Abstracts

1 MOLECULES CRAFTED TO SPY ON CELLS AND TUMORS

Roger Y. Tsien^{1 1}University of California, San Diego, Pharmacology, School of Medicine, La Jolla, California

Genetically encoded tags and indicators are molecular spies that reveal specific gene products and biochemical processes in living cells and organisms. Fluorescent proteins from jellyfish and corals have been bred to eliminate multimerization and cover the entire visible spectrum. Somatic hypermutation in B lymphoma cells harnesses the immune system to produce a powerful new way to evolve protein properties. Indicators constructed from fluorescent proteins can report local dynamic signals such as redox potential, neurotransmitter concentrations, protein-protein interactions, and kinase vs. phosphatase activities. Although fluorescent proteins are powerful tools, they cannot be reduced below ~220 amino acids, and their only useful readout is fluorescence. Much shorter peptide sequences combine genetic encoding with the greater range of spectroscopic properties available by organic synthesis. Tetracysteine motifs of 6-12 amino acids can be labeled in live cells with membrane-permeant biarsenical dyes. Unique applications include green vs. red pulse-chase labeling of old vs. new copies of the same protein, electron-microscopic localization, chromophore-assisted light inactivation of a chosen protein without the problems of antibody penetration, and measurement of local Ca2⁺ within nanometers of proteins such as Ca2⁺ channels. For clinical applications one would prefer not to have to introduce genes or be limited to optical detection. Arginine-rich sequences are known to mediate uptake of a wide variety of contrast agents into cells and tissues in vivo. We find that such uptake can be prevented by appending certain polyanionic sequences and selectively re-activated by cleavage of the linker. This new mechanism offers the exciting possibility that radioactive, magnetic, and infrared contrast agents and therapeutic drugs may be concentrated in diseased tissues expressing particular extracellular proteases.

2 SINGLE CELL KINASE SIGNALING FOR MECHANISTIC AND CLINICAL ANALYSES

Garry P. Nolan^{1 1}Stanford University, Microbiology and Immunology, School of Medicine, Stanford, California

Intracellular assays of signaling systems has been limited by an inability to correlate functional subsets of cells in complex populations based on active kinase states or other nodal signaling junctions. Such correlations could be important to distinguish changes in signaling status that arise in rare cell subsets during functional activation or in disease manifestation. Simultaneous detection of activated kinases and phosphoproteins in simultaneous pathways in subpopulations of complex cell populations by multi-parameter flow cytometric analysis allows identification of signaling cascades for disease states by ordering of kinase activation and phosphoprotein status in signaling hierarchies. Importantly, we demonstrate that ordering of these activations requires multiple interrogations of cells, and that the networks discovered are reflective of deeper correlations. Using Bayesian Network analysis (a form of machine learning) one can infer pathway connectivity in an automated fashion, allowing for high throughput derivations of signaling system networks graphs in PRIMARY CELLS. Notably, when kinase inhibitors, previously selected in in vitro assays are tested on complex cell populations, single cell analysis of signaling states reveals shocking differences in the inhibition of kinase activity in different cell subsets that will be discussed. The approach has powerful applications in mechanistic understanding, drug screening, and patient stratification for prediction of disease outcome in cancer, autoimmunity, infection, based on signaling network status. (1) Irish J.M., Hovland R., Krutzik P.O., Perez O.D., Bruserud O., Gjertsen B.T., Nolan G.P. (2004) Single Cell Profiling of Potentiated Phospho-Protein Networks in Cancer Cells. Cell. 118:217-228. (2) Sachs K., Perez O., Pe’er D., Lauffenburger

D.A and Nolan G.P. 2005. Causal protein-signaling networks derived from multiparameter single-cell data. Science. 308:523-9.

3 LOCATION PROTEOMICS: IMAGE INFORMATICS FOR SYSTEMS BIOLOGY

Robert F. Murphy^{1 1}Carnegie Mellon University, Biological Sciences & Biomedical Engineering, Mellon College of Science & Carnegie Institute of Technology, Pittsburgh, Pennsylvania

Systems Biology requires comprehensive, systematic data on all aspects and levels of biological organization and function. In addition to information on the sequence, structure, activities, and binding interactions of all biological macromolecules, the creation of accurate, predictive models of cell behavior will require detailed information on the distributions of those molecules within cells and the ways in which those distributions change over the cell cycle and in response to mutations or external stimuli. Current information on subcellular location in protein databases is limited to unstructured text descriptions or sets of terms assigned by human curators. These entries do not permit basic operations that are common to other biological databases, such as measurement of the degree of similarity between the distributions of two proteins, and they are not able to fully capture the complexity of protein patterns that can be observed. The field of location proteomics seeks to provide automated, objective, high-resolution descriptions of protein location patterns within cells. The initial foundation for the field was the demonstration that automated classifiers could be trained to recognize all major subcellular patterns in fluorescence microscope images, and especially the critical finding was that such systems could discriminate location patterns better than visual examination. The very high accuracy (over 98% on single 3D images) of these systems gave confidence that the numerical features used to describe location patterns could form a basis for extending the methods to unsupervised learning of patterns. To this end, we have described grouping of proteins into statistically-indistinguishable location patterns using consensus clustering methods. The resulting clusters, or location families, are analogous to clusters found for other domains, such as protein sequence families. Our current work is focusing on extending this work in a number of new directions. These include analyzing the temporal dependence of subcellular patterns, generalizing pattern analysis across many cell types, analyzing mixed multicell images in cultures and tissues, and creating generative models of subcellular patterns that can be incorporated into comprehensive models of cell behavior. The combination of these methods with large scale, high throughput imaging approaches will allow realistic cell modeling that reflects detailed knowledge of the subcellular location of all proteins. We anticipate that work in this field will also lead to improved diagnostics based on subcellular pattern discrimination.

4 THE NIH CHEMICAL GENOMICS CENTER: ANNOTATING THE BIOLOGICAL ACTIVITY OF CHEMICAL LIBRARIES USING QUANTITATIVE HIGH THROUGHPUT SCREENING

Jim Inglese^{1 1}National Institutes of Health (NIH), NIH Chemical Genomics Center, Bethesda, Maryland

The NIH Chemical Genomics Center (NCGC; www.ncgc.nih.gov/) is the founding member of the Molecular Libraries Screening Center Network (MLSCN), a network of ten centers established as part of the NIH Roadmap for Medical Research (nihroadmap.nih.gov/). The mission of the NCGC is to make accessible the technologies and protocols of high throughput biomolecular screening and chemistry optimization, developed primarily in the pharmaceutical and biotech industries, to academic investigators. The ‘product´ emerging from the NCGC pipeline will not be drugs, but rather chemical probes to aid in the understanding of biology and validation of therapeutic targets. In refining the screening process, the NCGC has developed a strategy called “quantitative HTS” (qHTS) to generate concentration-response curves for a range of biochemical and cellular assays on large compound collections using existing technologies. Our process is highly refractory to false positives, easily identifies compounds of low potency and efficacy, and demonstrates the potential to build high-quality chemical genomic databases. Examples from this new paradigm will be presented.

THE LCOS-BASED PROGRAMMABLE ARRAY MICROSCOPE (PAM)

Thomas M. Jovin¹, Martin Thomas², Guy M. Hagen¹, Donna

J. Arndt-Jovin³, Keith A. Lidke⁴, Andrew Hill², A. Martyn Reynolds², Jeremy Graham², Rainer Heintzmann⁵, Quentin Hanley^{6 1}Max Planck Institute for Biophysical Chemistry, Molecular Biology, Goettingen, Niedersachsen, Germany; ²Cairn Research Ltd., Faversham, United Kingdom; ³Max Planck Institute for Biophysical Chemistry, Goettingen, Niedersachsen, Germany; ⁴Sandia National Laboratories, Biomolecular Materials and Interfaces Department, Albuquerque, New Mexico; ⁵King’s College London, New Hunt’s House, Randall Division of Cell and Molecular Biophysics, London, United Kingdom; ⁶Nottingham Trent University, School of Biomedical and Natural Sciences, Nottingham, United Kingdom

In 1998, we reported the realization of an optically sectioning fluorescence microscope (PAM) for rapid, light efficient 3D imaging of living specimens.^1,2 The design was based on conjugate structured illumination & detection implemented with a spatial light modulator (SLM) located at an image plane of a conventional fluorescence microscope. The major advantages of the PAM are: (1) compact design with no moving parts; (2) very significant speedup/improvement in sectioning due to a pixel duty cycle of up to 50%, and simultaneous detection & processing of both conjugate (in-focus) & non-conjugate (out-of-focus) images; (3) ultrasensitive detection via electron multiplying CCD cameras; (4) programmable & adaptive optical sectioning based on libraries of dot, line & pseudo-random (Sylvester) sequence patterns; (5) flexible light sources, including LEDs; (6) generation & detection of arbitrarily polarization patterns; (7) compatibility with hyperspectral & lifetime-resolved imaging; (8) incorporation of light sources for photo-destruction and/or transformation, modes compatible with FRAP, FLIP & FCS/ICS; (9) minimal photobleaching. We report here the design, operation & application(s) of a new generation commercial PAM created by the combined efforts of the MPIbpc (Mol Biol Dept) and Cairn Research Ltd. (UK). The stand-alone module, including light source(s) and detector(s), features an innovative catadioptric design and a ferroelectric liquid-crystal-on-silicon (LCoS) SLM instead of the original DMD. It can be attached to a side port of a(ny) unmodified fluorescence microscope. The Cairn prototype system currently operated at the MPI incorporates a 6-position high-intensity LED illuminator and an Andor iXon emCCD camera, and is mounted on an Olympus IX71 inverted microscope with 60-150x objectives and a PZ translator, a Cairn Optoscan monochromator, and a Prior Scientific ProScan stage. Further enhancements are being incorporated: (i) point- and line-wise spectral resolution, detecting with an Applied Scientific Imaging SpectraCube imager or a Specim ImSpector imaging spectrograph; and (ii) lifetime imaging (FLIM) using phase-modulation² and TCSPC modules. Multiphoton operation and other nonlinear techniques should be feasible. Real-time sectioning with 50-100 ms exposures and lag-free manual focusing have been achieved with a single LED source. Operation is insensitive to mechanical shock applied to the microscope table; thus, the system is suitable for ruggedized (field) use. It is currently being applied to developmental biology and signal transduction studies, e.g. based on quantum dot ligands³. ¹Verveer PJ et al. 1998. J. Microsc.189:192; ²Hanley QS et al. 2005. Cytometry 67A:112 (latest PAM paper); ³Lidke DS et al. 2005. J. Cell Biol. 170:619.

10 CHEMICAL CYTOMETRY-ANALYTICAL CHEMISTRY OF SINGLE CELLS

Norman Dovichi^{1 1}University of Washington, Chemistry Department, Seattle, Washington

It is now possible to generate one- and two-dimensional protein electrophoresis data from single mammalian cells. These separations are capable of resolving a hundred components from the cell, and the data can be correlated with cell cycle and other properties of the cells. Examples have been generated from breast, prostate, colon, and esophageal cancer cell lines, from macrophages, astrocytes and neurons, and from osteoprecursors and myoblasts. Instrumentation is under development that can characterize tens of thousands of cells per day in one-dimensional electrophoresis and thousands of cells per day in two-dimensional electrophoresis. Other instrumentation is able to detect specific proteins at the single copy level in single cells and determine post-translational modifications of that protein. Finally, metabollic pathways for both carbohydrates and glycolipids have been monitored in single cells.

11 MULTIMODE LIVE CELL IMAGING REVEALS A NOVEL METHOD OF CELLULAR COMMUNICATION IN THE IMMUNE SYSTEM

Simon C. Watkins^{1 1}University of Pittsburgh, Center for Biologic Imaging, School of Medicine, Pittsburgh, Pennsylvania

Engagement of cell surface receptors leads to intracellular signaling that has generally been shown to alter phenotype and function of individual cells. Using a variety of cutting edge optical imaging methods We show here that myeloid-lineage dendritic cells and monocytes can be triggered to flux calcium by mechanical contact with a microprobe, and that the signal can be propagated within seconds to other cells at distances up to a hundred microns away by membranous connections called tunneling nanotubules (TNTs). These form a complex and transient network in live cells, with individual TNTs exhibiting great variation in length and diameter. In addition to calcium fluxes, microinjected dye tracers can be transferred through these connections. Following TNT-mediated stimulation, spreading of lamellipodia occurs in dendritic cells characteristic of that seen during the phagocytic response to bacteria. These results demonstrate that functional signals generated in a single cell can be transmitted to other cells through a physically connected network.

12 FLUORESCENT SPECKLE MICROSCOPY

Gaudenz Danuser^{1 1}The Scripps Research Institute, Department of Cell Biology, La Jolla, California

Fluorescent speckle microscopy (FSM) is a method to analyze the movement and assembly/disassembly dynamics of macromolecular structures. It capitalizes on fluorescent analog cytochemistry, in which purified protein is covalently linked to a fluorophore and microinjected or expressed as a GFP fusion in cells. Fluorescent protein coassembles with unlabeled, endogeneous protein, visualizing the localization and architectural organization of the structure inside a living cell. Despite its broad application, this classic approach has been limited in reporting subcellular dynamics because of the inherently high background fluorescence from unincorporated and out-of-focus incorporated fluorescent subunits and the uniform labeling of the target structure. However, if the labeled subunits make up less than 1% of total pool of subunits, fluorophore patterns of high non-uniformity accumulate. They provide a unique random code identifying specific subareas of the target structure in space and time. Addition of new fluorophores to the code indicates the local association of subunits; subtraction of fluorophores the local dissociation. These molecular events can be monitored by diffraction-limited imaging using wide-field, total internal reflection, or spinning disc confocal fluorescence microscopy equipped with sensitive CCD cameras. The resulting images display punctate patterns of local maxima, called speckles, over a dimmer background.

17 HIGH CONTENT ANALYSIS IN DRUG DISCOVERY

Joseph Trask^{1 1}Abbott Laboratories, Target and Lead Discovery, Abbott Park, Illinois

This talk will describe past, current and future trends of the use of high-content analysis (HCA) and high-content screening (HCS) in the biopharmaceutical industry and the crossover of the technology into the academic arena. The term “HCA/HCS” typically describes automated fluorescent imaging, image analysis, data management and the applications thereof. The technology not only complements flow cytometry technology, it resembles flow cytometry during its infancy with many challenges and a rapidly changing future. The impact of HCA/HCS technology in the biopharmaceutical industry is being felt throughout the entire drug development process from early drug discovery, target identification, target validation, screening through preclinical biomarker discovery including toxicology and mechanism of action studies. The technology has cross-pollinated into many areas of science including basic science and even material sciences. The goal of the talk is to provide the audience with a brief history of HCA/HCS and case studies where the technology has made an impact.

18 RARE EVENT DETECTION IN SOLID CANCERS

Alison L. Allan¹, Michael Keeney^{2 1}University of Western Ontario, Oncology, Schulich School of Medicine & Dentistry, London, Ontario, Canada; ²London Health Sciences Centre, Hematology/Flow Cytometry, London, Ontario, Canada

Given the multi-step nature of cancer development, there should be several opportunities for therapeutic targeting of tumor cells and/or the tumor microenvironment. The ideal way to identify and monitor disease progression is through surrogate marker approaches that are minimally invasive and allow for longitudinal testing, such as blood tests. Our current research focuses on the development of such approaches, in particular rare event detection by image and flow cytometric methods. Identifying rare populations requires a different approach than standard cytometry techniques, which rely mainly on positive and negative decisions made in either one, or at most, two dimensional space. Recent interest in identification and quantification of rare cell populations such as circulating epithelial tumor cells and circulating endothelial cells in cancer patients has pushed the limits of detection even further. Flow and image cytometric methods are now being developed to identify cellular populations with frequencies as low as 1-20 cells/ml. This presentation will discuss the issues that must be addressed when designing an assay to accurately detect rare populations of cells in blood or bone marrow, with emphasis on detection of circulating endothelial cells and circulating tumor cells in patients with solid tumors and in experimental mouse models of cancer.

19 IMAGE-BASED SCREEN FOR CELL CYCLE AND CANCER TARGETS

Daniel R. Rines^{1 1}Genomics Institute of the Novartis Research Foundation (GNF), GNF Bio-Imaging Center, San Diego, California

Chromosome segregation during mitosis depends on the proper function of specialized structural and cytoskeletal machinery. The duplicated chromosomes are separated equally to daughter cells by the highly dynamic fibers of the mitotic spindle called microtubules. The spindle consists of a bipolar array of microtubules where the extreme ends of the spindle are each anchored to a centrosome. Kinetochores are large protein complexes that assemble onto centromeric DNA sequences and physically attach the replicated chromosomes to the spindle fibers. Ultimately, the maintenance of genomic integrity depends on the proper attachment of chromosomes to the spindle and on the generation of opposing tensile forces that pull the chromosomes apart. Failure in either of these processes leads to unequal partitioning of the genome. However, little is known about how chromatid-microtubule attachment is mediated, or how opposing forces are generated. Currently, two anticancer agents, paclitaxel (taxol) and camptothecin, are used in the treatment of various forms of the disease. Taxol promotes irreversible polymerization of microtubules, disrupting their inherent dynamic nature. Camptothecin functions by inhibiting topoisomerase I activity and leads to large scale chromosome breakage when opposing tensile forces are applied. The diverse action of these two anti-mitotic compounds and the mechanical complexity of the segregation process suggest that many protein components are involved. In fact, recent studies in more genetically tractable organisms such as the budding yeast, S. cerevisiae, have identified over 50 proteins involved in the chromatid-microtubule attachment process alone and many of the functional orthologues have yet to be identified in humans. Using a RNA library of 49,000 double-stranded (ds)RNA targeting approximately 24,000 genes, we performed a loss-of-function screen for essential mitotic chromosome segregation genes. We identified novel genes whose inactivation caused mitotic arrest. Multi-parametric analysis of image-based data derived from a high-content screen including phospho-histone H3 levels, cellular proliferation and nuclear morphology allowed us to isolate both checkpoint and independent segregation genes.

20 A HUMAN PROTEIN ATLAS FOR NORMAL AND CANCER TISSUES

Mathias Uhlen¹, Fredrik Ponten^{2 1}Stockholm, Sweden; ²Uppsala University, Uppsala, , Sweden

Antibody-based proteomics provides a powerful approach for the functional study of the human proteome involving the systematic generation of protein-specific affinity reagents. We have used this strategy to construct a comprehensive, antibody-based protein atlas for expression and localization profiles in 48 normal human tissues and 20 different cancers (1). The Human Protein Atlas is publicly available (www.proteinatlas.org) and contains, in the first version, approximately 400,000 high-resolution images corresponding to more than 700 antibodies towards human proteins. Each image has been annotated by certified pathologists to provide a knowledge base for functional studies and to allow queries about protein profiles in normal and disease tissues (2). Our results suggest it should be possible to extend this analysis to the majority of all human proteins thus providing a valuable tool for medical and biological research, in particular for biomarker analysis in various patient cohorts. Reference: (1) Uhlen M, Ponten F. (2005) Antibody-based Proteomics for Human Tissue Profiling. Mol Cell Proteomics 4(4): 384-393, (2) Uhlen et al (2005) A human protein atlas for normal and cancer tissues, Mol Cell Proteomics, 4(12):1920-1932.

21 FLOW AND IMAGE CYTOMETRIC FRET FOR DETECTING PROTEIN ASSOCIATIONS

János Szöllõsi¹, György Vereb¹, Gábor Horváth¹, Péter Nagy^{1 1}University of Debrecen, Department of Biophysics and Cell Biology, Medical and Health Science Center, Debrecen, , Hungary

Supramolecular organization of biomolecules at the cell surface or inside the cell has an important role in determining the function and integrity of cells. Specific techniques are available now for detecting molecular proximity and interactions in cells, such as flow or image cytometric variations of fluorescence resonance energy transfer (FRET). Flow cytometric techniques offer the advantage of rapid analysis on a large number of cells (~105 cells in some minutes) with a high statistical accuracy and a possibility for analyzing heterogeneity at the population level. Flow cytometry, however, does not provide any information about the spatial localization of fluorescent probes, but instead measures the fluorescence intensity averaged over each cell. In contrast, microscopic techniques provide a high spatial resolution: conventional fluorescence microscopies have a ~250 nm resolution limited by diffraction of the optics. Although microscopies have several further advantages in detecting molecular dynamics of changes in the distribution or intensity of fluorescent probes, they suffer from a low statistical reliability, especially in the case of quantitative measurements. Thus, a combined application of flow and image cytometry in resolving particular biological questions can be a very powerful approach. In flow cytometry we applied fluorescent probes with longer wavelength excitation and multiple wavelength detection in the emission regions so that autofluorescence correction could be performed on a cell by cell basis in FRET analysis. These facts improved the accuracy of the FRET method and cells with low receptor expression, such as HPB-ALL cells transfected with various CD45 isoforms were amenable to FRET investigation. Combination of various forms of flow and image cytometric FRET methods revealed distinctive expression and association pattern of ErbB receptor tyrosine kinases on the surface of various cancer cell lines sensitive or resistant to trastuzumab (Herceptin®). Simultaneous application of image cytometric FRET methods based on donor and acceptor photobleaching provided a useful dual FRET approach revealing a unique coassociation pattern of integrins, CD44 and ErbB2 on the surface of tumor cells. By measuring the distances between various monoclonal antibody epitopes on ErbB2 molecules and the distances between epitopes and the cell membranes useful information was provided for positioning the extracellular domain in molecular modeling the nearly full length ErbB2 dimer. In this model favorable dimerization interactions were predicted for the extracellular, transmembrane and protein kinase domains, which may act in coordinated fashion in ErbB2 homodimerization, and also in heterodimers of ErbB2 with other members of ErbB family.

24 QUANTITATIVE LIVE IMAGING DESCRIBES MORPHOGENETIC NUCLEAR MOVEMENTS IN EARLY DROSOPHILA EMBRYO

Cris Luengo¹, Soile Keränen², Charless Fowlkes³, Gunther Weber⁴, Min-Yu Huang⁴, Oliver Rübel⁴, Bernd Hamann⁴, Damir Sudar¹, Jitendra Malik³, Mark D. Biggin², David W. Knowles^{1 1}Lawrence Berkeley National Laboratory, Life Sciences Division, Berkeley, California; ²Lawrence Berkeley National Laboratory, Genomics Division, Berkeley, California; ³University of California, Berkeley, Computer Science Division, Berkeley, California; ⁴University of California, Davis, Department of Computer Science, Davis, California

The Berkeley Drosophila Transcription Network Project (bdtnp.lbl.gov) is conducting a system-wide analysis of the transcription network in the early Drosophila embryo. As part of this multidisciplinary effort, novel imaging, image analysis and visualization methods have been developed to construct the first three-dimensional (3D) atlas of gene expression and morphology in an embryo at cellular resolution. Our aim is to quantify the relative expression of hundreds of genes in wild type embryos and in a series of mutant embryos, and to map these results onto “stereotypical embryos”. Multiple-color in-situ hybridizations are used to fluorescently label gene products of interest, and total DNA is counter-stained. High resolution, multi-channel, 3D images are acquired of entire embryos using two-photon excitation. Individual nuclei are isolated from the DNA-stained images using novel, automated segmentation techniques, and the relative gene expression within and around each nucleus is then quantified. Novel techniques have been developed to register data from many images onto a “stereotypical embryo” and create exquisite quantitative visualizations of the data in 3D. One novel observation we have made is the systematic change in nuclear packing densities during stage 5 (interphase cycle 14, up to gastrulation), which can be seen by comparing fixed embryos of different ages. To understand these nuclear movements, we have studied the process in live histone2A-GFP embryos. Individual embryos were imaged from the 13th cleavage cycle to the start of gastrulation in approximately 3 minute intervals. Rapid temporal sequences were important for tracking individual nuclei but restricted the imaging to the portion of the embryo closest to the objective lens. Orientation of the imaged embryo portions were determined by morphological features during gastrulation and egg-length was determined from optical sections taken through the middle of the embryo. These measurements allowed images from multiple embryos to be combined into whole-embryomaps. The resulting patterns of nuclear packing density and flow-fields correlate with the dorsal-ventral orientation of the embryo and the future location of the ventral furrow. These results, which support our fixed embryo work, show complex 3D nuclear movements prior to gastrulation that will have to be considered to fully understand dynamics in gene expression patterns.

25 ACQUIRING MULITPLE IMAGES OF LARGE TISSUE SECTIONS IN BOTH FLUORESCENCE AND TRANSMITTED LIGHT USING THE LASER SCANNING CONFOCAL TISSUESCOPE

Trudey Nicklee¹, David Hedley^{2 1}Toronto, Ontario, Canada; ²Princess Margaret Hospital, Toronto, Ontario, Canada

Large tissue sections that overfill a microscope 10x objective field of view typically must be imaged using a tiling method. In tiling, sequential fields of view are acquired as separate image files. This is time consuming, especially when algorithms are employed to align edges of individual fields. Furthermore, depending on the quality of the stage movement and algorithms used, artifacts may be introduced from misaligned fields. The Biomedical Photometrics TISSUEscope (Waterloo, Ontario; confocal.com) is a laser confocal scanning microscope, able to acquire a low resolution image of an entire microscope slide as a single continuous 20 mm x 70 mm field in 30 seconds. Large subsections of interest can then be imaged at selected resolution to 1µm per pixel. The instrument is equipped with blue, green and red lasers, and can simultaneous acquire images from two lasers. Software allows for the same selected field of view to be repeatedly scanned with different excitation and emission parameters. Therefore, multiple fluorescent images of this same field of view are acquired aligned. This allows not only for quantification of several markers in a single section, but also co-localization studies of these markers. As well as allowing rapid image acquisition compared to tiled field imaging at comparable resolution using a conventional fluorescence microscope, the system also incorporates transmission detectors that allow RGB imaging based on absorbance from the three lasers. A fluorescently stained slide can therefore be imaged, then restained and imaged for transmitted light using the same optical system. This allows selected areas of interest to be identified by transmitted light, then directly linked to the fluorescence images. Wide field multispectral analysis of histological sections is a powerful technique for studying complex molecular processes at the intact tissue level. The laser scanning TISSUEscope is a novel instrument that offers several advantages that we are currently evaluating. Applications include the analysis of signaling pathways in sequential biopsies obtained from patients in clinical trials of molecular cancer therapeutics, and studies that address the problems of intratumoral heterogeneity.

26 COMPARISON OF FLUORESCENTLY AND CHROMATIC LABELED TISSUE MICROARRAYS ANALYZED BY LASER SCANNING CYTOMETRY

Ed Luther^{1 1}CompuCyte Corporation, Cambridge, Massachusetts

Tissue microarrays (TMAs) are becoming invaluable tools in biomarker development. In developing automated analysis systems, the choice between fluorescent or chromatic staining techniques is a primary consideration. We have performed an evaluation comparing the two staining techniques. Materials and Methods: TMAs consisting of 121 elements (0.5 mm cores) were stained with either fluorescent or chromatic dyes and then submitted for blind analysis on a laser scanning cytometry platform. Control samples and breast cancer tissues were included. Serial sections of the arrays were stained with antibodies against HER2/neu protein and counterstained for DNA, either with chromatic reagents (DAB counterstained with hematoxylin) or fluorescent reagents (Alexa Fluor 488 counterstained with propidium iodide). The slides were analyzed on an iCyte® Automated Imaging Cytometer. High-speed scans were used to locate individual core elements within the array. High-resolution (0.25 micron) scans were then performed on each element to obtain either fluorescence or laser light absorption signals. Spectral deconvolution was used to separate the HER2 and DNA staining. In the chromatically stained sections, compensation techniques were applied for tissue autofluorescence. Scanned images were segmented using both nuclear segmentation and random sampling analysis. Results: In both TMA sets, the nuclear segmentation and random segmentation techniques (chromatic R2 = 0.9017, fluorescence R2 = 0.8048) correlated well. The correlation between the values from the chromatic and fluorescent labeled slides was less satisfactory—R2 = 0.6595. To investigate the discordance, analysis regions were defined around core elements falling into the following categories, laser scan images of representative elements were analyzed, and the staining patterns of the cores were confirmed. Category Chromatic Fluorescent Diagnosis 1 ^{+ +} HER2/neu positive 2 ⁺ -HER2/neu positive – below fluorescence threshold 3

⁺ Autofluorescence artifacts 4 --Negative Conclusions: Our analysis showed greater sensitivity for TMAs stained with chromatic dyes than for those with fluorescent dyes. This is contradictory to the current belief that fluorescent dyes have greater quantitative capabilities than chromatic IHC dyes, and is probably due to complications from autofluorescence of the tissues. The ability to analyze both chromatic and fluorescent TMAs on the same instrument platform allows investigators great flexibility in selecting an appropriate staining technology - fluorescence, chromatic or both.

30 TWO AND THREE DIMENSIONAL SEGMENTATION OF WHOLE CELLS AND CELL NUCLEI IN TISSUE

Dean P McCullough¹, Daniel Baggett², Stephen J. Lockett^{3 1}National Cancer Institute,Advanced Biomedical Computing Center, Frederick, Maryland; ²Worcester Polytechnic Institute, Mechanical Engineering, Worcester, Massachusetts; ³NCI / SAIC-Frederick, Frederick, Maryland

Communications between neighboring cells in large part drive tissue development and function, as well as disease-related processes such as tumorigenesis. In order to understand the molecular basis of these processes, it is necessary to quantitatively analyze specific molecules in adjacent individual cells or cell nuclei of the intact tissue. A major bottleneck preventing widespread use of such analyses is the lack of an efficient method that assures correct segmentation of all individual, whole cells in a given intact tissue volume of interest from 3D images. Consequently, we have developed software for identifying the optimum border around each individual cell or cell nucleus, a process known as segmentation, from 2D and 3D microscope images of intact tissue labeled with a fluorescent cell or nuclear surface marker. In the 2D case the optimum border was defined as the border that has an average intensity per unit length greater that any other possible border around the same cell or nucleus. Implementation of the algorithm required the user to indicate two points for each cell, one inside the cell and the other on the border. Thereafter segmentation was automatic and was peformed using dynamic programming. The method correctly detected virtually 100% of cells, because determination of the optimum path is not significantly affected by intermittent labeling of the cell borders, by diffuse borders, or by spurious signals away from the borders. It also contains interactive tools, which allows subsequent visualization and correction of the image. The method is also highly efficient due to only minimal user interaction. We have extended this method to 3D segmentation, which begins with 2D segmentation in a user-selected plane approximately through the center of the cell. Then the automatic algorithm separately finds the two surfaces of the cell in the planes above and below the user selected plane using dynamic programming. The algorithm does not find the true optimal surface since this is too computationally expensive, but closely approximates the optimum by finding a succession of partially optimal surfaces using the greedy algorithm with a look ahead of n steps. Following segmentation, the user may add points that are required to be on the surface to correct any perceived errors. The algorithm has been tested on a wide variety of biological tissue samples and will segment moderately irregularly shaped cells containing concavities. Work performed under contract grant sponsor: NCI/NIH; Contract #: NO1-CO56000.

31 A GENERAL TECHNIQUE FOR SEGMENTATION OF INDIVIDUAL CELLS IN LIGHT MICROGRAPHS

Zachary Pincus¹, Julie A Theriot^{2 1}Stanford University, Biomedical Informatics, School of Medicine, Stanford, California; ²Stanford University, Biochemistry, School of Medicine, Stanford, California

Many ad hoc methods exist for separating cells from the image background (and, with less success, from each other) in light micrographs. Here we present a general method that is applicable to many cell types and imaging modalities. A persistent difficulty with cell segmentation tools is that assumptions about cell shape, imaging conditions, and cell packing are “hard-coded” into the cell-finding logic. We instead make these assumptions explicit and modular. By using a parameterized “shape model” learned from training examples, our tool makes no implicit assumptions about cell shape. Performing the principal components analysis on cell outlines provided by the user produces a set of linearly independent modes of shape variation (e.g. large vs. small, round vs. elongated, etc.). These shape modes can then be recombined to generate candidate cells from a distribution of expected shapes. Because the only input required is a small set of shapes, it is easy to reconfigure this module for different cell types. We free our method from assumptions about the relationship between pixel values and the presence or absence of cells by converting images into a canonical form. Specifically, we use machine learning tools to convert images into “probability maps” where the value of each pixel is the likelihood that that pixel is inside of a cell. Simple k-nearest-neighbor classification of pixels based on local texture statistics is sufficient to transform images of many different types (e.g. phase contrast, DIC, epifluorescence) into simple-to-interpret probability maps. The only required inputs are a few image regions which have been manually labeled as “cells” or “background;” thus this module is also easy to reconfigure for different imaging conditions. To find a single cell in an image, a shape model is iteratively fit to a probability map. We numerically optimize the parameters of a given shape model (creating different candidate shapes) and its pose (x- and y-position and rotation) to maximize a goodness-of-fit function which evaluates how well the model fits to the probability map. This function evaluates the prior likelihood of the candidate shape, the distance between candidate shape edges and image edges, and the fraction of pixels with low likelihood of being in a cell that are nevertheless inside the candidate shape. We then refine the initial fit by deforming the shape to better fit the image within a level-set framework. This process is repeated to find all cells in an image. This method has shown human-competitive results in separating out individual cells from the background and from each other on many different image and cell types, including fixed S2 cell cultures, growing bacterial mats, and membrane-stained Drosophila epithelia.

32 FILO: AN UNBIASED SPATIAL ANALYSIS OF FISH SIGNALS IN INTERPHASE NUCLEI

Prabhakar Reddy Gudla¹, Addison Z Yee²