Characterization of the Mammalian CORVET and HOPS Complexes and Their Modular Restructuring for Endosome Specificity*

Background: The CORVET and HOPS complexes regulate endosomal cargo trafficking but have not been well characterized in mammals. Results: A detailed analysis of subunit interactions within the mammalian CORVET, HOPS, and VIPAS39/VPS33B complexes. Conclusion: Tethering complexes have adapted to the higher complexity of trafficking in mammalian cells. Significance: This work provides a detailed architectural insight into the mammalian endosomal tethering complexes. Trafficking of cargo through the endosomal system depends on endosomal fusion events mediated by SNARE proteins, Rab-GTPases, and multisubunit tethering complexes. The CORVET and HOPS tethering complexes, respectively, regulate early and late endosomal tethering and have been characterized in detail in yeast where their sequential membrane targeting and assembly is well understood. Mammalian CORVET and HOPS subunits significantly differ from their yeast homologues, and novel proteins with high homology to CORVET/HOPS subunits have evolved. However, an analysis of the molecular interactions between these subunits in mammals is lacking. Here, we provide a detailed analysis of interactions within the mammalian CORVET and HOPS as well as an additional endosomal-targeting complex (VIPAS39-VPS33B) that does not exist in yeast. We show that core interactions within CORVET and HOPS are largely conserved but that the membrane-targeting module in HOPS has significantly changed to accommodate binding to mammalian-specific RAB7 interacting lysosomal protein (RILP). Arthrogryposis-renal dysfunction-cholestasis (ARC) syndrome-associated mutations in VPS33B selectively disrupt recruitment to late endosomes by RILP or binding to its partner VIPAS39. Within the shared core of CORVET/HOPS, we find that VPS11 acts as a molecular switch that binds either CORVET-specific TGFBRAP1 or HOPS-specific VPS39/RILP thereby allowing selective targeting of these tethering complexes to early or late endosomes to time fusion events in the endo/lysosomal pathway.

recent data suggest that VPS33B and VIPAS39 form a separate complex (12).
The organization of these mammalian complexes has only partly been addressed with seemingly conflicting results and hampered the interpretation of functional experiments in which the roles of specific or multiple units are addressed (17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27). Furthermore, it is not known what drives membrane specificity of these complexes in mammals. Here we provide an extensive map of the mammalian CORVET, HOPS, and VPS33B-VIPAS39 complexes describing their intersubunit interactions and membrane recruitment. In agreement with Wartosch et al. (12) we find that VIPAS39 and VPS33B are not part of the CORVET or HOPS complex but assemble in a distinct complex. We find that the CORVET complex is prevented from recruitment to LE (in line with its function at the EE). Within the HOPS complex there are multiple RILP binding modules, and pathogenic (arthrogryposis-renal dysfunctioncholestasis (ARC) syndrome) mutations in VPS33B disrupt VPS33B-VIPAS39 complex assembly or RILP-dependent LE recruitment. Within the shared CORVET/HOPS core, VPS11 can bind TGFBRAP1 (CORVET) as well as VPS39 (HOPS), and these subunits likely compete for binding to VPS11 in the core of the CORVET-HOPS complex, thereby driving the membrane-specific assembly and tethering capability of these complexes.
Microscopy Sample Preparation-Transfected cells were fixed 24 h post-transfection with 4% formaldehyde in PBS for 15 min and permeabilized for 10 min with 0.05% Triton X-100 in PBS at room temperature. Nonspecific binding of antibodies was blocked by 0.5% BSA in PBS for 40 min, after which cells were incubated with primary antibodies in 0.5%BSA in PBS for 1 h at room temperature. After washing 3 times with PBS, primary antibodies were visualized with Alexa-Fluor secondary antibody conjugates antibodies in 0.5% BSA in PBS for 30 min at room temperature (Invitrogen). After three washes with PBS samples were mounted in Vectashield mounting medium (Vector Laboratories).
Protein Immunoprecipitation-MelJuSo cells were washed with ice-cold PBS and scraped into cell lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 5 mM MgCl 2 , 1% Nonidet P-40, 10% glycerol, pH 7.4) supplemented with complete EDTA-free protease inhibitor mixture. Cell lysates were obtained by incubation on ice for 10 min followed by centrifugation for clearing. The supernatants were incubated for 2 h with respective antibodies followed by capture with protein G-Sepharose 4 FF resin and washed extensively with wash-buffer (50 mM Tris, 150 mM NaCl, 5 mM MgCl 2 , 10% glycerol, pH 7.4). For GFP-tagged pulldown, GFP-TRAP beads were used (Chromotek). All experiments were repeated multiple times. To compose the figure panels, inserts from a single experiment containing all the experimental conditions (and all the transfected constructs) were composed as described in the figure legends.
Statistical Analysis-For calculation of correlation coefficients the signal intensity over a vector was plotted using the plot profile tool in ImageJ (http://imagej.nih.gov) in the respective fluorescence channels. Correlation coefficient for colocalization assays were calculated from plot profiles measuring intensity for the indicated fluorescence channels over a vector through the cells (as in van der Kant et al. (30)). Correlation coefficients for the two plots were calculated in Excel using the function, Statistical testing was performed in Graphpad Prism 6 using a one-way analysis of variance with Tukey's multiple comparison test for significance.

Results
Characterization and Definition of the Mammalian CORVET, HOPS, and VPS33B-VIPAS39 Complex-The yeast CORVET and HOPS complexes are assembled from eight different proteins (a Vps16/Vps33/Vps18/Vps11 core with Vps3/ Vps8 or Vps39/Vps41). Mammals do not possess 8, but 10 homologues to the yeast CORVET/HOPS subunits. Two mammalian homologues for yeast Vps33 exist (VPS33A and VPS33B) as well as an additional protein with weak homology to Vps16 (VIPAS39). To probe the interactions between the 10 putative CORVET/HOPS subunits and determine the topology of the different subunits in the CORVET-HOPS complexes we performed an extensive array of co-immunoprecipitations (IP) in human MelJuSo cells with antibodies against endogenous VIPAS39, VPS33B, VPS11, VPS16, VPS8, and VPS33A ( Fig. 1A) (antibodies specific for other components were either not available or did not work for IP). We observed extensive interactions between VPS11, VPS16, VPS8, TGFBRAP1, and VPS33A. In contrast, VIPAS39 and VPS33B interacted with each other but not with any of the other canonical CORVET/HOPS subunits in line with recent data (12).
To further map the HOPS complex and define possible interactions between VIPAS39 and VPS33B with HOPS components that we could not map in endogenous pulldown experiments, we ectopically co-expressed all six putative HOPS complex subunits as well as VPS33B and VIPAS39 as tagged proteins. We previously showed that expression levels of these tagged proteins in our systems are low (comparable to endogenous) and that the tagged subunits incorporate into functional complexes (30).
We pulled down GFP-VPS16, GFP-VPS11, or GFP-VPS33B and detected the associated proteins by Western blotting with antibodies against the different epitope tags (Fig. 1B). GFP-VPS16 specifically interacted with VPS41, VPS18, and VPS33A. In addition, GFP-VPS11 interacted with VPS39. In contrast to the endogenous co-IP (Fig. 1A), no interactions between the two groups (VPS16/41/18/33a and VPS11/39) were observed. A pulldown experiment with GFP-VPS11 also failed to detect interactions between GFP-VPS11 and endogenous VPS16/ VPS33a (data not shown). This may reflect a limitation of the overexpression system or an interference with normal interactions by the associated tags. In line with the endogenous IP ( Fig.  1A), we again observed strong interactions between VPS33B and VIPAS39 but not with any other subunit of the HOPS complex. This suggests that mammalian HOPS consist of VPS11/VPS16/VPS18/VPS33A/VPS39/VPS41, whereas VPS33B-VIPAS39 assembles in a distinct complex (Fig. 1C). Furthermore, the observation that tagged subunits recapitulate in large part endogenous interactions indicates we can use these constructs for further detailed mapping of the respective complexes.
A High Detail Interaction Map of the Mammalian HOPS Complex-The yeast HOPS complex resembles a seahorseshaped complex with a head (Vps16/Vps33/Vps41/Vps18) and a tail (Vps11/Vps39), and interactions within the complex have been extensively mapped (6, 7) ( Fig. 2A). No structure for the mammalian HOPS complex is available, yet our experiments ( Fig. 1) suggest that interactions within the head (VPS16/ VPS33/VPS41/VPS18) and tail (VPS39/VPS11) are conserved from yeast to mammal. To study the interactions within the proposed head of the mammalian HOPS complex, we expressed tagged truncation mutants of a particular subunit (domain organizations based on STRING 9.1 depicted in Fig.  2B and assessed the interactions with the three other head proteins by co-IP experiments ( Fig. 2C and summarized in Fig. 2D). We observed that VPS16 is a central protein in the head as its C terminus binds VPS33A, whereas the N-terminal segment binds to VPS41 and VPS18. The Ring finger (R) domains of VPS18 and VPS41 interact with the N terminus of VPS16 possibly acting in a tripartite interaction.
Different from yeast, the mammalian HOPS complex is not recruited to membranes by RAB7 (Ypt7 in yeast) but by binding to the RAB7 effector RILP that has no apparent ortholog in yeast (30,32). To evaluate membrane targeting of the mammalian HOPS complex, we mapped the minimal domains for membrane recruitment of distinct subunits by expressing tagged truncation mutants and assessed their recruitment to RILP-containing LE (Fig. 2E). In addition to VPS41 (32), we observed that small domains of various HOPS subunits could be recruited to RILP (summarized in Fig. 2, F and G). This suggests that RILP could act as a scaffold in the assembly of the HOPS complex or that interactions within the mammalian HOPS complex are further stabilized via secondary interactions of its subunits with RILP. In assembly, our data suggest that interactions within the HOPS complex are reasonably conserved from yeast to mammals, yet the complex has significantly evolved to accommodate a shift from RAB7 to binding to RAB7 effector RILP instead, possibly reflecting higher order regulation such as the concomitant binding of the dynein-motor complex (30).
VPS11 as the Core Receptor for HOPS-and CORVET-specific Subunits-We have shown that TGFBRAP1 and VPS8 endogenously bind core CORVET/HOPS subunits including VPS11 and VPS16 (Fig. 1A) and that VPS11 and VPS16 are recruited to LE by RILP (Ref. 30; Fig. 2E). This argued that overexpression of RILP via their secondary interactions with VPS11 or VPS16 might attract VPS8 and TGFBRAP1 to LE. However, when RILP and VPS8 or TGFBRAP1 were co-expressed, the latter were not recruited to RILP but remained localized to EE (Fig. 3A) similar to VPS8 and TGFBRAP1 expression alone (Fig. 3B). This indicates an active mechanism that prevents recruitment of TGFBRAP1 and VPS8 to LE even in the presence of RILP. We, therefore, asked how CORVETversus HOPS-specific protein binding to the core of the complex is regulated to assure correct targeting and sequential activity in endosomal maturation. Because TGFBRAP1 and VPS39 have high sequence homology, their competitive binding to the same subunit in the CORVET/HOPS core suggests a mechanism for sequential binding. A likely candidate for a common binding partner in the core of the COR-  DECEMBER     VET-HOPS complex would be VPS11, as we already identified interactions between VPS11 and VPS39 (Fig. 1). In line with our endogenous IP with VPS11 antibodies, an interaction between TGFBRAP1 and GFP-VPS11 was observed ( Fig. 3C), and co-expression of VPS11 and TGFBRAP1 resulted in the recruitment of cytosolic GFP-VPS11 to EE (Fig. 3D). To map the interactions between TGFBRAP1 and VPS39 with VPS11, we expressed tagged truncation mutants   . VPS11 interacts with CORVET specific subunits. A, MelJuSo cells co-expressing mRFP-RILP (red) and GFP-TGFBRAP1 or HA-VPS8 (green) were fixed, stained with antibodies against EEA1 (and anti-HA for VPS8), and imaged by CLSM. Scale bars: 10 m. Graphs show average correlation coefficient (CC) Ϯ S.E n Ͼ 25, for RILP or EEA1 and GFP, TGFBRAP1, or VPS8. Asterisks indicate significance to GFP control (****, p Յ 0.0001; * p Յ 0.05; ns ϭ not significant). B, MelJuSo cells co-expressing GFP-TGFBRAP1 or HA-VPS8 were fixed, stained with antibodies against EEA1 (and anti-HA for VPS8), and imaged by CLSM. Scale bars: 10 m. C, lysates of MelJuSo cells expressing GFP or GFP-VPS11 were immunoprecipitated (IP) with anti-GFP and analyzed by WB using anti-GFP and anti-TGFBRAP1 antibodies. Experimental conditions were run on the same blot with the same exposures for each detection antibody, and cutouts were taken and grouped for presentation purposes. TL, total lysate. D, MelJuSo cells expressing mRFP-VPS11 (blue) or co-expressing mRFP-VPS11 (blue) and GFP-TGFBRAP1 (green) were fixed, stained with antibodies against EEA1 (red), and imaged by CLSM. Scale bars: 10 m. Graphs show average correlation coefficient (CC) Ϯ S.E n Ͼ 25, between VPS11 and EEA1 Ϯ ectopically expressed TGFBRAP1 (****, p Յ 0.0001).

Architecture of the Mammalian HOPS and CORVET Complexes
of VPS11 and assessed putative interactions (Fig. 4, A and B). TGFBRAP1 and VPS39 bound to the same domain in VPS11 (Fig. 4, A and B), implying that these two proteins might compete for VPS11 binding (Fig. 4C). As TGFBRAP1 binds VPS11 and VPS11 binds RILP, we argued that a RILP-VPS11-TGFBRAP1 interaction could exist that could induce (erroneous) recruitment of TGFBRAP1 to LE. However, when VPS11 was co-expressed with TGFBRAP1 and RILP, VPS11 was exclusively recruited to EE and no longer recruited to LE (even though high levels of RILP were present) (Fig. 4D). This suggests a regulatory mechanism in which binding of TGFBRAP1 to VPS11 in the CORVET complex prevents VPS11 binding to RILP, thereby blocking the targeting of VPS11 to LE and premature assembly of the HOPS complex on LE. Such a mechanism would safeguard a sequential assembly of the CORVET and HOPS and thereby time the fusion events of EE and LE.

ARC Mutations Specifically Disrupt VPS33B-VIPAS39 Interactions or Complex Recruitment to Late Endosomes by RILP-
Published data regarding the role of VPS33B and VIPAS39 as members of the HOPS complex are seemingly conflicting (12,16). Our endogenous IP data (Fig. 1A) as well as our IPs with tagged subunits (Fig. 1B) indicated that VPS33B and VIPAS39 did not interact with other HOPS subunits. To further substantiate the existence of a separate VPS33B-VIPAS39 complex, we depleted CORVET/HOPS subunits by siRNA and assessed the effects on the stability of their endogenous interaction partners by Western blot (WB) (Fig. 5A). Silencing of VPS16 significantly decreased VPS33A and VPS11 protein levels, whereas VPS33B and VIPAS39 levels did not decrease (Fig. 5A). Similarly, silencing of VPS33B compromised VIPAS39 levels without affecting VPS16, VPS33A, or VPS11 (Fig. 5A). These findings reinforce the notion that VPS33B and VIPAS39 assemble into a separate complex.
Interestingly, autosomal recessive mutations in VPS33B or VIPAS39 cause ARC syndrome (23,33), a fatal multisystem disorder characterized by defects in apical transport in polarized cells. In line with this, VPS33B and VIPAS39 have previously been shown to function at RAB11 positive recycling endosomes (21,22). Under steady-state conditions VPS33B and VIPAR do not significantly localize to LE (22), but we have shown that these proteins can be recruited to LE by RILP (30). In line with a function of VPS33B and VIPAR at the LE, it was recently shown that VPS33B is important for the maturation of the ␣-granule, a specialized late endosomal compartment (34). Therefore, we investigated whether pathogenic mutations in VPS33B could still be recruited to LE by RILP. We studied two pathogenic mutations of VPS33B, a single amino acid substitution mutant (VPS33B L30P) and a truncation mutant (VPS33B 1-437), and mapped the effect on VPS33B-VIPAS39 interactions as well as RILP-dependent membrane recruitment. The truncated mutant of VPS33B (1-437) did not interact with VIPAS39 ( Fig. 5B; Ref. 22) but was still recruited to LE by RILP, indicating that VPS33B can bind RILP independent of VIPAS39   ( Fig. 5C). The L30P mutation in VPS33B had a different effect; although VPS33B L30P still interacted with VIPAS39 (Fig. 5B), it failed to be recruited to LE by RILP ( Fig. 5C; Ref. 30). Thus, our results indicate that two specific mutations in VPS33B, as found in ARC patients, affect assembly with VIPAS39 (VPS33B 1-437) or LE recruitment of the VPS33B-VIPAS39 complex (VPS33B L30P) by RILP (Fig. 5D).

Discussion
We have investigated the interactions between the proposed mammalian subunits of the multisubunit CORVET and HOPS tethering complexes and found, with modifications, that the organization of these complexes is largely conserved from yeast to mammals (Fig. 6A). As in yeast, the mammalian HOPS complex consists of VPS16, VPS11, VPS33A, VPS18, VPS41, and VPS39. We confirm that TGFBRAP1 and VPS8 are the mammalian CORVET-specific subunits (with TGFBRAP1 as the yeast vps3 ortholog) (11,35). In contrast to previous data (16), however, our data indicate that VIPAS39 and VPS33B, two CORVET/HOPS subunit homologues not present in yeast, are not part of CORVET/HOPS as also suggested by others (12,21,36  interactions of VPS33B and/or VIPAS39 with CORVET/HOPS or the existence of VPS33B-VIPAS39-HOPS interactions in other cellular systems. However, the existence of a VPS33/ VIPAS39 complex independent of other core subunits might explain why only mutations in VPS33B and VIPAS39 but not any of other HOPS subunits are associated to ARC syndrome (21)(22)(23)33) and why all HOPS complex subunits, but not VPS33B and VIPAS39, are required for Ebola-virus infection (19) tion at LEs (34) as well as recycling endosomes (16,21,22) and that both membrane binding and VIPAS39 binding are needed for correct VPS33B function. We show that HOPS subunits have evolved to accommodate binding to the RAB7 effector RILP (for a mammalian model superimposed on the yeast structure see Fig. 3G). The N terminus of yeast vps39 that binds Ypt7 is poorly conserved in the mammalian VPS39 ortholog, which might explain why the mammalian HOPS complex has shifted to bind RAB7 effectors such as RILP (30,32) and PLEKHM1 (14,15). The HOPS complex binds RILP on either side of the complex (VPS39/VPS11 in the tail and VPS41/VPS18 in the head) and may bridge RILP molecules on opposing vesicles. Because VPS33A in the head of HOPS also binds to SNAREs (41), a connection between two vesicles will then support tethering and subsequent fusion. It is interesting to note that the HOPS complex can also bind the GTPase Arl8b on lysosomes (20,42) and PLEKHM1 on LE, phagosomes, and autophagosomes (14,15), possibly tethering these vesicles with RILP-containing organelles (Fig. 6B). RAB7-RILP concomitantly binds the HOPS complex and the dynein-motor complex for retrograde transport (30), whereas the Arl8b effector SKIP also concomitantly regulates HOPS binding and anterograde transport by kinesin (42), indicating that timing of vesicle tethering is likely coupled to their transport.
We found that binding of TGFBRAP1 to VPS11 prevents the interaction of VPS11 with RILP and erroneous recruitment of CORVET complexes to RILP-bearing LE. This might provide an important regulatory step in the conversion of CORVET to HOPS complexes (Fig. 5) where TGFBRAP1 in CORVET would have to release VPS11, thereby allowing VPS11 to bind VPS39 and recruit the entire HOPS complex to RAB7-RILP during RAB5 (EEs) to RAB7 (LE) conversion. Specific targeting of tethering complexes by modulation of targeting subunits was recently also shown for GARP and EARP tethering on Golgi and EEs, respectively, indicating this might be a common mechanism in the regulation of vesicular fusion (43). TGFBRAP1 to VPS39 exchange might further be regulated by proteins such as MON1-CCZ1 (Fig. 6B), which are recruited by CORVET, interact with VPS39, and act as activators of RAB7 (44 -46) and signaling pathways such as the TGF␤ pathway, which has been shown to intersect with TGFBRAP1 and VPS39 function (47)(48)(49).
In conclusion, we describe the molecular architecture of the mammalian CORVET, HOPS, and VIPAS39-VPS33B complexes. We find that different ARC syndrome mutations affect VIPAS39-VPS33B or VPS33B-RILP interactions, indicating that this complex has a function at the LE. Through mapping of the CORVET and HOPS complexes we found that within the shared CORVET/HOPS core, VPS11 is a molecular switch that, depending on its interacting proteins (TGFBRAP1 or VPS39), determines targeting of CORVET and HOPS to EE and LE.
Author Contributions-R. v d K. designed and performed the experiments for Figs. 1-4, analyzed the data, and wrote the paper. C. T. H. J. designed and performed the experiments for Fig. 3, analyzed the data, and wrote the paper. R. H. W. (Fig. 4) and J. B. (Fig. 2E) performed the experiments. L. J. cloned the constructs used in the manuscript. J. K. and J. N. designed the experiments and wrote the paper.