Towards precision medicine-based therapies for glioblastoma: interrogating human disease genomics and mouse phenotypes
© The Author(s) 2016
Published: 22 August 2016
Glioblastoma (GBM) is the most common and aggressive brain tumors. It has poor prognosis even with optimal radio- and chemo-therapies. Since GBM is highly heterogeneous, drugs that target on specific molecular profiles of individual tumors may achieve maximized efficacy. Currently, the Cancer Genome Atlas (TCGA) projects have identified hundreds of GBM-associated genes. We develop a drug repositioning approach combining disease genomics and mouse phenotype data towards predicting targeted therapies for GBM.
We first identified disease specific mouse phenotypes using the most recently discovered GBM genes. Then we systematically searched all FDA-approved drugs for candidates that share similar mouse phenotype profiles with GBM. We evaluated the ranks for approved and novel GBM drugs, and compared with an existing approach, which also use the mouse phenotype data but not the disease genomics data.
We achieved significantly higher ranks for the approved and novel GBM drugs than the earlier approach. For all positive examples of GBM drugs, we achieved a median rank of 9.2 45.6 of the top predictions have been demonstrated effective in inhibiting the growth of human GBM cells.
We developed a computational drug repositioning approach based on both genomic and phenotypic data. Our approach prioritized existing GBM drugs and outperformed a recent approach. Overall, our approach shows potential in discovering new targeted therapies for GBM.
Glioblastoma (GBM) is one of the leading causes of cancer-related deaths in both the pediatric and adult populations . The standard treatment includes radiation plus chemotherapy following maximal safe resection of cancer mass . However, the prognosis of GBM patients remains poor even with optimal radio- and chemo-therapies: the mean survival is 15 months and most patients die within two years [2, 3]. In addition, GBM is not a priority for new drug development because of socioeconomic problems and medical difficulties . Both the grim prognosis and urgent clinical needs have motivated us to develop an “in silico” drug repositioning approach and pursue FDA-approved agents that has the potential to treat GBM but not previously identified as GBM therapeutics.
Since GBMs are highly heterogeneous at the genomic, histological and differentiation level, the lack of specific therapies contributes to the treatment failures. Cancer therapies that target on specific molecular profiles of individual tumors have the potential to maximize the efficacy . For example, Imatinib has been used to successfully treat a subtype of leukemia with mutations in the BCR-ABL fusion protein and has achieved a median survival of five years . Over the past two decades, extensive researches have identified hundreds of genetic mutations that likely drive the GBM formation [6, 7]. More recently, systemic multi-platform analysis of glioma and bioinformatic mining by The Cancer Genome Atlas (TCGA) has led to the classification of GBM into distinct molecular subtypes according to the genes altered during gliomagenesis [8, 9]. Here, we use the accumulated genomic data for GBM to guide the drug repositioning approach towards discovering precise targeted drugs for GBM.
In this study, we develop a novel GBM drug repositioning strategy leveraging both lower-level disease and drug genomics and higher-level mouse phenotypes. We first identify GBM-specific mouse phenotypes using a compiled list of GBM-associated genes identified by multiple TCGA studies [6, 8]. Then we screen all the FDA-approved drugs for candidates that share similar mouse phenotype profiles with GBM. We validate the approach using approved GBM drugs, and approximate the performance in detecting novel GBM drugs using two evaluation sets: a set of potential GBM therapies tested in clinical trials and a set of off-label GBM drugs in the post-marketing surveillance system. Finally, we investigate the top 10 % drug predictions. Overall, we combine the genomic and phenotypic data for diseases and drugs towards identifying novel targeted therapies for GBM.
Identify mouse phenotype profiles for GBM and drugs using disease genetics and drug target genes
TCGA Research Network provides a comprehensive catalog of genomic abnormalities driving tumorgenesis. We compiled a list of 102 genetic mutations that significantly differentiate GBM tumors and healthy tissue from several recent TCGA studies [6, 8]. The list primarily contains the genes in the core GBM-associated pathways, including p53, Rb, and receptor tyrosine kinase (RTK)/Ras/phosphoinositide 3-kinase (PI3K) signaling.
We extracted the mouse phenotypes that are linked to 102 GBM-associated genes from the MGI database. Each phenotype was weighted and ranked by the number of genes it is linked with. The mouse phenotype terms were removed from the list if their weights are smaller than the median of all weights. At last, we identified a list of 945 GBM-specific mouse phenotypes. We mapped the phenotype terms into 26 categories by tracing the isa relationship in the mammalian phenotype ontology. A score was calculated for each category as the sum of weights of all phenotypes in it. We ranked the phenotype categories by their scores and investigated the top five categories.
Then we identified the mouse phenotype profile for each of the 1348 FDA-approved drug. The drug target genes were first extracted from the STITCH database, and each drug-target link has a confidence score. Then we extracted the mouse phenotypes that are linked with the target genes for each drug. The phenotype terms are weighted by the sum of confidence scores of the corresponding target genes. Finally, we obtained a vector of weighted mouse phenotype features for each candidate drug.
Rank candidate drugs for GBM using mouse phenotype similarities between GBM and drugs
We calculated the phenotypic similarity between GBM and the drugs in order to rank the candidate drugs by their similarity to GBM. Phenotype terms associated with both GBM and the drugs were normalized by concepts in the ontology, which provides semantic relationships between concepts and has been widely used in biomedical applications [17, 21, 23, 24]. We calculated the semantic distances between the mouse phenotype vectors for GBM and the candidate drugs in the context of the mouse phenotype ontology.
We first quantified the information content for each phenotype term t as −l o g p(t), in which p(t) represents the frequency among phenotype annotations to all the 7568 mouse genes. In calculating the information content, if a gene is annotated by one phenotype term, we assumed that it is also annotated by the ancestors of this term in the hierarchy of mammalian phenotype ontology. Hence, a phenotype term has higher information content than its ancestors, which lie on higher levels in the ontology.
A similar definition of distance between a pair of concepts in an ontology was also used before .
Validate our approach through de novo prediction of approved, potential, as well as off-label GBM drugs
We tested whether our approach can prioritize the existing and novel GBM drug therapies in the top among 1348 candidates. We compiled three evaluation drug sets based on previous studies [25, 26]: the approved GBM drugs, potential GBM drugs that have been tested in clinical trials, and off-label GBM drugs identified from a post-marketing drug surveillance system. The approved GBM drug set contains temozolomide and carmustine, which are cytotoxic (non-targeted) chemical drugs, and bevacizumab, which is the first targeted drug approved for brain tumor. The potential GBM drug set contains 52 drugs collected from the clinical trials. In addition, the FDA drug surveillance system contains large-scale drug-disease data collected from hospitals, patients, and pharmaceutical companies. A total of 36 off-label uses for GBM were extracted from this system (containing zero overlap with the 52 potential GBM drugs). We have removed the approved GBM drugs from both the potential and off-label GBM drug sets, and evaluated the ranks for these two sets to approximate the performance of the proposed approach in predicting novel GBM drugs.
We compared the performance of our approach with a recent drug repositioning approach proposed by Hoehndorf  in ranking the above evaluation sets. The Hoehndorf’s method also used the mouse phenotype data, but did not incorporate the human disease genomics data. They matched the human phenotype ontology  and the mammalian phenotype ontology  to predict genes for a human disease using the gene-phenotype relationships in animal models. After that, they linked the predicted disease genes with the drug target genes to suggest candidate drugs for the given disease.
We first evaluated the ranks for approved GBM drugs, and the median ranks for the potential and off-label GBM drug sets. We tested the median rank instead of the average, because the median is not affected by individual large ranks. For example, if the rank for a GBM drug is 1348 using method A and 1000 using method B, both methods fail in detecting this positive example. But method B may achieve much higher average ranks than method A, affected by the large values of these two ranks. We also extracted the overlapping drugs between the ranked drug lists generated by the two methods, and performed the paired student’s t-test to evaluate the significance of their ranking difference. Then we combined the three evaluation sets, assumed all the drugs as the positive examples, and compared the precision-recall curve as well as the mean average precision between methods.
Identified mouse phenotypes are associated with GBM pathogenesis
The top-ranked categories of GBM-specific mouse phenotypes detected through disease genetics
Increased glioblastoma incidence
Nervous system phenotype
Abnormal astrocyte morphology
Decreased survivor rate
Immune system phenotype
Decreased leukocyte cell number
Our approach outperforms an existing drug repositioning approach in prioritizing approved, potential and off-label GBM drugs
Using the phenotype profiles detected through GBM and drug genomics, our approach prioritized the approved GBM drug bevacizumab in top 24.4 % among a total of 1348 chemicals, which is a much higher rank than Hoehndorf’s rank in 67.9 %. In addition, we identified the other two GBM approved drugs temozolomide and carmustine and ranked them within top 6.7 % and 10.7 %, respectively, while Hoehndorf’s ranking list does not contain these two drugs.
The ranks for GBM drugs in the three evaluation sets generated by our approach and Hoehndorf’s approach
Evaluation drug (set)
Potential drugs (clinical trials)
Off-label drugs (post-marketing surveillance)
Median ranks for different types of drugs in the combined evaluation set
Non-targeted cancer therapies (chemotherapies)
Targeted cancer drugs
Together, the result suggests that our approach performed significantly better than an existing method that also utilizes the mouse phenotype data in prioritizing all approved and novel GBM drugs, and specially in identifying potential targeted GBM drugs. One possible reason is that we used the most recent discoveries of GBM associated gene mutations and a more comprehensive drug-target database, which provides opportunities for discovering targeted therapies for GBM.
Examples in our top 5 % drug predictions for GBM
Type 2 diabetes
Symptoms of menopause
High cholesterol and triglyceride
In this study, we predict candidate targeted drugs for GBM through combining discoveries on disease genomics and large-scale mouse phenotype data. We currently have not considered the blood-brain barrier (BBB) permeability of the candidate drugs, which is a major challenge for drug discovery for CNS diseases. No readily available BBB permeable drug database can be publicly accessed to enable simple filtering among the candidate GBM drugs. Computational approaches based on decision tree have been developed to identify BBB permeable drugs . It is also possible to modify the drug chemically or pharmaceutically to increase its permeability . In summary, our future work contains further selecting the candidate GBM drugs that can be delivered into the brain.
TCGA recently classified GBM into four types: Proneural, Neural, Classical and Mesenchymal [6, 8]. Each class has distinct genomic profiles. The Classical GBM has increased EGFR expression and lacks TP53 mutations. The Proneural subtype shows alterations of PDGFRA and point mutations in IDH1. The Neural subtype is characterized by expressions of neuron markers. And the Mesenchymal GBM shows deletions of NF1, expression of mesenchymal markers, and high expressions of the TNF super family pathway and NF- κB pathway . Patients of the four types also respond differently to chemo- and/or radiotherapy . In the future, We will predict drugs for each of the four types targeting on their distinct genetic and genomic features towards achieving precision medicine for GBM. We expect specific and different drug predictions across the GBM subtypes.
In addition, human disease phenotypes, disease phenotypic similarities and drug similarities may also contribute to GBM drug repositioning. For the drug-gene interaction database, we currently use the STITCH database, but other sources like Cancer Cell Line Encyclopedia (CCLE)  may contain different knowledge. In the future, we will develop algorithms to seamlessly integrate more comprehensive data to further filter strong candidate GBM drugs. We will also test the candidate drugs in biomedical experiments and clinical studies.
We screened 1348 approved drugs and predicted targeted drugs for GBM through combining disease genomic and mouse phenotype data. Our approach prioritized the approved GBM drugs, and outperformed a recent drug repositioning method in identifying novel GBM drugs. For all positive examples of GBM drugs, we achieved a median rank of 9.2 %, comparing to 45.6 % generated by the earlier approach. In the paired t-test, our approach generated significantly higher ranks for the evaluation drugs than the baseline approach (p=3e −4). We found that many of our top-ranked predictions have been demonstrated effective in inhibiting the growth of human GBM cells. Overall, the results show that our drug repositioning approach has the potential in finding new targeted therapies for GBM.
GBM, glioblastoma; TCGA, the cancer genome atlas; FDA, food and drug administration; MGI, mouse genome informatics; RTK, receptor tyrosine kinase; PI3K, phosphoinositide 3-kinase; STITCH, search tool for interactions of chemicals; BBB, blood-brain barrier; CNS, central nervous system; ccle, cancer cell line encyclopedia; MAP, mean average precisio
YC, ZG and RX are supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under the NIH Director’s New Innovator Award DP2HD084068. BW is supported by NCI award R01CA155676 and R01CA152371.
The publication costs for this article were funded by the corresponding author.
This article has been published as part of BMC Genomics Volume 17 Supplement 7, 2016: Selected articles from the International Conference on Intelligent Biology and Medicine (ICIBM) 2015: genomics. The full contents of the supplement are available online at http://bmcgenomics.biomedcentral.com/articles/supplements/volume-17-supplement-7.
Availability of data and materials
Data is available by contacting Rong Xu at email@example.com.
RX conceived the study. YC designed the methods, performed the experiments and wrote the manuscript. All authors have participated study discussion and manuscript preparation. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Wen PY, Kesari S. Malignant gliomas in adults. New Engl J Med. 2008; 359(5):492–507.View ArticlePubMedGoogle Scholar
- Stupp R, Tonn JC, Brada M, Pentheroudakis G, Group EGW, et al.High-grade malignant glioma: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2010; 21(suppl 5):v190—v193.PubMedGoogle Scholar
- Omuro A, DeAngelis LM. Glioblastoma and other malignant gliomas: a clinical review. Jama. 2013; 310(17):1842–1850.View ArticlePubMedGoogle Scholar
- Bai RY, Staedtke V, Riggins GJ. Molecular targeting of glioblastoma: drug discovery and therapies. Trends Mol Med. 2011; 17(6):301–312.View ArticlePubMedPubMed CentralGoogle Scholar
- Fausel C. Targeted chronic myeloid leukemia therapy: seeking a cure. Am J Health Syst Pharm. 2007; 64(24 Supplement 15):S9—S15.PubMedGoogle Scholar
- Brennan CW, Verhaak RG, McKenna A, Campos B, Noushmehr H, Salama SR, et al.The somatic genomic landscape of glioblastoma. Cell. 2013; 155(2):462–477.View ArticlePubMedPubMed CentralGoogle Scholar
- Ostrom QT, Bauchet L, Davis FG, Deltour I, Fisher JL, Langer CE, et al. The epidemiology of glioma in adults: “state of the science” review. Neuro Oncol. 2014; 16(7):896–913.View ArticlePubMedPubMed CentralGoogle Scholar
- Verhaak RG, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, et al.Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer cell. 2010; 17(1):98–110.View ArticlePubMed CentralGoogle Scholar
- Network TCGAR. Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med. 2015; 372:2481–2498.View ArticleGoogle Scholar
- Sanseau P, Agarwal P, Barnes MR, Pastinen T, Richards JB, Cardon LR, et al.Use of genome-wide association studies for drug repositioning. Nat Biotechnol. 2012; 30(4):317–320.View ArticlePubMedGoogle Scholar
- Wang ZY, Zhang HY. Rational drug repositioning by medical genetics. Nat Biotechnol. 2013; 31(12):1080–1082.View ArticlePubMedGoogle Scholar
- Okada Y, Wu D, Trynka G, Raj T, Terao C, Ikari K, et al.Genetics of rheumatoid arthritis contributes to biology and drug discovery. Nature. 2014; 506(7488):376–381.View ArticlePubMedGoogle Scholar
- Chen Y, Xu R. Network-based gene prediction for Plasmodium falciparum malaria towards genetics-based drug discovery. BMC Genomics. 2015; 16(Suppl 7):S9.View ArticlePubMedPubMed CentralGoogle Scholar
- Dudley JT, Sirota M, Shenoy M, Pai RK, Roedder S, Chiang AP, et al. Computational repositioning of the anticonvulsant topiramate for inflammatory bowel disease. Sci Transl Med. 2011; 3(96):96ra76–96ra76.View ArticlePubMedPubMed CentralGoogle Scholar
- Jahchan NS, Dudley JT, Mazur PK, Flores N, Yang D, Palmerton A, et al.A drug repositioning approach identifies tricyclic antidepressants as inhibitors of small cell lung cancer and other neuroendocrine tumors. Cancer discovery. 2013; 3(12):1364–1377.View ArticlePubMedGoogle Scholar
- Chen Y, Li L, Zhang GQ, Xu R. Phenome-driven disease genetics prediction toward drug discovery. Bioinformatics. 2015; 31(12):i276—i283.View ArticlePubMedPubMed CentralGoogle Scholar
- Chen Y, Zhang X, Zhang Gq, Xu R. Comparative analysis of a novel disease phenotype network based on clinical manifestations. J Biomed Inform. 2015; 53:113–120.View ArticlePubMedGoogle Scholar
- Chen Y, Li L, Xu R. Disease Comorbidity Network Guides the Detection of Molecular Evidence for the Link Between Colorectal Cancer and Obesity. AMIA Summits Transl Sci Proc. 2015; 2015:201.PubMedPubMed CentralGoogle Scholar
- Xu R, Wang Q. PhenoPredict: A disease phenome-wide drug repositioning approach towards schizophrenia drug discovery. J biomed inform. 2015; 56:348–355.View ArticlePubMedPubMed CentralGoogle Scholar
- Eppig JT, Blake JA, Bult CJ, Kadin JA, Richardson JE, Group MGD, et al.The Mouse Genome Database (MGD): facilitating mouse as a model for human biology and disease. Nucleic Acids Res. 2015; 43(D1):D726—D736.View ArticlePubMedGoogle Scholar
- Hoehndorf R, Schofield PN, Gkoutos GV. PhenomeNET: a whole-phenome approach to disease gene discovery. Nucleic Acids Res. 2011; 39(18):e119—e119.View ArticleGoogle Scholar
- Hoehndorf R, Hiebert T, Hardy NW, Schofield PN, Gkoutos GV, Dumontier M. Mouse model phenotypes provide information about human drug targets. Bioinformatics. 2014; 30(5):719–725.View ArticleGoogle Scholar
- Chen Y, Ren X, Zhang GQ, Xu R. Ontology-guided organ detection to retrieve web images of disease manifestation: towards the construction of a consumer-based health image library. J Am Med Inform Assoc. 2013; 20(6):1076–1081.View ArticlePubMedPubMed CentralGoogle Scholar
- Robinson PN, Köhler S, Bauer S, Seelow D, Horn D, Mundlos S. The Human Phenotype Ontology: a tool for annotating and analyzing human hereditary disease. Am J Human Genet. 2008; 83(5):610–615.View ArticleGoogle Scholar
- Xu R, Wang Q. Large-scale combining signals from both biomedical literature and the FDA Adverse Event Reporting System (FAERS) to improve post-marketing drug safety signal detection. BMC bioinformatics. 2014; 15(1):17.View ArticlePubMedPubMed CentralGoogle Scholar
- Xu R, Wang Q. Large-scale extraction of accurate drug-disease treatment pairs from biomedical literature for drug repurposing. BMC bioinformatics. 2013; 14(1):181.View ArticlePubMedPubMed CentralGoogle Scholar
- Hoehndorf R, Oellrich A, Rebholz-Schuhmann D, Schofield PN, Gkoutos GV. Linking pharmgkb to phenotype studies and animal models of disease for drug repurposing. In: Pac Symp Biocomput. Pacific Symposium on Biocomputing: 2012. p. 388–399.Google Scholar
- Smith CL, Eppig JT. The mammalian phenotype ontology: enabling robust annotation and comparative analysis. Wiley Interdiscip Rev Syst Biol Med. 2009; 1(3):390–399.View ArticlePubMedPubMed CentralGoogle Scholar
- Watkins S, Robel S, Kimbrough IF, Robert SM, Ellis-Davies G, Sontheimer H. Disruption of astrocyte–vascular coupling and the blood–brain barrier by invading glioma cells. Nature communications. 2014:5. doi:10.1038/ncomms5196.
- Wei J, Barr J, Kong LY, Wang Y, Wu A, Sharma AK, et al.Glioblastoma cancer-initiating cells inhibit T-cell proliferation and effector responses by the signal transducers and activators of transcription 3 pathway. Mol Cancer Ther. 2010; 9(1):67–78.View ArticlePubMedPubMed CentralGoogle Scholar
- Wu A, Wei J, Kong LY, Wang Y, Priebe W, Qiao W, et al.Glioma cancer stem cells induce immunosuppressive macrophages/microglia. Neuro-oncology. 2010; 12(11):1113–1125.View ArticlePubMedPubMed CentralGoogle Scholar
- Dotti G, Gottschalk S, Savoldo B, Brenner MK. Design and development of therapies using chimeric antigen receptor-expressing T cells. Immunological reviews. 2014; 257(1):107–126.View ArticlePubMedGoogle Scholar
- Pyonteck SM, Akkari L, Schuhmacher AJ, Bowman RL, Sevenich L, Quail DF, et al.CSF-1R inhibition alters macrophage polarization and blocks glioma progression. Nature medicine. 2013; 19(10):1264–1272.View ArticlePubMedPubMed CentralGoogle Scholar
- Morosetti R, Servidei T, Mirabella M, Rutella S, Mangiola A, Maira G, et al.The PPAR γ ligands PGJ2 and rosiglitazone show a differential ability to inhibit proliferation and to induce apoptosis and differentiation of human glioblastoma cell lines. Int J Oncol. 2004; 25(2):493–502.Google Scholar
- Vlachostergios PJ, Hatzidaki E, Befani CD, Liakos P, Papandreou CN. Bortezomib overcomes MGMT-related resistance of glioblastoma cell lines to temozolomide in a schedule-dependent manner. Investig New Drugs. 2013; 31(5):1169–1181.View ArticleGoogle Scholar
- Altiok N, Ersoz M, Koyuturk M. Estradiol induces JNK-dependent apoptosis in glioblastoma cells. Oncology letters. 2011; 2(6):1281–1285.PubMedPubMed CentralGoogle Scholar
- Jiang P, Mukthavavam R, Chao Y, Bharati IS, Fogal V, Pastorino S, et al. Novel anti-glioblastoma agents and therapeutic combinations identified from a collection of FDA approved drugs. Journal of translational medicine. 2014; 12(1):1.View ArticleGoogle Scholar
- Turcan S, Fabius AW, Borodovsky A, Pedraza A, Brennan C, Huse J, et al.Efficient induction of differentiation and growth inhibition in IDH1 mutant glioma cells by the DNMT Inhibitor Decitabine. Oncotarget. 2013; 4(10):1729–1736.View ArticlePubMedPubMed CentralGoogle Scholar
- Suenderhauf C, Hammann F, Huwyler J. Computational prediction of blood-brain barrier permeability using decision tree induction. Molecules. 2012; 17(9):10429–10445.View ArticlePubMedGoogle Scholar
- Pardridge WM. Drug transport across the blood–brain barrier. J Cereb Blood Flow Metab. 2012; 32(11):1959–1972.View ArticlePubMedPubMed CentralGoogle Scholar
- Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, et al.The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012; 483(7391):603–607.View ArticlePubMedPubMed CentralGoogle Scholar