3D electron tomography of brain tissue unveils distinct Golgi structures that sequester cytoplasmic contents in neurons

Macroautophagy is a cellular process that is involved in protein and organelle degradation with diverse physiological implications. It is morphologically characterized by the formation of double-membraned vesicles termed autophagosomes (Cuervo and Macian, 2012; Mizushima et al., 2008). The deletion of genes coding for essential proteins in autophagosome formation causes neurodegeneration (Hara et al., 2006; Komatsu et al., 2006), revealing that the degradation of cytoplasmic contents is crucial for basal neuronal homeostasis. Surprisingly, the number of autophagosomes is very low in neurons (Mizushima et al., 2004; Nixon et al., 2005), which can be partly explained by the rapid clearance of autophagosomes in healthy neurons (Boland et al., 2008). Alternatively, pathways and/or structures that have not yet been identified might cooperate with neuronal autophagosomes in the degradation of cytoplasmic content. The Golgi complex has previously been identified as a membrane donor for conventional autophagosome formation both in yeast and mammalian systems (Ohashi and Munro, 2010; van der Vaart and Reggiori, 2010; Young et al., 2006).

For conventional electron microscopy, dehydrated specimens are prepared by chemical fixation, which is known to induce artefacts, to be inadequate for capturing dynamic events and to be particularly unsuitable for the preservation of membrane-rich structures (He and Fernandez, 2010; Korogod et al., 2015). Nowadays, optimal structural preservation can be achieved by preparing the specimens with rapid-freezing techniques. In particular, high-pressure freezing (HPF) can fully vitrify a bulk specimen (up to 200–400 µm) by means of delivering synchronized pressurization and cooling of the sample within milliseconds. Freeze-substitution involves the slow replacement of frozen cellular water with organic solvents at low temperature (−90°C range). At this temperature, large molecules and organelles are immobile but small molecules and solvents can penetrate cells (He and Fernandez, 2010). The combination of HPF and freeze-substitution techniques is actually the most adequate combination for the ultrastructural analysis of brain samples with high-resolution imaging (Korogod et al., 2015; Studer et al., 2014). Electron tomography is based on the acquisition of images from the specimen at different tilt angles, which are subsequently processed and combined to yield a three-dimensional (3D) reconstruction, or tomogram (Fernandez, 2012). The combination of HPF with freeze-substitution and electron tomography has been previously employed to analyze synaptic ultrastructure in brain tissue (Rostaing et al., 2006; Siksou et al., 2009).

We have set up a protocol to preserve mouse brain striatal tissue by using HPF and freeze-substitution. Use of electron tomography to analyze these samples has made it possible to identify and characterize distinct Golgi structures that coexist with neuronal autophagosomes and that might cooperate in the degradation of cytoplasmic contents.

Mouse striatal tissue was subjected to HPF and freeze-substitution. The high-quality preservation of the tissue was confirmed by performing transmission electron microscopy analysis (Fig. 1A). The tissue showed a uniform appearance with well-contrasted cell bodies and neuropil. At higher magnification, we identified specific Golgi structures composed of several concentric double-membraned layers (Fig. 1B–E). These Golgi structures engulfed cytoplasmic contents and were found to be at different stages of closure; they appeared as cup-shaped (Fig. 1B,E) or as completely closed structures (Fig. 1C,D). These engulfing structures were larger than nearby autophagosomes (Fig. 1F) and they coexisted in the neuronal soma (Fig. 1G). The electron microscopy images shown in this paper correspond to samples from 9-month-old animals. We also identified these engulfing structures at 2 and 6 months of age (data not shown) with no apparent difference across the ages (quantifications are shown in Table S1). The structures were also detected in samples that had been processed using conventional electron microscopy fixation methods (Fig. S1A). Engulfing structures coexisted with classic Golgi stacks in samples that had been preserved by using HPF (Fig. S1B).

We next performed electron tomography to visualize the 3D architecture of engulfing Golgi. Computational slices across the 3D volumes highlight the details of representatives of the open (Fig. 2; Fig. S2) and closed structures (Fig. 3; Fig. S2). Formation of these structures involves the simultaneous bending of several Golgi cisternae to form a cup-shaped isolation structure made up of several concentric layers of double membrane (Fig. 2; Fig. S2 and Movie 1). Although the precise mechanism of their formation still remains to be established, the presence of Golgi-derived vesicles in the vicinity of cisternal rims leads us to speculate that the elongation and sealing of the isolation structures is made possible by the fusion of these vesicles to the rims of each individual cisterna (Fig. 2A, arrow). These structures engulf different cytoplasmic components that are clearly identifiable (e.g. vesicles, Nissl bodies) (Fig. 2). In the closed state, the engulfed cytoplasmic material is completely isolated by concentric double-membraned layers (Fig. 3; Figs S2, S3 and Movie 2).

The appearance of the vesicles that could contribute to the sealing (thick membranes) (Fig. 2A) suggests that the trans-most cisternae are the direct isolation membranes. In accordance with this notion, immunolabelling for TGN38 (also known as TGOLN2; a marker for the trans-Golgi network and trans-most cisternae) and giantin (also known as GOLGB1; cis-mid Golgi marker) revealed these two proteins were preferentially associated with the inner and outer layers, respectively (Fig. 4A). Confocal microscopy analysis of immunolabelling of TGN38 identified discrete perinuclear structures with dimensions that were compatible with the central area of the engulfing structures. Consequently, staining for giantin labelled similar structures with larger diameters and non-labelled central areas. Double labelling for TGN38 and GM130 (also known as GOLGA2; a cis-Golgi marker) corroborated that the central part of these structures was derived from the trans-Golgi (Fig. 4B). A schematic representation of the model proposed for the formation of these Golgi structures is depicted in Fig. 4C.

Proteolytic enzymes are commonly provided to autophagosomes through fusion with late endosomes or lysosomes (Mizushima et al., 2008; Nixon et al., 2005). Both of these organelles are recognizable in the proximity of or in close contact with the engulfing Golgi (Fig. 1C,D and Fig. 3). Confocal microscopy analysis confirmed that the discrete structures that were labelled for giantin and TGN38 were found close to LAMP1 labelling (lysosomal marker) (Fig. 4B). The immunolabelling for cathepsin B (a typical lysosomal protease) revealed the presence of the protease in the structures (Fig. 4D). Moreover, confocal images confirmed that the labelling for cathepsin B was found to be enriched in the proximity of Golgi markers (Fig. 4D). Thus, degradative enzymes could be provided in the absence of a complete fusion between the Golgi structures and lysosomes through a kiss-and-run mechanism that is similar to that proposed for material transfer between endosomes and lysosomes (Luzio et al., 2007). It is well established that degradative enzymes are delivered to lysosomes through the Golgi complex (Kornfeld and Mellman, 1989; Traub and Kornfeld, 1997); therefore, it is also possible that enzymes that have not yet exited the Golgi complex could somehow accomplish their degradative function through an unconventional mechanism. The presence of engulfed cytoplasmic material was confirmed by the presence of COX IV labelling (a mitochondrial inner membrane protein) inside them (Fig. 4E). We confirmed that the number of conventional autophagosomes was low in striatal neurons by observing the diffuse cytoplasmic pattern of LC3 (the most widely monitored autophagosomal protein; MAP1LC3B isoform) in confocal images and the small ratio of lipidated to non-lipidated LC3 detected by western blotting (Fig. S1C). These results suggest that these Golgi structures might act as degradative intermediates, but further work is required to definitively establish this notion.

The identification and characterization of the engulfing Golgi would have been difficult by using conventional molecular and cell biology methods, thereby reinforcing the power of new 3D imaging technologies and optimal preservation of samples to address fundamental questions in cell physiology and architecture. Moreover, our findings have provided knowledge about the complexity and diversity of neuronal homeostasis. The fact that post-mitotic neurons need to adapt rapidly to changing conditions supports the implication of the Golgi complex, a dynamic organelle, in the degradation of cytoplasmic contents through the formation of the described engulfing Golgi structures.

Fig. S4 shows a scheme of the procedure for sample preparation. C57BL/6J mice (Jackson Laboratories) were kept as part of stable colonies in the animal facility with food and water available ad libitum. Experiments complied with Spanish and European legislation and Spanish National Research Council (CSIC) ethics committees on animal experimentation. Two animals of each age (2, 6 and 9 months) were processed in three different experiments. Mouse brains were dissected immediately post-mortem and 200-µm sagittal slices were obtained from the right hemisphere by using a tissue slicer (Stoelting, Co., Dublin, Ireland). Slices were promptly transferred to a device in which they were maintained under constant oxygenation (95% O₂ and 5% CO₂) in artificial cerebrospinal fluid (124 mM NaCl, 2.7 mM KCl, 1.25 mM KH₂PO₄, 10 mM glucose, 26 mM NaHCO₃ and 2 mM CaCl₂). Slices were then consecutively processed one by one. A sample punch provided by Leica Microsystems (Wetzlar, Germany) was used to extract striatal samples from the slices after incubating them in a cryoprotectant solution (10% Ficoll) for 2 min. Samples were placed on a flat specimen carrier (0.5-mm thick, 1.5 mm in diameter, 200-μm deep, Leica Microsystems, #16706898) and high-pressure-frozen in a Leica EMPACT2 device. For a given animal, the entire processing time (from the slicing to HPF) varied from 5 min (first slice processed) to a maximum of 30 min (last slice). HPF-processed samples were maintained in liquid nitrogen until further processing.

For control purposes, samples from selected slices were processed by using conventional electron microscopy methods. The protocol included fixation with 4% paraformaldehyde and 2% glutaraldehyde, followed by post-fixation with 1% OsO₄ and 0.8% potassium ferrocyanide, incubation with 2% uranyl acetate, dehydration and embedding in epoxy resin.

Flat carriers containing the frozen samples were placed inside a flow-through ring (Leica Microsystems, cat. # 16707157), and this was placed in a reagent bath (Leica Microsystems, cat. # 16707154) for the substitution and embedding steps in the Leica AFS2 EM FSP system. For the freeze-substitution to methanol, the following consecutive incubations in 0.5% uranyl acetate (UrAc) were performed: overnight at −90°C, 8 h at −90°C, overnight at −90°C and 8 h at −90°C, changing to fresh solution between incubations. Temperature was then raised from −90°C to −40°C for 7 h, and then samples were incubated at −40°C for 30 h. Finally, three washes with 100% methanol were made to remove remaining UrAc.

For resin embedding, the following consecutive incubations were performed: 3:1 mixture of methanol:Lowicryl resin HM20 (Lowicryl HM20 kit, Polysciences, PA; cat. #15924) for 1 h at −40°C, 1:1 mixture for 1 h at −40°C, 1:3 mixture for 2 h at −40°C, resin for 15 min at −40°C and fresh resin overnight at −40°C. After an extra change to fresh resin, the polymerization was performed at −40°C under UV light for 48 h. The temperature was then raised from −40°C to 20°C for 4 h and then the sample was incubated at 20°C for 48 h, all under UV light. Resin capsules containing the sample were detached from the flat carriers by hot (50°C) and cold (liquid nitrogen) cycles. 250-nm-thick sections were obtained using a Leica Ultracut EM-UC6 ultramicrotome. The sections were placed on Quantifoil^® S7/2-200 mesh gold (Au) grids (Quantifoil Micro Tools GmbH, Grossloebichau - Jena, Germany). We observed an average of 15 grids per animal, with four 250-nm-thick sections per grid. To check the integrity of the tissue and to identify structures of interest for further electron tomography analysis, samples were observed with a JEOL JEM-1011 100 kV transmission electron microscope.

Images from 250-nm sections were obtained by using a FEI Tecnai G2 200 kV transmission electron microscope. The sections were subjected to tilt-series acquisition, where a set of images was taken by tilting the sample around a single-tilt axis in the range ±60° at an interval of 1°. The acquired images were rescaled by a four-fold binning factor to yield a pixel size at the specimen level in the range 1.56–2.33 nm. Standard electron tomography processing then followed (Fernandez, 2012), using patch-tracking-based alignment with IMOD (Kremer et al., 1996) and Weighted-BackProjection reconstruction with Tomo3D software (Agulleiro and Fernandez, 2011). The resulting tomogram was subjected to noise reduction using slight Gaussian filtering (Martinez-Sanchez et al., 2011). 3D modelling relied on automated segmentation (Martinez-Sanchez et al., 2014) followed by manual refinement of membranes unambiguously identified in xy-slices of the tomograms. Faint membranes perpendicular to the electron beam direction were left out of the segmentation. Manual tracing and 3D visualization was performed using Amira software (FEI Visualization Sciences Group).

Grids containing 250-nm sections were floated on top of drops and sequentially incubated with 1% periodic acid for 10 min, three times with H₂O for 3 min, 5% sodium periodate for 20 min and three times with H₂O for 3 min. Grids were then blocked with a filtered solution containing 10% fetal bovine serum (FBS) and 0.1% Triton X-100 for 1 h. Primary antibodies – rabbit anti-giantin (PRB-114c, Covance), anti-TGN38 (NB110-40768, Novus Biologicals) and anti-LC3 (NB100-2220, Novus Biologicals), and mouse monoclonal anti-cathepsin-B (ab58802, Abcam) antibodies – were used at 1/100 dilution and incubated at 4°C overnight. An additional blocking step of 15 min was performed before the incubation at room temperature with a 1/40 dilution in 10% FBS of the secondary antibody for 45 min. A 10-nm-gold-conjugated secondary antibody [BB International EM, goat F(ab’)2 anti-rabbit IgG] and a 5-nm-gold-conjugated secondary antibody [BB International EM, goat F(ab’)2 anti-mouse IgG+IgM] were used. The immunolabelling of plastic sections is inefficient (58% of the analyzed structures were labelled for TGN38 and 46% for cathepsin B). The number of gold particles associated to each structure was also very low. Samples were observed by using a JEOL JEM-1011 100 kV transmission electron microscope.

Samples were processed as described previously (Fernandez-Fernandez et al., 2011). Primary antibodies used were: rat anti-LAMP1 (1D4B, Developmental Studies Hybridoma Bank; 1/50); rabbit antibodies against TGN38 (NB110-40768 and NBP1-03495SS, Novus Biologicals; 1/200), giantin (PRB-114c, Covance; 1/1000) and LC3 (NB100-2220SS, Novus Biologicals; 1/300); goat polyclonal anti-cathepsin-B (SC-6493, Santa Cruz; 1/250); and mouse monoclonal antibodies against GM130 (610822, BD Transduction; 1/200) and CoxIV (#11967, Cell Signalling; 1/300). The secondary antibodies used were: goat anti-rat IgG conjugated to Alexa-Fluor-594, goat anti-rabbit IgG conjugated to Alexa-Fluor-488 and goat anti-mouse IgG conjugated to Alexa-Fluor-594 for Fig. 4B; donkey anti-rabbit IgG conjugated to Alexa-Fluor-488 and donkey anti-goat IgG conjugated to Alexa-Fluor-546 for Fig. 4D; goat anti-mouse IgG conjugated to Alexa-Fluor-488 and goat anti-rabbit IgG conjugated to Alexa-Fluor-594 for Fig. 4E; and goat anti-rabbit IgG conjugated to Alexa-Fluor-488 for Fig. S1C. Secondary antibodies were obtained from Life Technologies and were used at a 1/500 dilution. Nuclei were stained with DAPI (Sigma, 1/200 dilution). Images were taken on a Leica TCS SP5 confocal multispectral system. All the images are single confocal sections taken with a 63×1.4 NA objective. The optical section thickness was 0.773 µm and the pinhole was 1 AU. Single confocal sections belong to series taken with a step size of 0.7 µm.

Samples were processed as described previously (Fernandez-Fernandez et al., 2011). Membranes were incubated with a rabbit-polyclonal anti-LC3 antibody (NB100-2220SS, Novus Biologicals) at a 1/500 dilution and then with horseradish peroxidase (HRP)-conjugated anti-rabbit IgG secondary antibody (NA9340V, GE Healthcare) at a 1/1000 dilution. The peroxidase reaction was developed with Clarity™ western ECL substrate (Bio-Rad).

We thank Dr Cristina Tomas-Zapico and Dr Enrique Gabande-Rodriguez for helpful discussions and the Electron Microscopy and Animal Facilities at our institute for invaluable help.