Publication
Article
Author(s):
The next leap forward in cancer treatment will see physicians tailoring treatments based on patients' individualized tumor profiles.
The next leap forward in cancer treatment will see physicians tailoring treatments based on patients’ individualized tumor profiles. To do this effectively will require powerful predictive tools. One group of researchers has taken the first step and created a novel algorithm that may be the Rosetta Stone used to translate data from gene expression models and other instruments into actionable information at the bedside.
Prognostic factors for response following chemotherapy of advanced cancer hold the promise for eventually tailoring such combinations to the “chemosensitivity” profile of tumors. Such “personalized medicine” approaches have used gene expression models (GEMs) to predict therapeutic response in cancer patients. These conventionally derived models are developed from a prior knowledge of outcomes and are then validated on independent clinical cohorts. Although promising, deriving models this way is expensive, time-consuming, and available only for a limited number of cancer types and drug combinations. Furthermore, such approaches are unable to use the same biomarkers when new agents are incorporated into established combinations.
COXEN = CO-eXpressions ExtrapolatioN
COXEN, a new approach to predicting drug responses in patients
At the University of Virginia, we began addressing this limitation by developing GEMs based on in vitro drug sensitivities and microarray analyses of the commonly used “NCI-60” cancer cell line panel. The US National Cancer Institute has used this panel of 60 diverse human cancer cell lines to screen more than 150,000 chemical compounds for anticancer activity. However, not all important cancer types are included on the panel, nor are drug responses on the panel predictive of clinical efficacy in patients. We asked, therefore, whether it would be possible to extrapolate from that rich database (or analogous ones from other drug screens) to predict activity in cell types not included or, for that matter, clinical responses in patients with tumors. Further development led to the formulation of a novel algorithm we term “Co-eXpression ExtrapolatioN” (COXEN; see figure).1,2 COXEN uses expression microarray data as a “Rosetta Stone” for translating between drug activities in the NCI-60 to drug activities in any other cell panel or set of clinical tumors.
COXEN-derived GEMs were evaluated in blinded fashion as predictors of tumor response and/or patient survival in nine independent cohorts of patients with breast (N=360), bladder (N=59), and ovarian (N=143) cancer treated with multi-agent chemotherapy. Of them, 233 were from prospectively enrolled studies.
Clinical results of drug response prediction using COXEN-derived biomarkers
In all studies, GEMs effectively stratified tumor response and patient survival, and this was independent of established clinical and pathologic tumor variables. In bladder cancer patients treated with neoadjuvant MVAC (methotrexate, vinblastine, doxorubicin, cisplatin), the three-year overall survival for those with favorable GEM scores was 81%, compared with 33% for those with less favorable scores (P=0.002). GEMs for breast cancer patients treated with FAC (fluorouracil, doxorubicin, cyclophosphamide) and ovarian cancer patients treated with platinum-containing regimens also stratified patient survival (five-year overall survival 100% vs. 74% (P=0.05) and three-year overall survival 68% vs. 43% (P=0.008, respectively). Importantly, GEMs effectively stratified tumor response and patient survival, independent of established clinical and pathologic tumor variables; this clinical prediction using in vitro GEM was superior to that of conventionally derived GEMs in bladder cancer.
Applying COXEN to drug discovery
As COXEN GEMs appeared useful in predicting outcome in patients, we reasoned that COXEN could also be used to predict which agents, previously screened on the NCI-60 panel, would be effective in specific tumor types. To test this hypothesis, we used COXEN and IC50 data from 45,545 compounds originally screened on the NCI-60 panel to predict chemosensitivity of these agents on 50 bladder cancer cell lines, a histology not present on the panel. COXEN was able to detect the majority of drugs used in bladder cancer. (However, the most exciting finding was of several drugs that were predicted to be several times more effective than known agents.) One such lead, C1311, an imidazoacridinone, was evaluated on these cell lines and found to be an active agent against most bladder cancer cells. Using yeast genetics, we have now found the primary mode of action and target of C1311.3—6
Clinical potential and future
Once extensively validated models are developed, the COXEN technique will have wide application. Validated models for several different treatment regimens used on a particular type of tumor can guide an oncologist toward selection of the optimal treatment for a specific patient. Validated models can also be used to predict response for tumors of particular tumor histology to approved drugs that have not previously been used to treat that particular tumor. This approach may prove very useful for patients with rare disease types or failure on established treatment regimens, and for whom no clear guidelines exist for salvage regimens. These models can also be used to increase the likelihood that a novel drug will be found efficacious in clinical trials through the selective accrual of patients who are predicted to respond to the drug by virtue of analysis of their tumor.
Importantly, since the COXEN technique has shown some promise in drug discovery, it could thus be used in the future to prioritize drug leads: after screening newly synthesized drugs on cell-line panels, estimates can be made about the effectiveness of treatment. Another important application is the use of this technique for drug “repositioning” or “salvage,” which may offer significant new applications for agents that have already been studied in clinical trials, but whose target cancer populations may not have been optimally identified in the past.
REFERENCES:
However, much work remains to be done on the development and improvement of these genomic drug response predictor models. Most chemotherapy regimens involve drugs that are administered in combinations, and therefore future work should devote particular attention to prediction of responses to these regimens. Although COXEN can predict combination therapy using models based on single drug effectiveness, understanding the synergistic effects would likely refine combination predictions for greater effectiveness. Furthermore, the large amount of biologic data outside the world of gene expression, such as mutational analysis, can also be integrated into these prediction models to further refine and improve prediction sensitivity and specificity. The combined application of these technologies and techniques may yet realize the promise of effective and individualized cancer therapy.
Dan Theodorescu, MD, PhD, is a professor in the Department of Urology at the University of Virginia, Charlottesville.