Correlative microscopy and electron tomography of GFP through photooxidation

© 2005 Nature Publishing Group http://www.nature.com/naturemethods ARTICLES Correlative microscopy and electron tomography of GFP through photooxidation Markus Grabenbauer1, Willie J C Geerts2, Julia Fernadez-Rodriguez3, Andreas Hoenger1, Abraham J Koster2 & Tommy Nilsson3 We have developed a simple correlative photooxidation method that allows for the direct ultrastructural visualization of the green fluorescent protein (GFP) upon illumination. The method, termed GRAB for GFP recognition after bleaching, uses oxygen radicals generated during the GFP bleaching process to photooxidize 3,3¢-diaminobenzidine (DAB) into an electron-dense precipitate that can be visualized by routine electron microscopy and electron tomography. The amount of DAB product produced by the GRAB method appears to be linear with the initial fluorescence, and the resulting images are of sufficient quality to reveal detailed spatial information. This is exemplified by the observed intra–Golgi stack and intracisternal distribution of a human Golgi resident glycosylation enzyme, N-acetylgalactosaminyltransferase-2 fused either to enhanced GFP or CFP. Photooxidation is a powerful tool in fluorescence light microscopy (FLM) and transmission electron microscopy (TEM) when extracting precise spatial information of fluorescently labeled molecules at the ultrastructural level1,2. Taking advantage of free oxygen radicals that form upon illumination, DAB precipitates into an electrondense reaction product through photooxidation. The DAB precipitate can then be visualized by TEM to reveal detailed spatial information. If performed in correlative mode (observing the same structure or cell at the light and ultrastructural level), fluorescently labeled molecules can first be recorded in living cells. These are fixed and processed through a photooxidation procedure permitting evaluation by TEM. In addition to spatial information, it is also possible to extract detailed temporal information, that is, when particular molecules exist in a given place (for example, molecules moving within a tubular structure). Preferably, such correlative microscopy should be quantitative so that the initial fluorescence is reflected in a proportional manner at the ultrastructural level. One of the most frequently used fluorophores in recent cell biology studies is based on the GFP from Aequorea victoria. Along with spectral mutants and fluorescent proteins from other cnidarian sources, GFP fusion proteins have provided important insights into both the temporal and spatial organization of molecules. At the level of FLM, the accuracy of spatial resolution is usually limited to the wavelength used to illuminate the GFP and tends to be 300–500 nm, though this limit is constantly being pushed through new technologies such as 4Pi microscopy and stimulated emission depletion microscopy3. In one study, wild-type GFP fused to a peroxisomal targeting signal was visualized by electron microscopy upon photooxidation to reveal ultrastructural localisation4. No other investigations of GFP or variants using that protocol have been described. As a consequence, FLM is presently used in combination with immuno-based TEM, that is, the presence of the GFP molecule is revealed first at the fluorescence microscopy level and then at the ultrastructural level using thin frozen sections labeled with gold particles conjugated with antibodies to GFP. In a few high-end studies, the same membrane structure was investigated, both at the fluorescence microscopy level and at the ultrastructural level (see5–7, for example). Such correlative microscopy is very demanding as finding the same cell or part of the cell that was illuminated is difficult in thin sections. Alternative methods have therefore been developed to permit illumination-based correlative microscopy. These take advantage of fluorophores such as BODIPY-labeled ceramide analogs8, the membrane probe FM 1–43 (ref. 9) or tetracysteine tags labeled by a biarsenical chromophore, ReAsH2, all of which have high potential of generating free radicals upon illumination. As GFP and related variants are so commonly used, we have focused on developing conditions in which the free radicals generated by GFP can be harnessed to precipitate DAB. Our goal was to establish conditions that produce a DAB precipitate of sufficient quality to permit high-resolution investigations in both two and three dimensions. Visualization in three dimensions is necessary to extract spatial and temporal information of GFP fusion proteins in highly convoluted membrane structures such as those involved in the exocytic and endocytic pathway (for example, on the Golgi architecture10). As established GFP markers, we used the human Golgi-resident enzyme N-acetylgalactosaminyltransferase-2 fused to the enhanced green or cyan GFP mutants, EGFP (GalNAc-T2GFP) and ECFP (GalNAc-T2CFP), respectively. Both are stably expressed 1Cell Biology and Cell Biophysics Program, EMBL, Meyerhofstr. 1, D-69117 Heidelberg, Germany. 2Department of Molecular Cell Biology, Institute of Biomembranes, Utrecht University, 3584 CH Utrecht, The Netherlands. 3Department of Medical Biochemistry and Swegene Center for Cellular Imaging, Göteborg University, Medicinaregatan 9A, 413 90 Göteborg, Sweeden. Correspondence should be addressed to T.N. (tommy.nilsson@medkem.gu.se). RECEIVED 30 AUGUST; ACCEPTED 24 SEPTEMBER; PUBLISHED ONLINE 21 OCTOBER 2005; DOI:10.1038/NMETH806 NATURE METHODS | VOL.2 NO.11 | NOVEMBER 2005 | 857 ARTICLES © 2005 Nature Publishing Group http://www.nature.com/naturemethods in human cells (HeLa) and have been extensively characterized both at the light and ultrastructural level11. RESULTS Development of the protocol The development of the GRAB protocol focused on improving the sensitivity and the preservation of ultrastructure, such that generated DAB products could be observed in the correct context, both in two-dimensional as well as three-dimensional (3D) electron microscopy (applying electron tomography). We achieved the main improvements by using ammonium chloride and sodium borohydrate to reduce aldehyde-induced autofluorescence and by blocking endogenous enzyme activities, both of which cause nonspecific DAB precipitation. An elevated oxygen content in the working DAB solution is necessary to generate sufficient amounts of oxygen radicals upon GFP illumination (detailed protocol is available in Supplementary Methods online). GFP recognition after bleaching (GRAB) By FLM, HeLa cells expressing GalNAc-T2GFP show a juxtanuclear distribution of this Golgi marker (Fig. 1a). After performing the GRAB reaction, DAB precipitation was predominantly found in Golgi stack–like profiles (Fig. 1b,c and Supplementary Fig. 1 online). At higher magnification, DAB staining appears in several cisternae giving the impression of a gradient-like distribution across the stack. In areas of slightly dilated cisternae, DAB precipitate appears restricted to the luminal side (Fig. 1c). This is consistent with the presence of EGFP on the luminal, carboxyterminal part of GalNAc-T2. It also suggests a locally restricted DAB precipitation. We detected no DAB precipitate in Golgi-like structures in the absence of illumination (not shown) or in control cells devoid of GalNAc-T2GFP (Fig. 1d). On occasion, we observed nonspecific DAB precipitates as well as osmiophilic autophagosome-like structures in both GFP-expressing cells and in control cells devoid of GFP expression. This is an inherent limitation of DAB-based protocols12 and must be taken into account when interpreting TEM micrographs. The observed gradient-like labeling for GalNAc-T2GFP is consistent with our earlier immuno-based TEM studies showing gradient-like distributions of Golgi resident glycosylation enzymes including GalNAc-T2 (refs. 13–15). The average number of gold particles in those studies was 5–10 per stack, making it impossible to detect an enzyme gradient over a single stack. Instead, 50–100 profiles had to be averaged to reveal a gradient. In this study and using the GRAB technique, a gradient-like distribution appears in a single stack. On occasion, we also observed GalNAc-T2GFP–induced DAB precipitate in what appeared as vesicular or tubular profiles, mostly close to the dilated rims of the cisternae but sometimes also in discrete round profiles away from the Golgi stack. Correlative light and electron microscopy To test whether GRAB can also be used with spectral variants of GFP, we performed similar experiments with a cell line stably expressing GalNAc-T2 coupled to ECFP (GalNAc-T2CFP). We performed this experiment in correlative mode to illustrate the time course of photooxidation. We imaged the initial Golgi-like fluorescence of GalNAc-T2CFP and overlaid this with a bright field channel (Fig. 2a,b) to follow the time course of the reaction. At the start of the experiment, cells were barely detectable by bright field 858 | VOL.2 NO.11 | NOVEMBER 2005 | NATURE METHODS a b ER N c d ER AV Figure 1 | Photooxidation of EGFP polymerizes DAB to an electron-dense precipitate. (a) Fluorescence light microscopy shows the Golgi resident enzyme GalNAc-T2GFP stably expressed in HeLa cells and exclusively localized to the juxta-nuclear region. (b) After photooxidation and epon embedding, the DAB staining can be identified by electron microscopy resembling the EGFP localization at the juxtanuclear Golgi stack. At higher magnification, the DAB polymerization sites can be precisely localized to the cisternal lumina with a gradient-like distribution across the Golgi stack. (c) The luminal precipitation of DAB confirms the predicted topology of GalNAc-T2GFP (arrow). (d) Control cells without EGFP expression do not show any precipitation in the Golgi apparatus after illumination. The osmiophilic autophagosomes (AV) show generally dark staining independent of illumination, GFP expression or presence of DAB. ER lumen is not labeled. N, nucleus; AV, autophagic vacuole; ER, endoplasmic reticulum. Scale bars, 5 mm (a), 500 nm (b), 100 nm (c), and 200 nm (d). microscopy (Fig. 2c,d). After 6 min, a cytosolic background staining appeared, which increased over time. We stopped the reaction after 8 min when ECFP fluorescence was bleached completely (Fig. 2e,f). We next prepared ultrathin epon sections (50–70 nm) and examined them by TEM, correlating Golgi stacks with their FLM images (Fig. 2). At higher magnification, we also observed a gradient-like distribution across the stack for this fusion protein (Fig. 2j). We observed osmiophilic structures along with staining in mitochondria, the latter suggesting nonoptimal blocking of endogenous oxidation reactions. We also tested other marker proteins fused to either EGFP or ECFP, in some of which the fluorescent protein was exposed in the cytosol (see Supplementary Figs. 2 and 3 online). In general, EGFP fusion proteins required longer reaction times compared to ECFP. All fusion proteins tested gave rise to various levels of DAB staining, suggesting that the GRAB technique must be adapted and optimized for different fusion proteins. Quantification of signal-to-noise ratio To better ascertain whether the final density of the DAB product correlates with the fluorescence of the EGFP or ECFP fusion protein, we quantitatively analyzed the signal-to-noise-ratios of DAB-stained Golgi and compared this to their corresponding signal-to-noise ratios as measured by FLM. Background staining was reduced by the addition of pottassium cyanide (see Methods). To generate a nonhomogeneous population of cells with respect to © 2005 Nature Publishing Group http://www.nature.com/naturemethods ARTICLES a d b e c g i h j f Figure 2 | The bleaching process and correlative light and electron microscopy of the same Golgi resident enzymes. (a) Hela cells expressing the cyan mutant ECFP fused to GalNAc-T2 are shown by fluorescence light microscopy. The indicated cell (arrow) was then followed during the incubation and finally analyzed at the ultrastructural level. (b) The fluorescence is combined with the bright field channel. (c–f) During the bleaching process in the presence of DAB, fluorescence fades away and background staining appears: 0 min (c), 4 min (d) and 6 min (e). After 8 min (f), the reaction was stopped. At this time-point, the fluorescence was completely erased. (g–i) One of the cells (highlighted by arrows in f and g) is identified in the electron microscope. A juxtanuclear Golgi stack becomes visible at higher magnifications (arrows in h,i). (j) The Golgi stack shows specific and very precisely localized DAB precipitation in the lumen of two cisternae. This staining is induced by ECFP and correlates directly to the observed fluorescence (arrow in a). Other dark structures like mitochondria, autophagic-like vacuoles, background precipitation as well as osmium stained membranes were distinguished from the specific staining by comparing with control cells without the fusion proteins. Scale bars, 25 mm (a–f), 10 mm (g), 5 mm (h), 2 mm (i) and 500 nm (j). fluorescence levels, we prebleached cells in the absence of DAB before the GRAB procedure (Fig. 3). Analyzing cells at different levels of prebleach, we found a linear correlation of DAB and GFP signal-to-noise ratios with a Pearson’s coefficient of r ¼ 0.976. This suggests that the observed density of DAB precipitate produced during the reaction is a linear reflection of the fluorescence intensity produced by the ECFP fusion protein. Electron tomography We next examined whether the GRAB technique could be used to highlight intracisternal distribution. As the Golgi apparatus consists of partly convoluted and fenestrated membranes, it was necessary to perform this analysis in thick sections by electron tomography. By tilting 250-nm-thick epon sections from
65 to +65 degrees around two perpendicular axes, a part of the Golgi stack containing DAB precipitate could be reconstructed into an electron tomogram. In virtual 5-nm cross-sections of the reconstructed tomographic volume, the electron-dense DAB precipitate is visible throughout two of the cisternae (Fig. 4a). By digital contrast enhancing, we obtained sufficient membrane contrast without lead citrate or uranyl acetate treatment. The photooxidation reaction produces a DAB precipitate of sufficient density to allow a spatial representation of GalNAc-T2GFP, even to the point of detecting the product in peripheral buds and vesicles (Fig. 4b–d and Supplementary Video 1 online showing the complete tomogram, the membrane tracing and the final model). We assigned peri-Golgi vesicles with great care; only when there was clear discontinuity or when the continuity was of such size that membrane continuity could be ruled out with a high degree of certainty, we assigned the structures as vesicles. As observed previously, we noticed on several occasions that vesicle-like structures were attached to the cisternal membrane through small continuities (Fig. 4). Based on their apparent thickness, we think it is unlikely that they correspond to membrane structures. Rather, we suggest that these connections correspond to protein structures that keep vesicles in close proximity to the cisternal membrane16,17. Additional views of peri-Golgi vesicles are available in Supplementary Video 2 and Supplementary Fig. 4 online. We finally examined the distribution of the DAB precipitate in the plane of a cisterna by thresholding of gray values representing GalNAc-T2GFP content in the tomogram volume of a DAB-filled cisterna (lower cisterna in Fig. 4d). The cisterna is colored gray and the DAB content is green (Fig. 4e). This highlights what appears as a nonhomogeneous distribution pattern showing local regions of increased DAB precipitation, suggestive of local GFP-labeled Golgi enzyme clusters. It also suggests that the reaction product does not NATURE METHODS | VOL.2 NO.11 | NOVEMBER 2005 | 859 a b c d e f g h 5 i Pearson's r = 0.967 4 DAB S/N (a.u.) © 2005 Nature Publishing Group http://www.nature.com/naturemethods ARTICLES 3 2 1 Figure 3 | Quantitative analysis of the signal-to-noise ratio by correlative light and electron microscopy. (a,b) Part of GalNAc-T2CFP– expressing cells (a) were prebleached in the absence of DAB (cells to the right of the dashed line) and completely bleached in presence of DAB. (c–g) The same cells imaged by bright-field light microscopy (red dashed box in c), are retraced in a thin section by electron microscopy (d). By increasing the magnification (d–g), individual Golgi stacks are revealed (g), which are directly correlated to fluorescent Golgi apparatus (arrow in a). The dashed squares (c–f) outline the proximate areas subsequently revealed at higher magnification (d–g). Transverse-cut Golgi stacks of the same ultrathin section are imaged in the same magnification and correlated to individual Golgi apparatus fluorescence signals after prebleaching (b). (h) The signal-to-noise ratios of specific ECFP Golgi fluorescence against cytosol and specific DAB precipitation against cytosolic background is plotted in a graph and shows a linear regression. (i) Strongly prebleached cells do not show any DAB precipitation in their cisternal lumen. Note the absence of background staining in mitochondria using 100 mM KCN to block endogenous oxidases (see Methods). Scale bars, 25 mm (a–c), 10 mm (d), 5 mm (e), 2 mm (f), 200 nm (g,i). 0 0 1 2 3 4 EGFP S/N (a.u.) 5 proceed throughout the cisternae. In support of this, bud-like profiles with high and low (Fig. 4e) amounts of precipitate occur side by side originating from the same cisterna. We propose that the GRAB technique as applied here can be used to reveal intramembrane distributions both in two and three dimensions. DISCUSSION The GRAB technique offers a method for quantitative correlation of GFP fluorescence with the corresponding putative protein distribution (represented by DAB precipitate) at the ultrastructural level. The protocol overcomes the low capacity of GFP molecules to produce sufficient amounts of oxygen radicals needed for illumination-based electron microscopy (detailed protocol is available in Supplementary Methods). We found that the DAB precipitate appears to be locally distributed, indicating a close spatial relationship between the chromophore (GFP) and the final precipitate. The DAB precipitate formed upon GFP illumination is somewhat finer, but seems to be qualitatively on par with that formed when horseradish peroxidase is expressed as a fusion protein or used in immuno-based labeling. Because the density of the DAB product generated by this method appears proportional to the initial fluorescence, this suggests that contrary to previous reports18,19 Golgi enzymes (as exemplified by GalNAc-T2GFP) do not appear excluded from peri-Golgi vesicles. This is in line with in vivo20 as well as in vitro21–23 findings. In immuno-based electron microscopy, the effective spatial resolution is hampered by the primary (and secondary) antibodies bound to an antigen within10
15 nm. In comparison, the resolu860 | VOL.2 NO.11 | NOVEMBER 2005 | NATURE METHODS tion with the GRAB technique is expected to be somewhat higher (5
8 nm) provided that the reaction product does not diffuse away from the point of origin. This appears not to be the case under the conditions used in this study. This is most easily explained by having very short diffusion distances of reactive oxygen species and DAB (Supplementary Fig. 3). Upon precipitation the reaction product becomes immobile. Another potential advantage compared to immuno-based electron microscopy is that DAB is much more membrane permeable and smaller in size. In other words, it has a higher capacity to function in small and closed membrane structures such as vesicles and buds where antigenic access in combination with immuno-gold might be limited. Immunobased electron microscopy also relies on having exposed antigens, thus necessitating thin sectioning, but illumination-based electron microscopy can be performed on thick epon sections, which constitutes a major advantage for electron tomography24. Conversely, it can be difficult to construct GFP fusion proteins with native activity and localization from which stable transfectants with near-native amounts of fusion protein must be created. As immuno-based electron microscopy relies solely on antigenic recognition of the endogenous protein, this is not a problem. The use of previously well-characterized fusion proteins can ease this requirement in GRAB. Hence, illumination-based electron microscopy could be the method of choice when performing 3D structural analysis. In conclusion, the possibility of performing correlative and quantitative illumination-based electron microscopy with EGFP or ECFP and their variants should be of great value for the community working with live cells as it offers a direct way of extending EGFP and ECFP
based studies to the ultrastructural © 2005 Nature Publishing Group http://www.nature.com/naturemethods ARTICLES a b c d Figure 4 | Electron tomography of a Golgi stack containing GalNAc-T2GFP. (a) In a virtual 5-nm section through the tomogram, DAB precipitation representing GalNAc-T2GFP is visible in two Golgi cisternae, a peri-Golgi vesicle (arrow) and a coat protein I (COPI)-coated bud (arrowheads). (b) In the 3D model extracted from the tomogram, the GalNAc-T2GFP–containing cisternae are colored green, GalNAc-T2GFP–containing vesicles are red, other vesicles are white, unlabeled cisternae are blue and the endoplasmic reticulum is purple. (c) In a different-angle view, the localization of a red vesicle containing resident enzymes adjacent to GalNAc-T2GFP–labeled cisternae becomes more evident (compare to arrow in a). A virtual section of the tomogram is set as background image. (d) Unlabeled cisternae are omitted. (e) By density thresholding, a domain-like pattern of GalNAc-T2GFP across a cisterna is highlighted. (f–h) A GalNAc-T2GFP–containing vesicle (arrow in a) was reconstructed by gray-level thresholding as an alternative representation method to manual tracing (shown in perspective views with and without virtual sections). Scale bar, 100 nm. 3D movies of the tomogram and the vesicle, are available (Supplementary Videos 1 and 2, respectively). e f g h level. We anticipate that variants of EGFP that are more directly tailored to illumination-based electron microscopy will soon emerge, thus improving the reliability of this method. METHODS Cell culture and transfection. We generated stable transfectants as described previously11. Briefly, we routinely kept monolayer HeLa cells (CCL 185; American Type Culture Collection) in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal calf serum (FCS) and 2 mM L-glutamine (GIBCO BRL, Life technologies) at 37 1C and 5% CO2. For generation of stable transfectants, we transfected plasmids encoding GalNAc-T2GFP or GalNAc-T2CFP into HeLa cells cultured in 10-cm tissue culture dishes in the presence of 5% FCS using the calcium phosphate protocol as described11,25. We selected for B3 weeks in the culture medium supplemented with 300 mg/ml geneticin (G-418 sulfate; Sigma) or 1 mg/ml puromycin (Sigma). After cell death, we recultured positive clones and checked for the presence of appropriate fluorescence. Fluorescence microscopy and imaging. We grew cells in glassbottom Petri dishes (1.5; MatTek) 24 h to 48 h before use and observed on a Zeiss Axiovert (100 TV) fitted with a 63 planapochromat 1.4 NA or a water-corrected 40 apochromat 1.2 NA. We imaged live cells with a Hamamatsu 3-chip color charge-coupled device (CCD) camera with Open Lab software (Improvision). GFP recognition after bleaching (GRAB) technique. We carried out all incubation steps including bleaching, dehydratation and resin polymerization in glass-bottom tissue culture dishes. We briefly washed cells with prewarmed calcium- and magnesium- free phosphate-buffered saline (PBS; pH 7.4) and fixed them for 30 min with prewarmed fixative containing 2% glutaraldehyde (25% stock solution; Merck) and 2% sucrose (USB) in PBS. After washing 3 with PBS, we blocked samples with 100 mM glycine and 50–100 mM potassium cyanide in PBS for 1 h followed by 40 min 100 mM ammonium chloride and 10 mg/ml sodium borohydrate in PBS. Potassium cyanide eliminates background DAB oxidation by mitochondrial respiration1,2 and glycine, ammonium chloride and sodium borohydrate reduce glutaraldehyde autofluorescence and consequently background DAB precipitation9,26. For photoconversion, we washed samples in Tris-buffered saline (TBS; pH 7.4) and then incubated them in a freshly prepared and oxygen-saturated solution of 1– 2 mg/ml DAB (Polysciences) in TBS at 10 1C or below. The low temperature maintains a high oxygen content and supports effective and specific DAB polymerization. To bleach, we illuminated samples using the appropriate filter settings for EGFP (excitation filter BP 470/40) or ECFP (excitation filter BP 436/20) using a 100-W mercury lamp (AttoArc; Zeiss). We monitored the development of DAB reaction product, and stopped photooxidation when cytosolic background staining occurred, typically after 30 min using GFP or 10 min using CFP (See Supplementary Methods for a detailed protocol). Electron microscopy. After photoconversion, we washed samples with distilled water and postfixed for 30 min on ice in 1% osmium tetroxide reduced by 1.5% potassium ferrocyanide27. We dehydrated the samples in graded ethanol series and embedded them in Epon 812 (Serva). After overnight polymerization, we removed the glass bottom of tissue culture dishes by hydrofluoric acid. We cut ultrathin sections (30–50 nm) of flat, embedded cells parallel to the surface on an Ultracut S ultramicrotome (Leica) and mounted them on Formvar-coated grids. Occasionally, we counterstained sections with 1% aqueous uranyl acetate and lead citrate (data not shown). These salts seem to have a higher affinity to osmicated compounds than to DAB precipitates; thus a higher-contrast preparation lowers the signal-to-noise ratio. We examined samples on a Philips BioTwin CM 120 at 100 kV accelerating voltage. We obtained digital images using a top-mount Gatan TEM camera and processed with Photoshop 7 (Adobe). Quantification of signal-to-noise ratios. To generate nonhomogeneous GRAB preparations, we partially photobleached cell NATURE METHODS | VOL.2 NO.11 | NOVEMBER 2005 | 861 © 2005 Nature Publishing Group http://www.nature.com/naturemethods ARTICLES groups expressing fluorescent GalNAc-T2 in the absence of DAB, imaged them and then bleached in the presence of DAB. After epon embedding, we identified individual cell groups and cells in the electron microscope using their shapes as a marker. To avoid any variations by Golgi stack orientation, section thickness or microscope and camera settings, we consecutively imaged only transverse-cut Golgi cisternae at the same magnification. The signal-tonoise ratios of raw data were measured and analyzed on DAB-stained Golgi cisternae versus cytosolic background using ImageJ (US National Institutes of Health) and plotted against similarly analyzed fluorescence data of exactly correlated cells and organelles. Electron tomography, 3D reconstruction and modeling. Epon sections of deep purple interference color (250-nm thickness) were decorated on both sides with 10 nm–sized colloidal gold markers. For electron tomography, we acquired dual axis tilt series on a FEI Tecnai 20 LaB6 transmission electron microscope at 200 kV. Digital images were recorded on a 2k  2k CCD camera mounted on the TEM (a Temcam F214 setup with 14-mm-sized pixels; TVIPS GmbH). The series were automatically recorded using the Open Tomo program28 over a tilt range of –651 to +651 with 11 increments around two orthogonal axes. Tomograms of the two tilt series were generated and combined using the IMOD program package29. The dimensions of the tomogram were 1,024  1,024  100 voxels, with 1.62 nm per voxel. We modeled Golgi stacks and vesicles by computer-assisted tracing of membrane contours based on gray-value thresholding. The presence or absence of DAB precipitation in cisternae and vesicles was decided based upon local intensity differences (thresholding) within the tomogram, and the modeled objects were annotated (green or red) to visualize GFP localization. For annotation and visualization we used several packages: the annotation was done using the Amira 3.0 (TGS) software package and the general 3D image handling with Priism/IVE (UCSF). Note: Supplementary information is available on the Nature Methods website. ACKNOWLEDGMENTS We wish to thank R. Pepperkok, J. Rietdorf and K. Miura (Advanced Light Microscopy Facility, EMBL Heidelberg) for discussions and help in image analysis, M. Lebbink (Utrecht, The Netherlands) for his help using Amira software, M. Axelsson for help in preparing mitotic populations of HeLa cells and Swegene for its support of the Center for Cellular Imaging in Gothenburg. HeLa cells stably expressing a-tubulin-EGFP were a generous gift of J. Lipp and J.-M. Peters (IMP Vienna, Austria). This work was supported by an EMBO fellowship (M.G.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/naturemethods/ Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/ 1. Maranto, A.R. Neuronal mapping: a photooxidation reaction makes Lucifer yellow useful for electron microscopy. Science 217, 953–955 (1982). 2. Gaietta, G. et al. Multicolor and electron microscopic imaging of connexin trafficking. Science 296, 503–507 (2002). 3. Egner, A. & Hell, S.W. Fluorescence microscopy with super-resolved optical sections. Trends Cell Biol. 15, 207–215 (2005). 862 | VOL.2 NO.11 | NOVEMBER 2005 | NATURE METHODS 4. Monosov, E.Z., Wenzel, T.J., Luers, G.H., Heyman, J.A. & Subramani, S. Labeling of peroxisomes with green fluorescent protein in living P. pastoriscells. J. Histochem. Cytochem. 44, 581–589 (1996). 5. Svitkina, T.M. & Borisy, G.G. Correlative light and electron microscopy of the cytoskeleton of cultured cells. Methods Enzymol. 298, 570–592 (1998). 6. Castejon, O.J. & Castejon, H.V. Correlative microscopy of cerebellar Golgi cells. Biocell 24, 13–30 (2000). 7. Mironov, A.A., Polishchuk, R.S. & Luini, A. Visualizing membrane traffic in vivo by combined video fluorescence and 3D electron microscopy. Trends Cell Biol. 10, 349–353 (2000). 8. Pagano, R.E., Sepanski, M.A. & Martin, O.C. Molecular trapping of a fluorescent ceramide analogue at the Golgi apparatus of fixed cells: interaction with endogenous lipids provides a trans-Golgi marker for both light and electron microscopy. J. Cell Biol. 109, 2067–2079 (1989). 9. Harata, N., Ryan, T.A., Smith, S.J., Buchanan, J. & Tsien, R.W. Visualizing recycling synaptic vesicles in hippocampal neurons by FM 1–43 photoconversion. Proc. Natl. Acad. Sci. USA 98, 12748–12753 (2001). 10. Marsh, B.J., Mastronarde, D.N., Buttle, K.F., Howell, K.E. & McIntosh, J.R. Organellar relationships in the Golgi region of the pancreatic beta cell line, HITT15, visualized by high resolution electron tomography. Proc. Natl. Acad. Sci. USA 98, 2399–2406 (2001). 11. Storrie, B. et al. Recycling of golgi-resident glycosyltransferases through the ER reveals a novel pathway and provides an explanation for nocodazole-induced Golgi scattering. J. Cell Biol. 143, 1505–1521 (1998). 12. Lubke, J. Photoconversion of diaminobenzidine with different fluorescent neuronal markers into a light and electron microscopic dense reaction product. Microsc. Res. Tech. 24, 2–14 (1993). 13. Nilsson, T. et al. Overlapping distribution of two glycosyltransferases in the Golgi apparatus of HeLa cells. J. Cell Biol. 120, 5–13 (1993). 14. Rabouille, C. et al. Mapping the distribution of Golgi enzymes involved in the construction of complex oligosaccharides. J. Cell Sci. 108, 1617–1627 (1995). 15. Rottger, S. et al. Localization of three human polypeptide GalNAc-transferases in HeLa cells suggests initiation of O-linked glycosylation throughout the Golgi apparatus. J. Cell Sci. 111, 45–60 (1998). 16. Orci, L., Perrelet, A. & Rothman, J.E. Vesicles on strings: morphological evidence for processive transport within the Golgi stack. Proc. Natl. Acad. Sci. USA 95, 2279–2283 (1998). 17. Shorter, J., Beard, M.B., Seemann, J., Dirac-Svejstrup, A.B. & Warren, G. Sequential tethering of Golgins and catalysis of SNAREpin assembly by the vesicle-tethering protein p115. J. Cell Biol. 157, 45–62 (2002). 18. Cosson, P., Amherdt, M., Rothman, J.E. & Orci, L. A resident Golgi protein is excluded from peri-Golgi vesicles in NRK cells. Proc. Natl. Acad. Sci. USA 99, 12831–12834 (2002). 19. Kweon, H.S. et al. Golgi enzymes are enriched in perforated zones of golgi cisternae but are depleted in COPI vesicles. Mol. Biol. Cell 15, 4710–4724 (2004). 20. Martinez-Menarguez, J.A. et al. Peri-Golgi vesicles contain retrograde but not anterograde proteins consistent with the cisternal progression model of intraGolgi transport. J. Cell Biol. 155, 1213–1224 (2001). 21. Lanoix, J. et al. GTP hydrolysis by arf-1 mediates sorting and concentration of Golgi resident enzymes into functional COP I vesicles. EMBO J. 18, 4935–4948 (1999). 22. Lin, C.C., Love, H.D., Gushue, J.N., Bergeron, J.J. & Ostermann, J.E.R. Golgi intermediates acquire Golgi enzymes by brefeldin A–sensitive retrograde transport in vitro. J. Cell Biol. 147, 1457–1472 (1999). 23. Malsam, J., Satoh, A., Pelletier, L. & Warren, G. Golgin tethers define subpopulations of COPI vesicles. Science 307, 1095–1098 (2005). 24. McIntosh, R., Nicastro, D. & Mastronarde, D. New views of cells in 3D: an introduction to electron tomography. Trends Cell Biol. 15, 43–51 (2005). 25. Paabo, S., Weber, F., Nilsson, T., Schaffner, W. & Peterson, P.A. Structural and functional dissection of an MHC class I antigen-binding adenovirus glycoprotein. EMBO J. 5, 1921–1927 (1986). 26. Baschong, W., Suetterlin, R. & Laeng, R.H. Control of autofluorescence of archival formaldehyde-fixed, paraffin-embedded tissue in confocal laser scanning microscopy (CLSM). J. Histochem. Cytochem. 49, 1565–1572 (2001). 27. Hayat, M.A. Principles and techniques of electron microscopy. (Van Nostrand Reinhold Company, New York, 1970). 28. Ziese, U. et al. Automated high-throughput electron tomography by precalibration of image shifts. J. Microsc. 205, 187–200 (2002). 29. Kremer, J.R., Mastronarde, D.N. & McIntosh, J.R. Computer visualization of threedimensional image data using IMOD. J. Struct. Biol. 116, 71–76 (1996).