Three-dimensional structure of a human connexin26gap junction channel reveals a plug in the vestibuleAtsunori Oshima†‡, Kazutoshi Tani†, Yoko Hiroaki†‡, Yoshinori Fujiyoshi†‡§¶, and Gina E. Sosinsky¶�

†Department of Biophysics, Faculty of Science, Kyoto University, Oiwake, Kitashirakawa, Sakyo-ku, Kyoto 606-8502, Japan; �National Center for Microscopyand Imaging Research, Department of Neurosciences, University of California at San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0608; ‡Core Researchfor Evolution Science and Technology (CREST), Japan Science and Technology Agency (JST), Oiwake, Kitashirakawa, Sakyo-ku, Kyoto 606-8502, Japan;and §Japan Biological Information Research Center (JBIRC), National Institute of Advanced Industrial Science and Technology (AIST), 2-41-6, Aomi,Koto-ku, Tokyo 135-0064, Japan

Communicated by David J. DeRosier, Brandeis University, Waltham, MA, April 24, 2007 (received for review January 29, 2007)

Connexin molecules form intercellular membrane channels facili-tating electronic coupling and the passage of small moleculesbetween adjoining cells. Connexin26 (Cx26) is the second smallestmember of the gap junction protein family, and mutations in Cx26cause certain hereditary human diseases such as skin disorders andhearing loss. Here, we report the electron crystallographic struc-ture of a human Cx26 mutant (M34A). Although crystallizationtrials used hemichannel preparations, the density map revealedthat two hemichannels redocked at their extracellular surfaces intofull intercellular channels. These orthorhombic crystals containedtwo sets of symmetry-related intercellular channels within threelipid bilayers. The 3D map shows a prominent density in the poreof each hemichannel. This density contacts the innermost helices ofthe surrounding connexin subunits at the bottom of the vestibule.The density map suggests that physical blocking may play animportant role that underlies gap junction channel regulation. Ourstructure allows us to suggest that the two docked hemichannelscan be independent and may regulate their activity autonomouslywith a plug in the vestibule.

connexin channels � electron crystallography � intercellularcommunication � membrane protein structure � two-dimensional crystals

Gap junctions contain intercellular communication channelsthat allow a wide variety of solutes with different sizes to betransferred between the cytoplasm of adjacent cells. Thesesolutes include ions, metabolites, nucleotides, peptides, andsecondary messengers. Gap junction channels have critical rolesin many biologically important processes including cardiac de-velopment, fertility, the immune system, and electrical signalingin the nervous system (1). The diversely expressed connexin26(Cx26) is the second-smallest member of the conserved mam-malian gap junction protein family. Hereditary mutations inhuman Cx26 cause macroscopic symptoms such as certain skindisorders and nonsyndromic and syndromic deafness (2).

Early electron microscopic studies suggested that closure ofthe gap junction channel occurs by rotating all six subunits ineach hemichannel (3, 4). More recently, a 3D structure of atruncated form of Cx43 reported on the arrangement and theassignments of the four transmembrane �-helical bundle in eachconnexin (5, 6). Electrophysiological experiments have shownthat the voltage sensor involves charged residues in the connexinN terminus and that it appears to face the aqueous pore (7, 8),and studies of hemichannel conductivity favor ‘‘an individualsubunit gating model’’ (9). In addition, the selectivity of gapjunctions for permeation of small molecules under �1 kDa (alsoknown as ‘‘permselectivity’’) depends on the connexin isoform.Each isoform has unique physiological responses to ions, phos-phorylation, and pH. Permselectivity is hypothesized to occurindependently from voltage gating, implying that gap junctionchannels possess multiple gating mechanisms (10).

In this study, we focus on the structure of Cx26 gap junctionchannels, because the short cytoplasmic tail of Cx26 makes itmore amenable for the formation of 2D crystals. It is believed

that the relatively unordered carboxyl terminus of larger iso-forms interferes with the necessary tight packing in a crystal. Inaddition, we used a site-specific mutant of human Met-34,hCx26M34A, because this mutant expresses in baculovirus in-fected Sf9 cells at higher quantities than wild-type Cx26-infectedcells. The hCx26M34A mutant is a single-site mutation at thesame position as the hCx26M34T mutant, which can causeprelingual nonsyndromic hereditary deafness (11). Although notas well characterized as hCx26M34T (12), this single-pointmutation of a Met to Ala decreases dye coupling in exogenouslytransfected HeLa cells and forms structures indistinguishablefrom wild-type gap junctions (13). We succeeded in making 2Dcrystals of the hCx26M34A suitable for cryoelectron crystallog-raphy. The 3D structure of hCx26M34A gap junction channelsreveals a prominent density in the pore of each hemichannel,suggesting that the channel is blocked by a physical obstruction.

ResultsTwo-Dimensional Crystallization of Connexin26 Complexes.hCx26M34A gap junction channels were expressed in and iso-lated from Sf9 insect cells [see supporting information (SI) Fig.7A]. Hemichannels (connexons) were isolated by affinity puri-fication using a C-terminal hexa-histidine tag (SI Fig. 7B).Purified hemichannels were mixed with the lipid dioleoylphos-phatidylcholine (DOPC) at a lipid-to-protein ratio of 1 (wt/wt).Reconstitution into lipid bilayers by dialysis produced 2D crys-tals �1 �m in diameter (SI Fig. 8A). Although the purifiedhemichannel is hexameric, the 2D arrays obtained by dialysisshowed an orthorhombic crystal lattice (SI Fig. 8B).

We recorded images of the hCx26M34A crystals embedded in0.05–1% tannic acid, 2–40% trehalose, or the combination ofthem. Computed diffraction patterns of one of the best 0° imagesshowed reflections to a resolution of �11 Å (SI Fig. 8C). Afterimage processing to correct crystal distortions, the resolutionimproved to 7 Å (SI Fig. 8D). Image processing revealed that thecrystals had p22121 symmetry and unit cell parameters of a �112.4 Å, b � 111.2 Å, and � � 90°.

Author contributions: A.O. and Y.F. designed research; A.O., K.T., Y.H., Y.F., and G.E.S.performed research; A.O., K.T., Y.H., Y.F., and G.E.S. analyzed data; and A.O., Y.F., andG.E.S. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

Abbreviations: Cx: connexin, Cx26M34A: connexin26 site-specific mutant Met34Ala.

Data deposition: The cryoEM structure reported in this paper has been deposited in theMacromolecular Structure Database (MSD), www.ebi.ac.uk/msd-srv/emsearch/index.html(accession no. EMD-1341).

¶To whom correspondence may be addressed. E-mail: gsosinsky@ucsd.edu or yoshi@em.biophys.kyoto-u.ac.jp.

This article contains supporting information online at www.pnas.org/cgi/content/full/0703704104/DC1.

© 2007 by The National Academy of Sciences of the USA

10034–10039 � PNAS � June 12, 2007 � vol. 104 � no. 24 www.pnas.org�cgi�doi�10.1073�pnas.0703704104

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021

http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1http://www.pnas.org/cgi/content/full/0703704104/DC1

Three-Dimensional Structure Determination and Organization ofhCx26M34A Orthorhombic Crystals. To determine a 3D structure,we collected images of samples tilted up to 45° and combinedthem to produce a density map at a resolution of 10 Å in themembrane plane and 14.1 Å normal to the membrane plane (Fig.1, SI Table 1, and SI Fig. 8E). A side view of the 3D map reveals

that the crystals have a thickness of �240 Å and contain threelipid bilayers (labeled Mem-1, Mem-2, and Mem-3 in Fig. 1).Remarkably, the map also shows that the hemichannels re-docked through their extracellular surfaces, forming completegap junction channels (Fig. 1 A and B). This is consistent withpublished results proposing extensive hydrophobic surfaces in

Fig. 1. Three-dimensional structure of Cx26 orthorhombic crystals.(A) Molecular packing of Cx26 in the 2D crystal. The gap junction channels are incorporatedin three lipid bilayers (Mem-1–Mem-3) with 21 symmetry along Mem-2. (B) View of the Cx26 density map perpendicular to the membrane plane. The threemembranes, indicated by gray bars, surround two extracellular gap regions. The map is contoured at 1.0� (light blue) and 2.4� (wheat color) above the meandensity. The transmembrane �-helical ribbon model (6) is docked into the density for one of the hemichannels. The four helices are color-coded as in Fig. 2. Twohelices D make contact with adjacent gap junction channels (red arrows). (Scale bar, 40 Å.) (C and D) Forty-angstrom-thick sections through the density mapparallel to the membrane plane, showing protein embedded in membranes Mem-1 (identical to Mem-3) (C) and Mem-2 (D). Tail ends of two helices D areindicated by red arrows as in B. (Scale bars, 40 Å.)

Oshima et al. PNAS � June 12, 2007 � vol. 104 � no. 24 � 10035

BIO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021

http://www.pnas.org/cgi/content/full/0703704104/DC1

the gap region (14). In bilayers Mem-1 and Mem-3, thehemichannels show poorer density than in Mem-2 (Fig. 1 C andD), presumably because of variability in the molecular packingbecause of the large lipid areas between individual hemichannelsand/or the flexibility of the cytoplasmic domains of the connexinsubunits. The cytoplasmic structures in Mem-1 and Mem-3 mayalso be deformed by their contact with the carbon film to whichthe crystals are adsorbed in the sample preparation procedurefor cryoelectron microscopy. By contrast, the hemichannels inMem-2 are protected from any forces such as the surface tensionupon specimen drying and mechanical interactions with thecarbon film. Therefore, the structural features of the hemichan-nels in Mem-2 should be the most accurate and, in particular,preserve the structure of the flexible cytoplasmic domains of theconnexins. Thus, the following description of the gap junctionstructure is based on the hemichannels in Mem-2 unless notedotherwise.

Each connexon in a gap junction membrane channel containsfour transmembrane segments, usually referred to as M1, M2,M3, and M4 (15). The combination of our x–y resolution of 10Å and the well defined cytoplasmic and transmembrane struc-

ture made it possible to unambiguously dock the proposedribbon model of the transmembrane domain (PDB accession no.1TXH) with slight modifications (Fig. 2). A comparison of thetransmembrane cylinder models for Cx43 (6) with our Cx26structure (Fig. 3) show that the dimensions of the channel, thesize of the pore constriction, and the positions of the helices areall essentially the same. It should be noted that Cx26 is part ofthe �-subgroup of the connexin family, whereas Cx43 is amember of the �-subgroup, yet the overall structure of these twoconnexin isoforms is very similar. Small shifts in the positionsand tilts of these superimposed helices are more likely becauseof the different crystal forms used in the structure analysis forCx43 and Cx26 (hexagonal versus orthorhombic) than differ-ences between the isoforms.

There are three protrusions on the cytoplasmic surface for eachfour-helix bundle, A�, B�, C, and D (Figs. 2 and 4). The cytoplasmicextension of helix D makes contact with the one from the adjacentgap junction channel, stabilizing the crystal packing (red arrows inFig. 1 B and D). Four-helix bundles are connected by a bridge-likedensity forming the largest of the three cytoplasmic protrusions(white arrows in Fig. 2; and see Fig. 4). This new density could

Fig. 2. Structural details of the Cx26 gap junction. The map is contoured as in Fig. 1. (Inset) Twenty-angstrom-thick section perpendicular to the membraneplane through the density map of a hemichannel in Mem-2. This section corresponds to the region enclosed by the white lines shown in A. The arrowhead pointsto the large density in the pore. The inner cytoplasmic protrusions (white arrows) extend from the cytoplasmic ends of helices B and C. (A–C) Thirty-angstrom-thickslabs through the density map corresponding to the position of the lines shown in Inset. The four helices are labeled A (cyan, A�), B (green, B�), C (yellow), andD (pink) as in the original Cx43 structure (5). The arrowhead and white arrows represent the plug and the inner cytoplasmic protrusions, respectively, as in Inset.

10036 � www.pnas.org�cgi�doi�10.1073�pnas.0703704104 Oshima et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021

represent part of the cytoplasmic loop of Cx26, because this domaincaps the top of helices B and C. The cytoplasmic loop has a largernumber of amino acid residues than the N or C terminus, and thusshould have a larger mass density. This implies that the connexinpolypeptide boundary may be across adjacent four-helix bundles.However, flexible loops are often invisible in crystallographic maps.The two loops in the extracellular gap domain are not clearlyseparated (Fig. 2C) and are postulated to form a differentlyorganized subdomain (16). At this time, the resolution is insufficientto assign these �-helical segments and the polypeptide boundary tospecific sequences within the transmembrane domain.

The Prominent Pore Density in the Vestibule. The 3D map shows adensity in the center of the pore (white arrowheads in Figs. 2 and4). Because this density is observed in both projection maps oftannic acid- and trehalose-embedded specimens (data notshown), it is most likely that the plug is part of the connexin andnot an artifact of the embedding agent. This plug density isclearly visible even in a map contoured at 2.4� (wheat-coloredcontour in Fig. 2). The plug is located inside of the membranelayer and forms contacts with the surrounding channel wall,which, at the constricted part of the vestibule, is formed by theinnermost helices C (Fig. 2B). This density strongly suggests that

a plug physically blocks the channel within the membrane.Density for the plug is also observed in the pores of thehemichannels in membranes Mem-1 and Mem-3 (Fig. 1 B andC). Each hemichannel has its own plug, conferring on it theability to gate its pore autonomously. It is possible that thetransjunctional voltage sensor and the physical gate resideexclusively within a single hemichannel (7, 17).

The connection of the plug to the top of the channel wall is notperfectly resolved, presumably because of flexibility of the linker(Fig. 2 Inset). We applied a B-factor of �700 to the map toenhance the higher-resolution weak amplitude data. This mapshows that the plug consists of six �-helix-like features and thatit has four protrusions, implying connections from the plug (Fig.5). It is not surprising that only four protrusions rather than sixappear from the plug because 6-fold symmetry is lost in thisorthorhombic crystal form. These features strongly suggest thatthe plug is formed by either the N- or C-terminal helices of Cx26.

DiscussionKey findings of this study are that hydrophobic interactions at theextracellular domains drive hemichannels to redock into do-decameric channels, the transmembrane domain structure isfairly conserved between �- and �-connexins and differentcrystal forms, and each hemichannel has its own plug in itsvestibule. The crystal form shown here is unprecedented in thegap junction structure literature because it contains three mem-branes and two sets of symmetry-related intercellular channels.This could occur only because removal of the detergent fromhemichannels causes the hydrophobic extracellular surfaces tobe exposed. Rather than have these surfaces face an aqueousenvironment, hemichannels redocked into an intercellular chan-nel. Comparison of the shape, size, and arrangements of thetransmembrane helices in the hCx26M34A structure with thetruncation mutant of Cx43 show that the architecture of the gapjunction channel is conserved between these two isoforms.Without the resolution to unambiguously trace the primarysequence of the connection from the plug to the surroundingchannel wall, we can only speculate that the plug arises from oneof the termini, in particular, the N terminus.

Possible Candidates for Components of the Plug. Our Cx26 gapjunction crystal structure shows that the channel vestibule isblocked by a physical obstruction we call the ‘‘plug.’’ Whetherthis structure reflects a functionally closed nonpermeant channelor an aberrant mutant channel remains to be determined.However, it is unlikely that a single amino acid change by itselfwould give rise to such a prominent feature. We have obtainedseveral 0° projection micrographs of wild-type 2D crystals whosecrystallinity is not as good as crystals of the hCx26M34A mutant.These 2D reconstructions have also revealed the density in thepore (data not shown). This last finding strongly implies that thephysical channel closure may be regulated independently in eachof the hemichannels and that this plug may make a channelimpermeant to larger molecules such as those used in ourprevious dye-transfer studies (13).

The likely candidates for the plug and its connector are the Nterminus, the cytoplasmic loop, and the C terminus of Cx26. TheN terminus is the most probable candidate of the three. A wealthof electrophysiological studies has established that the transjunc-tional voltage sensor resides in the N terminus and that theresidues sensing changes in the voltage field are located in thevestibule of the pore (7–9, 18). It has also been proposed thatmovement of the charges in the N terminus initiate channelgating (7, 19). This notion is supported by an NMR solutionstructure of an N-terminal peptide of Cx26 (20), although it hasnot yet been shown definitively that the N terminus forms thephysical assembly. We propose that the plug, which is locatedwithin the pore region, and is therefore ideally located to detect

Fig. 3. Superimposition of the transmembrane helices of Cx26 (yellow) andCx43 (red) (6). The arrangement of the helices differs slightly between the twoconnexins, but all of the corresponding helices more or less overlap with eachother. The channel dimensions and the pore diameters are approximately thesame in gap junctions formed by Cx26 and Cx43.

Fig. 4. Stereoview of the surface structure of Cx26 perpendicular to themembrane plane. The cytoplasmic protrusions are clearly defined. The map iscontoured at 1.0� (solid surface) and at 2.4� (wheat-colored mesh) above themean density. Helices B (green) and C (yellow) are color coded as in Fig. 2.

Oshima et al. PNAS � June 12, 2007 � vol. 104 � no. 24 � 10037

BIO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021

the transjunctional voltage field, contains the N termini of theconnexin subunits. This is different from voltage-sensitive K� orNa� channels, where the S4 domain is the voltage sensor butdoes not form an assembled plug (21). It should be noted thatsequence analysis has shown that only N termini have verysimilar lengths and a highly conserved sequence, but both thecytoplasmic loops and the C termini show variable lengths andless sequence conservation (22). This sequence conservation ofthe N terminus might, in turn, result in conservation of the plugfeature and closing mechanism within the connexin channelfamily.

The C terminus is the next most likely candidate for the plug.The Cx26 construct in this work has a hexa-histidine tag with athrombin digestion recognition sequence linker that results in�30 aa residues assigned to the C-terminal tail and makes itslightly longer than the N-terminal tail (21 aa residues). It hasalso been proposed that the C terminus and the part of cyto-plasmic loop is involved in pH regulation of gap junction gating,which is referred to as ‘‘particle-receptor’’ model (23), althoughthere is no direct evidence that these domains interact within thevestibule. Whereas Cx26 channels do gate at low pH (24), thismodel has not been demonstrated for Cx26 channels because ofthe very short C-terminal tail. Another mechanism has beenproposed for pH gating of Cx26 hemichannels (25, 26), althoughit has been reported that addition of tags to the Cx26 C terminuseliminates the ligand binding that promotes pH-induced channelclosure (27, **). Furthermore, we found that connexins withoutthe C-terminal tag never crystallized, and removal of the tagafter crystallization destroyed the crystals (data not shown). It ismore likely that the end of helix D corresponds to the hexa-histidine tag that serves to stabilize crystal formation rather thanforming plug. The idea that the C terminus extends from helixD is consistent with the assignment of helix D as M4 as suggested(6, 28).

The last candidate for the plug would be the cytoplasmic loopof Cx26, however, the number of amino acids (�35 residues) andrequirement that this part of the sequence makes a loop cannotaccount for connecting densities from the plug (Fig. 5). Previ-ously published work has suggested that the N terminus andcytoplasmic loop of Cx26 may interact directly (18). Consideringthat even the largest cytoplasmic protrusion could be too smallto cover the length of the cytoplasmic loop (Figs. 2 and 4), it ispossible for the N terminus and the cytoplasmic loop to coop-erate with each other, resulting in forming the plug.

The Plug Creates a Blocked State of Cx26 Gap Junction Channel. Sincethe M34T substitution of hCx26 was reported to be the cause ofnonsyndromic hearing loss (11), the functional role of theposition 34 has been studied, and it has been suggested thatM34T leads to a constriction of the channel pore (12). In ourwork, it is conceivable that the plug in the pore is locked intoplace by the hCx26M34A mutation because the physical obstruc-tion is consistent with the decreased permeability of thehCx26M34A mutation (13). In this case, the mutation at the M34site in M1 could affect the movement of the M1 helix in eachconnexin in each hemichannel, thus resulting in a change of theplug position that makes the channels nonfunctional or poorlyfunctional.

It should be noted that not only have we used a permeabilitymutant for this study but also that we have used crystallizationconditions (low pH, aminosulfonate buffer, carbenoxolone, andhigh Ca2� and Mg2�) that generally promote a closed Cx26channel. The remaining pore openings at the constriction site are�8 Å in diameter (Fig. 2B), indicating that these openings aretoo small for fully hydrated ions to pass (29). If our structure ofthe blocked state resembles a functionally closed state becauseof those factors in the crystallization condition, it allows us tosuggest a plug gating mechanism. Specifically, the two interact-ing Cx26 hemichannels can gate their channel directly andautonomously with a plug (Fig. 6). In this model, a Cx26 gapjunction is open only when both hemichannels release their plugstoward the cytoplasmic side. Once open, the pore can immedi-ately conduct large molecules such as peptides with a molecularmass of up to 1.8 kDa (30), because the large pore size (�15 Å

**Tao, L., Harris, A. L. (2005) Biophys J 88:201a (abstr.).

Fig. 5. Stereo top view of the Cx26 density map to which a B-factor of �700 was applied. The B-factor was applied to enhance the amplitudes of thehigh-resolution reflections, revealing six �-helix-like features in the plug density and protrusions that probably reflect the loops connecting the plug to thesurrounding channel wall. The four helices are color-coded A (cyan, A�), B (green, B�), C (yellow), and D (pink) as in Fig. 2.

Fig. 6. Hypothesized plug gating mechanism of gap junctions. Eachhemichannel (green) can regulate its channel activity autonomously. The gapjunction is open only when the plugs (red) in both hemichannels are displacedfrom the channel constriction formed by the innermost helices C (yellow)toward the cytoplasmic side. The flexible connections of the plug with thechannel are shown as red dashed lines.

10038 � www.pnas.org�cgi�doi�10.1073�pnas.0703704104 Oshima et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021

in the most constricted region) virtually does not change duringthe gating cycle. Given that our structure and others (15) haveindicated that the cytoplasmic domains of connexin are verylabile, the movement of the plug could also be regulated byinteractions with other cytoplasmic connexin domains, whichmight account for the diversity of channel properties betweendifferent connexins.

In either case, the fact that each hemichannel has its own plugsupports the argument that the relationships between two op-posing hemichannels can be independent of one another. Thisnotion well reconciles the observation of functional conductivehemichannels such as Cx46, Cx56, and Cx32 chimera designatedCx32*Cx43E1 (31–33). Because the current Cx43 structure doesnot have any plug density in the pore (5, 6), further studies arenecessary to verify whether the plug structure can be generalizedto all connexins. Structures at higher resolution of the open andclosed states will provide more structural detail and will beneeded to fully understand the functional role of a plug andgating mechanisms in these widely expressed channels.

Materials and MethodsProtein Purification and 2D Crystallization. hCx26M34A hemichan-nels were purified as described (13), with the minor modificationthat the protein was eluted with 300 mM L-histidine instead of300 mM imidazole. Purified hCx26M34A hemichannels weremixed with decyl maltoside-solubilized DOPC (Avanti PolarLipids, Alabaster, AL) at a lipid-to-protein ratio of 1. Themixture was dialyzed against 10 mM Mes (pH 5.8), 100 mMNaCl, 50 mM MgCl2, 5 mM CaCl2, 2 mM DTT, 100 �Mcarbenoxolone (Sigma, St. Louis, MO), 0.005% NaN3, and 1%glycerol. Dialysis was performed at 20°C for the first 24 h, 37°Cfor the next 96 h, and 20°C for the final 24 h.

Sample Preparation. Samples were negatively stained with 2%uranyl acetate for screening of 2D crystals. For cryoelectronmicroscopy, crystals were mixed with dialysis buffer containing0.05–1% tannic acid, 2–40% trehalose, or their combination andapplied to molybdenum grids covered with a thin carbon film.The different concentration of tannic acid or trehalose did notaffect the plug in the pore because all of the 0° projectionsobtained from any preparations showed the plug density as well(data not shown). After removal of excess liquid, the grid wasblotted with filter paper and plunged into liquid nitrogen.

Cryoelectron Microscopy and Structure Determination. Frozen spec-imens were transferred into a JEM-3000SFF electron micro-scope (JEOL, Tokyo, Japan) operated at 300 kV and equippedwith a field emission gun and a superfluid helium stage (34). Thespecimens were cooled to a temperature of 4K, and images ofspecimens tilted up to 45° were recorded on SO-163 film (Kodak,Rochester, NY) at a magnification of �60,000 with an electrondose of 25 electrons per Å2. The quality of images was checkedby optical diffraction, and selected images were digitized with aSCAI scanner (Zeiss, Thornwood, NY) by using a step size of 7�m. Lattice distortions and the contrast-transfer function werecorrected with the MRC package (35–37). The final density mapis based on 254 images that were combined to generate a mergedphase and amplitude data set. The 3D density map was visualizedwith the program Pymol (http://pymol.sourceforge.net/).

This work is supported by Grants-in Aid for Specially Promoted Re-search, Grant-in Aid for 21st Century Centers of Excellence, KyotoUniversity, Japan Science and Technology Corporation and New Energyand Industrial Technology Development Organization (to Y.F.), Na-tional Institutes of Health Grant GM065937 (to G.E.S.), and NationalScience Foundation Grants MCB-0543934 (to G.E.S.) and RR04050 (toMark Ellisman, National Center for Microscopy and Imaging Research,University of California at San Diego).

1. Harris AL (2001) Q Rev Biophys 34:325–472.2. Kelsell DP, Di WL, Houseman MJ (2001) Am J Hum Genet 68:559–568.3. Unwin PNT, Zampighi G (1980) Nature 283:545–549.4. Unwin PNT, Ennis PD (1984) Nature 307:609–613.5. Unger VM, Kumar NM, Gilula NB, Yeager M (1999) Science 283:1176–1180.6. Fleishman SJ, Unger VM, Yeager M, Ben-Tal N (2004) Mol Cell 15:879–888.7. Verselis VK, Ginter CS, Bargiello TA (1994) Nature 368:348–351.8. Purnick PE, Oh S, Abrams CK, Verselis VK, Bargiello TA (2000) Biophys J

79:2403–2415.9. Oh S, Abrams CK, Verselis VK, Bargiello TA (2000) J Gen Physiol 116:13–31.

10. Peracchia C (2004) Biochim Biophys Acta 1662:61–80.11. Kelsell DP, Dunlop J, Stevens HP, Lench NJ, Liang JN, Parry G, Mueller RF,

Leigh IM (1997) Nature 387:80–83.12. Bicego M, Beltramello M, Melchionda S, Carella M, Piazza V, Zelante L,

Bukauskas FF, Arslan E, Cama E, Pantano S, et al. (2006) Hum Mol Genet15:2569–2587.

13. Oshima A, Doi T, Mitsuoka K, Maeda S, Fujiyoshi Y (2003) J Biol Chem278:1807–1816.

14. Ghoshroy S, Goodenough GA, Sosinsky GE (1995) J Memb Biol 146:15–28.15. Sosinsky GE, Nicholson BJ (2005) Biochim Biophys Acta 1711:99–125.16. Foote CI, Zhou L, Zhu X, Nicholson BJ (1998) J Cell Biol 140:1187–1197.17. Harris AL, Spray DC, Bennett MV (1981) J Gen Physiol 77:95–117.18. Oh S, Rubin JB, Bennett MVL, Verselis VK, Bargiello TA (1999) J Gen Physiol

114:339–364.19. Oh S, Rivkin S, Tang Q, Verselis VK, Bargiello TA (2004) Biophys J

87:912–928.20. Purnick PE, Benjamin DC, Verselis VK, Bargiello TA, Dowd TL (2000) Arch

Biochem Biophys 381:181–190.

21. Jan LY, Jan YN (1992) Annu Rev Physiol 54:537–555.22. Cruciani V, Mikalsen SO (2006) Cell Mol Life Sci 63:1125–1140.23. Delmar M, Coombs W, Sorgen P, Duffy HS, Taffet SM (2004) Cardiovasc Res

62:268–275.24. Stergiopoulos K, Alvarado JL, Mastroianni M, Ek-Vitorin JF, Taffet SM,

Delmar M (1999) Circ Res 84:1144–1155.25. Bevans CG, Harris AL (1999) J Biol Chem 274:3711–3719.26. Yu J, Bippes CA, Hand GM, Muller DM, Sosinsky GE (2007) J Biol Chem

282:8895–8904.27. Tao L, Harris AL (2007) Mol Pharmacol 71:570–579.28. Skerrett IM, Aronowitz J, Shin JH, Cymes G, Kasperek E, Cao FL, Nicholson

BJ (2002) J Cell Biol 159:349–360.29. Miyazawa A, Fujiyoshi Y, Unwin N (2003) Nature 423:949–955.30. Neijssen J, Herberts C, Drijfhout JW, Reits E, Janssen L, Neefjes J (2005)

Nature 434:83–88.31. Paul DL, Ebihara L, Takemoto LJ, Swenson KI, Goodenough DA (1991) J Cell

Biol 115:1077–1089.32. Ebihara L, Berthoud VM, Beyer EC (1995) Biophys J 68:1796–1803.33. Pfahnl A, Zhou XW, Werner R, Dahl G (1997) Pflügers Arch 433:733–

779.34. Fujiyoshi Y, Mizusaki T, Morikawa K, Yamagishi H, Aoki Y, Kihara H, Harada

Y (1991) Ultramicroscopy 38:241–251.35. Henderson R, Baldwin JM, Downing KH, Lepault J, Zemiln F (1986)

Ultramicroscopy 19:147–178.36. Henderson R, Baldwin JM, Ceska TA, Zemlin F, Beckmann E, Downing KH

(1990) J Mol Biol 213:899–929.37. Crowther RA, Henderson R, Smith JM (1996) J Struct Biol 116:9–16.

Oshima et al. PNAS � June 12, 2007 � vol. 104 � no. 24 � 10039

BIO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Mar

ch 3

0, 2

021