High-Content Imaging and Informatics: A Joint Special Issue with Society for Biomolecular Imaging and Informatics and SLAS

This Society of Biomolecular Imaging and Informatics (SBI ² ; https://sbi2.org/) special issue of SLAS Discovery was originally planned to coincide with the seventh annual SBI ² conference and exhibit show (SBI ² High Content 2020) from September 15 to 17 at the University of Pittsburgh. Due to the COVID-19 pandemic and our commitment to the health and safety of the community, the SBI ² board of directors canceled the in-person SBI ² High Content 2020 meeting in Pittsburgh. SBI ² will instead be hosting the annual meeting in a virtual setting, as a free event, going live on September 16 and 17, 2020 (https://sbi2.org/2020-virtual-meeting-letter/), registration open June 1, 2020.

This special issue of SLAS Discovery contains 13 peer-reviewed papers, including 10 original research articles, 1 perspective, a technical note, and an application note. Many of these papers were recruited directly from the SBI ² membership and high-content screening (HCS) vendors, although some were submitted in response to a general request for papers issued to both SLAS and SBI ² . The papers cover a wide range of HCS and analysis applications, including three-dimensional (3D) cell culture models, phenotypic screening strategies, deep learning for image analysis, androgen receptor (AR) screening of environmentally relevant chemicals, use of convergent biosciences to study cancer, an assay to quantify misfolded protein accumulation on aggresomes, and computational identification of kinases that control axon growth in mouse.

Shortly after its founding in 2012, SBI ² began co-hosting the HCS/HCA Data and Informatics Special Interest Group (SIG) session from 2013 to 2020 at SLAS’s annual convention and exhibition meeting. In this edition, Fong et al. ¹ present “A Perspective on Expanding Our Understanding of Cancer Treatments by Integrating Two Approaches from the Biological and Physical Sciences,” the discussion topic of the 2019 SIG led by Dr. Shannon Mumenthaler. Multicellular systems such as cancer suffer from immense complexity, and it is important to capture the heterogeneity to achieve a deeper understanding of the underlying biology and develop effective treatment strategies. At the SIG, academic and industry leaders engaged in discussions surrounding what biological model systems are appropriate to study cancer complexity, what assays are necessary to interrogate this complexity, and how physical sciences approaches may be useful to detangle this complexity. The authors highlight the utility of mathematical models in predicting cancer progression and treatment response when tightly integrated with reproducible, quantitative, and dynamic biological measurements achieved using HCS concentration–response screening (quantitative high-throughput screening [qHTS]) and analysis. They discuss the impetus for convergent biosciences bringing new perspectives to cancer research and therapeutics.

Two of the papers in this issue focus on the AR, a member of the nuclear receptor family of transcription factors (TFs) and central regulator of male reproduction that is involved in many pathophysiological processes. Stossi and coworkers describe a single-cell distribution analysis of AR levels by high-throughput microscopy in cell models. ² Cell-to-cell variation of protein expression in genetically homogeneous populations is a biological trait often neglected in HTS campaigns and rarely used to characterize chemicals. The authors describe the capture of single-cell distributions of AR nuclear levels to evaluate assay reproducibility and to characterize responses to a small subset of chemicals. The authors describe a quality control metric that accounts for the distribution of single-cell data, and how it changes after treatment. Exposure to 45 Environmental Protection Agency (EPA)-provided chemicals that included many endocrine disrupting chemicals (EDCs) was then studied. Data from six compounds were then integrated with orthogonal assays to show that quantitative single-cell distribution analysis of AR protein levels is a sensitive and reproducible method to detect the potential androgenic and antiandrogenic action of environmentally relevant chemicals. Human health is at risk due to environmental exposures to a wide range of chemical toxicants and EDCs. The U.S. EPA has been pursuing new in vitro HTS assays and computational models to characterize EDCs. Szafran and coauthors describe the application of a mechanistic high-content analysis assay using a chimeric androgen receptor (ARER) that rapidly characterizes androgenic chemicals. ³ The high-content assay (HCA)-based GFP-ER:PRLHeLa assay allows for direct visualization of estrogen receptor binding to DNA regulatory elements. The authors describe the characterization of a modified functional assay based on the stable expression of ARER, wherein a region containing the native AR DNA-binding domain (DBD) was replaced with the ERα DBD. They demonstrate that an AR agonist DHT induces GFP-ARER nuclear translocation, PRL promoter binding, and transcriptional activity at physiologically relevant concentrations (<1 nM). In contrast, the AR antagonist bicalutamide induced only nuclear translocation of the GFP-ARER receptor and estradiol failed to induce chromatin binding, indicating androgen specificity. An HCS screen of reference chemicals from the EPA and ATSDR identified and mechanistically grouped known (anti-)androgens. The GFP-ARER cell model was then used to screen for potential antiandrogens in environmental samples in collaboration with the Houston Ship Channel/Galveston Bay TAMU EPA Superfund Research Program. The chromatin-binding GFP-ARER model represents a selective tool for rapidly identifying androgenic activity associated with drugs, chemicals, and environmental samples.

3D cell culture models that better recapitulate the physiological environments of human diseases are being deployed for drug discovery and development. 3D models exhibit improved morphology characteristics, cellular complexity, longer life spans in culture, and higher physiological relevance than two-dimensional (2D) cell culture models. HCS of 3D models has the potential to provide valuable information to untangle disease pathophysiology and interrogate drug responses. However, the practicalities of transitioning from 2D to 3D cultures for HCS studies poses challenges with respect to model and assay development, automation and instrumentation, image acquisition and analysis, and resources. Wardwell-Swanson et al. present a framework for optimizing high-content imaging of 3D models for drug discovery. ⁴ They describe three case studies of scaffold-free, multicellular spheroid models with scalable, automation-compatible plate technology enabling image-based applications: investigation of lipid droplet accumulation in a human liver nonalcoholic steatohepatitis (NASH) model, real-time immune cell interactions in a multicellular 3D lung cancer model, and HTS in a 3D co-culture model of gastric carcinoma. The proof-of-concept studies presented provide evidence of the potential for high-resolution image-based analysis of 3D spheroid models for drug discovery, and the application of cell-level and temporal-spatial analyses to explore the multicellular features of 3D models by HCS and more complex, lower-throughput microphysiological systems of organ networks. Kessel and Chan describe a high-throughput image cytometry method for the formation, morphometric, and viability analysis of drug-treated mammospheres (MMSs). ⁵ Nonadherent MMS assays have been used to investigate cancer stem cell (CSC) activities in breast cancers that can form tumorspheres (TSs) and maintain tumor growth. They have also been used to assess the effects of drug treatment on TSs formed from primary cancer cells or cell lines. The authors describe the development of a method using the Celigo Image Cytometer to characterize MMS formation and morphology in controlled conditions, and to evaluate drug effects on viability using calcein AM and propidium iodide staining. MMS formation and morphometric analysis of adherent and nonadherent propagation of four breast cancer cell lines are presented together with the effects of exposure to four drugs. They describe MMS growth and morphometric differences between adherent and nonadherent cell line propagation, among the four cell lines, and in response to drug exposure. Esquer and colleagues report on advanced HCS applications of clonogenicity in cancer. ⁶ Clonogenic assays have been used to observe the effects of radiation on cell survival, particularly cancer cells, and as a measure of CSC stemness where a single CSC has the ability to grow into a colony. CSCs are highly tumorigenic, have an unlimited proliferation potential with the capacity to generate tumor metastasis, and exhibit resistance to cytotoxic and targeted chemotherapy. The authors describe an HCS method to quantify the number and size of colonies in 2D and 3D culture models and to distinguish colonies based on fluorescent markers imaged on an Opera Phenix HCS platform. They describe how to scan at low magnification and rescan at a higher magnification to capture in greater detail colonies or even single cells of interest. Given the complexity of CSCs and their clinical relevance, these new methods provide a way to more effectively study and characterize CSC mechanisms that allow them to proliferate and persist, and for the discovery and development of therapies that can more effectively target these populations. Kim et al. present a comparison of cell and organoid-level analysis of patient-derived tumor organoids (PDTOs) to evaluate cell growth dynamics and drug response. ⁷ These organoids possess unique features given their heterogeneity in size, shape, and growth patterns. Using dynamic confocal live-cell imaging and image analysis methods, the authors examined tumor cell growth rates and drug responses in colorectal cancer (CRC) PDTOs. High-resolution imaging of H2B-GFP-labeled CRC organoids with DRAQ7 vital dye permitted tracking of cellular changes in individual organoids. Morphological features of 3D organoids, including volume, sphericity, and ellipticity, were used to evaluate intra- and interpatient tumor organoid heterogeneity. A strong correlation between organoid live-cell number and volume was found, and linear growth rate calculations based on volume or live-cell counts were used to determine differential drug effects such as cytotoxic versus cytostatic responses in PDTO cultures. Overall, the HCS analysis of CRC PDTOs produced multiple parameters that provided both patient- and drug-specific information.

There has been a resurgence in phenotypic drug discovery strategies. Cell painting is an image-based multiplexed phenotypic assay that uses relatively simple fluorescent dyes to stain cellular components or organelles together with image analysis to identify cells and extract numerous morphological features that can be organized into profiles that correspond with distinct cellular phenotypes. The cell painting assay has been used to compare the phenotypic profiles of cells from normal and diseased tissues, to detect altered phenotypic profiles elicited by chemical or genetic perturbation, and to group compounds and/or genes into functional pathways. The cell painting assay can be used to infer mechanism of action (MoA) by comparing the phenotypic profiles of compounds of unknown MoA with the phenotypic signatures or fingerprints of drugs and probes with known MoAs. The next-generation blueprint of computational toxicology at the U.S. EPA advocates the use of nontargeted HTS profiling assays for an initial characterization of the biological activity of environmental chemicals. Profiling assays should be compatible with concentration–response screening (qHTS) and provide HCS data that identify potency thresholds for the perturbation of cellular biology and provide information on putative mechanisms of toxicity. Willis et al. present a report on phenotypic profiling of reference chemicals across biologically diverse cell types using the cell painting assay. ⁸ The authors describe the use of the cell painting assay to characterize the phenotypic effects of 16 reference chemicals in qHTS mode across six biologically diverse human-derived cell lines. All cell lines were labeled using the same cytochemistry protocol, and the same set of phenotypic features were calculated. While it was necessary to optimize image acquisition settings and cell segmentation parameters for each cell type, the authors did not need to adjust the cytochemistry protocol. For some reference chemicals, similar subsets of phenotypic features corresponding to a specific organelle were associated with the highest effect magnitudes in each affected cell type, and the cell painting assay yielded qualitatively similar biological activity profiles across the group of diverse, morphologically distinct human-derived cell lines. Hughes and coworkers present a report on high-content phenotypic profiling in esophageal adenocarcinoma (EAC) that identifies selectively active pharmacological classes of drugs for repurposing and chemical starting points for novel drug discovery. ⁹ EAC is a highly heterogeneous disease, dominated by large-scale genomic rearrangements and copy number alterations that have hindered target-directed drug discovery and personalized medicine strategies. The authors describe the implementation of an HCS cell painting assay to profile the phenotypic responses to 19,555 compounds across a panel of six EAC cell lines and two tissue-matched control lines. They showcase the automated HCS image analysis pipeline used to identify compounds that selectively modified the phenotype of EAC cell lines and trained a machine learning model to predict the MoA of EAC-selective compounds using phenotypic fingerprints from a library of reference compounds. Several phenotypic clusters were enriched by similar pharmacological classes, including methotrexate and three antimetabolites that were highly selective for EAC cell lines. A small number of hits that exhibited potent and selective activity for EAC cell lines did not cluster with the phenotypic profiles of reference compounds and may be selectively targeting novel esophageal cancer biology. Steigele et al. present an application note on deep learning-based HCS image analysis for the enterprise. ¹⁰ Extracting information from complex imaging data poses substantial challenges that may prevent the broad adoption of more sophisticated phenotypic assays. Deep learning-based image analysis can reduce the effort required to analyze large volumes of complex image data at a quality and speed compatible with phenotypic HCS. The authors describe five essential design principles that should guide deep learning-based analysis of HCS images and multiparameter data: (1) insightful data representation, (2) automation of training, (3) multilevel quality control, (4) knowledge embedding and transfer to new assays, and (5) enterprise integration. Genedata Imagence is deep learning-based software that embodies these principles, and the authors used imaging data from a cell painting assay and a receptor internalization assay to illustrate the attributes of the software. The deep learning-based software retains expert knowledge from training on one assay and can reapply it to different, novel assays in an automated fashion. In a technical note, Howarth and coauthors describe a multiparametric live-cell transmitted light and fluorescent imaging HCS protocol using three fluorescent probes (Hoechst, Yo-Pro-3, and annexin V) to determine IC₅₀ values for cytotoxicity and provide insight into mechanisms of cell death, ranging from apoptosis to necrosis. ¹¹ The authors screened a small library (~200) of narrow-spectrum kinase inhibitors and 101 clinically approved oncology drugs to validate the method in four tumor cell lines on the Operetta HCS platform.

Lesire et al. present a report on high-throughput image-based aggresome quantification. ¹² Aggresomes are subcellular perinuclear structures where misfolded proteins accumulate by retrograde transport on microtubules. Proteostat is a red fluorescent molecular rotor dye, which becomes brightly fluorescent when it binds to protein aggregates. The authors describe the miniaturization of an imaging assay using Proteostat in 384-well plates and the development of an HCS method for the quantification of aggresomes on the InCell 6000 platform. The authors compared two image analysis methods, traditional image segmentation and analysis versus machine learning. Both methods produced robust quantification of cells with aggresomes, with satisfactory Z′ factor coefficients and reproducible EC₅₀ values for known aggresome inducers such as proteasome inhibitors. They screened the 1280-compound Prestwick library of approved drugs to find aggresome inducers and found hits with similar structural and physicochemical properties, some of which were previously described to modulate autophagy.

Devkota et al. present a report describing the computational identification of kinases that control axon growth in mouse. ¹³ The determination of signaling pathways and transcriptional networks that control complex biological processes is a major challenge from both basic science and translational medicine perspectives. Since this type of analysis may identify critical disease driver nodes to target for therapeutic purposes, the authors combined data from phenotypic HCS experiments with gene expression studies of mouse neurons to determine information flow through a molecular interaction network using a network propagation approach. They hypothesized that differences in information flow between control and injured conditions might prioritize relevant driver nodes and identified paths likely taken from potential source nodes to a set of TFs, called sinks. They found that kinases were enriched among source genes sending significantly different amounts of information to TFs in an axonal injury model. TFs that were differentially active during axon growth were also enriched in the set of sink genes. Notably, the enrichment was observed when the source genes were restricted to kinases shown to support or hamper neurite growth. A set of 71 source genes were identified that send significantly different levels of information to axon growth-relevant TFs. Analysis of information flow changes in response to axonal injury and their effects on TFs predicted to promote or inhibit axon growth enabled the authors to create a network diagram between source genes and their axon growth-relevant sink TFs.

We hope that you find this SBI ² special issue of SLAS Discovery illuminating and of interest. The field of HCS continues to evolve with the adaptation and implementation of new imaging technologies, image analysis methods, and informatics strategies. The articles in this issue illustrate how HCS is being applied to more complex and physiologically relevant biology for phenotypic drug discovery and how machine learning is being integrated to extract and analyze the information contained therein. Beyond drug discovery, HCS is providing strategies to test environmentally relevant chemicals for endocrine disrupting activity and toxicological effects. Combining phenotypic HCS data with gene expression studies to develop computational models to determine information flow through molecular interaction networks holds great promise for systems biology research.