Abstract
Day 1: Wednesday September 13, 2017 - Educational Courses
This course will review the concepts and technologies related to instrumentation (e.g. optics, camera, and illumination), infrastructure, and image acquisition software features and workflow. The level of instruction is geared to those new to HCS and to those interested in learning what's inside the “black box.” An understanding of the key components will enable you to optimize for the quality and throughput of HCS experiments.
This course will provide an introduction to the underlying concepts behind and challenges to the application of new technologies to HCS. These include the use of light sheet microscopy to image 3D cultures, hyperspectral detectors and dyes that promise increased multiplexing, automated FLIM systems for quantifying protein-protein interactions in live cells and SPAD array multifocal microscopes that allow high speed imaging while rejecting signal from autofluorescence.
For this course we are going to use the definition of phenotypic screening in which a substance is used to alter a measurable trait in a cell or organism. This differs from when we screen against a purified protein, or even the activity of a single target within a high-content screen, either endogenous or overexpressed. Hit selection and assay quality measures used in traditional screening (such as Z') may be difficult to apply to phenotypic endpoints that rely on multiple measures. Positive controls might not be readily available or interesting leads having a phenotype that differs from controls. We will use examples from our groups and literature with advice and best practices for designing, implementing and selecting hits from phenotypic high-content screens. From a strategic perspective we will discuss key success factors of historic phenotypic drug discovery projects.
It is critical the scientific community reproduces experiments to advance science; however, a recent Nature article showed biologists could only reproduce 40% of their experiments. This dismal display wastes time and money to bring new findings, cures, therapies, and treatments to humans. Reproducibility begins with developing and validating an experimental assay that is robust and repeatable over time, and provides relevant statistically significant biological information. Basic cell models and the more recent advancements of more complex cell prototypes for physiological relevant models to recapitulate the in vivo environment or response requires special considerations during assay development that includes a multitude of advance planning methods due to time constraints and financial considerations using primary cells, 3D cell models, or coculture cell models. In this workshop, attendees will learn basic and advance strategies, caveats, considerations, and solutions to develop robust and statistically reproducible high content imaging experiments for target-based screening or for phenotypic analysis.
HCS is a very ‘feature-rich’ approach to cell biology. These features enable complex methods for data analysis that reduce these features to a few experimentally tractable observations. How do all these data get reduced to key trends? How are these processes evaluated for significance and at what point does over-fitting become a distraction? This course will introduce the most common methods of data reduction and analysis, approaches for defining significance and conclude with a set of case studies from the literature. The course will be pitched to scientists who are not currently practicing in these data-mining approaches, but see themselves as consumers of such information and want to be better prepared to engage in conversations and experiment planning based on these methods.
The disease of cancer is complex and heterogeneous requiring many different lines of concurrent therapy. This ranges from surgery, to radiation, and ultimately always involves combination therapies in the form of small molecules, immuno-oncology approaches and treatment with biologics. To address this heterogeneity major efforts are underway within oncology drug discovery to find novel targets which could result in more durable and deeper responses. The evolution of novel targets also presents the need for more complex and relevant assays in which to study targets from the immune system, microenvironment and regulatory networks of transcription/translation. For years, the use of biochemical assays has enabled experimental determination of compound on-rates and off-rates that underline the affinity of a certain target with the limitation of studying systems using a highly reductionist approach. The evolution of high content imaging (HCI), however, is one method where these complex systems can be studied in a more physiologic context with the goal of understanding the function of the target, and more importantly how we can interrupt aberrant signaling in the cancer cell. Complexed with HCI is the ability to study things in real time using live imaging or in conjunction with signaling kinetics over multiple time points. This course will provide some real world examples of high content kinetic assays being employed today to elucidate these complex and novel drug targets.
This introduction will acquaint attendees with the concepts, methods, software and workflows behind automated image analysis. We will introduce the researcher to the basic principles behind determining which pixels in an image belong to each cell and/or cellular compartments and measuring properties of interest, with the intent of providing a fuller understanding of the rich information available for discerning phenotypes of interest. No prior knowledge is assumed, though attending the companion introductory sessions is recommended.
Flow cytometry is a legacy technology for single cell analysis that shares many of the same underlying fundamental principles with quantitative image analysis. Flow cytometry and quantitative image analysis are both used to derive high content data from single cells. This course will review the concepts and fundamentals of flow cytometry, terms, operations and processes, and will compare and contrast flow and image cytometry to help the attendees better understand how flow and image cytometry can complement and inform each other.
Imaging based assays for cellular phenotypes are central to high content screening and analysis. Due to the need to segment and measure specific cellular features, overexpression of fluorescently tagged proteins has been a staple of high content assay design. While antibodies are useful for measuring endogenous protein levels and localization, relatively few reagents exist that work well for immunofluorescence applications. This course will focus on the synergy that exists between the rapidly expanding genome editing toolbox (such as CRISPR/Cas9) and high content applications to measure cellular phenomena at the endogenous level. In addition to covering case studies that exemplify this synergy, practical workflows will be discussed for innovative assay design.
This workshop will review and interpret the screening measures of assay quality such as Z' (Z-prime) and Sw (assay window) and will demonstrate how they are related to well-established and broadly-used statistical techniques for reporting effect sizes (Cohen's d, Hedges g). The speaker will discuss the origins of the familiar HT/HC assay quality indices, their strengths and applicability, but also limitations and shortcomings. The talk will demonstrate the connection between the effect size metrics, commonly used statistics (such as Student t, and Hotelling T2), and performance measures employed in machine-learning (sensitivity, specificity, predictive values, F1 score, and AUC).
Although the presentation will mostly focus on the practical problem of assay quality quantification, it will also touch upon other important aspects of data analysis in phenotypic screening. It will reintroduce the important yet often misunderstood concepts of significance, replication, statistical power, fixed and random effects, and meta-analysis, and link those exotic-sounding terms to the everyday praxis of assay design, optimization, and use. The intended audience includes the screening practitioners working with all the types of HT or HC screens (bulk assays, image-based system, and flow cytometry instruments).
This seminar will highlight 3D cell culture model principles essential for HCS drug discovery including assay design, throughput, and imaging and analysis. Examples of applications will include profiling the activity of therapeutics utilizing scaffold-based and scaffold-free tumor organoid model systems.
A guided roundtable discussion focused on Best Practices & Unmet Needs to address these selected challenges in high content imaging practices
Day 2: Thursday, September 14, 2017 - Scientific Program
Charles H. Best Chair of Medical Research Director, Professor, Department of Molecular, Genetics, University of Toronto
We have developed experimental and computational pipelines which combine array-based yeast genetics and automated microscopy for systematic and quantitative cell biological screens or phenomics. In one project, we use the Synthetic Genetic Array (SGA) method to introduce fluorescent markers of key cellular compartments or cell cycle progression, along with sensitizing mutations, into yeast mutant collections. We then perform live cell imaging on the mutant arrays using HTP confocal microscopy to quantitatively assess the abundance and localization of our fluorescent reporters, providing cell biological readouts of specific pathways and cellular structures in response to thousands of genetic perturbations. For automated image analysis, we developed a hybrid computational pipeline that combines outlier detection and classical SVM-driven phenotype labeling, as well as a neural network-based approach. Our neural network, DeepLoc, was able to classify high divergent image sets, highlighting deep learning as an important tool for expedited analysis of high-content microscopy data.
Co-founder & Managing Director, Mimetas, Netherlands
Organ-on-a-chip has recently emerged as the new paradigm in enhanced, 3D tissue culture. The field builds on almost 26 years of developments in microfluidic and associated microfabrication techniques on the one hand and an urge towards ever more physiologically relevant cell and tissue culture approaches on the other hand. Application of microengineering techniques in cell culture enables structured co-culture, 3D culture, the use of flow and associated shear stress and application of controlled gradients. MIMETAS develops a commercially available platform based on a microtiter plate format that harbors up to 96 chips and enables perfused 3D co-culture in a membrane-free manner. The OrganoPlate® facilitates growth of tubules and blood vessels under continuous flow of medium, it allows engineering of organ complexity without usage of artificial membranes. The OrganoPlate® is fully compatible with liquid handling equipment and high-content readers and is easily adopted by end-users. Current flagship models in OrganoPlates® comprise the human kidney proximal tubule, central nervous system, colon, liver and blood vessels. These models are unsurpassed in terms of physiological relevance and throughput.
Stadtman Investigator, National Eye Institute, NIH
Pathological angiogenesis of capillaries located in the back of the eye (choroid) leads to an eye disease “wet” age-related macular degeneration (wet-AMD), one of the leading causes of blindness among elderly. In wet-AMD, choroidal capillaries grow and leak into the eye by breaching through the outer blood-retina-barrier that is formed by the tight junctions of the retinal pigment epithelium (RPE) cell layer located adjacent to these capillaries. Antibodies against Vascular Endothelial Growth Factor (VEGF) provide a temporary treatment by stopping capillary growth but do not cure the underlying disease. This is because there is no good model to identify mechanism of disease initiation. We have combined bioprinting and tissue engineering with induced pluripotent stem (iPS) cell technology to develop a 3D in vitro model of wet-AMD. Using a collagen-based gel for encapsulation of patient-specific iPS cell-derived endothelial cells, choroidal fibroblasts, and pericytes, we successfully bioprinted a microvascular network on one side of a ten-micron thick biodegradable scaffold. On the other side of the scaffold, we grow a RPE monolayer differentiated from the same patient's iPS cells. The scaffold serves as a transient support for RPE and choroid to secrete extracellular matrix and forms a membrane similar to Bruch's membrane in the back of the eye. This 3D tissue shows electrical properties that are reminiscent of the outer blood-retina-barrier of the eye. Furthermore, similar to VEGF induced vascular growth in wet-AMD, the in vitro microvascular network also responds to VEGF. The use of patient-specific iPS cells allows us to dissect genetic pathways associated with wet-AMD initiation. This work provides a platform to discover disease inducing pathways and the possibility of identifying potential therapeutic drugs for wet-AMD.
11:40 AM–12:05 PM: Speaker 1
12:05–12:30 PM: Speaker 2
• Chroma Technologies
• PerkinElmer
• ThermoFisher Scientific
University of California, Irvine
The human body is a complex, 3-dimensional assembly of over 200 different cell types held together by a network of extracellular matrix (ECM) molecules. Sadly, our current laboratory models of human disease do not even begin to reflect this complexity, contributing to the >85% failure rate of FDA-approved trials of new drugs. Mouse studies, though helpful, are also fraught with difficulties, often including poor modeling of human drug metabolism. For these reasons, we have created a human Vascularized Micro-Organ (VMO) platform. These “organs-on-chips” contain multiple human cell types embedded in a 3D ECM that are supplied with nutrients (and drugs) through a living blood vessel network, just as they are in the body. This is a unique platform—no other system offers human micro-organs on a chip that are supported by true blood vessels—that represents an optimal balance between simplicity of use and physiological complexity. We have developed a version of the VMO—the Vascularized Micro-Tumor (VMT) platform—that provides unprecedented opportunities to study drug responses of tumors in a more natural environment than a plastic dish. In the VMT a vascular network forms in a central tissue chamber and anastomoses (connects) with outer channels that represent arterioles (high pressure) and venules (low pressure). A blood substitute flows from the arteriole, through the vascular network, where it nourishes the surrounding tumor, and then out through the venule. The basics of the platform have been developed and its utility has been established as we have incorporated cells from several tumor types within the VMT platform, including colon, breast, melanoma, prostate and glioma, and shown its effectiveness for drug screening. The VMT platform provides a new and superior way to screen drugs for efficacy and toxicity.
Scripps Research Institute Florida, Department of Neuroscience, Jupiter, FL
The effective treatment of mitochondrial diseases requires new therapeutics that target mitochondrial dynamics and function. Processes included under the rubric of “mitochondrial dynamics” that become dysfunctional in mitochondrial disease involve mitochondrial biogenesis (birth), fission (division), fusion, trafficking to subcellular compartments, and mitophagy (turnover). Major functions of individual mitochondria that can become dysfunctional include ATP generation and calcium buffering. Since the central nervous system is targeted in several mitochondrial diseases and neurons have their own unique physiology, we have established a high throughput, high content, and multiplexed assay for mitochondrial dynamics in primary neuronal cultures. The assay system employs mouse cortical neurons that conditionally expresses a mitochondrial-tagged fluorescent reporter; high content imaging of somatic, axonal and dendritic mitochondria; and an image analysis/bioinformatics pipeline that extracts critical features about the mitochondrial system in neurons with and without drug treatment or in the background of genetic disease. With this assay system, we are able to simultaneously quantify mitochondrial mass in the soma and neurites along with morphological features such as length, width, area, and circularity. These features allow us to determine the effects of manipulations on biogenesis, fission, fusion, and mitochondrial health. In parallel, we have established a high throughput assay of mitochondrial function that measures the potential gradient across the inner mitochondrial membrane as a surrogate measure of ATP synthesis. We have used these two assays to screen through several thousand small molecules and have identified compounds that alter mitochondrial biogenesis and health in neurons, and compounds that promote ATP synthesis. Since proper neuronal function is highly dependent on normal mitochondrial dynamics and mitochondrial generated energy, small molecules identified through these screens may improve neuronal health and stave off mitochondrial disease.
3:00–3:25 PM: Speaker 1
4:00–4:25 PM: Speaker 2
4:25–4:50 PM: Speaker 3
The general meeting is a business meeting required for non-profits. This is an excellent opportunity to learn more about the society and find out how you can participate.
Day 3: Friday, September 15, 2017 - SBI2 Conference Scientific Program
Novartis Institutes for Biomedical Research, Basel, Switzerland
Identifying phenotypes in high-content cellular images is challenging. We present a deep learning approach based on a multi-scale convolutional neural network (M-CNN) that, without segmentation and in one cohesive step, classifies cellular images into phenotypes by using directly and solely the images' pixel intensity values. The only parameters in the approach are the weights of the neural network, which are automatically optimized based on a small number of training images annotated with phenotypic labels. The approach requires no object segmentation, no manual customization, and is applicable to single- or multi-channel images displaying single or multiple cells. We evaluated the classification performance of the approach on diverse cellular image datasets. The approach yielded overall a higher classification accuracy compared to state-of-the-art results, including those of other deep CNN architectures. In addition to using the network to simply obtain a yes-or-no prediction for a given phenotype, we use the probability outputs calculated by the network to quantitatively describe the phenotypes. Our study shows that these probability values correlate with chemical treatment concentrations, thus enabling chemical treatment potency estimation via CNNs. Our current supervised deep learning approach requires phenotypic labels to build the classification model. Future directions involving unsupervised deep learning methods that require no phenotypic labels for training will also be discussed.
Vice President of Biology at Recursion Pharmaceuticals
There are approximately 6,000 rare diseases affecting an estimated 25 million people in the United States. Rare diseases disproportionately impact children, and many of these children do not live to see their 5th birthday. Development of therapeutic treatments for rare diseases has been slow, with less than 5% with an FDA-approved treatment. There is a clear unmet need for innovative approaches to rapidly develop new medicines for the millions of patients suffering from rare diseases.
Scientists at Recursion Pharmaceuticals have developed a highly efficient, broadly applicable, and readily scalable approach to drug discovery that simultaneously leverages automated biology and artificial intelligence. By using high-throughput microscopy to measure hundreds of sub-cellular structural changes, caused by pathogenic perturbations, we have been able to generate data-rich “biomarker-less” high-dimensional in vitro phenotypes. Our approach is amenable for use with complex in vitro disease models that are more translatable, increasing the potential for identifying relevant therapeutics.
Using our platform, high-throughput drug screens can be rapidly executed to uncover and repurpose promising drug candidates that rescue disease signatures. This unique approach allows us to efficiently model and find potential treatments for hundreds of traditionally refractory diseases, including Spinal Muscular Atrophy (SMA), Ataxia Telangiectasia (AT), and Neurofibromatosis Type 2 (NF2), making it ideally suited to tackle the urgent unmet medical need of patients with rare diseases.
10:10–10:35 AM: Speaker 1
11:05–11:30 AM: Speaker 2
11:30–11:55 AM: Speaker 3
St. Jude Children's Research Hospital, Department of Developmental Neurobiology, TN
Cell polarity is a driving force that coordinates the choreography of neural development. How polarity signaling organizes the behavior of immature neurons and how polarity signaling cascades are regulated remain key questions facing the field of developmental neurobiology. These questions are critical to understand the pathology of neurodevelopmental diseases, where the production of neurons or their subsequent migration is defective. Studies combining necessity-sufficiency testing, cutting edge imaging technology and high throughput quantitative image analysis in the developing cerebellum show that a conserved polarity-signaling module, called the Pard complex, is essential for neuronal progenitor germinal zone exit by regulating cytoskeletal dynamics and cell-cell interactions needed for neuronal migration. I will present our progress identifying an upstream regulator of the Pard complex: an E3 ubiquitin ligase, Seven in Absentia, which mediates proteosomal degradation of Pard3; to control a shift from tangential to radial migration when cerebellar granule neurons leave their mitogenic niche, and drebrin; to control microtubule-actin crosslinking during CGN differentiation. While it is hypothesized that microtubule-actomyosin crosstalk is required for a neuron's “two-stroke” nucleokinesis cycle, the molecular mechanisms controlling such crosstalk are not defined. By using the drebrin microtubule-actin crosslinking protein as an entry point into the cerebellar granule neuron system in combination with super resolution microscopy, we investigated how these cytoskeletal systems interface during migration. Lattice light-sheet and structured illumination microscopy revealed a proximal leading process nanoscale architecture wherein f-actin and drebrin intervene between microtubules and the plasma membrane. Functional perturbations of drebrin demonstrate that proximal leading process microtubule-actomyosin coupling steers the direction of centrosome and somal migration, as well as the switch from tangential to radial migration. Finally, drebrin function is antagonized by the Siah2 E3 ubiquitin ligase, suggesting a model for control of the microtubule-actomyosin interfaces during neuronal differentiation.
Biophotonics Program Institute Nationale Optic INO, Quebec City, Canada
The combination of Fluorescence Lifetime Imaging Microscopy (FLIM) and hyperspectral imaging in drug screening research can provide information on the molecular specificity as well as the mechanism of action of candidate molecules. The Main drawbacks of such techniques are that they are typically time consuming, and are limited by the number of photons coming from the sample and the required spectral resolution. Conventional FLIM systems can record images of mCerulean3-Venus fluorescence protein pairs in live cells at a maximum rate of 1 image per 80 seconds for a 400x400 pixel image. Companion hyperspectral systems typically have 32 channels which either limit the spectral resolution or bandwidth of the system. Given these limitations, we have developed a new FLIM and hyperspectral system for quantitative high content screening that images at a speed of greater than 0.1 frames per second, and without trade-off on the image quality or lifetime resolution, as encountered with commercial time correlated single photon counting. Our design and implementation of new FLIM-fluorescence resonance energy transfer (FRET) instrument is based on a parallelized detection scheme to overcome the pile-up limitation of conventional Time Correlated Single Photons Counting systems. Our instrument is complimented with an intelligent region of interest selection software and enhanced binning scheme to achieve better lifetime(s) accuracy. An interleaving method enables integrated hyperspectral detection with 64 spectral channels resulting in a spectral resolution of 8nm over a broadband range from 450nm to 850nm. A flexible supercontinuum source further provides flexibility for optimal tuning to any desired chromophore target.
In this presentation we explain lifetime measurement limitation, and how this new high speed confocal FLIM-FRET hyperspectral microscope, achieved eight times faster speed than the reference system without compromise to image resolution and accuracy in florescence decay lifetimes for quantitative high content screening. Recent results of benchmarking studies using live cells tested with this new apparatus will be shown.
2:30–2:55 PM: Speaker 1
3:10–3:35 PM: Speaker 2
3:35–4:00 PM: Speaker 3
Departments of Biochemistry, Chemistry, Pharmacology and Medicine, and the Mass Spectrometry Research Center, Vanderbilt University School of Medicine, Nashville, TN
MALDI Imaging Mass Spectrometry (IMS) produces molecular maps of peptides, proteins, lipids and metabolites present in intact tissue sections. It employs desorption of molecules by direct laser irradiation to map the location of specific molecules from fresh frozen and formalin fixed tissue sections without the need of target specific reagents such as antibodies. Molecular images of this nature are produced in specific m/z (mass-to-charge) values, or ranges of values. Each imaged specimen gives rise to many hundreds of specific molecular images from a single raster of the tissue. In a complementary approach where only discrete areas within the tissue are of interest, we have developed a histology-directed approach that integrates mass spectrometry and microscopy.
We have employed IMS in studies of a variety of biologically and medically relevant research projects, several of which will be presented including studies in diabetic nephropathy involving both proteins and lipids and the differentiation of benign skin lesions from melanomas. In addition, IMS has been applied to drug targeting and metabolic studies both in specific organs and also in intact whole animal sections following drug administration.
This presentation describes recent technological advances both in sample preparation and instrumental performance to achieve images at high spatial resolution (1–10 microns) and at high speeds so that a typical sample tissue once prepared can be imaged in minutes. Instrumentation used in these studies includes both MALDI FTICR MS and MALDI TOF mass spectrometers. Applications utilize MS/MS, ultra-high mass resolution, and ion accumulation devices for IMS studies. Finally, new biocomputational approaches will be discussed that deals with the high data dimensionality of IMS and our implementation of 'image fusion' in terms of predictive integration of MS images with microscopy and other imaging modalities.