New research tool from Penn Medicine scientists could lead to more effective, less toxic dosing for many existing drugs.
Scientists from the Perelman School of Medicine at the University of Pennsylvania have developed a powerful tool for detecting and characterizing those molecular rhythms – a tool that could have many new medical applications, such as more accurate dosing for existing medications.

The tool is a machine learning–type algorithm called CYCLOPS that can sift through existing data on gene activity in human tissue samples to identify genes whose activity varies with a daily rhythm. (The acronym CYCLOPS stands for “CYCLic Ordering by Periodic Structure.)

"We can take advantage of that information potentially in many ways, for example to find times when it is easier to detect cancers and other diseases, and also to improve the dosing of many existing drugs by changing the time of day they are given," said lead author Ron C. Anafi, MD, PhD, an assistant professor of Sleep Medicine.

Described in the Proceedings of the National Academy of Sciences journal, CYCLOPS at least partly overcomes what has been one of the major obstacles to studying circadian rhythms in humans.

CYCLOPS instead is meant to use the enormous amount of existing data on gene activity in different human tissues and cells – data obtained from people at biopsies and autopsies, in scientific as well as medical settings.

Such data almost never includes the time of day when tissue samples were taken. But CYCLOPS doesn’t need to know sampling times. If the dataset is large enough, it can detect any strong 24–hour pattern in the activity level of a given gene, and can then assign a likely clock time to each measurement in the dataset.

In an initial demonstration, Anafi and colleagues used CYCLOPS to analyze a dataset on gene activity levels in mouse liver cells – a dataset for which sampling times were available. The algorithm was able to put data on cycling genes into the correct clock–time sequence even though it had no access to actual sampling times.

The algorithm performed best when restricting its analysis to genes whose activity is known to cycle in most mouse tissues – and under this condition it was able to correctly order samples for all mouse tissues. Focusing on human genes that are related to strongly cycling mouse genes, CYCLOPS also was able to correctly order samples taken from human brains at autopsy. "It effectively provided an independent, accurate prediction of the time of death," Anafi said.

Next the researchers used CYCLOPS to generate new scientific data on human molecular rhythms. In a first–ever analysis of human lung and liver tissue, the algorithm revealed the strongly cyclic activity in thousands of lung–cell and liver–cell genes. These included hundreds of drug targets and disease genes.

Anafi and his colleagues applied CYCLOPS to liver cell gene activity data, and again found many genes with strong circadian rhythms. Comparing normal liver tissue samples with those from primary liver cancers, they found that about 15 percent of the normally cycling genes they identified lost their rhythmic activity in the cancerous cells – which suggests that there are times of day when cancer cells can be more readily targeted while avoiding injury to normal tissue.

One of the strongly cycling genes CYCLOPS detected in liver cells was SLC2A2, which encodes a glucose transporting protein, GLUT2. The pancreatic cancer drug streptozocin interacts with GLUT2 in a way that tends to be toxic to cells that express it – sometimes toxic enough to kill patients receiving the drug. Anafi and colleagues showed that by giving mice streptozocin at a time of day when liver GLUT2 levels are lowest, they were able to significantly reduce the drug’s toxicity.