Making Discoveries: Find the Anomaly

In science class, students are often taught to follow the well-trodden path of existing hypotheses, ongoing data collection efforts, and established model systems. Advisors, thesis committee members, and fellowship grant reviewers certainly value this approach to doing science. However, I’ve found there is another way that has greater potential for making transformative discoveries that requires seeing beneath the surface that everyone else is looking at.
At the end of the movie The Matrix, the character Neo, played by Keanu Reeves, gains the ability to see through the virtual reality simulation that almost everyone else is trapped within. As he stares down a hallway in an abandoned building at three agents of the machine that has enslaved humanity, Neo sees the floor, walls, ceiling, and the agents themselves as greenish glowing computer code, dancing and flickering in real time. He has in this moment gained the ability to see the deeper structure of his reality. With this insight, he is finally able to defeat the agents, by simply reprogramming his nemesis Mr. Smith into oblivion. While fictional, this scene illustrates a truth about the discovery process — seeing deeper patterns that others overlook is a key to making breakthroughs. Anomalies can be the lens through which to see these deeper structures.
Some of the most exciting discoveries in my lab at Columbia have come from an initial anomaly. Almost 15 years ago, one of my students and I were trying to create a means of systematically uncovering the mechanisms by which thousands of drugs kill cells. We knew that there was an available chemical inhibitor called ZVAD that would block the effects of drugs that act through a cell death process known as apoptosis. If we treated cells with an apoptosis-inducing drug, then the cells would die, but, if we simultaneously treated cells with such a drug and ZVAD, then the cells would survive.
Using this strategy, we found we could classify drugs into two groups — those for which their killing activity was blocked by ZVAD, and those for which their killing activity was not blocked by ZVAD.
As we discussed this idea further, we realized that we could extend this strategy to inhibitors other than ZVAD. We collected dozens of additional inhibitors. By classifying each lethal drug using the pattern of how cells responded in the presence of each of the death inhibitors, we created a unique signature for each lethal drug, akin to a fingerprint.
We found that this fingerprint method was a powerful way to classify drugs by their mechanism of action. For example, numerous drugs in clinical use, such as Taxol, act by disrupting the microtubule skeleton within cancer cells. Whttp://www.columbia.edu/cu/biology/StockwellLab/index/publications/Wolpaw_PNAS_2011.pdf
We tested thousands of drugs using this system and organized the resulting data into a large a tree showing different mechanisms by which drugs kill cells. The tree started with its trunk on the bottom of the page, splitting repeatedly into smaller branches as it migrated up the page. While there were many fine branches terminating on the top, we could discern three large parts of the trunk. The left trunk represented the drugs that cause necrosis, or unregulated cell death. The middle trunk represented drugs that cause apoptosis. These two trunks represented the two major forms of cell death. But there was a third trunk on the right, which seemed to be an anomaly.e found that all of these microtubule-disrupting drugs had similar signatures, and that a pesticide that was thought to act through other mechanisms actually worked by disrupting the microtubule skeleton. Here is the paper we published describing this method:
We noticed that the third trunk contained the chemical erastin. Almost nine years earlier, my lab had been trying to find a chemical that would selectively kill cancer cells containing the RAS oncogene by screening thousands of chemicals for this activity. As the results had rolled off the machine that detected the number of live cells in each miniature test tube, I had examined the patterns. Row after row of testing data spooled up the computer screen. Near the end of the analysis, I saw it: just one chemical out of the 20,000 tested that selectively killed RAS mutant cells. It was a true needle in a haystack, and we named this chemical erastin. One curious observation about erastin stuck in my mind: we had evidence that erastin was not activating apoptosis. Beyond that, I wasn’t able to say at that time how erastin killed cells.
We found that erastin was located on this third trunk of our cell death tree. Our embrace of erastin’s potential was made possible through two anomalies — erastin itself killing through a new mechanism, and seeing an unexpected third trunk on the cell death tree. These anomalies ultimately led to our 2012 Cell paper proposing ferroptosis as a new form of cell death:
Ever since, I’ve found that searching for and exploiting anomalies is a valuable way to make transformative discoveries. For anyone getting started in science, I recommend paying particular attention to the data you generate that doesn’t fit your expectations — therein may be the key to an exciting breakthrough.