In last decade, a number of high-throughput large scale cancer genomic technologies have generated very comprehensive and complex datasets (Cline et al., 2013; Robbins et al., 2013; Van Allen et al., 2013; Friedman et al., 2015; Ledford, 2015). Some of key features of each type of tumor and individual patient’s cancer including a specific pattern of DNA mutations, deletion, amplification, gene rearrangements, translocation, microsatellite instability, and epigenetic alterations have been revealed, but not fully understood (Hanahan and Weinberg, 2011; Cline et al., 2013; Robbins et al., 2013; The Cancer Genome Atlas Research et al., 2013; Van Allen et al., 2013; Lawrence et al., 2014; Forbes et al., 2015). Numerous web-based oncogenomic portals for assessment of tumor genetic profiling derived from a very large number of patients have been served as powerful tools for discovery and implementation of personalized cancer medicine (Ciriello et al., 2013; Van Allen et al., 2013; Friedman et al., 2015; Klonowska et al., 2015). The use of genomic information to identify and develop innovative therapies depends on development of statistical, mathematical and computational methods. Translational bioinformatics is an emerging field dedicated on applying informatics to find genomic alterations that can be used to development of precision medicine strategies using multi-omics datasets (Ciriello et al., 2013; Dienstmann et al., 2015; Friedman et al., 2015). There are several large-scale cancer genomics platforms available for cancer researchers querying molecular profiles and clinical drug response from experimental and clinical trial studies (Van Allen et al., 2013; Dienstmann et al., 2015). These rich data sets allow clinical applications of genomic markers derived from treatment response and/or adverse events into biologic insights and new treatments. Familiarity with bioinformatics tools is one of present challenges to explore genomics projects aiming to reveal cancers’ gene drivers and thereby planning a rational sequence of targeted cancer therapies.

Further comprehensive histological and molecular characterization of various types of cancer cells depends on systems biology strategies to measure, create models and manipulate appropriated tissue cell culture, animal models and patient cohorts (Floor et al., 2012; The Cancer Genome Atlas Research et al., 2013; Belizário et al., 2015). Initially, system-level approaches have revealed a number of genetic and epigenetic alterations that contribute to oncogenesis, progression and metastasis of cancer cells (Hanahan and Weinberg, 2011). Recent studies on tumor pathobiology in hematological and solid cancers have revealed that heterogeneity is the major cause of poor drug efficacy and response duration, within and between individual patient groups (Visvader, 2011; McGranahan and Swanton, 2015). Because of heterogeneity at cellular, molecular pathways and pathophysiological levels, cancer cells that respond to particularly therapeutic treatment can rapidly adapt and develop extrinsic and intrinsic resistance changing their driver mutation signaling pathways (Luo et al., 2009; Carragher et al., 2012; McGranahan and Swanton, 2015). The targeted inhibition of a unique protein (e.g., protein kinase) in a driver cancer pathway many times results in the activation of pre-existing genetic alterations in tumor cell clones (Nijhawan et al., 2012; Holohan et al., 2013). Thus, the dynamic of variations in cancer pharmacogenomics is much more complex and require identification of target-shifting players, such as molecules and pathways at a certain time point along with the treatment (Bozic et al., 2013; Dienstmann et al., 2015; Xia et al., 2014). Therefore, only a high-resolution and broader view of signaling systems of the cancer genome could allow us to understanding the complex mechanisms that make tumor cells to subvert single- and multi-agent therapies.

In 1946, Louis Goodman and Alfred Gilman Goodman at Yale University were the first researchers to explore a synthetic molecule called “synthetic lymphocidal chemical” in a patient (J.D) with massive lymphoma. The patient presented partial response to the treatment. Afterward, the nitrogen mustard molecules were developed as chemical alkylating agents to treat human patients by Haddow et al. (1948). Until today, the screening of chemicals and natural products derived from microbial and plant species have been a preferential route to discover new candidates for cancer therapy (Neidle and Thurston, 2005; Carragher et al., 2012). With advances in the protein purification and crystallography methodologies, the so-called in silico approach for predicting and designing ligands to target structure using large virtual libraries have also been used (Neidle and Thurston, 2005). Research Collaboratory Structural Bioinformatics Protein Data Bank (RCSB PDB database¹) and PDBbind database² are examples of repositories of proteins structures, nucleic acids, and complex assemblies for in silico experiments (Neidle and Thurston, 2005). The open-source target validation programs for large scale protein kinase inhibitor screening have been a fruitful way for sharing knowledge and reagents to understanding protein kinases signaling pathways and drug discovery (Edwards et al., 2015; Campbell et al., 2016; Elkins et al., 2016). The development of highly selective ATP-competitive inhibitors for intracellular and membrane tyrosine and serine-threonine kinases (RTKs) has had a great impact in cancer therapy. For instance, Imatinib for the Abelson kinase (ABL), Lapatinib for epidermal growth factor receptor (EGFR) and ERBB2 transmembrane protein kinases, and novel inhibitors such as Vemurafenib to mutated BRAF^{V 600E} protein kinase (Elkins et al., 2016). However, the extensive redundancy of RTK-transducing pathways, cross-reactivity, toxicity and tumor resistance remain major challenges and limitations of targeting key protein kinases for cancer treatment (Fabbro et al., 2015). Currently, 33 protein kinase inhibitors approved by Food and Drug Administration (FDA) are available for patient clinical treatment (Fabbro et al., 2015).

In vitro and in vivo assays based on cancer cell lines, or tumor xenograft in immunodeficient mice, are often used to high-throughput screening and to assess the cytotoxic effects of small molecules (Sharma et al., 2010). Various morphological and biochemical methods are usually used as read-out for therapeutic efficacy of compounds targeting cell death modulators of apoptosis, necroptosis, pyroptosis, ferroptosis, and autophagy in cancer cell lines (Wolpaw et al., 2011; Dixon et al., 2012; Wolpaw and Stockwell, 2014). More recently, new in vivo mouse model named Patient-Derived Xenografts model (PDX), or avatar, has been developed to validate the activity and efficacy of novel agents for systemic therapies (Hidalgo et al., 2014). NOD/scid/IL2rg^null (NSG) mice, which lack T, B and NK cells, have been used to generate collections of human tumor PDX models (Hidalgo et al., 2014; Shultz et al., 2014). PDXs form well-organized three-dimensional architecture and display phenotypic and functional heterogeneity of tumors as well as cancer stem cells (CSC) repopulation (Chaffer et al., 2011; Gupta et al., 2011; Williams et al., 2013). This system also provides immune cells and non-malignant components of stromal microenvironment that have direct implications in CSC’s escape from immune surveillance (Chaffer et al., 2011; Gupta et al., 2011; Östman and Pietras, 2013; Zitvogel et al., 2013) as well as metastasis (Chaffer and Weinberg, 2011). There is convincing data supporting can stem cell origin and their possible de novo rise from non-CSCs in vitro and in vivo (Chaffer et al., 2011; Gupta et al., 2011). Conditionally reprogramming of epithelial cells (CRC) induced by a Rho kinase inhibitor (Y-27632), in combination with fibroblast feeder cells, is a cell culture technique that enable the growth and establishment of rare cancer cell lines, including CSCs, from animal and patient’s small biopsy (Liu et al., 2012; Gach et al., 2013). These new systems will greatly accelerate the identification of new biomarkers for cancer progression and small molecules candidates to cancer therapy.

Considerable effort has been put in system pharmacology approaches to discovery of new anticancer-drug lead compounds. It has been facilitated by the creation of small molecular profiles and chemical-structure databases and a complete annotation of chemical-genetic profiles based on well-genomic characterized cell lines. Weinstein et al. (1997) at the U.S. National Cancer Institute (NCI) developed the first high-throughput assay-based on 60 cancer cell lines, named NCI-60, for screening cancer candidate drugs. The access to the NCI-60 database is via the CellMiner web-based application³ (Reinhold et al., 2012, 2015). Various other researchers’ groups developed similar pharmacologic and biochemical approaches to screening compounds in large number of cell lines derived from various types of cancers. The results of these large pharmacogenomics studies are compiled in the following platforms: the Cancer Cell Line Encyclopedia (CCLE⁴; Barretina et al., 2012), Connectivity Map (CMAP⁵; Lamb et al., 2006), Genomics of Drug Sensitivity in Cancer (GDSC⁶; Garnett et al., 2012) and the Cancer Target Discovery and Development Project⁷ Advances in high-throughput technologies have allowed improvements and expansions in these data sets as described in details elsewhere (Dan et al., 2002; Basu et al., 2013; Yang et al., 2013; Covell, 2015; Klijn et al., 2015; Iorio et al., 2016). Together, these website platforms display genomic and pharmacogenomics datasets of over 1,000 cancers cell lines and their responses to more than 25,000 therapeutic agents.

This review will consist of an introduction of pharmaco genomics datasets for exploring potential inter-relationships among cancer genomics and drug discovery for precision cancer medicine. First, we will introduce cBio Cancer Genomics Portal (cBioPortal) and discuss the key molecular and clinical features using some examples of cancer genomics datasets. Second, we will exemplify how to query cancer candidate genes and small molecules and chemical with advanced web bioinformatics profiling tools available in CellMiner and Connectivity Map platforms. We illustrate in each section examples of relevant studies showing potential inter- relationships among cancer genomics and drug discovery. Finally, we discuss on challenges, limitations and new directions to personalized gene-targeted cancer therapy.

cBioPortal⁸ provides bioinformatics tools for gene-based visualization and analysis of molecular profiles and clinical attributes of cancer patients obtained in the large-scale clinical studies (Cerami et al., 2012; Gao et al., 2013; Schroeder et al., 2013). The portal was developed at Memorial Sloan-Kettering Cancer Center (MSKCC) Computational Biology Center (cBio) in partnership with The Cancer Genome Atlas (TCGA) and the International Cancer Genomics Consortium (ICGC). Currently users can access data from more than 10,000 tumor samples of 126 studies (as July 2016) and many provisional cancer studies are continuously updated. The data set includes DNA copy number variation, DNA methylation values, mRNA and microRNA expression based on microarrays, mutation profiles, protein and phosphoprotein levels when available for that study. The clinical data available include overall survival and disease-free survival intervals, gender, age, stage, and tumor grade. Recently, the portal has added digital images of each patient tumor biopsy displaying their histological and proliferative patterns. The cBioPortal provides Pathway Commons tools for exploring biological networks and interactions with up to 50 of the most highly altered neighboring genes in the selected cancer study. Biological networks can be viewed using cytoscape⁹ tolls for visualizing molecular interaction networks and integrating these interactions with gene expression profiles of interest. Furthermore, using the Genes and Drugs menu, users can predict drug-target interactions in a network.

The cBioPortal platform presents a list of over 70 scientific articles reporting on the growing number of clinical studies available in portal database for interactive data analyses (Cerami et al., 2012; Gao et al., 2013). CBioPortal has been cited in hundreds studies to describing mutational patterns of most common oncogene and tumor suppressor, insertion, deletion, and amplification that drive specific tumor-type, and prognosis in patients and entire cohort (Van Allen et al., 2013; Lawrence et al., 2014; Forbes et al., 2015).

The Connectivity Map¹² uses the concept of connectivity based on compound-gene signature developed by Todd Golub’s group at Broad Institute of MIT, Whitehead Institute and Harvard Medical School, Massachusetts (Lamb et al., 2006; Lamb, 2007). In their first study, the group examined the effects of exposure cancer cell line MCF-7, HL-60, SKMEL5 and PC3 to 164 perturbagens on gene expression using Affymetrix GeneChip microarrays. Perturbagens are a small molecule or a genetic interference such as knockdown or overexpression of a gene, using a genetic reagent such as shRNA (short hairpin RNA) and CRISPR/cas9 system. The instance is the basic unit of data obtained in one treatment, the source, the concentration, the cell line used, and the scan numbers for the treatment and its control (Lamb, 2007). For comparing two samples and to determine cumulative probability function by the null hypothesis, CMAP uses Kolmogorov–Smirnov test. CMAP database contains the results obtained with 1309 compounds and more than 7,000 core reference expression profiles. To date, the Broad Institute’s LINCS program has expanded the compound collection to over 20,000 using 50 types of cells, which gave over 1,800,000 perturbation profiles¹³ Differentially expressed genes between disease and normal conditions were used to form a signature for the disease. Some of these experiments were done using a hybrid capture sequencing method, which examine mRNA expression levels of 1,000 landmarker genes.

These genes are minimally redundant and widely expressed in different cellular contexts. The CMAP 2.0 software identifies chemicals with similarity in gene profiling among the matched genes or query gene signatures that were previously identified as common gene-expression changes to one or more known compounds of the CMAP database. The software reveals both compounds with positive and negative connectivity using the up-regulated query genes and down-regulated query genes representing a biological process. As a result of the comparisons, all drug profiles in the reference database will be given a connectivity score range from 1 to -1 representing their connections to the query signature. The method also allows the exploration of gene signatures of cellular states, development, and disease. Thus, CMAP can predict the regulatory networks and molecular interactions that take place in different types of cells under various conditions (Qu and Rajpal, 2012).

Yu et al. (2015) use CMAP to find key apoptosis genes induced after exposure to 191 anticancer drugs. The authors identified BCL2L11 (also named as BIM) and TNFAIP3 (also named zinc finger protein A20) among 13 top critical regulators of anticancer agents induced cell death. Cancer cells resistant to many cytotoxic compounds have been linked to widespread occurrence of deleted or down-regulated gene of the extrinsic signaling pathways, such as TNF family members and up-regulation or amplification of BCL2 protein family members of the intrinsic mitochondria pathways. In fact, a recent study using CMAP showed that LY294002, a PI3K inhibitor, and gossypol and AT-101, MCL1 or other anti-apoptotic BCL-2 family member inhibitors, were capable of reversing the prednisolone-resistance of MLL-rearranged Acute Lymphoblastic Leukemia (ALL) in infants (Spijkers-Hagelstein et al., 2014). Finally, CMAP database has been used to drug repositioning which is a rational screening to identify and compare drug efficacy and side effects of existing drugs to specific diseases in concert with gene signature (Hu and Agarwal, 2009; Qu and Rajpal, 2012).

Over the past 15 years, technological and bioinformatics advances have made possible the integration of multiple omics datasets and application of meta-analysis bioinformatics and systems biology approaches to analyze complex biological networks of cancer genomic datasets. As results, hundreds of novel oncogenes, tumor suppressor genes, DNA repair genes, focal adhesion, integrin signaling, extra-cellular matrix, actin/cytoskeleton genes have been retrieved from these cancer genome databases (Cline et al., 2013; The Cancer Genome Atlas Research et al., 2013; Lawrence et al., 2014; Forbes et al., 2015). The molecular, clinical and epidemiological features of over 126 large scale cancer types have been studied (as July 2016) and they are available for research community in cBioPortal platform. The well-known oncogenes and tumor suppressors TP53, EGFR, PIK3CA, PTEN, APC, KRAS, and BRAF have broadly used to classify tumor pathologies and tumor subtypes (Lawrence et al., 2014). This has been the basis for development of global new patient-centric clinical targeted therapy and precision oncology in development nowadays (Biankin et al., 2015). A large number of molecular aberrations and multiple recurrent chromosomal gains and losses identified in these oncogenomic databases remain unknown and represent important open question for investigation (Cline et al., 2013; Lawrence et al., 2014).

The tumor’s molecular signature varies widely in several human malignancies and spectrum of gene mutations that may have potential as predictive biomarkers continues to grow (Chang et al., 2016). From a clinical point of view, a major challenge in the interpretation of mutation profiles is to assess genomic variants that impact on drug treatment and clonal evolution. Recently, COSMIC Web portal updates on 2 million mutated variants in over 1 million tumor samples in the cancer genome examined (Forbes et al., 2015). We know that a very small number of rare mutations substantially contribute to oncogenesis and cancer progression. How many mutations could be made druggable? We do not know. What we know is that pathway-targeted therapies for many diverse variants converge on similar deregulated sub-pathways. For instance, cutaneous skin melanoma cells bearing BRAF^{V 600E} mutation that respond positively to Vemurafenib, the first BRAF kinase inhibitor approved to treat metastatic melanoma, progress to resistant cells (Bollag et al., 2012; Sullivan and Flaherty, 2013). The resistant cells display activation of parallel signaling pathways, such as ERK signaling, through KRAS and BRAF^{V 600E} amplification and mutations in the MAP2K1 and MAP2K2 genes (Solit and Rosen, 2014). Clonogenic growth of resistance BRAF-mutant melanoma cells treated with a combinatory drug therapy using BRAF and MEK inhibitors activate parallel P13K/AKT pathway (McCubrey et al., 2012; Solit and Rosen, 2014). There are many other similarities in clonogenic resistance mechanisms induced by receptor tyrosine kinase inhibitors in different types of tumors and cell lines (McCubrey et al., 2012; Wilson et al., 2012; Obenauf et al., 2015; Campbell et al., 2016; Kirouac et al., 2016). It is interesting that not only genetic mutation and amplification of these driver oncogenes, but also changes in cellular environment increase dependency in Ras/Raf/MEK/ERK and PI3K/PTEN/Akt/mTOR cascades (Campbell et al., 2016; Kirouac et al., 2016). How many sub-pathways are there? It is not yet imaginable. These results drive up some concerns on the promises of predictive genetics of drug sensitivity in the current personalized cancer medicine (Rubio-Perez et al., 2015).

The CellMiner and Connectivity Map are two examples of the multi-dimensionality cancer genomic platforms for predicting and developing potential targeted strategies to cancer cells. Using a variety of well-established cancer cell lines and in vitro and in vivo experiments and systems pharmacology analysis, several molecular mechanisms, pathways and cellular processes directly affecting drug response to a larger number of anticancer compounds have been revealed (Yang et al., 2010; Sharma et al., 2010; Heiser et al., 2012; Covell, 2015; Goodspeed et al., 2015; Reinhold et al., 2015). To further evaluate the suitability of any particular cell line as a model, CCLE and CDSC consortiums’ groups have undertaken deeply Integrative genomic and transcriptomic analyses of more than 1000 cell lines (Barretina et al., 2012; Klijn et al., 2015; Stransky et al., 2015; Iorio et al., 2016). From their reports it was clear that most cell lines display stable genomes and conserve similarly to molecular subtypes seen in patient’s tumor (Barretina et al., 2012; Klijn et al., 2015; Stransky et al., 2015; Iorio et al., 2016). Nonetheless, studies of RNA-Seq gene expression profiles demonstrated the presence of over 2,200 gene fusions in 232 cell lines. Among them, 168 are well known canonical oncogenic fusions, and more than 1,400 are new fusions involving a wide variety of gene patterns. In addition, it was observed that many human cell lines have acquired DNA and RNA fragments from human and murine viruses. More important, these studies confirmed a co-expression of transcripts to MET, EGFR, ITGA3 (Integrin α3), EPHA2 (the ephrin-A receptor 2), and CAV2 (Caveolin 2) gene among many cell lines (Klijn et al., 2015). This suggests that these cell lines have a constant activation of PI3K/AKT or MAPK/ERB signaling pathways, as described in many studies (Campbell et al., 2016; Kirouac et al., 2016). These popular cancer cell line models are of fundamental importance and reliable tools for implantation of systems pharmacology approaches (Keith et al., 2005; Yang et al., 2010). Nonetheless, predictive models for the complex combination of genetic alterations and biochemical pathways that translate in drug sensitivity and resistance in real solid tumors remain a challenge (Keith et al., 2005; Yang et al., 2010; Domcke et al., 2013; Covell, 2015).

In order to make tumor cells sensitive to drugs, we need to understand the role of programmed cell death proteins and mechanisms they use for re-wiring signaling pathways that control metabolism, cell growth, proliferation, invasion and cell-to-cell variability (Yang et al., 2010; Xia et al., 2014). Drug sensitivity varies according to the levels of concentration of paracrine and autocrine ligands, nutrients and metabolites present in the tumor microenvironment (Yang et al., 2010; Xia et al., 2014; Obenauf et al., 2015). The tumor microenvironment consists of mix population of cancer cells, stromal cells, vascular cells, and inflammatory cells growing under an acidic and low oxygen metabolic condition (Dang, 2012). Few de facto chemotherapeutic agents exist for modulating tumor metabolism (Galluzzi et al., 2013). Much remains to be learned about cancer metabolic rewiring and the impact of oncogenes and tumor suppressor genes, as examples c-Myc, Ras and p53, in the pentose phosphate pathway, redox homeostasis and mitochondrial respiration in cancer cells, and non-malignant cells in fully tumor conditions in vivo (Galluzzi et al., 2013). Translating these in vitro results to in vivo is not easy because of biochemical and genetic differences between cultured cell lines and heterogeneity within tumor cell populations.

Novel in vitro and in vivo approaches have been developed that reproduce heterogeneous behavior and genetic diversity of tumor cells that determine the evolution of resistance clones in response to chemotherapy (Garnett et al., 2012; Goodspeed et al., 2015). Conditional reprogramming (CR) induced by a Rho kinase inhibitor (Y-27632) is an emerging cell culture technology for generation of patient-derived stable cell lines and explore their genetic diversity in organoid 2D and 3D cell culture assays (Liu et al., 2012). Patient-derived tumor xenografts are potential complementary models for screening and development of anti-cancer agents in vivo (Williams et al., 2013; Hidalgo et al., 2014). The proof of concept that it is possible to successfully apply PDXs as model has been provided in many large-scale pharmacogenomics PDX studies using various cancer types (Dey et al., 2013; Hidalgo et al., 2014). The development of breast carcinoma subtypes-in-mouse PDX models has helped the identification of small chemical inhibitors to PI3K (Lehmann et al., 2014), checkpoint kinase 1 (Ma et al., 2012), aurora kinase (Romanelli et al., 2012), BCL-2 family-BH3 mimetic (Whittle et al., 2015), and many other drugs approved for treatment of breast cancers (Zardavas et al., 2013). In a recent study, Krepler et al. (2015) generated 12 PDXs of BRAF resistant melanoma patients. The authors applied genomic and proteomic methods to reveal the major signaling pathways and possible drivers of resistance in the PDXs. For instance, NRAS mutations were found in 3 PDXs, MAP2K1 (MEK1) mutations in 2, BRAF amplification in 4, and aberrant PTEN in 7 (Krepler et al., 2015). Furthermore, the authors analyzed potential combination of BRAF/MEK inhibitors Encorafenib and Binimetinib, or a triple combination of both inhibitors plus pan-PI3K inhibitor BKM120. They observed that only triple combination exceptionally abrogated tumor growth in PDX models (Krepler et al., 2015). Thus, this co-clinical model can be used to refine precision medicine approaches and identification of biomarkers for patient clinical outcome.

The CSCs, or cancer initiating cells, has emerged as target of new therapeutic modality aiming at overcome drug resistance in clinical cancer treatment (Frank et al., 2010; Egeblad, 2011; McGranahan and Swanton, 2015; Xie and Bourne, 2015). CSCs are characterized by their increased drug efflux capacity, which is mediated by ATP-driven ABC drug transporters (Dean et al., 2005). Verapamil and Cyclosporine, used to block P-glycoprotein-mediated efflux, and second generation ABC transporter inhibitors, such as PSC 833 and VX-710, have failed in cancer clinical trials (Dean et al., 2005). In addition, CSCs drug resistant have increased expression of repairing enzyme to chemical lesions made by reactive oxygen species and DNA damaging agents such as the topoisomerase inhibitors Etoposide, Adriamycin, and radiation therapy (Wang et al., 2015). In response to these agents, CSCs enter in a quiescent state, or senescence, and become insensitive to cell-cycle active chemotherapy (Campisi, 2013; Giuffrida et al., 2016). It is known that loss of PTEN induces senescence via a mechanism that requires p53. Survival factors released by cancer secretomes, and stress signal produced by damaged cells in tumor microenvironment, have key roles in cancer acquired resistance (Obenauf et al., 2015; Wang et al., 2015). Thus, to further dissection of these mechanisms we need to pursue new technical ways and use multiple target therapies to reverse senescence and kill proliferating CSCs.

Cancer stem cells exert multicellular functions in tumor tissue-specific networks and immune surveillance. More important, CSCs display differentiation-state plasticity that allow cancer cells to undergo epithelial to mesenchymal transition (EMT), which is a process that cancer cells gain migratory and invasive properties (Egeblad et al., 2010; Chaffer et al., 2011; Clevers, 2011; Gupta et al., 2011; Wang et al., 2015). The molecular profiling of CSCs are based on the expression of stem cell markers (CD34, CD44, CD133, ALDH, etc) and EMT gene drivers such as ETV5 (Ets Variant 5), NOTCH1, SNAI1 (Snail family zinc finger 1), TGFB1, among other genes (Chaffer and Weinberg, 2011; Clevers, 2011). Tumor-associated stromal cells and immune cells secrete soluble and insoluble glycoproteins in microenvironment that confer cell adhesion-mediated drug resistance (Obenauf et al., 2015; Wang et al., 2015). Some of these proteins may display potential biomarkers for targeted therapies (Egeblad, 2011; McGranahan and Swanton, 2015; Obenauf et al., 2015). The identification of microenvironmental changes that take place during tumor regression and resistance could be used for the design of more effective cancer treatment strategies (Pietras and Ostman, 2010; Almendro et al., 2014). Recent studies have used integrated proteogenomic approaches on tumor fragments to investigate intratumor heterogeneity and changes during chemotherapy in distinct cancer subtypes that favor cancer resistance and tumor evolution (Almendro et al., 2014; Mertins et al., 2016). Given the success of orthotopic implantation of human tumors in humanized mice models (Hidalgo et al., 2014; Shultz et al., 2014), we envisioned that future PDX studies will help us to understanding how conventional drugs drive selection of CSC clones within cancer population at different human-mouse tissue microenvironments. A significant promise is the use of multidimensional profiling of individual patient-CSC lines in in vivo models. Culture system of CTCs (circulating tumor cells) from individual patients is now being applied to test the drug susceptibility to diverse treatment regimens (Clevers, 2011; Almendro et al., 2014). In this context, several biological agents and small molecules targeting distinct components that control self-renewal and differentiation of CSCs, such as Notch, Hedgehog and WNT signaling pathway, are undergoing clinical trials (Zardavas et al., 2013). In this direction, K⁺ ionophore antibiotic salinomycin is one promising candidate under validation to CSC targeted therapy to triple-negative breast cancer patients (Gupta et al., 2009).

The validation of relevant compounds in clinical setting takes long time and typically results in enormous costs and failure rates. Co-clinical studies using matched cancer cell lines and PDX models are likely to have routine and successful applications that could bypass barriers to next generation cancer targeted gene therapies. Finally, oncogenomic portals and cancer informatics tools are ideal approaches for querying and retrieval of cancer’s gene biomarkers coupled with drug sensitivity and resistance. We anticipate that working in collaborative spirit, lab scientists, and clinical oncologists will ultimately find, design and validate novel and effective therapeutic strategies to improve personalized cancer medicine.

JB selected and reviewed the literature articles and wrote the manuscript. BS, MP-S, JN, and DM contributed to design and performed the experiments. All authors read and approved the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer VB and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.