SNX31: A Novel Sorting Nexin Associated with the Uroplakin-Degrading Multivesicular Bodies in Terminally Differentiated Urothelial Cells

Uroplakins (UP), a group of integral membrane proteins, are major urothelial differentiation products that form 2D crystals of 16-nm particles (urothelial plaques) covering the apical surface of mammalian bladder urothelium. They contribute to the urothelial barrier function and, one of them, UPIa, serves as the receptor for uropathogenic Escherichia coli. It is therefore important to understand the mechanism by which these surface-associated uroplakins are degraded. While it is known that endocytosed uroplakin plaques are targeted to and line the multivesicular bodies (MVBs), it is unclear how these rigid-looking plaques can go to the highly curved membranes of intraluminal vesicles (ILVs). From a cDNA subtraction library, we identified a highly urothelium-specific sorting nexin, SNX31. SNX31 is expressed, like uroplakins, in terminally differentiated urothelial umbrella cells where it is predominantly associated with MVBs. Apical membrane proteins including uroplakins that are surface biotin-tagged are endocytosed and targeted to the SNX31-positive MVBs. EM localization demonstrated that SNX31 and uroplakins are both associated not only with the limiting membranes of MVBs containing uroplakin plaques, but also with ILVs. SNX31 can bind, on one hand, the PtdIns3P-enriched lipids via its N-terminal PX-domain, and, on the other hand, it binds uroplakins as demonstrated by co-immunoprecipitation and proximity ligation assay, and by its reduced membrane association in uroplakin II-deficient urothelium. The fact that in urothelial umbrella cells MVBs are the only major intracellular organelles enriched in both PtdIns3P and uroplakins may explain SNX31's MVB-specificity in these cells. However, in MDCK and other cultured cells transfected SNX31 can bind to early endosomes possibly via lipids. These data support a model in which SNX31 mediates the endocytic degradation of uroplakins by disassembling/collapsing the MVB-associated uroplakin plaques, thus enabling the uroplakin-containing (but ‘softened’) membranes to bud and form the ILVs for lysosomal degradation and/or exosome formation.

Given the functional importance of the apical surface-associated uroplakins, their endocytic retrieval is likely to be tightly regulated [31][32][33]. It has been a central dogma of the urothelial biology field that the uroplakin-delivering fusiform vesicles can be induced by bladder distension to fuse with the apical surface, and that the surface-associated plaques can later be retrieved, upon bladder contraction, to re-form cytoplasmic fusiform vesicles thus achieving reversible adjustment of the apical cell surface area [2,3,33,34]. This view is not supported, however, by tracer studies indicating that internalized luminal plaques mainly become associated with the multivesicular bodies followed by lysosomal degradation [35,36]. A critical and as yet unanswered question about the endocytic degradation of uroplakins is: how can the rigid-looking uroplakin plaques, once become incorporated into the limiting membranes of the MVBs, bud into the lumen of MVB to form the small intraluminal vesicles, that have an extremely high curvature, for lysosomal degradation [37]?
To identify urothelium-specific genes, we previously generated a urothelium-specific cDNA library by suppression subtractive hybridization to eliminate cDNAs that are also present in ten other non-urothelial tissues including kidney, lung, spleen, skeletal muscle, esophagus, stomach, intestine, colon, brain, and liver [38]. This subtraction library achieved ,1,000-fold enrichment of urothelial-specific cDNA, as evidenced by a .10-fold increase in the cDNA of uroplakin Ib, a urothelial marker, and a .100-fold reduction in actin cDNA [38]. From this cDNA library, we originally identified several secretory proteins including tissue-type plasminogen activator, urokinase and a serine protease inhibitor PP5, that are made as major urothelial differentiation products that are secreted into the urine, thus establishing that bovine urothelium can function not only as a permeability barrier but also as a source of some urinary proteins [38].
Here we report the identification, from this urothelium-enriched cDNA library, of a novel sorting nexin, SNX31, that is coexpressed with uroplakins in terminally differentiated urothelial umbrella cells. The sorting nexins family proteins are involved in regulating protein trafficking, and their defects have been linked to several diseases including Alzheimer's and several infections [39,40]. SNX31 binds the uroplakin-as well as PtdIns3P-enriched membranes of the multivesicular bodies (and their intraluminal vesicles) that are the most prominent endosomal compartment in urothelial umbrella cells. Biotinylated, surface-associated uroplakins are endocytosed into the SNX31-positive mutlivesicular vesicles (MVBs), where uroplakins and SNX31 co-localize not only at the limiting plaques, but also at the highly curved membranes of the small intraluminal vesicles. We propose a model in which the binding of SNX31 to the cytoplasmic domain of uroplakins may cause the dissociation and collapse of the uroplakin particles located at the periphery of the MVB-associated plaques, thus facilitating the budding of the uroplakin-containing membranes to form intraluminal vesicles for lysosomal degradation and/or exosome formation.

SNX31 is co-expressed with uroplakins in terminally differentiated urothelial umbrella cells
From a previously established urothelium-enriched cDNA library [38], we identified a cDNA clone that encoded a sorting nexin that was highly enriched in mouse (Fig. 1A) and bovine bladder (Fig. S1). This cDNA encoded sorting nexin 31 (SNX31) whose expression, like that of uroplakin IIIa, was greatly upregulated when cultured bovine urothelial cells reached confluence and became more stratified and differentiated ( Fig. 1B; [41]), indicating a differentiation-dependent expression.
We generated several affinity-purified rabbit (peptide) antibodies against mouse and bovine SNX31. When used to immunoblot the total mouse and bovine urothelial proteins resolved by SDS-PAGE, these antibodies recognized a single band of SNX31 (51-kDa or 53-kDa for mouse or bovine, respectively; Fig. 1C, lanes 1 and 3); this band was specific as it was undetectable by preimmune antibodies. Immunofluorescent staining of mouse urothelial (vertical) sections using these antibodies revealed that SNX31 was associated with cytoplasmic vesicles in terminally differentiated, superficial umbrella cells ( Fig. 1D and E). Double-staining showed that the SNX31-positive vesicles were located lower than uroplakins in the cytoplasm of the umbrella cells (Fig. 1D). Staining of horizontal sections of urothelium in a whole-mount mouse bladder preparation, which offered a better spatial resolution, showed that keratin K20 formed a subapical, chicken wire-like network ( Fig. 1F; [42]), and that SNX31-positive vesicles were below this subapical K20-network and the uroplakinenriched upper cytoplasm ( Fig. 1F-H).

SNX31 is associated with multivesicular bodies in terminally differentiated urothelial umbrella cells
The SNX31 staining pattern in the umbrella cells was distinct from those of endocytic markers clathrin ( Fig. 2A), caveolin 2 ( Fig. 2B), EEA1 (Fig. 2C), Rab5a and GTPase Rho A, and the exocytic marker Rab11a (data not shown). Although in vertical sections of mouse urothelium the SNX31-vesicles seemed to be partially overlapping with the late endosome marker Lamp1 (Fig. 2D), high-resolution whole mount microscopy clearly distinguished the SNX31-and Lamp1-positive vesicles (Fig. 2E). Electron microscopy identified these SNX31-positive vesicles as multivesicular bodies (MVBs), which were lined with maturelooking uroplakin plaques and contained many small intraluminal vesicles (ILVs; Figs. 3 and 4).
In a few cases, we found that uroplakin IIIa was localized on membrane budding from the limiting membranes of the MVBs into the lumen (Fig. 3E). To better understand this process, we examined the ultrastructure of urothelial MVBs using high pressure freezing and freeze-substitution which offers a superior resolution over conventional TEM (Fig. 4). Our data indicate that the MVBs are lined with urothelial plaques morphologically indistinguishable from the apical plaques interconnected by hinge areas (Fig. 4A and B), and that ILVs were formed by invagination of the MVB membrane in hinge areas ( Fig. 4C and D; open arrows). The fact that such images were quite rare suggests that the invagination process was either rapid, or rare, or both.
To see whether the apical membrane proteins can be delivered to MVBs, we biotinylated the apical surface of mouse bladder urothelium in situ using a membrane-impermeable biotinylation reagent, sulfo-N-hydroxysuccinimide [45]. The internalized, biotin-tagged surface proteins, that we showed earlier are mostly uroplakins [45], initially form SNX31-negative vesicles ( Fig. 5A and B; open arrows), and later became co-localized with the SNX31-positive MVBs ( Fig. 5B and C). Electron microscopy showed that the internalized apical proteins are largely associated with the limiting plaques of the MVBs (Fig. 5D-F). These results indicate that the SNX31-positive MVB plaques are accessible from the apical surface.

SNX31 interacts with uroplakins
The above data indicate that SNX31 and uroplakins co-existed in not only the peripheral membranes, but also the ILVs of the MVBs. To determine whether SNX31 directly binds uroplakins, we performed co-immunoprecipitation experiments. Total pro- teins of mouse urothelial cells were solubilized in a detergentcontaining buffer, and immunoprecipitated using a monospecific antibody to SNX31 (Fig. 1C). Immunoblotting of the immunoprecipitated proteins resolved by SDS-PAGE showed the expected 51-kDa SNX31 ( Fig. 6A and B). Additional immunoblotting of the same fraction using antibodies to individual uroplakins detected UPIIIa and Ib (of the UPIIIa/Ib pair; Fig. 6A and B); which were not detected in proteins pulled down by the control (quenched) beads. UPIa was not detected and the results of UPII were inconclusive due to nonspecific UPII binding even to the control beads ( Fig. 6A and B). Similar results were obtained in three independent experiments ( Fig. 6A and B, and data not shown).
To determine the in situ subcellular sites of SNX31:uroplakin interaction, we performed proximity ligation assay (PLA; see Methods), which can localize and quantify protein-protein interactions at the single molecule level using standard fluorescent microcopy [46]. In this assay, urothelial sections were doublestained using a rabbit antibody to SNX31 and a newly generated mouse monoclonal antibody to uroplakin Ib, followed by incubation with nucleotide-conjugated species-specific secondary antibodies. Close physical proximity of the two secondary antibodies (within 15-20 nm) allows the hybridization of two ''linking'' nucleotides and the ligase-catalyzed amplification of a concatemeric product that can be detected and quantified digitally [46]. Negative controls, including the replacement of the primary antibodies by pre-immune sera or phosphate-buffered saline or the staining of intestinal epithelial sections, yielded no positive signals. As a positive control, we showed that UPIIIa and Ib, known to form a heterodimer [22][23][24], indeed generated strong PLA signal in the umbrella cells (Fig. 6C, G and K). The urothelial umbrella cells of the Buff mouse (Fig. 6E, I and K), which carries a Vps33a mutation that blocks the fusion of MVBs with lysosomes, are known to accumulate MVBs [44]. Consistent with this, the Buff mouse urothelium showed a tremendous increase in SNX31 staining in its umbrella cells (Fig. 6F vs. D) and a corresponding increase in the PLA signals of SNX31:UPIIIa interaction (Fig. 6J vs. 5H; 5K). In addition to confirming that SNX31 interacted with uroplakins, these PLA results indicated that such SNX31:uroplakin interactions occurred mainly at the MVBs.
As another independent way to assess the interactions between SNX31 and uroplakins, we fractionated normal mouse urothelial membranes by flotation equilibrium on a 0.4-2 M sucrose density gradient (Fig. 7). Under this condition, uroplakin IIIa floated as a major peak (1.0-1.1 M sucrose). Some of the SNX31 (30-40%) was associated with this uroplakin peak, with the rest forming a smear in the rest of the gradient presumably reflecting a soluble state. Uroplakins of the Vps33a mutant (Buff) mouse urothelium similarly formed a major peak (1.1-1.2 M sucrose), but, in this case, almost all of the SNX31 co-floated with the uroplakin peak consistent with an enhanced SNX31:uroplakin association as To visualize the subcellular distribution of SNX31 in cultured cells, we studied its expression in transfected MDCK cells (Fig. 8). We found that the ectopically expressed SNX31 bound to cytoplasmic vesicles that were positive for EEA1, an early endosome marker (Fig. 8A), without or with co-transfected uroplakins (Fig 8B-D). This result indicates that the targeting of SNX31 to early endosomes in MDCK cells is uroplakin-independent, and that it is probably mediated through the interaction with phospholipids.

Membrane association of SNX31 decreases in the uroplakin II-knockout urothelium
We next compared membrane-association status of SNX31 in normal vs. UPII-deficient urothelium (Fig. 9). We described earlier that UPII-knockout leads to (i) a decrease in the protein levels of its partner uroplakin UPIa, as well as the other uroplakin pair UPIb/ IIIa ( Fig. 9A and B), (ii) the loss of fusiform vesicles, apical urothelial plaques, and the plaque-lined MVBs (Fig. S2); and (iii) a change in the shape of umbrella cells from squamous to cuboidal or even columnar (Figs. 9B and S2; [10,12]). Immunofluorescent staining showed that the UPII-deficient urothelium lacked the large SNX31-positive MVBs; instead, SNX31 was associated with fine cytoplasmic vesicles (

SNX31 binds to PtdIns3P in a PtdIns3P kinase-dependent manner
The ability to bind distinct phosphoinositides through their PX domain is a hallmark of the SNX proteins and allows SNXs to target their cargoes to specific cellular membranes [39,47]. To determine the lipid-binding specificity of the SNX31, we  transfected 293T cells with a cDNA encoding SNX31 that was Cterminally tagged with HA. We then affinity-purified this protein and used it to overlay a nitrocellulose membrane with spots containing 100 pmol of various lipids (Fig. 10A). The lipid binding-specificity of SNX31:HA and the positive control, 'PtdIns(4,5)P 2 grip', was then detected using antibodies against the HA antigen or GST, respectively. We found that SNX31 preferentially bound PtdIns(3)P (Fig. 10A). In another experiment, we found that SNX31 co-localized precisely with EEA1 in the vesicular endosomes of transfected MDCK cells ( Fig. 10B; also see Fig. 8A). However, this association was abolished by wortmannin ( Fig. 10C; [47,48]). Hence, the association of SNX31 with early endosomes is largely dependent on PtdIns(3)P.

SNX31 is a highly urothelium-enriched and differentiation-dependent sorting nexin
Sorting nexins are a family of PX (phox-homology) domaincontaining proteins that link various protein cargoes to endosomes. These proteins usually have a broad tissue-distribution (see Fig. 1A, e.g., for the wide tissue-distribution of the closely related SNX17) performing general functions in membrane trafficking, cell signaling and organelle motility [39,40,49,50]. The urotheliumspecificity (Figs. 1A and S1) and differentiation-dependent expression of SNX31 (Fig. 1B, D and E) is quite exceptional, suggesting that it must perform an important function in terminally differentiated urothelial cells.
Possible roles of Get1 gene in the co-regulation of uroplakin and SNX31 genes Andersen and co-workers showed that the ablation of the Get1/ Grhl3 gene that encodes grainyhead, a transcription factor that binds to the promoter of UPII gene, led to the down regulation of not only genes encoding uroplakin II (7.56 fold reduction) and other uroplakins, but also the SNX31 gene (30.17 fold; [51]), suggesting that Get1 may contribute to the co-expression of the SNX31 and uroplakin genes in the urothelial umbrella cells.

SNX31 binds to uroplakin proteins in urothelial umbrella cells but not in transfected MDCK cells
Our data indicate that SNX31 interacts with uroplakins, based on immunoprecipitation ( Fig. 6A and B), proximity ligation assay (Fig. 6H, J and K), EM co-localization on the limiting membranes of MVBs and ILVs (Fig. 3), and co-floating during sucrose gradient centrifugation (Fig. 7). It is interesting that, in extracts of the in vivo urothelium, SNX31 co-immunoprecipitated with uroplakins Ib and IIIa that form the UPIb/IIIa pair (Fig. 6). Since UPIb has very little cytoplasmic domain [5,18,20]), we hypothesize that SNX31 pulled down UPIIIa of the (UPIb/IIIa) heterodimer via UPIIIa's cytoplasmic tail.
Although urothelial cells cultured in the presence of 3T3 feeder cells can stratify and differentiate [41], their differentiation deviates significantly from that of in vivo urothelium [52]. Unlike in vivo urothelium whose uroplakins are fully assembled to form 2D crystals of 16-nm particles covering the apical urothelial surface, the uroplakins in cultured urothelial cells do not form 16-nm particles let alone 2D crystals ( [9,24,41,52,53]; cf. [54,55]). It is therefore perhaps not surprising that even though uroplakins of the in vivo urothelium can co-immunoprecipitate with SNX31 (Fig. 6), those of cultured urothelial cells (or uroplakin dimercotransfected MDCK cells) fail to do so (data not shown) -possibly due to a lack of certain conformation or secondary modifications that are unique to uroplakins of in vivo urothelium.
Functional diversities of the FERM-containing sorting nexins SNX31, SNX17 and SNX27 form a small subgroup of sorting nexins sharing a 4.1/ezrin/radixin/moesin (FERM)-like domain [56]. Collins and co-workers showed recently that the FERM-like domain of SNX17 and SNX27 can recognize the NPxY motif in vitro, and that all three sorting nexins bind Ras GTPase [56]. While these three sorting nexins are structurally related, they seem to target different classes of protein cargos. For example, the FERM-and/or C-terminal domain of SNX17 can bind the NPx(F/Y) motif in the cytoplasmic tails of several membrane cargo proteins including P-selectins [57], LDL receptors [58], and  . 300 mg of total mouse urothelial proteins were incubated with the beads for 4h at 4uC, the immunoprecipitated proteins were resolved by SDS-PAGE and detected using antibodies as denoted (all monospecific recognizing a single band that is shown). Beads whose activated crosslinking side chains were quenched (Q), or control beads (C), were used as negative controls. Note that the anti-SNX31 pulled down not only SNX31, but also UPIIIa and UPIb (of the UPIIIa/Ib pair). UPII was pulled down, but the negative controls indicated that this was nonspecific. (B) A separate co-immunoprecipitation assay with anti-SNX31 crosslinked to the beads under two different crosslinking conditions (aSNX31a and aSNX31b); anti-Gst beads (Gst) were used as an additional negative control. (C) Colocalization of UPIIIa (red) and UPIb (green) in normal mouse urothelium. Note that UPIb and IIIa (UPIb/IIIa pair) co-localized precisely, as expected. the amyloid precursor protein [59]. Functionally, SNX17 promotes recycling and suppresses the degradation of its membrane cargo proteins [60,61]. In the case of SNX27, it contains an additional PDZ domain that targets proteins containing a PDZ-binding motif, including beta-Pix (p21-activated kinase-interactive exchange factor) [62] and multidrug resistanceassociated protein 4 (MRP4/ABCC4) [63]. Functionally, SNX27 in general enhances the recycling of plasma membrane (cf. [63]). Finally, we show here that SNX31 may facilitate multivesicular body-mediated uroplakin degradation in urothelial umbrella cells (see below). Together, these data indicate that, despite their structural relatedness, these three sorting nexins target different classes of cargo proteins and perform diverse tissue-specific and cellular context-dependent functions.
Possible roles of SNX31 in the uroplakin-degrading multivesicular vesicles: a model Figure 11 is a schematic diagram that provides a unifying model tying together all our data, and proposes a possible SNX31 function. As indicated in the diagram, the apical surface of umbrella cell is covered by uroplakin plaques (step a), which are endocytosed by a clathrin-and caveolin-independent mechanism (step b; [31]). It is unclear, however, whether this proceeds via the formation of small early endosomes (EE) or large fusiform-like vesicles (FV'; hence the question mark at step b; unpublished data).
Existing evidence suggests that MVBs are involved in uroplakin degradation (step c). (i) MVBs in umbrella cells are lined with urothelial plaques morphologically indistinguishable from the apical plaques (Figs. 3 and 4; [35,64,65]). (ii) Lectin-, peroxidaseas well as biotin-tagged apical uroplakins are delivered to MVBs ( Fig. 5; also see [28,35,36]). (iii) Vsp33a mutation, which blocks the fusion of MVBs to lysosomes, leads to the accumulation of MVBs and hindered uroplakin degradation [44]. Finally, (iv) overexpression of MAL, which facilitates the fusion of the exocytic uroplakin vesicles with the apical surface, stimulates uroplakin exocytosis and subsequent compensatory endocytosis, which in turn leads to a massive accumulation of MVBs and a decrease in uroplakin content [45]. Since proteins remaining on the limiting membranes of MVBs are known to be primarily recycled while those incorporated into the ILVs are largely destined for lysosomal destruction [66,67], most of the uroplakins sorted into the SNX31positive ILVs are probably destined for lysosomal degradation. However, one cannot rule out the possibility that some of the ILVassociated uroplakins are released to the extracellular space via the UPIb/UPIIIa, punctate signals were detected corresponding most likely to MVBs. (I) PLA signal of UPIb and UPIIIa interaction in the Buff (Vps33a mutant) mouse urothelium. (J) PLA signal of UPIIIa/SNX31 interaction in the Buff mouse urothelium. Note the tremendously increased PLA signals between UPIIIa and SNX31. (K) Quantification of the PLA signals using the Duolink Image Tool software: The average numbers of signals per umbrella cells are presented. Arrows mark the apical surface; other symbols are the same as in Fig. 1. Scale bars = 10 mm (panels C to J) doi:10.1371/journal.pone.0099644.g006 Figure 7. Association of SNX31 with uroplakin IIIa during sedimentation equilibrium centrifugation. Total bladder urothelial proteins from wild type and buff mice were subjected to a flotation analysis on a sucrose density gradient. The total homogenate was loaded at the bottom of a continuous sucrose gradient (0.4 to 2.0 M sucrose), and membranes were separated (fraction 1 represents the lightest and 20 the heaviest). LZ (loading zone) and P (pellet). Following centrifugation, proteins from various fractions were immunoblotted for SNX31, UPIIIa and Lamp1. Note that in normal mouse urothelium (A), SNX31 largely co-distributed with the uroplakin peak with the remaining forming a smear, and that, in Buff mouse (B), almost all the SNX31 co-floated with UPIIIa. doi:10.1371/journal.pone.0099644.g007 exosome pathway because small amounts of uroplakins have been detected in the urine-associated exosomes [68].
Our data indicate that SNX31 can bind to both uroplakin proteins (Fig. 6) which are mainly associated with apical surface and MVB, as well as the PtdIns(3)P-lipids (Fig. 10), which are associated with early endosomes, MVBs and their ILVs [66,69]. The fact that MVBs (and ILVs) are the only organelles that contain both SNX31 ligands (uroplakins and PtdIns(3)P) may account for SNX31's preferential binding to MVBs (Figs. 1 and 5; step c in Fig. 11).
An interesting feature of SNX31 is that it has an ESCRT-like domain. Protein sorting to MVBs and their entry into ILVs requires ubiquitination and ESCRT machinery [66,70]. The ESCRT complexes have ubiquitin-binding domains (UBDs) that can recognize the ubiquitinated cargo for it's sorting into the ILVs of the MVBs, followed by lysosomal degradation [71]. Although the C-terminus of UPIIIa is the major cytoplasmic domain of the uroplakin complex, we could not detect UPIIIa ubiquitination (data not shown). It will be interesting to determine in future studies whether uroplakin sorting to the MVBs occurs via a ubiquitin-and ESCRT-independent mechanism [70,72,73]. Alternatively, it is possible that the ESCRT machinery can recognize the ubiquitin-like domain of the SNX31, thus allowing the endocytosed uroplakin vesicles to target to MVBs and enter into ILVs (step d).
The invagination of the MVB-associated uroplakins to form small ILVs (step d) requires a significant bending of the rigidlooking urothelial plaques, which presents a major energetic barrier [37]. This invagination process can be greatly facilitated, however, if the uroplakin particles can be disassembled. Existing evidence indicates that uroplakin particles can indeed dissociate and undergo major conformational changes in response to mechanical stress or protein-protein interactions. First, we showed earlier by quick-freeze deep-etch electron microscopy that the hinge areas interconnecting the neighboring uroplakin plaques are not just particle-free 'regular-looking membranes', as was thought earlier. Rather, these areas contain numerous dissociated and partially collapsed uroplakin particles [7] and are, like the plaques per se, detergent-insoluble suggesting that these are also uroplakincontaining membranes [74]. Moreover, when isolated apical urothelial membranes reseal in vitro to form an outside-out vesicle, the plaques (normally 600-900 nm in diameter containing 1,400-3,000 particles) can break up into smaller ones (100-400 particles) in response to an increased membrane curvature. The newly formed hinge areas that separate the smaller plaques again contain dissociated and collapsed uroplakin particles morphologically similar to those found in the hinge areas of normal bladder urothelium [28]. These data suggest that the dissociation and collapse of uroplakin particles may be the normal mechanism underlying hinge formation [7]. Second, we showed by cryoelectron microscopy that binding to the extracellular domain of uroplakins by a single protein, i.e., the bacterial lectin FimH, can induce uroplakin particle to undergo global conformational changes [75]. Third, as mentioned earlier, uroplakins synthesized by cultured urothelial cells do not form 16-nm particles [9,24,52]. Taken together, these data suggest that the structure of the 16-nm uroplakin particles in 2D plaques is quite flexible as they can readily dissociate and undergo major conformational changes or even collapse in response to physical stress, ligand binding and cell culture conditions. Based on these data, we hypothesize that SNX31, once bound to the cytoplasmic tail of the MVB-associated UPIIIa, may cause the uroplakin particle located at the edge of a plaque to dissociate from the plaque and to collapse, thus facilitating the invagination of the uroplakin-containing membranes to form ILVs that are destined for lysosomal degradation (step e). We further hypothesize that continued binding of SNX31 to the cytoplasmic tail of uroplakins is required in order to maintain a collapsed uroplakin configuration in early ILVs (Fig. 11,  step d). Additional work is needed to test the validity of this model.

Mice
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of NYU Langone Medical Center (protocol #'s 090905-01 and -02). All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering. All studies were performed with the approval of the NYU Medical Center Institutional Animal Care and Use Committee (IACUC).

Tissue distribution of SNX31
Total RNAs were isolated using the RNeasy midi kit (Qiagen cat 75144), following the manufacture's instruction, from 1-3 g of mouse and bovine tissues. Mouse cDNA was synthesized by RT-PCR using a Superscript first-strand system (Invitrogen) with the primers: RTSNX31 Fwd and RTSNX31 Rev; length: 226 bp; cycle protocol: 95C 10 min; 95C 30 s, 58C 30 s, 72C 30 s 25 cycles; 72C 10 min. Bovine RNA was separated on a 1% agarose gel, transferred to a Hybond nitrocellulose membrane (Amersham Biosciences) and probed with a bSNX31 specific probe labeled with

Generation and characterization of rabbit antibodies to SNX31
Rabbit antisera were raised against synthetic peptides corresponding to (i) amino acid residues 423-439 (CKRSEG-DYVWDTLMEEGL), the C-terminus of the mouse SNX31; (ii) amino acid residues 387-402 (C-SPEMQIEVPEQGRSKK) located near the C-terminus of the bovine SNX31. The antibodies were affinity purified using the original peptide antigen and shown to be mono-specific when used to immunoblot the total proteins of mouse and bovine urothelia, or the total proteins of 293t cells that had been transfected using cDNA's encoding mouse or bovine SNX31.

Fractionation of urothelium or transfected 293T cells
Total cell extracts of bovine bladder epithelial cells were fractionated by centrifugation (60 min at 100006g) to produce a pellet containing nuclei and all cellular membranes and a soluble fraction [21,43]. Equivalent amounts of samples were loaded on SDS polyacrylamide gels (17%) followed by immunoblotting using an anti-SNX31 antibody. Transfected 293T cells or mouse and bovine bladder epithelial cells were washed in PBS pH 7.4 and lysed for 10 min at room temperature with RIPA buffer (1% NP-40, 0.5% DOC, 0.1% SDS, in 150 mM NaCl, 50 mM Tris (pH 7.5)), supplemented with protease inhibitor cocktail (Roche, Germany) [45]. Protein quantification was accomplished by using the BCA protein assay (Thermo Scientific, USA) and BSA used as a protein standard. Equal amounts of protein were separated by SDS polyacrylamide gel electrophoresis in 17% acrylamide gels and electroblotted onto a Transblot nitrocellulose membrane (BioRad, USA). Membranes were blocked in PBS pH 7.4 with 5% non-fat dry milk and probed with monoclonal anti-HA antibody (Sigma). Horseradish peroxidase-conjugated anti-mouse or antirabbit immunoglobulin G was used as secondary antibodies (Sigma). Immunoreactive bands were detected by chemiluminescence (ECL).

Construction of pcDNA3:HA vector
In order to construct a C-terminal HA-tagged pcDNA3 vector, primers CHAF/CHAR were boiled for 5 min in annealing buffer (10 mM Tris HCl, pH 7.5-8.0, 50 mM NaCl) and cooled to room temperature to get annealed through their complementary sequences. The vector pcDNA3 (Invitrogen) was digested by ApaI (NEB). The annealed double-stranded oligonucleotide was ligated into the digested vector in the molar ratio of 3:1 by T4 ligase (NEB) at 16uC overnight. The recombinant DNA was transformed into competent bacteria host cell, Top10. Confirmation polymerase-chain reaction was performed with the primers: 59TAC-GATGTTCCAGATTAC39 and 59 GTAAAGCACTAAATCG-GAA39 and sequenced. Mouse SNX31 cDNA was then used as a template to amplify the mSNX31 ORF sequence with the primers: CFLF/CFLR. The C-terminal HA tagged pcDNA3 vectors were digested with HindIII and XbaI (NEB) and ligated to the amplified mSNX31 PCR products. Top10 competent E. coli was transformed with recombinant plasmids. Positive clones were subsequently sequenced.

Transfection
293T and MDCK cells were cultured in MEM medium (Cellgro, Mediatech Inc.) supplemented with 1% PS (Gibco) and 10% FBS (Fisher/Hyclone), at 37uC in a humidified 5% CO 2 chamber. 80% confluent growing suspensions seeded 18 h before were transfected with the PolyJet reagent (SignaGen Laboratories). A mixture of DNA, PolyJet 1:3 in serum free DMEM was added to cells after 10 min of incubation at room temperature. Cells were analyzed 42 h post transfection [45].  Figure 8 in [45]). Although the binding of SNX31 to some EE cannot be excluded, such structures have not yet been detected. (c) The multivesicular bodies (MVBs) of bladder urothelial umbrella cells are highly specialized as they are lined by SNX31-associated uroplakin plaques. (d) SNX31 and uroplakins co-enter into the intraluminal vesicles (ILVs). It is hypothesized that the binding of SNX31 to the cytoplasmic tail of UPIIIa can cause the uroplakin particles to dissociate from the plaque and to collapse (changing the shape of the luminal 16nm particles from circle to triangle in the diagram) thus facilitating the invagination of the uroplakin-containing membrane to form ILVs. (e) MVBs fuse with lysosomes for uroplakin degradation [35,36]; this process is blocked in Buff mouse carrying a Vps33a mutation. This Immunofluorescence microscopy Paraffin sections of urinary bladders from 8 to 12 week-old mice were de-paraffinized, rehydrated, and microwaved to expose the antigenic sites. Transfected MDCK cells were cultured on cover slips, fixed in 4% paraformaldehyde solution in PBS and permeabilized in 1% Triton-X 100. All samples were then incubated with primary antibodies: monoclonal anti-UPIII

Electron microscopy
For ultrastructural study, mouse bladder urothelial samples were high pressure frozen in a Balzers HPM 010 apparatus, and freeze-substituted first with acetone containing 2% OsO 4 and then with 100% acetone in a Leica AFS apparatus. The samples were then embedded in Epon, and ultrathin (60 nm) sections were cut and counterstained with lead citrate and uranyl acetate [28].
For EM localization using the Lowicryl procedure, mouse bladders were excised and fixed for 4 h in 0.1 M sodium cacodylate buffer (pH 7.4) containing 3% (w/v) paraformaldehyde, 0.1% glutaraldehyde (v/v) and 4% sucrose. The samples were embedded in Lowicryl K4M (Polysciences, Inc., Warrington, PA) and polymerized under UV light (360 nm) at 235uC. For single labeling, ultra-thin sections were incubated with primary antibodies (AU1 for UPIII and 1:10 for anti-SNX31) at room temperature for 2 h and then 4uC overnight. Protein A-conjugated 15-nm gold particles (Cell Microscopy Center, University Medical Center Utrecht, 35584 CX Utrecht, The Netherlands) were used as secondary antibody. Double labeling was performed as previously described, with minor modifications by using 15-nm protein A gold particles for SNX31 and 5-nm protein gold for uroplakin III [45].
Ultrathin sections of Lowicryl K4M embedded biotinylated mouse bladder were cut, mounted on formvar-carbon coated nickel grids. After incubation with Nanogold conjugated streptavidin (1:50, Nanoprobes, Yanphank, NY), silver enhancement was performed in the dark for 8 minutes using HQ Silver Enhancement kit (Nanoprobes). Grids were stained with 3% uranyl acetate and examined under Philips CM-12 electron microscope and photographed with a Gatan (4kx2.7k) digital camera.
For cryoEM localization, mouse bladders were fixed with 4% paraformaldehyde in PBS for 1 h, rinsed in PBS, embedded into 10% gelatine blocks, cryoprotected by incubation in cold 2.3 M saharose and frozen in liquid nitrogen. Cryosections (55 nm thick) were cut with Leica FCS cryo-ultramicrotom at -120uC and collected on gold EM grids. Primary rabbit polyclonal antibodies against AUM (diluted 1:10,000) were incubated overnight at 4uC, washed in PBS, and incubated with goat anti-rabbit secondary antibodies, conjugated with 10 nm colloidal gold (Sigma, diluted 1:40) at room temperature for 1.5 h. Cryosections were counter-stained with methyl cellulose/uranyl acetate solution and examined with Philips CM100 transmission electron microscope, running at 80 kV.

Immunoaffinity purification
Monoclonal anti-HA agarose conjugate clone HA-7 beads (Sigma) were used to immunoaffinity-purify mSNX31:HA. Beads were packed in empty chromatography column with PBS pH 7.4. Column was then washed 3x with Glycine-HCL 0.1 M pH 2.5 (5x the column volume), followed by three sequential aliquots of PBS pH 7.4 (5x the column volume). Total protein lysate was loaded into the column and incubated overnight at 4uC. The flow through was collected and the column washed with PBS pH 7.4 till OD 280nm reached 0.01. mSNX31:HA was subsequently eluted with 8 aliquots (column volume) of Glycine-HCL 0.1 M pH 2.5 into vials containing 1 M of Tris HCl pH 8.0, in order to immediately neutralize the pH. The column was promptly regenerated by washing with Glycine-HCL 0.1 M pH 2.5 and PBS pH 7.4.
Lipid binding specificity assay PIP strips spotted with 100 pmol of different phospholipids were purchased from Echelon Research Laboratories. Membranes were blocked with PBST with 3% (w/v) BSA fatty acid free (Sigma, A7030) for 1 h, at room temperature, with gentle agitation. Blocked membranes were incubated with 0.5 ug/ml of mSNX31:HA or PIP2Grip (positive control from Echelon Research Laboratories) in PBST 3% (w/v) BSA fatty acid free, for 2 h at RT and then washed in PBST 6x over 30 min. Membranes were sequentially incubated with anti-HA (Sigma) or anti-GST (Sigma), respectively, at 1:5000 dilution in PBST 3% (w/v) BSA fatty acid free, for 50 min. Washes were performed as previously described. Membranes were incubated with secondary mouse HRP antibody (Sigma), 1:500 for 45 min also in PBST 3% (w/v) BSA fatty acid free. Following final washing, enhanced chemiluminescence was used to detect binding of HA or GST tagged proteins to phospholipids.

Wortmannin treatment
MDCK cells were transfected with mSNX31 and 48 h posttransfection incubated for different periods of time at 37uC in the presence or absence of 100 nM of wortmannin (SIGMA). Cells were then washed in PBS pH 7.4, fixed in 4% paraformaldehyde and permeabilized in 1% Triton X-100 to pursuit with an immunofluorescence assay.

Biotinylation
In order to biotinylate apical membrane proteins, Swiss Webster female 8-12 weeks old mice were anesthetized with isofluorane, instilled with 75 ml of PBS, and massaged to induce urination. One more wash was performed with 75 ml of biotinylation reagent and the bladder voided again. 75 ml containing Biotinylation sulfo-long chain biotin reagent at 1 mg/ml (2.1 mM) and 75 ml of PBS pH 8 were then added for 15 min. Bladder was voided and sequentially washed with 75 ml of PBS containing 5 mM of lysine. Mice were kept for 0, 30, 90 min and 3 hrs. Bladder was then removed and fixed in 3.7% formaldehyde in PBS pH 7.4 for 2 hr at RT, and processed for paraffin embedding; or fixed with 3% paraformaldehyde, 0.1% glutaraldehyde (v/v) and 4% sucrose in 0.1 M sodium cacodylate buffer, and embedded with Lowicryl K4M for immuno-electron microscopy [45]. Immunofluorescent staining for biotin was then performed with antibody Anti-Biotin (Rockland Immunochemicals, Inc) (TOPO3 for nuclear staining).

Co-immunoprecipitation assay
For immunoprecipitation of SNX31 and UPIIIa an activated aminolink plus resin (Thermo Scientific, Pierce, 26149) was precrosslinked with 60 mg of immunoaffinity-purified SNX31 antibody. Total urothelial cells were lysed in RIPA buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.1% SDS, 0.5% DOC, 1% NP40, supplemented with Roche Proteinase inhibitor) and 300 mg of total proteins were incubated overnight with either: control agarose beads (non activated), quenched resin (200 ml of quenching buffer were added to antibody coupling resin instead of SNX31 antibody) or with aminolink activated resin (all 25 ml beads). The beads were sedimented (5000 g, 2 min) and washed for 6x with cold lysis buffer. The antigens were then eluted with 0.2% SDS sample buffer, analyzed in denaturing-reducing SDS-PAGE and detected with respective antibodies.
In situ proximity ligation assay (in situ PLA) Paraffin sections of wild-type mouse bladder (0.5 mm) were blocked in 1% Fish gelatin in PBS, incubated overnight with the same blocking solution supplemented with antibodies for: UPIb (1:50), UPIIIa (1:5) and SNX31 (1:200). Samples were thereafter subjected to in situ PLA using Duolink Detection kit (Olink, Bioscience, Uppsala, Sweden). Briefly, samples were incubated with PLA probes (secondary antibodies) and the detection was done according to manufacturer's protocol. Slides were mounted with Duolink Mounting media with DAPI and analyzed using a Zeiss LSM 510 confocal microscope [46]. The cell images were exported using LSM software in TIF format for further analysis and the determination of signals performed with the Duolink Image Tool. Signals were normalized with the control signals. Quantifications are given as a number of signals per umbrella cell. Figure S1 Tissue distribution of SNX31 in bovine tissues: 20 mg of total RNAs from bovine urothelium and other tissues were separated by agarose gel electrophoresis and probed for (A) SNX31, (B) glyceraldehyde phosphate dehydrogenase (G), and (C) the 28S and 18S ribosomal RNAs (ethidium bromide (EB) staining; as a loading control). Lanes are: (1) bladder, (2) kidney cortex, (3) kidney medulla, (4) lung, (5) spleen, (6) esophagus, (7) stomach, (8) intestine, (9) brain, (10) liver, and (11) skeletal muscle. Note the bladder urothelium-specific expression of SNX31. (TIF) Figure S2 Uroplakin II-deficient mouse urothelium lacks the uroplakin-plaque lined multivesicular bodies. Note in (A to C) a lack of fusiform vesicles, apical uroplakin plaques, and multivesicular bodies, that are typical of normal urothelial umbrella cells (cf. Figs. 3A and 4A; [11,12]). Rather, the cytoplasm of UPII-deficient superficial cells is filled with small MAL-positive vesicles (SV) that are involved in delivering the remaining UPIb/IIIa uroplakin pair to the apical surface [45]. Note in (D and E) that, despite the lack of MVBs, late endosomes (LE) and lysosomes (Lys) are clearly identifiable. Magnification bars = 1 mm in A-C, and 0.2 mm in D and E. (TIF)