W. Howard Evans, Department of Medical Biochemistry, University of Wales College of Medicine, Heath Park, Cardiff CF4 4XN, UK. Fax: + 44 122 274 3133; Tel: + 44 122 274 3133; E-mail: firstname.lastname@example.org
Guinea-pig liver gap junctions are constructed from approximately equal amounts of connexins 26 and 32. The assembly of these connexins into connexon hemichannels and gap junctions was studied using antibodies specific to each connexin. Intracellular membranes were shown to contain low amounts of connexin 26 relative to connexin 32 in contrast to the equal connexin ratios detected in lateral plasma membranes and gap junctions. Assembly of gap junctions requires oligomerization of connexins into connexons that may be homomeric or heteromeric. Immunoprecipitation using antibodies to connexins 26 and 32 showed that liver gap junctions were heteromeric. A chemical cross-linking procedure showed that connexons solubilized from guinea-pig liver gap junctions were constructed of hexameric assemblies of connexin subunits. The intracellular site of oligomerization of connexins was investigated by velocity sedimentation in sucrose–detergent gradients. Oligomers of connexins 26 and 32 were extensively present in Golgi membranes and oligomeric intermediates, especially of connexin 26, were detected in the endoplasmic reticulum–Golgi intermediate subcellular fraction. Two intracellular trafficking pathways that may account for the delivery of connexin 26 to the plasma membrane and explain the heteromeric nature of liver gap junctions are discussed.
Key issues in studying the arrangement of protein subunits in multimeric channels are the cellular location where subunit oligomerization occurs and the underlying mechanisms ensuring subunit correct stoichiometry during channel assembly. Gap junctions are pervasive intercellular channels providing direct signalling pathways integrating and co-ordinating cellular activities in tissues and organs [1,2]. They are constructed of apposed connexon hemichannels that have been shown, primarily on the basis of the structural features revealed by X-ray scattering and by electron microscopy of paracrystalline arrays , to consist of a hexameric assembly of connexin protein subunits arranged around a central aqueous channel. Connexon hemichannels, located in the contacting plasma membranes dock, provide a direct intercellular communication pathway across which electrolytes, metabolites and signalling molecules, with a size of ≈1 kDa, are able to pass . Fourteen connexin isoforms have been identified in rodents [5,6] and nine in humans [5,7,8] with many homologues from other species .
Previous work has shown that connexin (Cx) 32, the major constituent of rat liver gap junctions, was present at low levels in subcellular fractions deriving from the constituent compartments of the secretory pathway and it was concluded that these results reflected the synthesis, trafficking and rapid turnover of gap junctions . In the present work guinea-pig liver was selected as a model system to study, using subcellular fractionation techniques, aspects of the intracellular trafficking and assembly of gap junctions because the tissue expresses approximately equal amounts of Cx32 and Cx26. The distribution of the two connexins in intracellular membranes, plasma membranes and gap junctions was studied to shed light on where oligomerization of the two connexins occurs and to determine whether the gap junctions were homomeric or heteromeric in composition. The results suggest that the formation of connexons is a gradual process that is completed in the Golgi apparatus. However, the low Cx26-to-Cx32 ratio observed in microsomal and Golgi fractions suggested that the bulk of Cx26 may be delivered to gap junctions via a different route to that followed by Cx32, which bypassed the Golgi apparatus.
Preparation of subcellular membrane fractions
Plasma membranes and gap junctions. Lateral plasma membrane fractions were prepared as described previously [9,10] by resuspending into 87% (w/v) sucrose a low-speed ‘nuclear’ pellet of a liver homogenate prepared in 1 mm NaHCO3. After flotation in a step-wise sucrose gradient consisting of 54, 48, 43 and 8% sucrose (w/v), plasma membranes were collected from the 43–48% (w/v) sucrose interface. To purify the gap junctions, plasma membrane preparations were resuspended in an equal volume of 40 mm NaOH, sonicated three times for 15 s and centrifuged to obtain a pellet. Gap junctions were then purified by sucrose density gradient centrifugation .
Sinusoidal plasma membranes and endoplasmic reticulum membranes. As described previously , the pellets from postmitochondrial supernatants were resuspended in 2 vol. of 72% (w/v) sucrose and layered stepwise with 39% (w/v) and 8% (w/v) sucrose. After rate zonal centrifugation, endoplasmic reticulum (ER) membranes were collected from the sample–39% (w/v) sucrose interface and sinusoidal plasma membranes were collected from the 39–8% (w/v) sucrose interface.
Endoplasmic reticulum–Golgi intermediate compartment and Golgi membrane fractions. An endoplasmic reticulum–Golgi intermediate compartment (ERGIC) fraction was prepared as described by Dunkle et al.  in which postnuclear supernatants were loaded on top of step sucrose gradients consisting of 2.0, 1.5 and 1.3 m sucrose layers. After centrifugation at 100 000 g for 16 h, the 1.3 m sucrose–supernatant interface was collected. Golgi stacks consisting of Golgi light, Golgi intermediate and Golgi heavy were obtained by a modification  of the classical procedure of Ehrenreich et al. .
Enzymatic analysis revealed that plasma membrane fractions were enriched more than 10-fold in 5′ nucleotidase activity, whereas Golgi and microsomal fractions were enriched 1.5–2-fold and 1.2-fold in 5′ nucleotidase activity compared with the tissue homogenate, respectively. Golgi fractions were enriched more than 60-fold in galactosyl transferase activity relative to the tissue homogenate. Similarly, ERGIC fractions were enriched, relative to the tissue homogenate 20-fold in galactosyl transferase, 12-fold in mannosyl transferase and threefold in 5′ nucleotidase activities.
Western blots were carried out after transferring electrophoretically proteins to Hybond ECL nitrocellulose membrane (Amersham, UK) at 100 V in 48 mm Tris, 39 mm glycine (pH 9.2) and 20% (v/v) methanol for 90 min. The membranes were blocked at room temperature for 1 h with 5% (w/v) fat-free milk in 25 mm Tris/HCl pH 7.5, 140 mm NaCl, 2.7 mm KCl (TBS) and 1% (v/v) Tween-20 (TBS-T). Membranes were incubated with connexin-specific antibodies diluted in TBS-T containing 5% milk for 2 h at room temperature or alternatively overnight at 4 °C. Antibodies were generated in rabbits (Table 1) and were used either as immunoglobulin fractions or affinity purified; these have been characterized extensively by Western blotting and immunocytochemistry [9,14–17]. In addition, a high-titre antibody  was generated to a sequence in human Cx26, Gap 28, using the multiple antigen-presentation procedure . For immunoprecipitation experiments, a mAb generated against Cx32 (Zymed) was used. After incubation with connexin-specific antibodies, membranes were washed three times with TBS-T, blocked for 30 min with TBS-T containing 5% milk and incubated for 1 h at room temperature, with the anti-(IgG) conjugated to horseradish peroxidase. After washing three times with TBS-T and once with TBS or NaCl/Pi pH 7.4, bands were visualized by chemiluminescence and detected by autoradiography.
Table 1. Properties of connexin anti-(peptide) antibodies used.
Subcellular fractions were solubilized by agitation for 2 h in 20 mm triethanolamine pH 9.2, 20 mm EGTA, 10 mm dithiothreitol and 2% (w/v) dodecyl maltoside. After solubilization, samples were centrifuged at 12 000 g for 15 min and supernatants loaded onto linear sucrose gradients, 10–40% (w/v), containing 0.1% dodecyl maltoside. After centrifugation at 150 000 g for 20 h at 4 °C, up to 12 fractions were collected, dialysed against distilled water overnight, concentrated in vacuo and connexin distribution analysed by SDS/PAGE and immunoblotting. The protein markers carbonic anhydrase (29 kDa) and myosin (200 kDa) were fractionated in sucrose gradients under identical conditions and corresponded to proteins sedimenting at 5S and 9S, respectively [19,20].
Connexons were dissociated from gap junctions by extraction with 20 mm triethanolamine pH 9.2, 20 mm EGTA, 10 mm dithiothreitol and 2% (w/v) dodecyl maltoside for 2 h with orbital rotation. Solubilized samples were centrifuged at 12 000 g for 15 min and the pellet discarded . A small aliquot of the supernatant was saved as a control sample containing noncross-linked connexins. The remainder of the supernatant was incubated in 10 mm dimethyl suberimidate (Pierce) for 20 h with orbital rotation. The reaction was stopped by adding an excess of glycine (50 mm final concentration). Samples were dialysed against distilled water and concentrated (Centricon-10 tube, Amicon). The cross-linked products were electrophoresed in a linear 5–20% (w/v) polyacrylamide gradient SDS/PAGE. Connexins in gels were detected by Coomassie blue staining (noncross-linked samples) or by immunoblotting as described above.
Gap junction plaques were solubilized by agitation at 4 °C for 2 h in 20 mm triethanolamine pH 9.2, 20 mm EGTA, 10 mm dithiothreitol and 2% (w/v) dodecyl maltoside or SDS. Samples were centrifuged at 12 000 g for 15 min and the pellet discarded. Supernatants were diluted 10 times in RIPA buffer (1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 50 mm Tris/HCl pH 8.0 and 150 mm NaC1) and incubated overnight with Gap 28 or Gap 34 antibodies (Table 1) at 1/100 dilution at 4 °C. Antibodies bound to connexins were incubated with protein A–Sepharose beads (Pharmacia, Uppsala, Sweden) for 2 h under the same conditions. Beads were washed three times with RIPA buffer and connexins extracted with Laemmli buffer  prior to analysis by SDS/PAGE. Connexins were detected by immunoblotting using Gap 28 antibodies or mAb against Cx32.
Analysis of guinea-pig liver gap junctions by SDS/PAGE and Coomassie blue staining showed that they were constructed of approximately equal amounts of Cx26 and Cx32, with ≈ 55% of total protein consisting of Cx26 (Fig. 1). Guinea-pig liver thus provided an appropriate model system to analyse the subunit composition and mechanisms of assembly of connexons.
Cx26 to Cx32 ratios in intracellular membranes and gap junctions are different
Membranes originating from different compartments of the secretory pathway of guinea-pig liver were isolated by subcellular fractionation. Connexin distribution in microsomal, Golgi and plasma membrane fractions was compared by Western blotting using antibodies specific to Cx26 or Cx32 (Table 1). Connexin 32 was found to be present in all subcellular fractions, with the highest relative amount detected in lateral plasma membranes (Fig. 2A). However, using a Western blotting procedure adjusted to demonstrate the approximately equal amounts of Cx26 and Cx32 in gap junctions, and lateral plasma membranes, the Cx26 antibody failed to recognize this connexin in microsomes, ERGIC fractions, Golgi membranes and sinusoidal plasma membranes, although the Cx32 antibody recognized its corresponding antigen in these fractions, albeit at lower levels than in gap junctions and plasma membranes (Fig. 2A). To investigate why equivalent expression of both connexins was not observed in microsomal, Golgi and sinusoidal plasma membranes, another Cx26-specific antibody that was of higher titre (Gap 28) was used. This further antibody identified Cx26 in all the subcellular fractions (Fig. 2B), with lower amounts detected in microsomal and Golgi fractions than in lateral plasma membranes. Analysis of the distribution of Cx32 in the various fractions (Fig. 2A) showed the highest intracellular levels in Golgi with low amounts in ER and ERGIC; as expected, lateral plasma membranes contained the highest levels of both connexins. These results agree with those obtained by Rahman et al.  in rat liver with Cx32. In view of the low relative amounts of Cx26 found in intracellular membranes (Fig. 2A), a direct comparison with Cx32 was not possible, but independent staining with a stronger antibody again indicated that it had a similar distribution between the intracellular fractions; the Cx26 present in Golgi fractions may reflect a small amount of Cx26 incorporated into connexons that are trafficked along the secretory pathway. In summary, the immunoblotting approach showed that although similar levels of Cx32 and Cx26 immunoreactive protein were present in guinea-pig liver gap junctions and lateral plasma membranes, Cx26 was detected by antibodies in the intracellular membranes and sinusoidal plasma membranes.
Oligomeric status of connexins in guinea-pig liver subcellular fractions
Liver subcellular fractions were analysed using an oligomerization assay that separated oligomeric connexins (migrating at ≈ 200 kDa) from individual connexins (migrating at ≈ 29 kDa). The sedimentation positions in the sucrose gradients of connexins and connexons were standardized using either oligomeric connexins (gap junctions solubilized in n-dodecyl maltoside) or individual connexins (gap junctions solubilized in SDS). Figure 3 shows that both connexins, present in approximately equal ratios, were clearly separated at the extremities of the sucrose gradient used.
The relative positions of Cx26 and Cx32 in the sucrose gradient velocity sedimentation analysis of microsomes (ER), ERGIC, Golgi and gap junction (GJ) plaques were compared (Fig. 4). In these experiments, an antibody to Cx26 (Gap 28) was used to identify this connexin even when present in low amounts relative to Cx32 in the ER, ERGIC and Golgi fractions. In microsomes (ER panel), Cx32 was mainly in the monomeric form because it migrated into the linear sucrose gradients at a position corresponding to connexins dissociated from gap junctions using SDS (Fig. 3); Cx26 distribution in microsomes was different with oligomeric intermediates and fully oligomerized products were detected (ER panel). In the ERGIC fractions, Cx26 and Cx32 migrated further into the sucrose gradient in a position that may correspond to oligomeric intermediates because they did not sediment as far as the Cx32 or Cx26 connexons (IC panel). In Golgi fractions, connexons were detected with antibodies to Cx32 or Cx26 (Golgi panel), and these showed a similar sedimentation profile to connexons in plasma membranes (Fig. 3) and gap junctions (Fig. 4, GJ panel).
Guinea-pig liver connexons constructed of Cx26 released from gap junctions are hexameric
The oligomeric composition of guinea-pig liver connexons containing Cx26 and Cx32, was studied. For chemical cross-linking, purified gap junction plaques were solubilized at high pH in mild detergents, conditions demonstrated to generate intact hexameric connexons . Connexons were incubated with dimethyl suberimidate, a bifunctional cross-linker that covalently links adjacent amino groups. Cross-linked oligomeric products of Cx26 were observed, extending from dimeric to hexameric forms with the latter being the major oligomeric product (Fig. 5A). The oligomeric states showed a linear relationship with their relative mobility on a gradient SDS/PAGE confirming that bands correspond to oligomers of Cx26 (Fig. 5B).
Guinea-pig liver connexons released from gap junctions are heteromeric
The oligomeric composition of guinea-pig liver gap junctions was studied by an immunoprecipitation approach. Gap junctions were dissociated using dodecyl maltoside into connexons, or using SDS into connexins, and examined by immunoprecipitation with antibodies to Cx26 (Gap 28) or Cx32 (Gap 34; Fig. 6). With SDS-solubilized connexins (Fig. 6, lanes 1 and 2), each antibody immunoprecipitated exclusively its corresponding connexin. However, when connexons solubilized from gap junctions with dodecyl maltoside were examined using this approach, the two antibodies immunoprecipitated both connexins (lanes 3 and 4). These results lead to the conclusion that Cx26 and Cx32 were located in the same connexon.
The functional reasons why cells express two or more connexins that may account for selectivity of gap-junction-mediated intercellular communication are not well understood [2–4]. Studies in which the specific recombinant connexins were expressed in mammalian and amphibian model systems have shown that gap junctions with characteristic electrophysiological  and dye-transfer properties  are generated and it is possible that by varying the ratio and arrangement of connexins in heteromeric connexon channels, which subsequently dock to yield homotypic or heterotypic gap junctions, a range of intercellular channel permeability properties result [25,26]. Adult mammalian liver expresses Cx32 and Cx26, although in embryonic and malignant liver, Cx43 is also present . The ratio of Cx26 to Cx32 proteins in gap junctions varies according to the source of the liver; in rat and human liver, over 90% of connexin expression is contributed by Cx32, whereas in mouse liver, higher amounts of Cx26 are present [28,29]. In the present work guinea-pig liver was used where gap junctions are composed of a slight excess of Cx26 over Cx32.
The present work addresses the distribution of two connexins expressed by guinea-pig liver in the context of the intracellular trafficking properties and their contribution to connexon makeup. The distribution of Cx32 in gap junctions and intracellular membranes was similar to that observed with rat liver . However, by adjusting the concentration of antibodies used to identify the two connexins, a similar ratio of Cx32 to Cx26 present in gap junctions and lateral plasma membranes was not observed in Golgi, ERGIC and microsomal fractions; use of a higher titre antibody to Cx26 allowed detection of the low relative amounts of Cx26 present in these subcellular fractions relative to plasma membranes and gap junctions. The reasons for the lack of approximate equivalence in the intracellular membranes of the two connexins assembled into heteromeric gap junctions that have approximately equimolar amounts is puzzling. The failure to detect Cx26 in intracellular membranes by antibody Des 3 and the low levels detected by Gap 28, a stronger antibody, are unlikely to be due to these antibodies recognizing a connexin conformational state in gap junctions but not in connexins or connexons hemichannels, because denaturing SDS/PAGE gels were used. Another possible explanation for the results is that the low levels of Cx26 are a reflection of faster kinetics through the Golgi apparatus to the plasma membrane and gap junctions. However, because oligomerization of connexins to generate connexons was shown to be completed in the Golgi, this explanation seems unlikely. A further possibility is that most Cx26 synthesized in the ER takes an alternative route to the plasma membrane that bypasses the conventional secretory pathway followed by Cx32. Examples of proteins taking unorthodox trafficking pathways exist. Fibroblast growth factors 1 and 2, interleukins and thioredoxin, proteins that lack a classical secretory signal sequence, follow an alternative direct route to the plasma membrane [30,31]. Herpes virus VP22, MHC class I and II and HIV Tat proteins also follow a direct nonclassical pathway to the cell surface for export [30,32,33] and ER amyloid-binding protein was shown to translocate from the ER to the cell surface in the presence of β amyloid . In yeast, up to four secretory routes involved in surface membrane growth and biogenesis have been described .
The absence of conventional signal sequences in connexins, combined with minimal ER/Golgi post-translational modifications, are properties that would be in keeping with the option of taking more than one trafficking route to gap junctions. In the context of post-translational modifications, connexins are not glycosylated, but phosphorylation of amino acids on the carboxyl tail of Cx43 [36,37] and Cx32 [38,39] occurs. In contrast, Cx26 has a short cytoplasmic carboxyl tail, and no evidence exists for its phosphorylation . Therefore, without any known post-translational modification requirements, the data can be reconciled with the majority of the Cx26 expressed following a more direct route to the plasma membrane and/or gap junction. Such nonclassical direct routing presupposes that substantial oligomerization of Cx26 occurs prior to transfer to the plasma membrane, possibly in the ER as demonstrated in the present work.
Independent support for the proposition that Cx26 and Cx32 follow largely independent trafficking routes to the plasma membrane emerges from results obtained in live COS-7 cells expressing a chimeric Cx26 in which the carboxyl tail was replaced by that of Cx43. This chimeric connexon was targeted to the plasma membrane via a trafficking route that was independent of Golgi involvement as ascertained by its insensitivity to brefeldin A . In contrast, the trafficking of Cx32 to the plasma membrane was dependent upon the cells maintaining an intact Golgi apparatus . Evidence supporting the possibility that different mechanisms account for the movement to the cell surface of Cx26 in female rat liver also comes from studies of Kojima et al.  showing a rapid and specific appearance of Cx26, but not of Cx32, in gap junctions after brief perfusion of liver, an event occurring without changes in connexin mRNA or protein expression levels. Also, hormones, e.g. glucagon and glucocorticoids, influence production of Cx26 gap junctions in rat hepatocytes independent of Cx32 gap junctions . These examples of the modulation of Cx26 expression independent of Cx32 expression at gap junctions of hepatocytes would be explained by the existence of connexin intracellular trafficking pathways that are independent. Differences in the in vitro synthesis of Cx26 and Cx32 have also been reported with both connexins inserted cotranslationally into membranes [20,28], but Cx26 also inserted post-translationally into canine pancreatic microsomal membranes, an unconventional observation for membrane proteins in eukaryotic cells [28,43].
The present results also show that Cx32 and the Cx26 minor component detected in Golgi fractions were fully oligomerized in the Golgi apparatus. Partially oligomerized products were detected in ERGIC fractions suggesting that this segment of the secretory pathway functions as a site where Cx32 oligomerization commences. The oligomerization of Cx26 seemed to occur even earlier in ER and ERGIC. In contrast, oligomerization of Cx43 in normal rat kidney (NRK) cells was reported to occur in the trans-Golgi . The present results concur with the general view that oligomerization is a post-translational serial process that is initiated early in the secretory pathway [44,45]. It is worth noting that the current in vivo approach is less likely to lead to artefacts, for cultured cells overexpressing connexins generate gap junction-like structures in the ER .
The oligomeric nature of Cx26 of connexons in guinea-pig gap junctions was also examined. The cross-linking experiments demonstrated that a ladder of oligomers leading to a hexamer was obtained with Cx26. This result extends the data of Cascio et al. , which showed hexamers of Cx32 in gap junctions. An immunoprecipitation approach showed a close association of the two connexins within individual connexons. Similar results supporting the presence of heteromeric connexons were obtained by immunoprecipitation of mouse liver connexons dissociated from gap junctions (data not shown). Evidence for gap junctions constructed of heteromeric connexons have been reported in insect cells expressing recombinant Cx26 and Cx32 , in mouse neuroblastoma cells transfected with rat Cx43 and human Cx37 , in ovine lens expressing Cx46 and Cx50 , and in rat hepatocytes induced for Cx26 expression . Heteromeric connexons were synthesized using a cell-free system supplemented with Golgi membranes . Heteromeric gap junctions with unusual electrophysiological properties constructed of Cx32 and Cx32–aequorin were also assembled in paired Xenopus oocytes . The present results appear to be at odds with evidence for homomeric connexons in mouse liver , which distinguished between two populations of connexons according to their mass (6 × 26 kDa and 6 × 32 kDa). We propose that guinea-pig liver gap junctions contain two populations of connexons. The first is one in which Cx32 is the predominant connexin in connexons trafficked to the plasma membrane via the Golgi apparatus. Heteromeric connexons with a high Cx32-to-Cx26 ratio would not differ significantly in mass from homomeric Cx32 connexons, thereby explaining the data of Sosinsky . The second population of homomeric connexons constructed of Cx26 can be generated if a Golgi-bypassing route is followed and these would be inserted directly into the lateral plasma membrane and/or gap junctions in view of the low Cx26 levels detected in sinusoidal plasma membranes.
In summary, the heteromeric nature of connexons in guinea-pig liver and the evidence suggesting two trafficking routes for Cx26 and Cx32 can be reconciled by proposing that connexons constructed of Cx26 are mainly homomeric and are inserted into plasma membranes via a pathway that does not implicate the Golgi apparatus. The details of this pathway remain to be explored. Connexins are likely to have arisen early in metazoan evolution, and it is possible that the first connexins were initially trafficked to gap junctions via a more direct pathway. Studies in live cells showed that Cx26 was delivered to the plasma membrane in 5 min, whereas Cx32 took 10–15 min . The classical secretory pathway in which membrane proteins traffic via the Golgi would thus account for the biosynthetic throughput of Cx32 but for only a small proportion of Cx26 captured into heteromeric connexons. The two connexins would ultimately be located in the same large gap junction plaques which may be composed of homomeric or heteromeric connexons that were observed by double-labelling immunocytochemical approaches in liver tissue .
This work was supported by an Medical Research Council Programme Grant and a grant (to JDI) from the Welsh Office for Research and Development. S.A. is a Commonwealth Scholar.