Reconstitution of Functional Voltage-gated Chloride Channels from Complementary Fragments of CLC-1*

We investigated the effect of truncations on the hu- man muscle chloride channel CLC-1 and studied the functional complementation from partial proteins. Al- most complete deletion of the cytoplasmic amino terminus did not affect currents, but truncating the intracel- lular COOH terminus after Leu 720 abolished function. Currents were restored by coexpressing this membrane- embedded part with the lacking cytoplasmic fragment that contains domain D13, the second of the two con- served cystathionine (cid:98) -synthase (CBS) motifs present in all eukaryotic CLC proteins. However, if the cut was after Gln 597 before the first CBS domain, no functional complementation was seen. Complementation was also obtained with channels “split” between transmembrane domains D7 and D8 or domains D8 and D9, but not when split between D10 and D11. Specificity of currents was tested by inserting point mutations in NH 2 -terminal (G188A and G230E) or COOH-terminal (K585E) fragments. In contrast to G188A and K585E, split channels did not tolerate the D136G mutation, suggesting that it may impede association from nonlinked fragments. Duplication, but not a lack of domain D8 was tolerated in “split” channels. Membrane domains D9–D12 can insert into the membrane without adding a preceding signal peptide to ensure the extra- cellular amino terminus of D9. Eventually, we succeeded in reconstituting CLC-1 channels from three separate polypeptides: the amino-terminal part up to D8, D9 through CBS1, and the remainder of the cytoplasmic carboxyl terminus.

CLC 1 chloride channels, originally identified by the expression cloning of CLC-0 from Torpedo electric organ (1), form a large gene family with at least nine members in mammals and conservation down to organisms like Escherichia coli and yeast (for review, see Ref. 2). Their importance is underscored by human inherited diseases; mutations in the muscle channel CLC-1 lead to myotonia (3,4), and those in CLC-5 lead to proteinuria and kidney stones (5).
CLC proteins are structurally unrelated to other channels, including Cl Ϫ channels like ␥-aminobutyric acid and glycine receptors or the cystic fibrosis transmembrane regulator CFTR. Hydrophobicity analysis indicated 13 hydrophobic domains (D1-D13 (1)). However, newer experimental findings (6 -8) suggest the presence of only 10 (or 12) transmembrane domains ( Fig. 1, top). The topology of the D9 -D12 region, a long hydrophobic stretch interrupted only once by a short hydrophilic segment, still poses problems. Both the amino-and the carboxyl terminus reside in the cytoplasm, and the loop between D8 and D9 is glycosylated.
CLC-0 forms homodimers with one pore per subunit (7,9,10), and this may also apply for CLC-1 (11). This one-proteinone-pore architecture distinguishes the CLC channels from voltage-gated cation channels. In shaker K channels, four homologous subunits form a single pore, with equivalent parts of each subunit contributing to it. Although the ␣-subunit of sodium and calcium channels consists of a single polypeptide chain, it shows the same 4-fold repetition of the K channel subunit motif. Ligand-gated anion channels like the ␥-aminobutyric acid A receptor are pentamers, with the second TMD of each monomer contributing to the single common pore. By contrast, and in this respect similar to CLC channels, CFTR also contains 12 TMDs forming one pore per protein. In contrast to CLC channels, it functions as a monomer and consists of two halves that are interrupted by a large cytoplasmic loop. It has two nucleotide binding folds and is a member of the ABC transporter superfamily. In channels with one pore per subunit, several different parts of the polypeptide chain must contribute to the pore, as has been shown both for CFTR (12)(13)(14)(15) and CLC channels (9, 16 -18). These parts must be positioned correctly by intramolecular interactions.
In the present work, we show that some of these interactions are strong enough to enable the functional expression of CLC-1 channels from individual polypeptides representing complementary parts of the channel protein. Surprisingly, this not only functions when the "cut" lies between transmembrane spans, but also if an otherwise inactive, COOH-terminally truncated channel is coexpressed with a carboxyl-terminal part lacking any TMD. Here, functional complementation is observed if the cut lies between the first and the second CBS domain (19,20) but not if the COOH-terminal part comprises both CBS domains. In addition, we show that "split" channels with a duplicated D8 domain are functional, indicating that D8 lies at the periphery of the channel.

EXPERIMENTAL PROCEDURES
Mutagenesis and Channel Expression-Human CLC-1 (GenBank TM accession number Z25884) was cloned between NcoI and EcoRI sites of pTLN (21), yielding pTLNH1. Carboxyl-terminally truncated channels (⌬C 390 , ⌬C 451 , ⌬C 520 , ⌬C 597 , and ⌬C 720 ; numbers indicate last amino acid position) were generated by PCR using 3Ј-primers that contained two stop codons and an EcoRI restriction site (GCGAATTCTTATCA . . .) in addition to the template-specific nucleotides. The 5Ј-primer was chosen upstream of the nearest single-cutter restriction site so that after digestion the PCR product could be ligated into pTLNH1 cut with the same enzymes. Amino-terminal truncations (⌬N 110 , ⌬N 369 , ⌬N 413 , ⌬N 509 , ⌬N 598 , and ⌬N 721 ; number indicates first amino acid position) were constructed using an upstream PCR primer containing the first two codons of CLC-1 (methionine and glutamate) including an NcoI site (GCGATACCATGGAG . . . ) plus template specific sequences of the new NH 2 terminus. The 3Ј-primer was placed downstream of an appropriate restriction site. Again, PCR products and vector pTLNH1 were cut with the same enzymes and ligated. Since construct ⌬N 413 starts in an extracellular loop, we constructed ⌬N 413 ϩ SP in which the signal peptide of the rat nicotinic acetylcholine receptor ␣-subunit (amino acids 1-47) replaces the first two residues of mutant ⌬N 413 to direct the amino terminus to the extracellular side. Construct ⌬N 413 ϩ SP ⌬C 720 is truncated at either end and was obtained by replacing the COOH terminus of mutant ⌬N 413 ϩ SP with the corresponding NdeI/EcoRI fragment from mutant ⌬C 720 .
Point mutations D136G, G188A, K585E, and R496S were introduced into full-length proteins or truncated channels ⌬C 451 or ⌬N 413 and ⌬N 413 ϩ SP by recombinant PCR. All PCR-derived fragments were sequenced. The plasmids were linearized, and capped cRNA was transcribed using SP6 polymerase (mMessage mMachine cRNA synthesis kit, Ambion). cRNA was injected into manually defolliculated Xenopus oocytes, which were kept in modified Barth Electrophysiology-Currents were measured at room temperature in ND96 (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl 2 , 1 mM MgCl 2 , 5 mM Hepes, pH 7.4) using a two-electrode voltage clamp with a Turbotec amplifier (Npi Instruments) and pCLAMP 5.5 software (Axon Instruments). A typical pulse protocol is depicted in Fig. 2. Currents were recorded in response to 150-ms voltage steps ranging from ϩ80 to Ϫ160 mV after a 25-ms prepulse to ϩ80 mV (to open the gate) from the resting potential of the individual oocyte. This was followed by a constant step to Ϫ60 mV and a return to the resting potential. For mutant G188A, the voltage was stepped for 4 s from the resting potential to voltages from ϩ20 to Ϫ160 mV in steps of Ϫ20 mV, followed by a constant pulse at Ϫ85 mV. Leakage or capacitive currents were not subtracted, but in the figures capacitive transients were cut off for clarity. Mutants were measured in at least two batches of oocytes (n Ն 6).

The Amino Terminus of CLC-1 Is Dispensable, while the Carboxyl Terminus Is Needed and Can Function as a Separate
Polypeptide-Both the amino and carboxyl termini differ in length among CLC family members and are poorly conserved except for few regions in the carboxyl terminus, including domain D13. We asked whether they serve specific functions and at first deleted the amino terminus between glutamate 2 and valine 110 (Fig. 1). This leaves only about 10 amino acids before the first TMD. Expression in Xenopus oocytes resulted in currents ( Fig. 2, ⌬N 109 ) indistinguishable from wild type (WT). CLC-1 WT currents have a chloride Ͼ iodide ion selectivity, are inwardly rectifying, and display a rapid, partial deactivation upon hyperpolarization (Ref. 3 and Fig. 2). When CLC-1 is truncated behind the conserved domain D13, as in the myotonic mutant R894X, currents expressed in Xenopus oocytes are reduced but qualitatively unchanged (22). By contrast, truncating the protein before the conserved region D13 (constructs ⌬C 597 and ⌬C 720 , Fig. 1) abolishes currents ( Fig. 2 for ⌬C 720 ). The complementary NH 2 -terminal truncations ⌬N 598 and ⌬N 721 did not give any currents either when injected alone (data not shown). Surprisingly, coexpression of the carboxylterminal part with the complementary membrane-anchored part of CLC-1 reconstituted robust WT currents for the combination ⌬C 720 plus ⌬N 721 but not with ⌬C 597 plus ⌬N 598 . In another combination (⌬C 720 plus ⌬N 598 ), we duplicated amino acids 598 -720. This resulted in typical CLC-1 currents, although currents were significantly lower than with the complementary pair ⌬C 720 plus ⌬N 721 (Fig. 2).
CLC-1 Can Be Split at Different Positions in the Membraneembedded Region-We next asked whether similar comple-FIG. 2. Currents expressed from WT CLC-1, truncated channel proteins, and channels split in the carboxyl-terminal cytosolic portion. Top, voltage protocol. Chloride currents were recorded in response to 150-ms voltage steps ranging from ϩ80 to Ϫ160 mV after a 25-ms prepulse to ϩ80 mV from the resting potential of the individual oocyte. This prepulse opens the gate of CLC-1, which closes upon hyperpolarization. These test pulses were followed by a constant step to Ϫ60 mV and a return to the resting potential. Note the differing scales for individual recordings, as indicated by the angles. Capacitive transients were cut off for clarity in figures. For diagrams of constructs and nomenclature of truncated channels see Fig. 1 and "Experimental Procedures." WT, currents of the full-length channel protein. ⌬N109, deletion of the cytoplasmic NH 2 terminus. ⌬C 720 , truncation of the COOH terminus at position 720. Bottom, different combinations of amino-and carboxyl-terminal portions. All current traces are representative examples from at least two batches of oocytes (n Ն 6). mentations might be possible if we "cut" the channel protein between TMDs and designed further constructs (Fig. 1). Several loops connecting CLC-1 transmembrane domains are short and highly conserved among CLC-0, CLC-1, and CLC-2 (especially D2-D3, D5-D6, and D6-D7), suggesting that they may be important for function. This has been confirmed by missense mutations found in human myotonia (23) and point mutations introduced in structure-function studies (9). Therefore, we focused on the large, poorly conserved extracellular loop between D8 and D9 and on the shorter, but also poorly conserved, D7-D8 and D10-D11 stretches. Further, since sequences flanking TMDs may be important for their correct membrane insertion, we decided to use slightly overlapping constructs. To ensure the extracellular position of the amino terminus of construct ⌬N 413 , a signal peptide (SP) from the rat acetylcholine receptor ␣-subunit was fused in front of it.
Expression in Xenopus oocytes of single truncated polypeptides (⌬C 390 , ⌬C 451 , ⌬C 520 , ⌬N 369 , ⌬N 413 , ⌬N 413 ϩ SP , ⌬N 509 ) did not yield currents different from negative controls (see Fig. 3 for ⌬N 369 ). When we coexpressed ⌬C 390 plus ⌬N 369 and ⌬C 451 plus ⌬N 413 ϩ SP , respectively, we observed currents with WT characteristics. The amplitude of ⌬C 451 plus ⌬N 413 ϩ SP was 20 -30% of WT currents ( Fig. 2 and Table I), and only 4% were obtained when coexpressing ⌬C 390 and ⌬N 369 (Fig. 3 and Table  I). The apparently increased outward current at ϩ80 mV for the latter combination is probably due to endogenous oocyte channels (compare H 2 O-injected control in Fig. 2) and was still present when CLC-1 currents were specifically blocked with 9-anthracene-carboxylic acid (data not shown). When the boundary was placed in the small hydrophilic loop between hydrophobic domains D10 and D11, we could not detect currents upon coexpression. Obviously, partial CLC-1 proteins can insert correctly into the membrane and form channels by associating with the corresponding counterpart. Voltage-dependent gating, rectification, and ion selectivity are all conserved (ion selectivity data not shown). Thus, they do not need an uninterrupted peptide backbone.
Several point mutations characteristically change CLC-1 currents. We wanted to check whether these effects are retained when expressed in "split" CLC-1 proteins. Mutating glycine 188 in D2 to alanine shifts the voltage dependence. Currents activate at positive potentials and exhibit steady state outward rectification (Fig. 4). 2 This phenotype is retained when the mutation is introduced into ⌬C 451 and coexpressed with the carboxyl-terminal half ⌬N 413 ϩ SP . The D136G mutation (in D1) produces strongly inward rectifying currents that slowly activate at voltages more negative than Ϫ60 mV (18). In this case, however, no currents were detectable when we coexpressed ⌬C 451 D136G with ⌬N 413 ϩ SP (Fig. 4). Replacing lysine 585 at the end of D12 with glutamate leads to channels deac-tivating more slowly than WT and displaying larger outward currents (24). When inserted into ⌬N 413 ϩ SP and coexpressed with ⌬C 451 , the same properties were found (Fig. 4).
D8 Is Essential for Channel Function, but an Additional D8 Is Tolerated-We also co-injected some of the "half-channels" in different combinations. As shown in Fig. 1, this produces channels that either lack domain D8 (⌬C 390 ϩ ⌬N 413 ϩ SP ) or contain it twice (⌬C 451 ϩ ⌬N 369 ). While no currents were obtained when D8 was lacking (data not shown), a duplication of this TMD was tolerated. As can be seen in Fig. 5B for ⌬C 451 ϩ ⌬N 369 , currents were lower but did not differ qualitatively from WT.
Membrane Domains D9 -D12 Can Insert Correctly without a Signal Peptide-When expressing ⌬C 451 together with ⌬N 413 , which lacks the signal peptide added in construct ⌬N 413 ϩ SP , one may expect nonfunctional proteins because D9 may have inserted with an inverted orientation. However, we again observed WT currents (Fig. 5C). Their size was reduced as compared with ⌬N 413 ϩ SP (Fig. 3C). Thus, a significant proportion of ⌬N 413 proteins have inserted correctly even without an additional signal peptide.
Carboxyl-terminal "half-channels" Cannot Displace the Equivalent Part in the Full-length Protein-As shown above, an additional transmembrane domain D8 was tolerated in split channels. This suggests that either the one in the NH 2 -or the COOH-terminal fragment is displaced from its normal position. We wondered whether a carboxyl-terminal half-channel can replace its equivalent from the full-length channel after coexpression, which would imply that the channel structure is flexible enough to allow such a replacement after the assembly of the complete channel from the full-length protein. We expressed the CLC-1 mutant R496S (in the D9/D10 block) together with constructs ⌬N 369 or ⌬N 413 ϩ SP . This mutation was identified in patients with recessive myotonia. It abolishes chloride currents in the physiological voltage range (25). If this part of the protein can be displaced by the nonmutated equivalent of the carboxyl-terminal "half-channel," one should be able to detect CLC-1 currents. However, coexpression of fulllength CLC-1 R496S with constructs ⌬N 369 or ⌬N 413 ϩ SP in molar ratios of 1:2 to 1:6 did not yield currents. Coexpression of WT ⌬C 520 with ⌬N 413 ϩ SP , which duplicates the D9-D10 stretch, did not yield functional channels either (data not shown).
Functional CLC-1 Channels Can Be Expressed from Three Separate Polypeptides-We could obtain functional channels when "splitting" CLC-1 between TMDs D8 and D9 or after residue 720 in the cytoplasmic carboxyl terminus. This suggests that also the stretch extending from D9 to the end of CBS1 may fold correctly to an independent "module." We therefore co-injected cRNA for the three constructs ⌬C 451 , ⌬N 413 ϩ SP ⌬C 720 , and ⌬N 721 (see Fig. 1, bottom), and again observed typical CLC-1 currents (Fig. 5D). Current amplitudes 2 M. Pusch and T. J. Jentsch, unpublished observation.

TABLE I
Comparative summary of currents Slope conductance at Ϫ20 mV for wild type CLC-1, truncation constructs, and coexpressed mutants was calculated. Results of n Ն 6 measurements are given in microsiemens (S) as mean Ϯ S.E. Data of nonfunctional truncated constructs other than ⌬C 720 and ⌬N 721 are not listed, but they gave similar conductances. Leak currents were not subtracted, but water-injected controls are included. a Despite the low absolute values, these currents were clearly recognized as typically CLC-1 by their characteristic kinetics (see Fig. 5).

Functional Coexpression of CLC-1 Channel Fragments
were lower than for the pair ⌬C 451 ϩ ⌬N 413 ϩ SP . This is probably due to decreasing efficiency of assembly with an increasing number of fragments.

DISCUSSION
Reconstitution of membrane proteins from artificial fragments has been demonstrated in prokaryotic (26,27) as well as eukaryotic systems. For example, the human ␤ 2 -adrenergic receptor could be functionally expressed in oocytes from two separate polypeptides (28). Similar results were obtained for M2/M3 muscarinic acetylcholine receptors (29) and the anion exchanger AE1 (30). With CFTR, surprising results by Sheppard et al. (31) show that its NH 2 -terminal half can dimerize with itself and form Cl Ϫ channels without the COOH-terminal half. Similar truncated CFTR proteins have been described in kidney (32). This probably reflects the symmetry inherent to ABC transporters, although, with the exception of the nucleotide binding folds, the first and the second part of the protein are not highly homologous. In contrast, and not surprising given its structure, none of the CLC-1 proteins lacking TMDs could form functional channels when expressed by itself.
In many transmembrane proteins, internal topogenic sequences ensure a correct integration even if the normal translation start is absent. The folding of ␣-helical membrane proteins is proposed to be a two-stage process (33). First, hydrophobic domains are established in the lipid bilayer, which then interact to form the final three-dimensional structure. Thus, there seems to be no principal difference between interand intramolecular assembly of transmembrane domains, apart from the facilitated encounter of helices in a single polypeptide chain.
Expression of truncated proteins lacking one or more transmembrane domains may, however, result in misfolding if a certain succession of the amino acid sequence is required to achieve the correct insertion of the nascent polypeptide chain. Moreover, newly synthesized TMDs may need previously translated TMDs for "guided" folding and assembly. The successful assembly of functional proteins from individual parts suggests that these form structurally independent subdomains or "modules" whose tertiary structure does not differ too much from the one found within the native protein. Further, the intermolecular interactions between these different modules is strong enough to allow their association in the absence of a covalent link. Examples of modular structures of membrane channels and transporters are provided by some gene superfamilies: voltage-gated sodium and calcium channels have four modules covalently linked in a single polypeptide, whereas they are encoded separately in K channels; ABC transporters FIG. 4. Effects of point mutations on CLC-1 currents mediated by full-length or coexpressed truncated channel proteins. Top, recordings from full-length proteins. Bottom, coexpression of CLC-1 halves ⌬C 451 ϩ ⌬N 413 ϩ SP with equivalent mutations. Typical current traces are shown, which were recorded as described in Fig. 2 except for mutant G188A, where the following pulse protocol was used. The oocyte membrane potential was stepped from the resting potential to voltages ranging from ϩ20 to Ϫ160 mV (for 4 s in steps of Ϫ20 mV), followed by a constant pulse at Ϫ85 mV. No functional complementation is observed with split channels containing the D136G mutant. Current traces are representative examples from at least two batches of oocytes (n Ն 6).  Fig. 1 and current traces in Fig. 3). D, coexpression of three cRNAs coding for amino acids 1-451, signal peptide ϩ413-720, and 721 to the COOH terminus (compare Fig. 1). Chloride currents were recorded as described in Fig. 2. Current traces are representative examples from at least two batches of oocytes (n Ն 6).
can be encoded by one, two, or four genes, whose individual products will assemble and form the complete transporter (34).
In the CLC-1 chloride channel, at least three regions are structurally autonomous: an amino-terminal one comprising D1-D8, a central one containing the D9 -D12 block and part of the cytoplasmic carboxyl terminus, and the rest of the hydrophilic tail, which includes D13. A cut between D10 and D11, however, was not tolerated. This suggests that either the NH 2or the COOH-terminal portion is misfolded (or both) or that interactions between these different parts are too weak to allow for their association even if folded correctly. The D9 -D12 region is a broad hydrophobic region that is important for permeation and gating (17,23). In the absence of clearly separated hydrophobic domains, it seems possible that it folds correctly only if translated as a continuous polypeptide. This view is strengthened by the fact that no cleavable signal peptide was necessary to correctly express fragment ⌬N 413 . Its intrinsic topogenic activity cannot be easily explained by the "positiveinside" rule (35)(36)(37), since there is no conspicuous charge asymmetry in this region. Exact topology is still unknown here except for the fact that the region before D9 is extracellular and the region after D12 is intracellular (8).
The CLC-1 protein could also be split between D7 and D8, but expression was less efficient than with the D8-D9 cut. The D7-D8 linker is shorter than the one connecting D8 and D9 and rather poorly conserved. However, it is always highly positively charged, and mutagenesis showed that it is important for CLC-2 gating (38). D8 is probably located at the channel periphery, since split channels functionally tolerated an extra D8 copy. Thus, the channel structure must allow for an extrusion of D8 domain into the lipid bilayer.
Channels deleted for D13 could be functionally complemented by coexpressing the lacking hydrophilic COOH-terminal portion. Similar results were recently reported for the Torpedo channel CLC-0 (39). D13 was originally identified as a region of intermediate hydrophobicity (1) and is now known to be cytoplasmic (8,40). It is conserved in all eukaryotic CLC proteins, suggesting some important, as yet unknown function. By contrast, it is absent from the two E. coli CLC proteins. Mutations in human disease already hinted at the importance of D13: point mutations truncating CLC-5 before or within D13 lead to Dent's disease, and these truncated proteins were nonfunctional in Xenopus oocytes (5,41). A myotonic mutation (R894X) that truncates CLC-1 shortly after D13 leads to reduced but qualitatively unchanged currents (22).
Recently, Bateman (19) and Ponting (20) identified a more general structural motif by data base screening to which the D13 domain conforms. Based on the crystal structure of inosine-5Ј-monophosphate-dehydrogenase, this so-called CBS (cystathionine ␤-synthase) domain probably consists of two short ␣-helical and three ␤-strand stretches in the order ␤␣␤␤␣. It is found in various other proteins like a certain protein kinase subunit and ABC transporters, but its functional role is unknown. Interestingly, the CBS domain is found twice in every eukaryotic CLC protein. The second CBS domain (CBS2) roughly coincides with D13, while the other one (CBS1) lies between D12 and D13. Strikingly, only constructs ⌬C 720 ϩ ⌬N 721 , which contain CBS1 and CBS2, respectively, could associate to form functional channels. No currents could be seen when ⌬C 597 was coexpressed with ⌬N 598 , which contains both CBS domains. The successful complementation of ⌬C 720 with ⌬N 721 suggests that CBS2 forms a correctly folded module, which can bind somewhere to the protein truncated behind CBS1. It is presently unclear whether this is by direct interaction between CBS1 and CBS2. When we coexpressed the overlapping constructs ⌬C 720 and ⌬N 598 (which contains CBS1 and CBS2), we again observed functional complementation, which, however, was less efficient. Thus, if CBS1-CBS2 binding is necessary, their intramolecular interaction in construct ⌬N 598 is weak enough to allow intermolecular competition with CBS1 from the channel backbone. On the other hand, coexpressing a nonconducting full-length channel, CLC-1 R496S (25), together with ⌬C 720 (lacking CBS2) did not yield currents. This could mean that CBS2 interacts too strongly with the rest of the CLC-1 R496S channel protein and therefore does not swing over to a nearby ⌬C 720 -protein in a heteromultimer and does not complement this truncated channel. A displacement of larger intramembranous parts of the channel also failed when coexpressing CLC-1 R496S (D1-D13) with constructs ⌬N 369 (D8 -D13) or ⌬N 413 ϩ SP (D9 -D13) or coexpressing ⌬C 520 (D1-D10) with ⌬N 413 ϩ SP (D9 -D13).
Channel properties like gating kinetics and rectification were preserved in functional split channels, and two of the point mutations introduced into partial channels retained their characteristic effects. However, mutation D136G prevented functional reconstitution from "half-channels." Possibly, neutralization of the negative charge in D1 at position 136 affects the contact surfaces between the two parts, thereby disturbing their association. In the full-length protein, tight spatial proximity is ensured by covalent links, leading to a functional pore whose gating, however, is drastically changed.
In summary, we could show that functional CLC-1 chloride channels can be formed from fragments, indicating the presence of at least three modular domains that can insert and fold independently and then associate. This includes the cytoplasmic domain D13, which conforms to the newly identified CBS motif. The subunits formed by the association of these partial proteins are then likely to further associate to form a homodimeric channel (11), as unambiguously shown for the CLC-0 chloride channel (9,10).