Live Cell Analysis of Aquaporin-4 M1/M23 Interactions and Regulated Orthogonal Array Assembly in Glial Cells*♦

Aquaporin-4 (AQP4) can assemble into supramolecular aggregates called orthogonal arrays of particles (OAPs). In cells expressing single AQP4 isoforms, we found previously that OAP formation by AQP4-M23 requires N terminus interactions just downstream of Met-23 and that the inability of AQP4-M1 to form OAPs involves blocking by residues upstream of Met-23. Here, we studied M1/M23 interactions and regulated OAP assembly by nanometer-resolution tracking of quantum dot-labeled AQP4 in live cells expressing differentially tagged AQP4 isoforms and in primary glial cell cultures in which native AQP4 was labeled with a monoclonal recombinant neuromyelitis optica autoantibody. OAP assembly was assessed independently by Blue Native gel electrophoresis. We found that OAPs in native glial cells could be reproduced in transfected cells expressing equal amounts of AQP4-M1 and -M23. Mutants of M23 that do not themselves form OAPs, including M23-F26Q and M23-G28P, were able to fully co-associate with native M23 to form large immobile OAPs. Analysis of a palmitoylation-null M1 mutant (C13A/C17A) indicated palmitoylation-dependent OAP assembly only in the presence of M23, with increased M1 palmitoylation causing progressive OAP disruption. Differential regulation of OAP assembly by palmitoylation, calcium elevation, and protein kinase C activation was found in primary glial cell cultures. We conclude that M1 and M23 co-assemble in AQP4 OAPs and that specific signaling events can regulate OAP assembly in glial cells.

Aquaporin-4 (AQP4) 2 is the predominant water channel in the mammalian central nervous system, where it is localized mainly in endfoot processes of pericapillary glial cells and in ependymal cells lining the ventricles (1,2). AQP4 provides the principle pathway for water entry into the brain and spinal cord through an intact blood-brain barrier in cytotoxic edema (3)(4)(5) and the pathway for removal of excess brain water in vasogenic edema and hydrocephalus (6,7). In addition to its central role in maintaining in brain water balance, AQP4 is also involved in neuroexcitation (8,9) and glial cell migration (10,11). AQP4 also acts as a receptor for autoantibodies found in serum from patients with the severe demyelinating autoimmune disease neuromyelitis optica (NMO) (12,13).
Freeze-fracture electron microscopy of glial cell endfoot membranes has revealed densely packed square arrays of intramembrane proteins known as orthogonal arrays of particles (OAPs) (14,15). Our laboratory first proposed that AQP4 forms OAPs based on its expression in the specific cell types where OAPs are seen (16). The involvement of AQP4 in OAP formation was proven by showing OAPs in AQP4-transfected cells (17), the absence of OAPs in tissues from AQP4 knock-out mice (18), and OAP labeling by AQP4 antibodies (2). The biological relevance of AQP4 assembly into OAPs is at present unknown. It has been speculated that OAPs might enhance AQP4 water permeability (19 -21), stabilize AQP4 polarization in glial cell foot-processes (22), and enhance glial cell-cell adhesion (23). The presence of OAPs has been correlated with neurological and neuromuscular diseases (24,25), and a recent study suggested that OAPs contribute to the pathogenesis of NMO (26).
The AQP4 transcript contains at least two alternate translation initiation sites (27), yielding a "long" (M1) isoform of ϳ34 kDa, and a "short" (M23) isoform of ϳ31 kDa (28 -30) (Fig. 1A). Both AQP4 isoforms form stable tetramers and function as water channels, with brain expressing both isoforms and tissues outside of the central nervous system expressing predominantly AQP4-M23 (16,31). A recent study of the rat brain genome suggested the presence of additional AQP4 isoforms, including a larger 39-kDa isoform (AQP4e) with 41 extra residues at the AQP4-M1 N terminus that forms functional water channels but is expressed in very low abundance (32). Freezefracture electron microscopy of cells transfected with AQP4-M23 shows large OAPs, sometimes containing Ͼ100 particles, whereas OAPs are largely absent in cells transfected with AQP4-M1 (33). OAP size is reduced when AQP1-M1 is coexpressed with AQP4-M23, suggesting an interaction between these two isoforms in vivo (21,33).
We have developed fluorescence methods to visualize OAPs in live cells to define the determinants of OAP formation. Our approach is to follow the diffusion of individual AQP4 tetramers in cell membranes after labeling a small subset of AQP4 molecules with quantum dots, which allows precise, rapid tracking with spatial precision of better than 20 nanometers. In an initial study (34), we introduced a Myc epitope into an extra-cellular loop of AQP4 for quantum dot labeling and showed free, rapid diffusion of AQP4-M1 and slow, highly restricted diffusion of OAP-associated AQP4-M23. Analysis of single molecule trajectories allowed determination of the fraction of AQP4 molecules in OAPs and their diffusion characteristics. We also investigated the molecular determinants of OAP formation by measuring the diffusion and assembly state of a series of AQP4 truncation mutants, point mutants, and chimeras (35). By expressing single AQP4 isoforms, we found that OAP formation by AQP4-M23 requires N-terminal hydrophobic interactions at residues 24 -26 (just downstream of Met-23) and that the absence of OAP formation by AQP4-M1 involves non-selective blocking of these hydrophobic interactions by seven residues just upstream of Met-23. In support of the conclusion that OAP formation by AQP4-M23 is stabilized by hydrophobic intermolecular interactions involving N-terminal residues, we found that AQP1, an AQP4 homolog that does not form OAPs, could be immobilized when its N-terminal domain was replaced with that of AQP4-M23.
Here, we investigated a different set of questions about whether and how the M1 isoform of AQP4 can disrupt M23 OAPs and whether AQP4 OAP assembly is regulated in native brain glial cells by lipid modification or phosphorylation. For these studies, a two-color tracking approach was developed in which two co-expressed AQP4 isoforms were labeled with different color quantum dots. We also developed an approach to track native AQP4 molecules in primary glial cell cultures using a purified, recombinant antibody against an extracellular epitope(s) on native AQP4. The single-molecule tracking data together with Blue Native gel electrophoresis suggest that M1 and M23 co-assemble in AQP4 tetramers with differential abilities to further assemble into OAPs. We also show that OAP assembly in glial cells can be regulated by post-translational modification of N-terminal cysteines in AQP4-M1 and by phosphorylation signaling.

EXPERIMENTAL PROCEDURES
DNA Constructs, Cell Culture, and Transfections-DNA constructs encoding rat AQP4 (M1 and M23 isoforms and mutants thereof) and human AQP1, each containing a 10-residue Myc epitope (NH 2 -EQKLISEEDL-COOH) in the second extracellular loop, were previously described (35). M23 with a hemagglutinin (HA) epitope (NH 2 -YPYDVPDYA-COOH) in place of the Myc epitope was generated by PCR amplification using rat M23 as template. A C-terminal serine residue was included to introduce a cloning restriction site and to generate the epitope tags of equal length (Fig. 1A). Constructs encoding non-epitope-tagged AQP4-M1 and M23 from rat, mouse, and human were PCR-amplified using whole-brain mRNA as template. All PCR fragments were ligated into mammalian expression vector pcDNA3.1 and fully sequenced.
Primary cultures of astrocytes (glial cells) from the neocortex of wild-type and AQP4 knock-out mice (36) were generated and grown as previously described (10). Primary astrocyte, COS-7 (ATCC #CRL-1651), and HEK293 (ATCC #CRL-1573) cell cultures were maintained at 37°C in 5% CO 2 , 95% air in appropriate medium containing 10% fetal bovine serum, 100 units/ml penicillin and 100 g/ml streptomycin. COS-7 cells were grown on 18-mm glass coverslips, and glial cells were grown in 35-mm glass-bottom dishes coated with poly-D-lysine (MatTek, Ashland, MA) for microscopy. Cells were transfected with DNA in antibiotic-free medium 12-24 h before the experiment using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol.
NMO Patient Serum and Recombinant AQP4 Autoantibody-NMO serum was obtained from an NMO-IgG seropositive individual who met the revised diagnostic criteria for clinical disease (37). Control (non-NMO) human serum was purchased from the University of California, San Francisco cell culture facility. Recombinant NMO antibody (rAb-53) was generated from a clonal-expanded cerebrospinal fluid plasma blast clone as described by Bennett et al. (38). Briefly, heavy-and lightchain constructs were co-transfected into HEK293 cells, and the supernatant was harvested, centrifuged to remove any cells and debris, and then incubated overnight with protein A-Sepharose (Sigma) at 4°C. The rAb was eluted in 0.1 M glycine, 1 M NaCl (pH 3.0) and immediately adjusted to pH 7.5 by the addition of 0.1 M Tris-HCl (pH 8.0). Recombinant IgG was subsequently exchanged and concentrated in storage buffer (PBS plus 0.1% IgG/protease-free bovine serum albumin) using Ultracel YM-30 microconcentrators (Millipore, Billerica, MA). Antibody integrity was confirmed by gel electrophoresis, and IgG concentration was determined by a human IgG capture enzyme-linked immunosorbent assay. Purified rAb-53 showed specific binding to human and mouse AQP4 in multiple independent assays (38).
Immunostaining-COS-7 cells transfected with Myc-and/or HA-tagged AQP4 isoforms were fixed with 4% paraformaldehyde and incubated for 20 min in blocking buffer (PBS containing 6 mM glucose, 1 mM pyruvate, 1% bovine serum albumin and 2% goat serum) and then for 20 min in 70 g/ml mouse anti-Myc (Covance, Emeryville, CA) and rabbit anti-HA (Zymed Laboratories Inc., South San Francisco, CA) antibodies in blocking buffer. Cells were then rinsed and incubated another 20 min with 10 g/ml goat anti-mouse IgG-conjugated Alexa Fluor 488 and goat anti-rabbit IgG-conjugated Alexa Fluor 555 (Invitrogen) in blocking buffer. For staining of non-epitopetagged AQP4 isoforms in primary glial cells and transfected COS-7 cells, live cells were incubated in blocking buffer for 10 min followed by 20 min in either 50 g/ml rAB-53, 1% NMO patient serum, or 1% control (non-NMO) human serum in blocking buffer. Cells were then rinsed extensively with PBS, fixed in 4% paraformaldehyde, and permeabilized with 0.25% Triton X-100. Cells were blocked again and incubated for 20 min with 2 g/ml rabbit anti-AQP4 antibody (Santa Cruz Biotechnology, Santa Cruz, CA), then rinsed with PBS. Finally, cells were incubated for 20 min with 10 g/ml goat anti-human IgG-conjugated Alexa Fluor 488 and goat anti-rabbit IgG-conjugated Alexa Fluor 555 (Invitrogen) in blocking buffer. After incubation with secondary antibodies, cells were rinsed extensively in PBS, and cover glasses were mounted with Vecta-Mount (Vector Laboratories, Burlingame, CA) containing 4Ј,6diamidino-2-phenylindole to stain nuclei.
Quantum Dot Labeling and Cell Treatments-Before labeling of AQP4 with quantum dots (Qdots), cells were washed with 2 ml of PBS containing 6 mM glucose and 1 mM pyruvate (GP buffer) and incubated for 5 min in blocking buffer. For two-color labeling of COS-7 cells co-transfected with Myc-and HA-tagged AQP4 isoforms, cells were incubated for 8 min with 70 ng/ml mouse anti-Myc (Covance) and rabbit anti-HA (Zymed Laboratories Inc.) antibodies in blocking buffer. Cells were then rinsed 5 times with GP buffer and incubated for 5 min with 0.1 nM goat F(abЈ) 2 anti-mouse IgG-conjugated Qdot 605, and goat F(abЈ) 2 anti-rabbit IgG-conjugated Qdot 655 (Invitrogen) in blocking buffer. Qdot labeling of non-epitope-tagged AQP4 isoforms in COS-7 or glial cell cultures was done using the same method. Cells were incubated for 8 min in 5 g/ml rAb-53 in blocking buffer, rinsed 5 times with GP buffer, and incubated the 5 min with goat F(abЈ) 2 anti-human IgG-conjugated Qdot 655 (Invitrogen) in blocking buffer. After incubation with Qdots, cells were rinsed extensively and maintained throughout experiments in GP buffer.
In some experiments S-palmitoylation was disrupted by incubation for specified times with 100 M 2-bromopalmitate (Sigma) in cell culture medium beginning 6 -12 h after transfection. To produce Ca 2ϩ -or kinase-mediated signaling, cells were incubated with 1 M thapsigargin (Sigma), 20 M forskolin (Sigma), or 0.2 M phorbol 12-myristate-13-acetate (Sigma) in GP buffer for 10 min at 37°C after Qdot labeling just before measurements.
Single Particle Tracking-SPT was done on a Nikon Eclipse TE2000S inverted epifluorescence microscope (Nikon, Melville, NY) equipped with a Nikon 100ϫ TIRF oil immersion objective (numerical aperture 1.45) and a deep-cooled CCD camera (Hamamatsu EM-CCD, Bridgewater, NJ), with additional technical specifications described previously (35). Qdot fluorescence was excited using an E460SPUV excitation filter and 475DCXRU dichroic mirror and detected through D605/ 40m or D655/40m emission filters (Chroma, Rockingham, VT). Data were acquired continuously at 11 ms per frame (91 Hz) for 6 s at 37°C within 30 min of the final wash step. For 2-color SPT, cells expressing many Qdots of each color were identified, then two successive 6-s acquisitions were made at each wavelength set.
Image sequences were analyzed, and trajectories were constructed as described previously (39). Briefly, the mean-squared displacement as a function of time ͗r 2 (t)͘ was constructed for each trajectory, and the diffusion coefficient D and offset (due to noise) were determined by a linear fit of the first three time steps of the mean-squared displacement, The offset was subtracted from each point, and the first quarter of the mean-squared displacement curve was fitted using a weighted nonlinear least-squares fitting algorithm (39). The range of an individual particle at selected time t was computed as range͑t͒ ϭ ͗r 2 ͑t͒͘ fit Only trajectories greater than 200 steps were analyzed, with data sets composed of at least 200 trajectories from at least 20 different cells. Diffusion data are primarily reported in the form of the cumulative distribution of ranges at 1 s, where P(range) is defined as the probability that a particle range is less than or equal to a given distance at t ϭ 1 s. The fraction of M23 and mutants assembled in OAPs was estimated as P(79 nm), the range at which P(range) ϭ 0.95 for pure M23 in COS-7 cells, as described (35).
Histograms of ranges of native AQP4 in glial cell cultures were fitted to a sum of distributions, each of the form f, where x is the range at 1 s, and ⌫ is the gamma function. In fitting the sum, shape parameters ␣ and ␤ in the component corresponding to tetramer (non-OAP) diffusion were fixed at those values determined for pure M1. The remaining fitted parameters describe the distribution of AQP4 diffusion in OAPs. Statistical significance (p value) for differences between distributions after deconvolution was determined using Pearson's 2 test. Electrophoresis and Immunoblotting-Whole brains were removed from a euthanized adult mouse and a juvenile rat. A 1-cm 3 fragment of human adult male brain cortex was obtained from the University of California, San Francisco tissue core. Brain tissues were homogenized in 10 mM Tris buffer containing 250 mM sucrose and protease inhibitors. Homogenized brain as well as COS-7 and glial cell cultures were lysed with NativePAGE sample buffer (Invitrogen) containing 0.5% dodecyl-␤-D-maltoside (EMD Chemicals, Gibbstown, NJ) for 10 min on ice. Lysates were centrifuged at 20,000 ϫ g for 30 min at 4°C, and the pellet was discarded. Total protein content was determined by Bradford assay. For Blue Native gel electrophoresis (BN-PAGE), 1-5 g of protein was mixed with Coomassie G-250 at a detergent:Coomassie ratio of 4:1, loaded onto a NativePAGE 3-12% Bis-Tris gel along with NativeMARK molecular weight markers, and run with NativePAGE running buffers according to the manufacturer's protocol (Invitrogen). For SDS-PAGE, the same cell lysates were denatured with NuPAGE lithium dodecyl sulfate sample buffer, loaded onto a NuPAGE 12% Bis-Tris gel along with SeeBlue Plus2 molecular weight markers, and run with NuPAGE MES SDS running buffer at 200 V (Invitrogen). Proteins were blotted onto polyvinylidene difluoride membranes (Millipore). After transfer, native proteins were fixed by soaking membranes for 15 min in 8% acetic acid and destained with methanol. Membranes were blocked with 5% nonfat milk and incubated with rabbit anti-AQP4 (Santa Cruz), rabbit anti-Myc (Santa Cruz), or rabbit anti-HA (Zymed Laboratories Inc.) primary antibodies at 4°C overnight. Membranes were then rinsed, incubated for 1 h with horseradish peroxidase-conjugated goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories, West Grove, PA), rinsed extensively, and labeled proteins were detected by ECL Plus enzymatic chemiluminescence kit (Amersham Biosciences).

Aquaporin-4 Orthogonal Arrays
of M1 and M23 produced smaller OAPs than transfection with M23 alone. However, a limitation of these prior studies to characterize M1/M23 interactions has been the inability to distinguish M1 from M23 in the membrane. To overcome this limitation, we generated AQP4 constructs containing Myc or HA epitopes inserted in the second extracellular loop of AQP4 (Fig.  1A). We found previously that insertion of a Myc tag at this position did not affect AQP4 plasma membrane trafficking or water channel function (34). Fig. 1B shows that epitope-targeted immunofluorescence of AQP4 in transfected COS-7 cells is highly specific. For two-color SPT, we labeled Myc-tagged M1 (M1.Myc) with Qdots with emission maximum at 605 nm and HA-tagged M23 (M23.HA) with Qdots emitting at 655 nm. As expected from our previous results, M1 diffusion was rapid and free, whereas M23 diffusion was highly confined due to OAP formation (Fig. 1C). Mean-squared displacement plots (Fig. 1D) and cumulative probability plots of range at 1 s (Fig.   1E) show that AQP4 diffusion was independent of Qdot color and epitope tag, with diffusion of M23.HA indistinguishable from that of M23.Myc.
Having established the two-color SPT method, we measured the diffusion of AQP4-M1 and M23 isoforms in COS-7 cells expressing AQP4 at different M23-to-M1 ratios. Upon co-expression with M1, the diffusion of M23 increased significantly, whereas expression of M23 produced a marked reduction in M1 diffusion ( Fig. 2A). When co-expressed, both isoforms exhibited very similar diffusion, indicating a high level of interaction. Quite different from the experimental data are the dashed curves in Fig. 2B, which are the predicted curves if M1 and M23 do not co-mingle (summed distributions for pure M1 and pure M23 weighted by their relative amounts). The broad distributions seen when M1 and M23 are co-expressed likely correspond to a broad size distribution of OAPs containing M1 and M23 (Fig. 2B).  DECEMBER 18, 2009 • VOLUME 284 • NUMBER 51 BN-PAGE was done to further investigate the co-mingling of M1 and M23 (Fig. 2, C and D). In cells expressing M23 alone, a large diffuse band migrating at Ͼ1200 kDa was visible by AQP4 immunoblot (Fig. 2C, top) along with a band at ϳ300 kDa corresponding to AQP4 tetramers. With increasing M1, the larger bands disappear, and the smaller bands and the tetramer band become more prominent. SDS-PAGE (lower panel) confirmed that the altered M23-to-M1 transfection ratio produced approximately proportionate changes in M1 and M23 protein expression. To identify AQP4 isoforms individually on native gels, BN-PAGE was done using Myc and HA antibodies to probe immunoblots (Fig. 2D). As probed by anti-HA antibody to visualize M23 (left panel), a reduced M23-to-M1 ratio resulted in the appearance of M23-containing OAP bands of smaller molecular size. Probing with anti-Myc antibody to visualize M1 (right panel) showed M1 presence in the OAP bands as well as more M1 in the tetramer band with a reduced M23to-M1 ratio. M1, therefore, partitions at least partially into OAPs in the co-transfected cells.  (40) found that the palmitoylation-null mutant M1-C13A/C17A, when expressed alone in transfected cells, formed OAPs in freeze-cleaved membranes. We found recently that this mutant expressed alone does not form OAPs in live cells at 37°C but can form OAPs at reduced temperatures (41).

Aquaporin-4 Orthogonal Arrays
The involvement of S-palmitoylation in OAP assembly was found only at reduced temperature, where incubation of native M1 with 2-bromopalmitate (BrPA) resulted in OAP formation at 10°C. We postulated that palmitoylation is only partially destabilizing to OAPs, and although it has no effect on the diffusion of M1 (when expressed alone) at 37°C, it might affect the ability of M1 to disrupt OAPs (at 37°C) when M1 and M23 are coexpressed, as in native glial cell membranes. Fig. 3A shows the effect of (de)palmitoylation of M1 on the diffusion of AQP4 in the presence of equimolar M23 at 37°C. Maximum effect was achieved after 8 -16 h of BrPA exposure, which resulted in slowed diffusion of both M1 and M23. BrPA had no effect on the diffusion of M1 or M23 when each was expressed alone (Fig. 3B). We then tested the ability of M1 cysteine mutants to disrupt M23 OAP formation. Single cysteine mutants M1-C13A and M1-C17A had a similar ability to disrupt M23 OAPs as native M1 (Fig. 3C), and as seen with native M1, BrPA treatment reduced this ability (Fig. 3D). After exposure to BrPA, the median range of AQP4 displacement at 1 s decreased from 84 to 49 nm with M1-C13A and from 87 to 43 nm with M1-C17A, as compared with a reduction from 130 to 48 nm with native M1 (Fig. 3E). These data implicate palmitoylation as a critical factor in M1-mediated regulation of OAP formation. As a key control, SPT was done in cells co-expressing M23 with the double cysteine (palmitoylation-null) mutant M1-C13A/C17A (Fig. 3C, bottom) at 37°C. When expressed alone, the diffusion of M1-C13A/C17A was nearly identical to that of native M1. Remarkably, when co-expressed with M23, the M1-C13A/ C17A mutant became largely immobilized, as seen for M23 alone. The palmitoylation-null mutant of M1 is, thus, unable to disrupt OAP formation by M23 but, instead, fully associates with M23 in large immobile OAPs.
We previously characterized point mutants in the AQP4-M23 N terminus with diminished ability to form OAPs (35,41). When expressed in COS-7 cell membranes, Ͻ20% of M23-F26Q and Ͻ10% of M23-G28P tetramers assembled into arrays at 37°C, whereas nearly all (non-mutated) M23 is present in assays. Interestingly, like the double cysteine (palmitoylationnull) mutant of M1 studied above, the M23-F26Q and M23-G28P M23 mutants completely lack the ability to actively disrupt array formation but fully assemble in immobile OAPs when co-expressed with native M23 (Fig. 4, A and B). As a control, we tested the effect of a non-OAP forming aquaporin, AQP1, when co-expressed with native AQP4-M23. As antici-

Aquaporin-4 Orthogonal Arrays
pated, AQP1 did not disrupt OAP formation, with M23 showing the same highly confined diffusion. However, unlike the AQP4 mutants, AQP1 did not co-associate with M23 in OAPs, as its diffusion was only slightly slowed in the presence of M23 (Fig. 4C), which is probably a consequence of molecular crowding in the membrane. Fig. 4D summarizes the percentage of AQP4 tetramers in OAPs for each of these sets of measurements.

Diffusion of Native AQP4 in Primary Glial Cell Cultures Is Similar to That of Co-expressed M1 and M23 AQP4 in COS-7
Cell Cultures-Although expression of epitope-tagged AQP4 isoforms and mutants in cell models has been useful in elucidating the structural and cellular requirements for OAP assembly, it is important to test key conclusions using native AQP4 in glial cells. We labeled native AQP4 in primary glial cell cultures by exploiting the discovery that patients with the disease NMO produce autoantibodies (NMO-IgG) against the extracellular epitope(s) on native AQP4 (13). A monoclonal recombinant NMO antibody (rAb-53) (38) that labels both M1 and M23 was selected for these studies. Other recombinant antibodies showed differential M1 versus M23 selectivity (data not shown). Fig. 5A shows that the rAb-53 stained native M23 and M1 from mouse, rat, and human. Likewise, whole serum from a patient with NMO also stained both isoforms. This finding contradicts a recent study reporting exclusive staining of M23 by NMO-IgG (26). rAb-53 also stained AQP4 on the surface of primary glial cell cultures from mouse neonatal brain cortex (Fig. 5B). BN-PAGE confirmed the presence of OAPs in mouse glial cells, with a similar pattern to that seen in whole brain homogenates (Fig. 5C, top). SDS-PAGE indicated an approximate 1:1 ratio of M23-to-M1 protein in the glial cell cultures (Fig. 5C,  bottom). Fig. 6A, left, shows SPT of native (non-epitope-tagged) AQP4 isoforms in transfected COS-7 cells, in which Qdot labeling was done with rAb-53. The expected distributions of diffusion range were found for M23 and M1. As an additional control, we found that the diffusion of native mouse M23 and M1, when transiently transfected into primary glial cell cultures from AQP4 knock-out mice, was similar to that seen in COS-7 cells (Fig. 6A, right). Notably, AQP4 diffusion in (nontransfected) glial cell cultures from wild-type mice was similar to that seen in COS-7 cells that were transfected with a 1:1 M23-to-M1 ratio (Fig. 6, A and B), supporting the conclusion that native M23 and M1 interact strongly in live glial cells at 37°C.
We developed a numerical method to resolve in native glial cells the number of AQP4 molecules in OAPs versus non-OAPs and the apparent size distribution of OAPs. This method is an extension of the threshold range analysis, as done in Fig. 4 and in prior studies (35,41), which provides a good description of one-component systems. However, a simple threshold analysis provides limited biological insight in complex systems where M1 and M23 co-expression produce broad distributions of AQP4 diffusion range as in Figs. 2 and 6A. We began by examining the distributions of pure M23 and M1 in transfected glial cell cultures from AQP4 knock-out mice. As expected for SPT data, the populations obey a gamma distribution (42) (Fig. 6C). The distribution of ranges for native AQP4 in wild-type glial cells fits well to a sum of two gamma distributions (Fig. 6D, left), one of which was fixed to the canonical M1 distribution corresponding to free AQP4 tetramers (see "Experimental Procedures"). The second gamma distribution in the sum represents the diffusion of AQP4 in OAPs. As summarized in Fig. 6D, right, there is a significant shift (p Ͻ 0.01) in the distribution of AQP4 in OAPs in wild-type glial cell cultures when compared with  DECEMBER 18, 2009 • VOLUME 284 • NUMBER 51 JOURNAL OF BIOLOGICAL CHEMISTRY 35855 pure M23 OAPs, likely corresponding to a shift to smaller OAP size. Integration of the two distributions indicated that 51 Ϯ 4% of AQP4 molecules were confined within OAPs in glial cell cultures from wild-type mice.

Aquaporin-4 Orthogonal Arrays
Regulated OAP Assembly in Primary Glial Cell Cultures-We tested in glial cell cultures the key prediction from M1/M23-co-transfected COS-7 cells that post-translational palmitoylation of M1 affects OAP assembly. Incubation of glial cell cultures overnight with BrPA produced a marked shift of the entire distribution of AQP4 diffusion to a lower range (Fig.  6E, top). Mathematical deconvolution of the OAP and non-OAP populations as described above indicated an increase to 81 Ϯ 7% of AQP4 molecules within OAPs after treatment with BrPA and a small although significant (p Ͻ 0.01) increase in the distribution of OAP size, with the median range of the OAP fraction decreasing from 51 to 40 nm (Fig. 6, F and G). Palmitoylation, thus, affects OAP assembly in native glial cells.
We further tested whether OAP organization in primary glial cell cultures can be modified by the major cellular signaling pathways. A wide range of inter-and intracellular events in glial cells are regulated by Ca 2ϩ oscillations and propagation of Ca 2ϩ waves (43). Thapsigargin is a cell-permeable inhibitor of the sarco-endoplasmic reticulum Ca 2ϩ -ATPase that increases cytosolic Ca 2ϩ concentration in all cell types including glial cells. Incubation of Qdot-labeled glial cell cultures with thapsigargin produced a modest increase in AQP4 diffusion (Fig. 6E,  bottom), with the median range at 1 s increasing by 44% from 109 to 157 nm. We also tested activators of two major phosphorylation pathways; forskolin, which increases cyclic adenosine monophosphate and thereby activates protein kinase A, and phorbol 12-myristate-13-acetate (PMA), which activates protein kinase C. AQP4 contains one consensus protein kinase A phosphorylation site at Ser-111 and two protein kinase C sites at Thr-107 and Ser-180 (Fig. 1A). Although forskolin did not affect AQP4 diffusion, PMA produced a comparable effect to that of thapsigargin, with an increase in median range to 154 nm at 1 s (Fig. 6E, bottom). Interestingly, mathematical deconvolution of the distributions did not show significant changes (p Ͼ 0.05) in the apparent OAP size distribution with thapsigargin or PMA (Fig. 6F, bottom) but instead showed a 20% decrease in the overall fraction of OAP-assembled AQP4 tetramers (Fig. 6G). Thapsigargin, forskolin, or PMA had no effect on AQP4 diffusion in cells transfected with M1 or M23 alone (data not shown). Possible mechanisms of Ca 2ϩ -and protein kinase C-mediated regulation of AQP4 OAPs are discussed below.

DISCUSSION
Strong M1/M23 Interactions Are Responsible for Regulation of OAP Size-As mentioned above, co-expression of M1 with M23 reduces the size of AQP4 OAPs when compared with those formed by M23 alone, and large OAPs of Ͼ 100 particles, which are routinely seen in M23-transfected cells, are rare in glial cell membranes (33). The nature of the putative interactions between M1 and M23 has been largely unknown. Using a dual-labeling approach, we conclude here that M1 interacts strongly with M23, probably forming M1/M23 heterotetramers, and that M1 effectively partitions into OAPs. We found here that the diffusion of Myc-tagged M1 and HA-tagged M23 became nearly identical upon co-expression over a range of M23-to-M1 molar ratios and that M1 partially migrated in OAP bands on native gels (Fig. 2).
The M23-to-M1 molar ratio had a strong effect on AQP4 diffusion (Fig. 2), likely reflecting a shift in the distribution of OAP sizes and the fraction of AQP4 molecules diffusing as free tetramers. Short-range diffusion of OAP-assembled AQP4 observed by SPT represents the summed movements of con-

Aquaporin-4 Orthogonal Arrays
fined oscillations by AQP4 tetramers and diffusion of whole AQP4 arrays of various sizes, with smaller OAPs likely diffusing faster than larger OAPs. We and others have found by freezefracture electron microscopy that M23-only OAPs can vary in size from Ͻ10 to Ͼ100 intramembrane particles (33,34). The distribution of ranges for pure M23 (Figs. 1D and 6A), thus, includes contributions from M23 diffusion in OAPs of different size. In Fig. 4 and in previous studies we used the range at which P(range) ϭ 0.95 for pure M23 as a threshold to estimate the fraction of AQP4 isoforms or mutants residing in OAPs under various conditions (35,41). Such analysis is useful in simple one-component systems having relatively narrow OAP size distributions, as found for various AQP4 mutants, but has limited utility in describing more complex systems where AQP4 diffuses as free tetramers as well as OAPs with a wide range of sizes. For such heterogeneous systems such as glial cell cultures we developed a new method based on a sum of gamma distributions (Equation 3) to deconvolve the broad distribution of AQP4 diffusion into free tetramer-only and OAP-only fractions. This approach provided useful information about whether selected maneuvers alter OAP size distribution and/or the fraction of AQP4 molecules within OAPs, as discussed further below.
Regulation of OAPs by S-Palmitoylation-A role for posttranslational lipid modification of AQP4-M1 in OAP formation was first proposed by Fujiyoshi and co-workers (40), who showed formation of OAPs in cells transfected with cysteine mutants of M1. We reported conflicting data indicating that M1 cysteine mutants diffuse identically to native M1 in live cells at 37°C (35). Recently, we were able to resolve these apparently contradictory results by showing that OAP formation by M1 cysteine mutants in live cells was temperature-dependent, with OAPs seen only at reduced temperature (41). By studying coexpressed M23 and M1 cysteine mutants here, we found that M1 palmitoylation does significantly alter AQP4 assembly at physiological temperature, not by altering the ability of M1 alone to form OAPs but, instead, by regulating the formation of mixed M1/M23 arrays. The physiological relevance of M1 palmitoylation on OAP regulation was shown here directly in primary glial cell cultures.
OAP formation is driven by hydrophobic interactions near the M23 N terminus, likely occurring at or near the plasma membrane. Prevention of OAP formation by M1 tetramers requires seven residues upstream from Met-23, although the specific sequence or palmitoylation state of these residues is not important (35). In a mixed AQP4 heterotetramer, the number of "active" N termini involved in OAP formation is dependent on the M1/M23 ratio, but in the absence of palmitoylatable cysteines M1 cannot modulate OAP formation, allowing growth into large immobile OAPs similar to those formed by M23 alone (Fig. 3C). We propose that in mixed heterotetramers the extra residues comprising the M1 N terminus actively block tetramer-tetramer contacts between adjacent M23 N termini, but to do so they must be anchored in the cytoplasmic face of FIGURE 5. A recombinant AQP4-specific antibody (rAb-53) stains native AQP4 within and outside of OAPs. A, immunofluorescence of COS-7 cells transfected with human, mouse, and rat M23 and M1 AQP4 is shown. Cell surface was stained (green) with rAb-53 (left), NMO patient serum (middle), or control serum (right). Also shown is the staining of fixed, permeabilized cells with C terminus-targeted anti-AQP4 (red). B, immunofluorescence shows surface staining of primary glial cell cultures from wild-type (left) and AQP4 knock-out (right) mice with rAb-53 (green). Nuclei were stained with 4Ј,6-diamidino-2-phenylindole (blue). Bar, 50 m. C, AQP4 immunoblots are shown after BN-PAGE (top) and SDS-PAGE (bottom) of lysates from cultured astrocytes from wild-type (ϩ/ϩ) and AQP4 knock-out (Ϫ/Ϫ) mice (left) and brain tissue lysates (right).
the plasma membrane near the site of interaction. Lipid modification provides the anchor, thereby allowing M1 to modulate OAP formation and growth.
The extent to which M1-C13A/C17A associated with OAPs when co-expressed with M23 was initially surprising. One might expect a significant fraction of AQP4 tetramers in the membrane to be M1-C13A/C17A homotetramers, which do not from OAPs on their own, and yet no population of freely diffusing M1-C13A/C17A was observed (Fig. 3C). However, we recently showed that M1-C13A/C17A homotetramers can form OAPs at reduced temperatures (41). Together, these data suggest that the additional residues in the M1 N terminus inhibit nucleation of M1 OAPs at 37°C. Meanwhile, lipid anchoring of the N terminus in the plasma membrane further prevents M1 homotetramers from associating with prenucleated OAPs. Removing the barrier to nucleation either by lowering the temperature or by co-expression of M23 allows nonpalmitoylated M1 homotetramers to assemble in OAPs.
S-Palmitoylation is a reversible modification that is highly regulated and involved in cellular signaling, but little is known about the mechanisms controlling the process (44). It remains to be determined whether and how glial cells actively regulate AQP4 assembly by altering the palmitoylation of M1.
Regulation of OAPs by Cellular Signaling- Fig. 6E shows that Ca 2ϩ -and protein kinase C-mediated signaling affect AQP4 assembly in glial cells. OAP assembly in live cells at 37°C is controlled by at least three factors: M1/M23 ratio, M1 palmitoylation, and the amount of AQP4 in the membrane. Signaling events could potentially affect each of these processes. Earlier work showed increased endocytosis of AQP4 after exposure to phorbol esters (45) and apparent rearrangements of OAPs after exposure to histamine, which elevates Ca 2ϩ (46). However, these previous experiments were performed in cells expressing M23 only, in which essentially all AQP4 is locked in large immobile arrays. It is unclear how phosphorylation of M23 alone could lead to OAP rearrangement. Indeed, we found no effect of PMA or thapsigargin in cells expressing only M23. In glial cells AQP4 exists in a dynamic state, with M1 appearing to be the major active modulator of OAP assembly. In mixed M1/M23 systems, preferential endocytosis of OAP-associated AQP4 could explain our observation of increased diffusion after treatment with thapsigargin or PMA. Nonspecific depletion of AQP4 from the membrane could have a similar effect because OAP assembly is driven by thermodynamic processes of mass action (41).
Biological Relevance of Regulated OAP Assembly-The potential biological consequences of regulated AQP4 assembly are largely unknown. M1 expression in glia is higher than in other cell types where AQP4 is found, suggesting that AQP4 assembly is more dynamically regulated in the central nervous system than in peripheral tissues, where M23 is the predominant isoform. Assembly in small-to-medium-sized OAPs that can diffuse in the membrane may allow glial cells to more efficiently sort and traffic large numbers of water channels to specific locations, such as endfoot processes or lamellipodia. Upon arrival, further assembly into large, immobile arrays could prevent diffusion away from these sites where high water permeability is required, without the need for continuous energy-dependent regulation.
Recently, much AQP4 research has focused attention on the inflammatory demyelinating disease NMO, which produces optic nerve and spinal cord lesions (47). The majority of NMO patients contain in their serum autoantibodies (NMO-IgG) recognizing AQP4 on the surface of glial cells (12). A recent study suggested that NMO-IgG specifically targets AQP4 OAPs, based on imaging data showing preferential staining of M23-transfected cells but not M1-transfected cells (26). This finding suggests that OAPs are the site of NMO pathogenesis.
We show here, however, that whole serum from an NMO patient as well as a monoclonal AQP4-specific recombinant antibody derived from an NMO patient (rAb-53) stains M1-transfected cells quite well (Fig. 5A). In general, serum from a single NMO patient contains many variants of NMO-IgG with differential abilities to stain AQP4 isoforms. The incorrect conclusion of the prior study that NMO-IgG stains only OAPs may be related to biased serum sampling or to inadequate sensitivity in detection of NMO-IgG staining of M1, perhaps because cellular M1 expression is uniform rather than punctate. The role, therefore, of AQP4 OAPs in the pathogenesis of NMO remains to be determined.
Conclusion-Our data from two-color SPT in transfected cells and SPT of native AQP4 in primary glial cell cultures have revealed new information about the nature of interactions between M1 and M23 isoforms and how these interactions modulate the assembly of AQP4 into OAPs. We found that M1 and M23 isoforms of AQP4 interact strongly, so much so that they are nearly indistinguishable when co-expressed in the same cell membrane. We confirmed suggestions based on direct evidence (48) that M1 co-associates with M23 in small OAPs. We found that AQP4 assembly is strongly dependent on the M1-to-M23 molar ratio. We also found that OAPs can be actively regulated in primary glial cell cultures by altering reversible S-palmitoylation of N-terminal cysteines in M1 or by Ca 2ϩ -and phosphorylation-mediated signaling. However, these active processes of regulation require co-expression of M1 and M23 isoforms, as no effect was found on the assembly of either isoform alone at 37°C. Our findings, thus, mandate a shift in thinking about AQP4 assembly in the central nervous system away from models concerning the behaviors of distinct M1 and M23 isoforms but, instead, to models representative of real glial cells in which M1 and M23 are co-expressed, and OAP formation is regulated through signaling processes that control AQP4 M1/M23 ratio, S-palmitoylation, and phosphorylation state.