A tyrosine‐based YXXΦ motif regulates the degradation of aquaporin‐4 via both lysosomal and proteasomal pathways and is functionally inhibited by a 10‐amino‐acid sequence within its C‐terminus

Aquaporin‐4 (AQP4) is a dominant water channel in the brain and is expressed on astrocytic end‐feet, mediating water homeostasis in the brain. AQP4 is a target of an inflammatory autoimmune disease, neuromyelitis optica spectrum disorders (NMOSD), that causes demyelination. An autoantibody recognizing the extracellular domains of AQP4, called NMO‐IgG, is critically implicated in the pathogenesis of the disease. Complement‐dependent cytotoxicity (CDC) and antibody‐dependent cellular cytotoxicity (ADCC) in astrocytes are the primary causes of the disease, preceding demyelination and neuronal damage. Additionally, some cytotoxic effects of binding of NMO‐IgG to AQP4, independent of CDC/ADCC, have been proposed. Antibody‐induced endocytosis of AQP4 is thought to be involved in CDC/ADCC‐independent cytotoxicity induced by the binding of NMO‐IgG to AQP4. To clarify the mechanism responsible for antibody‐induced endocytosis of AQP4, we investigated the subcellular localization and trafficking of AQP4, focusing on its C‐terminal domain, by making a variety of deletion and substitution mutants of mouse AQP4. We found that a tyrosine‐based YXXΦ motif in the C‐terminal domain of AQP4 plays a critical role in the steady‐state subcellular localization/turnover and antibody‐induced endocytosis/lysosomal degradation of AQP4. Our results indicate that the YXXΦ motif has to escape the inhibitory effect of the C‐terminal 10‐amino‐acid sequence and be located at an appropriate distance from the plasma membrane to act as a signal for lysosomal degradation of AQP4. In addition to lysosomal degradation, we demonstrated that the YXXΦ motif also functions as a signal to degrade AQP4 using proteasomes under specific conditions.


Introduction
Aquaporin-4 (AQP4) is the main water channel of the central nervous system (CNS), exclusively expressed in perivascular and subpial astrocytic end-feet [1][2][3]. It plays a major role in water homeostasis in the brain [4]. Recently, AQP4 has been attracting great attention as a key molecule involved in the mechanism for waste clearance in the brain, known as the glymphatic system. Diminishing the function of AQP4 has been hypothesized to impair the system and lead to many age-dependent diseases of the central nervous system, such as Alzheimer's [5].
AQP4 is a six-transmembrane protein that functions as a tetramer [6,7] and has two dominant isoforms, M1 and M23, produced by differences in transcriptional start sites as well as leaky scanning and out-of-frame uORF and re-initiation mechanisms (reviewed in Ref. [8]). The difference between the two isoforms is that the M23 isoform lacks intracellular N-terminal 22 amino acids of M1 [2,[9][10][11]. A critical feature of AQP4 is the formation of supramolecular aggregates named orthogonal arrays of particles (OAPs) [12,13]. When either of them is solely expressed in a cell line, only M23 homotetramers form OAPs since the N-terminal 22 amino acids of M1 have an inhibitory function against OAP formation [14,15]. Under physiological conditions, these isoforms are expressed simultaneously, incorporated randomly into a functional tetramer and form relatively small OAPs as compared to cells expressing M23 alone, implying that the expression ratio of M1 to M23 defines the size of OAPs [14,15].
The role of the cytosolic C-terminal domain of AQP has been best documented in AQP2 expressed in principal cells of the kidney collecting duct, in which phosphorylation at specific serine residues, located in the C-terminal domain, in response to vasopressin stimulation regulates its exocytosis to the apical plasma membrane to concentrate urine [16,17]. Multiple sequences in the C-terminal domain of AQP4 play a role in polarized membrane distribution. The C-terminal three amino acids of AQP4 (321-SSV-323) are putative motifs that interact with proteins possessing postsynaptic density 95/disk large/zonula occludens protein-1 (PDZ) domain. a-syntrophin (a PDZ-domaincontaining protein)-dependent incorporation of AQP4 into the dystrophin complex has been observed in skeletal muscle and astrocytic end-feet [18][19][20][21]. In addition, Madrid et al. [22] demonstrated that a tyrosine-based YXXΦ (amino acids with a bulky hydrophobic residue are placed at Φ) motif (277-YMEV-280) and a dileucine motif (288-ETEDLI-293) are collaboratively responsible for the basolateral distribution of AQP4 in MDCK cells, a kidney epithelial cell line. They also demonstrated the role of the YXXΦ motif in endocytosis and degradation of AQP4 in lysosomes via interaction with adaptor protein (AP) complexes AP-2 and AP-3 [22]. We have previously demonstrated that replacing the C-terminal 65 amino acids of mouse AQP4 (mAQP4) (Fig. 1A, Lys 259 -Val 323 ) with the corresponding domain of mAQP1 (Fig. 1A, Phe 238 -Lys 269 ) influences the steady-state subcellular localization of AQP4 [23]. These observations indicate an essential role for the Cterminal domain in the subcellular localization and intracellular trafficking of AQP4.
The tyrosine-based YXXΦ motif is involved in the intracellular trafficking of membrane proteins by directly interacting with AP complexes AP-1, AP-2, AP-3 and AP-4, through their medium subunits l1, l2, l3 and l4, respectively [24]. AP-1, AP-2 and AP-3 are components of clathrin-coated vesicles [24,25]. AP-1 is responsible for the bidirectional transport between the trans-Golgi network and endosomes [25]. AP-2 localizes to the plasma membrane and mediates the rapid internalization of a wide range of membrane proteins, whereas AP-3 localizes to endosomes and is involved in protein transport from endosomes to lysosomes [25].
AQP4 is a target of inflammatory autoimmune disease of the CNS, neuromyelitis optica spectrum disorders (NMOSD), in which a unique autoantibody recognizing the extracellular domains of AQP4, named NMO-IgG [26,27], has been implicated in the pathogenesis of this disease. The binding of NMO-IgG induces complement-dependent cytotoxicity (CDC) and antibody-dependent cellular cytotoxicity (ADCC) on the surface of astrocytic end-feet, which has been considered the initial event of this disease [28][29][30]. The mechanism for demyelination and neuronal damage after the disruption of astrocytes remains to be elucidated; however, some indirect effects on oligodendrocytes and neurons have been proposed. One such mechanism involves antibody-induced endocytosis of AQP4 accompanied by excitatory amino-acid transporter 2 (EAAT2), which increases extracellular glutamate concentration, followed by excitotoxicity to oligodendrocytes and neurons [31][32][33]. This observation prompted us to investigate the mechanism underlying antibody-induced endocytosis of AQP4.
In the present study, we examined the molecular basis of the mechanisms responsible for steady-state subcellular localization and degradation, as well as antibody-induced endocytosis of AQP4, focusing on its C-terminal domain, making a variety of deletion and substitution mutants of mAQP4.

Results
The intracellular C-terminal domain of AQP4 plays a role in endocytosis and trafficking to lysosomes after the binding of an antibody against the extracellular domains We have previously demonstrated that E5415A, a monoclonal antibody recognizing the extracellular   [7]. Amino acids possessing acidic, basic and hydroxyl groups are indicated in red, blue and green, respectively. (B) Schematic illustrations of the structure of cDNAs inserted into the expression constructs used in this study. (C-F) Effect of monoclonal antibodies against the extracellular domains of AQP4 on levels of wildtype mAQP4 (C, E, WT) and mAQP4 whose C-terminal domain has been replaced by that of mAQP1 (D, F, AQP1-C). Typical blots with anti-FLAG (top), anti-GFP (middle) and anti-actin (bottom) antibodies against lysates of cells expressing either WT (C) or AQP1-C (D) treated with vehicle (2% glycerol) for WT/human AQP4-specific C9401 for AQP1-C (lane 2), rodent-AQP4-selective E5415A (lane 3), E5415A in the presence of 100 nM bafilomycin A1 (Baf) (lane 4) and E5415A in the presence of 1 lM MG132 (lane 5) compared with that of nontreated cells (lane 1) are shown. Ratios of the mAQP4 amount and its derivatives (FLAG) to that of EGFP were calculated and shown as % of nontreated cells (E, F). Values are mean AE SEM of five independent experiments. ** (P < 0.01) and $$ (P < 0.01) indicate significant difference versus cells treated with vehicle and E5415A, respectively. domains of mAQP4, induces endocytosis and lysosomal degradation of the mAQP4 M1 isoform [34]. Moreover, replacing the C-terminal domain with the corresponding region of mAQP1 (Fig. 1A) influences the steady-state subcellular localization of AQP4 [23]. These observations suggest that the C-terminal domain of AQP4 plays a role in its trafficking to lysosomes after antibodies bind against its extracellular domains. To examine this assumption, we treated cells expressing mAQP4 M1 with the C-terminal domain of AQP1 (AQP4/AQP1-C) and wild type (AQP4/WT) with a monoclonal antibody, E5415A, for 24 h. The amount of the protein was measured by western blotting. To quantitatively detect each AQP4 derivative, we added a FLAG tag to its N-terminus (Fig. 1B). As observed in a previous study, in which a CHO-cell clone stably transfected with mAQP4 M1 was used [34], we detected a significant reduction in AQP4/WT after binding of E5415A ( Fig. 1C (lane 3), E), which was protected by bafilomycin A1 (Baf), an inhibitor of lysosomal function (Fig. 1C (lane 4), E). We also observed that MG132, an inhibitor of proteasomes showed an effect on the stability of AQP4 after treating with E5415A, although it was not statistically significant ( Fig. 1C (lane 5), E). Interestingly, AQP4/AQP1-C did not undergo degradation upon binding to E5415A (Fig. 1D (lane 3), F), suggesting the presence of a signal that regulates transportation to lysosomes in the C-terminal domain of AQP4.
The intracellular C-terminal domain of AQP4 is responsible for stabilizing and destabilizing AQP4 To investigate the role of the C-terminal domain in antibody-induced endocytosis and lysosomal transportation of AQP4, deletion mutants were made in mAQP4 M1 ( Fig. 2A). We transfected these constructs into two cell lines, CHO and HeLa and selected them with puromycin ( Fig.  2B,C, respectively). Unexpectedly, deleting 23 amino acids at the Cterminus (D301-323) destabilized AQP4 in both cell lines (Figs 2B,C (lane 3) and 5A). This was also the case with another deletion mutant, D292-323 ( Fig. 2B,C (lane 4)). However, further deletion of the C-terminal domain up to 53 amino acids (D271-323) re-stabilized the protein (Fig. 2B,C (lane 5)). These results suggest that a signal stabilizing AQP4 exists within the region between Lys 301 and Val 323 . Additionally, a signal destabilizing AQP4 is present between Gln 271 and Asp 291 . To further narrow down the regions, amino acids were deleted one by one from the C-terminus. As shown in Fig. 2D, the deletion of up to 10 amino acids from the C-terminus gradually destabilized AQP4, indicating that the signal responsible for stabilizing AQP4 is localized within this region. By contrast, the deletion of 44 amino acids (D280-323) or more stabilized AQP4, whereas AQP4 with a 43-amino-acid deletion (D281-323) remained destabilized, indicating that the signal responsible for destabilizing AQP4 is located immediately before Glu 281 (Fig. 2E).
To identify amino acids that contribute to the stabilization of AQP4, we introduced substitution mutations into full-length AQP4 (Fig. 3A). We focused on two serine residues (Ser at position 315 is Gln in human AQP4), which are sites for constitutive phosphorylation by protein kinase CK2 [35][36][37], and two acidic amino acids (Fig. 3A). Replacing both Asp 314 and Glu 318 with Ala significantly reduced the amount of AQP4 to the level seen in D314-323 ( Fig. 3B (lanes 4 and 6), D), although substituting Ala for either one of the two acidic residues had little effect on the stability of fulllength AQP4 (Fig. 3B (lanes 2 and 3), D). By contrast, replacing Ser 315 and Ser 316 residues with Gln and Ala, respectively, did not affect the stability of full-length AQP4 (Fig. 3B (lane 5), D), indicating that phosphorylation at these residues does not contribute to the stability of AQP4. It should be noted that deleting 10 amino acids at the C-terminus (D314-323) produces a di-lysine endoplasmic reticulum (ER)-retention motif [38] at its C-terminus (310-KKGK-313). Therefore, this mutant is unlikely to reach the cell surface or be transported to the lysosomes. Consequently, we also constructed another mutant lacking the C-terminal 14 amino acids (D310-323) and measured the amount of this protein.
Like the D314-323 and D314A/E318A mutants, D310-323 was destabilized in CHO cells (Fig. 3C (lane 4), E). To exclude the possibility that deleting the cluster of basic amino acids (310-KKGK-313) also contributed to the destabilization of AQP4 in the D310-323 mutant, we replaced all three lysine residues of full-length AQP4 with Ala residues. As expected, eliminating the three positively charged residues slightly enhanced the stability of full-length AQP4, excluding this possibility (Fig. 3C (lane 2), E). Taken together, the C-terminal 10-amino-acid sequence of AQP4 acts as a signal for stabilizing AQP4, critically implicating two negatively charged amino acids in its function.
A tyrosine-based sorting motif (YXXΦ) present just before Glu 281 (277-YMEV-280) has been found to act as a lysosomal targeting signal for AQP4 in the epithelial MDCK cell line [22]. Therefore, we examined whether this motif plays a role in destabilizing AQP4 in CHO cells. We introduced substitution mutations that disrupted this motif into an unstable mutant D282-323 ( These results indicate that the YXXΦ motif is responsible for the Cterminal deletion-induced degradation of AQP4.
The C-terminal 10-amino-acid stabilizing signal competes with YXXΦ motif-mediated transportation of AQP4 to lysosomes Since at least part of the D282-323/Y277F mutant reached the cell surface ( Fig. 5A,B, D282-323/Y277F;     6, D282-323/Y277F; and Movie S7), we assumed that the lack of the C-terminal domain enhanced constitutive endocytosis and lysosomal degradation of AQP4. Therefore, we examined the effect of the YXXΦ motif on the stability of full-length AQP4 (Fig. 4A).
To directly observe whether AQP4 and its derivatives were transported to lysosomes, we co-transfected cDNAs for AQP4 or its derivatives and the lysosomalassociated membrane protein 1 (Lamp1)-enhanced green fluorescent protein (EGFP) fusion protein (Fig. 1B) into HeLa cells and stained them with an anti-FLAG antibody. To label the plasma membrane of transfected cells, we also used myelin oligodendrocyte glycoprotein (MOG)-EGFP. Surface expression of MOG has been demonstrated in multiple cell lines, including HeLa cells [39,40]. The Y277F mutant was widely localized on the plasma membrane labelled with MOG-EGFP (Figs 1B and 5B) and in intracellular compartments, including lysosomes labelled with Lamp1-EGFP, which resembles the subcellular localization of the AQP1-C mutant (Fig. 6A,B, Y277F and Movie S4 and S8). Consistent with these observations, both Y277F and AQP1-C mutants were insensitive to treatment with bafilomycin A1 (Fig. 7C,L (lane 3), G,P), confirming that the YXXΦ motif plays a crucial role in the subcellular localization and degradation of AQP4.
By contrast, treating wild-type and full-length D314A/E318A mAQP4 M1 with bafilomycin A1 (Fig. 7A,B (lane 3), E,F) significantly increased each protein compared with the vehicle control (Fig. 7A,B (lane 2)), indicating that these proteins are degraded by lysosomal proteases. MG132 also tended to stabilize these proteins (Fig. 7A,B (lane 4)). Nevertheless, this result was not statistically significant (Fig. 7E,F), indicating that both wild-type and D314A/E318A mutants were dominantly degraded in lysosomes. The effect of bafilomycin A1 treatment on the D314A/E318A mutant was greater than that on the wild type, indicating that the lack of two acidic amino acids accelerates the degradation of AQP4 in lysosomes.
Consistently, strong punctate signals visualized with an anti-FLAG antibody were observed near lysosomal markers in cells expressing wild-type AQP4, regardless of treatment with bafilomycin A1/MG132 (Fig. 6A,B, WT (À) and WT (+), and Movies S1 and S2). Similarly, in cells expressing full-length AQP4 with D314A/ D318A mutations, strong punctate signals were detected near lysosomal markers when the cells were treated with bafilomycin A1/MG132 (Fig. 6A,B, D314A/E318A (+) and Movie S3). Cell-surface AQP4 was not detected with an anti-FLAG antibody (Fig. 5B); however, flow cytometric analysis demonstrated that the D314A/E318 mutant, as well as the wild type, reached the cell surface in both CHO and HeLa cells (Fig. 5A). Therefore, it is likely that they are transported to the plasma membrane before being distributed to lysosomes.
Taken together, these results indicate that a part of AQP4 in the plasma membrane constitutively undergoes endocytosis and is transported to lysosomes for degradation. The YXXΦ motif in the C-terminal domain is responsible for this process, and the Cterminal 10-amino-acid sequence of AQP4, including Asp 314 and Glu 318 , competes with this function.

Deleting upstream 12 amino acids prevents AQP4 from degradation induced via YXXΦ motif
We observed that the deletion of residues from a portion proximal to the sixth transmembrane domain ( Fig. 2A) stabilized AQP4, especially in CHO cells (Fig. 2B (lanes 6-8)). Interestingly, the deletion of 12 amino acids ranging from Arg 259 to Ala 270 ( Fig. 2A,  D259-270; Fig. 2B (lane 6)) still contained the YXXΦ motif. Importantly, this mutant was insensitive to bafilomycin A1 (Fig. 7D (lane 3), H) and showed subcellular distribution, as observed in the AQP1-C and Y277F mutants (Figs 5A and 6, D259-270 (À) and Movie S5). There are two possible explanations for this observation. First, the distance from the plasma membrane is essential for the YXXΦ motif to function as a signal for the lysosomal degradation of AQP4. Second, another signal is implicated in destabilizing AQP4, and the sequence is required for the YXXΦ motif to function. To verify these hypotheses, we replaced the 12-amino-acid region of the wild-type and D314A/E318A mutant with the corresponding region of AQP1 (239-TDRMKVWTSGQV-250, Fig. 8A). As shown in Fig. 8B,F, deletion of the sequence from Arg 259 to Ala 270 (D259-270) significantly increased wild-type AQP4 to a level comparable to that of the Y277F mutant; replacing the 12-amino-acid sequence with the corresponding region of AQP1 did not alter AQP4 levels. Similar results were observed with the D314A/E318A-mutant of AQP4 (Fig. 8C,G). These observations support the idea that the distance from the plasma membrane is essential for the YXXΦ motif to function as a signal for the lysosomal degradation of AQP4.
YXXΦ motif also functions as a signal for proteasomal degradation of AQP4 As mentioned above, deleting the C-terminal 10-aminoacid sequence produces a di-lysine ER-retention motif (Fig. 9A). This mutant was unstable, and bafilomycin A1 was ineffective (Fig. 9C (lane 3), E). By contrast, MG132 stabilized this protein (Fig. 9C (lane 4), E), indicating that the D314-323 mutant is degraded exclusively by proteasomes. Subcellular localization of the D314-323 mutant in the presence of MG132 showed a pattern resembling that of the ER marker calreticulin (Fig. 9B, Movie S9). We have demonstrated that deleting C-terminal 53 amino acids also form a di-lysine motif-like sequence (Fig. 9A, D271-323) and prevents the mutant from localizing to the plasma membrane [23]. It should be noted that this mutant was quite stable and insensitive to protease inhibitors (Figs 2B,C (lane 5) and 7K,O), suggesting that a sequence implicated in destabilizing the protein exists between Gln 271 and Lys 313 . We then focused on the YXXΦ motif again. Substitution of Tyr 277 or Val 280 with Phe or Arg, respectively, greatly stabilized the D314-323 mutant (Fig. 9D (lanes 2 and 3), F), strongly suggesting that the YXXΦ motif also contributes to proteasomal degradation of the D314-323 mutant. We also examined the effect of deleting 12-amino-acid upstream of the motif (D259-270). Like wild-type and D314A/ E318A mutant, deleting the sequence from Lys 259 to Ala 270 significantly increased the protein level (Figs 8D (lane 2) and 9H). Nonetheless, the deletion was less effective in protecting the protein from degradation than substituting Tyr 277 with Phe (Fig. 9D (lane 4), H). We also noticed that bafilomycin A1 and MG132 cooperatively protected the D282-323 mutant from degradation (Fig. 7J,N). Similar results were observed with the D301-323 deletion mutant (Fig. 7I,M). These observations and the results obtained using the D314-323 mutant suggest that C-terminal deletion enhances the proteasomal degradation of AQP4. Substitution of Tyr 277 with Phe stabilized the D282-323 mutant (Figs 4B and 8E,I), indicating that the YXXΦ motif contributes to both lysosomal and proteasomal degradation of the mutant. However, we did not observe any apparent effect of deleting the 12-amino-acid (Arg 259 -Ala 270 ) upstream of the motif on this mutant (Fig. 8E (lane 2), I).

YXXΦ motif is also responsible for antibodyinduced endocytosis of AQP4
Finally, we examined whether the YXXΦ motif is involved in lysosomal degradation following antibodyinduced endocytosis of AQP4. CHO cells expressing Y277F and D259-270 mutants were treated with E5415A for 24 h. Similar to the AQP1-C mutant, these mutants were resistant to antibody treatment (Fig. 10A,B (lane 2), D,E). By contrast, a mutant with the 12-amino-acid sequence between Leu 258 and Gln 271 replaced with the corresponding region of mAQP1 behaved like the wild type after 24-h incubation with E5415A (Fig. 10C (lane 2), F).
We also examined the localization of endocytosed E5415A in cells expressing mAQP4 and its derivatives. Cells were treated with Alexa Fluor 555-conjugated E5415A for 6 h (Fig. 11, green), followed by Alexa Fluor 647-conjugated E5415A for 18 h (Fig. 11, magenta). Although endocytosed fluorescent-labelled E5415A was exclusively localized in punctate structures (Fig. 11A) mainly localized inside the lysosomes labelled with Lamp1-EGFP in cells expressing wildtype mAQP4 (Fig. 11E,I, Movie S10), it was distributed not only in intracellular compartments but also in the plasma membrane of cells expressing AQP1-C (Fig. 11B,F,J, Movie S11), D259-270 (Fig. 11C,G,K, Movie S12) and Y277F (Fig. 11D,H,L, Movie S13). These results indicate that the YXXΦ motif also contributes to the antibody-induced endocytosis of AQP4 and its transportation to lysosomes.

Discussion
In the present study, we demonstrated that a tyrosinebased YXXΦ motif in the C-terminal domain of  AQP4 plays a crucial role in both the steady-state localization/turnover and antibody-induced endocytosis/lysosomal degradation of AQP4. Our results indicate that to fully act as a signal for lysosomal degradation of AQP4, the YXXΦ motif needs to escape from the inhibitory effect of the C-terminal 10-amino-acid sequence and be located at an appropriate distance from the plasma membrane. In addition to lysosomal degradation, we demonstrated that the YXXΦ motif also functions as a signal to degrade AQP4 using proteasomes under specific conditions. To the best of our knowledge, ours is the first report implicating the YXXΦ motif in the proteasomal degradation of membrane proteins.
YXXΦ motif mediates steady-state localization and turnover of AQP4, and the C-terminal 10-amino-acid stabilizing signal competes with its function The C-terminal 10-amino-acid sequence of AQP4 competes with the function of the YXXΦ motif in the steady-state localization and turnover of AQP4. A putative PDZ-binding motif (the last three amino acids of AQP4) is included within this sequence. Neely et al. [21] demonstrated that deletion of this motif accelerates the degradation of AQP4 in HEK293 cells. Consistent with the results of Neely et al., we also observed that deletion of the last three amino acids of AQP4 reduced the amount of AQP4; however, the effect of the deletion on the stability of AQP4 was limited (Fig. 2D (lane 4)). Here, we also demonstrated that the substitution of Ala for two acidic amino acids (Asp 314 and Glu 318 ) located upstream of the putative PDZ-binding motif significantly reduced the amount of AQP4 when the PDZbinding motif was intact (Fig. 3B (lane 4)). Therefore, the PDZ-binding motif is not the only functional signal, but a wider range of sequences is necessary to fully function against the YXXΦ motif. It should be noted that the last 10-amino-acid sequence of AQP4 ([D/E]-X-[S/T]-X-E-V-L-S-S-V) is highly conserved across vertebrates from mammals to fishes, which may also support this idea. Interestingly, similar opposing effects of the YXXΦ motif and PDZbinding motif on endocytosis and lysosomal degradation of human T-lymphotropic virus type-1 (HTLV-1) envelope (Env) protein have been reported [41]. Only two amino acids occur between the YXXΦ and PDZ-binding motifs of HTLV-1 Env protein; therefore, it is possible that the binding of PDZ proteins to the PDZ-binding motif interferes with the interaction between AP-2/AP-3 and the YXXΦ motif. However, this possibility was excluded because inserting an additional 10 amino acids between the two motifs did not alter their behaviour [41]. The authors speculated that docking of Env to the PDZ protein scaffold creates an environment in which AP-2/AP-3 hardly binds to the YXXΦ motif with high affinity [41] (or might be difficult to access). In the case of AQP4, 33 amino acids are inserted between the YXXΦ motif and the C-terminal 10-amino-acid sequence (40 amino acids between the YXXΦ and putative PDZ-binding motifs), which is consistent with the latter idea. Notably, the distance between the YXXΦ motif (Y-[M/I/V]-E-V) and the last 10-amino-acid sequence is highly conserved in mammals through reptiles (32-33 amino acids), suggesting the importance of the spatial relationship between the motifs in AQP4. Additionally, we cannot exclude the possibility that intramolecular interactions with the C-terminal domain prevent the association of AP complexes with the YXXΦ motif.
Further studies are necessary to elucidate the precise mechanism underlying the opposing effects of the YXXΦ motif and the C-terminal 10-amino-acid sequence of AQP4.

YXXΦ motif is responsible for antibody-induced endocytosis of AQP4
The YXXΦ motif also plays a role in antibodyinduced endocytosis and transportation to lysosomes because disruption of this motif protects the protein from degradation (Fig. 10A,B,D,E), similar to a mutant with the C-terminal domain of mAQP1 (Fig. 1D,F). Importantly, administering a fluorescentlabelled monoclonal antibody (E5415A) to cells expressing D259-270 and Y277F mutants only for the initial 6 h during a 24-h incubation localized it not only to intracellular compartments but also to the plasma membrane (Fig. 11C,D). This observation is in stark contrast to the localization of the labelled antibody only in vesicular structures in cells expressing wild-type mAQP4 after a 24-h incubation (Fig. 11A), suggesting that little antibody-bound AQP4 is transported to lysosomes but recycled back to the plasma membrane. Overall, intracellular localization of the exogenously added antibody in cells expressing Y277F, D259-270 and AQP1-C mutants (Fig. 11) resembled their steady-state subcellular localization (Fig. 6). By contrast, cells expressing wild-type AQP4 (Fig. 11) showed steady-state localization of the D314A/E318A mutant (Fig. 6). Thus, it is conceivable that the antibody binding to the extracellular domains of AQP4 releases the inhibitory effect of the C-terminal 10amino-acid sequence against the YXXΦ motif. It remains unclear how antibody binding to the extracellular domains accelerates endocytosis and lysosomal degradation of the wild-type M1 isoform. One possible explanation is that cluster formation of M1 homotetramers crosslinked by a monoclonal antibody releases the YXXΦ motif from the inhibitory effect of the Cterminal 10-amino-acid sequence. It should be noted that the exogenously expressed M23 isoform of AQP4 is unstable owing to degradation by lysosomal and proteasomal proteases, as treatment with bafilomycin A1 and MG132 significantly increases the amount of the M23 isoform [34,42]. When the M23 isoform is solely expressed in cells, the M23 homotetramer forms huge OAPs, naturally occurring clusters of AQP4 tetramer [14,15]. This fact supports the idea that cluster formation of AQP4 tetramers is a trigger to accelerate the degradation of AQP4, at least in part, in the lysosomes.

Deleting upstream 12 amino acids prevents AQP4 from degradation induced via YXXΦ motif
Another interesting finding in this study is that deleting 12-amino-acid upstream of the YXXΦ motif (D259-270) stabilizes AQP4 and makes it behave like a mutant lacking the YXXΦ motif. There are two possible explanations for this phenomenon: the distance from the plasma membrane is essential for the YXXΦ motif to function as a signal for lysosomal degradation of AQP4, and the 12 deleted amino acids are another signal for lysosomal degradation of AQP4 required for the function of the YXXΦ motif. Inserting an unrelated sequence (a corresponding region of mAQP1) restored the function of the YXXΦ motif (Figs 8B-D and 10C), clearly supporting the former possibility. Consistent with our observations, a similar positional effect on the activity of the YXXΦ motif was observed in the subcellular localization of Lamp1 [43]. In the case of Lamp1, both deletion and insertion of extra amino acids abolished lysosomal targeting of Lamp1; the Lamp1 mutants were recycled to the plasma membrane [43].

YXXΦ-motif functions as a signal for proteasomal degradation of AQP4
In addition to lysosomal degradation, we found the involvement of the YXXΦ motif in the proteasomal degradation of AQP4 using a mutant D314-323, in which a di-lysine ER-retention signal [38] is formed by the deletion (Fig. 9). Interestingly, the C-terminal deletion of AQP4 (D310-323 and D282-323) tended to be degraded by proteasomes (Fig. 7I,J,M,N). Since AQP4 is a highly hydrophobic protein and easily aggregates, especially under denaturing conditions, the C-terminal domain may play a role in stabilizing the conformation of AQP4 tetramers and monomers. However, this is unlikely because the C-terminal domain is dispensable for the proper folding of AQP4 [44][45][46]. Alternatively, since the C-terminal domain plays multiple roles in trafficking and/or localization of AQP4, deleting a part of the domain, including the PDZ-binding motif, may impair trafficking. This phenomenon could lead to the accumulation of AQP4 in unusual intracellular compartments where AQP4 evokes undesired functions that degradation pathways must remove. Since disruption of the YXXΦ motif rescued truncated AQP4 (D282-323 and D314-323) from degradation, we concluded that the mutants escaped recognition by molecules that mediate proteasomal degradation. However, because the patterns of localization between D282-323 (under treatment with both bafilomycin A1 and MG132) and D282-323/Y277F were different, especially the lack of localization of the D282-323 mutant on the cell surface ( Fig. 6 and Movie S6), we cannot exclude the possibility that disruption of the YXXΦ motif also modifies the trafficking of the mutants to subcellular compartments where molecules that mediate proteasomal degradation are not easily accessible.

Plasmid constructions
The mAQP4 M1 isoform, in which Met 23 was changed to Leu [34], avoids the expression of the M23 isoform by a leaky scanning mechanism [47]. A FLAG tag was fused to the N-terminus to quantitatively evaluate the expression level of WT AQP4 and its C-terminally mutated derivatives (Fig. 1B). PCR-based mutagenesis was performed to introduce nonsense and/or substitution mutations. All PCR products were inserted into the pGEM-T vector (Promega, Madison, WI, USA) for sequencing. cDNAs encoding the WT or mutant AQP4 and enhanced green fluorescent protein (EGFP) linked by an internal ribosomal entry site (IRES) (Fig. 1B) were inserted between the SalI and NotI sites of the episomal mammalian expression vector pEBMulti-Puro (FUJIFILM Wako Pure Chemical Corporation, Osaka, Japan). To visualize the plasma membrane and lysosomes, we cloned rat MOG cDNA from the total RNA extracted from the adult rat cerebellum and mouse Lamp1 cDNA from the total RNA extracted from the adult mouse brain, respectively, by PCR using KOD-Plus-Neo (TOYOBO Co., LTD, Osaka, Japan encoding rat MOG and mouse Lamp1 lacking the termination codon were fused C-terminally to EGFP, and its N-terminus was connected in-frame with IRES (Fig. 1B). The fragment containing IRES-MOG-EGFP and IRES-Lamp1-EGFP was inserted into the pEBMulti-Puro vector.
To examine the pathways responsible for AQP4 degradation, transfected cells were treated with 100 nM bafilomycin A1 (Merck Millipore Corporation, Burlington, MA, USA), an inhibitor of vacuolar-type and/or 1 lM MG132 (Merck Millipore Corporation), an inhibitor of proteasomes, at 37°C for 24 h. To examine the effect of antibody-induced endocytosis of AQP4, cells were treated with C9401 (2 lgÁmL À1 ), a monoclonal antibody that exclusively recognizes the extracellular domains of human AQP4 [48] or E5415A (2 lgÁmL À1 ), a monoclonal antibody that preferentially recognizes the extracellular domains of rodent AQP4 [34], at 37°C for 24 h.

Flow cytometry
Transfected CHO and HeLa cells were trypsinized and resuspended in PBS with 0.1% bovine serum albumin (BSA). The cells were then incubated with 2 lgÁmL À1 E5415A at 4°C. After an hour of incubation, cells were washed with 0.1% BSA in PBS and subsequently stained with PE-conjugated goat anti-mouse IgG (1 : 100; South-ernBiotech, Birmingham, AL, USA) for 1 h at 4°C. Flow cytometry was performed on a BD Accuri C6 Plus flow cytometer (BD Biosciences, Franklin Lakes, NJ, USA), and the data were analysed using FLOWJO software (Becton, Dickinson and Company, Franklin Lakes, NJ, USA).
To examine the localization of AQP4 or its derivatives, transfected cells were incubated in the presence or absence of bafilomycin A1 and MG132 for 24 h and fixed with 4% paraformaldehyde. The fixed cells were permeabilized with 0.1% Triton X-100 in PBS and stained with anti-FLAG (1 : 200, clone M2; Merck Millipore Corp.) antibody, followed by Alexa Fluor 555-conjugated anti-mouse IgG (1 : 200; Thermo Fisher Scientific).
To visualize the endocytosed monoclonal antibody, E5415A was labelled with Alexa Fluor 555 and 647 using Alexa Fluor 555 and Alexa Fluor 647 antibody labeling kits (Thermo Fisher Scientific), respectively, according to the manufacturer's instructions. The transfected cells were first incubated with 2 lgÁmL À1 Alexa Fluor 555-conjugated E5415A for 6 h and then washed twice with PBS. Subsequently, the cells were incubated with 2 lgÁmL À1 Alexa Fluor 647-conjugated E5415A for 18 h. The cells were fixed with 4% paraformaldehyde.

Statistical analysis
Statistical analyses were performed using JMP ver. 16.2.0 (SAS Institute Inc., Cary, NC, USA). Data were analysed using one-way ANOVA followed by the Tukey-Kramer method, except for Fig. 10D-F, which were analysed using a two-tailed Student's t-test.