimens (Fig. 4G and fig. S5B). We further analyzed the Ivy Glioblastoma Atlas Project multiregional glioblastoma RNA-seq dataset (26) and observed that identified IC-specific DPY30 target genes were primarily up-regulated in microvascular proliferation regions of tumors (Fig. 4H). Targeting DPY30 reduced expression of genes involved in G₂-M checkpoint, mRNA splicing, histone acetylation, and angiogenesis, whereas it elevated the expression of apoptosis-associated genes, as revealed by GSEA (fig. S5C). Collectively, these data show that DPY30 regulates angiogenesis and hypoxia pathways essential for tumor growth in the IC TME but is dispensable in culture.

Hypoxia induces rapid rewiring of H3K4me3 landscape regulating cellular transcriptional response (25). We observed that IC-specific H3K4me3 peaks were enriched for programs of transcriptional regulation and hypoxia (Fig. 3E). Because we found that DPY30 expression was associated with genes involved in response to hypoxia in TCGA glioblastoma tumors (Fig. 2D) and DPY30 regulated hypoxia pathway genes in IC tumors (Fig. 4D), we explored the function of DPY30 in cultured GSCs under hypoxic conditions (1% O₂). We detected decreased viability of DPY30-depleted GSCs under hypoxia (Fig. 5, A and B). The expression of DPY30 target genes identified in IC tumors was also down-regulated upon DPY30 knockdown under hypoxia (Fig. 5, C and D). The expression of PDE4B and ANGPTL4 was induced under hypoxia in a DPY30-dependent manner (Fig. 5, C and D). These results indicate that DPY30 regulates viability and gene expression in GSCs in response to hypoxia in vitro. To evaluate the effect of DPY30 on brain tumor angiogenesis under hypoxic TME, we first measured the migration of endothelial cells in the presence of tumor-conditioned medium from control or DPY30-depleted GSCs, both under hypoxia and normoxia. DPY30 depletion in GSCs compromised hypoxia-induced endothelial cell motility but had no effect under normoxia (Fig. 5, E and F). We then quantified blood vessel density in control and DPY30-depleted xenografts derived from patient-derived GSCs. DPY30-depleted xenografts showed reduced blood vessel density (Fig. 5, G and H, and fig. S6). Overall, these findings demonstrate that DPY30 regulates in vivo GSC tumor formation by supporting survival and angiogenesis in a hypoxic TME. Our results establish a unique angiogenic function for DPY30 in glioblastoma.

To determine the contribution of the H3K4me3 modification to DPY30-mediated gene expression, we performed H3K4me3 ChIP-qPCR analysis to measure occupancy of DPY30 target genes under control- and DPY30-depleted conditions both in IC tumors and cultured GSCs. In contrast to CC conditions, the promoter regions of PDE4B, NSUN7, and other DPY30 target genes were decorated with H3K4me3 in IC xenograft samples. ChIP-qPCR quantification of DNA showed that the DPY30 target genes were enriched for the activating H3K4me3 modification, which was depleted upon DPY30 knockdown only in IC tumors obtained from 3565 and 1919 GSCs (Fig. 6, A to F). H3K4me3 modification was not affected by DPY30 perturbation in CC samples (Fig. 6, A to F). These results establish DPY30 as an IC-specific regulator of H3K4me3 modification and gene expression in glioblastoma.

Comparative analysis of gene expression and H3K4me3 profiles of IC tumors with CC GSCs identified several genes with both high IC-specific mRNA expression and an IC-specific H3K4me3 peak (Fig. 1I). PDE4B was one of the top overexpressed genes under the IC condition in all four GSCs (Fig. 1A) with substantial gained H3K4me3 peaks (Fig. 1F). We observed a decrease in PDE4B expression upon DPY30 depletion in IC tumors (Fig. 4, E and F) in an H3K4me3-dependent manner (Fig. 6A). PDE4B has been linked to the regulation of angiogenesis and hypoxia-inducible factor signaling (27). Thus, we hypothesized that PDE4B was a downstream effector of DPY30 in glioblastoma tumors. We therefore investigated the contribution of PDE4B to glioblastoma tumor growth. Analysis of a TCGA glioblastoma dataset revealed that PDE4B mRNA expression was up-regulated in patient tumors compared to normal brain (Fig. 7A). PDE4B expression was also associated with poor prognosis in the TCGA glioblastoma dataset (Fig. 7, B and C). To predict the function of PDE4B in glioblastoma, we used TCGA glioblastoma gene expression data and identified programs correlated with PDE4B expression. PDE4B-correlated genes were positively enriched for processes of neuron and glia development and differentiation and ion transmembrane transport signatures (fig. S7A), whereas PDE4B expression negatively correlated with cell junction and adhesion, actin cytoskeleton, and epithelial development (fig. S7B). PDE4B-correlated genes were highly enriched for classical glioblastoma and glioblastoma plasticity along with synaptic signaling (Fig. 7D). Furthermore, PDE4B expression highly correlated with DPY30 expression in histologically confirmed high-grade glioblastoma tumors (fig. S7C) in TCGA glioblastoma dataset. PDE4B expression was elevated in IC xenografts compared to cultured cells (Fig. 7E) with a gained H3K4me3 peak on the promoter region of PDE4B (Fig. 7F). Furthermore, we knocked down PDE4B expression in GSCs and implanted these cells in the brains of immunocompromised NSG mice. Knockdown of PDE4B prolonged tumor latency, as measured by time to onset of neurological signs, and reduced tumor volumes compared to mice bearing GSCs transduced with a control shRNA (Fig. 7, G to J). Because PDE4B regulates microvascular density by cyclic adenosine 3′,5′-monophosphate (cAMP) (28), we measured cAMP concentrations in DPY30-depleted cells and IC tumors. DPY30 knockdown did not alter cAMP concentrations in vitro or in vivo (Fig. 7, D and E). These results suggest that DPY30 and PDE4B functions in IC tumor growth are independent of cAMP signaling.

Because DPY30 does not have direct enzymatic activity, DPY30 function has been targeted indirectly through a peptidomimetic inhibitor of its interaction with ASH2 like (ASH2L), but this agent is not amenable for IC treatment. Therefore, we sought DPY30-based druggable therapeutic dependencies leveraging the PRISM (profiling relative inhibition simultaneously in mixtures) dataset, which contains the results of pooled cell line chemical perturbation viability screens for 4518 compounds screened against over 500 cell lines (https://depmap.org/repurposing) (29, 30). Correlation of drug sensitivity [area under the curve (AUC)] with DPY30 mRNA expression in cancer cells revealed that high DPY30 expression correlated with sensitivity to a phosphatidylinositol 3-kinase (PI3K) inhibitor, a phosphodiesterase inhibitor, and a microtubule inhibitor (Fig. 8A). Because we had identified the phosphodiesterase PDE4B as one of the top up-regulated genes with gained H3K4me3 peaks in IC tumors, we prioritized phosphodiesterase inhibitor therapy for further investigation. The phosphodiesterase inhibitor rolipram has already been used in humans and rodents for its anti-inflammatory and antidepressant actions (31–33). Rolipram potentiates bevacizumab-induced cell death in human glioblastoma stem-like cells (34). We found that rolipram treatment had mild effects on the in vitro cell viability of a panel of GSCs at millimolar concentrations (Fig. 8B). In contrast, in vivo intraperitoneal administration of rolipram (75 mg/kg body weight) prolonged tumor latency, as measured by time to onset of neurological signs, and reduced tumor volumes compared to mice bearing xenografts from patient-derived GSCs (Fig. 8, C to F). Thus, the phosphodiesterase inhibitor rolipram can target GSCs grown in IC xenografts and may serve as a useful therapy in glioblastoma.

The purpose of this study was to critically evaluate the functions of the H3K4me3 regulator DPY30 in glioblastoma, because it was identified as an in vivo–specific dependency in our previous study (7) and subsequently to identify previously unidentified druggable targets. We used H3K4me3 and transcriptome analyses of paired cultured GSCs and IC tumors derived from these GSCs implanted in mice after 21 days of cell/tumor growth under respective conditions to identify specific targets of DPY30 under TC and IC conditions. We used limiting dilution assays and inducible RNAi to establish the role of DPY30 in regulating stemness and tumor propagation potential of GSCs in IC tumors. In all animal experiments, cages of mice were randomly assigned to treatment groups. The personnel injecting and imaging the mice were blinded to the names of treatment groups. By leveraging the PRISM dataset of pooled cell line chemical perturbation viability screens, we identified rolipram as a therapy for glioblastoma.

Glioblastoma tissues were obtained from excess surgical resection samples from patients after review by a neuropathologist and used in accordance with an approved protocol by the Institutional Review Board at Cleveland Clinic. As previously described (5), tumor cells were derived immediately after dissociation of primary patient tumor or after transient xenograft passage. All GSCs were cultured in Neurobasal medium (Invitrogen) supplemented with B27 without vitamin A (Invitrogen), epidermal growth factor, and basic fibroblast growth factor (20 ng/ml each; R&D Systems), sodium pyruvate, and GlutaMAX. To decrease the incidence of CC-based artifacts, patient-derived xenografts were produced and propagated as a renewable source of tumor cells for study. Short tandem repeat analyses were performed (conducted by the Duke University, Cell Line Authentication Service) to authenticate the identity of each tumor model used in these studies on a yearly basis. Cells were frozen and stored at −196°C (liquid nitrogen) when not being actively cultured. Mycoplasma testing was performed by qPCR on cellular supernatants on a yearly basis. Cells were grown for fewer than 20 in vitro passages from xenografts. DGCs were cultured in 10% serum Dulbecco’s modified Eagle’s medium to allow cell attachment and survival. To induce hypoxia, cells were cultured in hypoxia chamber (Sanyo) to maintain 1% O₂.

For DPY30 shRNA (shDPY30-263, TRCN #0000131112; shDPY30-403, TRCN #0000129317) and PDE4B shRNA (shPDE4B-1083, TRCN #0000048818; shPDE4B-525, TRCN #0000048819) experiments, IC xenografts were generated by implanting 5000 human-derived GSCs into the right cerebral cortex of NSG mice (the Jackson Laboratory) at a depth of 3.5 mm under a University of California, San Diego Institutional Animal Care and Use Committee–approved protocol. Brains were harvested after the first mice in any cohort (eight mice per cohort) presented neurological signs and fixed in 4% formaldehyde, dehydrated in 30% sucrose, and then cryosectioned. Hematoxylin and eosin (H&E) staining was performed on sections for histological analysis. In parallel survival experiments, mice were observed until the development of neurological signs.

For inducible shRNA experiments, 3565 GSCs were transduced with either TRIPZ-inducible lentiviral nonsilencing shRNA control (RHS4743, Horizon Discovery Biosciences Limited) or TRIPZ-inducible lentiviral human DPY30 shRNA (RHS4696-20067912020, clone ID: V2THS_159340, Horizon Discovery Biosciences Limited), and successfully transduced cells were selected by puromycin (1 μg/ml) for 3 days. To monitor tumor growth in living animals, all GSCs used for animal studies were transduced with firefly luciferase through lentiviral infection. Five thousand puromycin-selected cells were implanted into the right cerebral cortex of NSG mice. The mice implanted with GSCs transduced with shDPY30 shRNA were distributed across two arms, 10 days after implantation: (i) in which mice were left uninduced (no shRNA expression) and (ii) in which mice were treated with doxycycline to induce shRNA expression. To examine tumor growth, mice implanted with GSCs were monitored by bioluminescent imaging. Animals were treated with 120 mg/kg body weight d-luciferin intraperitoneally and anesthetized with isoflurane for the imaging analysis. The tumor luciferase images were captured using an IVIS imaging system (Spectrum CT, PerkinElmer).

For in vivo drug treatment studies, IC xenografts were generated by implanting 5000 patient-derived GSCs (3565 and 1919) into the right cerebral cortex of NSG mice as described above. Mice recovered for 7 days and then were randomly assigned into control versus treatment groups by a blinded investigator. Mice were treated with either rolipram (catalog no. R6520, MilliporeSigma) or an equivalent concentration of dimethyl sulfoxide resuspended in 0.9% saline by intraperitoneal injection daily for 3 weeks. Healthy, wild-type male or female mice of NSG background, 4 to 6 weeks old, were randomly selected and used in this study. Mice had not undergone prior treatment or procedures. Mice were maintained in 14-hour light/10-hour dark cycle by animal husbandry staff with no more than five mice per cage. Experimental animals were housed together. Housing conditions and animal status were supervised by a veterinarian. Animals were monitored daily until neurological signs were observed, at which point they were sacrificed. Neurological signs or signs of morbidity included hunched posture, gait changes, lethargy, and weigh loss. For in vivo LDA experiments, 3565 and 1919 GSCs, transduced either with nontargeting control shRNA or shDPY30-263 vectors, were implanted at different numbers (50,000, 20,000, 10,000, 5000, 1000, and 100) in the right cerebral cortex of NSG mice as described above. Mice were maintained up to 4 months or until the development of neurological symptoms. ELDA was performed using software available at http://bioinf.wehi.edu.au/software/elda, as previously described (46, 47).

Mice were deeply anesthetized by intraperitoneal injections of ketamine (60 mg/kg) and xylazine (40 mg/kg) and transcardially perfused with 15 ml of cold 1× phosphate-buffered saline (PBS) (pH 7.4), followed by 30 ml of cold neutral buffered paraformaldehyde [4% (w/v) in 1× PBS (pH 7.4)]. Tissues were postfixed in the same solution overnight at 4°C, followed by 30% sucrose [in 1× PBS (pH 7.4)] for 72 hours. Twenty-micrometer cryosections were prepared and mounted onto SuperFrost Plus slides and stored at −80°C until processing for lectin staining. For lectin staining, sections were washed three times, 5 min each, with 1× PBS (pH 7.4), and incubated with DyLight 594–labeled Lycopersicon esculentum (Tomato) lectin (catalog no. DL-1177, RRID:AB_2336416, Vector Laboratories; 1:200) diluted in 1× PBS (pH 7.4) with 0.2% Triton X-100 for 2 hours at room temperature. Sections were then washed three times, 10 min each, with 1× PBS with 0.2% Triton X-100 and cover-slipped using ProLong Diamond Antifade Mountant with 4′,6-diamidino-2-phenylindole (DAPI; P36962, Life Technologies). Optical sections z stacks were imaged using ×60 magnification on Leica Confocal SPE (Sanford Consortium, University of California, San Diego facility) and processed using ImageJ software (National Institutes of Health). For von Willebrand factor (vWF) staining, sections were washed three times, 5 min each, with 1× PBS (pH 7.4) and blocked in PBS with 0.6% Triton X-100 and 10% normal goat serum (blocking buffer) for 2 hours. Sections were incubated with vWF antibody (catalog no. AB7356, RRID:AB_92216, Millipore) at a 1:250 dilution in blocking buffer at 4°C overnight. The next day, the sections were washed three times with PBS with 0.4% Triton X-100 (wash buffer) and incubated with goat anti-rabbit secondary antibody (#A11008, Life Technologies). Sections were washed three times in wash buffer, and coverslips were mounted using ProLong Gold Antifade (Life Technologies).

For DPY30 knockdown RNA-seq in in vivo xenografts and in vitro culture, patient-derived GSCs (3565 and 1919) were transduced with either control or two independent nonoverlapping shRNAs for DPY30 (shDPY30-263, TRCN #0000131112; shDPY30-403, TRCN #0000129317), and successfully transduced cells were selected by puromycin for 3 days. Fifty thousand selected cells were implanted in mouse brain and were allowed to grow for 3 weeks. After 3 weeks, mice were euthanized, and the tumor was harvested, dissociated to single cells, and depleted of any infiltrating mouse cells using magnetic-activated cell sorting (MACS; Mouse Cell Depletion Kit, Miltenyi). Tumors from three mice were pooled together as one replicate. A fraction of selected cells was also maintained in standard GSC culture for 3 weeks in presence of puromycin before harvesting. Cells from in vivo xenografts and in vitro culture were processed for RNA isolation simultaneously.

RNA-seq was performed as previously described (49). Total RNA was extracted from flash-frozen cells or pulverized tissue using the miRNeasy Mini Kit (catalog no. 217004, QIAGEN) in accordance with the manufacturer’s protocols. Libraries were prepared and sequenced by BioGene. Stranded RNA library preparation was performed with ribosomal RNA depletion according to instructions from the manufacturer (Epicentre). Paired-end sequencing was performed on the Illumina HiSeq 2500 with 2× 150–base pair (bp) paired-end read configuration. FASTQ sequencing reads were trimmed using Trim Galore (www.bioinformatics.babraham.ac.uk/projects/trim:galore/), and transcript quantification was performed using Salmon (RRID:SCR_017036) in the quasi-mapping mode (50). Salmon “quant” files were converted using tximport (https://bioconductor.org/packages/release/bioc/html/tximport.html), and differential expression analysis was performed using DESeq2 (DESeq, RRID:SCR_000154) (51). GSEA was performed by selecting differentially expressed genes, generating a preranked list, and weight by the inverse of the false discovery rate (FDR) multiplied by the sign of the log₂ fold change and inputting the preranked list into the GSEA desktop application (http://software.broadinstitute.org/gsea/downloads.jsp) (52). Pathway enrichment bubble plots were generated using the Bader Lab Enrichment Map application (53) and Cytoscape (RRID:SCR_003032) (www.cytoscape.org).

Cells were collected and lysed in radioimmunoprecipitation assay buffer [50 mM tris-HCl (pH 7.5), 150 mM NaCl, 0.5% NP-40, and 50 mM NaF with protease inhibitors) and incubated on ice for 1 hour. Lysates were centrifuged at 4°C for 20 min at 14,000 rpm, and supernatant was collected. The BCA (Bicinchoninic acid) assay (Bio-Rad Laboratories) was used for determination of protein concentration. Equal amounts of protein samples were mixed with SDS Laemmli loading buffer, boiled for 5 min, electrophoresed using NuPAGE Bis-Tris Gels, and then transferred onto Nitrocellulose membranes. TBS-T (Tris-buffered saline with 0.1% Tween 20 detergent) supplemented with 5% nonfat dry milk was used for blocking for a period of 1 hour, followed by blotting with DPY30 (catalog no. ab137690, Abcam) and glyceraldehyde-3-phosphate dehydrogenase (catalog no. HRP-60004, RRID:AB_2737588, ProteinTech) primary antibodies at appropriate dilution at 4°C for 16 hours. Blots were washed five times for 5 min each with TBS-T and then incubated with appropriate secondary antibody in 5% nonfat milk in TBS-T for 1 hour. For all Western immunoblot experiments, blots were imaged using Bio-Rad Image Lab software (Image Lab Software, RRID:SCR_014210) and subsequently processed using Adobe Illustrator to create the figures.

For survival analyses, TCGA data (54) were downloaded using the “TCGA2STAT” R package (55). The Kaplan-Meier survival analysis with the log-rank analysis was used to assess prognostic significance of every gene in the TCGA GBM HG-U133A microarray and GBM or GBMLGG RNA-seq datasets (56). The processed UCSC TOIL analysis of TCGA and GTEx (The Genotype-Tissue Expression) RNA-seq data were used to determine genes that were differentially expressed between glioblastoma specimens and normal brain specimens (57). The Cox proportional hazards model and log-rank analysis were used to assess prognostic significance of every gene in the TCGA GBM HG-U133A microarray dataset regardless of IDH (isocitrate dehydrogenases) mutation status.

Therapeutic sensitivity and gene expression data were accessed through the Cancer Therapeutics Response Portal (https://portals.broadinstitute.org/ctrp/) (29, 30, 58). Gene signature scores were calculated for each cell line in the dataset using the single-sample Gene Set Enrichment Analysis Projection module on GenePattern (https://cloud.genepattern.org). The gene signature score was then correlated with AUC values for drug sensitivity for each compound tested. Correlation r value was plotted, and statistical analyses were corrected for multiple test correction.

Paired IC tumors and TC cells were used for comparative H3K4me3 ChIP-seq in IC versus CC samples. Five GSC models (3565, 1919, IN528, 387, and GSC23) were implanted in mouse brains (20,000 cells per mouse) and allowed to grow for 3 weeks. Cells were also maintained under standard serum-free CC conditions at the same time. After 3 weeks, mice were euthanized, and tumors were harvested, dissociated to single cells, and depleted of any infiltrating mouse cells using MACS (Mouse Cell Depletion Kit, Miltenyi). Tumors from three mice were pooled together. Cells from culture were collected at the same time and processed for ChIP assay. H3K4me3 ChIP assays were performed with 3 million cells using a Magna ChIP-seq kit (17-10085, MilliporeSigma) according to the manufacturer’s instructions. Briefly, cells were fixed with 1% formaldehyde at room temperature for 10 min, and the reaction was subsequently quenched with 0.125 M glycine. Then, the cells were washed with cold PBS and suspended with lysis buffer supplemented with Protease Inhibitor Cocktail II. Chromatin was sonicated to obtain DNA fragments with an average length of 200 to 500 bp by a Bioruptor Pico sonication device (Diagenode), followed by centrifugation for clearing. The soluble fraction of sheared chromatin was diluted 10-fold with ChIP dilution buffer and incubated with 5 μl of H3K4me3 antibody (catalog no. 39159, Active Motif) conjugated to EZ-Magna ChIP A/G magnetic beads overnight at 4°C with rotation. Chromatin-captured beads were washed once with low salt wash buffer, once with high salt wash buffer, once with LiCl wash buffer, and once with TE (Tris-EDTA) buffer for 10 min each by rotation at 4°C. Beads were resuspended in elution buffer, and the supernatant was removed by magnet. Proteinase K was added to the eluted DNA, followed by incubation at 62°C for 4 hours and 95°C for 10 min with shaking. Reverse cross-linked DNA was purified using purification columns and quantified using PicoGreen 628 (Invitrogen). Eluted DNA was used for sequencing library preparation or qPCR. Sequencing libraries were prepared using ThruPLEX NGS Library Preparation Kits (R400427, Rubicon Genomics) according to the manufacturer’s instructions (49). After library amplification, DNA fragments were agarose gel (1.0%) size–selected (<1 kb), analyzed using a Bioanalyzer (Agilent Technologies), and sequenced at Genewiz using Illumina HiSeq 2500 2×150-bp paired-end reads. ChIP-qPCR was performed using the SYBR Green PCR Master Mix (4309155, Applied Biosystems). Two primer sets were used for each target gene tested using the primers listed in table S2.

ChIP-seq reads were trimmed using Trim Galore v0.4.3 (www.bioinformatics.babraham.ac.uk/projects/trim:galore/) and cutadapt 1.14 (http://cutadapt.readthedocs.io/en/stable/guide.html). Reads were aligned to the hg19 human genome with BWA-MEM v0.7.17. BAM files were processed using SAMtools, and PCR duplicates were removed with Picard Tools (http://broadinstitute.github.io/picard/). H3K4me3 peaks were called using MACS2 (v2.1.1) using a ChIP input file as a control using default settings. bigWig track coverage files were generated from merged BAM files using the deepTools (v2.4.1) bamCoverage command with RPKM (per million mapped reads) normalization (http://deeptools.readthedocs.io/en/latest/index.html). DiffBind was used to identify differential H3K4me3 ChIP-seq peaks between IC and CC samples and to perform principal components analysis.

TRIzol reagent (Sigma-Aldrich) was used to isolate total cellular RNA from cell pellets according to the manufacturer’s instructions. The high-capacity cDNA reverse transcription kit (catalog no. 4368814, Thermo Fisher Scientific) was used for reverse transcription into cDNA. Quantitative real-time PCR was performed using Applied Biosystems 7900HT cycler using a Radiant Green Hi-ROX qPCR kit (catalog no. QS2050). qPCR primers used in this study are listed in table S3.

All statistical analyses are described in the figure legends. For TCGA glioblastoma versus normal brain RNA-seq calculations, four-way analysis of variance (ANOVA) controlling for sex, age, and ethnicity with Benjamini and Hochberg FDR method was used. For survival analyses, the Cox proportional hazards and log-rank analyses were used. For qPCR analyses, Student’s t test was used to assess statistical significance, when appropriate. For proliferation and gene expression assays, one-way ANOVA was used for statistical analysis with Dunnett’s multiple hypothesis test correction.