Effective Strategies for Personalized Cancer Therapy: Lessons from the 2012 US Presidential Elections
Original article published on Soapbox Science – Nature Blogs, 05 Dec 2012 here.
The 2012 election season in the US that ended a few weeks ago witnessed a never-before-seen barrage of “targeted” television advertisements, phone calls and door knocks. The use of “big data” by campaigns to effectively micro-target voter groups was particularly striking. Data companies gathered over 500 attributes from individual records including voting histories, demographics, hobbies, income etc. These data points were plugged into sophisticated algorithms on computer models to generate scores that identified undecided voters most likely to “swing”. The “persuasion scores” thus obtained drove campaign strategies to target these swing voters. Research shows that such voter targeting likely yields huge successes in terms of persuading unconvinced voters. Another layer to political campaigning was the use of transactional data to evaluate how opinions would change after interactions with campaign volunteers. Together, big data and an efficient use of technology have radically changed the nature of campaigning. It is remarkable how similar this approach is to that of personalized cancer medicine. Or at least to how an ideal personalized medicine approach should be for cancer management.
In spite of an increased spending on cancer research, the survival for cancer patients has not improved proportionately. The search continues for the elusive “magic bullet”. Cancer, with all its inter- and intratumoral heterogeneity represents a highly complicated microenvironment to target therapeutically, not unlike voter groups targeted by political campaigns. A decade ago, election campaigns consisted primarily of blanket advertising over television and radio. In retrospect, this method of campaigning seems very primitive and almost ineffective. Similarly, the blanket use of cytotoxic chemotherapy to treat cancer, though effective in some cases, seems ridiculously simplistic in today’s age of “-omics”. Not surprisingly, initial clinical studies for targeted agents using unselected patient groups failed miserably for lack of patient stratification. In this era of targeted therapeutics, identifying the “right” patient population for drug therapy is critical for treatment success. Importantly, rapid advancements in technologies for genomics and proteomics are providing us with tools to stratify patients for newer anti-cancer agents.
Initial failures encountered in clinical studies raised questions on how best to identify cancer patients likely to respond to targeted agents. This spurned a whole new field of predictive biomarker identification and development. But how many biomarkers are really necessary and sufficient to stratify cancer patients for successful personalized therapy? One? Four? Forty? Amidst the continuing attempts to develop targeted therapeutic agents came imatinib (better known as Gleevec) that proved to be a magic bullet for patients with chronic myelogenous leukemia (CML). Unfortunately, this highly successful response to Gleevec set the stage for the use of a single biomarker as a standard for directing targeted therapy. However, is it really judicious to select patients based on the absence or presence of a single biomarker? Are we effectively exploiting data obtained from genomic, proteomic, transcriptomic and metabolic profiling to predict cancer cell responses to drugs? The answer is a resounding “no”.
To use the election campaign analogy, would a strategist really consider it prudent to predict how a person is likely to vote based only on one data point, such as preference for type of car, movies, magazine subscriptions, neighborhood, hobbies, etc.? Not really! Hence the need for campaigns to collect over 500 data points for each voter. Similarly, cancer researchers can attest to the fact that the “one biomarker-one cancer” correlation is a gross over-simplification and remains an exception rather than the rule. Of course, using even one predictive biomarker to determine drug therapy is a big leap from the era of cytotoxic chemotherapy where almost every patient received only a standard cocktail of drugs. However, an ideal scenario would be to incorporate multiple data points in our decision making process for cancer management. For this purpose, the vast amounts of “-omics” data that are generated for each cancer patient need to be meaningfully interpreted and integrated in order to gain a holistic insight into cancer. Further, translation into clinical practice necessitates development of computationally modeled predictive algorithms based on bioinformatics and systems biology driven approaches.
Another challenge in cancer therapy is the prediction of cancer cell behavior following targeted therapy that rewires cellular networks. This is especially a concern for development of secondary resistance to targeted agents. To revert to the analogy of election campaigns, transactional data are used to track changes in voter opinions following interactions with campaign staff. These data, in turn, help plan future interactions with undecided voters. Similarly, a priori predictions of how cellular networks respond to treatment with targeted agents may help develop drug combinations that prevent emergence of resistance. To this end, two studies have analyzed interactions of drugs with different cellular pathways and networks. Researchers at the Broad Institute of MIT and Harvard University, Cambridge, MA have elegantly described a “connectivity map”. A similar approach was used by researchers at the Dana-Farber Cancer Institute, Boston MA and the University of Notre Dame, Notre Dame IN to generate a “drug-target network”. Analyses such as these may predict development of resistance to targeted drugs and help design effective combination therapies for cancer management.
Thus, personalizing cancer therapy calls for an inter-disciplinary approach with collaborations between disciplines such as biological sciences, physical sciences and oncology. A recent edition of Nature Outlook highlights this fact very eloquently. An ideal approach to personalized medicine for cancer management would begin with the profiling of tumor cells using genomic, proteomic, transcriptomic and metabolic analyses. Analyzing these data using systems biology-based algorithms would predict acute and long-term responses of tumors to different targeted agents administered singly or in combinations. Hopefully this approach will help design effective personalized treatment options for cancer patients at various stages of the disease.