The University of Texas MD Anderson Cancer Center consistently ranks as one of the two best cancer hospitals in the world, with 20,000 employees dedicated to the mission of “making cancer history.” The center treated more than 120,000 people last year. “For me,” says Dr. Philip Lorenzi, supervisor of laboratory and research in the Department of Bioinformatics and Computational Biology at MD Anderson, “walking alongside those patients and their families in our hallways serves as a constant reminder of why the experiment I designed today cannot wait until tomorrow. The environment is highly motivating.” DDNews asked Lorenzi to share his views on the challenges he sees standing in the way of actually achieving the goal of making cancer history.
DDNews: Dr. Lorenzi, how are you and your group contributing to cancer research?
Philip Lorenzi: Our group uses a variety of technologies for drug discovery and development. For example, we’re using large-scale screens to identify biomarkers capable of predicting patient response to chemotherapy, and we’re coupling those efforts with a new strategy to develop high-specificity diagnostic tests. Because our experiences suggest that many large data sets include a sizeable amount of “junk” and/or “miss the best targets,” we place strong emphasis on technical development and data quality.
DDNews: What therapies are you involved in evaluating?
Lorenzi: We’re developing “version 2.0” of L-asparaginase—a drug that has been used for more than 40 years to treat leukemia by removing the nutrient asparagine, which is often required by cancer cells. The new version of the drug is anticipated to reduce side effects and may be useful against additional cancer types beyond leukemia. We’re also developing a companion diagnostic test for which the goal is to separate patients into responders and non-responders prior to treating them.
DDNews: Please describe your work to stop cancer from using autophagy pathways.
Lorenzi: We first got into studying autophagy because our colleagues at the U.S. National Cancer Institute published an article reporting that L-asparaginase (the drug that our laboratory studies) induces autophagy; that prompted us to learn more about it. Autophagy is a stress response that recycles damaged cell components in an effort to cleanse cells and the body. Autophagy is thought to be highly active while you’re sleeping at night as a result of the temporary nutrient deprivation that follows many hours without a meal. The story is complex, though, because autophagy has been found to both prevent and support cancer development. Which autophagy genes should be targeted therapeutically to promote an anticancer effect? Although it was rapidly becoming a popular topic in the cancer field, there was no comprehensive list of autophagy-modulating genes and the direction in which they modulate the autophagy process, so we sought to build such a resource. To do that, we downloaded published data sets on autophagy and performed a parallel “pathway analysis” that involved three different software packages.
DDNews: How does Pathway Studio fit into this picture?
Lorenzi: Pathway Studio was one of the three different software packages that we used to construct the autophagy pathway. We also used previously published data generated using another technology called RNA interference. In total, seven different data sets helped us piece together a global picture of the autophagy process. The results were disturbing. A Venn diagram of the seven independent data sets showed very little overlap.
To explain what that means, consider a fictitious game of dart with a dart board of 1,000 scoring zones, and the goal of each team is to land a dart in all 1,000 zones with, say, 20,000 darts. Each team represents a method for studying the autophagy pathway. Some of the seven teams that we tested were terrible, landing darts in only one or two scoring zones. The best method, Pathway Studio, landed darts in 570 of the 1,000 scoring zones. So one conclusion is that if we could choose only one team for studying a pathway, we should choose Pathway Studio.
DDNews: Tells us how your work addresses the issue of Big Data and contributes to organizing, mining and extracting useful data from huge data sets?
Lorenzi: Data quality is an important determinant of the ability of Big Data to provide solutions to medical challenges. Our work on the autophagy pathway suggests a terrifying conclusion on the topic of Big Data—in our dart game analogy, no single team could hit all 1,000 scoring zones despite 20,000 attempts, suggesting that the technologies used to generate Big Data may have serious limitations. There are implications for drug discovery; if you’re a pharmaceutical company trying to develop an autophagy-targeted therapy, the technology that you use to identify a drug target may miss the best target. That might contribute to drug attrition rates these days. The human genome was sequenced more than 10 years ago and was anticipated to revolutionize the way drugs are discovered and developed. Nobody would disagree that it has, but I don’t think its full potential has been realized. Our findings suggest that technical limitations could be to blame. By either combining multiple technologies and/or by making significant improvements to existing technologies, we will come closer to realizing the full potential of the human genome and to extracting useful data from huge data sets.
DDNews: Please describe your progress to date in looking at drugs and the genes that mediate sensitivity to them.
Lorenzi: Despite the apparent limitations of ‘omic technologies, our results suggest that combining multiple technologies will improve our odds of finding the best drug targets and biomarkers. Adding Pathway Studio to the workflow may significantly improve those odds. RNA interference technology, on the other hand, has major limitations. For example, we validated a gene called ATF4 to be a key mediator of sensitivity to the drug L-asparaginase. But in genome-wide RNA interference screens, ATF4 does not show up as a hit, because some of the reagents used to knock down ATF4 exhibit off-target toxicity, meaning some reagents erroneously kill the cancer cell, masking our ability to identify ATF4 as a hit. Now that we understand this phenomenon, we’re developing improved algorithms for mining the resulting data set. The best solution to the problem, however, would be to fix the problem upstream at the experimental level by developing highly specific, on-target reagents. But development of such a resource will require major investment.
DDNews: Any final thoughts?
Lorenzi: There’s currently a major emphasis on analyzing and making sense of Big Data, but our findings suggest that we need to allocate more effort to upstream technical development and optimization of the technologies used to generate the data. I would speculate that these conclusions are applicable to Big Data issues beyond the realm of medicine and cancer; regardless of which field you’re in, it’s important to understand the technical limitations involved so that you can interpret the results appropriately.
Philip Lorenzi, Ph.D., is supervisor of laboratory and research in the Department of Bioinformatics and Computational Biology at MD Anderson Cancer Center, where he oversees several projects that use high-throughput ‘omic approaches in the context of biomarker identification and drug development. He is conducting genome-wide siRNA screens; metabolomic screens; antibody screens to develop highly specific, diagnostic-ready antibodies; lead optimization using computational chemistry coupled with high-throughput mutagenesis to engineer therapeutic proteins with improved therapeutic index; and chemical library screens to identify synergistic drug combinations. He is also co-director of the Proteomics and Metabolomics Core Facility, where he oversees development of the metabolomics platform consisting of tandem mass spectrometry and high-resolution mass spectrometry systems. Since 2013, Lorenzi has authored 10 peer-reviewed publications, including two in Nature, one in Nature Genetics and one in Cancer Cell.