In this minireview, Drs. Jayme L. Dahlin and Michael A. Walters summarize the different mechanisms by which molecules can be AICs (assay interference compounds), a subset of which can be flagged as PAINS (pan-assay interference compounds) by a popular set of substructure filters. They bin these into non-technology-related interference mechanisms (e.g. nonspecific reactivity, surfactants, aggregation..) and technology-related interference mechanisms (e.g. light scattering, fluorescence, quenching..). An awareness of these mechanisms can help alert drug hunters to red herrings and prioritize the best starting points while reducing risk.
Aches and Pains in Drug Discovery
By Jayme L. Dahlin and Michael A. Walters
In 2010 Baell and Holloway described pan-assay interference compounds (PAINS) as substructure filters to aid high-throughput screen triage. Since not all compounds with PAINS substructures are false positives and “pan-assay” may be an overbroad term, we typically employ the general term AICS (assay interference compounds) and “nuisance compounds” to refer to compounds that match these substructures and/or are known to interfere in assay readouts.
Many of these AICS interfere with assay readouts and/or biology by common mechanisms, which can be broadly grouped into non-technology-related interference and technology-related interference (Figure 1). Note the potential for overlap between the two categories.
Non-Technology-Related Assay Interference Mechanisms
Examples of non-technology-related interferences (compounds that affect biology but through unwanted mechanisms). These compounds can show activity in orthogonal assays and even apparent selectivity if other targets are not as susceptible to the same types of interference. Due to their ability to disrupt many targets, including those necessary for normal function, many compounds acting through these mechanisms may lead to cellular injury (“cytotoxicity”).
Nonspecific reactive compounds covalently modify biological targets without any functional specificity (in contrast to targeted covalent modifiers, which are usually aided by specific non-covalent interactions). Example: 1,2,4-thiadiazoles undergo ring-opening transformations by nucleophiles such as biological thiols. Note that purposeful screening of electrophilic fragments can lead to selective inhibitors.
Redox compounds can produce reactive oxygen species, which can nonspecifically oxidize biological targets sensitize to the redox environment. Example: quinolinediones will produce hydrogen peroxide under reducing conditions, often showing activity in cellular assays sensitive to oxidation.
Aggregators are compounds that classically form colloids at low micromolar compound concentrations in cell-free assays, which can then nonspecifically perturb protein structure and function. Example: proposed SARS-CoV-2 repurposed drugs, traditional Chinese medicines, and even cannabidiol (CBD) have shown activities in cell-free assays, but this activity will be significantly diminished by the addition of aggregate-busting detergents
Surfactants and other detergent-like compounds can nonspecifically disrupt cellular membranes and membrane-bound targets. Related interferences include nonspecific membrane perturbation, which can affect membrane-based targets and lead to activity in cellular assays.
Metal and synthetic impurities from chemical synthesis can directly modify certain biological systems, and activity is abolished with pure compound. Example: copper contaminants from cross-coupling reactions can inhibit metal-sensitive targets such as TET dioxygenases
Chelators can coordinate with metal ions, thereby disrupting metal-dependent biological systems. Example: 8-hydroxy-naphthyridines can nonspecifically chelate divalent metals, and their biological activities can vary according to metal content in the assay.
Cytotoxins can interfere in cellular assays by many mechanisms: decreasing the readout by simply killing or detaching cells, producing apparent on-target phenotypes due to off-target effects, or producing significant morphologies in high-content assays. Of course, in certain cases like cancer therapeutics, toxicity is desired (though this is often accompanied by selectivity and a sound mechanistic basis).
Technology-Related Assay Interference Mechanisms
Examples of technology-related interferences (compounds that primarily interfere with the underlying assay technology). Compounds that interfere with technology can still represent viable hits IF they show activity in orthogonal assays and are thoroughly and continuously vetted:
Light scattering compounds form tiny precipitates in solution that scatter light, disrupting the expected path and intensity of light. Example: light scatterers will lose activity when the particulates are partially solubilized with detergent-like Tween.
Singlet oxygen quenchers are compounds that effectively reduce the concentration of singlet oxygen (1O2), either by direct reaction (chemical quenching) or non-reactive interactions (physical quenching); this type of interference is most relevant in AlphaScreen-based assays, which utilize singlet oxygen for signal generation.
Light-absorbing compounds can interfere with assay readouts by absorbing light in regions critical to the assay technology; colored compounds can mimic activity in colorimetric assays, while UV-absorbing compounds can mimic activity in NADH/NADPH-based absorbance assays. Example: bromophenol blue absorbs light in the same spectral channels used for AlphaScreen signal transmission.
Quenching compounds effectively decrease the intensity of fluorescent light in an assay, and can occur through a variety of processes
Autofluorescent compounds can fluorescence in the same wavelengths as the assay readout, effectively mimicking the production of assay signal without perturbing the underlying biology; compounds that fluoresce in the GFP channel can even appear as active in GFP-reporter assays
Reagent mimetics can disrupt the expected formation of a capture reagent system by mimicking the chemical structure of the ligand. Example: biotin-like compounds can interfere with biotin-streptavidin tag systems.
Reporter stabilizers can directly interact with reporter proteins and enhance their cellular stability to mimic increased production. Example: some 1,2,4-oxadiazoles with structural similarities to luciferin can potently inhibit firefly luciferase and make it difficult to interpret their mechanism of action in luciferase reporter assays.
The Best of the Worst
Here are some of the worst offenders in drug discovery:
Compounds 1–6. The most promiscuous compounds in PubChem. (1. Nucleophilic aromatic substitution with biological nucleophiles, 2. Redox active, reaction with biological nucleophiles, 3. Redox active, reaction with biological nucleophiles, 4. Redox active, 5. Reaction with thiols, 6. Redox active.)
14. Reacts with thiols (n.b. similarity to 5)
Dealing with AICS and PAINS
- Always consider possible mechanisms by which a hit compound may interfere with the assay readout
- Flag problematic compounds early. Use the aggregated medicinal chemistry knowledge embedded in cheminformatic tools like PAINS, REOS (Rapid Elimination of Swill), APT (Abbott Physicochemical Tiering). Understand the strengths and limitations of such approaches.
- If appealing chemical matter is not available among the actives, carefully characterize the flagged compound(s) further to determine whether their pharmacology is real or purely driven by assay interference. A knowledgeable and careful consideration of interference mechanisms may lead to new drug discovery strategies for the target. (e.g., reactive compounds vis-à-vis electrophilic fragment screening)
- Learn the experimental rigor that is often required to identify nuisance compounds by reading the following comparative case studies: benzisothiazolones (assay interference) or useful inhibitors of TPH1 and HIV-1 Nef? See also: “PAINS in the Assay”, the cautionary tale of aminothienopyridazines as tau fibrillization inhibitors (even the correct counter-screens weren’t enough), and the careful work associated with the mechanism of action of SJ-172550 and the characterization of its chemical instability.
- The NIH Assay Guidance Manual details best practices for addressing interferences, and periodically issues updates based on new evidence. Current chapters include nonspecific reactivity, aggregation, light interferences, homogenous proximity assay interferences, and luciferase reporter interferences.
- Some compounds have pain-full looking substructures (ALS) (e.g. ponesimod (S1P modulator for multiple sclerosis), uristat (urinary tract pain), and eltrombopag (thrombocytopenia) but, for a variety of reasons are approved and useful drugs. Don’t put your brain or other targeted organs on the shelf during triage!
- This virtual HTS and its follow-up provides a cautionary tale and demonstrates the importance of considering AICS and PAINS in virtual libraries.
- Critics point out that consistently flagging PAINS is “…like someone who was bitten by a dog named Fido assuming that every dog named Fido is dangerous.” Of course, no scientist likes to be told that he or she needs to follow rules or guidelines. But obvious dross shows up in even the most high-profile journals quite regularly, (primarily from academic research, in our opinion, where screeners lack the extensive compound metadata that is available in industry, or are under more pressure to publish) so perhaps a little more care on everyone’s part is warranted. Counter-screens and confirmatory experiments should be the minimum required prior to publication. And to be fair, biotech and industry are not immune to the effects of interference compounds.
- We suggest that consistently flagging PAINS or AICs or REOS or ATP (no single, reliably available cheminformatic tool exists that covers all the compounds flagged by these methods) should be considered part of the standard HTS triage process. It is more like assuming that every rottweiler is dangerous until it consistently and affectionately licks your hand.
- The use of substructure and whole-molecule flagging methods are here to stay for researchers in the HTS triage trenches since it is most readily available source of distilled hit-to-drug knowledge. Machine-learning and artificial intelligence built on robust industrial databases can help with the challenges of HTS triage. But in the absence of access to well-curated data it may be best to err on the side of caution when choosing to commit expensive medicinal chemistry resources.
- Nuisance compounds have the potential to show tantalizing activity in both cellular and target engagement assays, making them all that much harder to triage. Be careful of the “eureka effect” and the lure of confirmation bias.
- Purposefully challenging an assay with interference is in fact required when validating clinical assays. It can be helpful to purposefully challenge biological assays with known interference compounds. This characterizes the phenotypes from undesirable modes of action (which are often hard to predict in cellular assays) and estimates the risk for each type of interference.
About Jayme L. Dahlin
Jayme L. Dahlin serves as a preclinical medicine and high-throughput biology lead for the National Center for Advancing Translation Sciences (NCATS). Prior to joining NCATS, Dr. Dahlin completed postdoctoral training in chemical biology at the Broad Institute in the laboratories of Drs. Stuart L. Schreiber and Bridget Wagner, as well as a clinical residency in clinical pathology at Brigham & Women’s Hospital/Harvard Medical School. He serves on the editorial board of the NIH Assay Guidance Manual and the Scientific Expert Review Panel of the Chemical Probes Portal. His research interests include chemical mechanisms of biological assay interference, high-throughput screening, and chemical biology.
Michael A. Walters
Michael A. Walters is a Research Associate Professor of Medicinal Chemistry in the Department of Medicinal Chemistry at the University of Minnesota (UMN). He is also a director in the Institute for Therapeutics Discovery and Development in the same department. His primary focus is on advancing therapeutics to treat the cognitive dysfunction that is present in many neurological diseases. He has been at the UMN for almost 15 years. Prior to that he was part of the Chemistry Department at Pfizer (Ann Arbor Laboratories). His first career was in the Chemistry Department at Dartmouth College. He was the communicating author on “The Essential Medicinal Chemistry of Curcumin”, (Dr. Dahlin is a co-author) a manuscript that currently has been viewed over 120,000 times and cited over 800 times.