Deep learning uncovers why synthetic cannabinoids cause harmful effects

New psychoactive substances, originally developed as potential analgesics but abandoned due to adverse side effects, may still have pharmaceutical value if researchers could nail down the causes of those side effects. A new study from the University of Illinois Urbana-Champaign used deep learning and large-scale computer simulations to identify structural differences in synthetic cannabinoid molecules that cause them to bind to human brain receptors differently from classical cannabinoids.

"The largest class of NPS are often sold as the street drugs Fubinaca, Chimica and Pinaca," said chemical and biomolecular engineering professor Diwakar Shukla. "In addition to the adverse side effects, the formulas used to produce NPS vary, making them challenging to detect in standard drug screenings."

New psychoactive substances are synthetic compounds; one class mimics the effects of classical cannabinoids. However, the study found that NPS tend to activate distinct signaling pathways in the human brain compared to classical cannabinoids. Specifically, they often trigger what's called the "beta arrestin pathway" rather than the "G protein pathway." This switch in signaling can lead to more severe psychological effects.

The study's findings are published in the journal eLife.

"New psychoactive substances bind very strongly to cannabinoid receptors in the brain and are slow to unbind, making them difficult to observe and simulate in standard laboratory or computer experiments," Shukla said. "It can take a huge amount of computer time to see these rare binding and unbinding events."

In the lab, graduate student Soumajit Dutta used a new simulation approach, the Transition-Based Reweighting Method, to estimate the thermodynamics and kinetics of slow molecular processes. The team found that TRAM can also be used to observe the rare, slow molecular processes involved in the unbinding of NPS from cannabinoid receptors - by efficiently sampling these events that would otherwise require massive computing resources.

The researchers also used the Folding@Home platform, which enables millions of volunteers worldwide to donate computing power. This approach allowed the team to run many simulations in parallel, stitching the results together and using algorithms to decide which simulations to run next. It allows for the study of very long or rare events that would be nearly impossible with a single computer or a small cluster.

Together, these methods allowed the researchers to uncover new physical insights into how NPS interact with receptors - insights that were previously out of reach due to computational limitations - pointing the way toward the design of safer cannabinoid-based drugs that could avoid harmful side effects.

By revealing the NPS signal via pathways associated with more adverse effects, researchers can now focus on designing new molecules that avoid triggering these pathways for medical use. Shukla said their findings could direct more researchers to aim for compounds that bind less tightly or unbind more readily, potentially reducing the drugs' harm.

The National Institutes of Health award R35GM-142745 and the National Science Foundation supported this research. Shukla is also affiliated with chemistry, bioengineering, the National Center for Supercomputing Applications, the Center for Digital Agriculture and the Carl R. Woese Institute for Genomic Biology.