Abstract
Abstracts
High Content Imaging in Genetic Toxicology, a.k.a. Tox Testing - We Don't Just Kill Dreams Anymore
The screening of chemical drug development candidates for genotoxicity is usually accomplished through the miniaturization of regulatory accepted in vitro assays. While this approach can result in highly predictive results for the outcome of the regulatory testing battery, it provides little mechanistic information and the molecular target(s) often remain undefined. The lack of mode of action (MoA) information often leads to the unnecessary attrition of potentially valuable drug development candidates and drives the need for approaches that allow for a better mechanistic understanding and the risk to exposed humans. In this presentation we will show data from a screening flow DDR (DNA Damage Response) assay and the use of mechanistic biomarkers to discriminate between clastogens, aneugens and cytotoxicants. Using real life screening candidates, we will present how high content imaging and other technologies can be used to further improve the mechanistic understanding of aneugenic and clastogenic compounds. Finally, we will integrate the results from these assays to develop an Adverse Outcome Pathway (AOP) starting with the initial molecular targets and subsequent cellular events leading to the positive result in the in vitro micronucleus assay outcome. The value of this approach and its impact on the risk assessment of pharmaceutical compounds will be discussed.
Cell Optimization in 3D Aligned Microfiber Plates for In Vitro Cardiac Toxicity Screening with Automated High Content Imaging
Department of Movement Disorders & Translation 4, MRL, West Point, PA
Human iPSC- derived cardiomyocytes (hiPSC-CMs) are increasingly used for in vitro toxicity screening and disease modeling using various assays, including automated high content imaging. Two of the prerequisites for a reliable plate-based compound screening assay are the consistency of cell health across and between the plates with low variability and optimized positive control compounds. The goals of this study were: 1) to identify the hiPSC-CM cell culturing conditions (cell density and plate format) that would yield maximal percentage of healthy cells and minimal well-to-well variability, and 2) to identify the optimal positive control conditions (compounds and concentrations) for different time points. The cells were plated in 3D aligned 96 or 384-well microfiber plates (NanoAligned) at 25,000, 30,000, 50,000, 70,000 cells/cm2 in 96-well plates, and 40,000, 50,000, 80,000, and 100,000 cells/cm2 in 384-well plates. Cell health was monitored every 3–4 days up to D15 after seeding. Cells were stained with calcein AM (healthy/necrotic cells), EthD-1 or RedDot (dead cells), Hoechst (nuclei), and TMRM (mitochondria) and imaged with the Operetta CLS confocal system (Perkin Elmer). Antimycin, staurosporine, and kinase inhibitors (sunitinib, crizotinib, and regorafenib) were tested as positive controls for different cellular features at 0.1–100μM for 4, 24, 48, and 72 hrs. Cellular features were extracted and percentages of healthy, necrotic and dead cells were quantified in each condition. Improved overall cell health and more homogeneous cell distribution were observed with increasing cell seeding density. Based on the results of this study, cell seeding density of 70,000 cells/cm2 in 96-well format and 100,000 cells/cm2 in 384-well format were the most optimal conditions with maximum percentage of healthy cells (92% and 94%) and lowest well-to-well variability; and the optimal positive controls were 0.1 μM antimycin at 4 hrs, 3 μM staurosporine at 24 hrs, and 0.78 μM staurosporine at 48 and 72 hrs.
Autophagy is a complex pathway regulated by numerous signalling events that recycles macromolecules and may be perturbed in lysosomal storage disorders (LSDs). During autophagy, aberrant regulation of the lysosomal Ca2+ efflux channel TRPML1 [transient receptor potential mucolipin 1 (MCOLN1)], also known as MCOLN1, is solely responsible for the human LSD mucolipidosis type IV (MLIV); however, the exact mechanisms involved in the development of the pathology of this LSD are unknown. In the present study, we provide evidence that the target of rapamycin (TOR), a nutrient-sensitive protein kinase that negatively regulates autophagy, directly targets and inactivates the TRPML1 channel and thereby functional autophagy, through phosphorylation (phospho-events). To demonstrate that these phospho-events regulate autophagy through the Ca2+ efflux activity of the TRPML1 channel, we performed real-time calcium imaging using the Ca2+ -specific reagent Fura-2 AM along with cells transfected with either wild-type or mutant channel constructs. This technique allowed for an easy demonstration of the Ca2+ influx activity of the channel from the media when activated using an appropriate agonist. Mutating our identified phosphorylation sites to unphosphorylatable residues proved to block TOR regulation of the TRPML1 channel, suggesting a mechanism for how TOR activity may regulate the TRPML1 channel.
Image Analysis of Cellular Epigenetic Landscapes Identifies a Signature of Glioblastoma Differentiation
We have pioneered a new approach Microscopic Analysis of Epigenetic Landscape (MIEL) to analyze epigenetic landscapes at single cell level by profiling alterations in epigenetic organization using automated imaging and computer-assisted multivariate analysis algorithms. Taking advantage of MIEL capabilities, we established a signature of differentiated glioblastoma cells that could be employed to identify compounds that differentiate glioblastoma into non-tumorigenic cells using high-throughput drug screening. As testimony to the breakthrough that we have achieved, none of the previous applications can use a single universal set of markers applicable to any mammalian cell. Moreover, our approach is rooted in the nuclear pattern of epigenetic marks, which have defined biological meaning. Indeed, the topography of epigenetic marks in the nucleus is directly related to the topology of active/inactive chromatin. Epigenetic marks are universal determinants of cellular fate and are present in all eukaryotic cells. Our discovery that these marks provide unique interpretable information about the epigenetic landscape of single cells is not an incremental advance. Instead it led to the novel insight that similar distributions of nuclear epigenetic marks likely equate with similar cellular identity. In contrast, in conventional phenotypic screening, a similar distribution of arbitrary markers is not necessarily equated with similar cellular identity. We argue that our work will have a broad impact on: 1. Analysis of cellular fates during reprogramming into iPSCs as well as direct reprogramming into defined cellular fates. 2. Discovering new drug targets based on signatures of cellular epigenetic states. 3. Understanding heterogeneity in human cancers and cancer stem cells including differentiation of cancer stem cells into non-tumorigenic cells.
High-Throughput, Open-Source Analysis of Three-Dimensional Structures Using CellProfiler
Increasingly, researchers create complex biological model systems involving three-dimensional structures, including organoids, tumor spheroids, and even whole organisms. High-content screening systems enable acquisition of 3D images in high-throughput yet the software to analyze crucial metrics from these images has lagged behind. We have recently added 3D segmentation, registration, and measurement capabilities to CellProfiler, open-source software for high-throughput image analysis. As a result, most operations possible for 2D images now also work for 3D images. We have begun testing these new features in several experimental situations including 3D images of live gene-edited human induced pluripotent stem cells labeled with cellular and nuclear dyes. The project is a collaboration with the Allen Institute for Cell Science, who provided data, beta-testing and partial support. Additional support was provided by the National Institutes of Health.
A Novel Screening Probe to Identify Regulators of Short Half-Life Proteins
University Health Network, University of Toronto
Short half-life proteins regulate a wide variety of essential cell processes, from cell cycle control to transcription. For many of these critical short half-life proteins, only a handful of well-characterized regulatory pathways have been identified. We have developed a new tool to identify novel regulators of short half-life proteins in a high-throughput manner. We generated a novel stability probe consisting of the highly regulated c-MYC protein (t½ = 30 min) fused to the Venus fluorescent protein. This fusion protein was ectopically expressed in a non-transformed cell line and then treated with an inhibitor known to decrease MYC turnover. The nuclear intensity of the cells did not markedly change, however the percentage of cells with elevated Venus signal above background (% Venus-positive cells) doubled, providing a read-out of increased stability. We applied this probe to screen 320 kinase inhibitors at two doses over three biological replicates, using the Opera confocal high-content imaging platform. Based on Z-score cutoffs calculated from the % Venus-positive cells in response to each treatment, we identified 16 inhibitors that increased the % Venus-positive cells. To further investigate this image-based dataset, we identified additional regulators of MYC-Venus based on morphology features. We performed unsupervised clustering of compounds, identifying four additional inhibitors that clustered with compounds that increased the % Venus-positive cells. Twenty inhibitors were validated by secondary screening at a range of doses, followed by measurement of MYC RNA for a subset of these. Finally, two inhibitors were found to increase endogenous MYC half-life. The fusion of a short half-life protein to Venus enabled us to conduct high-throughput screening for novel regulators of the fusion partner. We are currently testing Venus fusions to other short half-life proteins, with more comprehensive libraries, to better characterize the regulation of protein turnover in non-transformed human cells.
Evotec specializes in the development and implementation of high-content based principles for target validation, assays for high-throughput screening and compound validation to probe the pathophysiological environment. The systems developed at Evotec encompass both a target centric as well as phenotypic approach with emphasis on tissue samples and primary cells from appropriate hosts. Our offering epitomizes the identification of new drugs using disease-relevant models with disease-relevant endpoints. In the area of neurodegenerative disease, Evotec uses brain tissue samples; primary rodent cells as well as embryonic stem cell or iPS cell derived neuronal models throughout the hit identification cascade from target validation to mechanistic studies. Case studies will be used to illustrate how Evotec uses high content assays of different complexity and throughput through different stages of drug development and how Evotec is specially using neuronal, synaptic and dendritic spine characterization approaches as high sophisticated analysis principles.
Real-Time Kinetic Viability and Apoptosis Detection and Screening of 3D Multicellular Tumor Spheroids Using the Image Cytometer
In the recent years, three-dimensional (3D) tumor spheroid models have been increasingly used for small molecule or antibody-based cancer therapy research. The use of 3D multicellular tumor spheroid (MCTS) models may have better representation of the complex in vivo tumor microenvironments for cancer research. Previously, we have published a novel MCTS screening method using the Celigo Image Cytometer, where we described a method for measuring of end point viability and apoptosis of MCTS in 384-well ultra-low attachment (ULA) U-bottom microplates. It would be important to also obtain real-time viability and apoptosis MCTS results, however, there are no established methods. Herein we demonstrate the use of PI and caspase 3/7 as a method to detect viability and apoptosis in 3D MCTS. The method was initially validated by comparing the bright field kinetic growth rates of MCTS in the presence or absence of PI and caspase 3/7 stains for 16 and 21 days. Furthermore, the potential toxicity of PI was analyzed by digesting the tumor spheroids that were exposed to the reagent into single cell suspension and comparing their viabilities to untreated MCTS. By monitoring PI and caspase 3/7 fluorescent intensities in combination with spheroid size over time, the growth inhibition, viability, and apoptosis can be used to characterize MCTS in real-time. Finally, we performed a high-throughput 3D MCTS screening experiment to screen the real-time kinetic viability and apoptosis effects of 14 drug compounds (NIH/NCAT) on U87MG spheroids at different concentrations. Real-time kinetic viability and apoptosis assays are highly important for developing proper 3D cancer models, which can allow researchers to determine time-dependent drug effects that usually are not captured by end point assays. This would allow the improvement MCTS analysis method to better identify more qualified drug candidates for cancer drug discovery research.
The cADDis Assay for cAMP Detects Gi-Mediated Decreases in cAMP in Living Cells
Cell-based assays to detect Gi signaling have involved complex pharmacological interventions that can yield unreliable results. We previously reported our fluorescent cADDis assay and showed that it reliably detects Gs-mediated increases in cAMP levels. Agonists that stimulate a Gs-coupled receptor produce rapid changes in the intensity of a bright green or red fluorescent protein sensor that can be followed over time in living cells using automated fluorescence plate readers or imaging systems. Since cADDis can monitor Gs-mediated increases in adenyl cyclase activity, in theory it should also be capable of detecting Gi-mediated decreases. Here we apply our green fluorescent cADDis to the detection of Gi-mediated inhibition of adenyl cyclase activity. We validated and optimized the assay in living HEK293 cells using several known Gi-coupled receptors and agonists and we report robust Z' statistics and consistent EC 50 responses.
Development of Human Primary Cell Models Used for Toxicology Assessment and Metabolic Disease Drug Discovery
Biomedical researchers and pharmaceutical companies are in urgent need of novel research platforms and tools to accelerate their drug development programs. Traditional immortalized cell lines and animal models offer drug screening throughput at the cost of low translational success. ZenBio, Inc. is a leading provider of human cell tissue and cell systems, and we have focused our expertise to develop drug discovery platforms using normal and disease state models for screening purposes. Our goal is to generate more predictive cell models using sophisticated culture systems while integrating advanced biological readouts that generate meaningful phenotypic data using high content analysis (HCA) approaches. Our programs currently develop and validate 3D skin models, co-cultured and 3D airway models, metabolic disease models and toxicological assessment models; two projects will be described herein. (1) Our toxicology program has developed a high content analysis screening platform using primary human stem/progenitor cells (Endothelial Colony Forming Cells (ECFC)) that not only identifies environmental agent-induced cell death and/or proliferation, but also evaluates the disruption of human developmental processes, specifically cellular differentiation. (2) The primary component in the metabolic syndrome, is a set of risk factors highly correlated with type 2 diabetes and cardiovascular disease. Thus, investigation into therapeutic targets that induce weight loss has become a critical strategy in combatting the metabolic syndrome. A key approach in this fight against obesity and metabolic diseases is research into the stimulation and recruitment of brown adipose tissue. Currently at ZenBio, we are developing a novel adult human brown adipocyte cell model from donors of differing demographics including BMI, age, gender and diabetic state. We are multiplexing high content imaging assays that provide physiological relevance and predictive ability to identify novel human biomarkers and validate existing drug targets for the development of anti-obesity therapeutics.
Using Infrared Thermography to Develop a Novel Imaging Platform to Measure Thermogenesis at a Cellular Level
The activation and recruitment of brown adipose tissue has become an exciting target in the fight against obesity and its related metabolic diseases. The primary function of brown adipose tissue is adaptive thermogenesis driven by the ability to burn energy as heat through UCP-1 directed mitochondrial activity. Activation of brown adipose tissue and stimulation of the browning of white adipose tissue in rodent models results in increased energy expenditure, plasma triglyceride clearance and decreased adiposity. To date, screening efforts have been limited to rodent models leaving a void in the therapeutic development pipeline. ZenBio has addressed this need by providing the first characterized human brown adipocyte cell culture system to the research community. We have developed high content/high throughput screening platforms that multiplex phenotypic assays to screen potential candidates for stimulation of human brown adipogenesis. We have established protocols for individual assays that measure UCP1 protein expression, mitochondrial mass, lipid content, and glucose uptake and are currently validating multiplexed analysis according to standard high content screening practices. To examine the function of brown fat thermogenesis, we are using infrared thermography to develop a novel imaging platform. With specialized optics and camera setup, we have, for the first time, visualized and measured thermogenesis at a cellular level. Combining our sophisticated biological assays with a true physiological readout will provide a high demand service to identify agents that can increase differentiation and enhance activity of human brown adipocytes as potential treatments for obesity, T2D and metabolic disease.
Machine Learning and Computer Vision Approaches for Phenotypic Profiling in Yeast
One of the primary goals of functional genomics is to elucidate the complex relationships between an organism's genotype and its corresponding phenotype. Systematic assessment of subcellular phenotypes using high-content screening is a powerful method to study the genotype-to-phenotype problem. High-throughput microscopy and automated image analysis can increase the depth of functional information about the defects occurring inside the cell that are missed by more commonly used fitness-based assays. In our labs, we have imaged genome-wide genetic perturbations in Saccharomyces cerevisiae using GFP-labeled subcellular compartments to identify many possible morphological defects. My overarching goal is to develop new computational methods for quantitative cellular image analysis to explore a comprehensive list of subcellular morphological defects at both the cellular and population levels in response to thousands of genetic perturbations. This dataset will provide the subcellular phenotypic profile of any genetic perturbation in yeast. So far, I have been developing and optimizing each step of the computational image analysis pipeline separately. I have been optimizing computer vision and machine learning approaches for identifying cells with mutant morphologies and further separating them into specific mutant phenotypes. I propose to produce phenotypic profiles for each mutant background by reporting the percentage of cells with defective morphology, therefore reporting the penetrance of each mutation in yeast genome-wide perturbation screens. My analysis will allow for the identification of connections between discrete biological processes, the prediction of novel gene function, and the generation of a clearer understanding of basic eukaryotic cell biology.
Deep Learning for High Throughput Histology
Deep learning models have shown super human performance in a range of image classification challenges, and are set to revolutionise biomedical research, clinical practice, and the healthcare industry. One ideal application for deep learning type approaches is the analysis and histological grading of tumour biopsies; a task currently performed by human pathologists. Images of tumour sections are notoriously complex to analyse, as there exists extensive spatial, cellular and phenotypic heterogeneity in any tumour section. Thus even highly trained experts can have difficulty making relevant decisions based on the visual examination of these sections. A number of very recent studies have sought to use deep learning approaches to analyse histological sections. However, these approaches have required time consuming and expensive labelling of tissue regions. We thus sought to develop models that can label specific tumour regions based on whole sample classifications alone that could be deployed in high-throughput image analysis workflows. Specifically, we trained a deep neural network to predict breast cancer subtype, using a dataset that contained over a thousand samples. By then using saliency mapping, we developed a method for labelling specific regions of a tumour with a score of how invasive that region was; this highlighted how the network made the classification, and gave an interpretable and clinically relevant readout of tumour heterogeneity. Importantly, via transfer learning we can demonstrate that these models: 1) generalize to whole slide clinical specimens; 2) can be used to quantitatively profile the performance of immunohistochemically staining workflow; and 3) are broadly applicable to microscopy imaging datasets, such as those acquired in high-throughput compound screening studies. Taken together these workflows highlight the potential for deep learning in the analysis of in vivo imaging data, and are paving the way towards automated approaches to cancer diagnosis.
Three-Dimensional Time Lapse Imaging and Quantitative Analysis of the Active Migration of Human Vascular Endothelial Cells into a Multilayered Cell Sheet with Yokogawa's Confocal Image Cytometer CQ1
In the development of the tissue engineering, time lapse imaging using confocal laser scanning microscopy has been an essential methodology to investigate the three-dimensional (3D) dynamic behaviors of cells. To achieve long-term live cell imaging, it is critical to control the phototoxic damage to the cells caused by repeated laser scanning. The CQ1 from Yokogawa Electric Corporation is an all-in-one confocal quantitative imaging cytometer using a microlens-enhanced dual Nipkow disk confocal optical system which has been shown to be less harmful to living cells compared to conventional single beam scanning devices. Here we report the results of 3D time lapse imaging of live cells in a multilayered cell sheet with CQ1.
Magnetic 3D Bioprinting: A PDX-Derived Ex-Vivo Tumor Array Platform for High-Throughput Compound Screening
University of Texas MD Anderson Cancer Centre
Great efforts have been undertaken in the field of cancer to develop clinically relevant pre-clinical models to accelerate the translation of research advances into therapies and diagnostics that can directly improve patient outcomes. Shortcomings of this discovery/research translation effort result from significant discrepancies between simplified in-vitro 2D cell culture models used in drug screening and the complex 3D tumor microenvironment targeted by these same drugs. Patient-derived xenograft (PDX), in which patient tumor tissue implanted into immune-deficient mice, can provide improved models of human tumor biology relative to cell line ones. Therefore, PDX models can offer improved preclinical testing of new therapies for the treatment of cancer. However, a drawback associated with the use of animal systems is the requirement for immense resources needed to undertake such projects. Moreover, animal models have little relevance as a platform for high-throughput drug screening (HTS). To overcome these limitations, we applied magnetic 3D bioprinting to develop a PDX-derived ex-vivo tumor tissue platform for HTS. The PDX-derived tissue generated using magnetic 3D bioprinting mirrors the original PDX tumor in both tissue architecture and genetic signature. We confirm the predictive value of our screening template by demonstrating a similar response outcome between our template and a PDX mouse system generated from the same original tumor tissue and treated with the same panel of drugs. Utilizing an Inflammatory Breast Cancer (IBC) PDX system as the model, we screened our ex-vivo bioprinted tumor samples with a library of 340 anti-cancer compounds and identified lead candidates that had as strong anti-tumor activity as when used to treat IBC patients in the clinic. Incorporating our ex-vivo tumor tissue high-throughput screening platform as an early step in a PDX system will permit identification of new and effective tumor specific therapies in a time, cost and resource efficient manner.
There are over 7,000 known rare diseases and more than 50% have a simple Mendelian pattern of inheritance. We already know the molecular basis of most Mendelian diseases, and the pace of disease gene discovery has only increased with the advent of whole-exome sequencing. However, despite our knowledge of the genetic mutations that cause rare diseases, the development of effective treatments remains a challenge. The most realistic approach to this problem is to repurpose existing drugs. Drug repurposing cuts down the cost of drug development and there are a few Mendelian diseases currently undergoing clinical trials with repurposed drugs. It is possible that many more Mendelian diseases could benefit from drug repurposing. In order to find which drugs will work for each rare disease, I will characterize a thousand diseases in parallel using one method. High-content microscopy can detect the subcellular localization and aggregation of proteins, as well as abnormal subcellular morphology caused by mutant proteins. Therefore, I will systematically phenotype a set of 2,890 pathogenic protein variants underlying over 1,140 diverse Human Gene Mutation Database (HGMD) annotated phenotypes using immunofluorescence and high-content automated confocal microscopy. So far, I have characterized the localization of 616 mutant proteins, and discovered 45 mislocalized proteins (∼8%), indicating a substantial role for protein mislocalization in disease pathology. Protein stability and transmembrane domains appear to be key contributing factors of protein mislocalization. The data collected from this study, combined with previously generated data on protein-protein interactions, protein-DNA interactions and protein stability, will generate a unique, publicly available database for molecular phenotypes in Mendelian disease. Importantly, this resource will provide a platform for future efforts in drug discovery.
Machine Learning Enables Live Label-Free Phenotypic Screening in 3D
Cancer Research UK Edinburgh Centre, Institute of Genetics and Molecular Medicine, University of Edinburgh
There is a large amount of information in brightfield images that was previously inaccessible using traditional microscopy techniques. This information can now be exploited using machine learning approaches for both image segmentation and the classification of objects. We have combined these approaches with a label-free assay for growth and differentiation of leukemic colonies, to generate a novel platform for phenotypic drug discovery. Initially a supervised machine learning algorithm was used to identify in-focus colonies growing in a 3D methylcellulose gel. Once identified, unsupervised clustering and principle component analysis of texture based phenotypic profiles were applied to identify novel phenotypes. In a proof of concept study we successfully identified a novel phenotype induced by a compound that is currently in clinical trials for the treatment of leukaemia. We believe that our platform will be of great benefit for the utilization of patient-derived 3D cell culture systems for both drug discovery and diagnostic applications.
We present a concept of a drug-screening platform to test ∼10,000 drugs in one 96-well plate with 8 repeats per drug. The platform is based on high-density hexagonal arrays of conical microwells and bar-coded porous monodisperse spherical beads, which are used as slow-release drug delivery vehicles. When monodisperse beads are loaded onto an array, only one bead can enter each microwell, and once in a microwell, a bead rests on the conical sidewalls, effectively sealing the microwell. Conical microwells open towards the top and merge at the top of the array. Therefore, nearly all cells loaded onto an array fall into microwells. Microwells are coated to either prevent or promote cell adhesion and spreading. Drug screening assay starts with loading a cell suspension onto an array and letting cell fall into microwells. Next, bar-coded porous beads carrying drugs are loaded onto the array. A drug slowly released from a bead into the small sealed volume beneath the bead is retained in that volume for an extended time, providing prolonged exposure of cells in the microwell to the drug. At the end of an assay, beads can be removed from microwells to facilitate cell staining. An array of 200 μm deep conical microwells with an opening angle of 35° and a bottom area of 20,000 μm2 (to accommodate 50 adherent cell at 400 μm/per cell) has ∼800 microwells in one well of a 96-well plate. With 100 different bead barcodes, 100 drugs can be tested in each well, and a 96-well plate can be used to screen 9,600 drugs, with an average of 8 microwells per drug.