Advertisement

If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.

ENJOY UNLIMITED ACCES TO C&EN

Pharmaceuticals

Getting Rid Of Painful Compounds

ACS Meeting News: Computational method removes artifacts from high-throughput screening assay hit lists

by Celia Henry Arnaud
September 1, 2014 | A version of this story appeared in Volume 92, Issue 35

CORRECTION: This story was updated on Sept. 25, 2014, to correct the structure of the typical triple-threat PAIN compound shown.

The more often certain compounds show up as hits in high-throughput screening assays, the less they can be trusted. They are compounds that register as hits not because they selectively bind to a target but because they nonselectively react with it or with something else in the system.

SOURCE OF PAINS
Scientist performing high-throughput assay at University of Minnesota’s Institute for Therapeutics Discovery and Development.
Credit: Institute for Therapeutics Discovery & Development
High-throughput screening frequently results in the identification of pan assay interference compounds, or PAINS.

These compounds have a name that not only sums up their behavior but also describes how medicinal chemists feel about them: PAINS (pan assay interference compounds). The real problem with these compounds is that they waste time and resources as researchers try to optimize them.

The term was coined in 2010 by Jonathan B. Baell, a professor of medicinal chemistry at Monash Institute of Pharmaceutical Sciences, in Parkville, Australia. Since then, he and others have worked to make the community aware of these “subversive compounds.” The troublemakers were a topic of conversation at a symposium sponsored by the Division of Medicinal Chemistry at the American Chemical Society national meeting held last month in San Francisco.

PAINS can react with screening targets through a number of mechanisms, including covalent labeling, photoreactivity, oxidation/reduction, or chelation, said symposium organizer Michael A. Walters, of the Institute for Therapeutics Discovery & Development at the University of Minnesota. “They can be difficult to recognize because they show up in binding assays, but the mode is actually reactive,” Walters said.

Baell originally defined PAINS on the basis of results he got from six high-throughput screening campaigns. Some compounds popped up as hits in all six screening assays, but they led the researchers down dead ends that they couldn’t optimize their way out of.

To remove such problematic compounds from screening results, Baell and Georgina A. Holloway developed a set of computational filters (J. Med. Chem. 2010, DOI: 10.1021/jm901137j). They tried to distinguish between dirty and clean classes of compounds. The researchers wanted filters to catch PAINS but not be so stringent that they eliminated actual hits.

To do that, the team looked at the worst compound they had worked with that had nonetheless gone on to form the basis of a successful medicinal chemistry program. The molecule was a lead candidate against Bcl-xL, a mitochondrial protein that is a cancer target involved in the cell death signaling pathway and had registered as a hit in four of the assays. The team then designed the filters by using substructures that hit at least five out of the six assays.

For each substructure, they calculated an enrichment value, defined as the number of analogs that hit between two and six assays divided by the number of analogs that hit zero assays, and expressed it as a percentage. Classes of PAINS have enrichment factors of at least 30%, but most of them have much higher enrichment factors. The researchers refined each substructure to maximize the enrichment value and wrote text-based filters, which are publicly available.

The worst PAINS tend to be rhodanines, catechols, and quinones. “Just a simple rhodanine core ended up with an enrichment factor of more than 200%,” Baell said.

Because the filters were based on Monash’s compound library at the time, they aren’t exhaustive. “There might be a compound that looks like a PAIN and is a PAIN but isn’t recognized by the filter, just because that particular analog wasn’t in our library,” Baell said.

But Baell has no plans to expand the PAINS filters. Writing the text strings that make up the filters is “a huge amount of work,” he told C&EN. “I could probably do it more quickly now, but I’ve still got active drug discovery research” to run.

Some PAINS are particularly bad. Walters described a set of “triple threat” PAINS that hit a fungal lysine acetyltransferase known as Rtt109. Walters and his coworkers found a series of hits in their high-throughput screening, but they were unable to optimize them. When they looked more closely at these compounds, they found three problems: The compounds were involved in redox cycling, they had electrophilic reactivity, and they decomposed in buffer to compounds that also had electrophilic reactivity. All three of these characteristics give false-positive results.

Walters and his collaborators published one paper describing the successful development of an Rtt109 assay (PLOS One 2013, DOI: 10.1371/journal.pone.0078877). When they realized they were finding numerous PAINS, they decided not to publish the screening data until they could sort out the true actives. “We could have published all these data,” Walters said, describing the compounds as Rtt109 inhibitors. “But we decided it would be unethical to do so,” because they would be contributing false positives to the literature.

But just because compounds look like PAINS doesn’t mean they really are PAINS. For more than four years, R. Kiplin Guy, a researcher at St. Jude Children’s Research Hospital in Memphis, ignored an electrophilic hit—a chloronitrobenzamide—from a screen for antitrypanosomal compounds.

“I thought I knew what it was going to do,” he said. But he was wrong, and the initial data have been good enough to justify continuing with the project after all (J. Med. Chem. 2013, DOI: 10.1021/jm301687p). An important lesson, Guy told C&EN, is that computational models are good for warning scientists about potential issues but that experiments need to be carried out before dumping compounds.

A major hurdle to eradicating PAINS is that many have already been reported as inhibitors of various targets. When they show up again in other assays, people often assume that the compounds interact with multiple targets and have a core structure that just needs to be modified to get any needed selectivity. It leads to a vicious circle that Walters calls “the PAINS cycle.”

NOT SO PAINFUL
A reaction scheme showing a chlorobenzamide being optimized to a drug candidate.
Sometimes hits that look like PAINS, such as this chlorobenzamide, can still be optimized into viable lead compounds, such as this potential antitrypanosomal compound.

Speakers offered advice on how to break that cycle. First and foremost is to “check the compound’s natural history,” Walters said. “When you get a hit, do a literature search,” said Daniel A. Erlanson, president of San Francisco-based Carmot Therapeutics and editor-in-chief of the blog Practical Fragments. “You don’t want someone like me putting your paper up on my blog.”

Or on his slides. At the symposium, Erlanson devoted part of his presentation to what might be called “PAINS shaming,” calling out examples from the recent literature of papers that reported toxoflavin as an inhibitor of various targets.

Toxoflavin is a particularly notorious compound that falls under the PAINS umbrella because it has been known for more than 10 years to be a redox cycler, Erlanson said. Under the conditions used in many assays, toxoflavin is reduced and then spontaneously reoxidizes, he explained. In the process, it converts oxygen into hydrogen peroxide, which nonspecifically modifies proteins. Last year was a “bumper year” for toxoflavin because it spawned a number of papers identifying the compound as an inhibitor against various targets, he said.

The question often comes up, Baell said, of whether it would be better to physically remove PAINS from screening libraries rather than computationally filtering them. Commercial compound libraries are made up of 5–12% PAINS, he said. Although Baell recommends against adding more PAINS during library expansions, he does not advocate removing ones that are already there.

“I’m hesitant to ask vendors to remove them until we know more about PAINS mechanisms,” he said. “Chemical diversity is so limited in its representation anyway. Why get rid of something for a specific role when we don’t know what other things it might be useful for?” At the least, such compounds are needed for investigating PAINS mechanisms.

The main danger of tolerating the presence of PAINS, Baell continued, “is that it can be very difficult, and even awkward, to dissuade inexperienced researchers” from wanting to move such compounds forward in the drug discovery pipeline on the basis of “superficial, early, and misleading data.”  

Article:

This article has been sent to the following recipient:

0 /1 FREE ARTICLES LEFT THIS MONTH Remaining
Chemistry matters. Join us to get the news you need.