Exposure to plastic additives suggests potential androgen receptor agonism in prostate cancer

Introduction: Plastic is the most abundant human-made substance in the world, and routine ingestion represents a global public health threat. Plastics are comprised of solid hydrocarbon polymers with over 10,000 chemical “additives” that augment the material properties of plastic. These additives include known carcinogens, endocrine disrupting compounds, and DNA damaging agents. Endocrine disruptors are of particular concern, as they may promote growth of hormone-dependent cancers or inhibit hormone therapies used in treatment. Perhaps more alarming is how little we know about the thousands of additives not yet characterized. Here we developed a computational platform to identify additives with potential binding activity to androgen receptor (AR), the key hormone receptor used for prostate cancer growth signaling. The goal for this work is to better understand how these routine exposures may impact endocrine signaling and potentially fuel prostate cancer growth. Methods: We developed a pairwise deep learning model to enhance AR agonist prediction accuracy by leveraging relative potency relationships among compounds. After training this model on 469 compound structures and half-maximal effective concentration (EC 50 ) values, we screened 2,712 plastic additives and identified eight commonly used additives as potential AR agonists. We treated AR-dependent LNCaP cells engineered with a fluorescent prostate-specific antigen (PSA) reporter with each additive, ranging from 100 µM to 0.0001 µM. Cell growth rate and AR activity via PSA reporter were quantified using the analysis features on the Incucyte S3® (Sartorius). Results: Upon ranking predicted candidates from our deep learning model, we selected eight compounds for experimental validation based on favorable prediction values, cost, and availability. Out of seven additives initially screened, four showed significant increases in PSA reporter expression upon additive exposure at 10 µM (edetic acid, sebacic acid, 2,4,6-tribromophenol, and TTBP-TAZ). Two additives (TTBP-TAZ and 2,4,6-tribromophenol) showed significant growth proliferation at doses ranging from 1-100 µM. Cross-checking these additives with human biomonitoring data shows that decabromodiphenyl ether and 2,4,6-triboromophenol have been detected in human samples and are being monitored for potential detection. Conclusions: We have developed and deployed a pairwise deep learning approach to screen a database of plastic additives for predicted AR activity and validated experimentally the AR activation for three of these additives. This work could help identify areas of environmental exposure to AR agonists that increase risk of prostate cancer development or progression. Future work is focused on expanding this platform to identify compounds acting as regulators of other hormones relevant in cancer and developing assays to detect these additives in patient samples.