Chemical genomics and combo therapy

In this issue of Blood, Hassane and colleagues demonstrate that chemical genomic approaches can successfully identify rational combinations of small molecules for human malignancy.¹  Specifically, they determine that the combination of parthenolide and inhibitors of the PI3K/mTOR pathway synergize to eradicate not only the bulk population, but also tumor-initiating AML cells.

At an ever-increasing pace, with the explosive development of genomic technologies, the molecular basis of malignancy is unfolding, and in parallel, a pharmacologic toolbox to address validated cancer targets is growing. Yet, for many malignancies, we have failed to alter outcomes for patients with these diseases, despite an improved understanding of the mechanisms of disease pathogenesis. Numerous roadblocks still exist in the translation of laboratory-based findings to effective cancer therapies. One critical challenge is the identification of rational combinations of compounds among the expanding cancer-directed pharmacopeia. A critical question is whether we can bring to bear more directly emerging genomic approaches on the drug-discovery process.

It is now more than a decade since the pivotal work demonstrating the predictive power of genome-wide expression patterns to distinguish acute lymphoblastic leukemia (ALL) from acute myeloid leukemia (AML) was published, heralding a revolution in the study of malignancy.²  What then followed was the validation of this technology for a host of other indications from refinement of disease classification, to identification of new drug targets, to prediction of response to cancer therapies. More recently, the direct application of gene-expression profiling to drug discovery has been explored in the Connectivity Map (C-Map). The C-Map is based on the notion, initially demonstrated as feasible by Hughes et al in Saccharomyces cerevisiae,³  that one can create a compendium of gene-expression profiles for small-molecule discovery.

The C-Map is a publicly available reference database of more than 6100 genome-wide expression profiles from cultured human cancer cell lines treated under standard culture conditions with more than 1300 small molecules.^4,5  This compendium is accompanied by a pattern-matching software to enable the connection of investigator-derived gene-expression query signatures with small molecules profiled in the C-Map collection. This approach has been successfully applied in a wide range of applications, such as the identification of small molecules which render glucocorticoid-resistant ALL cells sensitive to glucocorticoid treatment,⁶  and to the identification of the protein target of small molecules, such as heat shock protein 90 as a target of the natural products gedunin and celastrol in the modulation of an androgen-resistance signature in prostate cancer.⁷  In the studies by Hassane et al, the application of C-Map is taken to yet another level with the successful identification of rational combinations of small molecules in silico for AML, including the AML stem cell (see figure).

Previous work by the authors identified parthenolide as a small molecule with activity against both the bulk population of AML cells and the leukemia-initiating cell by simultaneously inhibiting NF-κB and inducing oxidative stress.⁸  The pharmacologically superior analog, dimethyl-amino-parthenolide (DMAPT), is currently in clinical trials.⁹  Successful cancer-directed therapy, however, almost always involves combinations of molecules, whether using cytotoxic chemotherapy or newer, so-called targeted therapies. For example, it has been demonstrated that inhibition of both extracellular signal-regulated kinase and phosphatidylinositol 3-kinase (PI3K)/AKT is necessary in tumors with oncogenic activation of both pathways.^10,11  Here, Hassane et al validate the hypothesis that chemical genomic approaches can identify compounds that synergize with either parthenolide or DMAPT.¹  From the genome-wide expression profiling of parthenolide, it was noted that parthenolide treatment induces a cytoprotective response in part through activation of the Nrf2 pathway. The authors reasoned that compounds in the C-Map negatively correlated with a 150-gene parthenolide signature might counteract the cytoprotective signature induction, and hence diminish the cytoprotective response induced by parthenolide. They found that in the 6100 drug instances in the C-Map, parthenolide itself, not surprisingly, was the most positively connected molecule with the 150-gene signature query. Among compounds achieving a negative connection score, there was a striking enrichment for compounds inhibiting PI3K and mammalian target of rapamycin (mTOR). The hypothesis that activation of the PI3K/mTOR pathway is a response to treatment with parthenolide was next confirmed with biochemical assays. In addition, inhibitors of the PI3K/mTOR pathway prevented the activation of Nrf2.

In an elegant series of experiments using state-of-the-art approaches to evaluating primary human AML bulk and progenitor cells, Hassane and colleagues demonstrate synergism with PI3K and mTOR inhibitors in combination with parthenolide in vitro and with DMAPT in vivo in primary human AML orthotopic models. The combination of the mTOR inhibitor temsirolimus with DMAPT impaired engraftment of AML stem cells and decreased the tumor burden of established AML in orthotopic prima-graft models. There was no evidence of toxicity or impairment of normal hematopoiesis in normal mice, suggesting a therapeutic window for this combination in treating AML. Moreover, this “negative connection” with PI3K and mTOR inhibitors was not merely a general response to cytotoxicity as it was not enriched with chemotherapy agents used to treat AML, such as daunorubicin or etoposide.

With DMAPT already in clinical trial for patients with AML, these results have immediate translational relevance in second-generation clinical trials. It will be important to explore whether actual human patients treated with DMAPT have up-regulation of PI3K/mTOR signaling in their AML cells either with initial treatment or at the time of relapse in those who have an upfront response to DMAPT treatment. Beyond the excitement of this combination of DMAPT with PI3K/mTOR inhibitors specifically for AML, the studies by Hassane et al suggest a generalizable approach for identifying rational combinations of small molecules for cancer using the currency of transcriptional profiles. Rather than the laborious trial-by-error experiment to test all possible combinations of clinically relevant molecules in combination with a lead molecule, one can now envision an in silico analysis querying publicly available datasets, such as the C-Map, to identify lead combinations for testing. With the dissection of human malignancy into increasingly granular molecular subtypes and the plethora of potential small molecules for testing as single agents, let alone in combination, improved ability to identify rationale combinations of molecules to guide more molecularly informed clinical translation is essential. The studies by Hassane et al are an important step toward that goal.