Time-lapse imaging of fluorescent proteins in living cells has become an indispensable tool in biological sciences. However, its application at the organismal level still faces a number of obstacles, such as large specimen sizes preventing illumination of internal tissues, high background fluorescence and uncontrollable movement of target tissues or embryos. Here we describe our solutions for these issues to obtain 4-D fluorescent images from living Drosophila embryos using confocal microscopes. A computational procedure that detects and corrects the shift of moving objects to virtually stabilize them in time-lapse movies (iSEMS) is presented. We discuss the importance of postimaging treatment of raw image stacks for the discovery of novel phenotypes that have previously escaped attention from the analyses of fixed specimens.
Time-lapse imaging of cells and organs has greatly increased our capacity to understand the cellular processes in living organisms. This approach is particularly powerful when combined with the use of fluorescent proteins expressed in specific cells and/or subcellular compartments in cells in culture that is accessible to cutting edge technology of fluorescent probes and imaging equipment (Miyawaki 2005; Shaner et al. 2005). To apply these techniques to tissues, organs and living animals, additional issues must be overcome before seriously addressing specific questions. First, the large size of animal embryos (thickness: ~200 µm for Drosophila, 300–1000 µm for Xenopus), compared with cultured cells (~20 µm thick), interferes with illumination and light emission, and considerably reduces the intensity and quality of images. Thus the fluorescence must be bright enough for detectors, while limiting the amount of protein expression to avoid interference of the cellular processes of interest. Second, the type of fluorescent proteins and intracellular localization tags, and the methods of protein expression must be carefully selected so as to achieve the best signal-to-noise ratio. Third, appropriate imaging devices with lasers and filter sets should be used. Fourth, care must be taken to handle specimens to avoid damage during preparation and imaging condition should be gentle enough to minimize photo toxicity while ensuring maximum optical resolution for given fluorescent signals. Fifth, raw 4-D image stacks are not convenient for browsing multiple images and are often insufficient for revealing detailed morphological information. Thus, image stacks should be z-projected and converted to movie files. In addition, contrast enhancement and other image processing techniques are often desirable to highlight specific signals of interest for further quantitative analyses. Finally and most importantly, the throughput of time-lapse recording is normally limited by availability of equipment and the long time required for recording. To avoid the risk of making erroneous conclusions based on a small sample size, it is necessary to increase the throughput of imaging. This task requires automation of image acquisition, and capabilities to store and manage large amounts of data.
Here we discuss the above-mentioned issues based on our experience with imaging of Drosophila embryos using conventional confocal microscopes. This article begins by providing basic tips for developmental biologists who are interested in starting or improving fluorescent imaging of embryos. We then describe our solution for obtaining and handling large amounts of image data for comprehensive understanding of dynamic biological events.
Flies were cultured with standard yeast-corn meal-agar medium and the eggs were collected on apple juice-agar plates. Embryos were dechorionated with 1.75% sodium hypochloride for 3 min and washed extensively with distilled water. Embryos at a proper stage were selected under a fluorescent stereomicroscope (Olympus SZX16) and picked up with forceps (FONTAX, INOX No5). Embryos were placed on a glass-bottomed Petri dish (35 mm, AGC Techno Glass, Chiba, Japan) that was coated with a thin film of rubber cement (Maruni LTD, Osaka, Japan, http://www.maruni-ind.co.jp/e; diluted with heptane). Embryos were gently compressed to the glass surface to flatten the contact area and were covered with distilled water.
Image acquisition and processing
Time-lapse images were obtained using an inverted confocal microscope Olympus FV300 (10 mW Ar 488 nm and 1 mW HeNe 543 nm lasers) or FV1000 (15 mW LD 473 and 15 mW LD 559 nm lasers), with oil immersion objectives (Plan Apo 60x objective lens numerical aperture (NA) 1.42, Olympus, Japan). Details of imaging conditions are described in the legends for each movie. Confocal aperture was generally opened to the maximum size (300 µm). After acquisition, the image stacks in multi TIFF format (Olympus) were transferred to a Mac OS X server where an in-house developed software converted the stacks into z-projections and QuickTime movies.
The fluorescent recovery after photobleaching (FRAP) experiment was carried out on the dorsal branch of stage 12 embryos (btl > Dα-catenin-GFP) using a confocal microscope equipped with two scanners (Olympus FV1000). The GFP signal in a part of the adherens junction (AJ) was photo bleached by 405 nm diode laser (50 mw) irradiation and images were taken by continuous recording (256 × 192 pixel, 3 z-stacks/frame/2.6 s). Although lasers of longer wavelength (such as 488 nm) can be used for photobleaching, 405 nm laser was more efficient for GFP, which has major excitation wavelength at 395 nm. Quantification of FRAP data was carried out with a custom written software.
iSEMS compare a pair of successive frames in a x-y-t image stack and shift each frame to optimal positions that minimize the drift of objects (Fig. 1). The procedure is divided into three steps: (i) sliding frame n +1 in 25 different patterns; (ii) calculating ‘similarity (see below)’ of overlapping regions of image n and n +1; and (iii) shifting frame n +1 in the slide pattern that gives the minimal differential value. iSEMS then repeats the procedure with a quarter of the step size of the previous round until the step size becomes 1.
To rapidly match two successive frames n and n +1, the 25 different sliding patterns of overlap were selected as follows (Fig. 1B,C). For an image stack of 512 × 512 pixels (Fig. 1B), the initial step size was 32 pixels (= 512/16) and image n +1 was shifted in 25 different patterns to fit to the grid 1 (128 × 128 pixels). In the subsequent rounds, the step size was reduced to eight pixels (grid size 32 × 32) and then two pixels (grid size 8 × 8). The final round explored the remaining nine patterns with the step size of one pixel. The total number of patterns explored was 84 (25 + 25 + 25 + 9), which is far less than all possible matching patterns of the entire field (511 × 511 = 261 121) or in grid 1 (128 × 128 = 16 383). To estimate the similarity of the overlapping images, an evaluation function where the variance of pixel values subtracted between the adjacent frames was used. The matching pattern that gave the smallest variance was judged to be most similar.
Results and discussion
Live imaging of Drosophila embryos
Figure 2A shows our set up for culturing Drosophila embryos for observation. The specimens are observed using an inverted microscope, which is equipped with a confocal scanner unit and a motorized stage that allows us to take multiple time-lapse images from several positions. We have chosen the conventional confocal microscope for our study because it is widely available from many suppliers with a variety of options with relatively affordable cost compared with two-photon microscopes. For wavelength separation, we prefer standard dichroic mirrors and filter sets rather than more sophisticated prisms, diffraction grating and variable slits due to higher light recovery. The Drosophila egg is covered by a vitelline membrane and chorion, and is fertilized within its mother. It completes embryonic development in ~22 h at 25°C. The egg is elongated in shape: approximately 500 µm in anterior-posterior axis and 200 µm in dorsal-ventral axis. To prepare embryos for observation, they are dechorionated and placed on glass-bottomed Petri dishes using a glue to ensure stable and tight attachment to the glass surface. x-y and x-z view of a fixed embryo viewed in such configuration is shown in Figure 2B. The dishes were subsequently filled with distilled water, submerging the embryos: gas-permeable Halocarbon oil could have been used; however, we preferred water due to its superb properties in preventing desiccation and serving as a heat sink. This configuration allowed good optical clarity for observation under high NA objectives, and should be applicable for other organisms with comparable-sized eggs, and organs or slice cultures.
Choice of fluorescent protein
Among various fluorescent proteins available in the market, Aquiora GFP derivatives such as enhanced green fluorescence protein (EGFP) (Zhang et al. 1996) and Venus (Nagai et al. 2002) remain the best choices for high-resolution imaging due to their excellent quantum yield and rapid maturation into fluorescently active oxidized forms. The latter feature is particularly important for applications in organisms that undergo rapid development. In addition, as Drosophila embryos develop in relatively anaerobic conditions, they are also capable of developing normally under water. Based on our experience, it takes approximately 2 h for accumulation and maturation after the initiation of EGFP fusion protein expression, although this time varies considerably depending of the level of transcription and the nature of the fluorescent proteins. In contrast, maturation time of DsRed in the eye imaginal discs was estimated to be 22.5 h (Akimoto et al. 2005). For two-color labeling we have been using a red-shifted, monomeric version of DsRed called mCherry (Shaner et al. 2004) (Fig. 2E). It should be noted that although a number of new types of fluorescent proteins are being discovered and engineered to adapt for specific uses, it is always wise to carry out pilot experiments to check fluorescence levels and maturation times of the fluorescent proteins of interest in the tissue one plans to study: fluorescent proteins that worked well in some cell types or in vitro, do not necessarily work in the same way in other situations.
The second consideration is how and where to express the fluorescent proteins. The rule of thumb is to limit the expression to only the cells you want to observe. Fluorescence in any nearby tissues or cell types will interfere with the signal of interest and consequently, reduce the contrast in fluorescence. Cell type-specific expression is normally achieved by driving expression of fluorescent proteins under control of cell-specific promoters (Chalfie et al. 1994). In Drosophila, tissue-specific expression can be achieved by crossing a strain with tissue-specific Gal4 expression and a strain with fluorescent protein genes under control of Gal4 responsive promoter (Brand & Perrimon 1993). To obtain optimal image clarity, target tissues should ideally be located within approximately 30 µm from the surface: image clarity declines if the tissue of interest is located deeper in the specimen (> 100 µm).
To further increase the image clarity and information content in fluorescent signals, it is desirable to use subcellular localization tags to fluorescent proteins. We have used GFP fusion to tag cytoskeleton-binding proteins Moesin (Edwards et al. 1997) (F-actin, Fig. 2C) and Tau (Micklem et al. 1997) (Microtubule, Fig. 3) with satisfactory results. These polymer binding fusion proteins generally gave good signal-to-noise ratios due to the low abundance of proteins when unassociated with cytoskeleton polymers. On the other hand direct GFP fusions to cytoskeletal components (GFP-Actin and GFP-Tubulin) tend to accumulate high monomeric forms in the cytoplasm and have some toxicity due to reduced efficiency to polymerize.
Dual color imaging
We currently use a combination of two diode lasers (LD 473 nm, LD 559 nm) for two-color imaging of EGFP and mCherry. Due to a wider spacing of excitation wavelength, long lifetime and smaller wattage requirement, diode lasers are a better choice over a combination of two gas lasers (Ar 488 nm, HeNe 543 nm). In an example shown in Figure 2E, a pair of dorsal branches was visualized with the general cell membrane marker mCherry-CAAX and the luminal marker Srp-GFP (Kakihara et al. 2008). The cells at the branch tips called fusion cells contacted with each other and formed an intracellular luminal cavity (Tanaka-Matakatsu et al. 1996). The high-speed video images (supplemental movie 4 of Kakihara et al. ) revealed the process of luminal cavity formation took place in a 30-min period.
Ideally one would like to obtain high-resolution images of 3D stacks at high speed over a long period. In reality, however, a compromise must be made to balance the two conflicting requirements: maximizing the sensitivity to detect weak fluorescence signals while minimizing the intensity of excitation light intensity to avoid photo toxicity. We normally achieve this by opening the confocal aperture to maximum (300 µm). Although this reduces confocality and thus sacrifices the image resolution, a stronger signal intensity usually improves the overall quality of images. The number of z-sections and acquisition intervals should also be carefully determined to balance the information content of images and reduce damage to the specimen. We normally confirm the lack of phototoxicity by monitoring the developmental time of embryos and viability of observed embryos after overnight incubation.
Despite the careful handling of specimens during imaging, some perturbations of cellular and developmental processes due to tissue-specific expression of GFP fusion proteins, photo toxicity, and embryo manipulation are unavoidable. For example we have observed that expression of E-cadherin-GFP in the tracheal system delayed extension of dorsal branch of the trachea, most likely due to higher levels of E-cadherin preventing rapid cell rearrangement (Shindo et al. 2008). Although this phenotype was mild and did not affect the final morphology of the tracheal system, the viability, or fertility, one should be aware of such potential problems when drawing any conclusion from such observations. Thus it is important to assess the potential toxicity of the fluorescent protein of interest. Inclusion of proper controls such as the use of additional marker proteins and observation in other cell types would be necessary to back up any conclusions you might want to make. When assessing the effect of gene perturbation by mutations or expression of dominant-negative effecter molecules, we normally compare images of mutant animals and controls taken by simultaneous recording using multipoint imaging set up.
Once 4-D image stacks are obtained, the images are usually z-projected after adjustment of the contrast, brightness and gamma, and are saved in conventional movie formats such as QuickTime for browsing. Most of software that comes with confocal microscopes includes these functions. However, it is sometimes worthwhile to apply filters to further reduce the noise and intensify the signals of interest. One example is shown in Figure 2D. A variety of filters are available as ImageJ plugins (http://rsb.info.nih.gov/ij/). One should keep in mind that the choice of filters should be carefully selected to improve, but not to over intensify the signals.
Storage and management of large amounts of data set
The high-precision motorized x-y stage has allowed simultaneous time-lapse recordings of multiple specimens and has greatly increased the throughput, making the management of large amounts of data set being a serious issue. For example, the present multipoint time-lapse system can easily generate several gigabytes of image data per day, which has to be stored and encoded into compact movie files for quick browsing. We manage this issue by using a dedicated server system, which stores and processes the raw images into movie files and places them into a browsable database.
Stabilizing rapidly moving objects: iSEMS
One commonly encountered problem in live imaging of living organisms is the drift of objects in the field. Factors contributing to the drift include: migration of individual cells and movement and shape change of tissues due to morphogenetic movement. In addition, shift of x-y-z positions of the specimens due to thermal deformation of equipment and adjustment by operators during long-term recording also contribute to abrupt and unpredictable drift of objects. The net effects of these factors contribute to the ‘shaky’ movement of objects, which is often disturbing for observers. Automated image alignment software, such as TurboReg (http://bigwww.epfl.ch/thevenaz/turboreg/) is available. This software works on the assumption that some objects (spot, edge or shape) in image stacks remain invariable, and that those objects can be used as landmarks for tracking and for image alignment. However, biological specimens often lack such discrete landmarks, making this approach difficult.
To overcome the above-mentioned problems, we have developed an image alignment tool iSEMS (inter-frame shift correction; Materials and methods and Fig. 1) to process x-y-t image stacks. In short, iSEMS compares one image with another at previous time points and shifts its x-y position to best match the two images. The procedure is repeated for all consecutive image pairs and produces time-lapse image stacks.
As shown in Figure 3A (upper panel; supplemental movie 2), a pair of dorsal tracheal branches approaches from each side of the embryo (frame 0, 24) and fuse at the dorsal midline (frame 74). During the 100-min period of recording the embryo gradually rotated and the dorsal midline and fusion point drifted out of the observed area (upper side, frame 99). Processing of the image stack by iSEMS compensated for the drift: the tips of the tracheal branches now stayed near the center of the field (Fig. 3, lower panels, supplemental movie 3). Since both branches are moving, there is no static object that can be used as an absolute landmark. iSEMS tracked and maintained the mid-point between the two tips of the tracheal branches approximately to the center of the field.
Another application of iSEMS is to virtually stabilize the imaged objects for signal quantification. FRAP is a technique to measure the rate of protein turnover in a defined area. Faithful tracking of photo-bleached area is a key element for successful quantification of fluorescent signals. Movement of cells in rapidly developing organisms often causes problems in image tracking. Application of iSEMS in photobleaching experiments of tracheal branches greatly improved the tracking efficiency (Fig. 3B, supplemental movie 4), and allowed for semi-automated image quantification using an in-house developed software (Takebayashi et al. 2007; Shindo et al. 2008).
Advantages and limitations of iSEMS
One advantage of iSEMS over other methods of image stabilization is the speed of computation due to the greatly reduced number of matching. For the case of 512 × 512 pixel size images, iSEMS searches matches from only 84 patterns, that is 0.03% of all possible patterns matching in the entire field (262 144 patterns) or 0.5% if the search is limited to grid 1 (16 383 patterns). Although iSEMS neglects a large majority of potential matches, repeating the search of 25 patterns in a roughly spaced grid (32 pixels for 512 × 512 pixel size images) in increasingly smaller grid sizes efficiently identifies good matches. We are now routinely applying iSEMS to time-lapse images of Drosophila embryos and in most cases we have obtained satisfactory results. The simple algorithm allowed iSEMS to process images with relatively slow CPUs in desktop computers.
Some limitations for application should be noted. First, iSEMS cannot track abrupt movement of objects greater than one quarter of the x or y length of the image (> 128 pixel for 512 × 512 pixel size images). If the frame interval is long enough, this problem may be avoided by manually adjusting the position of the specimens during recording. Second, iSEMS does not track rotation of objects.
4-D time-lapse microscopy
Three-dimensional tissue architectures are sometimes difficult to grasp by browsing optical sections or two-dimensional representations of projected 3-D stacks. This is especially important when assessing the interaction of different tissue layers undergoing rapid morphogenetic movements. Dorsal closure of Drosophila embryos is one such example, where dorsal ectoderm partially overlaps with the adjacent extra-embryonic epithelia (amnioserosa), and undergoes concerted epithelial expansion of epithelia to cover the entire dorsal surface of the embryo. We prepared 4-D (x-y-z-t) time-lapse stacks and formatted them to show simultaneously the z- and x- stacks (Wada et al. 2007). Comparison of the still images (Fig. 4A) and movies (video 2 in the supplement of Wada et al. ) revealed that part of the amnioserosa moves beneath the leading edge of dorsal ectoderm and forms a two-tiered structure with a broad basolateral adhesion interface.
Capturing key moments in morphogenesis
Embryonic development proceeds through a series of morphogenetic movements in which one process depends on proper execution of the previous one. Such a ‘history’ of morphogenetic processes cannot be appreciated by examination of fixed materials. Here we present two examples of the use of time-lapse recording in spotting crucial intermediate moments that are not obvious in the final stage of development.
Dorsal closure is driven by contractile force in amnioserosa, which pulls dorsal ectoderm toward dorsal midline through the tight linkage of basolateral cell adhesion (Kiehart et al. 2000) (Fig. 4A–C, video 4 of Wada et al. ). Jun N-terminal Kinase (JNK) signaling is activated at the leading edge of dorsal ectoderm and regulates expression of BMP2/4-like ligand Dpp, which influences contractile cell behavior in both the dorsal ectoderm and amnioserosa (Hou et al. 1997; Homsy et al. 2006; Wada et al. 2007). Mutations of JNK or Dpp signaling activity arrest dorsal closure and result in severe ‘dorsal open’ phenotypes. Another class of genes affecting dorsal closure is those encoding integrin cell-matrix adhesion receptors. The terminal phenotype of mutants of those genes was minor compared with JNK and Dpp signaling mutants. Many mutants of α3 integrin survived until the third instar with minor scars in the dorsal epidermis (Rohrbough et al. 2000). Although this relatively weak terminal phenotype may be taken as evidence of α3 integrin playing a minor role in dorsal closure, time-lapse analyses revealed that this is not the case (Wada et al. 2007). As shown in Figure 4D (see also video 5 of Wada et al. ), α3 integrin mutant embryos undergo near complete breakage of dorsal epidermis–amnioserosa boundary due to the reduction of integrin-mediated cell adhesion. However, in a subsequent stage the open holes were sealed by expansion of the free edge of dorsal epidermis by active filopodia movement. Thus α3 integrin mutants transiently exhibited severe dorsal open phenotypes, which were later repaired by a wound-healing reaction (Wood et al. 2002), which remained active in the mutants. JNK signaling mutants exhibited a more severe phenotype because they were defective in both dorsal closure and wound healing (Galko & Krasnow 2004).
Another example is tracing of the migratory behavior of the tracheal branch. The dorsal branch tip attaches to the internal surface of the dorsal epidermis and its position is tightly associated with the posterior compartment of the epidermis (Fig. 4E, no misplacement was observed in 175 cases; examined Kato et al. ). To perturb this precise positioning mechanism, Kato et al. (2004) blocked Wingless (Wg) signaling by ectopic expression of Axin, a negative regulator of Wg signaling required for tracheal branch patterning. This treatment caused bifurcation of branch tips (Fig. 4F): the bifurcated branch tips, however, were almost always located at the posterior compartment (3.2% misplaced branches in 126 cases), suggesting that the mechanism limiting the position of tracheal branch tips to the posterior compartment remained active. However, time-lapse observation of mutant embryos revealed that the tip of the dorsal branch sometimes wandered out of the posterior compartment to settle in the anterior compartment for a short period and then quickly returned to the posterior compartment (Fig. 4G, movie 2 of Kato et al. ). This observation suggested that the position of the tracheal branch tips is limited to the posterior compartment by constantly adjusting their position, demonstrating that an active cell migration ensures the robust mechanism for cell positioning.
The practical tips for time-lapse imaging described here are only a small fraction of a large number of possible alternatives. Scientists should try to be up-to-date on the latest version of fluorescent reporters, for possible applications to their specific needs. The server system we have described is suited for mass storage and processing of automatically acquired images. For analyses of images for specific purposes, image-processing tools such as ImageJ (http://rsb.info.nih.gov/ij/) are available. With a rich resource of publicly available plug-ins, researchers should be able to customize the use of ImageJ for their specific experimental needs. Some of the image processing tools described in this manuscript are available from http://www.cdb.riken.jp/signal/. Given the rapid progress in imaging technology and genetic tools to label specific cell types with multiple colors, live imaging will continue to stimulate our imagination to uncover new aspects of development and morphogenesis.
We are grateful to Ken Kakihara, Housei Wada, Atsushi Wada and the members of the Hayashi lab for their contribution to the works described here. We also thank Tatsuhiko Noguchi for comments on the manuscript. K.K. was supported by a Riken Special Postdoctoral Fellowship.