There are currently no targeted therapies available for the treatment of LMS. Treatment of LMS typically consists of surgery with doxorubicin-based chemotherapy and consideration for adjuvant ifosfamide and radiotherapy in selected cases. Doxorubicin-based therapy has only shown a marginal association with improved overall survival, thereby making it important to evaluate additional therapeutic molecules for the treatment of these tumors [1
Previously, we have shown that there exist three distinct subtypes of LMS, characterized by unique genomic, transcriptional, and protein expression characteristics [2
] and we hypothesized that the differences inherent to these tumor subtypes may underlie the heterogeneity in drug responses observed in LMS patients. Previously, gene expression profiles had been shown to be predictive of metastatic outcome in LMS, suggesting that evaluating the transcriptional features of these tumors may provide important insights into the biology of LMS [14
]. In the present work, we performed a comparative gene expression profiling study between samples from each of these three LMS subtypes and a set of benign leiomyomas, and we used this analysis to identify drugs predicted to turn each of the LMS subtypes from a “malignant” to a “benign” state.
Of the 11 small molecules evaluated experimentally in our current study, two drugs, Cantharidin and MG-132, were able to strongly inhibit cell viability in all three LMS cell lines tested. Cantharidin is an ancient Chinese medicine that has been demonstrated to have anticancer activity through its inhibition of protein phosphatases [10
]. Interestingly, in our functional gene set analysis, we previously found that all three LMS subtypes are highly enriched for phosphoproteins compared to the background full Homo sapiens
]. The in vitro
efficacy demonstrated in the current work may provide a starting point for a more rigorous in vitro
and in vivo
exploration of Cantharidin or other phosphatase inhibitors for the treatment of LMS.
MG-132 is a potent inhibitor of the proteasome with an ability to specifically reduce the degradation of ubiquitin-conjugated proteins in mammalian cells [13
]. Proteasome inhibitors have demonstrated clinical efficacy in several cancers, as is evidenced by bortezomib's 2003 FDA approval for the treatment of relapsed multiple myeloma and mantle cell lymphoma, and carfilzomib, a next-generation proteasome inhibitor, showing promising results in mid-stage clinical trials [15
]. Given the strong effect on cell viability observed with MG-132 in LMS cells, we evaluated whether bortezomib (which was not included in the original cmap screen) could similarly inhibit LMS cell viability and found that the drug had extremely potent antigrowth effects on the three LMS cell lines evaluated. Bortezomib had previously been investigated for the treatment of malignant soft-tissue sarcomas in a 21-patient Phase II clinical trial; while the authors concluded that bortezomib had limited activity as a single agent for the treatment of these cancers, it is interesting to note that only four LMS patients were included in this cohort, and that the single confirmed partial response observed in the study was in an LMS patient [16
]. Therefore, it may be worthwhile to further investigate the clinical potential of bortezomib for LMS treatment, either as a single agent or in combination with other molecules with demonstrated antigrowth activity in LMS.
Several reports in the literature have utilized comparative gene expression profiling studies between normal and diseased tissues to identify and validate drugs with therapeutic potential. Two such studies utilized cmap to identify potential therapeutic molecules for neuroblastoma and colorectal cancer [6
]. A similar approach was recently used to demonstrate that topiramate, an anticonvulsant used to treat epilepsy, showed therapeutic efficacy in a preclinical model of inflammatory bowel disease [16
]. While these studies highlight the potential utility of gene expression-based approaches to help identify a molecule that can be repurposed for novel therapeutic indications, little systematic analysis has been performed to evaluate the relationship between the cmap-derived enrichment scores and the actual responses observed in disease models; previous published reports have focused predominantly on documenting the positive associations discovered with little attention paid to predicted associations that were unable to be validated in follow-up experiments.
In the present work, we show that while using cmap we were able to identify two drugs that potently inhibited LMS cell growth, there was no overall statistically significant association between cmap enrichment scores and actual cell viability inhibition in vitro
in the 11 drugs that we tested. It is important to note that the primary goal of our analysis was to identify new therapeutic drugs for LMS and not to systematically evaluate cmap, and consequently the set of 11 drugs we selected was significantly enriched for drugs with highly nonzero enrichment scores. This design may have increased our ability to identify effective drugs, but gave us little statistical power to rigorously evaluate associations between drug response and cmap score, as we evaluated few drugs with enrichment scores near zero. The two drugs that showed in vitro
efficacy showed highly divergent enrichment scores, with Cantharidin showing a highly negative score while MG132 showed a strongly positive score. Although our study was not well-powered to identify significant overall associations between enrichment scores and in vitro
efficacy in LMS, the rate at which we were able to identify drugs that could inhibit cell viability using cmap (2/11 molecules evaluated, or 18.4%) was higher than similarly designed studies that screened entire chemical libraries without a priori
predictions of efficacy. For example, Rickardson and colleagues observed a 4.4% hit rate (56/1,266 molecules evaluated) for small molecules that could inhibit myeloma cell line growth, and Zhang and colleagues observed a 0.6% hit rate (16/2,816 molecules evaluated) in a screen of new therapeutic compounds that could inhibit thyroid cancer growth [17
]. While our data are far from definitive in answering the important question of cmap's utility for identifying effective therapeutic molecules, they do suggest that algorithmic approaches can be taken to increase the success rate of small molecule screens.
Further, while the gene expression profiles of the three LMS cell lines evaluated showed the strongest similarity to LMS Subtype II, it is likely still necessary to evaluate a broader spectrum of cell lines in order to ascertain whether, and to what degree, our LMS patient tumor subtyping is applicable to immortalized cultured cells. Unfortunately, there exist very few human LMS cell lines available for study, thereby limiting our ability to more rigorously investigate the relationship between subtype specificity and drug response. As more LMS clinical specimens are immortalized and made available to the research community for cell culture studies, it would be informative to ascertain their subtype specificity and to characterize their drug response profiles.
Our findings do suggest that further work is needed to systematically assess the relationship between cmap enrichment scores and drug effects in a variety of cancer cell model systems. It will be valuable for future studies of drug candidates identified using computational approaches to document both the identified candidates that are successfully validated as well as the candidates that do not show efficacy in follow-up experiments.
In conclusion, we utilized gene expression profiles to predict novel drug candidates for LMS, and we functionally tested 11 of these drugs using in vitro assays. Our study identified two drugs, Cantharidin and MG-132, that showed strong antigrowth effects in LMS cell lines and that may form the starting point for a more focused evaluation of these and similar drugs for the treatment of LMS.