Portal de Periódicos da CAPES

Imaging in Systems Biology

2007; Cell Press; Volume: 130; Issue: 5 Linguagem: Inglês

10.1016/j.cell.2007.08.031

1097-4172

Sean G. Megason, Scott E. Fraser,

Single-cell and spatial transcriptomics

Most systems biology approaches involve determining the structure of biological circuits using genomewide "-omic" analyses. Yet imaging offers the unique advantage of watching biological circuits function over time at single-cell resolution in the intact animal. Here, we discuss the power of integrating imaging tools with more conventional -omic approaches to analyze the biological circuits of microorganisms, plants, and animals. Most systems biology approaches involve determining the structure of biological circuits using genomewide "-omic" analyses. Yet imaging offers the unique advantage of watching biological circuits function over time at single-cell resolution in the intact animal. Here, we discuss the power of integrating imaging tools with more conventional -omic approaches to analyze the biological circuits of microorganisms, plants, and animals. Although we have nearly complete genome sequences for most model systems, we are very far from understanding how this genomic code is executed during development to build an organism. Implicit in the genome sequence of an organism is a complete set of instructions for constructing that organism (aside from epigenetic factors). The problem is that we as scientists have no way of directly deciphering the code. For example, we cannot predict a protein's function based on its sequence, we cannot predict when and where a protein will be expressed based on a gene's noncoding sequence, and we cannot predict the effect of removing a gene's function on the system as a whole. Currently, the only code that we can easily decipher is the parts list. Thus, the big challenge ahead is to answer the related questions with respect to the underlying genetic regulatory network: How do these parts interact as a system, and how does this system function to create an organism? We call this exciting area of research in the postgenomic era "systems biology." Because systems biology is still young, its boundaries remain fluid and its practitioners use a number of different tools from a variety of different perspectives. The birth of the field was made possible by genomics and other high-throughput approaches such as proteomics and microarrays. Not surprisingly, these -omic approaches are still important tools of systems biology and have proven invaluable for identifying and characterizing the components of biological systems on a comprehensive scale. Recently, these approaches have been refined to identify possible interactions among components, such as protein-protein interactions and protein-DNA cis-regulatory interactions. Yeast two-hybrid, biochemical pull downs, protein microarrays, and chromatin immunoprecipitation have been used to characterize the network of interactions on a large scale in several organisms (Cusick et al., 2005Cusick M.E. Klitgord N. Vidal M. Hill D.E. Interactome: gateway into systems biology.Hum. Mol. Genet. 2005; 14: R171-R181Crossref PubMed Scopus (297) Google Scholar, Harbison et al., 2004Harbison C.T. Gordon D.B. Lee T.I. Rinaldi N.J. Macisaac K.D. Danford T.W. Hannett N.M. Tagne J.B. Reynolds D.B. Yoo J. et al.Transcriptional regulatory code of a eukaryotic genome.Nature. 2004; 431: 99-104Crossref PubMed Scopus (1670) Google Scholar). These approaches are providing a rough draft of the components and interactions that comprise biological networks on an unprecedented scale. However, even the best rough drafts are relatively coarse, error prone, and uninformed by biological dynamics. Because of this, current high-throughput data provides only a starting point from which a number of hypothesis-driven "wet" experiments must be done to determine the function of a system. Recent work on the elaboration and testing of gene regulatory networks (Levine and Davidson, 2005Levine M. Davidson E.H. Gene regulatory networks for development.Proc. Natl. Acad. Sci. USA. 2005; 102: 4936-4942Crossref PubMed Scopus (481) Google Scholar) offer ample evidence of the hard work needed to validate direct cis-regulatory interactions among even a small set of target genes involved in a circumscribed phase of development. We argue here that imaging can play a vital role in systems biology, offering a path from rough static models to more refined, quantitative dynamic models. In vivo imaging can capture quantitative data at single-cell resolution and do so noninvasively as the biological circuit functions, offering insights that cannot be matched using in vitro approaches. With the emergence of automated instrumentation and advanced analysis tools, such intravital imaging has become practical for both hypothesis-driven research and high-throughput discovery science. In hypothesis-driven research, for example, multicolor imaging can be used to monitor the essential nodes of a biological circuit in real time. In high-throughput research, standardized reporters and imaging conditions permit intravital imaging to capture data approaching the same large scale as today's -omic approaches but with improved temporal and spatial resolution. We begin this review by comparing the relative roles for -omics and imaging in systems biology, then we discuss the unique requirements for applying imaging to systems biology, and finally we conclude by offering examples of imaging in systems biology. Conventional -omic approaches excel at the first goal of systems biology: characterizing the structure of biological networks. This requires the identification of all the components of the network as well as the identification of all the interactions between them. Systems level analysis works best when all the components and interactions are defined, as the lack of only a few can cause a model to fail. For this reason, comprehensive approaches such as sequencing, microarrays, and interactomic approaches are ideal. Genetics can help to identify the components and connections of a network, but redundancy and pleiotropy can mask important links. Once the structure of a network has been elucidated, how it functions as a circuit can be investigated. The advantages of imaging approaches in probing how biological circuits function (Figure 1) are presented below. In studies of biological circuit function, single-cell resolution is a key advantage (Figure 1A). In addition to being the basic building blocks of organisms, cells are the computational unit in biological circuits. Most components of regulatory networks, ranging from signal-transduction components, transcription factors, and cis-regulatory elements act within single cells. Although receptor-ligand interactions and secreted signaling proteins can influence distant cells, the interpretation of the signaling is executed within single cells. Thus the network guiding the development of a multicellular organism is best thought of as a network of networks—many intracellular networks linked by a few intercellular interactions. The "output" of many biological circuits is also at the cellular level such as changes in fate, proliferation, apoptosis, or cell shape. Most of these phenomena make little sense at either higher (organismal) or lower (molecular) levels or averaged over a population of cells (a cell cannot be partially dead or partially differentiated). Averaging fails even for populations of cells that at one time appear homogeneous, as genetically identical cells can have significant differences in the expression of components, their connections, and functional states that drastically alter the function of a biological circuit. Imaging can readily provide subcellular resolution. It is difficult to achieve single-cell resolution with most -omic approaches. Microarrays typically require material from thousands of cells to achieve sufficient signal. Single cells can be analyzed by amplifying the sample, but this might introduce artifacts into the data (Subkhankulova and Livesey, 2006Subkhankulova T. Livesey F.J. Comparative evaluation of linear and exponential amplification techniques for expression profiling at the single-cell level.Genome Biol. 2006; 7: R18Crossref PubMed Scopus (48) Google Scholar). It is difficult to accurately and reproducibly isolate single cells of a defined stage and position for -omic analysis from a field of cells, and even if successful, the very act of removing the cell can change its context. Because biological circuits function over time (Figure 1B), techniques are needed to assess the temporal dynamics of the circuit. Yet, most -omic and other biochemical approaches begin by homogenizing the sample, which disrupts the function of the circuit. A time course can be deduced by collecting samples at multiple time points. For example, microarray analysis of the yeast cell cycle has been accomplished by taking aliquots of a culture that was synchronized (through the use of α-factor, size fractionation, or a temperature sensitive cell-cycle mutant [Spellman et al., 1998Spellman P.T. Sherlock G. Zhang M.Q. Iyer V.R. Anders K. Eisen M.B. Brown P.O. Botstein D. Futcher B. Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization.Mol. Biol. Cell. 1998; 9: 3273-3297Crossref PubMed Scopus (3789) Google Scholar]). However, such approaches provide population averages that can obscure any fast or variable features of the circuit in the individual cells. Thus, there is simply no way to get adequate longitudinal data using a technique that requires the sample to be destroyed in order to be measured. Imaging is well suited to collecting longitudinal data because it can be done noninvasively on intact, fully functioning organisms. Time-lapse, fluorescent microscopy can monitor the status of a circuit in each cell of a population repeatedly for hours. Furthermore, imaging can match the timescale of biological computation, which can range from seconds for small molecule signaling (such as Ca2+ or cAMP) to hours for protein-based signaling (such as a morphogen-inducing target gene expression). Rather than being simple on-off switches, biological circuits can show graded responses (Figure 1C). Thus, it is critical to collect quantitative information about the circuit's components (such as concentration, localization, and posttranslational modification) to accurately model the function of a circuit. A limitation of many -omic approaches is that they do not accurately measure the quantity of a component. For example, microarrays do not directly provide a measure of quantity but instead provide a relative measure of concentration by comparing the binding of two samples, either through direct competitive hybridization to a single array or by comparing the results from two singly hybridized microarrays. Great effort has gone into developing statistical measures to help interpret such differential microarray measurements (Breitling, 2006Breitling R. Biological microarray interpretation: the rules of engagement.Biochim. Biophys. Acta. 2006; 1759: 319-327Crossref PubMed Scopus (41) Google Scholar). Yet, whether these measurements can be considered quantitative is widely debated in the literature (see Tan et al., 2003Tan P.K. Downey T.J. Spitznagel Jr., E.L. Xu P. Fu D. Dimitrov D.S. Lempicki R.A. Raaka B.M. Cam M.C. Evaluation of gene expression measurements from commercial microarray platforms.Nucleic Acids Res. 2003; 31: 5676-5684Crossref PubMed Scopus (606) Google Scholar, Miklos and Maleszka, 2004Miklos G.L. Maleszka R. Microarray reality checks in the context of a complex disease.Nat. Biotechnol. 2004; 22: 615-621Crossref PubMed Scopus (141) Google Scholar, Frantz, 2005Frantz S. An array of problems.Nat. Rev. Drug Discov. 2005; 4: 362-363Crossref PubMed Scopus (49) Google Scholar, MAQC Consortium, 2006MAQC ConsortiumThe MicroArray quality control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements.Nat. Biotechnol. 2006; 24: 1151-1161Crossref PubMed Scopus (1627) Google Scholar and references therein). Another limitation of microarrays is that they measure RNA, although proteins do most of the work in biological circuits and it is thus protein concentration that matters most for modeling. Differences in the rates of translation, posttranslational modification, and protein stability limit the correlation between RNA and protein levels (r ∼ 0.4–0.6). The correlation can vary wildly with the cell type, developmental stage, and category of proteins (Gygi et al., 1999Gygi S.P. Rochon Y. Franza B.R. Aebersold R. Correlation between protein and mRNA abundance in yeast.Mol. Cell. Biol. 1999; 19: 1720-1730Crossref PubMed Scopus (3077) Google Scholar, Greenbaum et al., 2003Greenbaum D. Colangelo C. Williams K. Gerstein M. Comparing protein abundance and mRNA expression levels on a genomic scale.Genome Biol. 2003; 4: 117Crossref PubMed Scopus (1173) Google Scholar). Protein and antibody microarrays could address this issue (Kung and Snyder, 2006Kung L.A. Snyder M. Proteome chips for whole-organism assays.Nat. Rev. Mol. Cell Biol. 2006; 7: 617-622Crossref PubMed Scopus (65) Google Scholar), but the increased difficulty of working with proteins has made it challenging to develop a robust, specific, and general platform for assaying protein concentration. Fluorescent imaging has the ability to obtain both qualitative and quantitative data. A fluorescent microscope can serve as a microspectrophotometer, measuring the concentration and distribution of fluorophores inside a cell. Unlike an RNA molecule in a complex mixture hybridizing to a probe on a microarray, there exists a simple linear relationship linking fluorescence and concentration given by the fluorescent dye's extinction coefficient and quantum yield. For example, Wu and Pollard, 2005Wu J.Q. Pollard T.D. Counting cytokinesis proteins globally and locally in fission yeast.Science. 2005; 310: 310-314Crossref PubMed Scopus (396) Google Scholar constructed yeast with yellow fluorescent protein (YFP) fusions to a number of different cytoskeletal proteins expressed from their endogenous genetic loci and showed a very strong linear correlation (r = 0.99) between fluorescence and quantitative western blots across a range of expression levels. Thus, the correlation between image intensity and actual number of protein molecules can be much better than that for microarray replicates. A final consideration in analyzing biological circuits is that anatomy matters (Figure 1D). The structure and function of biological circuits varies from tissue to tissue within an organism, from cell to cell within a tissue, and even between subcellular compartments. Indeed, it is these spatial differences in the function of biological circuits that make different parts of an organism unique. For example, a morphogen diffusing across a field of cells may cause differences in the circuitry depending on the distance from the source. The problem with -omic and other biochemical approaches is that the first step in the procedure is to "homogenize" the sample, destroying the anatomical context of the data, which can blur spatial differences. Imaging has a unique ability to capture data at all these spatial ranges from macro to nano from anatomically intact specimens. In summary, -omics and imaging can provide insights into many of the same problems, but they do so from very different angles. Typical -omic approaches provide high genomic resolution but low spatial and temporal resolution; in contrast, imaging provides high spatial and temporal resolution but low genomic resolution. To help compare these approaches, it is useful consider the hypothetical "xyztg molecular data universe." Such a five-dimensional data set could describe the expression patterns and subcellular localization patterns for all proteins in the genome (g) at all developmental times (t) and places (xyz) in three dimensions in an organism. Approaches based on -omics and imaging take slices though this data universe in different directions, as depicted in the cartoon in Figure 2. For example, using in toto imaging of a green fluorescent protein (GFP) transgenic organism as described below, it is possible to acquire a 4D (xyzt) slice through the data universe for a single gene. In contrast, microarrays can acquire data across the whole genome but for only a single time and place per array—that is, a 1D genomic slice through many genes but for a single point of x, y, z, and t. Compared with -omic techniques, imaging can acquire functional genomic data in higher throughput and in a format more amenable to integration. As an example, imagine a data set consisting of the expression pattern of all genes during embryonic development of zebrafish at high enough resolution for use in systems modeling. In rough numbers, such a five-dimensional molecular data universe would contain 25,000 points across the genomic dimension (one for each gene); 1,000 points across the time axis of development (one every 5 min); and 100 points across each of the spatial dimensions of x, y, and z. Time-lapse imaging can capture a 4D slice (xyzt) of this universe for a single gene, whereas -omic methods can only capture a 1D slice at a single point in x, y, z, and t. Which transect through the data universe is better for systems biology? Consider throughput, an area in which imaging in typically thought of as low throughput but high content. We believe that the high content can dominate in interrogating the data universe efficiently. Consider what it would take to actually measure the xyztg molecular data universe in zebrafish embryogenesis, where there are roughly 25,000 genes; more than 100 points across each spatial dimension; and 1,000 points across time (roughly cellular spatial resolution and a few minutes of temporal resolution). Because imaging a GFP reporter gene can acquire an entire xyzt slice for a single gene per experiment, it would take 25,000 experiments to capture a complete data set. This is a lot of experiments, to be sure, but not infeasible as the quality, control, and price of microscopes improve. In contrast, -omic techniques could capture the whole genome per experiment but for only a single point in the data cube. To get the same spatial and temporal resolution as 25,000 imaging experiments, it could take as many as 1,000,000,000 (100 x × 100 y × 100 z × 1000 t) -omic experiments. Although this calculation is an exaggeration, as different cells and time points might be pooled for the -omics analysis, it points to the size of the challenge. Even if this were to be optimized, there is the issue of data continuity. It would be very difficult to integrate the 1D genomic data across all the dimensions of x, y, z, and t. In contrast, it would be much easier to superimpose the 4D xyzt data generated for each gene by in toto imaging. It is unlikely that many researchers will attempt to acquire data on this scale, but similar considerations of resolution, throughput, and continuity apply to more typical, hypothesis-driven experiments. We believe that for many questions in systems biology such as understanding the function of a particular biological circuit, the high spatial-temporal resolution from imaging data for a handful of genes is more powerful than the high genomic resolution of -omic approaches for a handful of spatiotemporal points. Although imaging is a widely used technique in biology, its use for systems biology presents a number of challenges. The paradigm shift that started with genomics and continues with systems biology is that data are quantitative, standardized, comprehensive, and (most importantly) digital. Previously, data generated by the scientific community were compartmentalized: they were published in thousands of individual, copyrighted, hard-copy (i.e., analog) journal articles. This made it very difficult to search, reuse, or integrate the huge wealth of data. This was changed by centralized databases that store biological data in a standardized, digital format (Figure 3A). The power of being digital and available is that data can be constantly reused, synthesized, and grown (Negroponte, 1996Negroponte N. Being Digital. Vintage Publishing, New York1996Google Scholar). Imagine the loss of utility if sequence data only existed in the pages of individual journal articles, that there was no GenBank or BLAST. This power is being extended to some other types of data such as microarray data and in situ expression patterns, but most forms of biological data are still published in an analog form. The availability of standardized, reusable, digital data of many types is essential for the progress of systems biology. Thus, to be useful for systems biology, imaging must be able to capture data in a standardized, quantitative, digital form like that we now take for granted with sequence data (Figure 3B). Capturing data in such a form requires careful forethought of a number of special considerations throughout the pipeline of imaging from labeling to image analysis. One of the most important questions regarding image capture is "what to image"? There are a number of methods for generating contrast in optical imaging, but the most important approach by far is fluorescence. Fluorescent imaging permits a number of different channels to be imaged in one specimen and allows a wide range of very specific structures and even species of molecules to be imaged. Small molecule fluorophores can be used to measure pH and ion concentrations (e.g., Ca2+) and can be used to tag specific proteins by immunofluorescence or with the FlAsH/ReAsH system (Gaietta et al., 2002Gaietta G. Deerinck T.J. Adams S.R. Bouwer J. Tour O. Laird D.W. Sosinsky G.E. Tsien R.Y. Ellisman M.H. Multicolor and electron microscopic imaging of connexin trafficking.Science. 2002; 296: 503-507Crossref PubMed Scopus (774) Google Scholar). The most powerful approach to molecular imaging, however, is through the use of fluorescent proteins such as GFP (Shaner et al., 2005Shaner N.C. Steinbach P.A. Tsien R.Y. A guide to choosing fluorescent proteins.Nat. Methods. 2005; 2: 905-909Crossref PubMed Scopus (2155) Google Scholar). The main advantage of fluorescent proteins is that they are genetically encoded. This allows them to be used in intact animals for time-lapse imaging, expressed at endogenous levels for quantitative imaging, and genetically engineered to have a wide array of useful properties such as different colors or functional reporters. Fluorescent proteins can be used to mark a variety of things. Transcriptional reporters in which a specific enhancer/promoter is used to drive expression of a fluorescent protein can be used to mark when and where a gene is expressed or to mark specific cell types. Fusion of a fluorescent protein with a protein of interest can be used to characterize its subcellular localization, which can be very important in defining its function, as many proteins change their localization depending on their functional state (Figure 3C). Fluorescent protein fusions allow this change in functional state of a protein to be monitored noninvasively in living cells (Nelson et al., 2002Nelson G. Paraoan L. Spiller D.G. Wilde G.J. Browne M.A. Djali P.K. Unitt J.F. Sullivan E. Floettmann E. White M.R. Multi-parameter analysis of the kinetics of NF-kappaB signalling and transcription in single living cells.J. Cell Sci. 2002; 115: 1137-1148Crossref PubMed Google Scholar). One concern to keep in mind when working with fluorescent protein transgenics is that its dynamics of degradation may differ from the endogenous protein, especially for transcriptional reporters. There are other types of functional reporters that can be made with fusion proteins (Giepmans et al., 2006Giepmans B.N. Adams S.R. Ellisman M.H. Tsien R.Y. The fluorescent toolbox for assessing protein location and function.Science. 2006; 312: 217-224Crossref PubMed Scopus (2198) Google Scholar). In fluorescent resonant energy transfer (FRET), a photon can be absorbed by one fluorescent protein and nonradiatively transferred to a nearby fluorophore causing a red shift in emission and often a change in the fluorescent lifetime. FRET only occurs if the fluorescent proteins have overlapping emission and excitation spectra, are close together, and in the proper orientation. FRET reporters, consisting of two fluorescent proteins with a linker containing a recognition sequence for a protease, have been used as in vivo reporters for caspase activity (Xu et al., 1998Xu X. Gerard A.L. Huang B.C. Anderson D.C. Payan D.G. Luo Y. Detection of programmed cell death using fluorescence energy transfer.Nucleic Acids Res. 1998; 26: 2034-2035Crossref PubMed Scopus (185) Google Scholar). FRET reporters can also be made in such a way that a conformational switch in the intervening protein sequence will change the spacing or orientation of the two fluorescent proteins. This approach generally requires detailed structural knowledge of the target protein and some trial and error in the placement of the FPs. Such approaches have been used to generate Ca2+ reporters using a calmodulin linker (Miyawaki et al., 1997Miyawaki A. Llopis J. Heim R. McCaffery J.M. Adams J.A. Ikura M. Tsien R.Y. Fluorescent indicators for Ca2+ based on green fluorescent proteins and calmodulin.Nature. 1997; 388: 882-887Crossref PubMed Scopus (2473) Google Scholar) and for the detection of phosphorylation by PKC using a linker containing a phosphorylation site and phosphobinding site (Violin et al., 2003Violin J.D. Zhang J. Tsien R.Y. Newton A.C. A genetically encoded fluorescent reporter reveals oscillatory phosphorylation by protein kinase C.J. Cell Biol. 2003; 161: 899-909Crossref PubMed Scopus (442) Google Scholar). FRET can be used to monitor protein-protein interactions in vivo by fusing one protein with a donor fluorescent protein and the other with an acceptor. This approach has been extended to monitor interactions of 3 proteins using 3-color FRET (Galperin et al., 2004Galperin E. Verkhusha V.V. Sorkin A. Three-chromophore FRET microscopy to analyze multiprotein interactions in living cells.Nat. Methods. 2004; 1: 209-217Crossref PubMed Scopus (159) Google Scholar). It is also possible to construct activity-based reporters using only a single fluorescent protein. Siegel and Isacoff, 1997Siegel M.S. Isacoff E.Y. A genetically encoded optical probe of membrane voltage.Neuron. 1997; 19: 735-741Abstract Full Text Full Text PDF PubMed Scopus (329) Google Scholar inserted GFP into Shaker, a voltage-sensitive potassium channel, in such a way that voltage-dependent rearrangements of the channel would alter the fluorescence of the GFP. A final type of fluorescent protein-based reporter uses bimolecular complementation of "split fluorescent proteins," which by themselves are not fluorescent but regain their fluorescence when the two halves are brought into close proximity (Baird et al., 1999Baird G.S. Zacharias D.A. Tsien R.Y. Circular permutation and receptor insertion within green fluorescent proteins.Proc. Natl. Acad. Sci. USA. 1999; 96: 11241-11246Crossref PubMed Scopus (670) Google Scholar, Nagai et al., 2001Nagai T. Sawano A. Park E.S. Miyawaki A. Circularly permuted green fluorescent proteins engineered to sense Ca2+.Proc. Natl. Acad. Sci. USA. 2001; 98: 3197-3202Crossref PubMed Scopus (758) Google Scholar, Hu et al., 2002Hu C.D. Chinenov Y. Kerppola T.K. Visualization of interactions among bZIP and Rel family proteins in living cells using bimolecular fluorescence complementation.Mol. Cell. 2002; 9: 789-798Abstract Full Text Full Text PDF PubMed Scopus (1132) Google Scholar). One of the most significant advantages of imaging for doing systems biology is that it allows data to be captured longitudinally over time through the use of time-lapse imaging. A basic requirement of time-lapse imaging is that the specimen continues to function normally throughout the course of image acquisition. Successfully culturing a specimen is made much more difficult by the requirements of imaging: the specimen must be anesthetized so it does not twitch, immobilized so it does not drift, held in the proper orientation and within the working distance of the objective (which is often short for high numerical aperture objectives), and capable of resisting phototoxicity from the constant bombardment of the photons used for imaging. The requirement of immobilization during culture is fairly easy to meet for some species such as bacteria and yeast, which can be simply imaged on thin agar pads (Elowitz et al., 2002Elowitz M.B. Levine A.J. Siggia E.D. Swain P.S. Stochastic gene expression in a single cell.Science. 2002; 297: 1183-1186Crossref PubMed Scopus (3546) Google Scholar) but is more challenging in other species. Zebrafish embryos can be blocked from twitching with anesthesia and from drifting with dilute agarose (Koster and Fraser, 2004Koster R.W. Fraser S.E. Time-lapse microscopy of brain development.Methods Cell Biol. 2004; 76: 207-235Crossref PubMed Scopus (22) Google Scholar). Chick (Kulesa and Fraser, 2002Kulesa P.M. Fraser S.E. Cell dynamics during somite boundary formation revealed by time-lapse analysis.Science. 2002; 298: 991-995Crossref PubMed Scopus (103) Google Scholar) and mouse (Jones et al., 2005Jones E.A. Baron M.H. Fraser S.E. Dickinson M.E. Dynamic in vivo imaging of mammalian hematovascular development using whole embryo culture.Methods Mol. Med. 2005; 105: 381-394PubMed Google Scholar) embryos can be successfully filmed for 12–36 hr using more involved setups that maintain proper temperature and gas balances while holding the embryo still and near the objective. Photobleaching of the fluorophore and phototoxicity to the specimen are always concerns during time-lapse imaging. These effects can be mitigated by reducing the intensity of the excitation light until images with the minimal acceptable signal-to-noise ratio are produced and through the use of high numerical apertu