Intralumenal docking of connexin 36 channels in the ER isolates mistrafficked protein

The intracellular domains of connexins are essential for the assembly of gap junctions. For connexin 36 (Cx36), the major neuronal connexin, it has been shown that a dysfunctional PDZ-binding motif interferes with electrical synapse formation. However, it is still unknown how this motif coordinates the transport of Cx36. In the present study, we characterize a phenotype of Cx36 mutants that lack a functional PDZ-binding motif using HEK293T cells as an expression system. We provide evidence that an intact PDZ-binding motif is critical for proper endoplasmic reticulum (ER) export of Cx36. Removing the PDZ-binding motif of Cx36 results in ER retention and the formation of multimembrane vesicles containing gap junction-like connexin aggregates. Using a combination of site-directed mutagenesis and electron micrographs, we reveal that these vesicles consist of Cx36 channels that docked prematurely in the ER. Our data suggest a model in which ER-retained Cx36 channels reshape the ER membrane into concentric whorls that are released into the cytoplasm.

The intracellular domains of connexins are essential for the assembly of gap junctions.For connexin 36 (Cx36), the major neuronal connexin, it has been shown that a dysfunctional PDZ-binding motif interferes with electrical synapse formation.However, it is still unknown how this motif coordinates the transport of Cx36.In the present study, we characterize a phenotype of Cx36 mutants that lack a functional PDZ-binding motif using HEK293T cells as an expression system.We provide evidence that an intact PDZ-binding motif is critical for proper endoplasmic reticulum (ER) export of Cx36.Removing the PDZ-binding motif of Cx36 results in ER retention and the formation of multimembrane vesicles containing gap junctionlike connexin aggregates.Using a combination of site-directed mutagenesis and electron micrographs, we reveal that these vesicles consist of Cx36 channels that docked prematurely in the ER.Our data suggest a model in which ER-retained Cx36 channels reshape the ER membrane into concentric whorls that are released into the cytoplasm.
Electrical synapses serve as a fast means for signal transmission in the nervous system and provide unique functions such as neuronal synchronization or signal averaging (1, 2).Structurally they are defined as gap junctions, specialized cell junctions that contain clusters of intercellular channels establishing a conductive link for ionic currents and metabolites (3).Each gap junction consists of thousands of intercellular channels that are made of membrane proteins called connexins.A characteristic feature of these proteins is their ability to self-assemble into gap junction channels.To form such a channel, six connexins first oligomerize and assemble into a connexon (or hemichannel).Afterward, two opposing connexons on adjacent cell membranes dock via noncovalent interactions of the extracellular loops in each connexin to form a functional gap junction channel.Among all gap junction proteins that have been described in mammals, connexin 36 (Cx36) is regarded as the major neuronal connexin because its expression is virtually (except for the pancreas) (4) confined to neurons (5).Gap junctions that are made of this particular isoform are characterized by their low single channel conductance and their tremendous degree of plasticity (6,7).Especially, neuromodulators and second messenger systems have been shown to exert drastic effects on gap junction coupling via phosphorylation of Cx36 (8,9).Recent studies indicate that active transport and turnover mechanisms of Cx36 function as an additional means to adjust the strength of electrical synapses (10,11).
Like many membrane proteins, connexins follow a characteristic life cycle.Depending on the isoform, they either oligomerize in the endoplasmic reticulum or the Golgi apparatus.From here, newly assembled channels are transported to the plasma membrane by cytoskeleton-dependent mechanisms and added to the periphery of the gap junction.The internalization of gap junctions relies on the formation of so-called annular junctions (12)(13)(14).These structures are formed by the endocytosis of gap junction channels of both cells from the center of the plaque leading to the formation of a double membrane vesicle that is subjected to the proteolytic systems of the cells (15,16).Significant insight into the life cycle of gap junction proteins was gained by imaging studies of recombinant connexins in expression systems.Usually, a GFP-tag (or any other fluorophore) is fused to the intracellular domains of a connexin to monitor transport processes or simply to visualize the protein without an immunolabeling procedure (16)(17)(18).Protein tags, although they have proven to be powerful tools, can compromise protein function depending on their insertion site.Often, they act as a physical barrier and mask certain protein-protein interaction domains, resulting in a loss of association with important binding partners (19).Such effects have been described for Cx43 and Cx36.The expression of a Cx43-GFP fusion protein in which a GFP tag is linked to the C terminus of Cx43 leads to the formation of unusually large gap junctions in HeLa cells (20).For Cx36-EGFP, on the contrary, a C-terminal GFP tag prevents gap junction formation in transfected HeLa cells and results in a partial loss of electrical synapses in transgenic mice (19,21).These phenotypes can be explained by the position of the GFP tag.In each of the constructs, GFP is located in immediate proximity to the PDZ-binding motif, a short amino acid sequence at the C terminus of connexins that is necessary to recruit scaffold proteins containing PDZ domains.The PDZ-binding motif of Cx36 consists of the four C-terminal amino acids: SAYV: serine (S), alanine (A), tyrosine (Y), and valine (V) (22).Due to its proximity to the GFP tag, this sequence is inaccessible for PDZ domains preventing protein-protein interaction at the C terminus, which ultimately results in the transport deficit observed in Cx36-EGFP mice and transfected HeLa cells.These effects can be reproduced with Cx36 mutants that lack the PDZ-binding motif.Although these mutants show some capacity to form gap junctions, they only assemble into intercellular clusters that are significantly smaller than those of WT Cx36.This suggests that the transport deficit of Cx36-EGFP is based on a dysfunctional carboxyl terminus.
Accumulating evidence suggests that the PDZ-binding motif of Cx36 is critical for electrical synapse formation.Still, it is unclear how this sequence coordinates the transport of Cx36.Interestingly, Helbig et al. (19) reported that the Cx36-EGFP fusion protein not only fails to form gap junctions in HeLa cells, but it also seems to assemble into spherical cytoplasmic clusters.Similar observations were made for Cx36 mutants that lack the tubulin-binding motif.Brown et al. (10) have shown that the deletion of this motif (ranging from W277 to S292) interrupts the microtubule-dependent transport of Cx36 and causes the formation of large annular gap junctions.Each of these two studies describe a transport deficit that correlates with the increased formation of intracellular vesicles.How these structures are formed and why they occur in Cx36 mutants with a dysfunctional carboxyl terminus is currently unknown.
In the present study, we describe a transport deficit that is observed in PDZ-binding deficient Cx36 mutants (throughout the paper, referred to as Cx36/S318ter).We identify a mechanism that explains the formation of cytoplasmic connexin aggregates that have been reported in earlier studies and reveal how deleting the PDZ-binding motif affects the intracellular transport of Cx36.Overexpression of the Cx36/ 318ter mutant results in the formation of multimembrane vesicles that originate in the endoplasmic reticulum (ER).Our data suggest that deleting the PDZ-binding motif prevents Cx36 from exiting the ER.Because of this retention mechanism, connexons on adjacent ER membranes are allowed to dock prematurely, leading to the formation of gap junctionlike ER sheets that coil up into multimembrane vesicles called connexin whorls.Here, we reveal an unanticipated function of the classical head-to-head docking mechanism described for gap junction channels and show that it can function as a release mechanism to isolate ER-retained connexins.

Removal of the PDZ-binding motif of Cx36 causes ER retention
Previous studies have shown that the PDZ-binding motif in Cx36 is essential for the formation of electrical synapses.Fusion proteins such as Cx36-EGFP, in which an EGFP tag is fused to the C-terminal tip of Cx36, exhibit severe transport deficits and tend to accumulate in the cytoplasm (19).C-terminal tags are likely to compromise the functionality of the PDZ-binding motif because they block access to the extreme C terminus.Currently, it is unknown how exactly the inhibition of this motif affects the intracellular transport of Cx36 and why it results in the intracellular accumulation of the connexin in expression systems.To test if similar defects occur in vivo, we analyzed the distribution of transgenic Cx36-EGFP in WT and Cx36 KO mouse retinas.In the WT retina, Cx36-EGFP is present in punctate spots in expected locations, particularly in the synaptic inner plexiform layer where Cx36 electrical synapses are abundant (Fig. 1A left).In Cx36-EGFP/Cx36 KO retinas, far fewer Cx36-EGFP gap junctions are present, as was observed in an earlier study (21) as a result of the failure of Cx36-EGFP to traffic properly and form the homologous AII amacrine cell gap junctions.In addition to this defect, we observed Cx36-EGFP-positive aggregates in AII amacrine cell somas that reflect an accumulation of mistrafficked protein (Fig. 1A right and inset).This observation suggests that similar transport deficits occur in neurons such as the AII amacrine cell and simple expression systems as, for instance, HeLa cells.To shed light on the processes causing these defects, we generated a Cx36 mutant that lacks the PDZ-binding motif and a Cx36-SNAP construct in which a SNAP-tag is positioned at the C terminus (Fig. 1B).In an initial experiment, we expressed these constructs in HEK293T cells and compared their distribution to WT Cx36.The deletion of the PDZbinding motif had a striking effect on the distribution of the Cx36/S318ter mutant.This mutant did not assemble into compact perinuclear structures but instead formed large intracellular vesicles that resembled annular junctions (Fig. 1C).A similar phenotype was observed for the Cx36-SNAP construct.We quantified the frequency and size of these vesicles (here and throughout the paper referred to as whorls) and found that both parameters were significantly increased in the mutant and the fusion protein (Fig. 1, D and  E).None of the mutants influenced the frequency of gap junction clusters, whereas deletion of the PDZ-binding motif had a striking effect on the intracellular distribution of the Cx36/S318ter mutant and the Cx36 SNAP construct (Fig. 1F).The Cx36/S318ter displayed around 1.3 whorls per cell and the Cx36-SNAP construct had even more than twice as many.The size of these whorls was also increased in comparison to the WT but there were no size differences between Cx36/S318ter and Cx36-SNAP (Fig. 1E).
During image acquisition, we observed that Cx36 whorls were often quite bright and displayed intensities in the same order of magnitude as gap junctions, which suggests that they contain a high concentration of connexin molecules.One possibility to explain the whorl formation that is induced by the Cx36 mutants is an increased internalization rate.However, Cx36 whorls were rarely associated with gap junctions, but they instead localized around the nucleus and in the cytoplasm.This suggests that these structures are formed as a result of a premature release mechanism but not during internalization.To identify the subcellular compartment in which these vesicles originate, we analyzed the colocalization of Cx36, Cx36/S318ter, and Cx36-SNAP with ER and Golgi markers.Here we observed a striking colocalization of Cx36 whorls and the ER membrane marker GFP-Sec61B (Fig. 1G) (Fig. S1, around 50% colocalization with Cx36/S318ter and 20 % colocalization with Cx36-SNAP).On the contrary, we did not find any obvious differences in the association with the Golgi apparatus for Cx36 and the Cx36/S318ter mutant.These constructs rarely colocalized with the trans Golgi marker Golgin-97 (Fig. 1H).The lack of association with the Golgi and the extensive colocalization of Cx36 whorls with GFP-Sec61B suggest that the disruption of the PDZ-binding motif prevents Cx36 from exiting the ER.The resulting formation of connexin whorls is likely to reflect a spillover mechanism that induces a premature release of connexins from the ER.

Extracellular cysteines are required for whorl formation
Previous studies have shown that overexpression of GFPtagged ER resident proteins results in the generation of multilamellar vesicles that consist of stacked ER membranes.The formation of these structures is triggered by dimerization of GFP.Homophilic interactions of multiple GFP monomers along ER tubules cause the ER to coil up into concentric whorls, eventually leading to their release (23).As connexins have the ability to self-assemble into dense aggregates, we reasoned that a similar mechanism is responsible for the whorl formation we observed here.A premature docking of opposing connexons in the ER might have the same effects and result in a structural reorganization of ER membranes.This mechanism could explain the increased whorl formation we observed for ER-retained Cx36 mutants.To test if the formation of connexin whorls requires a gap junction-like docking mechanism, we substituted individual cysteines (cysteine 55 and 62, Fig. 2A) in the extracellular loop of Cx36 with serines.Previous studies have shown that mutations in this region prevent gap junction formation because they interfere with the intercellular docking of connexons (24).Eliminating the extracellular cysteines should also prevent the formation of ER whorls, provided a similar docking mechanism is responsible.To test this hypothesis, we expressed Cx36 cysteine mutants in HEK293T cells and quantified the frequency of whorls for each construct.As expected, the expression of these mutants entirely prevented gap junction formation but also resulted in a striking reduction of ER whorls (Fig. 2, C and D).For each of the cysteine mutants, we found hardly any whorls.This difference was most distinctive for the ER-retained Cx36/318ter and the Cx36-SNAP construct (Fig. 2E).These constructs displayed the highest number of whorls.Each of the corresponding cysteine mutants, however, formed hardly any whorls, indicating that the formation of connexin whorls requires intact extracellular loops.Surprisingly, we found that all cysteine mutants showed a marked decrease in protein expression (Fig. 2B) and apparent signs of cell damage.Especially, cells that expressed larger amounts of the mutant often had a rounded shape and deformed nuclei.To demonstrate that the formation of connexin whorls is independent of the cell type and mainly driven by functional docking motifs in the extracellular domains, we repeated the experiment in N2A cells.
Here the expression of Cx36-SNAP in N2A cells led to a clear formation of ER whorls that was prevented by transfection of the corresponding cysteine mutant (Fig. 2F).

Ultrastructure of ER whorls
Our initial experiments have shown that the formation of connexin whorls requires the extracellular loop cysteines.Provided that Cx36 whorls are formed by a gap junction-like docking mechanism, it is to be expected that they are identical to actual gap junctions in terms of structure and density.To test this hypothesis, we resolved the ultrastructure of ER whorls and gap junctions in Cx36-SNAP-transfected HEK293T cells using a correlative light and electron microscopy approach.Cx36-SNAP-expressing cells were labeled with tetramethylrhodamine (TMR) and imaged in a live cell imaging set up.Individual regions of interest (ROI) were then traced back and scanned with the high resolution of EM, which allowed us to compare the ultrastructure of whorls and gap junctions (Fig. 3A i-iii ).As expected, we found that Cx36 whorls consisted of multiple membranes (Fig. 3B i ).Each sheet within these whorls contained two parallel membrane layers with densely packed connexins.This configuration had obvious features of gap junctions, which suggests that ER whorls are indeed formed by a similar head-to-head docking mechanism.Interestingly, we also observed that some gap junctions were associated with stacks of intracellular membranes that seemed to consist of docked channels (Fig. 3B ii ).The configuration of this structure was indistinguishable from the gap junction.To further characterize the structure of Cx36 whorls, we reconstructed the entire volume of individual vesicles using serial block-face SEM imaging.These reconstructions illustrated the ellipsoid shape of Cx36 whorls (Fig. 3E i and E iii ) and their complex multimembrane structure (Transparent version in Fig. 3E ii ).In line with the correlative light and electron microscopy (CLEM) data, we observed direct conjunctions between ER sheets, gap junctions, and ER whorls.We labeled these structures with pseudo colors in single serial block-face scanning electron microscopy (SBF-SEM) sections (Fig. 3, C i -D iii , blue: gap junction, yellow: whorl, green: ER sheet) and in 3D reconstructions (Fig. 3E i-iii , color scheme same as in D).These scans showed that ER sheets were often aligned with the gap junction and simultaneously contacting the whorl.Interestingly, we observed a single vesicle that was seemingly about to be removed from the gap junction forming a continuous connection with an ER sheet (Fig. 3, C ii -D ii ; insets in C ii ).In cells with moderate expression of Cx36-SNAP, we often detected typical gap junctions that had no connections to intracellular whorls or ER sheets (Fig. 3F i -F iii ).

Pulse-chase labeling experiments reveal the mechanism of whorl formation
So far, our data have shown that the ability of connexins to self-assemble into dense aggregates serves as a basis for a premature ER release mechanism.The proposed docking mechanism, however, does not explain why connexin aggregates assemble into compact vesicles.To understand the structural changes underlying whorl formation, we conducted live cell imaging experiments.Cx36-SNAP-transfected cells were treated with the SNAP-specific TMR ligand and monitored in a live cell imaging set up.During an acquisition time of 30 min, we did not detect any obvious structural changes of individual whorls in the region of interest, except for minor movements (Fig. 4A).Interestingly, we recorded a vesicle that appeared to be removed from the gap junction (Fig. 4A, magnified inset).This structure was not internalized from the center of the gap junction like an annular junction, but it was seemingly coiling up along the cell membrane.Considering our previous observation, it seems possible that this structure represents a whorl that is about to be formed at the gap junction.To further resolve the time course of whorl formation with improved spatial resolution, we performed pulsechase labeling experiments using two different SNAP ligands.The SNAP-tag can be conjugated with interchangeable fluorescent dyes that form irreversible bonds with the enzyme (25).This technology is well suited for pulse-chase experiments because it allows a differential labeling of Cx36-SNAP proteins that were synthesized at different time points.As a pulse, we incubated Cx36-SNAP-transfected cells with the SNAP-Cell Oregon Green ligand.This initial incubation labels all preexisting SNAP molecules prior to the chase.At different time points after the initial incubation, we applied the TMR ligand  to label Cx36-SNAP molecules that were synthesized after the pulse.This strategy allowed us to distinguish two different generations of Cx36-SNAP proteins using confocal scans.The pulse-chase labeling experiment revealed that ER whorls consist of old and newly synthesized Cx36 proteins (Fig. 4B).We often observed that new material was incorporated to the lumen of preexisting whorls.In line with the SBF-SEM data, we found newer TMR-labeled vesicles in direct association with older Oregon Green-labeled gap junctions.We scanned a single whorl after a 30 min TMR chase using the high resolution of the Airy scan.Here, we found that Cx36 whorls consisted of several layers (Fig. 4C), which is consistent with our previous observation.Newly synthesized material (TMR labeled) always covered the inside of the whorl, whereas older Oregon Green-labeled material formed the outer layer.These observations suggest that connexin whorls are not isolated vesicles, as for instance connexosomes ( 16), but they appear to be dynamic structures that are constantly growing.Given the direct connections between gap junctions and ER sheets, it is likely that a certain percentage of Cx36 whorls is immediately formed at the cell membrane, creating the false impression of an internalization event.

Characterization of cysteine mutants
Our experiments suggest that overexpression of Cx36 cysteine mutant forms results in apparent cell damage causing morphological changes.This raises the question of whether disrupting whorl formation is toxic for transfected HEK293T cells.As connexin whorls function as a removal mechanism, we reasoned that this cell damage is caused by an accumulation of cysteine mutants that are not incorporated into whorls.
To test this, we cotransfected the ER-retained Cx36/S318-ter, Cx36-SNAP, and the corresponding cysteine mutants with the luminal ER marker DsRed2 ER5.We found that each of the cysteine mutants displayed strong colocalization with DsRed2 ER5 (Fig. 5, A ii , A iv , B ii and B iv ).Due to the low expression of these mutants, we had to increase the imaging laser intensity in these conditions.A similar degree of colocalization was observed when Cx36/S318-ter-and Cx36-SNAP-transfected cells were imaged with the same laser intensities.However, we worked with lower intensities in these conditions to avoid an oversaturation of connexin whorls.Although Cx36 whorls originate from the ER, they did not seem to colocalize with the luminal ER marker DsRed2 ER5 (Fig. 5, A i , A iii , B i and B iii ).This can be explained by the fact that docking of connexins in the ER lumen displaces the luminal content leaving no space for the ER marker protein (26).
To determine the extent of cell death, we carried out cell vitality assays.A live/dead assay was used to visualize dead cells.Although Cx36/C62S-SNAP transfectants often appeared to be damaged based on cell morphology and shape of their nuclei, our live/dead assay showed no apparent differences in the fraction of dead cells for each condition (Fig. 5C i ).In Cx36-SNAP and Cx36/C62S-SNAP, we observed around 10% of dead cells.This number was not affected by the cysteine mutant.We also tested for upregulation of ER stress markers and determined the level of phosphorylated eukaryotic initiation factor alpha (elf2α) in cells that were transfected with the cysteine mutant (Fig. 5C ii ).Here we did not observe any changes in the phosphorylation state of elf2α when we compared with Cx36-SNAP-and Cx36/C62-SNAP-expressing cells.As a positive control, we induced ER stress applying 2 mM DTT for 1h.This treatment, unlike overexpression of Cx36/C62-SNAP, caused an increase in the phosphorylation of elf2α (Fig. 5C ii ).
Another possibility to explain the increased cell damage we have observed for the cysteine mutants is the formation of leaky hemichannels.Previous studies have shown that disruption of extracellular cysteines in Cx43 is correlated with an increase in hemichannel activity (27).It is possible that the C62S substitution has similar effects on the permeability of Cx36 connexons and leads to aberrant channel activity resulting in increased uptake of extracellular Ca 2+ , which could be sufficient to induce apoptosis.To exclude this possibility, we measured the channel activity of the cysteine mutant and compared the ethidium bromide uptake of Cx36-SNAP and Cx36/C62S-SNAP transfectants.To identify cells that express the fusion proteins, we incubated the samples with SNAP Cell Oregon Green prior to the measurement.The ethidium bromide assay showed no differences in dye uptake between the two conditions (Fig. 5D ii ).For the two constructs, we observed the same of extent dye uptake after 5 min of perfusion, which is likely to originate from endogenous hemichannel activity.In line with these observations, surface biotinylation assays showed a relative decrease in the surface expression of the Cx36/C62S-SNAP mutant (actin was used as a negative control to exclude cytoplasmic contaminants) (Fig. 5D ii ).This confirms that aberrant hemichannel activity of plasma membrane-localized Cx36 mutants is unlikely to induce the cell damage we have seen in prior experiments.Although the overall dye uptake was similar for both constructs, we cannot exclude that the C62S mutation influences the permeability of Cx36 connexons since the Cx36/C62-SNAP mutant exhibited reduced surface expression, making direct comparisons impossible.

Cx36 whorls associate with ER phagy receptors
The preceding experiments have drawn a precise picture of how Cx36 is assembled into ER whorls.To understand how these structures are further processed and potentially degraded, we carried out a BioID screen using a recently described TurboID-destabilized GFP nanobody (dGBP) construct, which consists of a conditionally dGBP and TurboID.This strategy was developed as a modular system by Xiong et al. (28), enabling the targeting of TurboID towards any given GFPtagged protein of interest.In order to apply this tool for the identification of whorl-associated proteins, we co-expressed Cx36-EGFP as a target molecule alongside Cx36-SNAP.This strategy allowed us to direct the V5-TurboID-dGBP construct to connexin whorls that have incorporated Cx36-EGFP (Fig. 6A).To validate the specificity of the strategy, we tested streptavidin labeling in cotransfected cells 3h after biotin (50 μM) supplementation.Here, we found that V5-dGBP-TurboID and Cy3-Streptavidin reactivity (Fig. 6A) was mainly restricted to Cx36-EGFP-containing ER whorls and gap junctions to a lesser extent, suggesting that the nanobody construct is correctly targeted.To induce proximity biotinylation for a large scale pull down, we treated transfected cells with 50 μM biotin for 3 h and performed the pull-down afterward.The pull-down results were confirmed by testing streptavidin reactivity in the eluted samples (Fig. 6B).A sister gel was incubated with a V5 antibody (V5 epitope tag on the N terminus of TurboID-dGBP) to visualize the isolated fusion protein.A Cx36 antibody was used to confirm streptavidin affinity capture of Cx36-SNAP and Cx36-EGFP.Both proteins were highly abundant in the precipitates of the pulldown and occurred as a double band (Upper band: Cx36-EGFP, lower band: Cx36-SNAP, straight arrow: oligomer).The eluates of the BioID assay were subjected to mass spectrometry to screen for potential interactors of Cx36.Among the proteins that were identified in the precipitates (Fig. 6A), we detected the testis- expressed protein 264 (Tex264) and the p62 protein (Gen: pSQSTM1) (the p62 protein was also present in the negative control, but as it has a ubiquitous function for protein degradation, it is not surprising that it occurred in a complex with V5-dGBP-TurboID).The precipitates were also tested for Tex264 on western blots, which showed that this protein was only enriched in Cx36-EGFP/Cx36-SNAP/V5-TurboID-dGBP-transfected samples, suggesting a specific association with Cx36.These findings were of particular interest for us as both candidates function as ER phagy receptor molecules, an autophagosomal pathway that degrades fragments of the ER to maintain ER homoeostasis.To validate these hits, we tested the localization of p62 and Tex264 in HEK293T cells that were transfected with Cx36, Cx36/318ter, and Cx36-SNAP.In this experiment, we observed abundant colocalization between Tex264 and Cx36-SNAP or Cx36/S318ter (Fig. 6C).Tex264 labeling was weak throughout the cell but concentrated at ER whorls.Even whorls that were formed in Cx36 WT transfectants colocalized with Tex264.Likewise, the p62 protein colocalized with all three Cx36 variants.Among other candidate proteins that were identified by BioID, we also detected the endosomal protein rab11.Unlike, p62 and Tex264 however, it only showed little colocalization with Cx36-containing ER whorls (Fig. S2B).The specific association of Cx36-containing ER whorls with two autophagy receptors indicates that these structures are targeted by autophagosomes.Thus, ER whorls appear to represent a point of no return from which nonfunctional Cx36 channels are subjected to the proteolytic systems of the cell.The interaction of the ER phagy receptor Tex264 with Cx36 whorls confirms that these structures are formed in the ER.

Discussion
In the present study, we demonstrate that ER retention of Cx36 results in the formation of multimembrane vesicles containing gap junction-like connexin aggregates.Our data suggest a model in which premature docking of Cx36 connexons reshapes the ER membrane into concentric whorls that are released into the cytoplasm (Fig. 7).These findings are supported by site-directed mutagenesis and electron micrographs revealing similar structural profiles for gap junctions and connexin whorls.This type of whorl formation in cell culture appears to be an overexpression artefact that occurs because transfected cells produce enormous amounts of proteins that exceed the capacity of the ER export machinery.We even detected connexin whorls in Cx36 WT transfectants, which suggests that not all the synthesized proteins can leave the ER by a conventional export mechanism.This illustrates that the formation of connexin whorls is a concentrationdependent phenomenon.Each of the two Cx36 variants, Cx36/S318-ter and Cx36-SNAP, have an increased propensity to form whorls because these constructs are trapped in the ER, which ultimately promotes interactions between proximal connexons.The reduced gap junction formation and accumulation of intracellular aggregates of Cx36-EGFP in retinal AII amacrine cells in a Cx36 KO mouse background further argues that this mechanism functions in vivo.Another important conclusion that can be drawn from our observations is that Cx36 is most likely to oligomerize in the ER, as only fully assembled connexons are capable of forming gap junctions.Hence, Cx36 follows a similar pattern as the beta group connexins Cx32 and Cx26 and forms hemichannels before reaching the Golgi apparatus (29)(30)(31)(32).
Connexin whorls with similar features to the ones described here were reported for a cataract-associated mutant of the lens connexin Cx50.In this mutant, proline 88 is substituted by serine, which causes a conformational change that blocks the release of Cx50 from the ER.It has been suggested that Cx50containing ER whorls act as light scattering particles that are detrimental for the function of the lens.To study the mechanism of whorl formation, Lichtenstein et al. (26) used pulsechase labeling and found that Cx50 whorls are formed by the addition of newly synthesized proteins to the outer layer of the vesicle.Our experiments suggest, to the contrary, that newly synthesized protein is constantly added to the inside of the whorl, whereas older Cx36 molecules form the outer layers.These differences may be ascribed to the type of tag that was used in each study.Lichtenstein et al. (2009) (26) used a flash tag consisting of four cysteine, whereas our experiments were conducted with the SNAP-tag.Also, it is possible that intermolecular interactions between the cytoplasmic domains of the connexin affect the orientation of adjacent ER sheets while the whorl is being formed.Hence, there are a couple of factors that could explain the differences in the whorl formation mechanism.Still, in case of both connexins, Cx50 and Cx36, it is evident that ER retention and a premature docking are the main causes of whorl formation.
DsRed2ER5 (magenta) and the cysteine mutants (green) overlaps entirely.C i , representative confocal images of Cx36-WT-SNAP-and Cx36-SNAP/C62Stransfected cells stained with live/dead viability staining kit.NucBlue Live reagent: stains the nuclei of all cells; detected with a standard 4 0 ,6-diamidino-2phenylindole (DAPI) filter.NucGreen Dead reagent: stains only the nuclei of dead cells; detected with standard GFP (green) filter set.Scale bar represents 20 μm.Quantification of dead cells 24 h post staining with live/dead dyes.Plot with SEM; Mann-Whitney test (two-tailed).C ii , lysates of Cx36-SNAP-and Cx36-SNAP/C62S-transfected HEK293T cells were tested for phosphorylated elf2α which serves as an ER stress marker.No changes in elf2α phosphorylation were detected between Cx36-SNAP-and Cx36-SNAP/C62S-transfected cells.As a positive control, we induced ER stress in nontransfected cells using 2 mM DTT for 60 min.This treatment resulted in increased phosphorylation of elf2α.D i , EtBr dye uptake assays demonstrate that substitution of C62S does not affect the permeability of transfected HEK293T cells.Representative confocal images of EtBr uptake in Cx36-SNAP-and Cx36/C62S-SNAP-expressing cells labeled with SNAP-Cell Oregon Green dye.Scale bar represents 10 μm.Bar graph quantification of total increase in fluorescence 5 min after EtBr application.Only cells expressing Cx36 were used for dye uptake analysis.Plot with SEM; Mann-Whitney test (two-tailed).D ii , cell surface biotinylation assay displaying levels of Cx36-SNAP/C62S and Cx36-SNAP at the membrane.Total cell lysates (input) show expression of Cx36 and intracellular ß-actin.The streptavidin pull-down fractions (surface) show a reduction in membrane levels of Cx36-SNAP/C62S when compared to Cx36-WT.ß-actin served as an internal control.Bar graph with mean and SD; unpaired t test (two tailed), **< 0.01.n = 3 experiments.N.S., not significant; ROI, regions of interest; elf2α, eukaryotic initiation factor alpha; EtBr, ethidium bromide.Our ultrastructural examinations revealed that ER sheets were often closely connected to gap junctions and Cx36 whorls.These sheets exhibited similar structural profiles as gap junctions suggesting that they were formed by docked connexons.Comparable structures were mentioned in a review by Segretain and Falk (15), who reported that rough ER sheets are tightly associated with gap junctions in HeLa cells.Their exact function is unknown, but it was proposed that these structures function as a reserve pool from which newly synthesized channels can transfer to the plasma membrane by a diffusion-based mechanism.However, it is unlikely that this is the case for the structures we have observed in this study.The proximity of these sheets and their alignment along the plasma membrane suggests that the gap junction is attracting ER-incorporated Cx36 channels through an unknown mechanism.As already pointed out before, it is possible that homophilic interactions between the cytoplasmic domains of Cx36 channels on both sides (gap junction versus ER) lead to such an alignment.It also cannot be excluded that certain scaffold proteins at the gap junction interact with Cx36 in ER tubules that happen to be close enough.Both scenarios we have outlined are based on the abundance Cx36 in transiently transfected cells and it is unlikely that such events occur at synapses.Nevertheless, these data essentially indicate that Cx36 gap junctions can attract intracellular connexins, thereby promoting the formation of connexin whorls at the gap junction.

Mistrafficked connexin removal
ER whorls have been observed in different settings and serve as a stress response that is critical to maintain ER homeostasis (23,33).In a process termed ER-phagy, whorls are released from the ER and selectively degraded by autophagosomes to ensure the timely removal of potential stressors (34).The extensive association between Cx36 and the receptor molecules p62 and Tex264 implies that the degradation of connexin whorls requires the same mechanism.The p62 protein is an autophagic cargo adapter that can recruit ubiquitinated proteins into autophagosomes (35,36).Indeed, previous studies have shown that Cx36 is ubiquitinated by the ubiquitin ligases Lnx1 and Lnx2 (37).These enzymes might provide the correct degradation signal for p62 to bind Cx36 and attract the expanding autophagosome.Like p62, Tex264 interacts with the autophagosomal protein LC3 to direct substrates into the phagophore.While p62 has been described as a ubiquitous autophagy receptor, Tex264's function appears to be more ERspecific.Deletion of Tex264 causes a substantial inhibition of ER phagy, although the different ER phagy receptors are known to work cooperatively (38).Additionally, compared to the other receptors, Tex264 has the highest affinity for LC3, indicating that it might be the most efficient receptor involved in the degradation of Cx36 whorls.
Although there is no in vivo evidence for the existence of connexin whorls, it is noteworthy that these structures share features with cytoprotective functions that have been reported for other systems.Connexin whorls bear a resemblance to structures such as aggresomes (39) in the sense that they concentrate nonfunctional proteins that might otherwise interfere with important cell functions.Moreover, our data suggest that the formation of connexin whorls functions as a  mechanism to expose these proteins to a degradative pathway for cells to avoid an overabundance of protein in ER.Since overexpression systems have obvious disadvantages, it is difficult to predict whether connexin whorls are formed under in vivo conditions and whether they might be beneficial for cell survival.To test a potential cytoprotective function of connexin whorls, we performed cell viability assays and compared the frequency of dead cells between Cx36-SNAP-and Cx36-SNAP/C62S-transfected HEK293T cells.These tests revealed neither apparent differences in cell viability nor an upregulation of ER stress markers, which seems to exclude a cytoprotective function.However, we often observed that cells expressing larger amounts of the cysteine mutants displayed deformed nuclei and a rounded shape, which are typical indicators of apoptosis and cell detachment.It is important to point out that each of the cysteine mutant variants we have generated was expressed at a much lower level.It seems possible that those cells that expressed larger amounts of the mutant died immediately whereas cells producing lower concentrations survived.Our viability assays might have only included low expressing cells without any obvious signs of cytotoxicity, which could explain these conflicting results.The differences in the expression levels, however, may as well arise from different turnover rates.Pulse-chase experiments have shown that Cx50 whorls in HeLa cells are long lived and exhibit an estimated half-life time of 67 h, which is about 20 times longer than that of gap junction incorporated Cx36 (26,40).Since ER whorls contain multiple layers of densely packed connexins, they may be less prone to degradation, which would extend their half-life time and ultimately increase the expression level of Cx36.
Given the reduced surface expression of the Cx36-SNAP/ C62 mutant, we cannot entirely rule out that the C62S substitution has no effect on the permeability of connexons in the plasma membrane.
Furthermore, as the ER itself functions as a Ca 2+ reservoir, it is possible that ER-localized cysteine mutants trigger a release of Ca 2+ from the ER into the cytosol resulting in apoptosis.
Our data indicate that the PDZ-binding motif of Cx36 is critical for the export from the ER.The deletion of the SAYV sequence resulted in a striking phenotype that was characterized by the formation of ER whorls and an aggregation of Cx36 vesicles around the nucleus.This finding apparently raises the question of whether there is a certain PDZ protein that interacts with Cx36 in the early secretory pathway to control the delivery of connexons to the gap junction.Indeed, PDZ domain containing scaffold proteins such as syntenin1 have been shown to mediate the ER export of pro-TGF-α (41).A similar mechanism could apply to Cx36, which would explain the increased ER retention we observed for the Cx36/ S318ter mutant lacking the PDZ-binding motif.However, our data do not make a case that the PDZ ligand of Cx36 functions as a sort of ER export signal.Such export signals are often quite short and consist of one or two amino acids (42).It is possible that certain amino acids within in the sequence of the PDZ-binding motif constitute a separate export signal.Previous studies have shown that a single C-terminal valine residue at position −1 (C-terminal tip) is sufficient to drive COPIIdependent ER export of the major histocompatibility complex, class I, F (HLA-F) (43).Interestingly, valine is also the C-terminal amino acid of Cx36, suggesting that a similar export mechanism could be involved.The main task of ER export signals is to mediate an interaction between the cargo protein (Cx36) and specific cargo receptors, so called Sec24 proteins.These proteins recruit the cargo into COPII vesicles, which are subsequently released from the ER and transported to the Golgi (44).It is possible that the Cx36/318ter mutant is unable to interact with these cargo receptors because it is missing the corresponding binding motif.This would ultimately prevent the incorporation of the mutant into export vesicles resulting in ER retention and subsequent whorl formation.Surprisingly, we observed that the Cx36/318-ter mutant still formed gap junctions although a huge percentage of the connexin was trapped in the ER.This suggests that there must be an alternative mechanism to export Cx36 from the ER.Meyer et al. have previously reported that neurons such as the AII amacrine cell use different assembly mechanisms for gap junctions with different synaptic partners.One feasible explanation is a second export signal that might be sufficient to partially sustain the transport of the mutant.However, further studies will be required to elucidate the ER export mechanisms of Cx36.
The present study demonstrates that a premature docking of connexons functions as removal mechanism compartmentalizing Cx36 channels that are retained in the ER.Further studies will be necessary to confirm if such a mechanism operates in neurons to avoid an overabundance of connexins in the ER.As neurodegenerative diseases such as Parkinson's or Alzheimer have been suggested to involve ER stress (45), connexin whorls could represent a mechanism to support ER homoeostasis in stressed cells.

Cell culture and transient transfection
Human embryonic kidney 293T cells (HEK293T/17; catalog #CRL-11268; ATCC; RRID: CVCL_1926) were cultivated in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1% penicillin and streptomycin, and 1% nonessential amino acids (all Thermo Fisher Scientific) at 37 C in a humidified atmosphere with 5% CO2.For the cell surface biotinylation assay, 800,000 cells were seeded in 60 mm plates.For live cell imaging, 25,000 cells were seeded in 35 mm glass-bottom dishes (MatTek Corporation).For live microscopy experiments, the medium was changed to DMEM lacking phenol red 30 min prior to imaging.HEK293T cells were transiently transfected with Effectene Transfection Reagent Kit (Qiagen Inc.) according to the manufacturer's guidelines.Cells were transfected with a total of 1000 ng of DNA for each 60 mm plate.For 35 mm glassbottom dishes, cells were transfected with 200 ng of DNA.All assays were performed 48 h posttransfection.For immunocytochemical experiments, 270,000 to 450,000 cells were plated on to 35 mm dishes a day before transfections.One microgram of DNA were transfected using 5 μl Geneporter2 (Genlantis) or 1.5 μl Lipofectamine 2000 (Thermo Fisher Scientific).

Pulse-chase experiments
For pulse-chase labeling experiments, Cx36-SNAP transfectants were treated with 6 μM SNAP-Cell Oregon Green (New England Biolabs Inc.) in 10% FBS DMEM for 30 min at 37 C to quench all Cx36-SNAP molecules.Afterward, coverslips were washed three times in serum-free DMEM and incubated for 30 min at 37 C.At different time points after the initial incubation (0.25 h, 0.5 h, 1 h, 2 h, 3 h), the cells were incubated in 3 μM SNAP-Cell TMR-Star (New England Biolabs Inc.) in 10% FBS DMEM for 30 min and embedded using Vectashield (supplemented with 4 0 ,6-diamidino-2-phenylindole).

Cell surface biotinylation assay
HEK293T cells were seeded on 60 mm plates and transfected with Cx36-WT-SNAP or Cx36-C62S-SNAP.Biotinylation assay was performed 48 h posttransfection.Cells were washed once with PBS containing both calcium and magnesium and labeled with 0.5 mg of membraneimpermeable EZ-linkTM Sulfo-NHS-Biotin (Thermo Fisher Scientific) per plate for 30 min at room temperature.Plates were washed three times, 5 min each, with 50 mM glycine buffer to quench the reaction.Cells were then washed with PBS lacking calcium and magnesium and lysed with IP Lysis buffer (Thermo Fisher Scientific) supplemented with protease inhibitor cocktail kit (Thermo Fisher Scientific).Cell lysates were incubated overnight with 90 μl of Dynabeads MyOne Streptavidin C1 (Invitrogen) on a shaker at 4 C.The next day, beads were collected on a magnet and washed using the following buffers: twice with buffer 1 (2% SDS in dH 2 O), once with buffer 2 (0.1% deoxycholate, 1% Triton X-100, 500 mM NaCl, 1 mM EDTA, 50 mM Hepes pH 7.5), once with buffer 3 (250 mM LiCl, 0.5% NP-40, 0.5% deoxycholate, 1 mM EDTA, 10 mM Tris; pH 8.1), and twice with buffer 4 (50 mM Tris, 50 mM NaCl pH 7.4).To elute proteins and disrupt beadprotein complexes, beads were boiled for 5 min in 60 μl of 1×Laemmli buffer.

Dye uptake
HEK293T cells were seeded on 35 mm plates and transfected with Cx36-WT-SNAP or Cx36-C62S-SNAP.SNAP-Cell Oregon Green (New England Biolabs Inc.) was used to label and visualize Cx36-expressing cells.The labeling stock was dissolved in 50 μl of dimethylsulfoxid and diluted to 1:200 in complete clear medium to yield a labeling concentration of 5 μM.Cells were incubated in SNAP-tag labeling medium at 37 C, 5% CO 2 for 30 min.Cells were washed once and incubated in fresh clear medium for 30 min prior to imaging.Dishes were placed in a live-cell imaging chamber at 37 C with CO 2 and acquired using Zeiss LSM 700 confocal microscope under EC Plan-Neofluar 40×/1.30Oil M27 objective.Ethidium bromide was added to a final concentration of 10 μM immediately prior to imaging.Images were taken every 15 s for a duration of 5 min at a 1024 × 1024 pixel resolution.Total ethidium bromide uptake values were used for statistical analysis and were calculated by subtracting the initial fluorescence at T = 0 min from the final fluorescence at T = 10 min.Only cells expressing Cx36-SNAP were used for analysis.Images were analyzed using ImageJ.

Imaging
For live cell imaging experiments, 2.7 × 10 5 HEK293T cells were plated on 30 mm MatTek dishes and 24 h later transfected with 1 μg of Cx36-SNAP.At the next day, these cells were treated with 3 μM SNAP-Cell TMR-Star (NEB) for 30 min at 37 C and washed in 10% FBS DMEM for 30 min.Afterward, the DMEM was substituted with Ames medium (Sigma).Live cell imaging experiments were conducted with a Nikon Eclipse Ti microscope using a 60× objective.After TMR labeling, HEK293T cells were placed in an incubation chamber kept at 37 C, 5% CO 2 and imaged for 30 min in 30 s intervals.Fluorescence images were acquired with a Zeiss LSM800 confocal microscope using an 60× objective with a numerical aperture of 1.3.

Image analysis
The frequency and size of connexin whorls were measured with ImageJ.The number of whorls was quantified for individual cells (z-stack of 10 × 0.2 μm) using the cell counter plugin.Intracellular vesicles without a recognizable lumen were excluded from quantification.These were usually vesicles with an inner diameter smaller than 0.3 μm.A linear ROI and the measure function were used to determine the inner diameter of individual whorls.For colocalization analysis of Cx36 and Golgi97, cells were scanned with a Leica TCS SP8 confocal microscope using a 63×/1.4oil objective, at a pixel size of 45 nm.Image stacks (8 μm) were subsequently processed in Fiji (46): first, stacks were thresholded using the Moments algorithm of the Auto Threshold function.Then, colocalized pixels were identified using the Colocalization Highlighter.In the resulting 8 bit image, four ROI (ROI = 3.01 × 3.01 μm 2 ) were placed on the colocalized puncta and the colocalized area was measured using the Analyze Particle function which was set to count only puncta larger than 0.125 μm 2 .Subsequently, the same ROIs were placed on the Golgi97 image stacks, and again, the stained area was measured.From these data, the percentage of colocalization with Cx36 was calculated and averaged for each transfected construct (four ROIs analyzed in each stack, three stacks per coverslip, two coverslips per construct).In total, three independent transfection experiments were analyzed and the resulting data were compared by a two tailed Mann-Whitney test.

Conventional CLEM
HEK293T cells (1 × 10 5 ) were seeded in 35 mm μ-Dishes with a gridded polymer coverslip bottom (Cat.No.81166; Ibidi) coated with 0.01% poly-L-lysine.Next day, cells were transfected (FuGene) with either pCx36-SNAP or pCx36 C62S-SNAP.On the third day, cells were stained with 100 nM SNAP-Cell TMR-Star (New England BioLabs) for 30 min at 37 C and chemically fixed with a mixture of 4% paraformaldehyde and 0.1% glutaraldehyde in 0.2 M Hepes buffer, pH 7.4, for 15 min at 37 C. ROIs were observed with a confocal laser scanning microscope (Leica TCS SP5) and the software LAS AF (Leica).After imaging, cells were fixed with 2.5% glutaraldehyde (Science Services) in 0.2 M Hepes, pH 7.4, for 1 h at RT and postfixed with 1% osmium tetroxide (Science Services) in 0.1 M cacodylate buffer, pH 7.4, for 2 h on ice.After washing, cells were dehydrated in a graded ethanol series (30%, 50%, 70%, 90%) and finally two rinses in anhydrous ethanol and three rinses in anhydrous acetone at RT. Cells were infiltrated and flat-embedded in graded mixes of acetone and Epon 812 (Sigma).After resin polymerization, the gridded polymer bottom was removed, and the coordinates were transferred to the resin surface allowing trimming.Serial 70 nm sections were cut with an ultramicrotome (Leica EM UC7RT) and collected on formvar-coated EM copper slot grids.After automated staining with uranyl acetate and lead citrate (Leica EM AC20), samples were observed via either Zeiss TEM 902 A (50 kV, TRS slow-scan 2K CCD camera, software ImageSP by TRS image SysProg) or Jeol TEM JEM2100plus (200 kV, emsis XAROSA CMOS TEM camera, software RADIUS by Emsis).Stitching and overlay of CLSM and TEM images was done using Photoshop (Adobe).

Serial block-face scanning electron microscopy CLEM
After light microscopy, HEK293T cells selected for SBF-SEM were fixed in 2.5% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.4.Subsequently, samples were processed via adapted version of the NCMIR rOTO-postfixation protocol (47) and embedded in hard Epon resin, ensuring pronounced contrast and electron dose resistance for consecutive imaging.All procedures were performed on the Ibidi μ-Dish.In brief, after fixation, samples were postfixed in 2% osmium tetroxide (Science Services) and 1.5% (w/v) potassium ferrocyanide (Riedel de Ha€ en) in cacodylate buffer for 1 h on ice.The cells were then incubated in 1% (w/v) thiocarbohydrazide (Riedel de Ha€ en) in dH 2 0 for 20 min, followed by an additional 2% osmication step in water for 30 min at room temperature.After washing in dH 2 0, samples were submerged in 1% aqueous uranyl acetate overnight at 4 C. Cells were then incubated in freshly prepared Walton's lead aspartate (Pb(NO 3 ) 2 (Carl-Roth), L-Aspartate (Serva), KOH (Merck)) for 30 min at 60 C. Subsequently, cells were dehydrated through a graded ethanol (Carl-Roth) series (30%, 50%, 70%, and 90%) on ice for 7 min each, before rinsing in anhydrous ethanol twice for 7 min and twice in anhydrous acetone (Carl-Roth) for 10 min at room temperature.Afterward, cells were infiltrated in an ascending Epon:acetone mixture (1:3, 1:1, 3:1) for 2 h each, before an additional incubation in hard mixture of 100% Epon 812 (Sigma).Final curation was carried out in hard Epon with 3% (w/w) Ketjen Black (TAAB) at 60 C for 48 h.Once polymerized, the μ-Dish bottom was removed via toluene melting from the resin block including the attached cells and finder grid imprint.ROIs were trimmed and the sample blocks were glued to aluminum rivets using two-component conductive silver epoxy adhesive and additionally coated in a 30 nm thick gold layer.The rivet containing the mounted resin block was then inserted into the 3View Gatan stage, fitted in a Jeol JSM 7200F, and aligned parallel to the diamond knife-edge.The cells proved to be stable under imaging conditions of 3.1 kV accelerating voltage, high vacuum mode of 10 Pa, utilizing a 30 nm condenser aperture and a positive stage bias of 600 V. Imaging parameters were set to 6 nm pixel size, 3.4 μs dwell time, in between ablation of 60 nm and an image size of 10240 x 10240 pixels.Overall, an approximate volume of 60 × 60 × 19 μm (à 320 slices) was acquired.Image acquisition was controlled via Gatan Digital Micrograph software (Version 3.32.2403.0;https://www.gatan.com/products/tem-analysis/gatan-microscopy-suite-software).Further postprocessing, including alignment, filtering and segmentations were performed in Microscopy Image Browser (Version 2.7); (48).To keep the computational resources to a manageable limit, the final stacks were binned before exporting the files for volumetric visualization in Amira 3D (Version 2021.1,Thermo Fisher Scientific).

Statistical analysis
Statistical analysis and data presentation were performed using GraphPad Prism 8 (https://www.graphpad.com/).Values reported consist of mean ± SEM.Normality was tested using the Anderson-Darling and the D'Agostino-Pearson test.Significance was tested using a two-tailed t test, two-tailed Mann-Whitney U test, or a Kruskal-Wallis test for multiple comparisons.
acknowledges funding from the DFG (RTG 1885/2 Molecular Basis of Sensory Biology) and the European Union under the action of ERA-NET NEURON (JTC2020: Rethealthsi), financed by the German Federal Ministry of Education and Research (BMBF, 01EW2107).L. B. & V. L. were funded by the Deutsche Forschungsgemeinschaft (DFG iBiOs (No.PI 405/14-1), SFB 944 Z-Project).The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflicts of interests-The authors declare that they have no conflicts of interests with the contents of this article.

Figure 1 .
Figure 1.Removal of the PDZ-binding motif of Cx36 and C-terminal SNAP-tag leads to the formation connexin whorls.A, comparison of Cx36-EGFP expression in Cx36 WT and Cx36 KO retinas.Cx36-EGFP expression in the Cx36 KO leads to accumulation of the fusion protein in somas of AII amacrine cells.

Scale bar represents 20
μm. Magnified inset: 10 μm.B, illustration of different Cx36 mutants used in this study.C, perinuclear distribution of WT Cx36 in transfected HEK293T cells.Removal of the PDZ-binding motif or the addition of a C-terminal SNAP in Cx36 results in the formation of large annular inclusions (whorls).Scale bar represents 10 μm.Inset: 2.5 μm.D, frequency of intracellular whorls per cell.Whorls were counted in 9 to 11 cell clusters from two independent transfections.Compared to WT Cx36, Cx36/S318ter-and Cx36-SNAP-transfected cells show a significant increase in whorl formation.p < 0.05, *.Values are mean with SEM and significance was determined by a Kruskal-Wallis test followed by a post hoc Dunn's analysis.E, average size of connexin whorls in all three Cx36 variants.Cx36 whorls are increased in size in Cx36/S318ter-and Cx36-SNAP-transfected cells.Eleven to thirty whorls were measured per condition.p < 0.05 *.Significance was tested using a Kruskal-Wallis followed by a post hoc Dunn's analysis.F, frequency of gap junctions per cell.p > 0.05.Ten to eleven cell clusters from two independent transfections were quantified per condition.Values are mean with SEM and significance was determined by a Kruskal-Wallis test followed by a post hoc Dunn's analysis.G, colocalization of Cx36 whorls and the ER membrane marker Sec61B-GFP.Scale bar represents 10 μm.Inset 2.5 μm, (H) Cx36 and Cx36/S318ter show no differences in association with the Golgi marker Golgin-97.p = 0.1345.Scale bar represents 10 μm.Inset: 2.5 μm.Colocalization was quantified in an entire stack.Values are mean ± SD and were obtained from three independent transfection experiments.Significance was determined by a two tailed Mann-Whitney test.N.S., not significant.

Figure 2 .
Figure 2. Mutations of cysteines in the extracellular loop of Cx36 prevent whorl formation.A, cartoon illustrating the positions of the extracellular cysteines in Cx36.B, protein levels of cysteine mutants in transfected HEK293T cells.Arrows indicate the monomer, and asterisks indicate the dimer.Each of the cysteine mutants exhibits a marked reduction in Cx36 expression.C, HEK293T cells expressing WT Cx36 and the Cx36/S318ter mutant.Substitution of cysteines 55 or 62 results in a diffuse distribution of Cx36 in the cytoplasm.D, cysteine mutations in the Cx36/ S318ter or the Cx36-SNAP construct prevent whorl formation.Images are presented as maximum projections of 10 slides (2 μm).Scale bar represents 10 μm.Inset: 2.5 μm.E, frequency of intracellular whorls per cell.Whorls were counted in 9 to 11 cell clusters from two independent transfections.Compared to Cx36/S318ter and Cx36-SNAP, the corresponding cysteine mutants show a significant decrease in whorl formation.p < 0.05.Values are mean with SD and significance was determined by Kruskal-Wallis test followed by a post hoc Dunn's analysis.F, the formation of Cx36 containing ER whorls independent of the cell type is used in the experiment.Cx36-SNAP but not Cx36-SNAP/C62S transfected N2A cells form whorls. Scale bar represents 10 μm.Inset: 2.5 μm.

Figure 3 .
Figure 3. CLEM and SBF-SEM reveal the ultrastructure of connexin whorls.A i -B iii , fluorescence signals were correlated with electron micrographs.These experiments revealed that connexin whorls are multimembrane vesicles.Each layer within these vesicles consisted of two adjoined membranes.This configuration was indistinguishable from the structure of gap junctions.C i -D iii , single slices showing electron micrographs of gap junction (blue), ER whorls (yellow), and ER sheets (green).ER whorls are directly connected to gap junction via membranes of the ER.E i -E iii , 3D reconstruction of gap junctions and ER whorls via serial block face imaging.Note that ER sheets are continuous with ER whorls and come in close contact with the gap junction.F i -F iii , 3D reconstruction of a gap junction that is not connected to intracellular ER sheets.Length of scale bar is indicated in each panel.

Figure 4 .
Figure 4. Live cell imaging and pulse-chase experiments reveal the mechanism of Cx36 whorl formation.A, Cx36-SNAP expressing HEK293T cells were labeled with TMR and imaged in a live cell imaging chamber.Images are shown in 5 min intervals.Scale bar represents 10 μm.Inset: 2.5 μm.B, Cx36-SNAP-transfected HEK293T cells were pulsed with Oregon Green and chased with TMR at different time points: 15 min, 30 min, 1 h, 2 h, 3 h.Scale bar represents 10 μm.Inset: 2.5 μm.C, airy scan of pulse-chased whorl.Two different planes are shown.TMR-labeled material (magenta) is covering the inside of the whorl.A 3D reconstruction demonstrates that older Oregon Green-labeled protein is forming the outer layers of the whorl.Scale bar represents 2.5 μm.n = 3 independent transfections.

Figure 5 .
Figure 5. Characterization of Cx36 cysteine mutants.A i -A iv , cotransfections of Cx36 mutants and ER-localized DsRed2ER5 reveal that the cysteine mutants Cx36/C62S/S318ter and Cx36-SNAP/C62S are almost entirely confined to the ER.Whorls formed by Cx36/S318ter or Cx36SNAP transfectants do not colocalize with dsRed2ER5.Scale bar represents 10 μm.Magnified insets: 2 μm.B i -B iv , line scan depicting the intensity along each ROI.Intensity profile of

Figure 6 .
Figure 6.Cx36 associates with the p62 protein.A, cartoon illustrating whorl-specific proximity ligation.Co-expression of Cx36-SNAP and Cx36-GFP leads to the incorporation of Cx36-GFP into large connexin whorls.The GFP-tag on Cx36-EGFP functions as a bait to direct the V5-dGBP-TurboID to connexin whorls, allowing biotinylation of whorl-associated proteins.A heat map illustrates the abundance of proteins that were identified via BioID.Confocal scans were used to confirm that V5-GBP-TurboID is directed to Cx36 whorls.Scale bar represents 5 μm.B, Western blots confirming streptavidin affinity capture.Streptavidin-HRP was used to detect biotinylated proteins.A V5 antibody was used to detect the V5-dGBP-TurboID construct.Cx36-EGFP and Cx36-SNAP were co-isolated.The Cx36 antibody detects two bands in the eluates of Cx36-EGFP+ Cx36-SNAP condition (indicated with arrows).Upper band: Cx36-EGFP; lower band: Cx36-SNAP.n= 3 independent pull downs.C, colocalization of Tex264 and p62 with different Cx36 variants in transfected HEK293T cells.Scale bar represents 10 μm.Inset: 2.5 μm.n= 3 transfections.dGBP, destabilized GFP nanobody.

Figure 7 .
Figure 7. Cartoon illustrating the whorl formation mechanism.Connexin whorls are the result of ER retention.This retention promotes premature docking of connexons in the ER causing the formation of whorls.