Automatic identification of subcellular phenotypes on human cell arrays

Light microscopic analysis of cell morphology provides a high-content readout of cell function and protein localization. Cell arrays and microwell transfection assays on cultured cells have made cell phenotype analysis accessible to high-throughput experiments. Both the localization of each protein in the proteome and the effect of RNAi knock-down of individual genes on cell morphology can be assayed by manual inspection of microscopic images. However, the use of morphological readouts for functional genomics requires fast and automatic identification of complex cellular phenotypes. Here, we present a fully automated platform for high-throughput cell phenotype screening combining human live cell arrays, screening microscopy, and machine-learning-based classification methods. Efficiency of this platform is demonstrated by classification of eleven subcellular patterns marked by GFP-tagged proteins. Our classification method can be adapted to virtually any microscopic assay based on cell morphology, opening a wide range of applications including large-scale RNAi screening in human cells.     

Automatic analysis of dividing cells in live cell movies to detect mitotic delays and correlate phenotypes in time

Live-cell imaging allows detailed dynamic cellular phenotyping for cell biology and, in combination with small molecule or drug libraries, for high-content screening. Fully automated analysis of live cell movies has been hampered by the lack of computational approaches that allow tracking and recognition of individual cell fates over time in a precise manner. Here, we present a fully automated approach to analyze time-lapse movies of dividing cells. Our method dynamically categorizes cells into seven phases of the cell cycle and five aberrant morphological phenotypes over time. It reliably tracks cells and their progeny and can thus measure the length of mitotic phases and detect cause and effect if mitosis goes awry. We applied our computational scheme to annotate mitotic phenotypes induced by RNAi gene knockdown of CKAP5 (also known as ch-TOG) or by treatment with the drug nocodazole. Our approach can be readily applied to comparable assays aiming at uncovering the dynamic cause of cell division phenotypes.High-content image-based screening is a powerful technology for gene function studies or drug profiling. This technology is characterized by the combination of automated microscopy to rapidly acquire high-content images of treated cells and sophisticated computational methods to extract quantitative information in an automatic and unbiased way.Quantitative studies have been performed based on populations of cells to analyze high-throughput RNAi (Wheeler et al. 2004; Neumann et al. 2006; Goshima et al. 2007), protein overexpression (Harada et al. 2005), or drug profiling screens (Perlman et al. 2004; Loo et al. 2007). Such studies require methods for segmentation and feature extraction, and classification if different object classes are considered. Publicly available software platforms like CellProfiler (Carpenter et al. 2006) can be applied. For population-based studies, however, results are often limited to general features of entire cell populations at certain time points.By contrast, following single cells over time allows studying the inherent dynamics of cellular and molecular processes more accurately and is consequently widely used in state-of-the-art cell biology. To make time-lapse imaging of single cells applicable for high content screening, additional methods for tracking of cells throughout image sequences and recognition of their phenotypic changes are required. Tracking approaches have been used, e.g., to quantify the level of fluorescently tagged proteins over time (e.g., Sigal et al. 2006; Gordon et al. 2007) or to quantify cell–cell interactions and cell migration dynamics (e.g., Chen et al. 2009). Automated classification methods have also been used on static images to distinguish different phenotype classes, providing information on the structure and location of subcellular phenotypes at a single cell level (e.g., Conrad et al. 2004; Huang and Murphy 2004; Chen et al. 2007; Hamilton et al. 2007).Combining classification and tracking methods to study the temporal behavior of different cell classes at a single cell level enables a detailed analysis of the kinetics of a phenotype and allows putting different phenotypes in a causative order. This is ideal for many dynamic biological processes such as, the cell cycle. Automatic determination of cell cycle phases has been performed using phase-contrast (Yang et al. 2005; Li et al. 2008) and fluorescence (Chen et al. 2006; Wang et al. 2007, 2008; Padfield et al. 2009) microscopy image sequences. There, cells were classified into a maximum of four phases based on two-dimensional (2D) multicell images. However, none of these previous studies determined cell cycle phase lengths and abnormal morphologies, which are required for fully automated annotation of aberrant mitotic phenotypes.Here, we have overcome this limitation and present a fully automatic approach to determine morphological and temporal phenotypes by accurately computing and analyzing the lengths of normal and, if present, abnormal mitotic phases and their temporal correlation. Our approach is based on three-dimensional (3D) multicell confocal microscopy image sequences of unsynchronized cell populations expressing fluorescent markers of chromosomes. To analyze these images, we introduce an approach that is based on a multislice 2D strategy and consists of the following main steps: Segmentation and tracking of chromosome sets, extraction of static and dynamic image features, classification, phase length determination and parsing the cell division cycle by a finite state machine.We validated our approach on two sets of proof-of-principle experiments. First, we characterized the detailed cell division behavior of a human cell line (HeLa) with fluorescently marked chromosomes. Then, we compared this behavior to cells where mitosis was perturbed with low doses of the spindle poison nocodazole or by RNAi depletion of CKAP5 (also known as ch-TOG), a microtubule-associated protein (MAP) involved in spindle organization in diverse organisms (Gard and Kirschner 1987; Cullen et al. 1999). Our approach performed with comparable accuracy as very time-consuming manual annotation and allowed quantitative and detailed statistical analysis of the effects of drug or siRNA perturbations on cell division. In addition, we successfully applied our approach to images of a different cell line (NRK, normal rat kidney) and from a different screening platform. Thus, our approach is applicable to fully automated analysis of cell division in movies from living cells and can be applied on a large scale or adapted to other biological processes.     

Concise review: a high-content screening approach to stem cell research and drug discovery

High-throughput screening (HTS) is a technology widely used for early stages of drug discovery in pharmaceutical and biotechnology industries. Recent hardware and software improvements have enabled HTS to be used in combination with subcellular resolution microscopy, resulting in cell image-based HTS, called high-content screening (HCS). HCS allows the acquisition of deeper knowledge at a single-cell level such that more complex biological systems can be studied in a high-throughput manner. The technique is particularly well-suited for stem cell research and drug discovery, which almost inevitably require single-cell resolutions for the detection of rare phenotypes in heterogeneous cultures. With growing availability of facilities, instruments, and reagent libraries, small-to-moderate scale HCS can now be carried out in regular academic labs. We envision that the HCS technique will play an increasing role in both basic mechanism study and early-stage drug discovery on stem cells. Here, we review the development of HCS technique and its past application on stem cells and discuss possible future developments. Stem Cells2012;30:1800-1807.     

RNAi microarray analysis in cultured mammalian cells

RNA interference (RNAi) mediated by small interfering RNAs (siRNAs) is a powerful new tool for analyzing gene knockdown phenotypes in living mammalian cells. To facilitate large-scale, high-throughput functional genomics studies using RNAi, we have developed a microarray-based technology for highly parallel analysis. Specifically, siRNAs in a transfection matrix were first arrayed on glass slides, overlaid with a monolayer of adherent cells, incubated to allow reverse transfection, and assessed for the effects of gene silencing by digital image analysis at a single cell level. Validation experiments with HeLa cells stably expressing GFP showed spatially confined, sequence-specific, time- and dose-dependent inhibition of green fluorescence for those cells growing directly on microspots containing siRNA targeting the GFP sequence. Microarray-based siRNA transfections analyzed with a custom-made quantitative image analysis system produced results that were identical to those from traditional well-based transfection, quantified by flow cytometry. Finally, to integrate experimental details, image analysis, data display, and data archiving, we developed a prototype information management system for high-throughput cell-based analyses. In summary, this RNAi microarray platform, together with ongoing efforts to develop large-scale human siRNA libraries, should facilitate genomic-scale cell-based analyses of gene function.     

Human vs machine: evaluation of fluorescence micrographs

To enable high-throughput screening of molecular phenotypes, multi-parameter fluorescence microscopy is applied. Object of our study is lymphocytes which invade human tissue. One important basis for our collaborative project is the development of methods for automatic and accurate evaluation of fluorescence micrographs. As a part of this, we focus on the question of how to measure the accuracy of microscope image interpretation, by human experts or a computer system. Following standard practice we use methods motivated by receiver operator characteristics to discuss the accuracies of human experts and of neural network-based algorithms. For images of good quality the algorithms achieve the accuracy of the medium-skilled experts. In images with increased noise, the classifiers are outperformed by some of the experts. Furthermore, the neural network-based cell detection is much faster than the human experts.     

Analysis and recognition of touching cell images based on morphological structures

Automated analysis and recognition of cell-nuclear phases using fluorescence microscopy images play an important role for high-content screening. A major task of automated imaging based high-content screening is to segment and reconstruct each cell from the touching cell images. In this paper we present new useful method for recognizing morphological structural models of touching cells, detecting segmentation points, determining the number of segmented cells in touching cell image, finding the related data of segmented cell arcs and reconstructing segmented cells. The conceptual frameworks are based on the morphological structures where a series of structural points and their morphological relationships are established. Experiment results have shown the efficient application of the new method for analysis and recognition of touching cell images of high-content screening.     

RNA干涉技术在功能基因组学和医学研究中的应用

RNA干涉 (RNAi)能特异性降解 m RNA,在转录后水平抑制基因活动 ,作为新的技术平台将对功能基因组学研究和疾病的防治产生重要影响。RNAi能特异地“剔出”靶基因的表达及功能 ,可广泛地运用于基因功能的分析。 RNAi是动植物在抗御病毒、转座子等外来基因侵袭或内源性基因变异的进化过程中发展起来的 ,利用这种内在的防御机制可望发展成为征服感染、肿瘤等疾病的有效的手段。     

High-speed large-scale automated isolation of SARS-CoV-2 from clinical samples using miniaturized co-culture coupled to high-content screening

ObjectivesA novel coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is responsible for the current coronavirus disease 2019 global pandemic. Only a few laboratories routinely isolate the virus, which is because the current co-culture strategy is highly time-consuming and requires a biosafety level 3 laboratory. This work aimed to develop a new high-throughput isolation strategy using novel technologies for rapid and automated isolation of SARS-CoV-2.MethodsWe used an automated microscope based on high-content screening (HCS), and we applied specific image analysis algorithms targeting cytopathic effects of SARS-CoV-2 on Vero E6 cells. A randomized panel of 104 samples, including 72 that tested positive by RT-PCR and 32 that tested negative, were processed with our HCS strategy and were compared with the classical isolation procedure.ResultsThe isolation rate was 43% (31/72) with both strategies on RT-PCR-positive samples and was correlated with the initial RNA viral load in the samples, in which we obtained a positivity threshold of 27 Ct. Co-culture delays were shorter with the HCS strategy, where 80% (25/31) of the positive samples were recovered by the third day of co-culture, compared with only 26% (8/30) with the classic strategy. Moreover, only the HCS strategy allowed us to recover all the positive samples (31 with HCS versus 27 with classic strategy) after 1 week of co-culture.ConclusionsThis system allows the rapid and automated screening of clinical samples with minimal operator workload, which reduces the risk of contamination and paves the way for future applications in clinical microbiology, such as large-scale drug susceptibility testing.     

Laser scanning cytometry for automation of the micronucleus assay

Laser scanning cytometry (LSC) provides a novel approach for automated scoring of micronuclei (MN) in different types of mammalian cells, serving as a biomarker of genotoxicity and mutagenicity. In this review, we discuss the advances to date in measuring MN in cell lines, buccal cells and erythrocytes, describe the advantages and outline potential challenges of this distinctive approach of analysis of nuclear anomalies. The use of multiple laser wavelengths in LSC and the high dynamic range of fluorescence and absorption detection allow simultaneous measurement of multiple cellular and nuclear features such as cytoplasmic area, nuclear area, DNA content and density of nuclei and MN, protein content and density of cytoplasm as well as other features using molecular probes. This high-content analysis approach allows the cells of interest to be identified (e.g. binucleated cells in cytokinesis-blocked cultures) and MN scored specifically in them. MN assays in cell lines (e.g. the CHO cell MN assay) using LSC are increasingly used in routine toxicology screening. More high-content MN assays and the expansion of MN analysis by LSC to other models (i.e. exfoliated cells, dermal cell models, etc.) hold great promise for robust and exciting developments in MN assay automation as a high-content high-throughput analysis procedure.     

Identical mutations in the CSB gene associated with either Cockayne syndrome or the DeSanctis-cacchione variant of xeroderma pigmentosum

Xeroderma pigmentosum (XP) and Cockayne syndrome (CS) are two hereditary disorders in which photosensitivity is associated with distinct clinical and cellular phenotypes and results from genetically different defects. We have identified the primary molecular alteration in two patients in whom clinical manifestations strongly reminiscent of a severe form of XP were unexpectedly associated with the CS cellular phenotype and with a defect in the CSB gene. Sequencing of the CSB -coding region in both cDNA and genomic DNA showed that these patients had identical alterations to those in a patient with the clinical features of the classical form of CS. These data, together with fluorescence in situ hybridization analysis, demonstrated that the two siblings with XP as well as the CS patient were homozygous for the same CSB mutated allele, containing a silent C2830T change and a nonsense mutation C2282T converting Arg735 to a stop codon. The finding that the same inactivating mutation underlies different pathological phenotypes indicates that there is no simple correlation between the molecular defect and the clinical features. Therefore, alterations in the CSB gene give rise to the same repair defect at the cellular level but other genetic and/or environmental factors determine the pathological phenotype.     

Phenotype MicroArrays for High-Throughput Phenotypic Testing and Assay of Gene Function

The bacterium Escherichia coli is used as a model cellular system to test and validate a new technology called Phenotype MicroArrays (PMs). PM technology is a high-throughput technology for simultaneous testing of a large number of cellular phenotypes. It consists of preconfigured well arrays in which each well tests a different cellular phenotype and an automated instrument that continuously monitors and records the response of the cells in all wells of the arrays. For example, nearly 700 phenotypes of E. coli can be assayed by merely pipetting a cell suspension into seven microplate arrays. PMs can be used to directly assay the effects of genetic changes on cells, especially gene knock-outs. Here, we provide data on phenotypic analysis of six strains and show that we can detect expected phenotypes as well as, in some cases, unexpected phenotypes.     

Identification of novel mammalian growth regulatory factors by genome-scale quantitative image analysis

Functional profiling technologies using arrayed collections of genome-scale siRNA and cDNA arrayed libraries enable the comprehensive global analysis of gene function. However, the current repertoire of high-throughput detection methodologies has limited the scope of cellular phenotypes that can be studied. In this report, we describe the systematic identification of mammalian growth-regulatory factors achieved through the integration of automated microscopy, pattern recognition analysis, and cell-based functional genomics. The effects of 7364 human and mouse proteins, encoded by individually arrayed cDNAs, upon proliferation and viability in U2OS osteosarcoma cells were evaluated in a live-cell, kinetic assay using quantitative image analysis. Overexpression of more than 86 cDNAs (1.15%) conferred dramatic increases in the proliferation, as determined cell enumeration. These included several known growth regulators, as well as previously uncharacterized ones (LRRK1, Ankrd25). In addition, novel functional roles for two genes (5033414D02Rik, 2810429O05Rik), now termed Gatp1 and Gatp2, respectively, were identified. Further analysis demonstrated that these encoded proteins promoted cellular proliferation and transformation in primary cells. Conversely, cells depleted for Gatp1 underwent apoptosis upon serum reduction, suggesting that Gatp1 is essential for cell survival under growth-factor-restricted conditions. Taken together, our findings offer new insight into the regulation of cellular growth and proliferation, and demonstrate the value and feasibility of assessing cellular phenotypes through genome-level computational image analysis.     

Examination of the increased speed of detection of HER-2/neu gene amplification in breast cancer by fluorescence in situ hybridization (FISH) using image analysis software

The time required to count signals in the detection of HER-2/neu gene amplification in breast cancer by fluorescence in situ hybridization (FISH) has been a problem. To assess whether the amount of time necessary for counting could be reduced, image analysis using computer software (Win ROOF) was tested. Five photographs from each FISH sample were arranged into ten composite photographs. All ten composite photographs were necessary when using the conventional method of manual counting. However, using only four of the composite photographs and the image analysis method, the 60 necessary nucleus numbers could be measured, and a constant ratio of HER-2/neu / CEP 17 was obtained. In all 58 samples used, in the presence or absence of HER-2/neu gene amplification, there was agreement in counts between the conventional and image analysis methods, and a good correlation of r=0.961 (p<0.001) was obtained. Using the image analysis method, the necessary scoring time was reduced, particularly when the HER-2/neu gene had been amplified, where it was completed in about 1/4 of the time normally required. These results indicate that this image analysis method can be applied when using FISH in other areas of research, and may increase the speed of examination.     

A bioinformatics pipeline to build a knowledge database for in silico antibody engineering

A challenge to antibody engineering is the large number of positions and nature of variation and opposing concerns of introducing unfavorable biochemical properties. While large libraries are quite successful in identifying antibodies with improved binding or activity, still only a fraction of possibilities can be explored and that would require considerable effort. The vast array of natural antibody sequences provides a potential wealth of information on (1) selecting hotspots for variation, and (2) designing mutants to mimic natural variations seen in hotspots.The human immune system can generate an enormous diversity of immunoglobulins against an almost unlimited range of antigens by gene rearrangement of a limited number of germline variable, diversity and joining genes followed by somatic hypermutation and antigen selection. All the antibody sequences in NCBI database can be assigned to different germline genes. As a result, a position specific scoring matrix for each germline gene can be constructed by aligning all its member sequences and calculating the amino acid frequencies for each position. The position specific scoring matrix for each germline gene characterizes “hotspots” and the nature of variations, and thus reduces the sequence space of exploration in antibody engineering.We have developed a bioinformatics pipeline to conduct analysis of human antibody sequences, and generated a comprehensive knowledge database for in silico antibody engineering. The pipeline is fully automatic and the knowledge database can be refreshed anytime by re-running the pipeline. The refresh process is fast, typically taking 1 min on a Lenovo ThinkPad T60 laptop with 3G memory.Our knowledge database consists of (1) the individual germline gene usage in generation of natural antibodies; (2) the CDR length distributions; and (3) the position specific scoring matrix for each germline gene. The knowledge database provides comprehensive support for antibody engineering, including de novo library design in selection of favorable germline V gene scaffolds and CDR lengths. In addition, we have also developed a web application framework to present our knowledge database, and the web interface can help people to easily retrieve a variety of information from the knowledge database.     

Molecular correlates of genes exhibiting RNAi phenotypes in Caenorhabditis elegans

Understanding genome-wide links between genotype and phenotype has generally been difficult due to both the complexity of phenotypes, and until recently, inaccessibility to large numbers of genes that might underlie a trait. To address this issue, we establish the association between particular RNAi phenotypes in Caenorhabditis elegans and sequence characteristics of the corresponding proteins and DNA. We find that genes showing RNAi phenotypes are long and highly expressed with little noncoding DNA and high rates of synonymous site substitution (KS). In addition, genes conferring RNAi phenotypes have significantly lower rates of nonsynonymous site substitution (KA). Collectively, these sequence features explain nearly 20% of the difference between the sets of loci that display or lack a RNAi-mediated effect, and reflect aspects both of the RNAi mechanism and the biological function of the genes. For example, the particularly low rate of evolution of genes in the sterility RNAi phenotype class suggests a role of C. elegans life history in shaping these patterns of sequence and expression characteristics on phenotypes. This approach also allows prediction of a set of heretofore-uncharacterized loci for which we expect future RNAi studies to reveal phenotypic effects (i.e., false negatives in present screens).     

Drosophila under the lens: imaging from chromosomes to whole embryos

Microscopy has been a very powerful tool for Drosophila research since its inception, proving to be essential for the evaluation of mutant phenotypes, the understanding of cellular and tissue physiology, and the illumination of complex biological questions. In this article we review the breadth of this field, making note of some of the seminal papers. We expand on the use of microscopy to study questions related to gene locus and nuclear architecture, presenting new data using fluorescence in-situ hybridization techniques that demonstrate the flexibility of Drosophila chromosomes. Finally, we review the burgeoning use of fluorescence in-vivo imaging methods to yield quantitative information about cellular processes. Electronic Supplementary Material Electronic Supplementary Material is available for this article at and accessible for authorized users.     

A computerized cellular imaging system for high content analysis in Monastrol suppressor screens

In this paper, we describe a new bioimage informatics system developed for high content screening (HCS) applications with the goal to extract and analyze phenotypic features of hundreds of thousands of mitotic cells simultaneously. The system introduces the algorithm of multi-phenotypic mitotic analysis (MMA) and integrates that with algorithms of correlation analysis and compound clustering used in gene microarray studies. The HCS-MMA system combines different phenotypic information of cellular images obtained from three-channel acquisitions to distinguish and label individual cells at various phases of mitosis. The proposed system can also be used to extract and count the number of cells in each phase in cell-based assay experiments and archive the extracted data into a structured database for more sophisticated statistical and data analysis. To recognize different mitotic phases, binary patterns are set up based on a known biological mitotic spindle model to characterize cellular morphology of actin, microtubules, and DNA. To illustrate its utility, the HCS-MMA system has been applied to screen the quantitative response of 320 different drug compounds in suppressing Monastrol. The results are validated and evaluated by comparing the performance of HCS-MMA with visual analysis, as well as clustering of the drug compounds under evaluation.