Cross-over endocytosis of claudins is mediated by interactions via their extracellular loops

Claudins (Cldns) are transmembrane tight junction (TJ) proteins that paracellularly seal endo- and epithelial barriers by their interactions within the TJs. However, the mechanisms allowing TJ remodeling while maintaining barrier integrity are largely unknown. Cldns and occludin are heterophilically and homophilically cross-over endocytosed into neighboring cells in large, double membrane vesicles. Super-resolution microscopy confirmed the presence of Cldns in these vesicles and revealed a distinct separation of Cldns derived from opposing cells within cross-over endocytosed vesicles. Colocalization of cross-over endocytosed Cldn with the autophagosome markers as well as inhibition of autophagosome biogenesis verified involvement of the autophagosomal pathway. Accordingly, cross-over endocytosed Cldns underwent lysosomal degradation as indicated by lysosome markers. Cross-over endocytosis of Cldn5 depended on clathrin and caveolin pathways but not on dynamin. Cross-over endocytosis also depended on Cldn-Cldn-interactions. Amino acid substitutions in the second extracellular loop of Cldn5 (F147A, Q156E) caused impaired cis- and trans-interaction, as well as diminished cross-over endocytosis. Moreover, F147A exhibited an increased mobility in the membrane, while Q156E was not as mobile but enhanced the paracellular permeability. In conclusion, the endocytosis of TJ proteins depends on their ability to interact strongly with each other in cis and trans, and the mobility of Cldns in the membrane is not necessarily an indicator of barrier permeability. TJ-remodeling via cross-over endocytosis represents a general mechanism for the degradation of transmembrane proteins in cell-cell contacts and directly links junctional membrane turnover to autophagy.


Introduction
Tight junctions (TJs) are mainly formed by claudins (Cldns), transmembrane proteins that play a major role in limiting and regulating paracellular permeation in epithelia and endothelia. Cldns are tetraspanning membrane proteins, with both termini located in the cytoplasm, PLOS

Microscopy
To determine vesicle numbers, fixed cells were imaged using a confocal microscope (Zeiss NLO with a Plan-Neofluar 100x 1.3 Oil objective, Zeiss, Oberkochen, Germany). Z-stacks with 0.25 μm intervals between slices through the whole cell were evaluated using ImageJ (http:// imagej.nih.gov/ij/). To automatize the vesicle count, an ImageJ macro was created. Fluorescence thresholds were set to 70 for YFP and TRQ and to 40 for Alexa 647. Vesicles with a radius of 50 nm < r < 500 nm were counted in each slice and mean vesicle numbers per slice were calculated. Cross-over endocytosed vesicle numbers were normalized to the total length of the contacts shared with cells expressing the other fluorophore, while total vesicle numbers were normalized to the cell perimeter. For each experiment, vesicle numbers of treated cells were normalized to untreated samples. For live cell imaging, cells were seeded on glass coverslips (r = 15 mm) coated with PLL. Cocultures of monotransfected MDCK-II cells were washed with HBSS+/+ (Hanks balanced salt solution with Ca 2+ and Mg 2+ ; Thermo Fisher Scientific, #14065049) once and imaged in fresh HBSS+/+.
For fluorescence resonance energy transfer (FRET) measurements by photoacceptor bleaching, HEK-293 cells were transiently transfected with TRQ/YFP pairs, as indicated in the figures. After transfection, the cells were grown on glass coverslips for 2 d before measurement. Increase in TRQ-signal was measured after 10 cycles of YFP-bleaching. YFP/TRQ ratios were determined to have a linear effect on FRET-values between ratios 0.5 and 8, only cell-contacts with these ratios were used. FRET efficiencies were normalized to the YFP/TRQ ratio and then to TRQ-Cldn5/YFP-Cldn5.
Fluorescence after photobleaching (FRAP) measurements were performed in cocultures of MDCK-II cells monotransfected with TRQ-Cldn5 or YFP-Cldn5 (and accordingly the F147A or Q156E variants). Before bleaching, 5 pictures were taken of the contact to reach a stable starting intensity. In YFP/TRQ cell-contacts YFP was bleached in an area approx. 10% of the contact length (1 ms pulse, 100% laser intensity) and the signal intensity measured in 5 s intervals for 220 s. Curves were normalized to the range of bleached signal. Mobile fractions were determined by a fitted curve (Graphpad Prism5, http://www.graphpad.com/scientificsoftware/prism/).
For stimulated emission depletion (STED) images, the Leica TCS 3X gSTED, Leica LASX software and HC PL APO CS2 100x 1.4 Oil objective (Leica, Wetzlar, Germany) were used. TRQ was excited at 405 nm and emission was detected from 440 nm-513 nm. YFP was excited at 514 nm and emission was collected from 519 nm-581 nm. Emissions from both proteins were collected on HyD detectors in the time gated mode (0.3-6 ns) in the presence of a 592 nm depletion laser.
Transepithelial electrical resistance (TER) and paracellular permeation measurements TER was measured using Cellzscope (nanoAnalytics, Münster, Germany). MDCK-II cells were seeded on hanging PET cell culture inserts (Merck Millipore, #PIHT12R48) at a density of 80,000 cells/insert. After TER values had reached a stable plateau, the cells were transferred to a 24-well plate and washed twice with HBSS+/+ warmed to 37˚C. Lucifer yellow (m.w. 457 Da; Sigma-Aldrich, #L0259) was dissolved in warm HBSS+/+ (200 μM) and 200 μl solution were applied to the apical compartment of the insert, in order to determine the permeability coefficient of the cell barrier [23]. The basolateral compartment was filled with 1ml warm HBSS+/+ and the plate was incubated at 37% with 10% CO 2 for 10 min. The inserts were then moved to a fresh 24-well plate containing prewarmed HBSS+/+ and incubated for two more 10 min periods. From each basolateral compartment, 80 μl of sample were transferred to a 96-well plate in triplicate and Lucifer yellow fluorescence was measured using a Tecan microplate reader (Tecan, Männedorf, Switzerland); concentrations were determined from a standard curve.

Modeling
Modeling of murine Cldn5 and its mutants was performed using Iterative Threading ASSEmbly Refinement (I-TASSER) [24]. Amino acids 1 to 190 of murine Cldns were included in the modeling (transmembrane domains and extracellular loops, excluding the charged amino acids C-terminally flanking the fourth transmembrane domain) and the models of the mutants were aligned with the wild type model using PyMOL (https://www.pymol.org/).

Statistics
Data was analyzed using the Kruskal-Wallis test (one way ANOVA) followed by Dunn´s multiple comparison post-test or using a two tailed t test. Statistics and graphs were created using GraphPad Prism5. Further details are given in the figure legends; results are shown as mean ±SEM as indicated. If not stated otherwise, differences were considered significant if p<0.05.

Amino acid substitution in extracellular loop 2 of claudins influences their interactions and barrier properties
To elucidate the impact of Cldn-Cldn interactions on cross-over endocytosis, murine Cldn5 with amino acid (aa) substitutions in the ECL2 was transfected into MDCK-II cells. Substitutions Cldn5 F147A and Cldn5 Q156E were previously found to impair homophilic trans-interactions; cis-interactions as measured by fluorescence energy transfer (FRET) were not affected when coexpressed with wildtype (wt) Cldn5 [2]. After seeding, the transepithelial electrical resistance (TER) of MDCK-II cells developed a tightness curve which was characterized by a peak followed by a plateau. TER was increased in cells expressing Cldn5 wt compared to untransfected cells similar to earlier findings [22]. This increase was significant at the beginning of the plateau 90 min post seeding (Fig 2A, from 36.5±0.9 to 68.3±1. OÁcm 2 ). Cldn5 F147A and Cldn5 Q156E substituents showed lower TER values (44.9±0.8 and 47.9±0.9 OÁcm 2 , respectively) than Cldn5 wt transfected cells 90 h post seeding and at the peak (untransfected: 132.7±10.7, wt: 163.0±12.7, F147A: 73.1±2.4, Q156E: 61.3±4.0 OÁcm 2 ). These reduced values indicate that the positions F147 and Q156 are involved in the paracellular tightening function of Cldn5. Moreover, the peaks reached by Cldn5 F147A and Cldn5 Q156E was significantly lower than the peak reached by untransfected cells, which indicates that Cldn5 F147A and Cldn5 Q156E interfere with the function of endogenous Cldns. There was also a significant delay in the peak for Cldn5 F147A and a still greater delay for Cldn5 Q156E (Fig 2A). The barrier permeability for lucifer yellow (457 Da) was nearly unchanged for Cldn5 F147A transfected cells, while transfection of Cldn5 Q156E resulted in a strong increase in the permeability compared to Cldn5 wt -transfected cells (Fig 2B). That means Cldn5 F147A resulted in an opening for molecules smaller than 457 Da only.
FRET measurements revealed reduced cis-interactions for Cldn5 F147A /Cldn5 F147A , and for Cldn5 Q156E /Cldn5 Q156E . However, Cldn5 F147A was more severely affected ( Fig 2C). The membrane mobility of Cldn5 Q156E was slightly enhanced compared to that of wt Cldn5, while Cldn5 F147A showed an increased mobility in the plasma membrane as measured by fluorescence recovery after photobleaching (FRAP). Cldn5 displayed a mobile fraction of 19%, Cldn5 Q156E 25%, and Cldn5 F147A 51% ( Fig 2D). Cldn5 F147A and Cldn5 Q156E transfection led to lower peaks in TER measurement and, in case of Cldn5 Q156E , to higher permeability compared to untransfected cells (Fig 2A). This suggests that Cldn5 F147A and Cldn5 Q156E are incorporated into the TJs and interfere with the function of endogenous Cldns.

Impaired claudin-claudin interactions lead to reduced cross-over endocytosis
Modeling with I-TASSER revealed slight conformational changes in the secondary structure of the aa-substituents, i.e. elongation of the third and fourth β-strand in the ECL1 (Fig 3A,  arrows). In the surface model, changes in the binding interface became visible and are visualized by a white circle. Cldn5 Q156E led to a charged residue being introduced into the binding interface ( Fig 3B, right), which explains the altered binding properties. In Cldn5 F147A , only slight surface changes occurred ( Fig 3B, middle), consistent with the unchanged barrier properties measured by TER (Fig 2A). However, in the models of Cldn5 F147A and Cldn5 Q156E , the conformation of the aromatic side chains of Y148 and Y158 was altered (Fig 3C and 3D). F147 and Y148 are part of an aromatic binding interface involved in trans-interactions, Y158 is also involved in cis-interactions [2]. The conformational changes induced by the amino acid substitutions F147A and Q156E on Y148 could be responsible for the reduction of cis-interactions measured by FRET ( Fig 2C).
When transfected into MDCK-II cells, Cldn5 is mainly located in the plasma membrane, but many vesicles can be found intracellularly ( Fig 4A). Approximately 25% of these vesicles contained cross-over endocytosed Cldn5 (Fig 4B). Cldn5 F147A and Cldn5 Q156E were also located in the plasma membrane ( Fig 4A) but there were more total intracellular vesicles as compared to Cldn5 wt (Fig 4A, arrows; Fig 4C, right) while the number of cross-over endocytosed vesicles was dramatically decreased (Fig 4A, arrowheads; Fig 4C, left). Cldn5 F147A expressing cells were more severely affected than Cldn5 Q156E expressing cells regarding vesicle numbers. Cldn5 F147A showed less cross-over endocytosed and more total vesicles than Cldn5 Q156E (Fig 4C). This elevated total vesicle count may reflect a higher rate of single membrane endocytosis and recycling. In the case of Cldn5 F147A , vesicles seem to colocalize more frequently with Rab11 than wt Cldn5-containing vesicles ( Cross-over endocytosis is affected by chlorpromazine, an inhibitor of the clathrin pathway, and filipin III, an inhibitor of the caveolin pathway MDCK-II cells were treated with inhibitors of various pathways of endocytosis (Fig 5). Vesicles containing TRQ-Cldn5 were counted in TRQ-Cldn5-transfected cells (total vesicle number) and in YFP-Cldn5-transfected cells (cross-over endocytosed vesicles). For total vesicles, a decrease was found after treatment with all inhibitors tested, however that with dynasore (Dyn) was a trend only (Fig 5). The strongest (approx. 50% reduction) and statistically significant effect was observed after treatment with chlorpromazine (CPZ), an inhibitor of clathrin adaptor protein 2 (AP-2) [25]. Inhibition of caveolae formation using filipin III (F3) [26] led to decreased vesicle numbers within 1 h after treatment, and inhibition of dynamin using Dyn [27] resulted in a non-significant reduction in vesicle numbers. For TRQ-Cldn5 vesicles crossover endocytosed into YFP-Cldn5 transfected cells, a strong decrease in vesicle numbers was seen after treatment with CPZ and F3. F3 reduced the vesicle numbers transiently after 1 h of treatment, while CPZ led to a longer lasting reduction. Treatment with Dyn did not lead to a significant change in the number of cross-over endocytosed vesicles. A similar scenario is known from D. melanogaster during elongation of cell-cell contacts. Addition and local removal of adherence junctions was accompanied by accumulation of the clathrin machinery at the junctions, but Dyn inhibition had no effect on junctional remodeling [28].

Lysosomal inhibitor chloroquine leads to accumulation of cross-over endocytosed claudins
In live cell imaging experiments, vesicles containing TRQ-Cldn were frequently observed in YFP-Cldn expressing cells; however, fewer YFP-Cldn positive vesicles were visible in TRQ-Cldn cells (Fig 6A, arrows). This observation may reflect the pH-sensitivity of YFP [29], if cross-over endocytosed Cldns are transported to cellular compartments with acidic environment, such as lysosomes. Treatment with Bafilomycin A1 (baf, an inhibitor of the V-H + -ATPase) which prevents endosomal acidification, led to a time-dependent increase of cross-over endocytosed YFP-Cldns (Fig 6A, arrows). After 3 h incubation with baf there was an 8-fold increase in crossover endocytosed YFP-intensity, while there was no significant increase in TRQ-intensity. In fixed samples, the number of cross-over endocytosed TRQ-Cldn5 vesicles was increased after inhibition of lysosomal degradation using chloroquine (CQ, a weak, cell permeable base that leads to alkalinization of lysosomes, Fig 6B). This effect was observed for Cldn5 wt , as well as Cldn5 F147A and Cldn5 Q156E cells. However, much greater accumulation of cross-over endocytosed vesicles occurred for Cldn5 wt than for Cldn5 F147A and Cldn5 Q156E (Fig 6B, left). This accumulation supports the involvement of the lysosomal degradation pathway for cross-over endocytosed Cldn5 and confirms that there is less cross-over endocytosis for interactiondefective Cldn5 mut . In untreated MDCK-II cells, cross-over endocytosed Cldn5 was found in vesicles containing the lysosomal marker lysosomal-associated membrane protein 1 (Lamp1) (Fig 6C), which confirms the contribution of the lysosome in the degradation of cross-over endocytosed Cldns. However, there were also Lamp1-negative vesicles containing cross-over endocytosed Cldn5, which indicates an intermediate step in the trafficking of these vesicles.

Cross-over endocytosed vesicles colocalize with autophagosomal markers
Next, the prelysosomal Lamp1-negative vesicles containing cross-over endocytosed Cldn5 were further characterized. The endocytosis of integral membrane proteins from two neighboring cell membranes indicates a double-membrane morphology of the vesicle, reminiscent of the structure of autophagosomes. To test the involvement of the autophagosome, cocultured cells expressing either TRQ-Cld5 or the autophagosomal marker microtubule-associated protein 1A/1B-light chain 3 (LC3) [30], tagged with red and green fluorescent protein (RFP-GFP-LC3), were observed under live conditions. This construct monitors autophagic flux: in the cytoplasm and non-acidic autophagosomes both fluorophores of the tandem LC3 are visible, while in compartments with low pH, such as autolysosomes, only RFP is functional [31], as GFP is quenched at low pH [32]. Cross-over endocytosed TRQ-Cldn5 vesicles in RFP-GFP-LC3-transfected cells frequently colocalized with the RFP-signal alone (Fig 7A, arrows), and with both RFP and GFP, prominently near the plasma membrane (Fig 7A, arrowheads). Colocalization with GFP indicates the presence of TRQ-Cldn5 in autophagosomes before acidification, i.e. before fusion with the lysosome. Immunostaining of fixed YFP-Cldn5/TRQ-Cldn5 cocultures of monotransfected MDCK-II cells against autophagy related protein 16L (ATG16L) revealed a colocalization with cross-over endocytosed Cldn5 (Fig 7B, arrows) as well as with Cldn5 at the cell membrane (arrowheads).
Phosphoinositide 3-kinase (PI3K) and phosphatidylinositol-3-phosphate 5-kinase (PIKfyve) have been implicated in the formation of autophagic vesicles [33,34]. Inhibition of PI3K using LY294002 or of PIKfyve using YM-201636 resulted in a reduction in cross-over endocytosed vesicles by 35%, which further supports involvement of the autophagic machinery in the cross-over endocytosis of Cldns.

Super-resolution microscopy reveals double-membrane structure of cross-over endocytosed vesicles
Stimulated emission depletion (STED) microscopy was applied to visualize the membrane topology of cross-over endocytosed Cldns in endosomes of subdiffraction size in living cells. Cross-over endocytosis of claudins Imaging cocultured TRQ-Cldn5 and YFP-Cldn5 transfected cells using gated STED microscopy with approximately 50 nm resolution [35] showed clear separation of cross-over endocytosed Cldns and Cldns expressed by the observed cell within the same vesicle ( Fig  8A). In untreated cells, vesicles containing TRQ-Cldn5 were measured in YFP-Cldn5 transfected cells. The intensity profile across the vesicle clearly showed that YFP was distributed in the outer membrane of the vesicle, while cross-over endocytosed TRQ-Cldn5 was located inside the vesicle (Fig 8A, right). After treatment with the lysosomal inhibitor CQ, larger  YFP-Cldn5-containing structures became visible, with accumulation of TRQ toward the side of the vesicle (Fig 8B, arrows). As CQ leads to the dilatation of lysosomes [36], the outer membrane of cross-over endocytosed vesicles corresponds to the lysosomal membrane. The diameter of the outer membrane of cross-over endocytosed vesicles in untreated cells ranged between 400 and 900 nm (n = 9), while vesicles in CQ-treated cells were larger (0.6 to 2 μm, n = 8).
The results observed by STED microscopy were confirmed by a fluorescence proteinase protection (FPP) assay. In this approach, the plasma membrane (but not organelle membranes) of live cells was selectively permeabilized and proteinase K was added to the cells. Fluorophores directed toward the cytoplasm were digested, while those contained inside vesicles were protected from degradation. In YFP-Cldn5 transfected cells containing cross-over endocytosed TRQ-Cldn5, only YFP-Cldn5 was noticeably digested after 25 min of exposure to proteinase K (Fig 8C, arrows), while TRQ-Cldn5 was not. This confirms the differential orientation of Cldn termini. While fluorophores from neighboring cells were located inside the vesicle and protected from digestion, fluorophores of Cldns derived from the observed cell were oriented toward the cytosol (Fig 9A).

Discussion
In this study, a direct involvement of the autophagic machinery in TJ-remodeling is demonstrated, and for the first time the presence of Cldns is shown in large double membrane vesicles. Our results confirm a type of endocytosis of Cldns, where TJ-associated Cldn-oligomers are cross-over endocytosed into the neighboring cell in double-membrane vesicles in a clathrin-and caveolin-dependent manner. Moreover, cross-over endocytosis depends on Cldn interactions.
The ECLs of Cldns strongly interact with Cldns of neighboring cells; interactions between Cldn1/Cldn1, Cldn5/Cldn5 or Cldn1/Cldn5 have been shown to reach affinity values in the nanomolar range [18]. This very high affinity leads to strong adhesion between cells and effectively hinders their separation. The oligomerization of Cldns across the paracellular cleft is mediated by their ECLs [2,22]. Amino acid exchange in the ECL2 of Cldn5 in the positions F147 and Q156 leads to decreased trans-interactions, but was thought not to influence cisinteractions when Cldn5 F147A or Cldn5 Q156E was measured against Cldn5 wt [2]. However, we now find that cis-interactions were compromised in a FRET assay, when Cldn5 F147A / Cldn5 F147A or Cldn5 Q156E /Cldn5 Q156E pairs were analyzed. The earlier measurements for substitution/wt pairs show that impaired cis-interactions can be rescued by association with other Cldns. This means that other parts of the molecule compensate for the deficit in binding, probably the ECL1 which is well known to contribute to paracellular tightness [37].
In MDCK-II cells endogenous Cldn1 is expressed, which is able to interact strongly with Cldn5 in cis and trans [17]. However, the transfection of Cldn5 F147A or Cldn5 Q156E interferes with barrier properties compared to untransfected cells, indicating that the endogenously Cross-over endocytosis of claudins expressed Cldns were not able to rescue the defects in trans-interactions. Interestingly, Cldn5 F147A showed greatly increased mobility within the membrane, and cis-interactions and cross-over endocytosis are decreased more strongly than for Cldn5 Q156E . In contrast, the permeability of the cell layer is strongly increased in Cldn5 Q156E -expressing cells, while cis-interactions are moderately affected and mobility is not affected. This suggests that the mobility of Cldns in the membrane depends on Cldn-Cldn-interactions, but does not necessarily have an impact on barrier tightness.
Permeability changes demonstrate that the substituents are incorporated into the TJs however they show less cross-over endocytosis than Cldn5 wt . This is evidence that only the highly adhesive sections of the neighboring membrane are pulled into the neighboring cells. This adhesiveness is proven by testing the permeability, an equivalent of trans-associations closing the intercellular cleft. Our results also suggest a correlation of cis-interactions and cross-over endocytosis, as the substituent F147A, which exhibits weaker cis-interactions than Q156E, is also less cross-over endocytosed than Q156E. Colocalization of endogenous Cldn1, Cldn2, Cldn7 and occludin reveals that this mechanism of endocytosis is not targeting specific types of TJs-proteins for endocytosis, but rather specific parts of the TJs in which the proteins are complexed.
After endocytosis, Cldns can either be recycled back to the plasma membrane, or targeted for degradation [38]. Our experiments show that cross-over endocytosed Cldns are delivered to the lysosome using the autophagosomal machinery. Recently, it has been reported that the plasma membrane can be a membrane donor for autophagosome formation: Atg16L, a protein involved in the initiation of autophagosome formation, interacts with clathrin heavy chain to form autophagosomal precursors at the plasma membrane [39]. We find Atg16L to colocalize with Cldn5 in cross-over endocytosed vesicles as well as at the plasma membrane. The presence of GFP-LC3, which indicates an early autophagosome, in cross-over endocytosed vesicles confirms a role of the autophagosomal machinery in the induction of TJ-remodeling. This is further supported by the reduction of cross-over endocytosis after inhibition of PI3K and PIKfyve, kinases involved in autophagosome biogenesis [33,34]. Autophagy plays a crucial role in maintaining cellular homeostasis, by the removal of large particles and aggregates from the cytosol. The phagophore (precursor of the autophagosome) is derived from the single-membranous endoplasmic reticulum, among other sources [40]. It elongates and encloses intracellular particles, thereby forming typical double-membrane vesicles that fuse with lysosomes to deliver their cargo for degradation [41]. In contrast, cross-over endocytosis does not seem to involve such single-membranous stages as the phagophore. The molecular basis of the biogenesis of autophagosome-like vesicles that do not develop from a phagophore remains elusive. However, the effects of PI3K and PIKfyve inhibition suggest that phosphatidylinositol phosphates are involved in the regulation of this mechanism, possibly by specifying the part of the TJ targeted for endocytosis. This mechanism differs from conventional endocytosis in a critical way: instead of a single membrane being endocytosed, a patch of the TJ containing two cell membranes is removed as a double-membrane vesicle into one of the cells. In super-resolution images, clear separation of the Cldns from both cells is visible within the vesicles, which indicates that both plasma membranes are incorporated. Cldns derived from the cell containing the vesicle are directed toward the cytosol, while the termini of cross-over endocytosed Cldns of the neighboring cell are directed toward the lumen of the vesicle (Fig 9A). In a fluorescence protease protection (FPP) assay, cross-over endocytosed TRQ-mCld5 accordingly was protected from digestion, as only fluorophores in the cytosol are accessible to the proteinase. The luminal position of cross-over endocytosed Cldn is also supported by the disappearance of the cross-over endocytosed YFP-signal. As YFP is sensitive to an acidic environment [29], its loss of fluorescence after cross-over endocytosis is due to the localization of the fluorophore inside the acidic lysosomal compartment. Accordingly, inhibition of lysosomal acidification by bafilomycin rescued YFP-fluorescence in our experiments. The colocalization of cross-over endocytosed Cldns with the autophagosomal marker LC3 might be interpreted as engulfment of the cross-over endocytosed vesicle by an autophagosome. However, the FPP assay implies that this is not the case, as both fluorophores would be protected from digestion if surrounded by an autophagosomal membrane. This conclusion is supported by the finding that the distribution cross-over endocytosed TRQ-Cldn5 signal is not affected by inhibition of lysosome function by chloroquine, while the outer membrane containing YFP-Cldn5 from the same cell grows in size ( Fig  8B). Double membrane vesicles have been described in electron microscopy and assumed to contain Cldns. However, the presence of Cldns in these vesicles was not confirmed so far [19].
Our results now prove this assumption.
Inhibition of dynamin does not lead to a reduction in cross-over endocytosis, however inhibition of clathrin/AP-2-interaction and cholesterol depletion (filipin III treatment) does. Therefore scission of the nascent cross-over vesicle at the plasma membrane does not depend on dynamin, but the interaction with clathrin/AP-2 and caveolin/cholesterol has an impact on cross-over endocytosis. A similar scenario is known from in Drosophila during the elongation of cell-cell contacts. Addition and local removal of adherence junctions is accompanied by an accumulation of the clathrin machinery at the junctions but a dynamin inhibition has no effect on junctional remodeling [28]. Consequently further dynamin-independent mechanisms have to be considered. Thus, chlorpromazine causes relocation of clathrin and AP-2 from coated pits to endosomes [25]. However, the interaction of clathrin and AP-2 with ATG16L at the plasma membrane regulates the initiation of autophagy [39]. In this manner the clathrin pathway connects autophagy with cross-over endocytosis. The role of the caveolin pathway in autophagy is also documented: Cholesterol depletion leads to the induction of autophagy [42], as does caveolin knockdown [43]. Filipin III binds to cholesterol in the membrane and inhibits the interactions between cholesterol and caveolin, leading to inhibition of caveolae formation [26]. In addition to inhibition of endocytosis, filipin III can therefore reduce the number of cross-over endocytosed vesicles by increasing autophagic clearance of cross-over endocytosed Cldns.
Cross-over endocytosed vesicles contain various Cldn-subtypes as well as occludin. These are rather substantial and unspecific changes in TJ morphology, whereas the removal of specific Cldns from the TJ without eliminating the barrier function must be a strictly regulated process. Under nitric oxide stress, Cldn5 has recently been shown to be delivered for autolysosomal degradation in a caveolin-mediated process [44]. In the intestine autophagy has been shown to enhance barrier function by specifically down-regulating Cldn2 [45]. However, our results indicate, that the specificity of this down-regulation may not be achieved through selective internalization of a specific Cldn subtype, but rather by increased incorporation of other Cldns into the TJs after endocytosis of Cldn5-or Cldn2-rich TJ sections, respectively (Fig 9B). In addition, Cldn5 can be degraded by the proteasome pathway following ubiquitination, but also by an ubiquitin-independent mechanism [13], which still need to be clarified with respect to cross-over endocytosis.
The targeting of Cldns via a clathrin-dependent mechanism shares similarities with the internalization of some gap junction proteins. Connexin 31.1 can also be targeted by clathrin for degradation, via autophagy, in cell lines of non-small cell lung cancer [46]. For other connexins, the existence of small double-membrane vesicles has been proposed in cervix carcinoma cells [47]. In MDCK-II cells, cross-over endocytosed Cldn1 and -3 have been suggested to be involved in wound healing [19]. For adherence junctions, cadherin-dependent uptake of cell material from the neighboring cells has been described in fibroblasts [48]. This suggests a general mechanism for the endocytosis of transmembrane proteins that feature very strong affinity to proteins in neighboring cell membranes, such as Cldns [18]. This is further supported by the finding that, at the neuron-to-glia interface, the transmembrane EphB2 receptor is also cross-over endocytosed from neurons to neighboring glia cells [49].
In summary, our data illustrate the mechanism of TJ remodeling involving autophagy, where strongly adhering Cldns (via their extracellular loops) are not separated before endocytosis but internalized in an autophagosome-like double membrane vesicle (Fig 9B). In this process, both the interacting proteins and membranes from two neighboring cells are cross-over endocytosed into either cell. This vesicle then fuses with the lysosome, and the inner membrane (i.e. the membrane derived from a neighboring cell) is digested while the outer membrane becomes part of the lysosomal membrane. Cldn molecules not incorporated in the TJs, e.g. in the apical membrane, would be available for conventional endocytosis via the endosomal pathway (Fig 9B), while this is not possible for double membrane vesicles. Reports on other junctional proteins and transmembrane receptors support the assumption of a general mechanism for reorganizing cell-cell contacts. Collectively, the findings help to understand the highly dynamic behavior of cell-cell contacts, and demonstrate that autophagy is not only a pathological phenomenon but also of physiological relevance in cell-cell interaction across various cell types.