INTRODUCTION

Freeze-fracture replica electron microscopy reveals views thatreflect the macromolecular architecture of the membranes, sothat the number and distribution of a wide variety of trans-membrane macromolecules, revealed as intramembraneparticles, can be determined. However, this technique offers nodirect information on the biochemical composition of cellmembrane components. The combination of freeze-fracturereplica with cytochemical labeling of individual cell membranemolecules, freeze-fracture replica cytochemistry, can providedirect evidence of the chemical nature and topology of the cellmembrane components. Previous attempts for freeze-fracturereplica cytochemistry have relied on either: (a) labeling cellsuspensions before replication (Pinto da Silva and Branton,1970; Pinto da Silva and Kan, 1984); or (b) labeling splitmembrane after platinum/carbon (Pt/C) shadowing(Andersson-Forsman and Pinto da Silva, 1988b; Dinchuk et al.,1987; Gruijters et al., 1987). However, further applications ofthese techniques to various cells and tissues has been lacking,mainly due to methodological limitations. The most decisivelimitation is that these techniques are restricted to only thelabeling of external leaflets of membranes (outer membrane

halves) of isolated (free) cells or monolayers, and do not permitthe cytochemical characterization of the internal leaflets (innerhalves) of membranes in both isolated cells and tissue samples.Here we propose the use of freeze-fracture replica cytochem-istry combined with sodium dodecylsulphate (SDS) digestionfor cytochemical labeling of freeze-fractured and replicatedcell membranes of isolated cells and tissue samples. Fig. 1shows schematically the rationale of our approach: after quick-freezing (Fig. 1A), freeze-fracture, and Pt/C shadowing (Fig.1B), unfixed tissues are treated with 2.5% SDS to dissolveunfractured membranes and cell components (Fig. 1C). Thecytoplasmic and exoplasmic halves of the cell membrane,remain attached to the Pt/C cast. Although SDS dissolvesunfractured portions of the membrane, it would not reach,micellize, and extract split membrane halves, as their apolardomains are positioned against, and stabilized by, their Pt/Ccasts. We reasoned that it should therefore be possible to labelthe inner (cytoplasmic) and outer (exoplasmic) leaflets ofmembranes in both isolated cells and tissue samples (Fig. 1D).In this report, we tested the feasibility of the SDS-digestedfreeze-fracture replica labeling (SDS-FRL) technique inimmunolabeling of various intercellular junctions with theantibodies against their major component proteins, because

3443Journal of Cell Science 108, 3443-3449 (1995)Printed in Great Britain © The Company of Biologists Limited 1995JCS9370

We propose a new electron microscopic method, thesodium dodecylsulphate (SDS)-digested freeze-fracturereplica labeling technique, to study the two-dimensionaldistribution of integral membrane proteins in cellularmembranes. Unfixed tissue slices were frozen with liquidhelium, freeze-fractured, and replicated in aplatinum/carbon evaporator. They were digested with2.5% SDS to solubilize unfractured membranes andcytoplasm. While the detergent dissolved unfracturedmembranes and cytoplasm, it did not extract fracturedmembrane halves. After SDS-digestion, theplatinum/carbon replicas, along with attached cytoplasmicand exoplasmic membrane halves, were processed for cyto-chemical labeling, followed by electron microscopic obser-vation. As an initial screening, we applied this technique tothe immunogold labeling of intercellular junction proteins:connexins (gap junction proteins), occludin (tight junction

protein), desmoglein (desmosome protein), and E-cadherin(adherens junction protein). The immunogold labeling wasseen superimposed on the image of a fracture face visual-ized by platinum/carbon shadowing. The immunoreactionwas specific, and only the structures where the proteinswere expected were labeled. For instance, anti-occludinimmunogold complexes were observed immediatelyadjacent to the tight junction strands on the protoplasmicand exoplasmic fracture faces. No significant levels of goldlabel were associated with non-tight-junctional regions ofplasma membranes. The procedures of the SDS-digestedfreeze-fracture replica labeling and its potential signifi-cance are discussed.

Key words: freeze-fracture replica, immunogold electronmicroscopy, gap junction, tight junction, adherens junction,desmosome

SUMMARY

Freeze-fracture replica electron microscopy combined with SDS digestion for

cytochemical labeling of integral membrane proteins

Application to the immunogold labeling of intercellular junctional complexes

Kazushi Fujimoto

Department of Anatomy, Faculty of Medicine, Kyoto University, Sakyo-ku, Kyoto 606-01, Japan

3444

they are easily recognized under an electron microscope, andcan be labeled specifically. Our findings demonstrate that theimmunogold particles bound exclusively to the intercellularjunctions identified in the same Pt/C replica according toelectron microscopic criteria. The initial screening thussuggests the great potential of SDS-FRL for the in situ identi-fication of integral membrane proteins.

MATERIALS AND METHODS

The tissues (livers, small intestines, and kidneys from rats, mouse, orchicks) were cut into sections less than 100 µm-thick by a Microslicer(Dosaka EM Co., Kyoto, Japan), and collected in phosphate bufferedsaline (PBS), pH 7.4. Tissue slices were quick-frozen by contact witha copper block cooled with liquid helium (Heuser et al., 1979). Thefrozen samples were fractured in a Balzers BAF 400T freeze-ech unit(Balzers Union, Liechtenstein) at −110°C, replicated by deposition ofPt/C from an electron beam gun positioned at a 45° angle followedby carbon applied from overhead. To release the replicas from thespecimen carrier, the carrier was immersed gently in PBS. Afterfloating off, the pieces of replica were transferred to 5 ml of 2.5%SDS (Sigma Chemical Co., St Louis, MO) containing 10 mM Tris-HCl and 30 mM sucrose, pH 8.3. SDS-digestion was carried out for1-12 hours at room temperature, with vigorous stirring (according toour experience, SDS-digestion for 2 hours at room temperaturesuffices for specimens obtained from the liver and intestine). Aftertreatment with SDS, replicas (actually carbon-stabilized membranehalves) were rinsed for at least 1 hour, with four or more changes ofPBS, and placed on drops of 1-10% bovine serum albumin (BSA) inPBS (BSA-PBS) for 30 minutes at room temperature. The replicaswere then labeled with the following antibodies for 1 hour at roomtemperature: 1/200 diluted anti-connexin (Cx)32 monoclonalantibody (provided by Drs A. Takeda and T. Shimazu, Department ofBiochemistry, School of Medicine, Ehime University, Ehime, Japan),1/500 diluted anti-Cx26 polyclonal antibody (provided by Drs A.Kuraoka and Y. Shibata, Department of Anatomy, Faculty ofMedicine, Kyushu University, Fukuoka, Japan), 1/200 diluted anti-E-cadherin monoclonal antibody (ECCD-1; provided by Dr M.Takeichi, Department of Biophysics, Faculty of Science, Kyoto Uni-versity, Kyoto, Japan), 1/50 diluted anti-occludin antibodies (Oc-1and Oc-2; provided by Drs M. Furuse and S. Tsukita, Department ofMedical Chemistry, Faculty of Medicine, Kyoto University, Kyoto,Japan), 1/200 diluted anti-desmoglein monoclonal antibody (DG 3.10;Boehringer Mannheim GmbH, Germany). After labeling, the replicaswere washed three times with PBS and incubated for 1 hour at roomtemperature with the second antibody-conjugated colloidal gold(Janssen Pharmaceuticals, Piscataway, NJ) diluted 1:50 in BSA-PBS.After immunogold labeling, the replicas were immediately rinsedseveral times in PBS, fixed with 0.5% glutaraldehyde in 0.1 Mphosphate buffer (pH 7.4) for 10 minutes at room temperature,washed twice with distilled water, and picked onto Formvar-coatedgrids. Electron microscopy was done using a JEOL 1200EX. Fordouble-labeling experiments of Cxs, we have used two primary anti-bodies (anti-Cx26 polyclonal antibody and anti-Cx32 monoclonalantibody) simultaneously. The optimum concentration of each of theprimary antibodies was determined from the dilution series when thesingle-immunolabeling was performed, as mentioned above. As cyto-chemical controls, specimens were incubated with non-immune rabbitIgG or had the primary antibody omitted from the labeling process.

The replicas were washed by placing the PBS, BSA-PBS ordistilled water in successive wells of a porcelain spot plate and usinga platinum loop to transfer each replica. Immunolabeling was con-veniently accomplished by putting the replicas into drops of theimmunocytochemical reagents on Parafilm inside a moisture chambercreated within a Petri dish. The replicas were immersed in mediumduring the entire process including SDS-digestion, washing andimmunolabeling to prevent contamination; the replicas were neverallowed to float on the medium and dry out during the entire process.

RESULTS

The fine structure of the fracture face observed in the Pt/Creplicas of the preparation subjected to SDS-FRL was the same

K. Fujimoto

Fig. 1. SDS-digested freeze-fracture replica labeling. (A) Cell isfrozen. (B) Freeze-fracturing and platinum/carbon shadowing (Pt/C)reveal the exoplasmic fracture face (EF) and the protoplasmicfracture face (PF). (C) SDS dissolves the unfractured membrane andcell component, but does not extract the split membrane halves.(D) The replica is processed for labeling, followed by electronmicroscopic observation.

Pt/CPF

EF

A

B

C

D

3445Freeze-fracture replica immunolabeling

as that of the conventional freeze-fractured faces. Therefore,although the immunogold particles were actually localized onthe surface of carbon-stabilized membrane halves attached tothe replicas, we used here the conventional nomenclature offreeze-fractured faces.

Tight junctionsFreeze-fracture replica electron microscopy revealed the char-acteristic anastomosing network of the tight junction strand onthe lateral plasma membranes. A high density labeling withanti-occludin antibody Oc-1 (Furuse et al., 1993), recognizingthe C-terminal cytoplasmic domain, was always observed onthe tight junctions of the chick intestinal epithelial cells,

urinary tubule epithelial cells, hepatocytes (Fig. 2A) and endo-thelial cells. Because of its relatively flat surface, the chickendothelium was frequently fractured to reveal large areaswithin the luminal surface (Fig. 2B). The protoplasmic fracturefaces (P-faces) of the luminal cell membranes of adjoiningcells were separated by a narrow meandering band whichdisplayed a remarkably constant width (6-7 nm; Fig. 2B,arrowheads). The immunogold particles were observed alongthe band.

Close examination of the immunogold labeling revealed thegold particles specifically associated with tight junction strands(Fig. 2C, solid arrow) on the P- and exoplasmic fracture faces(E-faces) of the cell membrane. Interestingly, the immunogold

Fig. 2. SDS-FRL of chick hepatocytes(A and C) and endothelial cells (B)with anti-occludin antibody, Oc-1.(A) The immunogold labeling isobserved on the extended meshwork oftight junction strands and/or grooves atthe point where three cells arejuxtaposed. (B) The protoplasmicfracture faces of luminal surfacemembranes of adjoining endothelialcells are separated by a narrowmeandering band (arrowheads). Theimmunogold particles are exclusivelyassociated with the band. (C) Highmagnification of Oc-1 immunogold-labeled tight junction. The immunogoldparticles are associated with the strands(filled arrow) on the P-face and grooves(open arrow) on the E-face, but notwith the liner grooves (arrowheads) onthe P-face. bc, bile canaliculus. Bars,200 nm.

3446

labeling was also found over the linear grooves on the E-faces(Fig. 2C, hollow arrow), but not over the grooves on the P-faces (Fig. 2C, arrowheads). No significant difference in thestaining pattern was observed between the labeling with Oc-1and Oc-2 (data not shown).

Adherens junctions Adherens junctions (zonula adherens, intermediate junctions)are located adjacent to tight junctions on the lateral membraneand completely circumscribe the cell apex of simple epithelialcells. Further description of the adherens junction morphologyin freeze-fracture replica is not possible because the freeze-fracture faces of the junctional membranes are indistinguish-able from those of the non-junctional membranes. Since themonoclonal antibody, ECCD-1 (Yoshida-Noro et al., 1984),recognizes the N-terminal extracellular domain of E-cadherin,we expected the exoplasmic side of the junctional membraneto be labeled. In fact, the immunogold labeling was found inareas of the E-face in association with the tight junction strandsof the mouse intestinal epithelial cells (Fig. 3A) and hepato-cytes (Fig. 3B), but the complementary P-face was not labeled(data not shown).

DesmosomesThe desmosomes (macula adherens) are spot-like areas ofintercellular contact and are morphologically distinct. In thefreeze-fracture replica, desmosomes appear as focal intramem-brane particle (IMP) aggregates. Since the monoclonalantibody, DG3.10 (Koch et al., 1990), recognizes the cyto-plasmic domain of desmoglein, we expected the cytoplasmicside of the junctional membrane to be labeled. The P-face ofthe desmosome was specifically labeled (Fig. 4A and B). Nolabeling was observed on the non-junctional membrane.

Gap junctionsIn freeze-fracture replicas, gap junctions were characterized bypatches of densely packed IMPs of uniform size (9-10 nm indiameter) and shape on the P-face. The E-face of gap junctionsappears as closely spaced pits, which are complementary to theIMPs on the P-face. Both the P- and E-faces were often presentover a single gap junction and in these instances the fractureplane jumped between the membranes. In mouse liver, Cx26-and Cx32-associated gold particles could be identified in thesame gap junction plaque in doubly immunostained replicas(Fig. 5A). Since the antibodies for the C-terminal cytoplasmic

K. Fujimoto

Fig. 3. SDS-FRL of mouse intestinalepithelial cell (A), hepatocyte (B) withanti-E-cadherin antibody, ECCD-1. Onthe E-face, the immunogold labeling isobserved adjacent to the tight junctions,but not within the tight-junctionalmeshworks (tj). Bars, 200 nm.

Fig. 4. SDS-FRL of rat intestinalepithelial cell with anti-desmogleinantibody. The spot-likeimmunoreaction positive areas are seenon the P-face (A). Close examinationreveals that the labeling is localized onthe desmosome (B). tj, tight junctions.Bars, 200 nm.

3447Freeze-fracture replica immunolabeling

domain of Cx26 (Kuraoka et al., 1993) and Cx32 (Takeda etal., 1988) were used, we expected the cytoplasmic side of thegap junctional membrane to be labeled. In fact, the exposedjunctional particles on the P-face were immunogold-labeled(Fig. 5A, PF). Interestingly, junctional pits on the E-face werealso labeled with immunogold particles (Fig. 5A, EF).

We next examined the distribution of connexins in the arti-ficially altered liver gap junctions using SDS-FRL. We accom-plished this by perfusion for 5 minutes in situ with PBS madehypertonic by the addition of 0.5 M sucrose (Goodenough andGilula, 1974). Although this treatment is known to result insplitting of the hepatocyte gap junctions and tight junctions,we failed to encounter a half-gap junction on each of the two

separated cells in the freeze-fracture replicas. However, we didfind other morphological alterations of the gap junctions asfollows: the large cluster of IMPs was observed on the P-face,which was easily identified as a gap junction, but in someregions of these gap junctions, IMPs appeared to be dispersedto various degrees (Fig. 5B). Occasionally, small gap junctionswere seen in association with the large gap junctions. Inaddition, we observed structure-formation in areas of themembrane removed from morphologically identical gapjunctions, appearing as isolated small aggregates of IMPs.Although these IMP aggregations were often indistinguishableon pure morphological basis, they were exclusively labeledwith Cx antibodies (Fig. 5C).

Fig. 5. SDS-FRL of intact (A) andhypertonic saline-treated (B and C)mouse hepatocytes with anti-Cxantibodies. (A) The double-immunolabeling with anti-Cx26 (15 nmgold particles) and anti-Cx32 (10 nmgold particles) reveals that Cx26 andCx32 are colocalized to the same gapjunction plaque. (B) The disocciated gapjunction plaques are labeled with anti-Cx32 (15 nm gold particles) and anti-Cx26 (10 nm gold particles) antibodies.(C) SDS-FRL with anti-Cx32 antibody(15 nm gold particles) reveals that someIMP aggregates are clearlyimmunolabeled. These findings implythat SDS-FRL makes it possible toreveal gap junctions consisting of just afew IMPs (connexons) to manythousand. PF, protoplasmic fractureface; EF, exoplasmic fracture face. Bars,200 nm.

3448

Using non-immune IgG instead of immuno IgG for SDS-FRL, gold particles were rarely found in the replicas (data notshown).

DISCUSSION

We have introduced a novel cytochemical labeling technique,‘sodium dodecylsulfate-digested freeze-fracture replicalabeling’ (SDS-FRL), which allows the two-dimensionalobservation of the distribution of a cytochemical label on themembrane surface. This new technique reveals the surface dis-tribution of the label coincident with the Pt/C replica imagesof the exoplasmic and protoplasmic fracture faces. Our initialapplications indicated specific labeling and exceedingly lowbackground.

SDS-FRL was inspired from the ‘fracture-flip’ (Andersson-Forsman and Pinto da Silva, 1988a; Fujimoto and Ogawa,1991; Fujimoto and Pinto da Silva, 1988) and ‘fracture-flip/Triton X-100’ (Fujimoto and Pinto da Silva, 1989, 1992)techniques, both methods that provide electron microscopicviews of both exoplasmic and cytoplasmic membrane surfaces,and is based on the carbon-stabilization of the hydrophobicface of split membrane halves by freeze-fracturing and carbonshadowing. The rationale of the fracture-flip is as follows: afterfreeze-fracturing, carbon-shadowing (fixation), thawing andwashing, the replicas are picked from above with a Formvar-coated grid and inverted (flipped). Images of the cell surfaceare then obtained by evaporation of Pt/C. In principle, thefracture-flip cannot reveal the inner, cytoplasmic surface of themembrane because the cell body remains attached to the inner,cytoplasmic membrane half. We have developed ‘fracture-flip/Triton X-100’ to overcome this obstacle. We use unfixedcells that, after freeze-fracture and carbon fixation, are treatedwith Triton X-100. The detergent partially dissolves unfrac-tured areas of the plasma membrane, with the release of thecytoplasmic contents. The inner, cytoplasmic half of themembrane, remains attached to the carbon cast. Therefore,both the cytoplasmic and exoplasmic surfaces can then be visu-alized by Pt/C replication. We have applied this technique toreveal the cytoplasmic and exoplasmic surfaces of erythrocytes(Fujimoto and Pinto da Silva, 1992) and isolated nuclearmembranes (Fujimoto and Pinto da Silva, 1989). Althoughnon-ionic detergents such as Triton X-100 are less denaturingand can be used to remove a protein from a membrane whilstpreserving protein structure and protein-protein interactions,they cannot completely solubilize a bulk tissue sample. Thusthe fracture-flip/Triton X-100 technique can be applied to onlyisolated (free) cells or intracellular organelles. Although ionicdetergents (e.g. SDS) tend to denature proteins by destroyingtheir secondary, tertiary and quaternary structure, antigenicityis retained. This property is useful for our approach, but itmight not be helpful in the studies of native protein structure.

The question has naturally arisen as to whether or not themembrane halves including the lipids really are retained afterthe SDS-treatment. Even if the carbon film acts to stabilize thelipids, such a specific sparing of the membrane, bearing inmind the rather crude and harsh nature of the treatment, seemssurprising. Actually, we have succeeded in differential phos-pholipid analysis of cytoplasmic and exoplasmic membranehalves using SDS-FRL with a specific monoclonal antibody to

a phospholipid, indicating an asymmetric distribution of lipidsacross the bilayer membrane (K. Fujimoto and T. Fujimoto,unpublished). Our analyses of membrane proteins and lipidsprovide compelling evidence that the split membrane halvesremain attached to the Pt/C replicas even after SDS-treatment.

We expected that the immunolabeling of gap junctions andtight junctions with the antibodies against the junction proteinswas restricted to the P-faces of these structures, because theantibodies recognize the C-terminal cytoplasmic domain of theproteins. However, the immunogold labeling was observed onboth the P- and E-faces of these junctions. This is thought toinvolve an unfractured junctional component, which binds theantibody and remains attached to the junctional membrane halfrecognized as an E-face (Fig. 6).

In chicks, all types of simple epithelial cells were clearlylabeled with Oc-1 and Oc-2 at their tight junction strands onthe P-faces and the complementary structures on the E-faces.These findings suggest that the strand consists of occludin atleast as one component. The immunolabeling to the linergrooves on the E-faces is of particular interest. As mentionedabove, we reasoned that the labeling of the groove on the E-face is thought to involve an unfractured junctionalcomponent, that binds the antibody and remains attached tothe fractured junctional membrane in which the grooves arevisualized (see Fig. 6, hollow arrowhead). This findings leadto the simple idea that occludins interact with other identicaloccludin molecules in a homophilic manner. However, we donot exclude the possibility that occludins interact with somereceptor molecules, or directly interact with lipids in adjoiningcell membranes (Furuse et al., 1993). In addition, we speculatethat the strand consists of occludin at least as one component,

K. Fujimoto

Fig. 6. Interpretation of the results obtained from SDS-FRL.(A) Stippled areas represent cross-sectional views of integralmembrane proteins within junctional membranes of adjacent cells.(B) After freeze-fracturing along the heavy line (A) and Pt/Creplication, a replica reveals the two fracture faces: the P-face andthe E-face. SDS-FRL using antibody recognizing the cytoplasmicdomain of the integral membrane protein reveals the labeling on boththe P- and E-faces, if the unfractured junctional components (integralmembrane proteins, p) remain attached to the junctional membranehalf and bind the antibody. It stands to reason that the pit (or groove)revealed on the E-face shows labeling (open arrowhead), but the piton the P-face shows no labeling (open arrow).

P-face E-face P-face

A

Bp p p

3449Freeze-fracture replica immunolabeling

but we cannot exclude the possibility that tight junctions areregarded as stabilized instances of linear membrane fusionintermediates and that the strands consist of cylindricalinverted membrane micelles composed of a continuum oflipids and varying complements of integral proteins (Pinto daSilva and Kachar, 1982).

Herein, we have discussed the SDS-FRL technique and thetype of information it can provide. Transmembrane macro-molecules revealed as intramembrane particles, which areindistinguishable on a purely morphological basis, can beselectively labeled using SDS-FRL. Thus, we have beenstudying the dynamics of the intercellular junction proteins,e.g. connexins (K. Fujimoto et al., unpublished) and occludin(M. Furuse et al., unpublished) during junction formation.SDS-FRL is technically easy, but, we have to pay attention tothe partition of integral membrane proteins during freeze-fracture (see Pinto da Silva, 1984, for review). For instance,some membrane proteins that partition with the inner (cyto-plasmic) half of the membrane, can be labeled with theantibody against the cytoplasmic domain, but not labeled withthe antibody against the extracellular domain of the proteins(see Fig. 1D). By contrast, proteins heavily expressed at theouter surface (e.g. proteins of the Ig super family and thecadherin family) tend to partition with the outer half of themembrane. However, partition of these proteins with the outerhalf of the membrane does not necessarily imply dragging ofthe cytoplasmic domains through the inner half of themembrane. In general proteins of the Ig super family and thecadherin family have a single membrane-spanning hydropho-bic domain. Thus a simple breakage of the membrane-spanningdomain could occur. In this case, these proteins can be labeledon the P-faces of the membrane with the antibody recognizingthe cytoplasmic domain. This accounts for the labeling patternof desmoglein, which is known to be homologous withcadherin family proteins (Koch et al., 1990). In conclusion, westress that electron microscopy using SDS-FRL will bridge thegap between biochemistry and more morphologically orienteddisciplines in biomembrane research.

For generously providing the peptide-specific antibodies, I thankDrs Akira Takeda and Takashi Shimazu for the Cx32, Drs AkioKuraoka and Yosaburo Shibata for the Cx26, Dr Masatoshi Takeichifor the E-cadherin and Drs Mikio Furuse and Shoichiro Tsukita forthe occludin. I gratefully acknowledge numerous helpful discussionswith Drs Toyoshi Fujimoto and Toru Noda. I am grateful to Drs PedroPinto da Silva, Ross G. Johnson and Camillo Peracchia for their usefuladvice. I thank Dr Chizuka Ide for his encouragement throughout thisstudy. I also express my sincere gratitude to Dr Kazuo Ogawa for hisencouragement. I thank Ms Sawako Nakamura, Natsuko Hatanakaand Kasumi Nomura for their secretarial assistance, and Mr AkiraUesugi for his help with the photographic work. This work wassupported by a research grant from the Ministry of Education, Scienceand Culture of Japan.

REFERENCES

Andersson-Forsman, C. and Pinto da Silva, P. (1988a). Fracture-flip: newhigh-resolution images of cell surfaces after carbon stabilization of freeze-fractured membranes. J. Cell Sci. 90, 531-541.

Andersson-Forsman, C. and Pinto da Silva, P. (1988b). Label-fracture of cellsurfaces by replica staining. J. Histochem. Cytochem. 36, 1413-1418.

Dinchuk, J. E., Johnson, T. J. A. and Rash, J. E. (1987). Postreplicationlabeling of E-leaflet molecules: membrane immunoglobulins localized insectioned, labeled replicas examined by TEM and HVEM. J. ElectronMicrosc. Tech. 7, 1-16.

Fujimoto, K. and Pinto da Silva, P. (1988). Macromolecular dynamics of thecell surface during the formation of coated pits is revealed by fracture-flip. J.Cell Sci. 91, 161-173.

Fujimoto, K. and Pinto da Silva, P. (1989). Surface views of nuclear pores inisolated rat liver nuclei as revealed by fracture-flip/Triton-X. Eur. J. CellBiol. 50, 390-397.

Fujimoto, K. and Ogawa, K. (1991). Fracture-flip and fracture-flipcytochemistry to study macromolecular architecture of membrane surfaces:practical procedures, interpretation and application. Acta Histochem.Cytochem. 24, 111-117.

Fujimoto, K. and Pinto da Silva, P. (1992). Fracture-flip/Triton X-100 revealsthe cytoplasmic surface of human erythrocyte membranes. Acta Histochem.Cytochem. 25, 255-263.

Furuse, M., Hirase, T., Itoh, M., Nagafuchi, A., Yonemura, S., Tsukita, Sa.and Tsukita, Sh. (1993). Occludin: a novel integral membrane proteinlocalizing at tight junctions. J. Cell Biol. 123, 1777-1788.

Goodenough, D. A. and Gilula, N. B. (1974). The splitting of hepatocyte gapjunctions and zonulae occlusentes with hypertonic disaccharides. J. CellBiol. 61, 575-590.

Gruijters, W. T. M., Kistler, J., Bullivant, S. and Goodenough, D. A.(1987). Immunolocalization of MP70 in lens fiber 16-17 nm intercellularjunctions. J. Cell Biol. 104, 565-572.

Heuser, J. E., Reese, T. S., Jan, L. Y., Dennis, M. J. and Evans, L. (1979).Synaptic vesicle exocytosis captured by quick-freezing and correlated withquantal transmitter release. J. Cell Biol. 81, 275-300.

Koch, P. J., Walsh, M. J., Schmelz, M., Goldscmidt, M. D., Zimbelmann, R.and Franke, W. W. (1990). Identification of desmoglein, a constitutivedesmosomal glycoprotein, as a member of the cadherin family of celladhesion molecules. Eur. J. Cell Biol. 53, 1-12.

Kuraoka, A., Iida, H., Hatae, T., Shibata, Y., Itoh, M. and Kurita, T.(1993). Localization of gap junction proteins, connexins 32 and 26, in rat andguinea pig liver as revealed by quick-freeze, deep-ech immunoelectronmicroscopy. J. Histochem. Cytochem. 41, 971-980.

Pinto da Silva, P. and Branton, D. (1970). Membrane splitting in freeze-eching. Covalently labelled ferritin as a membrane marker. J. Cell Biol. 45,598-605.

Pinto da Silva, P. and Kachar, B. (1982). On tight-junction structure. Cell 28,441-450.

Pinto da Silva, P. (1984). Freeze-fracture cytochemistry. In Immunolabellingfor Electron Microscopy (ed. J. M. Polak and I. M. Varndell), pp. 179-188,Elsevier Science Publishing Co., New York.

Pinto da Silva, P. and Kan, F. W. (1984). Label-fracture: a method for highresolution labeling of cell sufaces. J. Cell Biol. 99, 1156-1161.

Takeda, A., Kanoh, M., Shimazu, T. and Takeuchi, N. (1988). Monoclonalantibodies recognizing different epitopes of the 27-kDa gap junction proteinfrom rat liver. J. Biochem. 104, 901-907.

Yoshida-Noro, C., Suzuki, N. and Takeichi, M. (1984). Molecular nature ofthe calcium-dependent cell-cell adhesion system in mouse teratocarcinomaand embryonic cells studied with a monoclonal antibody. Dev. Biol. 101, 19-27.

(Received 14 April 1995 - Accepted 25 July 1995)