A Search of Public Gene Expression Microarray Data Identifies Novel Agents That Selectively Eradicate AML Stem Cells.

Cancer stem cells are thought to initiate and perpetuate disease in several malignancies, including AML. The subpopulation of leukemic stem cells (LSCs) in AML is relatively resistant to clinical agents such as cytarabine (Ara-C), and thus likely represents an important target for improving therapeutic outcome. We have discovered previously that the sesquiterpene lactone, parthenolide (PTL), effectively ablates bulk AML cells as well as LSCs while causing no appreciable toxicity to normal hematopoietic cells. The mechanism of PTL action involves concomitant inhibition of NF-kappaB-mediated survival and induction of oxidative stress. However, the clinical utility of PTL is limited by poor bioavailability, necessitating the identification of additional “PTL-like” compounds that potentially circumvent this limitation. We hypothesized that the gene expression signature resulting from treatment of primary AML with PTL could be used to search through all publicly available gene expression profiles for similar chemical or molecular genetic perturbations that recapitulate the effects of PTL. To test this hypothesis, we obtained gene expression signatures for exposure of primary AML to PTL for 6 hours on Affymetrix U133+ 2.0 microarrays. Next, we acquired ∼23,000 multi-center gene expression microarray experiments performed on Affymetrix U133-based microarrays from the NCBI Gene Expression Omnibus (GEO). The PTL signature was systematically compared to each of these expression profiles using a novel correlation-based computational approach that is useful in the absence of manual curation. Among the top 15 hits resulting from our search were 4-hydroxynonenal (HNE) and celastrol (CEL), in addition to agents previously shown to have anti-LSC properties such as MG-132 and prostaglandin J2. Importantly, we then found that both HNE and CEL shared a therapeutic and mechanistic relationship to PTL. Following 24 h drug exposure, both agents successfully eradicated bulk AML in addition to phenotypically-defined LSCs (using flow cytometric assays) or functionally-defined LSCs (using NOD/SCID mouse xenotransplant assays). No appreciable toxicity was apparent to normal hematopoietic cells. Similar effects were observed at the progenitor level using colony-forming assays where leukemic colonies were completely absent with HNE exposure and dramatically reduced with CEL exposure. However, normal erythroid and myeloid colonies successfully formed in the presence of either agent at levels comparable to untreated controls. At the mechanistic level, HNE and CEL each demonstrated the PTL-like ability to inhibit NF-kappaB as indicated by diminished nuclear localization of NF-kappaB/p65 in confocal microscopy and decreased NF-kappaB binding in gel shift assays. In addition, both agents induced stress responses indicated by the nuclear accumulation of Nrf2 and activation of its transcriptional target, heme oxygenase 1. Thus, our findings indicate the utility of Internet-based multi-center collections of high-throughput data to facilitate discovery of leukemia drugs and drug targets. This data sharing and discovery paradigm will thus accelerate translation of diverse microarray experiments into useful therapies.