High-throughput approaches to uncover synergistic drug combinations in leukemia.

We report a comprehensive drug synergy study in acute myeloid leukemia (AML). In this work, we investigate a panel of cell lines spanning both MLL-rearranged and non-rearranged subtypes. The work comprises a resource for the community, with many synergistic drug combinations that could not have been predicted a priori, and open source code for automation and analyses. We base our definitions of drug synergy on the Chou-Talalay method, which is useful for visualizations of synergy experiments in isobolograms, and median-effects plots, among other representations. Our key findings include drug synergies affecting the chromatin state, specifically in the context of regulation of the modification state of histone H3 lysine-27. We report open source high throughput methodology such that multidimensional drug screening can be accomplished with equipment that is accessible to most laboratories. This study will enable preclinical investigation of new drug combinations in a lethal blood cancer, with data analysis and automation workflows freely available to the community.

Transforming growth factor-beta can suppress tumorigenesis through effects on the putative cancer stem or early progenitor cell and committed progeny in a breast cancer xenograft model.

The transforming growth factor-beta (TGF-beta) pathway has tumor-suppressor activity in many epithelial tissues. Because TGF-beta is a potent inhibitor of epithelial cell proliferation, it has been widely assumed that this property underlies the tumor-suppressor effect. Here, we have used a xenograft model of breast cancer to show that endogenous TGF-beta has the potential to suppress tumorigenesis through a novel mechanism, involving effects at two distinct levels in the hierarchy of cellular progeny that make up the epithelial component of the tumor. First, TGF-beta reduces the size of the putative cancer stem or early progenitor cell population, and second it promotes differentiation of a more committed, but highly proliferative, progenitor cell population to an intrinsically less proliferative state. We further show that reduced expression of the type II TGF-beta receptor correlates with loss of luminal differentiation in a clinical breast cancer cohort, suggesting that this mechanism may be clinically relevant. At a molecular level, the induction of differentiation by TGF-beta involves down-regulation of Id1, and forced overexpression of Id1 can promote tumorigenesis despite persistence of the antiproliferative effect of TGF-beta. These data suggest new roles for the TGF-beta pathway in regulating tumor cell dynamics that are independent of direct effects on proliferation.

Identification of an integrated SV40 T/t-antigen cancer signature in aggressive human breast, prostate, and lung carcinomas with poor prognosis.

Understanding the genetic architecture of cancer pathways that distinguishes subsets of human cancer is critical to developing new therapies that better target tumors based on their molecular expression profiles. In this study, we identify an integrated gene signature from multiple transgenic models of epithelial cancers intrinsic to the functions of the Simian virus 40 T/t-antigens that is associated with the biological behavior and prognosis for several human epithelial tumors. This genetic signature, composed primarily of genes regulating cell replication, proliferation, DNA repair, and apoptosis, is not a general cancer signature. Rather, it is uniquely activated primarily in tumors with aberrant p53, Rb, or BRCA1 expression but not in tumors initiated through the overexpression of myc, ras, her2/neu, or polyoma middle T oncogenes. Importantly, human breast, lung, and prostate tumors expressing this set of genes represent subsets of tumors with the most aggressive phenotype and with poor prognosis. The T/t-antigen signature is highly predictive of human breast cancer prognosis. Because this class of epithelial tumors is generally intractable to currently existing standard therapies, this genetic signature identifies potential targets for novel therapies directed against these lethal forms of cancer. Because these genetic targets have been discovered using mammary, prostate, and lung T/t-antigen mouse cancer models, these models are rationale candidates for use in preclinical testing of therapies focused on these biologically important targets.

Chemokine (C-C motif) ligand 2 mediates the prometastatic effect of dysadherin in human breast cancer cells.

Dysadherin, a cancer-associated membrane glycoprotein, down-regulates E-cadherin and promotes cancer metastasis. This study examined the role of dysadherin in breast cancer progression. Expression of dysadherin was found to be highest in breast cancer cell lines and tumors that lacked the estrogen receptor (ER). Knockdown of dysadherin caused increased association of E-cadherin with the actin cytoskeleton in breast cancer cell lines that expressed E-cadherin. However, knockdown of dysadherin could still suppress cell invasiveness in cells that had no functional E-cadherin, suggesting the existence of a novel mechanism of action. Global gene expression analysis identified chemokine (C-C motif) ligand 2 (CCL2) as the transcript most affected by dysadherin knockdown in MDA-MB-231 cells, and dysadherin was shown to regulate CCL2 expression in part through activation of the nuclear factor-kappaB pathway. The ability of dysadherin to promote tumor cell invasion in vitro was dependent on the establishment of a CCL2 autocrine loop, and CCL2 secreted by dysadherin-positive tumor cells also promoted endothelial cell migration in a paracrine fashion. Finally, experimental suppression of CCL2 in MDA-MB-231 cells reduced their ability to metastasize in vivo. This study shows that dysadherin has prometastatic effects that are independent of E-cadherin expression and that CCL2 could play an important role in mediating the prometastatic effect of dysadherin in ER-negative breast cancer.

Bone sialoprotein mediates the tumor cell-targeted prometastatic activity of transforming growth factor beta in a mouse model of breast cancer.

Transforming growth factor betas (TGF-beta) play a dual role in carcinogenesis, functioning as tumor suppressors early in the process, and then switching to act as prometastatic factors in late-stage disease. We have previously shown that high molecular weight TGF-beta antagonists can suppress metastasis without the predicted toxicities. To address the underlying mechanisms, we have used the 4T1 syngeneic mouse model of metastatic breast cancer. Treatment of mice with a monoclonal anti-TGF-beta antibody (1D11) significantly suppressed metastasis of 4T1 cells to the lungs. When metastatic 4T1 cells were recovered from lungs of 1D11-treated and control mice, the most differentially expressed gene was found to be bone sialoprotein (Bsp). Immunostaining confirmed the loss of Bsp protein in 1D11-treated lung metastases, and TGF-beta was shown to regulate and correlate with Bsp expression in vitro. Functionally, knockdown of Bsp in 4T1 cells reduced the ability of TGF-beta to induce local collagen degradation and invasion in vitro, and treatment with recombinant Bsp protected 4T1 cells from complement-mediated lysis. Finally, suppression of Bsp in 4T1 cells reduced metastasis in vivo. We conclude that Bsp is a plausible mediator of at least some of the tumor cell-targeted prometastatic activity of TGF-beta in this model and that Bsp expression in metastases can be successfully suppressed by systemic treatment with anti-TGF-beta antibodies.

Comparing genetically engineered mouse mammary cancer models with human breast cancer by expression profiling.

Breast cancer is a heterogeneous disease, and much of the molecular basis for this heterogeneity is being unraveled using advanced genomic technologies. More recently, global transcriptional profiling has proven to be an effective new tool for characterizing human tumors. Genomic "signatures'' have been developed that classify tumors with varying prognoses and responses to treatment. Recent studies have begun to extend the use of global transcriptional profiling to better characterize genetically engineered mouse (GEM) models of breast cancer, which will improve the ability to translate basic research advances into clinical advances. GEM models of mammary carcinoma have proven to be invaluable tools to gain insight into mechanisms underlying tumor initiation, progression, and therapeutic responses in an in vivo system where tumors spontaneously develop in an appropriate tissue environment. This review will discuss the use of transcriptional profiling of breast cancer in tumors from both human patients and GEM models to improve prognostic measures, examine mechanisms of tumor initiation and progression, identify novel therapeutic targets, and improve pre-clinical testing for drug development. Together, these advances offer a framework for classifying human tumors, identifying appropriate GEM models for specific experimental purposes, and utilizing the combined data to identify more specific and effective cancer therapies.

Gene expression profiling identifies a unique androgen-mediated inflammatory/immune signature and a PTEN (phosphatase and tensin homolog deleted on chromosome 10)-mediated apoptotic response specific to the rat ventral prostate.

Understanding androgen regulation of gene expression is critical for deciphering mechanisms responsible for the transition from androgen-responsive (AR) to androgen-independent (AI) prostate cancer (PCa). To identify genes differentially regulated by androgens in each prostate lobe, the rat castration model was used. Microarray analysis was performed to compare dorsolateral (DLP) and ventral prostate (VP) samples from sham-castrated, castrated, and testosterone-replenished castrated rats. Our data demonstrate that, after castration, the VP and the DLP differed in the number of genes with altered expression (1496 in VP vs. 256 in DLP) and the nature of pathways modulated. Gene signatures related to apoptosis and immune response specific to the ventral prostate were identified. Microarray and RT-PCR analyses demonstrated the androgen repression of IGF binding protein-3 and -5, CCAAT-enhancer binding protein-delta, and phosphatase and tensin homolog deleted on chromosome 10 (PTEN) genes, previously implicated in apoptosis. We show that PTEN protein was increased only in the luminal epithelial cells of the VP, suggesting that it may be a key mediator of VP apoptosis in the absence of androgens. The castration-induced immune/inflammatory gene cluster observed specifically in the VP included IL-15 and IL-18. Immunostaining of the VP, but not the DLP, showed an influx of T cells, macrophages, and mast cells, suggesting that these cells may be the source of the immune signature genes. Interestingly, IL-18 was localized mainly to the basal epithelial cells and the infiltrating macrophages in the regressing VP, whereas IL-15 was induced in the luminal epithelium. The VP castration model exhibits immune cell infiltration and loss of PTEN that is often observed in progressive PCa, thereby making this model useful for further delineation of androgen-regulated gene expression with relevance to PCa.

Identification of tumor-initiating cells in a p53-null mouse model of breast cancer.

Using a syngeneic p53-null mouse mammary gland tumor model that closely mimics human breast cancer, we have identified, by limiting dilution transplantation and in vitro mammosphere assay, a Lin(-)CD29(H)CD24(H) subpopulation of tumor-initiating cells. Upon subsequent transplantation, this subpopulation generated heterogeneous tumors that displayed properties similar to the primary tumor. Analysis of biomarkers suggests the Lin(-)CD29(H)CD24(H) subpopulation may have arisen from a bipotent mammary progenitor. Differentially expressed genes in the Lin(-)CD29(H)CD24(H) mouse mammary gland tumor-initiating cell population include those involved in DNA damage response and repair, as well as genes involved in epigenetic regulation previously shown to be critical for stem cell self-renewal. These studies provide in vitro and in vivo data that support the cancer stem cell (CSC) hypothesis. Furthermore, this p53-null mouse mammary tumor model may allow us to identify new CSC markers and to test the functional importance of these markers.

Statistical issues in the design and analysis of gene expression microarray studies of animal models.

Appropriate statistical design and analysis of gene expression microarray studies is critical in order to draw valid and useful conclusions from expression profiling studies of animal models. In this paper, several aspects of study design are discussed, including the number of animals that need to be studied to ensure sufficiently powered studies, usefulness of replication and pooling, and allocation of samples to arrays. Data preprocessing methods for both cDNA dual-label spotted arrays and Affymetrix-style oligonucleotide arrays are reviewed. High-level analysis strategies are briefly discussed for each of the types of study aims, namely class comparison, class discovery, and class prediction. For class comparison, methods are discussed for identifying genes differentially expressed between classes while guarding against unacceptably high numbers of false positive findings. Various clustering methods are discussed for class discovery aims. Class prediction methods are briefly reviewed, and reference is made to the importance of proper validation of predictors.

Effects of pooling mRNA in microarray class comparisons.

MOTIVATION: In microarray experiments investigators sometimes wish to pool RNA samples before labeling and hybridization due to insufficient RNA from each individual sample or to reduce the number of arrays for the purpose of saving cost. The basic assumption of pooling is that the expression of an mRNA molecule in the pool is close to the average expression from individual samples. Recently, a method for studying the effect of pooling mRNA on statistical power in detecting differentially expressed genes between classes has been proposed, but the different sources of variation arising in microarray experiments were not distinguished. Another paper recently did take different sources of variation into account, but did not address power and sample size for class comparison. In this paper, we study the implication of pooling in detecting differential gene expression taking into account different sources of variation and check the basic assumption of pooling using data from both the cDNA and Affymetrix GeneChip microarray experiments. RESULTS: We present formulas for the required number of subjects and arrays to achieve a desired power at a specified significance level. We show that due to the loss of degrees of freedom for a pooled design, a large increase in the number of subjects may be required to achieve a power comparable to that of a non-pooled design. The added expense of additional samples for the pooled design may outweigh the benefit of saving on microarray cost. The microarray data from both platforms show that the major assumption of pooling may not hold. SUPPLEMENTARY INFORMATION: Supplementary material referenced in the text is available at http://linus.nci.nih.gov/brb/TechReport.htm.