Erythroid expression and oligomeric state of the AQP3 protein.

Biochemical and biophysical studies have shown that the strictly water-permeable aquaporins have a tetrameric structure, whereas results concerning the oligomeric state of GlpF, the glycerol facilitator of Escherichia coli, are dependent upon the analytical technique used. Here, we analyzed the oligomerization of the AQP3 aquaglyceroporin, which presents a mixed selectivity for water, glycerol, and urea. At first, based on transcript detection by reverse transcription-PCR from human erythroid tissues and membrane expression detected by flow cytometry analysis, we demonstrated that AQP3 is expressed on human and rat but not on mouse red blood cells. Then, the quaternary structure of AQP3 was determined using as models human red blood cell membranes, which carry both AQP1 and AQP3, and two heterologous expression systems: Xenopus laevis oocyte, for density and size estimation of aquaporins, and Saccharomyces cerevisiae yeast, which expressed a non-glycosylated form of AQP3. By velocity sedimentation in sucrose gradient after non-denaturing detergent solubilization, AQP3 was essentially found as mono- and dimeric species in conditions under which AQP1 preserved its tetrameric structure. Freeze-fracture studies on oocyte plasma membranes gave a size of AQP3 particles in favor of a dimeric or trimeric structure. Finally, by cross-linking experiments with red blood cell membranes, AQP3 is visible as different oligomeric structures, including a tetrameric one.

Since the discovery of the first water channel, aquaporin-1 (AQP1), 1 which selectively transports water, others members of the major intrinsic protein (MIP) family have been identified. Among them, some are also permeable to small solutes such as urea and glycerol, and these were named aquaglyceroporins. In mammals, the first one to be characterized was AQP3 (1,2), and more recently, AQP7 (3) and AQP9 (4,5) were discovered. These proteins have a higher sequence identity with GlpF, the bacterial glycerol facilitator of Escherichia coli, than with strict aquaporins. The three-dimensional structure of AQP1 has been extensively studied and determined recently at high resolution (6). These analyses have confirmed the proposed hourglass model (7) in which each monomer is formed of six tilted helices spanning the membrane bilayer and surrounding a central density zone corresponding to the B and E loop folding back into the membrane to form the water pore. Structure analysis of AQP1 has also revealed its tetramerization (8 -11), first observed from freeze-fracture studies (12,13) and then by velocity sedimentation experiments after non-denaturing detergent solubilization (7,14). The latter approach has been used extensively to study the structural and functional properties of some other members of the MIP family and has confirmed the tetrameric form of strict aquaporins (15,16). The crystallization of some strict aquaporins has permitted the visualization of the tetrameric arrangement of AQP0 (17), AQPZ (18), AQPcic (19), and ␣-TIP (20). The role of subunit tetramerization is unknown, but it may provide a better stability of the protein, or it could be involved in the function, even if functional independence of AQP1 monomers has been suggested by radiation inactivation (21), directed mutagenesis, and chimeric molecule design (22)(23)(24). Although all proteins of the MIP family might have a similar quaternary structure due to their general topology, the low identity of some segments and the difference of selectivity observed between aquaporins and aquaglyceroporins suggest that they could not share the same organization. In fact, whereas the tetrameric organization has been demonstrated for most of the strict aquaporins by biochemical and biophysical approaches, only the quaternary structure of GlpF (for which the results are unclear) has been studied. By freeze-fracture structure studies and velocity sedimentation on sucrose gradients, GlpF appeared to be monomeric in Xenopus laevis oocyte (25,26). In contrast, crosslinking experiments (27) and studies of the three-dimensional crystal structure obtained at a 2.2-Å resolution (28) revealed a tetrameric organization of GlpF.
An aquaglyceroporin is not only a water-permeable channel but also a glycerol facilitator, and its oligomerization state should be resolved to elucidate the relationships between structure and selectivity properties of the MIP family members. For this purpose, we used velocity sedimentation on sucrose gradient of AQP3 solubilized from the native system, human erythrocytes, which possess both AQP1 (tetrameric) and AQP3, or from heterologous systems (Saccharomyces cerevisiae yeast and X. laevis oocyte). Indeed, in a previous work, we (29) underlined the presence of AQP3 in human red blood cell (RBC) membranes by immunological and functional studies. In this study, we confirmed these findings by identification of the transcripts in human fetal liver, bone marrow, and reticulocytes and by immunological analysis by flow cytometry with specific antibodies. In addition, freeze-fracture experiments were performed to compare the size of human AQP1 (hAQP1) and rat AQP3 (rAQP3) expressed in X. laevis oocyte membrane and to determine the degree of oligomerization of these molecules according to Eskandari et al. (30). Chemical cross-linking using disuccin-imidyl suberate (DSS) was also carried out to further analyze the oligomerization state of hAQP3 on RBCs.
PCR Amplification of the Human AQP3 Transcript-To determine whether the hAQP3 transcript is present in erythroid tissues, 0.5 ng of human fetal liver or human bone marrow Marathon-Ready cDNAs and 5 ng of single strand DNA of human reticulocytes were used in nested PCR amplifications (94°C for 1 min (1 cycle); 94°C for 30 s; 68°C for 2 min (25 cycles)) between sense primer SP-A (positions Ϫ25-Ϫ6) and antisense primer AS-B (positions 900 -879) according to the manufacturer's instructions. The second PCR was performed with 1 ⁄50 of the first PCR reaction in the same conditions using primers SP-C (positions 1-22) and the AS-D (positions 877-853). After agarose gel analysis and transfer, the final PCR products were identified by hybridization with the 32 P-labeled nucleotide probe (nucleotides 453-478). For primer designation, nucleotide ϩ1 was taken as the first nucleotide of the hAQP3 initiation codon (GenBank TM accession number D25280).
Flow Cytometry Analysis-RBC suspensions at 0.2% (v/v) in phosphate-buffered saline (PBS) were fixed with 1% (v/v) formaldehyde and 0.025% (v/v) glutaraldehyde for 15 min at room temperature and permeabilized for 15 min with 1% (w/v) n-octyl glucoside. hAQP1 and rAQP3 expressions were examined by flow cytometry (FACSCalibur, BD Biosciences, San Diego, CA) using the fixed permeabilized RBCs of human, rat, and mouse species resuspended in PBS with 10% normal donkey serum and incubated at room temperature with rabbit anti-hAQP1 (1:200) or anti-rAQP3 (1:200) antisera. After 60 min of incubation, the RBCs were washed and stained with 100 l of phycoerythrinconjugated F(abЈ)2 fragments of donkey anti-rabbit IgG (1:40) for 30 min in the dark (Beckmann Coulter, Villepinte, France). After another washing step, the cell suspensions were analyzed.
Aquaporin Detection by Western Blot Analysis-RBC membranes were prepared by hypotonic lysis in the presence of 0.5 mM AEBSF as described previously (33). From the membrane pellet, a part was conserved at Ϫ20°C in PBS for chemical cross-linking experiments and velocity gradient analysis. The other part of the pellet was resuspended in 10 mM Tris-HCl, pH 6.8, 1 mM EDTA, and 5% (w/v) SDS for deglycosylation. RBC membrane proteins (cross-linked or not) and renal homogenate were digested overnight at 37°C with N-glycosidase F according to the manufacturer's instructions before electrophoresis. A control reaction was carried out in enzyme-free buffer in otherwise identical conditions. Western blot analyses were performed from protein preparations separated by SDS-PAGE, transferred to polyvinylidene difluoride membrane (PerkinElmer Life Sciences, Boston, MA), blocked in PBS/milk 5%, and incubated for 1 h at room temperature with rabbit anti-hAQP1 (1:4,000) or anti-rAQP3 (1:500) antisera. Specifically bound antibodies were visualized with a peroxidase-conjugated goat anti-rabbit IgG (1:5,000) (Promega, San Luis Obispo, CA) revealed by chemiluminescence (ECLϩPlus, Amersham Biosciences, Inc.).
Protein Solubilization and Velocity Gradient Analysis-RBC and yeast membrane samples (S. cerevisiae strain W303-1A transformed with rAQP3-PYX 222 and prepared according to Ref. 34) containing 800 and 700 g of protein, respectively, were solubilized in 100 l of a buffer composed of 20 mM Tris, pH 7.4, 1 mM dithiothreitol with either 1% (w/v) SDS, 1% (w/v) Triton X-100, 2% (w/v) n-octyl ␤-D-glucopyranoside (OG), or 2% (w/v) N-laurylsarcosine (NLS) overnight at 4°C. The samples were centrifuged at 100,000 ϫ g for 30 min at 4°C to remove unsolubilized material. Velocity sedimentation by gradient centrifugation was performed as described (25). Solubilized membrane proteins (90 l) were loaded on a 5-ml sucrose gradient (2-20% (w/v)) of the different solubilization buffers (gradients with SDS were only at 0.1% (w/v) SDS) and subjected to a 100,000 ϫ g centrifugation for 16 h at 5°C. Twenty fractions of 250 l were collected from the bottom to the top, and 10 l of each were analyzed by SDS-PAGE and immunoblot. As sedimentation markers, a mixture of ␤-amylase, 8.9 S (200 kDa); bovine serum albumin, 4.3 S (66 kDa); and cytochrome c, 1.8 S (12.4 kDa) was similarly processed.
Rat AQP3 Expression in Xenopus Oocytes and Freeze-fracture Studies by Electron Microscopy-From the pSport1-rAQP3 construct, generously provided by Dr. M. Echevarria (Sevilla, Spain), the rAQP3 cDNA was excised by EcoRI digestion, blunt-ended, and cloned into EcoR V site of pT7TS vector (kindly provided by Dr. P. Krieg, Austin, TX) (35). After linearization of the pT7TS-rAQP3 construct with XbaI restriction enzyme, capped sense RNA was synthesized using T7 RNA polymerase with the mCAP mRNA capping kit (Stratagene, La Jolla, CA). hAQP1 in pT7TS was kindly provided by V. Laizé (CEA/Saclay). Expression studies were carried out by microinjection of 50 nl of either water or complementary rAQP3 or hAQP1 RNAs (0.2 mg/ml) in collagenasetreated oocytes, and functional tests were realized 3 days after injection as described previously (36).
Before fixation in 2.5% (w/v) glutaraldehyde in Barth solution for 2 h at 18°C, H 2 O or hAQP1 or rAQP3 cRNA-injected oocytes were rapidly emptied of their cytoplasm by aspiration with a Ͼ 100-m-diameter pipette. Emptied oocytes were cryoprotected by incubation in Barth solution supplemented with 30% glycerol for 1 h at room temperature, placed between two copper sample holders, and then frozen in melting 1, 1, 1, 2-tetrafluoroethane, a cryofreeze aerosol purchased from Agar (Essex, UK). Samples were fractured in a Balzers 300 apparatus at Ϫ150°C under a 10 Ϫ7 torr vacuum. Fractured surfaces were coated with platinum at 45°and carbon at 90°under the conditions of the manufacturer. Replicas were cleaned in bleach, washed in distilled water, and observed in a Philips EM400 microscope at 80 kV. Representative series of images of P face (protoplasmic face) fractures were enlarged at ϫ95,000 final magnification to determine the intramembrane particle (IMP) density (particles/m 2 ). Images were amplified by 800,000 to estimate the size of the particles according to Eskandari et al. (30).
Cross-linking Experiments-For cross-linking, RBC membrane pellets containing 660 g of protein in 100 l of PBS were mixed with 10 l of 50 mM DSS in Me 2 SO (Pierce), giving a final concentration of 4.5 mM. Incubations were carried 30 min at room temperature. The reaction was quenched by the addition of 5.5 l of 1 M Tris-HCl, pH 7.5 (20 mM final) followed by a 15-min incubation on ice. Cross-linked RBC membrane proteins were treated or not with N-glycosidase F (see above). After electrophoresis, Western blots were performed with IgG anti-rAQP3 and anti-hAQP1.

Analysis of AQP3 Transcripts in Human Erythroid
Tissues-We first demonstrated by nested PCR that the AQP3 transcript was present in fetal liver, bone marrow, and reticulocytes from human tissues. As expected, a PCR product of 876 bp was identified in each case (Fig. 1A). To further identify these products, PCR material was transferred to a Hybond membrane and hybridized with a radioactive oligonucleotide probe corresponding to a part of the sequence coding for the C external loop specific to AQP3. A band at the same expected position was revealed from the three different samples (Fig. 1B).
Flow Cytometry Analysis of AQP3 and AQP1 on RBCs-To confirm our previous studies indicating that hAQP3 is present in human RBC (29), we performed immunofluorescence analysis using permeabilized human, rat, or mouse RBCs stained with anti-rAQP3 (cross-reacting with human (29) and mouse AQP3, see immunoblot in Fig. 3) or anti-hAQP1 antibodies (Fig. 2). An increase of the fluorescence was observed when human and rat (but not mouse) permeabilized RBCs were incubated with the anti-rAQP3 (Fig. 2, heavy solid lines) as compared with incubation with the preimmune serum (Fig. 2,  dashed lines). In addition, no signal was observed when AQP3 antibodies were incubated with an excess of antigen peptide (not shown). As a positive control of these reactions, we used an antibody prepared against hAQP1 purified from RBCs, which reacted with the three permeabilized RBC species (Fig. 2, light  solid lines).
Glycosylated and Non-glycosylated AQP3 in RBC and Kidney Membranes-Immunoblots were performed using our polyclonal antibody directed against the 26 amino acids C-terminal of rat AQP3 (Fig. 3). In human RBC membranes, a single band at ϳ46 kDa, which was reduced to 26 kDa after N-glycosidase F treatment, was detected. AQP3, however, was absent from human platelets (data not shown). Examination of rat RBC membranes revealed two major bands: one at 25 kDa and a broader one at 32 kDa. After N-glycosidase F treatment, the 32-kDa component was reduced, whereas the 25-kDa band was increased. A similar profile was obtained with rat inner medulla kidney cells, but the glycosylated band was of greater size (38 instead of 32 kDa) as compared with that of RBC membranes. In rat and mouse preparations, a weak additional band of ϳ66 kDa that was reduced to 46 kDa after N-glycosidase F treatment was detected. No signal could be detected with mouse RBC membranes, although the anti-rAQP3 antibody cross-reacted with AQP3 present in the total mouse kidney homogenate. These results confirm that AQP3 is present in human and rat RBC membranes but is absent (or very reduced) in those of the mouse.
Velocity Gradient Centrifugation-To determine the oligomerization of AQP3 by comparison with the well known tetramerization of AQP1, we first performed velocity sedimentation on sucrose gradients of human RBC membranes, which possess both hAQP1 and hAQP3. To compare the oligomerization state of these two membrane proteins in detergent micelles, we assume that the amount of detergents bound to hAQP1 and hAQP3 is the same because the hydrophobicity profiles of the MIP proteins are highly conserved (25). Twenty fractions of 250 l were collected from each gradient of 5 ml (from 1 (heavy fraction) to 20 (light fraction)), solubilized in Laemmli buffer, and analyzed by immunoblot. As shown in Fig.  4, when human RBC membranes were solubilized in 1% SDS (left), the monomeric glycosylated and non-glycosylated forms of hAQP1 (Fig. 4, top) appeared in light fractions (fractions [15][16][17], corresponding to an S value close to 2 as determined previously for AQP1 monomer (11). Immunoblotting of the same fractions with anti-rAQP3 (Fig. 4, bottom) revealed a glycosylated monomeric form of hAQP3 in light fractions (around fractions 14 -16) as well as a dimeric glycosylated one in heavier fractions (around fraction 12), which was not dissociated in Laemmli buffer. A higher oligomeric SDS-insoluble form might be present in fractions 8 -9. RBC membranes were also solubilized in non-denaturing detergents before laying on gradients. Fig. 4 (right) shows the position of the fractions containing hAQP1 (top) and hAQP3 (bottom) after NLS 2% solubilization. As expected, hAQP1 in a non-denaturing detergent conserved its tetrameric structure; it was found in heavier sedimentation fractions (fractions 6 -8) with an S value around 8. We can notice that a very small part of hAQP1 was visible in a fraction that peaks at 14 corresponding to dissociated mono- , and mouse (C) RBCs were stained with anti-rAQP3 antibody (heavy solid lines) and anti-hAQP1 antibody (light solid lines) followed by phycoerythrin-donkey anti-rabbit IgG (F(abЈ) 2 fragments) and analyzed by flow cytometry as described under "Materials and Methods." Preimmune rabbit sera were used as negative controls (dotted lines). The fluorescence intensity subjected to linear amplification is shown on the x axis, and the cell number is shown on the y axis. The mean fluorescence intensity is indicated for sera. For preimmune serum, values were 55 and 59 for AQP1 and AQP3, respectively, for human RBCs; 59 and 61 for rat RBCs; and 61 and 62 for mouse RBCs. mers with an S value of 4.3. In contrast, hAQP3 was found in lighter fractions, around 13 (5S) for glycosylated monomers and around 10 (7S) for glycosylated dimers. The use of Triton X-100 and OG gave the same results (data not shown) regarding the difference of the sedimentation between hAQP1 and hAQP3, but the S values found were smaller, close to the S values determined previously (7,11). These sedimentation shifts to heavier fractions obtained with NLS as compared with SDS, Triton X-100, or OG are probably due to a higher amount of NLS bound to the proteins. Velocity gradient centrifugation was further performed with yeast membranes expressing rAQP3, and the fractions were analyzed as above. In SDS-or NLS-solubilized yeast membranes, two N-glycosidase F-resistant bands at 25 and 50 kDa corresponding to the same sedimentation coefficient obtained with RBC membranes were revealed with the antibody against rAQP3 (not shown).
Freeze-Fracture Studies on Oocyte Plasma Membrane-As the oligomeric state of AQP3 was not clearly resolved by velocity sedimentation, possibly due to the higher sensitivity of AQP3 to non-denaturing detergent, we carried out another approach associating freeze-fracture and electron microscopy. Measurements of density and diameter of IMPs were per-formed in the plasma membrane of hAQP1-, rAQP3-, or H 2 Oinjected Xenopus oocytes at 72 h after injection. Fig. 5A presents an example where an increase in IMP density can be observed on the P face of hAQP1 and rAQP3 oocyte plasma membrane as compared with control oocyte (723 Ϯ 165 and 406 Ϯ 76, respectively, versus 181 Ϯ 40). Total areas screened in each condition were respectively 37 mm 2 (30 pictures), 27 mm 2 (nine pictures), and 47.7 mm 2 (24 pictures) of membrane replicas from rAQP3-, hAQP1-, and H 2 O-injected oocytes. We then measured the diameter of particles (ϳ600 -800 particles in each condition) perpendicularly to the direction of the shadow using a software and a calibration with 10-nm gold beads. We found a distribution following Gaussian curves, represented in Fig. 5B. The mean diameters with standard deviations were found to be 8.4 Ϯ 1.6 nm (n ϭ 597) for endogenous particles. For hAQP1-and rAQP3-expressing oocytes, the mean diameters were 8.2 Ϯ 1.4 nm (n ϭ 777) and 7.4 Ϯ 1.3 nm (n ϭ 879), respectively. These three Gaussian curves are significantly different at p Ͻ 0.001. According to the oligomeric models constructed from a monomeric unit with a mean diameter of 3 nm (46) and assuming the thickness of the platinum-carbon layer to be ϳ0.75 nm (12), the oligomeric state of AQP3 in Xenopus oocyte plasma membrane is a trimeric form. However, FIG. 3. Immunoblot analysis of AQP3 from human, rat, and mouse RBC membranes and kidney tissues. Untreated (Ϫ) and N-glycosidase F-treated (ϩ) RBC membranes and kidney preparations were separated by SDS-PAGE and immunoblotted with anti-rAQP3 antibody as described under "Materials and Methods." Two g each of human, rat, and mouse RBC membranes were loaded on the gel. For rat and mouse kidneys, 5 g of inner medulla homogenate and 10 g of total homogenate, respectively, were loaded.

FIG. 4. Velocity sedimentation analysis on sucrose gradient.
Human RBC membrane proteins solubilized with different detergents were loaded on sucrose gradient, subjected to centrifugation, and analyzed on 12% SDS-PAGE immunoblots with anti-hAQP1 and anti-rAQP3 antibodies as described under "Materials and Methods." Left, human RBC membranes solubilized in 1% SDS; right, human RBC membranes solubilized in 2% NLS. Molecular markers are indicated.

FIG. 5.
Freeze-Fracture studies of hAQP1 and rAQP3 cRNAinjected oocyte plasma membranes. Analysis of P face replicas obtained after freeze-fracture of oocyte membranes. A, high magnification of unidirectional shadowed replica micrograph pictures representative of the analyzed membrane surface areas of oocyte injected with water and hAQP1, or rAQP3 cRNAs. B, size distribution of P face particles in oocytes expressing AQP1 (q), rAQP3 (f), and water-injected oocyte control (E). Particle diameter was figured out by measuring the width of the particle perpendicularly to the direction of the shadow. For the frequency histograms, particle diameter measurements were placed in 0.5-nm bins, and the histograms were plotted at the center of the bin. Results were presented as frequency histograms that were fitted to a single or multiple Gaussian functions shown in Eq. 1, where x is the particle size (nm) and f(x) is the relative frequency for a given size, A is an approximation of the area under the curve for a given particle population, m is the mean diameter for each population, and s is the standard deviation.
if we consider a platinum-carbon thickness of 1.5 nm (37), a dimeric structure seems to be the most representative of the oligomeric state of AQP3 in these plasma membranes.
Cross-linking of AQP3 from Human RBCs-Total human RBC membrane proteins were subjected to chemical DSS crosslinking followed (or not followed) by digestion with N-glycosidase F (Fig. 6). Western blot was performed using the anti-rAQP3 and anti-hAQP1 antibodies. In the samples untreated with the cross-linker, we observed with the anti-AQP3 a major band visualized around 46 kDa, presumably representing the glycosylated AQP3 monomer (Fig. 6, lane 1). When membranes were treated with DSS (Fig. 6, lane 2), the majority of the signal was present at high molecular mass, corresponding to glycosylated oligomers. When the DSS-untreated sample was digested with N-glycosidase F, there was an important shift to the monomeric form at 25 kDa (Fig. 6, lane 3 versus lanes 1 and  2). After DSS treatment and deglycosylation, the presence of oligomeric structures (including tetrameric ones) was observed (Fig. 6, lane 4 versus lane 3). AQP1 was not sensitive to DSS incubation as the same profile was observed in control and DSS-treated samples (not shown).

DISCUSSION
Our previous studies (29) have shown by biochemical and functional approaches that the AQP3 protein is present on human RBC membranes. Our present studies, including nested reverse transcription-PCR from human erythroid tissues (reticulocytes, bone marrow, fetal liver), flow cytometry, and immunoblot analysis, indicate that the AQP3 transcript is present in erythroid tissues and confirmed that the AQP3 protein is present on the human RBCs. These findings contradict those of Yang et al. (38) and might be explained by a difference in either the antibodies used or in sample preparation (RBC permeabilization). AQP3 was also present on rat RBCs but was absent from mouse RBCs as shown here by flow cytometry and immunoblot analysis with the anti-rAQP3 antibody. This antibody, however, was able to detect AQP3 in kidney homogenate from both rat and mouse. The absence of AQP3 in mouse RBCs was also observed by Yang et al. (38). The large disparity in the glycerol permeability of vertebrate RBCs (39) could be explained by either the high or low density of membrane AQP3 from one species to another or by some other protein component or by both.
The tetrameric structure of AQP1 in its native membranes and after expression in heterologous systems has been shown by a variety of methods including biochemical approaches (size exclusion chromatography and velocity sedimentation after solubilization in non-denaturing detergents) (7,11,14,40,41), freeze-fracture studies (12), and projection maps obtained from AQP1 crystals (8 -10, 42-44). Other aquaporins strictly permeable to water have also been shown to be tetrameric by biochemical analysis (AQP2 (15), AQP4 (16), AQPZ (45), AQPcic (26)) or by crystallographic analysis (AQP0 (17), AQPZ (18), AQPcic (19), and ␣-TIP (20)). An important issue related to these studies is to determine whether there is a possible link between the membrane organization and the selectivity of MIP proteins and more specifically whether or not aquaglyceroporins and glycerol facilitators are also tetrameric. Some studies have been done on GlpF by all three methods described before to answer this question. After solubilization in non-denaturing detergent and velocity sedimentation on sucrose gradient, GlpF was found to be monomeric when expressed in oocyte membranes (25,26). In freeze-fracture studies, the sizes of IMPs of oocyte membranes expressing GlpF were found to be smaller than those found for AQPcic, 5.8 versus 8 nm, also suggesting a monomeric form of GlpF in oocyte membranes (46). By contrast, the projection map of GlpF crystals showed that the protein is present as tetramers (28).
Here, we attempted for the first time to determine the membranous organization of the AQP3 aquaglyceroporin that possesses a mixed selectivity (water/solutes). We used the human RBC membranes, which bring the advantage to possess both AQP1 and AQP3 in their native states, and also two heterologous systems: yeast, expressing a non-glycosylated form of AQP3, and oocyte, used for size estimation of AQP3 by freezefracture. We first studied the migration of these proteins in a sucrose gradient after treatment with denaturing and nondenaturing detergents. The observation of monomeric and dimeric forms for AQP3 in all membranes led us to consider two possibilities: (i) AQP3 is present in monomeric and dimeric forms in membranes and (ii) AQP3 resides in tri-and/or tetrameric forms in membranes but shows a higher sensitivity to non-denaturing detergents than AQP1. Our results show that important differences of sensitivity to non-denaturing detergents exist between AQP1 and AQP3. Interestingly, a recent study by sucrose gradient sedimentation analysis showed that variation of the ionic strength can change the oligomeric state of GlpF and that Mg 2ϩ stabilizes its tetramer assembly (47). Thus, our results present some similarities with the results obtained on GlpF regarding the various oligomerization state of AQP3, which may vary according to the analytical technique used, notably the low ionic condition that favors monomer formation.
Freeze-fracture studies have already been used to estimate the IMP diameter into plasma membranes where aquaporins were expressed (12,30,37,46,48). Further analysis by freezefracture studies on oocyte plasma membranes revealed only a small difference between the IMP size of AQP3 and AQP1, thus suggesting an oligomeric state of AQP3. However, it is difficult to determine which oligomeric state of AQP3 predominates into oocyte plasma membrane due to the weak precision concerning the evaluation of the thickness of the platine-carbon layer. The mean size of a dimer and trimer of 3-nm-diameter units (size of AQP1 monomer (12)) are 4.5 and 6.0, respectively (46). The addition of a small layer of platine-carbon (0.75 nm) is compatible with a trimeric structure, whereas a more consequent replica (1.5 nm) should be in favor of a dimeric structure. The diameter size of the tetrameric AQP1 in our study (8.2) means that, with our apparatus, the platinum thickness is probably small, around 0.85 nm according to the mean diameter of 6.5 nm from a theoretical tetrameric complex (44). If we assume that AQP3 is a tetramer in membranes, the difference in the diameter of the particles seen by freeze-fracture could be due either to a difference in the shape of the tetramer, which is perhaps more compact with tight monomers, or to a more important embedding of AQP3 in lipid bilayer as proposed by van Hoek et al. (37) comparing aquaporin sizes in Chinese hamster ovary, or to both. We must also keep in mind that injecting exogenous cRNA in X. laevis could sometimes lead to the expression of endogenous proteins (49), resulting in underestimation of the AQP3 IMP density and perhaps of their size (37).
The most interesting result came from the cross-linking experiments performed with DSS. As observed, dimers, trimers, and tetramers were stabilized by DSS treatment, whereas AQP1 seemed to be unaffected by the DSS treatment, confirming different quaternary structures of AQP1 and AQP3. In conclusion, our results provide evidence that AQP3 exists in multiple oligomeric forms composed of weakly associated monomers. Analysis of channel and transport proteins favor oligomeric structures (50), sometimes with a regulated equilibrium between different forms (51). The exact role of the oligomeric arrangement of the proteins of the MIP family is not yet understood.