Troglitazone and Prochloroperazine Repositioned As An Anti-Leukemia Stem Cell Drug Combination Using Chemical Genomics-Based Prediction

Abstract 1885

Current therapies used in the treatment of AML are highly toxic and have been shown to be ineffective at targeting AML stem cells (AML-SCs), which are chemoresistant and thus likely to provide a surviving reservoir of cells that drives AML relapse. The clinical relevance of AML-SCs in relapse is supported by studies indicating that higher AML-SC proportions at diagnosis and during minimal residual disease correlate with worse outcome. Previous studies have demonstrated that survival of AML-SCs is heavily dependent on both NFkappaB activation and redox homeostasis, which are simultaneously disrupted by the known anti-AML-SC compound, parthenolide (PTL). However, PTL represents a suboptimal drug based on its poor pharmacokinetics. Derivatives of PTL have been created and are currently in clinical trials. However, the process of taking pre-clinical findings to clinic is lengthy and success rates in this process are low. As therapies amenable to rapid clinical translation are urgently needed in AML, we sought to determine whether chemical genomics could be used to predict combinations of FDA-approved drugs that target AML-SCs.

Previous studies have used chemical genomics to find PTL-like drugs by interrogating the Connectivity Map and Gene Expression Omnibus databases for drugs that perturb the cancer transcriptome so as to best mimic the transcriptional signature of PTL. However, no single agent found thus far is FDA approved. We hypothesized that we could extend the use this information to identify pairs of FDA approved drugs that together optimally mimic the anti-AML-SC transcriptional signature. To this end, we partitioned transcriptome changes caused by individual chemical perturbations of the Connectivity Map database into gene networks and sought to identify pairs of drugs that maximally perturbed the same networks as drugs targeting AML-SCs. This strategy revealed a combination consisting of the off-market anti-diabetes drug, troglitazone, and the anti-nausea drug, prochlorperazine.

Individually, neither troglitazone nor prochlorperazine was able to target primary AML-SCs ex vivo, consistent with the inability of each drug to achieve a favorable PTL-like signature at transcriptomic level. Indeed, ex vivo exposure of primary AML-SCs to either troglitazone or prochloroperazine produced 99% and 87% mean viability, respectively, 24 hours post-treatment (N=3). In contrast, the combination of the two drugs reduced mean viability to 16% (p < 0.05; N=3). We next ascertained the effects of these drugs alone or combination against AML progenitors using methylcellulose colony forming unit (CFU) assays. Individually, troglitazone and prochlorperazine each exerted a modest effect on colony formation (87% and 77% of untreated controls, respectively; N=3). In sharp contrast, the combination of both drugs proved toxic against AML progenitor/stem cells, reducing colony formation to 25% of untreated controls when measured 3 weeks post-treatment (p < 0.05; N=3). Toxicity to normal cord blood stem cells was modest when drugs were administered alone or in combination, preserving the selectivity of PTL. Finally, we corroborated these findings in functional studies using NOD/SCID xenotransplantation models. We found that 2 of 3 primary patient specimens showed significant eradication of AML-SC function (p < 0.05, 5 mice per patient sample). Together, these findings suggest that transcriptomic effects of pre-clinical drugs can be used in chemical genomic screens to define FDA approved drug combinations that can more rapidly be translated to clinic.