The Habc domain of syntaxin 3 is a ubiquitin binding domain

Syntaxins are a family of membrane-anchored SNARE proteins that are essential components required for membrane fusion in eukaryotic intracellular membrane trafficking pathways. Syntaxins contain an N-terminal regulatory domain, termed the Habc domain that is not highly conserved at the primary sequence level but folds into a three-helix bundle that is structurally conserved among family members. The syntaxin Habc domain has previously been found to be structurally very similar to the GAT domain present in GGA family members and related proteins that are otherwise completely unrelated to syntaxins. Because the GAT domain has been found to be a ubiquitin binding domain we hypothesized that the Habc domain of syntaxins may also bind to ubiquitin. Here, we report that the Habc domain of syntaxin 3 (Stx3) indeed binds to monomeric ubiquitin with low affinity. This domain binds efficiently to K63-linked poly-ubiquitin chains within a narrow range of chain lengths but not to K48-linked poly-ubiquitin chains. Other syntaxin family members also bind to K63-linked poly-ubiquitin chains but with different chain length specificities. Molecular modeling suggests that residues of the GGA3-GAT domain known to be important for ionic and hydrophobic interactions with ubiquitin may have equivalent, conserved residues within the Habc domain of Stx3. We conclude that the syntaxin Habc domain and the GAT domain are both structurally and functionally related, and likely share a common ancestry despite sequence divergence. Binding of Ubiquitin to the Habc domain may regulate the function of syntaxins in membrane fusion or may suggest additional functions of this protein family.

SNARE (soluble N-ethylmaleimide-sensitive factor attachment protein receptor) proteins are the indispensable mediators of membrane fusion reactions within the endomembrane system of eukaryotic cells [1][2][3][4][5] . The SNARE superfamily consists of several sub-families whose members contain one or two SNARE domains of ~ 60 residues in length 3,4,6 . Members of the syntaxin family of SNAREs are central to the formation of SNARE complexes. They contain a C-terminal transmembrane anchor, preceded by the SNARE domain. The latter engages in interactions with cognate SNAREs to form a SNARE complex in a 4-helix bundle arrangement that ultimately leads to membrane fusion [7][8][9] .
Syntaxins also contain an N-terminal regulatory domain that consists of three α helices (a, b, c) and has been termed the H abc domain. At least 16 syntaxins are encoded in the human genome and many more in divergent species. Amongst these syntaxins, the H abc domains are poorly-or not at all-conserved at the primary sequence level. However, in the cases of syntaxin family members whose H abc domains have been structurally studied it was found that they all share a highly conserved fold. This includes Stx1A 10,11 , Stx6 12 , Sso1 13 , Stx10 14 , Vam3p 15 and several others (see Protein Data Bank) whose H abc domains fold into essentially superimposable three-helix bundles despite limited or absent sequence similarity. The H abc domains of various syntaxins have been found to be binding sites to proteins that regulate SNARE function including those of the munc18, munc13 and synaptotagmin families 16 . In addition, in some-but not all-syntaxins the H abc domains have the ability to engage in an intramolecular interaction with the SNARE domain resulting in a tetrameric helical bundle. This "closed" conformation generally inhibits the formation of complexes with cognate SNARE proteins and thereby inhibits membrane fusion 16,17 . These findings clearly indicate that the H abc domains of syntaxins play a critical Scientific Reports | (2020) 10:21350 | https://doi.org/10.1038/s41598-020-78412-0 www.nature.com/scientificreports/ role in the regulation of membrane fusion and that this function depends on the conserved three-dimensional structure of these domains. It was found that a conserved domain in a very different family of proteins shares the same fold with the syntaxin H abc domain. The GAT (GGAs and TOM) domain of GGA1 was found to be nearly superimposable with the H abc domains of Stx1A and Stx6 despite their lack of sequence similarity 18 . GGA proteins are Golgi-and endosome-associated clathrin adaptor proteins involved in cargo recruitment in membrane trafficking pathways. At the time of the discovery of the structural similarity with the syntaxin H abc domain, relatively little was known about the function of the GAT domain. Subsequently, however, the GAT domains of GGA proteins were found to be ubiquitin binding domains [19][20][21][22][23][24] which helped to explain their function in recruiting ubiquitinated membrane proteins for targeting to multivesicular bodies (MVBs) 25 .
Ubiquitin, an 8 kDa protein, is covalently attached to lysine residues of target proteins via E3 ligases 26 . In the case of membrane proteins, reversible ubiquitination serves as a signal for targeting to endosomes, and then to intraluminal vesicles of MVBs. MVBs can subsequently either fuse with lysosomes leading to degradation 27 or they can fuse with the plasma membrane leading to extracellular secretion of membrane proteins in the form of exosomes 28 . Ubiquitin itself may be ubiquitinated at any of its seven lysine residues leading to target proteins being tagged with a chain of polyubiquitin molecules 29 . K48-linked polyubiquitin chains are a signal for proteasomal degradation whereas mono-ubiquitin and K63-linked polyubiquitin chains are another signal for trafficking of membrane proteins to the endosomal pathway, and especially into the MVB pathway [30][31][32] . The GAT domain of GGA proteins has been shown to be required for the sorting of membrane proteins tagged with K63-polyubiquitin chains into the MVB pathway 31,32 .
GGA proteins themselves are also mono-ubiquitinated in a manner dependent on the binding of ubiquitin to their GAT domain 19 . A large number of ubiquitin-binding proteins have been found to also be ubiquitinated themselves. The reasons for this are not always completely clear but it is thought that concurrent ubiquitinbinding and ubiquitination of sorting proteins aids in the establishment of protein networks to create sorting domains 25,33 . We have recently reported that syntaxin 3, a SNARE involved in membrane fusion at the apical plasma membrane of polarized epithelial cells, undergoes mono-ubiquitination at lysine residues adjacent to its transmembrane domain 34 . Ubiquitination of Stx3 leads to endocytosis from the basolateral plasma membrane, direction into the endosomal/MVB pathway and eventually excretion with exosomes 34 . Functional studies using a non-ubiquitinatable Stx3 mutant suggested that Stx3 may function to sort specific cargo proteins into the MVB/ exosomal pathway 34 . Such a function is unexpected for a protein thought to be involved in membrane fusion.
The structural similarity between the H abc domain of syntaxins and the GAT domain suggests a common ancestry and related function. Given this structural similarity, and given the finding that Stx3-like GGA proteins-is mono-ubiquitinated and appears to play a role in cargo sorting in the MVB/exosomal pathway, we hypothesized that the H abc domain of Stx3 may be a ubiquitin-binding domain. We report here that Stx3 indeed binds to ubiquitin and K63-linked polyubiquitin chains, but not K48-linked polyubiquitin chains. Structural modeling and mutagenesis experiments suggest that the mode of ubiquitin binding could be similar to that of the GAT domain and may involve conserved hydrophobic interactions and a salt bridge. These results suggest that syntaxin function may be regulated by ubiquitin binding, and that syntaxins may function in protein sorting in addition to their established role in membrane fusion.

Results
Stx3 binds non-covalently to ubiquitin. To investigate the possibility that Stx3 may bind to ubiquitin we incubated a purified GST-fusion protein of the entire cytoplasmic domain (1-265) of Stx3 with ubiquitincoated beads to observe any interaction in a pull-down assay (Fig. 1a, domain architecture map of Stx3). A GST-fusion protein of the GGA3-GAT domain served as a positive control. As shown in Fig. 1b, GGA3-GAT interacts strongly with ubiquitin-coated beads as expected. GST-Stx3 also interacts with ubiquitin-coated beads although with reduced efficiency as compared to GGA3-GAT. A GST-fusion protein containing only the H abc domain (1-146) of Stx3 pulls down with ubiquitin-coated beads much more efficiently than the entire Stx3 cytoplasmic domain (3.78 fold change increase) (Fig. 1c). This suggests that the H abc domain directly binds to ubiquitin, and that this interaction is inhibited by intramolecular binding between the H abc and SNARE domains in the "closed conformation" of Stx3. To test this possibility, we introduced the LE165/166AA mutation that has previously been shown to prevent the closed conformation in the highly conserved Stx1A leading to a constitutively open conformation 17 . The observed increase in binding of the open-mutant vs. wild-type Stx3 (1.73 fold change increase) (Fig. 1c) suggests that the H abc domain preferentially binds to ubiquitin when it is not engaged in binding to the SNARE domain.
To assess the affinity of the H abc domain of Stx3 for ubiquitin, we utilized a surface plasmon resonance assay using immobilized GST-Stx3-H abc in comparison with GST-GGA3-GAT as a positive control. The measured K D for the binding of GST-GGA3-GAT to ubiquitin is 0.211 mM (Fig. 1d), which is consistent with previously, published values of 0.231 mM 24 and 0.181 mM 23 . This interaction has been described as a "high-affinity" interaction for a ubiquitin binding protein 29 . In comparison, the measured K D for the H abc domain of Stx3 is 1.36 mM (Fig. 1e) indicating that the binding of Stx3 to mono-ubiquitin in solution is a low-affinity interaction and is weaker than the binding of GGA3 to ubiquitin.
Altogether, these results suggest that the similarity between the GAT domain of GGA proteins and the H abc region of Stx3 is not only structural, but also functional with respect to ubiquitin binding.

Stx3 binds to K63-linked but not K48-linked polyubiquitin chains. Ubiquitin-binding domains
commonly have weak affinities for mono-ubiquitin but the presence of multiple ubiquitin binding sites in the same molecule often results in much higher affinities for polyubiquitin chains 29   www.nature.com/scientificreports/ Next, we tested whether the ability to interact with K63-linked ubiquitin chains is unique to Stx3 or may also be a feature of other syntaxins. We compared GST-fusion proteins of Stx1A, Stx2, and Stx4 side-by-side with Stx3. As shown in Fig. 2b, Stx1A and Stx2 interact with K63-linked ubiquitin chains similarly to Stx3 albeit with somewhat differing preferences for different chain lengths. Stx4 only exhibits very weak interaction with Ubi 6 . Altogether, these data suggest that the H abc domains of several syntaxins are capable of binding to polyubiquitin chains and that ubiquitin-binding may be a conserved function among the syntaxin protein family.
Structural modeling. The GAT domain of GGA3 has been shown to have two distinct binding sites for ubiquitin. Site 1 has been studied in most detail and encompasses residues from the C-terminal half of helix A and the N-terminal half of helix B (Fig. 3a). X-ray structure analysis of ubiquitin in complex with the GGA3-GAT  To better understand how the H abc domain of Stx3 may interact with ubiquitin we constructed a model based on the X-ray structure of ubiquitin in association with site 1 of the GGA3 GAT domain 23,24 . We modeled the Stx3 sequence into the known structure of the H abc domain of the closely related Stx1A 10,11 , and then fitted this Stx3 H abc domain onto the GGA3 GAT domain. The resulting model is shown in Fig. 4. This allowed us to identify Stx3 residues that correspond most closely to the known hydrophobic and ionic interactions between GGA3-GAT and ubiquitin. As shown in Fig. 4, the Stx3 H abc domain has two glutamic acid residues (E78 and E83) in very similar positions as E246 and E250 of GGA3-GAT, and these residues may be predicted to engage in a salt bridge with R42 of ubiquitin. Similarly, L59, L77 and L80 of Stx3 would form a hydrophobic pocket that may interact with I44 of ubiquitin, analogous to the hydrophobic pocket formed by L227, M231 and L247 of GGA3 (Fig. 4).
Due to the lack of similarity between Stx3 and GGA3 at the primary sequence level (Fig. 3a) these predictions would have been difficult or impossible to make. However, we note that E78 and E83 of Stx3 and E246 and E250 of GGA3 could be aligned closely with each other (Fig. 3a). Sequence alignment of Stx3 orthologs from numerous species indicates that all of the residues that may potentially interact with ubiquitin are highly conserved (Fig. 3b). In addition, sequence alignment of human Stx1-4 showed that E78/E83 and L77/L80 are largely conserved between them (Fig. 3c).
Mutational analysis. Based on this model, we decided to mutate residues L77, E78, L80, and E83 of Stx3 to alanine residues and test any effects on the ability to interact with ubiquitin chains. GST fusion proteins with the cytoplasmic domain of Stx3 containing either double leucine (L77A/L80A) or double glutamate (E78A/E83A) mutations were generated. Introducing all four mutations simultaneously resulted in an insoluble GST fusion protein that could not be analyzed.
The ability of the mutants to bind to K63-linked polyubiquitin chains was assessed using the same assay as in Fig. 2. Introducing the E78A/E83A mutations had no discernible effect on polyubiquitin binding as compared to wild-type Stx3 (Fig. 5). Introducing the L77A/L80A mutations had only a seemingly minor effect in that it eliminated a very weak interaction with K63-linked Ubi 3 (Fig. 5). Given that similar mutagenesis experiments with GGA GAT domains frequently lead only to minor disruption of ubiquitin binding 19,21,23 , however, these results may not be surprising. First, the interactions between the GAT domain and ubiquitin involve numerous contacts with multiple residues. Second, the fact that the GAT domain has two separate ubiquitin binding sites suggests that mutations of one site alone might have little effect on overall ubiquitin binding, especially for the binding of polyubiquitin chains. We hypothesize that polyubiquitin chains may wrap around the entire surface  23,24 . The Stx3 sequence was modeled into the known structure of the H abc domain of the closely related Stx1A 10,11 and fitted onto the GGA3 GAT domain. R42 of ubiquitin (yellow) is known to engage in an ionic interaction with E246 and E250 of GGA3-GAT (green). I44 of ubiquitin (yellow) is known to engage in interactions with a hydrophobic pocket formed by L227, M231 and L247 of GGA3-GAT (green). Stx3 residues (blue) that correspond most closely to these residues are E78 and E83 for the ionic site and L59, L77 and L80 for the hydrophobic site.

Scientific Reports
| (2020) 10:21350 | https://doi.org/10.1038/s41598-020-78412-0 www.nature.com/scientificreports/ of the GAT domain, and by analogy also the H abc domain of syntaxins, and engage in numerous contacts that are difficult to completely disrupt by mutagenesis. Such a binding mode may also explain why similar structures (GAT and H abc domains) could bind to polyubiquitin even in the absence of highly conserved primary sequence similarity. In this regard, it is interesting that the L77A/L80A mutations in Stx3 appear to disrupt only the binding to K63-linked Ubi 3 , which may suggest that longer ubiquitin chains can compensate by interacting with additional residues simultaneously.

Discussion
This study illuminates a novel characteristic of syntaxins, ubiquitin binding. This is a surprising finding because, to our knowledge, ubiquitin-binding has not previously been reported for any SNARE protein, nor has there been any indication that ubiquitin-binding could affect SNARE-mediated membrane fusion events. On the other hand, the fact that the 3D structures of the GAT and H abc domains are highly similar, and the fact that hydrophobic and ionic residues known to mediate ubiquitin-binding of the GAT domain may have equivalent residues in the H abc domain of Stx3 (Fig. 4) makes it plausible that both domains may share a similar function. Besides in GGA proteins, GAT domains are also present in the more distantly related proteins TOM1 and TOM1-L1, both of which also bind to ubiquitin 37 . The degree of primary sequence similarity among these GAT domains is low but the overall structures of these 3-helix bundles are highly conserved. The finding that H abc domains of syntaxins not only share the same fold with GAT domains but also function in ubiquitin-binding suggests that these domains share a common ancestry. Our data support the conclusion that the H abc domain of Stx3 is a bona fide ubiquitin binding domain. The specificity for polyubiquitin-K63 chains could be influenced by the ability of GST proteins to form multimers 35 . Furthermore, it is known that SNARE complexes form dimers and higher-order multimers at the site of fusion 36 . We therefore hypothesized that the specificity for K63-linked chains could be favored when Stx3 is present in clusters (forming multimers). This may add a high level of complexity to the regulation of ubiquitin binding of Stx3, as previously suggested by Sims et al. 35 for other UBA domains. While the purpose of the ability of syntaxins to bind to mono-ubiquitin and K63-linked polyubiquitin chains remains to be elucidated, several possibilities can be envisioned. Binding of K63-linked polyubiquitin chains to the H abc domain of a syntaxin may interfere with the ability of that syntaxin to bind to other regulatory proteins that are known to interact with the H abc domain such as members of the munc13, synaptotagmin and munc18 families of SNARE regulators. Thereby, K63-linked polyubiquitin chains, possibly attached to specific regulatory proteins, may regulate SNARE function and therefore membrane fusion in certain vesicle trafficking pathways. Another possibility is that binding of K63-linked polyubiquitin chains to the H abc domain of a syntaxin would interfere with the ability of the H abc domain to engage in an intramolecular interaction with the SNARE domain of that syntaxin. This would result www.nature.com/scientificreports/ in a "constitutively open" conformation of that syntaxin and, again, may regulate membrane fusion functions. Another possibility emerges from our recent finding that Stx3 can undergo ubiquitination at a cluster of lysine residues located between its SNARE domain and transmembrane domain 34 . It may be possible that the H abc domain could engage in an intramolecular interaction with this covalently attached ubiquitin thereby locking such modified Stx3 in a "constitutively closed" conformation. Finally, it is possible that binding of K63-linked polyubiquitin chains to the H abc domain of Stx3 may be unrelated to a function in membrane fusion but rather relates to a different function. Such a possibility may be supported by our recent finding that Stx3 that is covalently ubiquitinated at the lysine cluster proximal to its transmembrane domain, enters the endosomal pathway, traffics to intraluminal vesicles of MVBs, and is eventually excreted with exosomes 34 . We reported that a nonubiquitinatable mutant of Stx3 (termed Stx3-5R) is not only unable to enter the MVB/exosomal pathway but also interferes with the recruitment of a specific apical exosomal cargo protein, the orphan G-protein coupled receptor GPRC5B, into this pathway. This suggested that Stx3 normally plays a role in cargo recruitment in a fashion that is dependent on its ability to be ubiquitinated. Interestingly, the Stx3-5R mutant was also found to disrupt the secretion of human cytomegalovirus (hCMV) virions, a result that-combined with other findings-suggests that hCMV exploits the MVB/exosomal pathways for virion production and secretion 34 . In this regard, Stx3 bears striking similarities to GGA proteins. Both contain a similarly structured ubiquitin-binding domain, both undergo ubiquitination themselves, and both are involved in recruitment of membrane proteins into the MVB pathway. In the case of GGA proteins, this sorting function requires cargo proteins to be tagged with K63-linked polyubiquitin chains 31,32 .
Altogether, these results suggest that syntaxins contain a ubiquitin binding domain similar to the GAT domain. The implications of this finding are yet to be elucidated but may relate to the regulation of membrane fusion functions and/or point towards a novel function of syntaxins in the sorting of membrane proteins.
Mono-ubiquitin binding assay. 55 µl (dry volume) of ubiquitin-coated agarose (Millipore-Sigma) or CL4B Sepharose (GE Healthcare Lifesciences) beads were washed three times with 1 ml of Buffer A (25 mM HEPES-KOH, pH 7.4, 125 mM potassium acetate, 2.5 mM magnesium acetate, 5 mM EGTA) + 1% FBS. Buffer was removed from pelleted beads using a Hamilton syringe and 55 µL of Buffer A + 1% FBS were added to beads to create a 50/50 slurry. 30 µL of the bead slurry were added to 1 ml of Buffer A + 1% FBS containing 1 µM of purified GST-fusion protein in an Eppendorf tube. Tubes were tumbled end-over-end for 2 h at room temperature. Beads were pelleted and washed four times with 1 ml of Buffer A + 0.005% Tween-20 before pelleting and removing residual buffer with a Hamilton syringe. Beads were re-suspended in 45 µl of SDS-PAGE sample buffer containing 100 mM DTT, boiled, separated on SDS-PAGE, and transferred to nitrocellulose membrane. Membranes were probed with a polyclonal goat anti-GST antibody (GE Healthcare Life Sciences) and a Stx3-antibody generated by us (available as MAB2258 from Millipore-Sigma) and a donkey anti-goat IgG HRP-conjugated secondary antibody (Jackson Immunoresearch).