Structural Characterization of Two Aquaporins Isolated from Native Spinach Leaf Plasma Membranes*

,

(AQP1), 1 from human erythrocytes and the demonstration of water transport in Xenopus oocytes expressing its complementary RNA (5) confirmed this hypothesis.
Since then, a large number of channel-forming integral proteins homologous to AQP1 have been found in all forms of life (6). This membrane protein family was initially named the MIP family after its first sequenced member, the major intrinsic protein (MIP) of bovine lens fiber cells (7). Multiple sequence alignment and phylogenetic analysis of 164 members of the MIP family, now frequently referred to as aquaporin super family, revealed 16 subfamilies that form two distinct clusters, the aquaporin (AQP) cluster and the glycerol facilitator-like cluster (8). The AQPs are highly specific for water, whereas the glycerol facilitators allow the passage of small, nonionic molecules such as glycerol and urea (9). In addition, ovine MIP tetramers have been found to form a groove and tongue contact via their extracellular surfaces, lending support to a dual function of the protein, as a water channel and as a cell to cell adhesion molecule in the eye lens (10).
Most members of the aquaporin super family have similar molecular masses, ranging from 25 to 31 kDa. Based on their sequence homology all members are predicted to comprise six hydrophobic, membrane-spanning ␣-helices connected by five loops of variable length and to have cytosolic N and C termini (7,11). Highly conserved regions are located on the loops B and E, which contain the NPA amino acid motifs (12). Site-directed mutagenesis experiments led to the hypothesis that loops B and E fold back into the membrane and that the NPA boxes must be involved in the selectivity filter of the channel (13). Other highly conserved residues are found in the helices, revealing the transmembrane helix-helix packing motif GXXXG (14), as well as conserved charged buried residues that were proposed to form ion pairs (15).
Velocity sedimentation, glutaraldehyde cross-linking, and gel filtration (16), as well as transmission and scanning transmission electron microscopy (STEM) (17,18), have shown that AQPs and glycerol facilitators are tetrameric proteins. When reconstituted into lipid bilayers, these proteins often form highly ordered two-dimensional (2D) crystals suitable for cryoelectron and atomic force microscopy (AFM). Consequently, a wealth of structural information is now available for several aquaporins from different organisms, ranging from single particle projections to the atomic model elucidated at 0.38-nm resolution (19) and subnanometer topographical data recorded by AFM (see Ref. 18).
Cell to cell water transport is required for many physiological processes in plants, including the transcellular movement of water in the transpiration stream, the circulation of water between the xylem and the phloem, the stomatal or organ movement, and cell enlargement. Aquaporins have been found in the plasma membrane (plasma membrane intrinsic proteins (PIPs)) and in the vacuolar membrane (tonoplast intrinsic proteins (TIPs)) of plants, being expressed in organ-, tissue-, and cell type-specific manners. At the cellular level PIPs act in water uptake and release, whereas TIPs are thought to mediate cellular turgor maintaining the structural integrity of the cell. Plant PIPs are further divided into two subfamilies named PIP1 and PIP2 (20). In addition to several single amino acid residue substitutions, PIP1s are characterized by a long N terminus and a shorter C terminus and PIP2s by a short N terminus and a longer C terminus (21). In contrast to PIP2, PIP1 homologs show poor water transport activity in oocytes (20,21). The interplay between these aquaporins is likely to be crucial in the regulation of the intracellular osmolarity. The first plant aquaporin to exhibit water transport activity when expressed in Xenopus oocytes similar to the human AQP1 was ␥-TIP from Arabidopsis (22). Evidence for the regulation of certain plant aquaporins by reversible phosphorylation was later shown for ␣-TIP from seeds (23) and PM28A from spinach leaves (24) (for recent reviews on plant aquaporins see Refs. 25 and 26).
The major integral proteins of spinach leaf plasma membranes migrate as a single band of 28-kDa molecular mass and were thus termed PM28. Edman degradation of endoproteinase Lys-C fragments originating from the 28-kDa band, polymerase chain reaction with oligonucleotide primers based on the obtained amino acid sequences, and subsequent screening of a spinach leaf cDNA library yielded two full-length clones of MIP homologs, pm28a and pm28b. The sequence information showed that PM28A belongs to the PIP2, whereas PM28B belongs to the PIP1 subfamily (1). In vivo, PM28 was identified as the major phosphoprotein of the spinach leaf plasma membrane (1). In vitro experiments using plasma membrane vesicles demonstrated that PM28A phosphorylates in a Ca 2ϩ -dependent manner by a plasma membrane-associated protein kinase (1). PM28A was shown to be an aquaporin by expression in Xenopus oocytes, and its water channel activity is regulated by phosphorylation at two different serine residues (24).
Here, we present the structure of PM28A and a new putative aquaporin isoform, PM28C, whose sequence is distinct from that of PM28B but is also a member of the PIP1 subfamily. A fast and efficient purification protocol likely to be applicable to other plant membranes has been developed. The oligomeric state of PM28A and PM28C particles is defined by STEM mass analysis, and their shape, dimensions, and distribution are defined by electron and atomic force microscopy. In addition, the exact mass and stochiometry of the mixture containing the two isoforms was determined by matrix-assisted laser desorption ionization-mass spectrometry (MALDI-MS). The absence of PM28B in our preparations and the ratio of PM28A to PM28C indicate that these PIPs are differentially expressed in spinach leaf plasma membranes.
Gels were electroblotted onto Immobilon polyvinylidene difluoride membranes (Millipore, Bedford, MA). Blots were blocked for 30 min in 2% milk powder dissolved in 10 mM Tris-HCl (pH 8), 150 mM NaCl, 0.05% Tween 20, 0.02% NaN 3 (TBST), and was subsequently incubated for 60 min with the primary antibody (serum-diluted 1:2000 in TBST) directed against the sequence ALGSFRSNPTN in the C-terminal region of PM28A (see Table I). Before incubation with the anti-rabbit IgG alkaline phosphatase-conjugated secondary antibody (Sigma; diluted 1:2000 in TBST), the blot was washed once for 15 min in TBST containing 2% milk powder and 3 times for 10 min in TBST. Blots were washed 3 times in TBST for 10 min, once with 100 mM Tris-HCl (pH 9.5), 100 mM NaCl, 5 mM MgCl 2 and were then developed with Western Blue™ (Promega) stabilized substrate for alkaline phosphatase. All washing and incubation steps of the blot were performed at room temperature.
Two-dimensional Crystallization-Purified protein (2 mg/ml) was mixed with E. coli lipids solubilized in OTG (mixed micelles stock solution, 4 mg/ml E. coli lipids in 20 mM Mes-NaOH (pH 6), 5% OTG, 0.01% NaN 3 ) to achieve a lipid to protein ratio of 1 (w/w). The final protein concentration was adjusted to 1.33 mg/ml, and the final OTG content was adjusted to 1.93%. The reconstitution mixture (60 l) was preincubated at room temperature for 30 min and dialyzed against 1.5 liters of 10 mM Mes-NaOH (pH 6), 100 mM NaCl, 100 mM MgCl 2 , 2 mM dithiothreitol, 0.01% NaN 3 for 24 h at room temperature, 24 h at 37°C, and another 24 h at room temperature.
Separation of the peptides was performed at a flow rate of 50 l/min on a C18 reverse-phase column (1 ϫ 250 mm; VYDAC™ 218TP51, Hesperia, CA) connected to an Applied Biosystems 120A pump. Bound peptides were eluted with a 60-min linear gradient from 0.1% trifluoroacetic acid (solvent A) to 80% acetonitrile containing 0.09% trifluoroacetic acid (solvent B). Elution of protein fragments was monitored at 214 nm.
N-terminal Peptide Sequencing-Automated Edman degradations were performed on an Applied Biosystems 473A protein sequencer according to the manufacturer's recommendations.
Mass Spectral Analysis-MALDI-TOF analysis was performed on a Bruker REFLEX III mass spectrometer (Bruker Daltonik GmbH, Bremen, Germany). PM28 reconstituted into lipid bilayers was solubilized for less than 20 s in 80% formic acid at room temperature (final protein concentration, 4 mg/ml). After solubilization of the membranes, 2 l were immediately mixed with 1 l of matrix solution (10 mg/ml ␣-cyano-4-hydroxycinnamic acid (Aldrich) in 80% acetonitrile, 0.1% trifluoroacetic acid) and placed on the sample plate to dry. Calibration of the instrument in the high molecular mass range was done with bovine serum albumin (Sigma) by using the molecular masses of the singly, doubly, and triply charged forms of bovine serum albumin. Calibration and mass measurements of the PM28 proteins were carried out in the linear mode. The [MϩH] ϩ values of PM28 isoforms shown in Fig. 7B, the ratio between the two peaks, and the corresponding standard deviations were calculated from 13 different spectra. Mass spectra for the endoproteinase Lys-C fragments of PM28 were acquired by mixing 1 l of the fractions collected by reverse-phase HPLC with 1 l of matrix solution and spotting the mixture onto the target plate. For mass measurement in the low molecular range the instrument was operated in the reflectron mode and calibrated using the monoisotopic masses of the adrenocorticotropic hormone (fragment 18 -39; Fluka), substance P (Fluka), and angiotensin (Fluka). Scanning Transmission Electron Microscopy Mass Measurement-PM28 isoforms solubilized in OTG were adsorbed for 1 min to glow discharged thin carbon films supported by a thick fenestrated carbon layer (directly after cation-exchange chromatography). The gold-plated copper grids were then washed on 8 drops of quartz double-distilled water and were freeze-dried at Ϫ80°C overnight in the microscope. For mass analysis, annular dark-field images were recorded in a STEM (VG-HB5) at 80 kV and doses of 325 Ϯ 35 electrons/nm 2 . Digital acquisition of the images and microscope parameters, system calibration, and mass analysis were carried out as described previously (32). The total experimental error was calculated as the standard error of the mean, plus 5% of the measured particle mass to account for the absolute calibration uncertainty.
Transmission Electron Microscopy and Image Processing-Detergent-solubilized particles eluted from the cation-exchange column were directly adsorbed for 1 min to parlodion carbon-coated copper grids rendered hydrophilic by glow discharge at low pressure in air. Grids were washed with 4 drops of double-distilled water and stained with 2 drops of 0.75% uranyl formate. Images were recorded on Eastman Kodak Co. SO-163 sheet film with a Hitachi H-7000 electron microscope operated at 100 kV. Electron micrographs of single particles adsorbed to the carbon film were digitized using a Leafscan-45 scanner (Leaf Systems, Inc., Westborough, MA).
All image processing steps described below were performed using the Semper image processing system (33) (Synoptics Ltd., Cambridge, United Kingdom). For single-particle analysis PM28 complexes were extracted, aligned, and classified. Briefly, a reference was established by selecting a well preserved particle and symmetrizing it 20-fold rotationally (34,35). Cross-correlation functions of this reference with images of digitized micrographs containing adsorbed particles of PM28 revealed correlation peaks at the particle positions, irrespective of their angular orientation. Using this particle picking method a gallery of 4096 particles was created. These were submitted to a multivariate statistical analysis (36) without alignment and were classified into clusters of particles with similar features. To this end, a program package kindly provided by J. P. Bretaudière (37) was used. The various cluster averages revealed square and round shaped particles at different angular orientations. These averages were taken as references for subsequent angular and translational alignment of the extracted 4096 particles. Aligned particles were classified again, and cluster averages were calculated. The resolution of the calculated averages was estimated according to the Fourier ring correlation function (FRC) (38), the phase residual (PHR) (39), and the spectral signal to noise ratio (SSNR) (40).
Atomic Force Microscopy-The stock solution of crystals (1.33 mg/ml protein) was diluted 20-fold in 10 mM Tris-HCl (pH 8.8), 150 mM KCl, 25 mM MgCl 2 or 10 mM Tris-HCl (pH 8.8), 50 mM MgCl 2 , depending on the experiment. A 30-l drop of this solution was deposited on freshly cleaved muscovite mica (Mica New York Corp., New York, NY) prepared as described previously (41). After 15 to 30 min, the sample was gently washed with the appropriate buffer to remove membranes that were not firmly attached to the substrate. Images were acquired with a commercial AFM (Nanoscope III; Digital Instruments, Santa Barbara, CA) equipped with a 120-m scanner (j-scanner) and a liquid cell.

RESULTS
The polypeptide pattern of spinach plasma membranes exhibits a dominant protein band of 28-kDa molecular mass (Fig.  1A, lane 1, arrow). In the first purification step, proteins adhering to the membranes were removed by urea/alkali treatment (Fig. 1A, lane 2). This stripping yielded a pure plasma membrane preparation highly enriched in membrane proteins. New bands appeared above the 42.7-and the 66.2-kDa markers, indicating aggregation of the very dominant polypeptide otherwise running at 28 kDa (Fig. 1A, lane 2). Moreover, the lipid bilayer not accessible to the detergent before the urea/ alkali treatment could now be solubilized by OTG and other detergents (e.g. dodecyl-␤-maltoside and octyl-␤-D-glucopyranoside; data not shown). When a membrane preparation solubilized in 3% OTG was applied to a cation-exchange column, a prominent peak was observed in the elution profile at a concentration of about 55 mM NaCl (Fig. 1B). Fractions showing UV absorbance at 280 nm were collected and subjected to SDS polyacrylamide gel electrophoresis and Western blotting using an antibody directed against the C terminus of PM28A. The polypeptide bands in the silver-stained SDS gel (Fig. 1C) corresponded to the bands from the Western blot (Fig. 1D) and could, therefore, be assigned to PM28A. As seen in Table I, the peptide used to raise the PM28A-specific antiserum is not present in the PM28B or PM28C protein. 2 The tendency of PM28A to form higher oligomers can also clearly be seen in the Western blot. If the cation exchange column was not overloaded by OTG-solubilized plasma membranes, SDS polyacrylamide gel electrophoresis of the flow-through showed complete depletion of the band at 28 kDa, indicating full binding of the PM28 proteins to the column matrix (data not shown).
To check the purity of isolated PM28A further, the protein pool eluted from the cation exchanger was cleaved with the endoproteinase Lys-C, and the digest was separated by reverse-phase HPLC. The m/z values measured by MALDI-TOF for the fractions of the various elution peaks are indicated in Fig. 2. Fractions that could be identified as fragments coming from PM28A (Fig. 2, underlined m/z values) are listed in Table  II. Fractions that could not be identified based on the measured mass were subjected to amino acid sequencing (Fig. 2, peaks  marked with an asterisk). The fragments at m/z 2406.3 and m/z 4165.5 yielded the N-terminal sequences a, GFQPG-PYQVGGGGSNYVHHGYTK and b, QINNWNDHWIFWVGP-FIGA, respectively. Edman degradation of the peptides at m/z 1594.5 and m/z 5526.6 yielded no amino acid sequences. Fig. 3 illustrates the high homology between the two sequenced fragments and parts of the PM28A sequence. Accordingly, the fragment at m/z 2406.3 forms the C loop, and that at m/z 4165.5 forms part of the E loop and helix 6 of PM28C. The sequence of new isoform, PM28C, shows that it belongs to the PIP1 subfamily of plasma membrane AQPs 2 (see "Experimental Procedures" for details on the sequence analysis and assign-ment of the fragments to the different domains of PM28A). No indications were found for the presence of PM28B in our preparations.
Mass measurement by STEM was used to assess the homogeneity of the purified PM28 isoforms and determine their aggregation state. A typical low dose dark-field image recorded from freeze-dried, unstained PM28 is shown in Fig. 4A. The mass analysis of 1169 particles yielded a single peak at 157 Ϯ 56 kDa after correction for beam-induced mass-loss. This is compatible with a tetrameric protein embedded in a detergent micelle of about 40 kDa. Trimeric or pentameric complexes could be excluded, because the total experimental error of the mean amounted to Ϯ 8 kDa (for details see "Experimental Procedures").
Detergent-solubilized PM28 particles were negatively stained and examined by transmission electron microscopy. Fig. 5A shows the homogeneity of the purified PM28 isoforms after the cation exchange chromatography step. However, two particle types with subtle differences could be distinguished upon close inspection of the electron micrographs, one type being larger and rather circular (Fig. 5B) and the other smaller and almost square (Fig. 5C). Single-particle and multivariate statistical analysis was applied to improve the signal to noise ratio. Top view projection averages for the two particle types 2 L. Fraysse and P. Kjellbom, unpublished results.

PM28A A A I K A L G S F R S N P T N PM28B
I P F K S R S PM28C I P F K S K FIG. 2. C18 reverse-phase chromatography of endoproteinase Lys-C-digested PM28. The numbers above the peaks indicate the masses of the peptides measured by MALDI-TOF in the corresponding fractions. Masses that could be assigned to the known PM28A isoform are underlined (for more details see Table II). Peaks labeled with B indicate components originating from the buffer-containing detergent (20 mM Bicine-NaOH (pH 8.75), 150 mM NaCl, 0.4% OTG, 0.008% NaN 3 ). The elution peaks labeled with N yielded blank spectra in the mass spectrometer. Elution was monitored at a UV absorbance of 214 nm. Peaks marked with an asterisk were subjected to amino acid sequencing. The fragments at m/z 2406.3 and m/z 4165.5 could be successfully sequenced by Edman degradation and yielded the amino acid sequences GFQPGPYQVGGGGSNYVHHGYTK and QIN-NWNDHWIFWVGPFIGA, respectively.
were calculated from a gallery of 4096 particles extracted from digitized electron micrographs by automated picking. Both populations were clearly tetramers as already predicted by STEM mass measurement, consequently the averages were 4-fold symmetrized. The projection of the larger and rather circular PM28 isoform (Fig. 5B, (-) unsymmetrized and (ϩ) 4-fold symmetrized particles) calculated from 444 motifs of 4096 had a side length of 8.8 nm and showed less density in the center of the tetramer than at the periphery. The smaller and almost square PM28 isoform was less abundant (Fig. 5C, (-) unsymmetrized and (ϩ) 4-fold symmetrized particles). The projection was calculated from 255 motifs of 4096, had a side length of 8.0 nm, and had a homogeneous density distribution over the whole particle. The resolution of the projection averages was determined using the following three criteria: (i) the FRC (38), (ii) the PHR (39), and (iii) the SSNR (40). The average of the larger and rather circular particle type (Fig. 5B) had a resolution of 1.8 (FRC), 2.4 (PHR), and 1.6 nm (SSNR), whereas the smaller and almost square one (Fig. 5C) had a resolution of 1.8 (FRC), 2.5 (PHR), and 2.6 nm (SSNR).
2D crystals formed when PM28 was reconstituted into lipid bilayers. Two different crystal types were found; one with a p42 1 2 symmetry and lattice vectors of a ϭ b ϭ 9.76 Ϯ 0.06 nm (Fig. 6A) and the other with a hexagonal lattice of closely packed tetramers and unit cell dimensions of a ϭ b ϭ 8.58 Ϯ 0.18 nm (Fig. 6B). Imaging in the AFM, which has a much higher signal to noise ratio than the transmission electron microscope, indicated segregation of the two isoforms into two different crystal types and confirmed the structural features of the isoforms revealed by single particle analysis. The tetramers in the p42 1 2 crystal could be correlated to the smaller and almost square PM28 isoform (Fig. 5C), whereas the particles in the hexagonal lattice corresponded to the larger and rather circular one (Fig. 5B). Crystals containing a mixture of both particle types were not observed. The surface topograph of the p42 1 2 crystal (Fig. 6A and inset) shows some similarity to the topograph of aquaporin-Z from E. coli (43). Also the unit cell dimensions of the crystal and the side length of the particles  forming it are almost identical to those found for aquaporin-Z (44). Particles forming the hexagonal crystal were less compact and were characterized by a deep indentation in the center of the tetramer. This central indentation correlates with the low density found in the larger and rather circular projection average obtained by single-particle analysis (Fig. 5B). Also in AFM topographs, the particles displayed a less pronounced square shape, although four corners could frequently be seen at higher magnification ( Fig. 6B and inset). The tetramers seem to be flexible, allowing the formation of a hexagonal lattice, which has never been observed before for aquaporin crystals.
Mass analysis by MALDI-TOF was performed on the reconstituted membranes to measure the exact mass of the mature proteins and to exclude the existence of additional isoforms, e.g. PM28B. Fig. 7A shows the MALDI spectrum of PM28 isoforms. Two singly charged molecular ions, [MϩH] ϩ , were recorded, one at m/z 29839 Ϯ 11 (n ϭ 13) and the other at m/z 30683 Ϯ 7 (Fig. 7B, n ϭ 13). In addition to the [MϩH] ϩ , the doubly charged ion, [Mϩ2H] 2ϩ , and the singly charged dimer, [2MϩH] ϩ , of PM28A and the new isoform were observed (Fig.  7A). No additional peaks, which would indicate the presence of other proteins, were found. Hence, the purification strategy described in the present report yields PM28A and PM28C preparations of high purity. DISCUSSION The urea/alkali stripping of spinach leaf plasma membranes proved to be a crucial step in our purification protocol. Not only was the amount of nonmembrane protein contaminants considerably reduced, but the lipid bilayers were also made accessible to the solubilizing detergent, enabling highly efficient solubilization. Because the plasma membranes of various plants and their tissues can be isolated by two-phase partitioning (27), this purification step should also facilitate the isolation of both their aquaporins and other membrane proteins.
The aquaporin PM28A (1) from spinach leaf plasma membranes was identified by Western blot and MALDI-MS analysis of endoproteinase Lys-C-digested PM28A peptide fragments. The latter analysis also led to the discovery of a new isoform, PM28C, that copurifies with PM28A. The amino acid sequences of two domains of this isoform were acquired by Edman degradation. Comparison revealed a high homology between corresponding domains of PM28A and PM28C, especially at the beginning of helix 6 (see Fig. 3). The deduced amino acid sequence of the new isoform, PM28C, has been shown to be distinct from that of PM28B. However, both PM28B and PM28C belong to the PIP1 subfamily, whereas PM28A belongs to the PIP2 subfamily of plasma membrane AQPs. 2 These PM28 proteins have very similar masses and high calculated pI values (pI ϳ10) and should therefore copurify together. PM28B, which was originally identified by screening of a spinach leaf cDNA library (1) and was never isolated as a protein, could not be detected in our preparations. This indicates that either the protein is present in very low amounts or is not located in the plasma membrane of spinach leaves. Another explanation for this observation could also be that PM28B is only expressed in plasma membranes under certain, unknown stress conditions. Both isoforms were shown to be tetramers by STEM mass analysis. The measured mass of 157 Ϯ 56 kDa is in good agreement with the values determined for other detergentsolubilized MIP homologs by STEM (17,28,45). The total experimental error of Ϯ 8 kDa is explained by counting statistics of the scattered electrons and calibration errors. Thus, the ϳ3.5-kDa mass difference for the tetramer of the two isoforms calculated from the MALDI-TOF measurement was not resolved (Fig. 4B). However, electron microscopy of negatively stained OTG-solubilized PM28 isoforms revealed two different populations of particles. The most abundant form consisted of the larger and rather circular tetramers, and the less abundant form consisted of the smaller and almost square projections with side lenghts of 8.8 and 8.0 nm, respectively. Because particles from digitized electron micrographs were picked automatically (see "Experimental Procedures"), the number of FIG. 6. Atomic force microscopy of PM28 isoforms after reconstitution into lipid bilayers. The tendency of the two isoforms to segregate into two different crystal forms is shown. A, the topography of the less abundant crystals can be assigned to the smaller and almost square particles found initially by single-particle analysis of negatively stained PM28 (see Fig. 5C). This isoform crystallizes with a p42 1 2 symmetry and lattice vectors of a ϭ b ϭ 9.76 Ϯ 0.06 nm. B, the larger and rather circular particles compatible with those in Fig. 5B can be recognized in the more abundant hexagonal crystal with lattice vectors of a ϭ b ϭ 8.58 Ϯ 0.18 nm. The insets display well preserved particles. Imaging buffers, 10 mM Tris-HCl (pH 8. particles used to calculate the averages in Fig. 5, B and C was a direct indication for their distribution on the grid. The ratio of the smaller and almost square to the larger and rather circular particles was 0.6 as estimated from single-particle classification. This estimate is in excellent agreement with the ratio of the two peaks observed by MALDI-MS (Fig. 7B) of 0.6 Ϯ 0.1 (n ϭ 13; m/z 29839:m/z 30683). Assuming similar desorption behavior during ionization in the mass spectrometer and similar adsorption to the carbon film of the transmission electron microscopy grid for the two PM28 isoforms, these data indicate that the smaller and almost square particles correspond to the isoform yielding a peak at m/z 29839. According to the cDNA sequence, this isoform is PM28A, which has a calculated average mass of 29931 (1).
Crystal formation was observed after reconstitution of PM28 into lipid bilayers. Structural analysis by AFM clearly showed that the two PM28 isoforms segregate into two different crystal types; amazingly, mixed crystals were never observed. In addition, the structural features indicated by single-particle analysis were confirmed. The smaller and almost square tetramers built highly ordered 2D arrays with a p42 1 2, whereas the larger and rather circular particles formed less well ordered hexagonal arrays. The quality of the two crystals is reflected by the standard deviations of the unit cell dimensions (p42 1 2, a ϭ b ϭ 9.76 Ϯ 0.06 nm; hexagonal lattice, a ϭ b ϭ 8.58 Ϯ 0.18 nm), which was three times higher for the hexagonal crystal. The overall shape of the particles forming the hexagonal arrays was more round than square. Although corners were often observed on AFM topographs recorded at high magnification ( Fig. 6B  and inset), there was considerable flexibility in the shape of these particles. This structural inhomogeneity, on the one hand, enabled the formation of a hexagonal crystal, which is unusual for aquaporins, and on the other hand, hindered the growth of highly ordered crystals.
MALDI-MS and more recently electrospray ionization-MS have been applied to assess the masses of full-length membrane proteins with high accuracy (46 -48). We have used a similar approach to test the purity of our reconstituted PM28 into 2D crystals. MALDI-MS analysis of entire aquaporins reconstituted into lipid bilayers yielded very sharp, symmetrical peaks and small standard deviations when the m/z values were averaged over several spectra. Single isoforms differing by only 844 Da could easily be resolved in the spectra. Broad, asymmetrical peaks in the MALDI spectra arise from formylated states of the protein. Thus, it was crucial to keep the incubation time of the sample in formic acid as short as possible. A very interesting perspective for the future is to demonstrate whether the resolution achieved by this method is sufficient to detect differently phosphorylated states of whole aquaporins or other membrane proteins.
PIP1 and PIP2 aquaporins were isolated from spinach leaf plasma membranes and were shown to be differentially expressed. These findings suggest that differential expression may provide a means to regulate the water flux across the plasma membrane, in addition to the known mechanism of regulation by phosphorylation of the PIP2 aquaporins (24).