The Society of Biomolecular Imaging and Informatics Second Annual Conference

High-content, image-based screens provide a powerful approach for identifying new compounds that induce similar cellular responses to known drugs, but with different chemical structures or targets. A key challenge for every phenotypic screen is the selection of biomarkers, which are typically picked based on domain knowledge. Here, we present an objective method for identifying “ORACLs” (Optimal Reporter cell lines for Annotating Compound Libraries), whose phenotypic profiles most accurately classify a training set of known drugs. We demonstrate that a single ORACL in a single-pass screen can functionally annotate large compound libraries accurately across diverse drug classes, including drug classes not included in the original training set. Our approach has the potential to change the design of phenotypic screens: desired collections of discovery drug classes can be matched objectively to ORACLs, which will greatly increase the efficiency, scale and accuracy of future phenotypic screens.

Program Sessions

Session I – HCA Assay Development and Implementation

Presentation I-1 Functional and Chemical Genomics to Decipher Heart Failure Mechanisms

Mercola Mark 1 2 Wahlquist Christine 1 2 Rojas-Muñoz Agustin 1 2 Colas Alexandre R. 1 2 McKeithan Wesley L. 1 2 Spiering Sean 1 2 Yu Shenghan Michael 1 2 Jeong Dongtak 3 Wei Ke 1 2 Serrano Ricardo 2 Hajjar Roger 3 del Alamo Juan Carlos 4 McDonough Patrick 5 Lozano Pilar Ruiz 6 Cashman John 7 Price Jeffrey H. 5

1 Department of Bioengineering, University of California, San Diego 2 Sanford-Burnham-Prebys Discovery Institute, La Jolla, CA 3 Icahn Mt. Sinai School of Medicine, New York, NY 4 Department of Mechanical and Aerospace Engineering, University of California, San Diego 5 Vala Sciences, San Diego, CA 6 Department of Pediatrics, Stanford University, San Diego, CA 7 Human Biomolecular Research Institute, San Diego, CA

Current drug therapies to treat heart failure act through relatively few targets, and heart failure continues to be a major cause of mortality. We have developed high-throughput assays to read out physiological function of cardiomyocytes derived from human patient induced pluripotent stem cells (hiPSC-CMs). Physiological effects are recorded optically using small molecule fluorescent probes, and image analysis is developed to analyze kinetics and morphology of action potential kinetics, calcium transient kinetics as well as quantify contraction and relaxation metrics. Use of benchmark compounds that alter kinetics and morphology of the various metrics reveals the ability to discriminate effects caused by diverse mechanisms. With patient-specific hiPSC-CMs, we present data quantifying prolongation of action potentials and after depolarization in response to a Na channel blocker, and we describe chemical characterization and removal of off-target effects of the channel blocker. Combining this technology with functional genomics and patient-specific hiPSC-CMs provides a powerful means to identify cellular proteins involved in the regulation of cardiomyocyte physiology during health and disease. Using this approach, we have identified microRNAs involved in the decline in calcium reuptake through suppression of SERCA2a in heart failure, and we found that blocking one miRNA, miR-25, halted the loss of cardiac function in a mouse heart failure model.

Presentation I-2 Memory & Mood Enhancers: Targeting Neuroplasticity Networks Using High-Content Imaging in Patient-Specific Stem Cell Models

Haggarty Stephen J. Chemical Neurobiology Laboratory, Harvard Medical School, Departments of Neurology & Psychiatry, Massachusetts General Hospital, Boston, USA

The causes of neuropsychiatric disorders are heterogeneous in nature and not fully understood. Despite the immense etiological heterogeneity, individuals often share common behavioral manifestations that increasingly are being shown to arise due to perturbations of common molecular and cellular mechanisms. Human genetic studies are also increasingly identifying genetic factors that influence disease susceptibility. However, despite these advances, there remains a paucity of well-validated, novel targets for therapeutic intervention. Here, I will summarize our recent efforts to advance a platform for modeling the pathophysiology and treatment of neuropsychiatric disorders based upon the use of patient-specific, human induced pluripotent stem cell (iPSCs). Using examples of single-gene disorders, such as Fragile X syndrome, Pitt-Hopkins syndrome, tauopathies, as well as complex polygenic causes, including bipolar disorder, we have developed a panel of high-content and pathway-based screens that report on diverse aspects of neuroplasticity, including networks critical to neurodevelopment, memory, and mood neurocircuits. Collectively, these studies have implicated new molecular targets and networks for developing next generation treatments for neuropsychiatric disorders. On-going generation of an expanded library of iPSCs as part of the ‘MGH iModels Project’, in conjunction with in-depth clinical and cellular phenotyping holds promise for providing insights into the pathophysiological substrates of neuropsychiatric disorders and is likely to inform the development of targeted therapeutics for their treatment and ideally prevention.

Session II – Image Analysis and Informatics

Presentation II-1 KNIME: Integrating Image Processing and Advanced Analytics

Dietz Christian KNIME, University Konstanz, Germany

State-of-the-art high-throughput microscopy calls for software solutions that can handle huge volumes of heterogeneous image data. These kinds of software solutions have to fulfill an extensive range of requirements (e.g. image analysis, machine learning, statistics, and visualization) and yet still be easy to use for the non-expert. There are many monolithic and a number of highly task-oriented platforms that can process and analyze biological image data. However, in view of the diverse nature of present problems, it is often necessary to integrate several applications. This involves going through many manual, time-consuming, and error-prone steps, which are hard to document and therefore incomprehensible to others. The open-source analytics platform KNIME provides the functionality to seamlessly integrate a diverse and powerful collection of software tools and libraries. KNIME is a widely-used (over 50k users), user-friendly data integration, processing, analysis, and exploration workflow system designed to manage huge amounts of heterogeneous data. In the spirit of interoperability and to avoid redundant development but benefit from achievements of other tools, KNIME Image Processing uses and integrates leading-edge libraries such as ImageJ2, ImageJ-Ops, SCIFIO, OMERO, ImgLib2, ClearVolume, CellProfiler and others. These image processing tools can not only exchange data and therefore be used in combination; in addition, they also inherit all advantages from KNIME, for example the smart-caching of data and the possibility to link their output to other KNIME extensions from completely different domains. With image processing, machine learning, statistics, and visualization extensions, KNIME therefore satisfies all software-solutions requirements for the processing and analysis of huge amounts of heterogeneous data.

Presentation II-2 Segmentation and quantification of subcellular structures in fluorescence microscopy images using Squassh

Rizk Aurélien Paul Scherrer Institut, Switzerland

Detection and quantification of fluorescently labeled molecules in subcellular compartments is a key step in the analysis of many cell biological processes. Pixel-wise colocalization analyses, however, are not always suitable, because they do not provide object-specific information, and they are vulnerable to noise and background fluorescence. We present a versatile method named “Squassh” (segmentation and quantification of subcellular shapes),¹ which is used for detecting, delineating, and quantifying subcellular structures in fluorescence microscopy images. Squassh makes use of a segmentation method that directly connects the image-segmentation task with biological reality through prior knowledge about the imaged objects, the image-formation process, and the noise present in the image. This segmentation method corrects for microscope blur and detector noise and provides optimally deconvolved segmentations.²

Squassh outputs location, size, length, perimeter, and intensity of segmented objects which can be used in conjunction with colocalization analysis or to classify objects. It is implemented as a user-friendly multithreaded plug-in for the free open-source framework ImageJ and comes with a set of functions for the free statistical software R to provide data filtering, figure generation, and statistical significance tests.

We then describe how we use Squassh to characterize the trafficking of G protein-coupled receptors (GPCRs). Despite the physiological and pharmacological importance of GPCRs, receptor activation and its translation into cytoplasmic trafficking and cellular response remain elusive. By using RAB GTPases as markers of intracellular compartments together with Squassh for automatic colocalization quantification we show how to monitor the dynamic distribution of receptors after stimulation by a given drug.

References

1. Rizk A, Paul G, Incardona P, et al.: Segmentation and quantification of subcellular structures in fluorescence microscopy images using Squassh. Nat Protoc 2014;9:586–96.

2. Paul G, Cardinale J, Sbalzarini IF: Coupling image restoration and segmentation: a generalized linear model/Bregman perspective. Int J Comput Vis 2013;104:69–93.

Session III – Phenotypic Drug Discovery

Presentation III-1 Embracing Complexity: A Drosophila Approach to Cancer Therapeutics

Cagan Ross Developmental and Regenerative Biology/Oncology, Mt. Sinai Hospital, New York, USA

Current efforts at ‘personalized medicine' primarily focus on using genomic analysis to identify the best therapeutic target. This has shown some success but, overall, precision therapies have struggled to stably reverse tumor progression. In recent years, my laboratory has used Drosophila to explore the importance of increased complexity in both our models and our therapies.

Regarding models, we find that building tumors with increasing genetic complexity leads to progressive resistance against targeted therapies. These fly models are built to mimic specific patients. For example, drugs that successfully suppressed progression of a Ras model of colorectal cancer failed to suppress Ras-Pten-Apc-P53 tumors. In general, the more variants captured by our fly models—which more completely capture a patient's variant profile—the more resistant they are to single agents. I will discuss how we use our models to develop drug cocktails that are designed to address this complexity while lowering whole body toxicity. Working with Owen Sansom's group, we have validated these cocktails in mammalian colorectal systems.

Regarding drug development, my laboratory has collaborated with multiple other laboratories to develop new generation lead therapeutic compounds. Combining fly genetics plus medicinal chemistry—in collaboration with the laboratories of K. Shokat, A. Dar, A Schlessinger—we have developed low-specificity kinase inhibitors that address several targets. I will discuss how 'balanced polypharmacology' can prove more successful at addressing tumor progression, resistance, and heterogeneity.

Finally, I have recently opened the Center for Personalized Cancer Therapeutics. Working with clinicians, genomicists, and bioinformaticists, we have developed a 'pipeline' in which sequencing data leads to personalized “fly avatars.” Using robotics-based screening, we develop candidate therapeutic cocktails for each patient. Our goal is to leverage the promise of genomics and to ask the question: how much complexity do we need to account for to successfully treat patients?

Presentation III-2 High definition functional and chemical genomics for drug discovery

O'Shea Patrick High Content Biology, Innovative Medicines, AstraZeneca, UK

Phenotypic-led drug discovery is a key strategy to accelerate new medicines development and identify novel targets with strong rationale for therapeutic treatment of disease. As an industry we are increasingly focused on tackling unprecedented disease areas. This has led to an increased demand to run phenotypic discovery programs for the identification and validation of novel targets. Whilst such approaches are not new, technologies are emerging that promise to enable phenotypic-led drug discovery in disease-relevant models at unprecedented scale and definition.

High-Content Analysis (HCA) has developed into an important platform, not least for phenotypic profiling. We describe approaches taken at AstraZeneca to develop and implement multiparametric high-content assay panels in physiologically relevant models of disease to identify and validate targets and characterize compound mode of action. We discuss the tools and workflows adopted to facilitate the efficient, consistent, and reliable application of HCA at scale for phenotypic discovery using both chemical and biological (e.g. miRNA, gRNA, protein) libraries. We believe a key component to target discovery is the integration of molecular genetics with phenotypic discovery to aid in the identification of the fundamental genetics responsible for disease and drug response. The Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) and CRISPR associated (Cas) system has been developed as an RNA-guided gene-editing tool based on components of a bacterial immune system that can make the process of adding, deleting or replacing specific genes in organisms and cells easier than previous approaches. We describe work we have undertaken to develop and deploy arrayed-CRISPR libraries for phenotypic discovery. Such approaches should enable functional readout in any cellular assay enabling the utilization of sophisticated models of disease in phenotypic screening and ultimately, high definition functional genomics for target discovery.

Session IV – Innovative Imaging

Presentation IV-1 Biodynamic Imaging for Phenotypic Profiling of 3D Tissues

Nolte David D. Purdue University

Interest in 3D tissue is growing because it provides a biologically relevant context for cellular signaling and growth. However, a serious challenge presented by 3D culture has been the difficulty to extract high-information content from deep inside tissue. Biodynamic Imaging (BDI) overcomes this barrier by combining principles of Doppler radar with light microscopy. It is a label-free laser-ranging technique that images volumetrically up to 1 mm deep inside living tissue, using intracellular motion as its image contrast. Intracellular motions, such as organelle motility, nuclear size changes, cell membrane fluctuations, and cell adhesion restructuring capture different drug mechanisms of action that affect cellular dynamics. These different mechanisms create characteristic drug-response Doppler spectrograms that act as unique fingerprints for each drug.

Biodynamic imaging has been applied to an array of xenobiotic compounds and environmental conditions in multicellular tumor spheroids, xenograft explants, and cancer biopsies. The results of phenotypic profiling show functional differences in the response of proliferating tissue relative to hypoxic regions of the tumors, such as drug-response spectrogram features that make the important distinction between drug-induced apoptosis and necrosis. Cellular mitosis also can be identified as high-amplitude events that can be correlated to the cell cycle by synchronizing and controlling the collective behavior in the tissue.

This talk explains the operational principles of biodynamic imaging, showing how BDI is able to perform phenotypic profiling in three-dimensional tissue culture. Potential applications include proliferation and toxicity studies in life science research, phenotypic profiling of drug response, and prediction of chemosensitivity of patient biopsies.

Presentation IV-2 Structured Illumination “Confocal” Microscopy for 3D High Content Screening

Price Jeffrey H. 1 2 3 Kellner Albert L. 1 Guigli Mirco 1 2 3 1 Scintillon Institute 2 Sanford-Burnham Medical Research Institute 3 Vala Sciences, Inc.

Structured illumination microscopy (SIM) as implemented by Tony Wilson's group increases 3D optical resolution. Although the term “confocal” is not often associated with SIM, the mathematical models are nearly identical as are the predicted and measured resolution enhancements. While classical confocal optics block out-of-focus blur with the pinhole, SIM subtracts out-of-focus blur computationally. Spinning disk confocal microscopes (which also sacrifice some resolution to gain parallelism) transmit only about 1/4 of the light of wide field microscopes, and spot-scanning confocal microscopes transmit only about 1/200^th, due to light blocked by the pinhole and optical complexity. However, optically-simpler SIM has light efficiency identical to wide field microscopy. SIM has also been reported to be less prone to lateral aberrations (variations in resolution across the field of view) than confocal microscopy. SIM was reported to be more prone to residual background artifacts imaging thick samples containing bright out-of-focus components; however, since HCS specimens are typically thin this may be rare. As with all confocal methods, SIM requires multiple recordings to construct a single image or optical section. While classical confocal recordings can be integrated to acquire the finished result in a single image (as is typical in spinning disk systems), the SIM recordings of each projected pattern must be acquired independently for computation of the final confocal image. SIM patterns can be changed very rapidly with the Texas Instruments Digital Light Processor^TM (DLP^TM) so that the slowest step is image acquisition – 100 fps on our Hamamatsu Flash CMOS camera. We experimented with 4-, 9- and 25-element repeating 2D patterns to create SIM images, compared the resolution and image quality to wide field images, and summarize the results relative to confocal and wide field microscopy to evaluate the suitability of this technique for 3D HCS.

Closing keynote

Bhatia Dr. Sangeeta Massachusetts Institute of Technology, Broad Institute, Harvard University

Engineered livers for modeling health