Gap junction internalization and processing in vivo: a 3D immuno-electron microscopy study Free

Gap junctions are arrays of permeable channels between two cells that have well-established roles in intercellular signaling (Nielsen et al., 2012). The basic structural unit is the transmembrane protein connexin. Six connexins assemble to form the connexon, which is a pore within the membrane. A connexon in one cell docks head-on with a connexon in a neighboring cell to form a channel between the two cells. Large gap junctions may consist of hundreds or thousands of channels packed densely in a patch a few micrometers in diameter (Larsen, 1977).

Connexons are added to the plasma membrane via small post-Golgi vesicles, followed by docking between two cells (Laird, 2006), but how connexons are removed from the membrane and their subsequent fate are incompletely understood. To turn over the gap junction, cells could undock connexons and then endocytose them in small parcels. Instead, connexons remain docked, and the gap junction is taken up by one of the two cells (Laird, 2006; Falk et al., 2016). This was first suggested by electron microscopists who interpreted circular gap junction profiles (first called annular gap junctions) as internalized gap junctions (Espey and Stutts, 1972; Merk et al., 1973). This interpretation was convincingly corroborated by live imaging of GFP-tagged connexins, which showed formation of vesicles of comparable size (Jordan et al., 2000; Piehl et al., 2007). The internalized gap junction structure is now often called a connexosome (Laird, 2006).

Gap junction internalization is therefore a type of endocytosis, in which the plasma membrane of the neighboring cell remains attached to the endocytosed plasma membrane (Heck and Devenport, 2017). Gap junction internalization can also be considered to be a form of trogocytosis (Joly and Hudrisier, 2003) in which a portion of the plasma membrane and cytosol of the neighboring cell is transferred to the engulfing cell (Fig. 1A). While there is evidence that internalized gap junctions are degraded (Leithe et al., 2006; Falk et al., 2016), there is a possibility that further processing after the initial engulfment is involved in other modes of cell–cell communication (see Discussion).

In a previous study (Norris et al., 2017), we addressed several methodological issues that have limited electron microscopic studies of gap junctions in the past. Mouse ovarian follicles were high-pressure frozen in order to preserve structure better than chemical fixation, which involves diffusion of aldehydes through the cell membranes and cross linking of proteins, during which abnormal processes may occur (Murk et al., 2003a). The frozen tissue was freeze-substituted, embedded in Lowicryl, sectioned, then immunolabeled with an antibody against connexin 43 (Cx43, also known as GJA1). This allowed us to unambiguously identify gap junctions. Serial sections were then imaged to obtain three-dimensional information; this distinguished between internalized connexosomes and gap junctions in the process of invagination. Likewise, a round profile in a single section could be a vesicle or an invagination, or an apparently empty vesicle could contain intraluminal vesicles; serial sections distinguished these alternatives.

Here, we investigate the fate of internalized gap junctions by examining Cx43 localization. Ovarian follicles were high-pressure frozen, serial sectioned and immunolabeled for Cx43. A volume corresponding to ∼35 cells was comprehensively documented; every Cx43-containing membrane was categorized, counted and measured. We found a large number of Cx43-labeled structures that we interpret as modifications of the connexosome.

Mouse ovarian antral follicles were high-pressure frozen 30 min after exposure to luteinizing hormone (LH). This tissue was used because it has large numbers of Cx43 gap junctions (Okuma et al., 1996; Norris et al., 2008, 2017; Baena et al., 2020), which are caused to internalize in response to hormone (Larsen et al., 1987). As in our previous study, the follicles were embedded in Lowicryl, serial sectioned, then labeled with primary antibody against Cx43 and gold-labeled secondary antibody (Norris et al., 2017). The anti-Cx43 antibody is a commercial polyclonal rabbit antibody generated against the cytoplasmic C-terminal tail, and is used for detecting total Cx43 (Sosinsky et al., 2007). We used a different scanning electron microscope in order to obtain higher resolution images than those in our previous study. This provided clearer images of the classic gap junction ultrastructure (Fig. 1B–D).

We will use the term connexosome to mean a double membrane vesicle completely detached from the plasma membrane in which the outer and inner membranes are contacting each other and are labeled with Cx43 throughout the periphery of the vesicle. Connexosomes were the most abundant Cx43-labeled structure in the cytoplasm, but there were also many other membranous structures that contained Cx43 (Movie 1), which we interpreted to be modifications of the connexosome.

We characterized these Cx43 structures by classifying and counting them in a defined volume. By doing so, we gained information on the relative abundance of various forms, which could be useful in deducing dynamics. We analyzed two volumes of mural granulosa cells that were each 60 µm×60 µm in the x-y plane (imaged at 5 nm per pixel), in 73 sections of 60 nm thickness. Thus each volume was 15,768 µm³. This volume corresponds to approximately 35 spherical cells with a 12 µm diameter.

In this volume, there were 86 gap junctions. Of these, 75 were located in a planar patch. The 11 others were present on highly in-folded membranes, which we refer to as invaginating gap junctions. There were 132 internalized structures. Of the 132 Cx43-labeled structures, 76 were connexosomes and 56 appeared to be connexosomes that had undergone processing.

Direct imaging of living cultured cells has shown that connexosomes are formed by internalization of entire gap junctions, and that connexosomes can undergo fission (Piehl et al., 2007; Bell et al., 2019). We examined the high-pressure frozen and serial-sectioned ovarian follicles for connexosome formation. We also measured the surface areas of all of the connexin 43 containing membranes, because this information could be relevant to connexosome formation (Fig. 1E).

The average surface area of the 75 ‘flat’ gap junction plaques in the volume was 1.9±0.1 µm² (mean±s.e.m.), with a range of 0.36–6.8 µm². Serial sections showed that most were disc shaped, corresponding to a disc of average diameter 1.56 µm. The 11 invaginated gap junctions looked like they could form connexosomes. These were larger than the ‘flat’ gap junctions, with an average surface area of 3.4±0.5 µm² and a range in area of 1.1–6.8 µm² (N=11; P<0.0001; one-way ANOVA with Tukey's correction).

Connexosomes, on the other hand, had an average surface area of 1.20±0.1 µm² (mean±s.e.m.) and a range in area of 0.14–2.9 µm² (N=76). The average connexosome area thus was 64% of the average gap junction area (Fig. 1E,F).

It seems likely that invaginated gap junctions were frozen in the process of internalization and are therefore the source of connexosomes. The surface areas of invaginated gap junctions are significantly larger than connexosomes, based on one-way ANOVA analysis. In live imaging of Cx43–GFP-expressing cells, connexosomes have been observed to undergo fission (Piehl et al., 2007; Bell et al., 2019). In ovarian follicle cells, the connexosomes may also undergo fission after internalization; alternatively, they could form from a portion of a larger invaginated gap junction.

Of the 132 Cx43-labeled internalized structures, 56 appeared to be modified connexosomes. To describe the modifications, it is necessary to keep track of the origins of the membranes and compartments of the unmodified connexosomes. The outer membrane comes from the plasma membrane of the host cell, whereas the inner membrane comes from the plasma membrane of the neighboring cell. The tiny ‘gap’ between the two membranes originally was the extracellular space. The compartment within the inner membrane is derived from the cytoplasm of the neighboring cell.

The initial modification appears to be fusion of a connexosome with another vesicle. In 7 of the 56 modified connexosomes, a patch of unlabeled outer membrane bulged outward from a labeled inner membrane (Fig. 2A,B; Movie 2). This is what would be expected if a vesicle had fused with the outer membrane. The space between the two membranes was either dark (Fig. 2B) or clear (Fig. 2C), suggesting that different types of vesicles were involved. Cx43 labels the inner membrane in areas where the inner and outer membranes have separated. In these regions, the connexons must have become undocked.

The inner and outer membranes had become more separated in 7 of the 56 modified connexosomes, and small vesicles were present in this enlarged space (Fig. 2C). The small vesicles were generally not labeled with Cx43, and their diameters averaged 72 nm. This corresponds closely to the reported diameters of intraluminal vesicles (50– 80 nm) found in multivesicular endosomes (Murk et al., 2003b; Hanson and Cashikar, 2012; Scott et al., 2014). The inner membrane was identifiable and labeled, whereas the outer membrane was much less labeled (Fig. 2B–D; see Movie 2 for the three sets of serial sections that document this pattern of Cx43 labeling). In some striking examples, the inner and outer membranes were completely separated, and no Cx43 label was present in the outer membrane (Fig. 2D). Although it is unlikely that Cx43 in the still intact outer membrane has been completely degraded, it has apparently become modified so that the antibody no longer recognizes it.

In the remaining 41 of the 56 modified connexosomes, there was an outer membrane similar in diameter to that of unmodified connexosomes that had little to no Cx43 labeling. Furthermore, instead of a single inner membrane labeled with Cx43, there were various different sized vesicles. Generally, the larger vesicles were labeled with Cx43, and smaller vesicles were unlabeled (Fig. 3A,B; Movie 3).

The diameters of the Cx43-labeled vesicles in modified connexosomes averaged 123 nm, and ranged from 38 nm to 366 nm (Fig. 3C). Based on the wide range of diameters and the presence of Cx43, these vesicles seem likely to have formed by fission of the inner connexosome membrane. For comparison, we also measured small vesicles within structures that lacked Cx43 labeling (Fig. 3D). Unlabeled vesicles were significantly smaller (P<0.0001; unpaired t-test), and also in the range of reported sizes for intraluminal vesicles, averaging 52 nm in diameter, with a range of 36–70 nm (Fig. 3C). In addition to subdivision of the inner membrane, some of the modified connexosomes had short tubules extending from the outer membrane into the cytosol (see arrows in Fig. 3A,B), as is often seen in endosomes (Klumperman and Raposo, 2014).

To test the possibility that the electron-dense regions of some modified connexosomes represented lysosomal fusion, we labeled adjacent sections for either connexin 43 or for cathepsin B, which is a degradative enzyme abundant in lysosomes (Klumperman and Raposo, 2014). We found that cathepsin B localized to electron-dense vesicles that resembled lysosomes in granulosa cells (Fig. 4A). Cathepsin B also localized to electron-dense regions of some modified connexosomes (Fig. 4B,C), indicating that lysosomes are involved in processing of some connexosomes.

By applying new methods for 3D protein localization and imaging with scanning electron microscopy, we categorized, counted and measured every Cx43-containing structure in a defined volume of ∼30,000 µm³. Compared to previous approaches, this provides comprehensive data on the types of structures that contain Cx43, and establishes several undescribed features of how internalized gap junctions are processed within an intact tissue.

Live-cell imaging studies of several cultured mammalian cell lines expressing Cx43–GFP chimeras have shown that either entire gap junctions are internalized, followed by fission (Piehl et al., 2007; Bell et al., 2019), or that a small portion of the gap junction center is internalized (Falk et al., 2009).

The 3D data from high-pressure frozen tissue allowed us to look for internalization intermediates and also to make the first systematic measurements comparing gap junction and connexosome areas. Most gap junctions were disk shaped on a flattened region of plasma membrane (N=75). There were 11 invaginated gap junctions, and their average areas were larger than those of the flat gap junctions. It seems likely that the largest gap junctions begin to invaginate and are the source of connexosomes in this tissue. The internalization of whole invaginated gap junctions should produce correspondingly large connexosomes, but connexosomes as a group are smaller than gap junctions. This is consistent with fission occurring soon after internalization or with the formation of multiple small connexosomes from a large invaginating gap junction.

The initial connexosome modification was a partial separation of the inner and outer membrane while the rest of the gap junction is intact. The separated outer membrane bulges out and lacks Cx43, while the inner membrane retains it (Fig. 2B,C). It seems likely that a vesicle has fused with the connexosome, perhaps at a bare patch left over from the internalization process (Falk et al., 2009, 2014). What vesicles initially fuse with the connexosomes? Many appear to be clear vesicles, which is consistent with endosomes (Murk et al., 2003a), but several vesicles were dark, which is characteristic of lysosomes (Klumperman and Raposo, 2014). This was confirmed by immunolocalization of cathepsin B to electron-dense regions of some processed connexosomes.

Further modifications involve two different processes. One is complete separation of the two membranes accompanied by a loss of Cx43 immunogenicity in the outer membrane. Endosome or lysosome fusion could cause the connexons of the gap junction to undock by lowering pH (Falk et al., 2014). Because the Cx43 antibody epitope is the cytosolic C-terminal tail, the membrane separation may result in cleavage or covalent modification of this region. The other process is the generation of numerous smaller Cx43-positive and -negative vesicles within the modified connexosome. It is important to note that these vesicles are all present in the space directly enclosed by the outer connexosome membrane. This space originates from the extracellular space between the two cells. The unlabeled vesicles seem likely to derive from the largely unlabeled, modified outer membrane, whereas the labeled vesicles seem likely to come from the labeled inner membrane. The unlabeled vesicles resemble in size the intraluminal vesicles of multivesicular endosomes that are produced by the ESCRT machinery (Hanson and Cashikar, 2012; Scott et al., 2014). The Cx43-labeled vesicles have a larger size range and their topology is consistent with an ESCRT-driven process operating from the cytosolic side of the inner connexosome membrane (either ‘fission’ to produce larger vesicles or ‘outward budding’ to produce small vesicles). It has been shown that ESCRT machinery can bind to ubiquitylated connexins (Auth et al., 2009), and this machinery could be transferred during connexosome formation.

Fig. 5 summarizes the connexosome modifications we have described. Fusion with a vesicle (Fig. 5, step 1) triggers the separation of the inner and outer connexosome membranes (Fig. 5, step 2). Meanwhile, the uncoupled inner membrane labeled with Cx43 undergoes fission (Fig. 5, step 3). The ultimate fate of the modified connexosome is either degradation or further modification (Fig. 5, step 4), as occurs in some endosomes (see below).

In contrast to previous studies, we did not find evidence for connexosome degradation by autophagy. Autophagosomes engulf internalized gap junctions in the equine hoof wall (Leach and Oliphant, 1984), canine ventricular myocardium (Hesketh et al., 2010), HeLa cells and mouse embryo fibroblasts (Lichtenstein et al., 2011; Fong et al., 2012). In mouse liver cells, connexins are degraded by autophagy (Bejarano et al., 2012). However, we did not observe intermediates resembling a phagophore or an autophagosome in this tissue under these conditions.

An earlier study by Leithe et al. (2006) investigated connexosome modifications in cultured rat liver epithelial cells treated with phorbol ester. As in our study, they observed that the connexosome transformed to have a single outer membrane with smaller enclosed vesicles. Based on immunolocalization of endosomal markers and Cx43 by fluorescence and by gold labeling of cryosectioned sucrose-embedded tissue, they concluded that this was a ‘multivesicular endosome’, and that it was destined to fuse with a lysosome and undergo degradation. Falk et al. (2012) later noted that phorbol ester induces phosphorylation of connexin and proposed that unphosphorylated connexins target a connexosome for autophagy while phosphorylated connexins target a connexosome to an endolysosomal pathway. Our observations are consistent with this proposal, because Cx43 is highly phosphorylated at the time point we studied (Norris et al., 2008).

The term multivesicular endosome is now associated with ‘signaling hub’ rather than being considered solely a vehicle for degradation. Intraluminal vesicle (ILV) formation within multivesicular endosomes can control levels of receptor signaling (Dobrowolski and De Robertis, 2012; Scott et al., 2014). In some situations, such as in antigen presentation (Kleijmeer et al., 2001), ILVs can fuse with an endosome-limiting membrane (‘back-fusion’; Bissig and Gruenburg, 2014). Membrane fusion events similar to back-fusion result in the release of nucleocapsids (Le Blanc et al., 2005; Grove and Marsh, 2011) or cargo from endocytosed extracelluar vesicles (Joshi et al., 2020) to the cytoplasm of a host cell. Notably, these forms of membrane fusion typically depend on acidic pH (Grove and Marsh, 2011; Joshi et al., 2020). This may relate to the broad roles of lysosomes, which are also now known to act as metabolic hubs (Settembre et al., 2013; Perera and Zoncu, 2016).

Modified connexosomes closely resemble multivesicular endosomes in appearance but differ in one remarkable way: the vesicles within are composed of membranes and cytoplasm from the neighboring cell. If, as occurs in multivesicular endosomes, enclosed vesicles fuse with the outer membrane, this would mix membranes and cytoplasm from the neighboring cell. Such a mechanism would provide a novel, gap junction-dependent pathway for two neighboring cells to interact (Fig. 5, step 4). For instance, this could explain a puzzling result in immune cells, where Cx43 gap junctions are implicated in the transfer of MHC-II-bound antigens from macrophages to dendritic cells (Mazzini et al., 2014).

C57Bl/6J female mice (Mus musculus) from The Jackson Laboratory (Bar Harbor, ME) were used in this study. The use of mice was in accordance with guidelines published by the National Institutes of Health and approved by the UConn Health Institutional Animal Care and Use Committee. Mice were housed in a temperature and humidity controlled environment with water and food available ad libitum.

Anti-connexin 43 polyclonal antibody produced in rabbit (#C6219; Lot 045M4882V; Sigma-Aldrich, St. Louis, MO) was used at 1:100 for a final concentration of ∼7 µg/ml. Anti-cathepsin B (#31718; Lot 1; Cell Signaling Technologies, Danvers, MA) was used at 1:100 for a final concentration of∼0.5 µg/ml. For secondary antibody, 10 nm gold particle-conjugated goat anti-rabbit IgG F(ab’) 2 fragment (#25362; Electron Microscopy Sciences, Hatfield, PA) was used at a dilution of 1:20.

Ovarian follicles, 360–400 µm in diameter, were manually isolated from 25-day-old mice. The follicles were cultured on Millicell membranes (Millipore; PICMORG50) in MEM alpha medium, as described by Norris et al., 2008, with the replacement of 3 mg/ml bovine serum albumin (MP Biomedicals; #103 700) for serum.

After 25 h of culture, follicles were treated with 10 µg/ml ovine luteinizing hormone (oLH-26; National Hormone and Pituitary Program) for 30 min, then transferred to brass specimen carriers and high-pressure frozen with an EMPACT 2 (Leica Biosystems, Buffalo Grove, IL).

Freeze substitution and embedding was done as described by Rubio and Wenthold (1997), with some modifications. Samples were freeze-substituted with 2.5% uranyl acetate (Electron Microscopy Sciences) in dry methanol for 32 h at −90°C in an AFS 2 freeze substitution unit (Leica Biosystems). The temperature was then raised 5°C per hour to −45°C. Samples were then rinsed in methanol and infiltrated with Monostep Lowicryl HM-20 (Electron Microscopy Sciences) in increasing concentrations of 1:1, 2:1, and pure Lowicryl HM-20 for 2 h each. After an overnight infiltration with Lowicryl HM-20, samples were polymerized in a fresh change of Lowicryl under ultraviolet light for 33 h at −45°C. Polymerization continued under UV light as the temperature in the AFS 2 was increased by 5°C per hour to 0°C, then held at 0°C for a further 33 h. When samples were removed from the AFS 2, they were pink in color and were left to polymerize at room temperature until the pink hue was gone 2 d later.

Ultrathin sections (60 nm) of Lowicryl HM-20-embedded follicles were cut on a UC-7 ultramicrotome (Leica Biosystems) with a diamond knife (Diatome, Hatfield, PA). The sections were picked up by an automated tape collector on glow-discharged kapton tape (Terasaki et al., 2013; Kasthuri et al., 2015; Baena et al., 2019).

For immunostaining, ribbons of follicle sections on kapton tape were cut to lengths of ∼7.5 cm and attached to a sheet of parafilm with double-sided carbon tape (Electron microscopy Sciences; #77816). Sections were rehydrated with 1× phosphate-buffered saline (PBS; Life Technologies, Grand Island, NY) and blocked in 5% normal goat serum (Invitrogen, Frederick, MD) in a solution of 1% bovine serum albumin in PBS.

Following an overnight incubation at 4°C in primary antibody, sections were rinsed three times for 5 min each in PBS, then rinsed once in 1% BSA in PBS. Next, secondary antibody diluted at 1:20 was applied to sections 1 h at room temperature. Sections were then rinsed with 1× PBS followed by Milli-Q filtered water and dried overnight. When adjacent sections of tissue were immunolabeled for Cx43 or cathepsin B, sections were stained in separate dishes to avoid any mixing of antibody (Norris et al., 2017). Sections placed back in their original order were post-stained with 5% uranyl acetate in 50:50 methanol:water for 5–7 min, then rinsed generously in water.

Immunolabeled sections on tape were attached to a 10 cm diameter silicon wafer (University Wafer, South Boston, MA) with double-sided carbon adhesive tape (Electron Microscopy Sciences). Wafers were carbon coated (Denton, Moorestown, NJ) and first imaged on a Sigma field emission scanning electron microscope (Zeiss, Thornwood, NY) using a backscatter detector, as described in Norris et al., 2017. Two volumes of mural granulosa cells were imaged at 5 nm/pixel resolution, with a field of view of 60 square micrometers.

Original low-resolution images obtained on the Zeiss Sigma were aligned using the Register Virtual Stack Slices macro (FIJI), then larger files were aligned and diced for convenient viewing with a custom program (Piet, provided by Duncan Mak and Jeff Lichtman, Harvard University). These files were used to track all Cx43-labeled internalized structures. When internalized structures looked complex, they were re-imaged using a higher resolution electron microscope as described below.

Higher resolution images of the sections were taken on an FEI Verios 460L field emission scanning microscope with a backscatter detector. Using immersion mode, sections were imaged at 5.0 kV, with a beam current of 0.4 or 0.8 nA. Images had a dwell time of 3.0 μs, with a line integration of 5 or 10, and drift correction. Image sizes of 2048×1768 pixels were taken with horizontal field widths between 2.76 and 7.89 µm to generate images with resolutions between 1 and 4 nm/pixel. Under these conditions, we could detect the characteristic pentalaminar appearance of gap junctions in some sections. To minimize beam damage, it was beneficial to first use a ‘beam shower’ over the sections by scanning at a lower magnification, with a higher beam current, and a faster scan rate for 10–20 s.

Image stacks were created by aligning serial images using the Register Virtual Stack Slices macro in Fiji. Stacks were imported into the TrakEM2 program in the Fiji software to measure surface areas. Separate ‘area lists’ were created for reconstructions and measurements of individual gap junction plaques or connexosomes. Images of 15 nm/pixel were traced. Using the ‘measure’ tool, the column labeled ‘AVG-s’ (average smooth) was used as an estimate of surface area for both connexosomes and gap junction plaques. Default units of pixels squared were converted into micrometers squared. To calculate the surface areas of gap junction plaques, the ‘AVG-s’ values were first divided by two, as the initial value represented both faces of the gap junction plaque.

Surface area measurements of gap junctions, invaginated gap junctions and connexosomes were compared using a one-way ANOVA with Tukey's correction (Prism software, GraphPad).

We thank Rindy Jaffe and Matthias Falk for critical review of the manuscript and useful discussions. We thank Art Hand, Maya Yankova and Maria Rubio for technical advice. We thank Valentina Baena, Tracy Uliasz and other members of the Jaffe and Terasaki laboratories for technical assistance and help with collecting samples.