Connexin-specific distribution within gap junctions revealed in living cells

A number of oligomeric proteins, including acetylcholine receptors, aquaporins, tight junction occludin and claudins, and gap junction channels, arrange into densely packed clusters, arrays or strands in the plasma membrane (McNutt and Weinstein, 1970; Friend and Gilula, 1972; Rash et al., 1974; Axelrod et al., 1976; Dunia et al., 1987; Peng, 1983; Yang et al., 1996; Tsukita and Furuse, 1999; Tsukita and Furuse, 2000). However, the signals that regulate their packing are poorly understood. Ultrastructural analyses revealed that gap junction channels, in particular, arrange extensively into two-dimensional sheets, or so-called gap junction plaques (McNutt and Weinstein, 1970; Friend and Gilula, 1972; Rash et al., 1974; Ginzberg and Gilula, 1979). Less than a dozen to many thousand individual channels can be combined into a single plaque that can extend from several nanometers to several micrometers in diameter (McNutt and Weinstein, 1970).

Gap junction channels represent double membrane protein structures that form by the head-to-head docking of two half-channels, termed connexons, to create hydrophilic pores across the membranes (Makowski et al., 1977). The channels mediate direct cell-to-cell communication by allowing the passage of small biological molecules (up to approximately 1 kDa) from one cell to the other (Gilula et al., 1972). Each connexon is composed of six polytopic transmembrane protein subunits, termed connexins (Cx). Therefore, only a single type of protein is required, and is sufficient, to construct gap junction channels. At least 14 different connexin isotypes have now been cloned and sequenced from mice. All represent structurally conserved non-glycosylated members of a multigene family that mainly differ in their C-terminal domain (see Bruzzone et al., 1996a; Bruzzone et al., 1996b; Goodenough et al., 1996; Kumar and Gilula, 1996, for recent reviews). Although the different connexin isotypes exhibit a distinct tissue distribution, many cell types express more than one connexin isotype. This may allow the assembly of hetero-oligomeric connexons constructed from different connexin isotypes, in addition to the assembly of homo-oligomeric connexons constructed from a single connexin isotype.

Previous studies have shown that the connexin subunits are cotranslationally integrated into the endoplasmic reticulum (ER) membrane in a process that is similar to that described for other transmembrane proteins (Falk et al., 1994). There, the connexins adapt their native transmembrane configuration, traversing the membrane bilayer four times with N- and C-termini located in the cytoplasm (Falk and Gilula, 1998). Further studies demonstrated that connexins can assemble into functional hexameric connexons in the ER membrane (Falk et al., 1997). Subcellular fractionation studies (Musil and Goodenough, 1991; Falk et al., 1994) and immunocolocalization analyses (Musil and Goodenough, 1991; Laird et al., 1995) suggest that connexins pass through the Golgi apparatus to reach the plasma membrane. Connexons are believed to dock in the plasma membrane to form the complete double-membrane channel structures, and channels cluster into gap junction plaques (reviewed by Yeager et al., 1998; Falk, 2000a; Falk, 2000b).

Many questions related to the biosynthesis and turnover of gap junction channels remain, however. Specifically, how and in which oligomeric or clustered stage are the connexins transported from the Golgi to the plasma membrane; where are the connexons integrated into the membrane; how do connexons in apposing membranes align and dock to initiate channel formation; what is the organization of the channels and why do they cluster; how is biosynthesis and function regulated and which signaling molecules are involved; and finally, what mechanisms are involved in gap junction turnover?

To address these questions in living cells, we tagged the gap junction proteins α₁(Cx43), β₁(Cx32) and β₂(Cx26) with the autofluorescent tracer green fluorescent protein (GFP), and its cyan (CFP) and yellow (YFP) color variants, and combined this reporter technology with high-resolution fluorescence deconvolution (DV) microscopy, dual-color live cell and time-lapse imaging and three-dimensional volume rendering. Here I present (1) an extensive analysis of the tagged connexins after expression in the HeLa model cell-line, (2) a detailed three-dimensional view of entire gap junctions as they appear in the plasma membranes of living cells, and (3) how channels assembled from different connexin isotypes distribute within gap junction plaques.

cDNA sequences of the autofluorescent reporter proteins CFP, GFP and YFP were fused in-frame to the C-terminus of rat α ₁ (Cx43), human β ₁ (Cx32) and rat β ₂ (Cx26) cDNAs. The authentic connexin stop codons were deleted by PCR mutagenesis using proofreading native PFU Polymerase (Stratagene, La Jolla, CA, USA). Amplified connexin cDNAs were cloned into the EcoRI and BamHI restricted vectors pEGFPN1, pECFPN1 and pEYFPN1 (Clontech, Palo Alto, CA, USA). The resulting connexin constructs contained the 5′-untranslated connexin sequence of variable length (187 untranslated nucleotides in the α ₁ [Cx43] construct, 62 untranslated nucleotides in the β ₁ [Cx32] construct, and 48 untranslated nucleotides in the β ₂ [Cx26] construct), the complete connexin specific coding sequence, a seven-amino-acid linker (Ala-Asp-Pro-Pro-Val-Ala-Thr) and the CFP, GFP or YFP coding sequence, respectively. All constructs were verified by restriction digest and automated DNA sequencing.

Human epitheloid cervix carcinoma cells (HeLa, ATCC CCL 2; American Type Culture Collection, Rockville, MD, USA) were used throughout this study. Chinese hamster ovary cells (CHO-K1, ATCC CCL 61), African green monkey kidney cells (COS-7, ATCC CRL1650), baby hamster kidney cells (BHK-21, ATCC CCL 10), embryonic mouse fibroblasts (3T3-Swiss albino, ATCC CCL 92), adenovirus 5-transformed human embryonic kidney cell line 293 (ATCC CRL 1573), liver oval cells (T51B, a gift of S. Shen, Northwest Hospital, Seattle, USA), and Madin Darby canine kidney cells (MDCK, ATCC CCL 34) were used as controls. Cell lines were maintained in standard cell culture medium supplemented with 10% fetal calf serum, and 100 i.u. penicillin and 100 U streptomycin per ml. For transfections, cells were split the day prior to transfection to reach 60-70% confluency. Cells were transfected with Superfect ® Transfection Reagent (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions. Transfection efficiencies of 10-30% were generally obtained with HeLa cells.

HeLa cells were transiently transfected with GFP-tagged α ₁ (Cx43) and incubated overnight. Cells were fixed in 2.5% glutaraldehyde, 0.1 M sodium cacodylate buffer, pH 7.3, 1 mM CaCl₂, for 45 minutes at room temperature (RT) and, after washing 3 times with 0.1 M cacodylate buffer, with 1% osmium tetroxide in 0.1 M cacodylate buffer (for 1 hour at RT). After washing in 0.1 M cacodylate buffer (3 times) cells were treated in 0.5% tannic acid (30 minutes at RT), 1% Na₂SO₄ in 0.1 M cacodylate buffer (10 minutes at RT), followed by a rinse in 0.1 M cacodylate buffer. Cells were dehydrated in an ethanol series following an initial incubation in 1% uranyl acetate in 10% ethanol (15 minutes at RT) followed by a final treatment in 2-hydroxypropylmethacrylate (Electron Microscopy Sciences, Fort Washington, PA, USA). Cells were flat-embedded in LX-112 (Ladd Research Industries Inc., Burlington, VT, USA) and incubated overnight at 60°C. Embedded cells were mounted, thin-sectioned and examined with a Philips CM100 electron microscope.

Total cell homogenates of transfected cells were separated without boiling on 10% SDS-Mini protein gels (BioRad, Hercules, CA, USA), transferred onto nitrocellulose membranes, and blocked with 5% dry milk. Purified α ₁ (Cx43) specific polyclonal antipeptide antibodies (α1S) (Milks et al., 1988), a monoclonal antipeptide antibody specific for the C-terminal amino acid residues of β ₁ (Cx32) (β ₁SB8), and a monoclonal antipeptide antibody specific for a portion of the intracellular loop domain of β ₂ (Cx26) (β ₂JD5) (Risek et al., 1994) were kindly provided by N. B. Gilula (The Scripps Research Institute, La Jolla, CA, USA). A polyclonal anti-GFP antibody was obtained from Clontech Laboratories. Primary antibodies were used at 1:1000 dilution, horseradish peroxidase-coupled secondary goat anti-rabbit (BioRad) and goat anti-mouse (Pierce, Rockford, IL, USA), respectively, were used at 1:15,000 dilution. Bound antibodies were detected by enhanced chemiluminescence (Pierce).

HeLa cells were seeded on microscope coverglasses treated overnight with poly-L-lysine solution (Sigma, St Louis, MO, USA), transfected, and incubated overnight. Cells were fixed for 10 minutes at RT in 2% formaldehyde (Polysciences Inc. Warrington, PA, USA), permeabilized in ice-cold acetone (5 minutes at −20°C), blocked (30 minutes at RT in 1× PBS containing 1% bovine serum albumin), washed 3 times in 1× PBS, and incubated with primary antibodies (described above) diluted 1: 100 in 1× PBS containing 0.5% BSA (1 hour at RT). Monoclonal anti-Golgi p58K protein (Sigma) and polyclonal anti-calnexin antibodies (StressGen, Victoria, BC, Canada) were used at 1:100 dilution. Cy3-coupled goat anti-rabbit and goat anti-mouse (Jackson Laboratories, West Grove, PA, USA), respectively, were used as secondary antibodies at 1:200 dilution. Nuclei were stained with DAPI before mounting in Fluoromount-G (Southern Biotechnology Associates Inc., Birmingham, AL, USA). Cells were examined with a Zeiss Axiophot fluorescence microscope equipped with a 100× (Plan-Neofluar NA 1.3 oil immersion) objective lens and standard filter sets. Single, double and triple exposures were imaged on Kodak 1600 ASA slide film.

To examine dye transfer capabilities of tagged gap junctions in HeLa cells, medium was exchanged with pre-warmed 1× PBS and dishes were placed on a stage heated to 37°C mounted on an inverted epifluorescence microscope (Zeiss Axiovert 100). Clusters and pairs of cells expressing GFP-labeled gap junction plaques were selected, single cells of the clusters with gap junction plaques visible by fluorescence illumination in their plasma membranes were microinjected with 5% solutions of either Lucifer Yellow, Cascade Blue or Sulforhodamine (Molecular Probes, Eugene, OR, USA) and imaged with a 60× long-distance objective lens (Olympus LC Plan FL, NA 0.70). Representative injection experiments (n>20 per connexin isotype) are shown. Non-transfected HeLa cells in the same dishes were injected as controls (n>20).

To obtain high-resolution images and three-dimensional volume reconstructions of gap junctions in live cells a deconvolution microscope system (DeltaVision Model 283, Applied Precision Inc., Issaquah, WA, USA) was implemented. DeltaVision’s deconvolution process computationally reassigns the fluorescent blur generated by out-of-focus fluorescence and lens aberration to its source using an inverse matrix algorithm, namely the constrained iteration method developed by Agard and Sedat (Agard et al., 1989) to noticeably enhance the resolution of the images (see Shaw, 1998; Falk and Lauf, 2000, for methodology reviews). HeLa cells were seeded on poly-L-lysine-treated coverglasses (Bioptechs Inc. Butler, PA, USA) and transfected with tagged connexins. The next day, coverglasses were mounted in a closed live-cell chamber (Focht Live-Cell Chamber System FCS2, Bioptechs Inc.) using Phenol Red-free DMEM (Gibco BRL, Grand Island, NY, USA). Temperature was kept constant at 37°C using the attached automatic FCS2 temperature controller. A Kodak Photometrics, CH350L liquid-cooled charged-coupled device (CCD) camera (500 KHz, 12 bit, 1317×1035 pixels) was used for imaging, attached to an inverted, wide-field fluorescence microscope (Olympus IX-70). The autofluorescent tracer proteins CFP, GFP and YFP were excited with a standard mercury arc lamp attached to a fiber optic illumination scrambler providing even illumination. GFP fluorescence was detected using a standard FITC filter set. CFP and YFP fluorescence was imaged alternating using an 86002 filter set (Chroma Technology Corp., Brattleboro, VT, USA). Optical sections of gap junction plaques were acquired at maximum magnification (1500×, PlanApo 100×, NA 1.35 oil immersion lens, plus 1.5× auxiliary magnification). Section planes were collected in 0.2 μm steps covering the entire gap junction in a z-axis field of 5-10 μm using the attached Applied Precision Inc. high-precision motorized stage. Images were acquired, setting binning and gain of the CCD camera to 1, resulting in a minimum pixel size corresponding to 44.8×44.8 nm (approx. 5×5 gap junction channels) of the specimen. In some experiments the relatively low signal and bleaching-tendency of CFP required setting the gain to 4 to obtain adequate exposure times. In general, 0.2-0.5 second exposure times were sufficient to obtain images suitable for deconvolution with GFP (≥1,100 gray levels; dynamic range of the cooled CCD camera was 0–4095). Exposure times were between 0.2-1 seconds for YFP and 2-6 seconds for CFP. Images were saved and processed on attached Silicon Graphics workstations (O2, Octane) using the DeltaVision software package softWoRx (Version 2.5), and stored on CD-ROM.

Five QuickTime movie sequences featuring animated three-dimensional volume renderings of entire gap junction plaques imaged in living cells are included with this paper.

To study the organization of gap junctions in live cells, the C-termini of three major connexin isotypes α₁(Cx43), β₁(Cx32) and β₂(Cx26) were tagged with the autofluorescent reporter GFP. GFP-tagged connexins assembled into typical gap junction plaques that were visible by fluorescence microscopy as green fluorescent puncta and lines in the adjoining plasma membranes of transfected HeLa cells (marked with arrows in Fig. 1A). GFP expressed alone was not targeted and filled the entire cell (Fig. 1A, bottom right panel). The size of the plaques correlated with the connexin isotype, and time of expression. Identical plaques were also assembled in other communication-competent and -deficient cell lines, such as CHO, COS-7, BHK, NIH 3T3, 293, T51B and MDCK cells, which were either transiently or stably transfected. HeLa cells were chosen for all further studies presented here since they have been shown to efficiently assemble functional gap junctions when transfected with connexin-specific cDNAs (Elfgang et al., 1995), they do not express endogenous α₁(Cx43), β₁(Cx32) or β₂(Cx26) (Eckert et al., 1993; Elfgang et al., 1995; Hülser et al., 1997), and their growth is not contact-inhibited. The latter results in the formation of gap junctions viewable onto their surface, besides the more common view onto the edge of the plaques.

Expression of GFP-tagged α₁(Cx43) was observed first 4-5 hours post-transfection as diffuse fluorescence throughout the cell body (cells marked with asterisks in Fig. 1B, 4h) that then accumulated in an area next to the nucleus (marked with arrows in the images 4h and 5h in Fig. 1B). Protein localization studies indicated that α₁(Cx43) localized to the ER and Golgi (see below). The first gap junctions were formed 4-6 hours post-transfection as a single (arrowhead in Fig. 1B, 5h) or a few small fluorescent spots (Fig. 1B, 5h and 6h). Time-lapse recordings (M. Falk, unpublished data) indicated that, over time, smaller plaques increased in size or fused by moving laterally within the plane of the membranes. After longer expression times plaques could grow very large, filling large areas of the adjoining membranes (Fig. 1B, 40h). 50-60 hours post-transfection, when transcription had diminished from transiently transfected plasmids, only small gap junction plaques were left (Fig. 1B, 60h), which were degraded after 80 hours post-transfection under these experimental conditions (Fig. 1B, 80h). Similar plaque assembly and degradation kinetics were obtained with GFP-tagged β₁(Cx32) and β₂(Cx26).

Gap junctions assembled from GFP-tagged connexins also appeared normal by high-resolution structural criteria. Typical gap junction plaques with continuously joined plasma membranes throughout the area of the junctional membrane, the central gap, the 7-striped staining pattern, and no obvious connections to cytoskeletal elements were visible in thin section electron micrographs of HeLa cells transiently transfected with α₁(Cx43) (Fig. 1C). Gap junctions assembled from GFP-tagged α₁(Cx43) appeared similar in width to gap junctions assembled from wild-type connexin (17-20 nm). This is a result of the staining technique that only insufficiently contrasts the GFP polypeptides. No gap junctions were detected in untransfected HeLa cells.

Tagged connexins were further examined by immunoblot analysis using connexin, as well as GFP-specific antibodies. Protein bands with an electrophoretic mobility of approximately 70 kDa, 59 kDa, 53 kDa, and 27 kDa, corresponding to α₁(Cx43)-GFP, β₁(Cx32)-GFP, β₂(Cx26)-GFP and GFP alone, respectively, were detected in transfected HeLa cells (Fig. 2A, lanes 1, 2, 4, 6, 8-11). No such protein bands were detected in mock transfected cells (Fig. 2A, lanes 3, 5, 7, 12). Also, no protein bands corresponding to endogenous connexins or GFP alone were detected (Fig. 2A, lanes 1-10). Slower migrating protein bands were detected in longer exposures of α₁(Cx43)-GFP transfected cells (marked with an arrow in Fig. 2A, lane 2). They correlate with the phosphorylated forms of wild-type α₁(Cx43) (Musil and Goodenough, 1991; Laird et al., 1991; Laird et al., 1995). Some faster migrating protein bands probably representing degradation products were detected with all three connexin isotypes in longer exposures as well (marked with asterisks in Fig. 2A, lane 2; shown for α₁[Cx43]-GFP only).

In immunostained HeLa cells transfected with GFP-tagged α₁(Cx43), β₁(Cx32), and β₂(Cx26), all detectable GFP fluorescence (Fig. 2B, column 1, green) corresponded to connexins fluorescently labeled with connexin-specific antibodies (Fig. 2B, column 2, red), resulting in a complete overlap of the two chromophores (Fig. 2B, column 3, yellow). Representative cells typically seen on the slides in repetitive experiments are shown.

Some intracellular localized connexin-GFP fluorescence was observed in the transfected cells as well (e.g. Figs 1A,B, 2B). To examine whether these connexins may represent transport intermediates in transit to the plasma membrane, double-labeling experiments of connexin-GFP and ER or Golgi membranes were performed. Antibodies directed against calnexin, an ER resident, and p58, a Golgi resident protein were used with secondary antibodies coupled to Cy3 (red) (Fig. 2C). A typical staining of ER and Golgi membranes in fixed HeLa cells is shown in the left panels. On the right, HeLa cells transiently transfected with α₁(Cx43)-GFP are shown. In transfected cells with green fluorescent gap junction plaques in their adjoining plasma membranes, intracellular located connexin-GFP (green) colocalized with ER and Golgi compartments (red), resulting in yellow, connexin-GFP-containing ER and Golgi membranes, respectively. Similar images were obtained for GFP-tagged β₁(Cx32) and β₂(Cx26), and with protein disulfide isomerase (ER-specific) or Mannosidase II (Golgi-specific) antisera.

To investigate whether connexins tagged with GFP form functional gap junction channels, classical dye transfer experiments using gap junction permeable dyes were performed. Confluent monolayers of HeLa cells were transiently transfected with GFP-tagged α₁(Cx43), β₁(Cx32) or β₂(Cx26) (shown in Fig. 3A,D,G,K). Individual cells in a group, or a pair of transfected cells were microinjected. Cells were imaged before (Fig. 3B,E,H), and shortly after (1-3 minutes) dye injection (Fig. 3A,C,F,I,K). The dye spread within seconds to the neighboring cells only when gap junctions assembled from GFP-tagged connexins were present. Dye did not spread to adjacent untransfected cells of the monolayer even after prolonged incubation times (16 hours at 37°C). Untransfected HeLa cells not expressing connexins (marked with an asterisk in Fig. 3A) did not transfer dye either. Dye transfer was immediate (1-2 seconds) in α₁(Cx43)-GFP and β₁(Cx32)-GFP expressing cells, and somewhat delayed (2-5 seconds) in β₂(Cx26)-GFP expressing cells. Similar results were also obtained with other gap junction permeable dyes, such as Cascade Blue or Sulforhodamine.

Taken together, these results demonstrate that connexins, tagged with GFP on their C-terminus, trafficked, assembled and clustered normally into functional gap junction channels and typical gap junction plaques.

To investigate the organization of gap junctions and to reconstruct three-dimensional views of entire gap junctions as they appear between living cells, GFP-tagged connexins were expressed in HeLa cells, and assembled gap junctions were imaged in living cells by high-resolution fluorescence deconvolution microscopy. Stacks of consecutive microscopic z-sections were acquired from gap junction plaques either visible onto their edge, or onto their surface, respectively, covering the entire depth of the plaques. Images were deconvolved, and three-dimensional volume reconstructions were rendered from the z-sections using softWoRx computer software that then were rotated around their longitudinal axis. A selection of such volume views of smaller (≤2 μm diameter) and larger (≥5 μm diameter) gap junction plaques assembled from GFP-tagged α₁(Cx43) is shown in Fig. 4, and animated in the QuickTime movie sequences 1, 2, 3, 4 and 5. Gap junction plaques were found to be quite diverse in appearance. In general, plaques were irregular in shape rather than circular. The edges of the plaques were usually wavy or jagged, and sometimes long protrusions (asterisk in Fig. 4D) or deep invaginations deprived of gap junction channels extended from their edges. Circular, as well as irregular-shaped areas (approx. 0.1-1 μm in diameter) lacking gap junction channels were detected within the gap junction channel plaques (arrows in Fig. 4D,E). Furthermore, fluorescent vesicles were detected in intracellular compartments such as rER and Golgi membranes (Fig. 4A,C,E), on the edges (arrowhead in Fig. 4E), and in close vicinity of the plaques (arrowheads in Fig. 4F,G). Note that some of the more complex gap junctions shown in Fig. 4D,F,G, and in the QuickTime movie sequences 3-5 appear more fluorescent in their center where the plaques are imaged onto their edges. This is due to the imaging technique and does not reflect a true increase in fluorescence. In the plane view areas of the plaques only two layers of GFP, located on either side of the channels, are present per section (see Discussion and Fig. 7). Approximately 20 channels including their GFP tags are present per section in the areas where the plaques were sectioned through their edges.

To investigate whether different coexpressed connexin isotypes would gather within the same junctional plaque and how they would distribute, the connexin isotypes α₁(Cx43), β₁(Cx32) and β₂(Cx26) were tagged with the GFP color variants CFP and YFP, respectively. Tagged connexins were coexpressed in all possible combinations in transfected HeLa cells. Both connexin isotypes gathered within the same junctional membrane areas and assembled into gap junction plaques that contained both connexin isotypes. When tagged β₁(Cx32) was cotransfected with tagged β₂(Cx26), the two proteins were distributed homogenously throughout the entire plaque, resulting in homogenously yellow plaques (Fig. 5, row 2). When tagged α₁(Cx43) and β₁(Cx32), or α₁(Cx43) and β₂(Cx26) were coexpressed, still both proteins gathered within the same gap junction plaques, however, completely segregated into well-separated domains (green, or red) consisting of either connexin isotype (Fig. 5, rows 3,4). Identical results were obtained with exchanged tags. To further eliminate the possibility that the protein tags were responsible for this specific channel distribution, cells were cotransfected with two cDNAs encoding the same connexin isotype, but tagged either with CFP or YFP, respectively. In all these instances, the tagged connexin isotypes were distributed evenly, resulting in homogenously yellow plaques (shown for α₁[Cx43]-CFP/α₁[Cx43]-YFP in Fig. 5, row 1). Segregation was independent of plaque size – small (Fig. 6A) and larger plaques (Figs 5, 6B) showed identical segregation and ratio of the two expressed connexins, as shown for coexpressed α₁(Cx43)-YFP/β₂(Cx26)-CFP in Fig. 6B.

Three different connexin isotypes, α₁(Cx43), β₁(Cx32) and β₂(Cx26), were tagged with the autofluorescent proteins CFP, GFP and YFP and this reporter technique was combined with single and dual-color, high-resolution fluorescence deconvolution (DV) live-cell microscopy and volume rendering. This approach permitted a study of the structural organization and the distribution of channels within gap junction plaques assembled from one or two different connexin isotypes in living cells.

When the tagged connexins were expressed in transfected HeLa or other communication-competent or -deficient cell lines, they assembled into functional gap junctions that appeared indistinguishable by structural and ultrastructural analysis from wild-type gap junctions. Dye microinjection analysis indicated that the tagged channels allowed the passage of gap junction permeable dyes, as described for wild-type gap junction channels. Further, Bukauskas et al. (2000) found that the electrical properties of channels assembled from GFP-tagged α₁(Cx43) were identical to those of wild type α₁(Cx43), except for a reduced sensitivity to transjunctional voltage.

Further, it was found that the size of the gap junction plaques, in general, correlated with the connexin isotype. Gap junctions assembled from α₁(Cx43)-GFP, in general, formed more solid, brightly fluorescent plaques, which over time could grow very large (≥10 μm diameter). Gap junctions assembled from β₂(Cx26)-GFP formed less fluorescent plaques, which grew to intermediate sizes (≤10 μm diameter). Gap junctions assembled from β₁(Cx32)-GFP generally remained small (≤2 μm diameter), often only visible as a line of relatively faint puncta. Whether the GFP tag may reduce efficiency of β₁(Cx32) transport and/or assembly, and by this influences β₁(Cx32) gap junction plaque size, was not further investigated. Interestingly, however, β₁(Cx32) assembled into larger plaques (5-10 μm in diameter) when it was coexpressed with another connexin isotype, such as α₁(Cx43) or β₂(Cx26) (see Fig. 5). Large gap junction plaques (≥10 μm diameter) were found in different tissues, such as the otocyst sensory epithelium (Ginzberg and Gilula, 1979), embryonic epidermis (Risek et al., 1994), and between pillar cells in the organ of Corti in the inner ear (Andrew Forge, University College London, personal communication).

Expression of GFP-tagged α₁(Cx43) monitored over time indicated that expression is initiated in the ER and then accumulated in the Golgi region before the first gap junctions were formed at specific sites of the adjacent membranes, similar to results obtained for wild-type connexins (Musil and Goodenough, 1991; Laird et al., 1991; Laird et al., 1995; Falk et al., 1994). Some intracellular fluorescence detected in α₁(Cx43)-GFP, β₁(Cx32)-GFP and β₂(Cx26)-GFP transfected HeLa cells, however, did not colocalize with ER and Golgi markers and, therefore, does not represent connexin cargo vesicles in transit to the plasma membrane (visible as green vesicular structures in Fig. 2C, panels 2 and 4). These vesicular structures may include internalized fragments of gap junction plaques destined for degradation (Larsen et al., 1979; Mazet et al., 1985; Severs et al., 1989); alternatively, trafficked connexins (George et al., 1999), mis-targeted fusion proteins or connexin-GFP degradation products. Time-lapse recordings (M. Falk, unpublished data) indicated that the plaques could grow by the addition of channels and by plaque fusion due to lateral movement in the plane of the membrane. Comparable results were also obtained by Jordan et al. (1999) and Holm et al. (1999).

GFP was fused to the C-terminus of connexins since earlier studies showed that deletions (Unger et al., 1999), as well as additions (Sullivan and Lo, 1995) to the C-terminus of connexin subunits were possible without interfering with the intracellular transport, or disrupting assembly of the mutated subunits into gap junction channels and gap junction plaques. The crystal structure of GFP (Ormö et al., 1996) revealed that its 238 amino acid residues (27 kDa) fold into a tightly packed can 24×42 Å in size. This unique tight structure is believed to be responsible for its inert behavior (Misteli and Spector, 1997). Fig. 7 shows a schematic representation of a gap junction channel drawn to scale that is assembled solely from GFP-tagged connexins. Dimensions were adapted from Makowski et al. (1977) and Unger et al. (1999). The scheme indicates that sufficient space for the six GFP tags is available within the connexon. The electron cryocrystallographic analysis of gap junction channels suggests that the C-terminal domains of the connexin polypeptides form the outer portion of the connexon that faces the cytoplasm (Unger et al., 1999). The N- and C-termini of the GFP polypeptide are located close to each other on one end of the barrel (Ormö et al., 1996). Therefore, it appears likely that the GFP tag simply extends the connexon into the cytoplasm without altering (1) the overall connexon structure, (2) its docking into complete gap junction channels and (3) packing of the channels with their natural spacing of 90Å center to center (Revel and Karnovsky, 1967). This results in two layers of GFP, one on each cytoplasmic side of the adjoining membranes, and in approximately 100,000 tags/μm². This explains the strong fluorescence of the gap junction plaques, which even allows imaging in the presence of phase-contrast illumination (see Figs 1A, 3). Taken together, our analysis demonstrates that GFP-tagged connexins assemble into functional gap junction channels, and cluster into bona fide gap junctions.

The detailed structural organization of gap junctions in living cells was investigated by employing high-resolution fluorescence deconvolution microscopy. Predominantly, larger gap junction plaques (2-15 μm in diameter) were investigated since they cannot be confused with vesicular structures and their structural characteristics can better be resolved. However, the same characteristics were also seen in smaller (≤2 μm diameter) gap junction plaques (compare Figs 4A,B, 5, rows 2 and 3 left panels and 6A), and in gap junction plaques assembled from stable transfected and endogenous expressed connexins (M. Falk, unpublished results). Gap junction plaques imaged with this technology were found to be diverse in shape (Fig. 4), typically with wavy or jagged edges and long extending protrusions and deep invaginations. Fluorescent vesicles detected at the edge and in the close vicinity of the plaques, as well as further away in the cytoplasm, probably represent transport vesicles and/or partial plaque degradation products. Striking circular and sometimes irregularly shaped areas (approx. 0.1-1 μm in diameter) deprived of gap junction channels that could move quite rapidly were found within the plaques. Such channel-free areas were first described by freeze-fracture electron microscopy and were termed nonjunctional membranes (Friend and Gilula, 1972). Currently, we are investigating the nature, significance and dynamics of these nonjunctional membrane areas.

Volume rendering of entire z-section stacks enabled us to transform gap junction plaques, so far only imaged as one-dimensional puncta, lines or fragmental sheets with conventional imaging techniques, into complete three-dimensional structures, thereby demonstrating for the first time how gap junctions appear in the adjoining membranes between living cells. The volume reconstructions were further animated to better demonstrate the gap junction plaque structure (see attached QuickTime movie sequences 1-5).

The distribution of connexins within gap junction plaques was investigated by tagging different connexin isotypes with the GFP color variants CFP and YFP, allowing simultaneous observation of both proteins in living cells. Coexpressed β₁(Cx32) and β₂(Cx26) distributed homogeneously throughout the plaques, as indicated by the yellow plaque color, while α₁(Cx43) and β₁(Cx32), or α₁(Cx43) and β₂(Cx26) segregated into well-separated domains. Segregation or mixing of channels was independent of plaque size and ratio of the two expressed connexin isotypes (Figs 5, 6). This demonstrates that the mode of channel distribution is strictly dependent on the connexin isotype.

Segregated areas consist of homomeric, homotypic channels that will either result in a red or green color. Yellow areas present in unprocessed images of some of these plaques could be interpreted as a mixing of the two connexin or channel types in these contact areas (see unprocessed images in rows 3 and 4 of Fig. 5). However, these yellow areas disappeared completely, or were reduced to narrow borders located only between green and red domains, after the images were deconvolved (Fig. 5, right panels). This indicates that domains consisting of mixed channel populations do not exist in these α/β plaques and the appearance of yellow is an imaging artifact, generated by the limited spatial resolution of light microscopic techniques. Ideally, the yellow borderline should disappear completely in the deconvolved images if homogenous domains solely adjoin each other. However, a thin yellow line will remain if pixels on the photochip of the CCD camera used for imaging overlap the border region and will sense both emission signals as is visible in the plaque images shown in Figs 5, row 3, 6A, bottom panel, and 6B, lower panels.

The mixed plaques can consist of heterotypic, mixed homotypic, or heteromeric channels. The pixel size corresponds to 45×45 nm, corresponding to approx. 5×5 gap junction channels when images were acquired at highest magnification (as in the plane-view images of Fig. 5). This requires that the two connexin isotypes must mix thoroughly throughout the entire plaques since the plaques appeared homogenously yellow. Although heterotypic channels can be formed between cells expressing β₁(Cx32) and β₂(Cx26) (Elfgang et al., 1995), the formation of heterotypic channels appears not to be very likely under the conditions of these experiments, since both cells forming the junction were transfected with a mixture of the two connexin cDNAs and, therefore, both cells should express both connexins. Mixed homomeric connexons were found in liver gap junction plaques assembled from two different connexin isotypes by scanning transmission electron microscopy (STEM) analysis (Sosinsky, 1995; reviewed in Sosinsky, 1996). Alternatively, the yellow plaques can be formed by heteromeric channels that contain both connexin isotypes (β₁[Cx32] and β₂[Cx26]). This would be consistent with previously published results on gap junction channel composition obtained by biochemical, immunological and electrophysiological approaches (Stauffer, 1995; Jiang and Goodenough, 1996; Falk et al., 1997; Brink et al., 1997; Bevans et al., 1998; He et al., 1999). Other methods capable of resolving connexon composition are, however, required to determine the channel composition in the β₁(Cx32)/β₂(Cx26) plaques.

How can connexin isotypes mix homogenously, or segregate? To my knowledge, this is the first time that such a clear separation has been described for a clustered membrane protein. The complete separation of the α and β channels suggests that the affinity between alike channels is greater than the affinity of channels that are not alike, and such channels composed of noncompatible connexin isotypes segregate into separated domains. Yet unknown characteristics shared by β₁(Cx32) and β₂(Cx26) could allow a homogenous codistribution, while connexins with different characteristics (α₁[Cx43] plus β₁[Cx32] and α₁[Cx43] plus β₂[Cx26]) would segregate from each other. Such characteristics could be based on the composition of the lipids that surround the channels (reviewed in Gil et al., 1998), or on structural motifs encoded by the connexin amino acid sequence. Further experiments should uncover such signals and elucidate their intrinsic characteristics. Furthermore, it remains to be established whether the observed channel segregation within the plaque is temporary, or permanent, and if the composition of the plaques correlates with their function in direct cell-to-cell communication.

I dedicate this work to the memory of Norton Bernie Gilula, my teacher, colleague and friend, who died on September 27, 2000 after his lengthy battle with cancer. I thank Kelly Moremen for providing Mannosidase II-specific antibodies, Norton B. Gilula for providing connexin-specific antibodies, equipment and encouragement, Malcolm Wood and Theresa Fassel for help with electron microscopy and initial training in deconvolution microscopy, Peijuan Shen for technical assistance, and Lucio Benedetti, Lukas Buehler, Jutta Falk-Marzillier, Danuta Balicki and Ben Giepmans for their scientific discussion and critical suggestions. This work was supported by grant RO1 GM55725 from the National Institute of Health to M.M.F. M.M.F is a former recipient of a Deutsche Forschungsgemeinschaft fellowship Fa 261/1-1.