Structure of aquaporin-2 vasopressin water channel.

Aquaporin-2 (AQP-2) is a vasopressin-regulated water channel in the kidney collecting duct. AQP-2 is selectively permeable to water molecule and is translocated between the apical membrane and subapical endosomes in response to vasopressin. To investigate the localization and structure of the aqueous pathway of the AQP-2 water channel, a series of site-directed mutants was constructed and functionally analyzed. Insertion of N-glycosylation reporter sequence into each hydrophilic loop (HL) indicated that AQP-2 has a six-membrane spanning topology and that insertional mutations in HL-2 or HL-5 do not alter water channel function. Mercury-sensitive site of AQP-2 is located near the second asparagine-proline-alanine (NPA) domain at cysteine 181, but not near the first NPA domain. Replacement of HL-3 or HL-4 with the corresponding part of Escherichia coli glycerol facilitator abolished water channel function without changing plasma membrane expression of the channel protein. Introduction of cysteine residues in His-122, Asn-123, Gly-154, Asp-155, or Asn-156 induced partial mercury sensitivity, and point mutations in asparagine 123 significantly altered water permeability. Our results implicate that the structure of AQP-2 is different from models previously proposed for AQP-1 and that HL-3 and HL-4 are closely located to the aqueous pathway.

Aquaporin-2 (AQP-2) is a vasopressin-regulated water channel in the kidney collecting duct. AQP-2 is selectively permeable to water molecule and is translocated between the apical membrane and subapical endosomes in response to vasopressin. To investigate the localization and structure of the aqueous pathway of the AQP-2 water channel, a series of site-directed mutants was constructed and functionally analyzed. Insertion of N-glycosylation reporter sequence into each hydrophilic loop (HL) indicated that AQP-2 has a six-membrane spanning topology and that insertional mutations in HL-2 or HL-5 do not alter water channel function. Mercury-sensitive site of AQP-2 is located near the second asparagineproline-alanine (NPA) domain at cysteine 181, but not near the first NPA domain. Replacement of HL-3 or HL-4 with the corresponding part of Escherichia coli glycerol facilitator abolished water channel function without changing plasma membrane expression of the channel protein. Introduction of cysteine residues in His-122, Asn-123, Gly-154, Asp-155, or Asn-156 induced partial mercury sensitivity, and point mutations in asparagine 123 significantly altered water permeability. Our results implicate that the structure of AQP-2 is different from models previously proposed for AQP-1 and that HL-3 and HL-4 are closely located to the aqueous pathway.
Aquaporin-2 (AQP-2, 1 previously reported as WCH-CD or AQP-CD) is a water channel in the apical membrane of the kidney collecting duct (1). Water permeability of this nephron segment is regulated by vasopressin through the membrane shuttle mechanism (2)(3)(4), in a mechanism by which AQP-2 is translocated between the apical membrane and endosomes under vasopressin regulation (5)(6)(7). Complementary DNAs for rat and human AQP-2 have been isolated (8,9), and the primary structure of AQP-2 has been identified. AQP-2 is a very hydrophobic membrane-integral protein of a molecular mass of ϳ29 kDa. It is a member of the MIP protein family (10) and is homologous to aquaporin-1 (AQP-1, previously reported as CHIP28) (11)(12)(13). Functional expression of AQP-2 showed that it is highly permeable to water molecule but not to urea, glycerol, and ions and that its water permeability is mercurysensitive and temperature-insensitive (8). Furthermore, it has been shown that mutations in AQP-2 gene are responsible for deficient vasopressin antidiuresis in some patients with neph-rogenic diabetes insipidus (14,15).
Despite accumulated knowledge of AQP-2 physiology implicating functional importance of AQP-2 in urine concentration and homeostasis of body fluid, the molecular structural basis of AQP-2 is not well known. The localization and higher order structure of the aqueous pathway of AQP-2 have to be elucidated to account for its selective permeability to water molecule and explain the mutation-related channel malfunction. The molecular structure of AQP-1, the first identified water channel, has been studied and partially resolved. It was shown that AQP-1 exists in plasma membrane with tetramer formation (16,17) but that each monomer is functionally independent, thus leading to the assumption that single aqueous pore spans each monomer (18,19). Regarding the structure of the aqueous pore in functionally active AQP-1 monomer, three structural models have been proposed (20): the hourglass model (21), the ␣-helical model (22), and the ␤-barrel model (23). According to the hourglass model, the aqueous pathway is formed by two domains with an NPA box, which is an asparagine-prolinealanine sequence highly conserved among the MIP family members. In the ␣-helical model, the aqueous pathway is located between transmembrane segments with ␣-helical conformation. In the ␤-barrel model, the aqueous pore is formed with ϳ16 antiparallel ␤-sheets analogous to a porin channel of bacteria species (24). The validity of these models has not been sufficiently examined. Moreover, there have been no investigations regarding the molecular structure of aquaporins other than AQP-1. In addition to being significant for understanding the general structure of channels, elucidation of the structure of aquaporins will provide insights for the development of an AQP-2 inhibitor, which is potentially of remarkable use as a water diuretic.
This study examined membrane topology, mercury-sensitive sites, and functional roles of the two of the NPA-containing domains and other hydrophilic loops of AQP-2. We found significant participation of hydrophilic loops other than NPAcontaining domains in the formation of the aqueous pathway of AQP-2. Based on our observations, a new structural model for AQP-2 is proposed.

EXPERIMENTAL PROCEDURES
Site-directed Mutagenesis and in Vitro cRNA Synthesis-AQP-2 mutants were constructed using polymerase chain reaction-based sitedirected mutagenesis. A fragment between SphI site at nucleotide 219 and StuI site at nucleotide 809 in pAQP-2/ev1 (8) was replaced by a polymerase chain reaction fragment coding for mutants. Potential glycosylation sites were produced by inserting an N-glycosylation reporter sequence (asparagine-threonine-serine, NTS) (25) into one of four hydrophilic loops of non-glycosylated mutant of AQP-2, N124D (Fig. 1). For hydrophilic loops (HL) 1, 2, 4, and 5, an NTS motif was inserted after serine 36 (producing 36NTS), alanine 65 (65NTS), glycine 154 (154NTS), and valine 194 (194NTS) of N124D, respectively. To directly compare the structure of AQP-2 and AQP-1, the insertion positions were chosen corresponding to the BamHI sequence insertion positions in the previous study (26). For determination of mercury sensitivity, some residues of mercury-resistant mutant of AQP-2, C181A, were substituted with cysteine. For recombination of HL-3 or HL-4, a cluster of amino acids in HL-3 of AQP-2, ALHNNATA, was replaced with YPNPHINF in the corresponding part of E. coli glycerol facilitator (GlpF), and ERRGDNLGSP in HL-4 of AQP-2 was replaced with DGNGVPRGP of GlpF, with RRRRDLGGSA of AQP-1, or with ERRN-GRLGSV of MIP. The nucleotide sequences of mutants were verified by a fluorescence sequencer (model 373A, Applied Biosystems). Mutated plasmid was purified by affinity columns (Qiagen, Germany) and linearized with NotI or XbaI. Capped cRNA was synthesized using T3 RNA polymerase and in vitro transcription system (Promega). Standard molecular biological procedures were used (27).
Preparation of Oocytes and Measurement of P f -Defolliculated stage V and VI oocytes from female Xenopus laevis were injected with 20 nl of water or cRNAs (0.1 mg/ml). After incubation in 200 mosm modified Barth's buffer at 18°C for 48 h, oocytes were transferred to 70 mosm Barth's buffer diluted with distilled water, and the time course of osmotic volume increase was monitored as described (8). Oocyte volume was calculated from oocyte projection area. Osmotic water permeability (P f ) was determined from initial oocyte volume (V 0 ϭ 9 ϫ 10 Ϫ4 cm 3 ), initial oocyte surface area (S ϭ 0.045 cm 2 ), molar volume of water (V w ϭ 18 cm 3 /mol), and osmolarity inside (osm in ) and outside (osm out ) the cell as shown: The effects of mercury reagents were examined by incubating oocytes in Barth's buffer containing 1 mM HgCl 2 for 10 min prior to P f measurements.
Rat Kidney and Oocyte Membrane Isolation and Glycosidase Digestion-Kidney medulla from 200-g Sprague-Dawley rats were dissected, homogenized in 9 volumes of homogenization medium (0.32 M sucrose, 5 mM Tris-HCl (pH 7.5), 2 mM EDTA, 20 g/ml phenylmethylsulfonyl fluoride) with 15 strokes of a Teflon homogenizer, and centrifuged at 3,000 ϫ g for 10 min to remove organelles. The supernatant was centrifuged at 100,000 ϫ g for 30 min to collect the membrane fractions. The pellet was suspended in 100 l of suspension buffer (5 mM Tris-HCl (pH 7.5), 2 mM EDTA). Total membrane fractions of oocytes were collected as follows (18): groups of 10 -15 oocytes prepared as for P measurements were homogenized in 9 volumes of homogenization medium. The yolk and cellular organelles were removed by centrifuging at 750 ϫ g for 5 min at 4°C. The membrane fractions were then pelleted from the supernatant at 16,000 ϫ g for 30 min at 4°C and suspended in 10 l of 1.25% SDS/oocyte. Plasma membrane fractions of oocytes were collected as described (28). Briefly, 20 -40 oocytes were homogenized, and organelles were removed by centrifugation at 350 ϫ g for 5 min. Supernatant was centrifuged at 15,000 ϫ g for 30 min, and membranes were suspended with ϳ100 l of 1.22 g/ml sucrose in phosphate-buffered saline. Membranes were overlaid onto layers of 1.22, 1.18, and 1.16 g/ml sucrose in phosphate-buffered saline solution and centrifuged at 85,000 ϫ g for 90 min. Plasma membrane fractions in mid-layer were collected, diluted seven times with phosphate-buffered saline, and pelleted by centrifugation at 15,000 ϫ g for 1 h. Plasma membrane fractions were enriched Ͼ10 times as assessed by alkaline phosphodiesterase activity (29). To remove the N-glycosylation side chains, 10 g of kidney membranes or membrane fractions from one oocyte were digested with 2 units of N-glycosidase F (Boehringer Mannheim) in 10 l of incubation buffer (0.5% (w/v) SDS, 50 mM B-mercaptoethanol, 1.3% Nonidet P-40, 10 mM phenanthroline) for 12 h at 37°C and analyzed by immunoblotting.
Immunoblotting of AQP-CD and Mutants-SDS-polyacrylamide gel electrophoresis (PAGE) was performed as described (8). Total membrane from 1 oocyte and plasma membrane from 20 oocytes were solubilized by heating at 70°C for 10 min in sample buffer (3% SDS, 60 mM Tris-HCl (pH 6.8), 5% ␤-mercaptoethanol, and 10% (v/v) glycerol). Protein was separated by electrophoresis through 10% SDS-PAGE slab gels and transferred to Immobilon membranes (Millipore). The membranes were blocked in blotting buffer (0.15 M NaCl, 0.05% Triton X-100, and 20 mM Tris-HCl (pH 7.4)) containing 2% nonfat dry milk and then incubated with either preimmune or immune serum (diluted 1:200 in blotting buffer) at room temperature for 1 h. The membranes were then washed in several changes of blotting buffer, incubated for 1 h with blotting buffer containing 3 Ci of 125 I-labeled protein A (ICN Biochemicals, Costa Mesa, CA), washed three times, and exposed to x-ray film for 12 h.
Quantitative Evaluation of Protein Expressed on the Oocyte Membrane-The AQP-2 protein was evaluated in intact oocyte membrane as reported (30). An anti-AQP-2 antibody against the external domain of AQP-2 was raised in rabbit using a synthetic peptide corresponding to amino acid residues at 113-127 as described (8). Oocytes, prepared as for P f measurements, were blocked in 2% bovine serum albumin in modified Barth's buffer for 1 h at 4°C and then washed with Barth's buffer. All subsequent steps were performed on ice. A group of 10 oocytes was incubated with 1:100 diluted antibody against an external domain in Barth's buffer for 1 h, washed with a large volume of washing buffer (0.2% bovine serum albumin in modified Barth's buffer) three times, incubated with 1 Ci/ml 125 I-protein A in Barth's buffer for 1 h, and washed several times in a large volume of washing buffer. The 10 oocytes were dissolved in 10% SDS, and the radioactivity was measured by a gamma counter.

AQP-2 Membrane Topology Determined from N-Glycosylation
Site-Immunoblot analysis of rat kidney membrane and oocyte expressing AQP-2 showed that AQP-2 consists of ϳ29 kDa and higher molecular mass components, which were removed by N-glycoside F digestion (Fig. 2). N-Glycosylation was not observed when asparagine 124 (Asn-124) was replaced with aspartic acid, showing that AQP-2 is N-glycosylated at Asn-124. Injection with N124D induced high osmotic water permeability that was comparable to that of wild type, suggesting that N-glycosylation is not required for water channel function of AQP-2 (Fig. 3). The sidedness of each hydrophilic loop was determined by the insertion of asparagine-threonine-serine (NTS) consensus motif and assessing the accessibility of Nglycosyltransferase. As shown in Fig. 2, AQP-2 mutants were glycosylated only when the NTS signal was inserted into the first or the fifth loop and not when NTS was inserted into the second or the fourth loop (Fig. 2, lanes 3-6). Molecular sizes of glycosylation in these mutants were the same as those in wild type of AQP-2. In both of these N-glycosylated mutants, the glycosylated Ϫ29 kDa protein was removed by digestion of N-glycosidase F (Fig. 2, lanes 7 and 8). These results indicated that the first, third, and fifth hydrophilic loops are localized extracellularly.
Role of Hydrophilic Loops 2 and 5 in the Formation of AQP-2 Aqueous Pathway-The effects of three amino acid insertions on the channel function were examined by looking at osmotic water permeability of oocytes expressing NTS-inserted mutants (Fig. 3). Insertions of a few amino acid residues into HL-2 and HL-5 had little effect on the function of AQP-2, in contrast to the fact that amino acid insertion into the corresponding position of AQP-1 completely abolished water channel function (26). Osmotic water permeability induced by NTS-inserted mutants was slightly lower than that induced by non-glycosylated mutant of AQP-2, N124D. Relative plasma membrane expressions assessed by radioimmunoassay were 91 Ϯ 3% (36NTS), 81 Ϯ 4% (65NTS), 94 Ϯ 4% (154NTS), and 90 Ϯ 2% (194NTS) of N124D, showing that the plasma expression of these mutants is slightly decreased from N124D. Taken together, osmotic water permeability of all glycosylation mutants was slightly lower than N124D, partly due to decreased plasma membrane expression. Since it was significantly higher than that of water-injected control, however, and since mercuric inhibition was apparent, it can be assumed that all of the NTS-inserted mutants preserve water channel function.
As the mercury-sensitive site is expected to be localized close to the aqueous pore (21), inhibition by mercury agent was examined for a series of cysteine mutants. When cysteine at 181 was replaced, mercury inhibition disappeared in contrast to other substitutions of C75S, C79S, and C144S; replacements of cysteine 181 with larger residues inhibited water permeability, suggesting cysteine 181 is localized near the aqueous pathway (Fig. 4). Substitution by cysteine of alanine at 65, an alternative mercury-sensitive site proposed by the hourglass model, in mercury-resistant mutant C181A did not induce mercury sensitivity. Localization of mercury-sensitive site in the first NPA domain, something which strongly supports the hourglass model for AQP-1, was not observed in AQP-2, implicating the difference in the structures of the aqueous pathways of AQP-1 and AQP-2. Observations from NTS insertion and cysteine substitution taken together indicated that cooperative participations of HL-2 and HL-5 in the formation of the aqueous pathway of AQP-2 is not as critical as proposed in the hourglass model for AQP-1.

Roles of HL-3 and HL-4 in the Formation of AQP-2 Aqueous
Pathway-Participation of HL-3 and HL-4 in the formation of the aqueous pathway was revealed by the replacement of these domains with the corresponding part of E. coli GlpF (Fig. 4). GlpF was chosen because its function has been examined in detail and because it is known to be permeable to glycerol but not to water (31). Recombinant HL-3 or HL-4 to GlpF did not induce osmotic water permeability (Fig. 4). In contrast, replacement of HL-4 with other water channels, AQP-1 and MIP, induced osmotic water permeability to a level comparable to that of the wild type. To exclude the possibility of the decrease in readily assembled protein and plasma membrane expression, localization of AQP-2 in total membrane fractions or in plasma membrane fractions was examined by immunoblots. Oocyte membrane was fractionated through a sucrose gradient by ultracentrifugation, and plasma membrane fraction was collected. Immunoblot of fractionated membranes from oocytes injected with GlpF recombinants cRNA indicated that the amounts of synthesized mutants and mutants expressed in plasma membrane were comparable to those of the wild type (Fig. 5). Independent analysis of the amount of the plasma membrane expression was performed by radioimmunoassay. Binding of antibody against external domain of intact oocytes expressing mutants was assayed. Specificity and sensitivity of the antibody were confirmed by immunoblotting (data not shown). Relative expression to the wild type was 65 Ϯ 10% for HL-3 mutant and 68 Ϯ 12% for HL-4 mutants, indicating substantial expression of immunoreactive protein on oocyte plasma membrane. Intact glycosylation of recombinant HL-4 suggested natural maturation of the mutants. When HL-3 was replaced with GlpF, glycosylation was not observed because of removal of the natural glycosylation site of AQP-2.
A series of single-residue substitutions in HL-3 and HL-4 was performed to investigate the interaction of these loops with the aqueous pathway. Replacement of native residues to cysteine has successfully been used to determine the localization of the aqueous pathway of aquaporins (21). The rationale is that there is a high probability that cysteine residues near the aqueous pathway interact with imposed mercury agents and inhibit water channel function. When a cysteine residue was introduced into amino acid 122 or 123 of mercury-resistant mutant of AQP-2, C181A, up to 40% inhibition by low concen- tration of HgCl 2 was observed, which was significantly higher than that for C181A (p Ͻ 0.001, n ϭ 20). The mercury inhibition, although it is not complete, implicated that Asn-123 is likely to be located near the aqueous pathway. Subsequently, a residue Asn-123 was replaced with a series of amino acids to examine the interaction of the lateral moiety with the aqueous pathway. Osmotic water permeability was decreased roughly in accordance with the size of the lateral moiety (Fig. 4). This may be interpreted as the localization of the constriction of the aqueous pathway at Asn-123 and the occlusion by the larger lateral moiety. Plasma membrane expressions of the series of Asn-123 mutants were comparable to those of the wild type as assessed by immunoblot (Fig. 6) and radioimmunoassay (60 Ϯ 10% for N123A, 65 Ϯ 12% for N123Q, 65 Ϯ 12% for N123D, 65 Ϯ 12% for N123W, relative to the wild type, n ϭ 3-5), showing substantial expression of N123 mutants in the plasma membrane. A similar analysis was done for HL-4. Cysteines in 154, 155, and 156 exhibited high mercury sensitivity, with the highest at Asp-155 (p Ͻ 0.001 versus C181A, n ϭ 20), implicating close localization of Asp-155 to presumed aqueous pathway. DISCUSSION In this study, we have examined the structure of the aqueous pathway of rat AQP-2 and found that the contribution of HL-3 and HL-4 is significant in the formation of the aqueous pathway. It was suggested that the structure of the aqueous pore is somehow different from that of AQP-1 proposed previously (21)(22)(23). A six-transmembrane topology and mercury-sensitive positions in AQP-2 are identical to AQP-1 hourglass model (21). However, mutations near NPA boxes in HL-2 and HL-5 did not significantly alter water channel function, and cysteine residues inserted near the first NPA box did not induce the alternative mercury-sensitive site. Alternatively, based on the findings that the replacement of HL-3 and HL-4 with hydrophobic residues decreased water permeability without affecting plasma membrane expression and that mercury sensitivity was found in these domains, it was suggested that HL-3 and HL-4 are located near the aqueous pore.
Membrane topology and mercury-sensitive cysteine residues have first been determined experimentally in this study, and it was shown that AQP-2 is similar to AQP-1 proposed as the hourglass model in its topological presentation. Membrane topology was partially determined from the presence of N-glycosylation, which indicates extracellular localization of the site. N-Glycosylation site insertion has been successfully used to map membrane topology of the cystic fibrosis transmembrane regulator (25) and the glutamate receptor (32). The N-glycosylation insertion was used in the present study because the glycosylation is natural and the consensus glycosylation sequence minimally perturbs the native sequence and structure (25). Immunoblot results indicated the extracellular localization of HL-1, HL-3, and HL-5, implicating that the ␣-helical model (22) and the ␤-barrel model (23) are not adequate for the structure of AQP-2. Sidedness of both termini, HL-2 and HL-4, cannot be determined because the extracellular NTS motif is not always glycosylated. However, when the evidence of phosphorylation of the carboxyl terminus of AQP-2 is also taken into account (30), it may be reasonable to conclude that AQP-2 has a six-transmembrane topology identical to that of AQP-1 proposed in the hourglass model (21,26).
Inhibition of osmotic water permeability by low concentration of mercury chemicals is characteristic of protein water channel, and mercury is believed to interact with cysteine residues located near the aqueous pore, perturbing the structure and function of the aqueous pore (21). Therefore, identification of the mercury-sensitive cysteine residue is critical for locating the aqueous pore. The results showed that the mercury-sensitive cysteine of AQP-2 is Cys-181, which corresponds to Cys-189 of AQP-1. In addition, molecular size-dependent im-pairment of water permeability was observed in Cys-181 replacement, which was similar to observations in AQP-1. When Cys-181 was replaced with tryptophan, P f had the lowest value. Because the plasma membrane expression of C181 mutants was similar as assessed by antibody binding and membrane fractionation (data not shown), it is likely, as proposed in the AQP-1 experiments (18), that replacement of Cys-181 with larger residues interfered with the aqueous pathway. These data suggested that the aqeuous pore of AQP-2 may be closely located to Cys-181.
Although identical topology and the mercury-sensitive site of AQP-2 and AQP-1 reasonably implicated the structural similarity of the two channels, our observation from mutational analysis indicated that the structure of the aqueous pathway of AQP-2 is somehow different from that of AQP-1. Although the contribution of HL-2 and HL-5 to the formation of the aqueous pathway has been strongly postulated in the hourglass model for AQP-1 (21), our results were not compatible with the previous observations. First, introduction of cysteine to Ala-65, which corresponds to Ala-73 of AQP-1, did not induce mercury sensitivity. Symmetrical localization of mercury-sensitive sites in Cys-189 and Ala-73 were one of the bases for the hourglass model for AQP-1. Second, insertion of a few amino acid residues into HL-2 and HL-5 of AQP-2 did not affect water channel function. Inhibition of water permeability by the insertion of the BamHI sequence into the corresponding sites of HL-2 or HL-5 supported the hourglass model of AQP-1 (26). In the hourglass model, it was claimed that HL-2 and HL-5, being folded into the lipid bilayer, partially form the aqueous pore (21). However, a glycosylation reporter sequence inserted into HL-5 was properly glycosylated with minimal effects on water permeability, showing extracellular localization of this part. Our results may indicate that the hourglass model is not adequate for describing the aqueous pore of AQP-2.
Functional analysis of replacement, mercury sensitivity, and point mutations in HL-3 and HL-4 implicated that these loops participate in the formation of the aqueous pathway of AQP-2. HL-3 and HL-4 are relatively hydrophilic compared to other segments of AQP-2. Hydropathy analysis of other aquaporins including AQP-1, AQP-3, MIP, and ␥-TIP showed that hydrophilicity of these domains is a common feature of water channels (33,34). Thus, it is reasonable to speculate that HL-3 and HL-4 contribute to the structure of the aqueous pore and to selective water permeability. Disappearance of water permeability by the replacement of HL-4 with GlpF but not with AQP-1 or MIP implicated that hydrophilicity of this domain may be critical for water permeability. Furthermore, the findings that mercury sensitivity was maximum at positions Asn-123 and Asp-155 and that substitutions of Asn-123 with larger residues decreased P f can be interpreted that Asn-123 and Asp-155 are the closest to the presumed constriction of the aqueous pore of an expected size of ϳ2 Å (21-23). However, our observations may raise several questions, which must be resolved before any definitive conclusions can be made. First, although mercury inhibition found in HL-3 and HL-4 was significant compared to C181A, it was lower than that for wild type. The lower mercury sensitivity can be explained by the possibility that HL-3 and HL-4 are located further to the aqueous pathway than Cys-181. Second, insertion of NTS after the residue 154, which should be the critical site for water channel function according to our data, did not alter P f . This may be because hydrophilicity of inserted residues minimally disturbed the pore structure. Further structural analyses will be required to interpret our current observations and those of the previous studies and to describe a more precise structure of the aqueous pore.
Care must be taken for the interpretation of mutational analysis of aquaporins because it has been well known that mutations in aquaporins sometimes disturb synthesis, assembling, and plasma membrane expression of mutant proteins in oocytes (19,21,35). Thus, functions of mutated channel have to be normalized with the amount of protein in the plasma membrane. Therefore, we undertook immunological analysis of synthesized mutants to ensure that mutants were readily synthesized and expressed in the plasma membrane. Plasma membrane expression of mutated channel was successfully resolved by two methods. For this purpose, we did not use immunohistochemical staining of oocyte membranes, which is less quantitative compared to the two methods we used. All mutations in HL-3 and HL-4 examined here did not significantly affect plasma membrane expression, confirming that the mutations only minimally perturb the channel structure.
On the basis of our results, we propose a structure model for AQP-2 water channel (Fig. 7). Between six transmembrane ␣-helices, a central channel may be formed, the diameter of which would be larger than the expected size for selective aqueous pore. Constrictions with a pore size of ϳ2 Å that determine selective water permeability may be assembled in the association of HL-3 and HL-5 in the extracellular side and in the association of HL-2 and HL-4 in the cytoplasmic side. These hydrophilic loops may be partially folded into the large pore, forming constrictions and hydrophilic apparatus that exclude charged ions and large molecules. Since N-glycosylation of HL-3 or HL-5 did not influence pore structure, it is likely that these domains are not folded deep into lipid bilayer as described in the hourglass model. The roles of the NPA boxes, which are strictly conserved among the MIP family, are not clear from our study. It is speculated that the NPA boxes are critical to correctly assemble three-dimensional structures of aquaporins because AQP-2 proteins with mutations near NPA boxes were suggested to be folded improperly (35).