ASSAY and Drug Development Technologies

Twenty percent of prostate cancer (PCa) patients develop a noncurable drug-resistant form of the disease termed castration-resistant prostate cancer (CRPC). Overexpression of Androgen Receptor (AR) coactivators such as transcriptional intermediary factor 2 (TIF2) is associated with poor CRPC patient outcomes. We describe the implementation of the AR-TIF2 protein-protein interaction biosensor (PPIB) assay in a high-content screening (HCS) campaign of 143,535 compounds. The assay performed robustly and reproducibly and enabled us to identify compounds that inhibited dihydrotestosterone (DHT)-induced AR-TIF2 protein-protein interaction (PPI) formation or disrupted preexisting AR-TIF2 PPIs. We used multiparameter HCS data z-scores to identify and deprioritize cytotoxic or autofluorescent outliers and confirmed the resulting qualified actives in triplicate. None of the confirmed AR-TIF2 PPIB inhibitors/disruptors exhibited activity in a p53-hDM2 PPIB counter screen, indicating that they were unlikely to be either nonselective PPI inhibitors or to interfere with the biosensor assay format. However, eight confirmed AR-TIF2 PPIB actives also inhibited the glucocorticoid receptor (GR) nuclear translocation counter screen by >50%. These compounds were deprioritized because they either lacked AR specificity/selectivity, or they inhibited a shared component of the AR and GR signaling pathways. Twenty-nine confirmed AR-TIF2 PPIB actives also inhibited the AR nuclear localization counter screen, suggesting that they might indirectly inhibit the AR-TIF2 PPIB assay rather than directly blocking/disrupting PPIs. A total of 62.2% of the confirmed actives inhibited the DHT-induced AR-TIF2 PPI formation in a concentration-dependent manner with IC₅₀s < 40 μM, and 59.4% also disrupted preexisting AR-TIF2 PPI complexes. Overall, the hit rate for the AR-TIF2 PPIB HCS campaign was 0.12%, and most hits inhibited AR-TIF2 PPI formation and disrupted preexisting AR-TIF2 complexes with similar AR-red fluorescent protein distribution phenotypes. Further secondary and tertiary hit characterization assays are underway to select AR-TIF2 PPI inhibitor/disruptor hits suitable for medicinal chemistry lead optimization and development into novel PCa/CRPC therapeutics.

The nucleolus is a dynamic subnuclear compartment that has a number of different functions, but its primary role is to coordinate the production and assembly of ribosomes. For well over 100 years, pathologists have used changes in nucleolar number and size to stage diseases such as cancer. New information about the nucleolus' broader role within the cell is leading to the development of drugs which directly target its structure as therapies for disease. Traditionally, it has been difficult to develop high-throughput image analysis pipelines to measure nucleolar changes due to the broad range of morphologies observed. In this study, we describe a simple high-content image analysis algorithm using Harmony software (PerkinElmer), with a PhenoLOGIC™ machine-learning component, that can measure and classify three different nucleolar morphologies based on nucleolin and fibrillarin staining (“normal,” “peri-nucleolar rings” and “dispersed”). We have utilized this algorithm to determine the changes in these classes of nucleolar morphologies over time with drugs known to alter nucleolar structure. This approach could be further adapted to include other parameters required for the identification of new therapies that directly target the nucleolus.

Skeletal muscle-associated toxicity is an underresearched area in the field of high-throughput toxicity screening; hence, the potential adverse effects of drugs and chemicals on skeletal muscle are largely unknown. Novel organotypic microphysiological in vitro models are being developed to replicate the contractile function of skeletal muscle; however, the throughput and a need for specialized equipment may limit the utility of these tissue chip models for screening. In addition, recent developments in stem cell biology have resulted in the generation of induced pluripotent stem cell (iPSC)-derived skeletal myoblasts that enable high-throughput in vitro screening. This study set out to develop a high-throughput multiplexed assay using iPSC-derived skeletal myoblasts that can be used as a first-pass screen to assess the potential for chemicals to affect skeletal muscle. We found that cytotoxicity and cytoskeletal integrity are most useful and reproducible assays for the skeletal myoblasts when evaluating overall cellular health or gauging disruptions in actin polymerization following 24 h of exposure. Both assays are based on high-content imaging and quantitative image processing to derive quantitative phenotypes. Both assays showed good to excellent assay robustness and reproducibility measured by interplate and interday replicability, coefficients of variation of negative controls, and Z′-factors for positive control chemicals. Concentration response assessment of muscle-related toxicants showed specificity of the observed effects compared to the general cytotoxicity. Overall, this study establishes a high-throughput multiplexed assay using skeletal myoblasts that may be used for screening and prioritization of chemicals for more complex tissue chip-based and in vivo evaluation.

Deep convolutional neural networks show outstanding performance in image-based phenotype classification given that all existing phenotypes are presented during the training of the network. However, in real-world high-content screening (HCS) experiments, it is often impossible to know all phenotypes in advance. Moreover, novel phenotype discovery itself can be an HCS outcome of interest. This aspect of HCS is not yet covered by classical deep learning approaches. When presenting an image with a novel phenotype to a trained network, it fails to indicate a novelty discovery but assigns the image to a wrong phenotype. To tackle this problem and address the need for novelty detection, we use a recently developed Bayesian approach for deep neural networks called Monte Carlo (MC) dropout to define different uncertainty measures for each phenotype prediction. With real HCS data, we show that these uncertainty measures allow us to identify novel or unclear phenotypes. In addition, we also found that the MC dropout method results in a significant improvement of classification accuracy. The proposed procedure used in our HCS case study can be easily transferred to any existing network architecture and will be beneficial in terms of accuracy and novelty detection.

In response to a variety of insults the unfolded protein response (UPR) is a major cell program quickly engaged to promote either cell survival or if stress levels cannot be relieved, apoptosis. UPR relies on three major pathways, named from the endoplasmic reticulum (ER) resident proteins IRE1α, PERK, and ATF6 that mediate response. Current tools to measure the activation of these ER stress response pathways in mammalian cells are cumbersome and not compatible with high-throughput imaging. In this study, we present IRE1α and PERK sensors with improved sensitivity, based on the canonical events of xbp1 splicing and ATF4 translation at ORF3. These sensors can be integrated into host cell genomes through lentiviral transduction, opening the way for use in a wide array of immortalized or primary mammalian cells. We demonstrate that high-throughput single-cell analysis offers unprecedented kinetic details compared with endpoint measurement of IRE1α and PERK activity. Finally, we point out the limitations of dye-based nuclear segmentation for live cell imaging applications, as we show that these dyes induce UPR and can strongly affect both the kinetic and dynamic responses of IRE1α and PERK pathways.