A team of researchers at UC San Francisco has devised a new approach for early stage drug discovery that uses techniques from the world of computer vision in combination with a powerful new tool: a lineage of genetically modified cancer cells in a dish that change their appearance when treated with drugs targeting common disease pathways.
The researchers hope to employ these cells -- in combination with algorithms akin to those used by face-recognition software -- to quickly organize the hundreds of thousands of molecules in institutional compound libraries according to their biological function, greatly facilitating the search for new and better drugs for cancer and other diseases. For this reason, they dubbed the cells ORACLs (Optimized Reporter cell lines for Annotating Compound Libraries).
Co-senior authors Steven Altschuler, PhD, and Lani Wu, PhD, are professors of pharmaceutical chemistry in the UCSF School of Pharmacy and conducted the initial research in this area while at the University of Texas Southwestern Medical Center. The pair envision making drug discovery more like the field of genomics, where researchers have spent decades annotating large genomic databases as more and more functions are discovered for individual genes and regulatory elements. As a result of this community effort, researchers can now rapidly identify particular genes of interest for more targeted investigations.
Using ORACLs to functionally annotate existing compound libraries could have similar benefits, enabling researchers to quickly identify drugs that are clinically relevant for specific disease pathways, and significantly streamlining the way scientists use the shared drug discovery facilities, or cores, which are common in academic research.
“Drug screening cores are very busy – they run continuously,” Wu said. “The problem is that screening results generally cannot be reused. When you have a new biological target you want to hit with a drug, you have to go and screen the whole compound library again.”
Shape-shifting cells reveal drug mechanisms
Researchers commonly screen for effective drugs using “reporter cells,” which are carefully designed to change their behavior or appearance in response to a successful treatment. But the creation of an ORACL – a single reporter cell line capable of distinguishing between multiple common drug classes – required a different approach.
“We didn't know how to design such a versatile reporter cell based on our existing biological knowledge,” said Chien-Hsiang “Charles” Hsu, a graduate student in the Altschuler and Wu lab and co-lead author on the new study. “So we thought, why not just randomly tag proteins and screen the cells with drugs to see which line could produce the response that fit our goal?"
The researchers created an assortment of 93 cell lines, based on an existing line of lung cancer cells, by tagging randomly selected genes with a fluorescent marker. They treated cells from each of these lineages with a panel of 30 compounds belonging to six commonly used classes of cancer drugs, then analyzed images of the cells with custom algorithms. As the team had hoped, different drugs caused cells to change their shape and pattern of fluorescence in distinct ways that enabled the software to deduce which type of drug had been applied.
As anticipated, some cell lines were more informative than others: To the researchers’ delight, one cell line proved capable of distinguishing between the six drug classes with 94 percent accuracy. Some of the more subtle aspects of the cells’ characteristic responses to different drug types required the computer algorithms to detect, but others were obvious under the microscope: DNA damaging drugs caused cells to swell, while HDAC inhibitors produced a spiky appearance. MTOR inhibitors produced dim fluorescence throughout the cell, while treatment with hsp90 inhibitors resulted in fluorescent speckles within the cytoplasm.
The researchers had found their ORACL.
Oracular cells could screen compound libraries for overlooked drugs
To test their ORACL cells, the team screened through more than 10,000 small molecules of unknown function from several institutional compound libraries. Their analysis identified 106 molecules whose effects on the cells matched those of the “training” drugs that had been used to initially select the ORACL, and further experiments confirmed that at least 90 of these matches affected the same biological pathway as the training drug.
To the researchers’ surprise, ORACL cells also produced characteristic responses to a slew of additional compounds, some of which turned out to belong to drug classes not included in the original training set, including ER and Aurora Kinase inhibitors, glucocorticoid steroids, ATPase inhibitors, and dihihydrofolate reductase inhibitors, as well as a dozen other groups of molecules whose shared biological function is still not known.
“What’s amazing,” said Altschuler, “is that we were able to do one screen, one time, and fish out molecules that were in many different diverse classes at once.”
The researchers are currently scaling up their screens to enable rapid annotation of compound libraries containing hundreds of thousands molecules. They hope that ORACLs will be adopted to identify new compounds that hit biological pathways where current drugs have too many side effects or miss certain patient populations. They could find compounds that affect pathways affected by few known drugs or even drugs whose effects seem far from any known biological pathway.
“These are really early steps in drug discovery,” Altschuler said. “We hope finding more high quality compounds will make all the later steps more efficient. Currently the process takes billions of dollars over many, many years. Wouldn't it be nice to make that easier?”