The Golgi-Associated Hook3 Protein Is a Member of a Novel Family of Microtubule-Binding Proteins

The positioning, trafficking, and architecture of various membrane compartments rely on microtubules (Schroer 2000). For example, the microtubule organizing center (MTOC) specifies the characteristic perinuclear position of secretory and endocytic pathway components (Matteoni and Kreis 1987; Ho et al. 1989; Bloom and Goldstein 1998). Within each pathway, vesicular transport between compartments is organized along microtubules: trafficking from early to late endosomes depends on microtubules, as does retrograde and anterograde trafficking between the Golgi complex and the ER (Gruenberg et al. 1989; Strous et al. 1991; Aniento et al. 1993; Lippincott-Schwartz 1998). Furthermore, after the dispersion of the Golgi complex during mitosis, appropriate reformation of the juxtanuclear Golgi ribbon requires microtubules (Lippincott-Schwartz 1998; Shima et al. 1998). This functional relationship between microtubules and Golgi membranes is reflected in their extensive interdigitation about the MTOC (Cole and Lippincott-Schwartz 1995).

Two families of motor proteins mediate organelle transport along microtubules (Hirokawa 1998). Different members of the Kinesin family carry specific membranous cargos along microtubules (Goldstein and Philp 1999). Dynein family members carry cargo from the cell periphery towards the juxtanuclear minus ends of microtubules located at the MTOC. Interference with dynein function results in the mislocalization of elements of the secretory and the endocytic compartments toward the cell periphery (Burkhardt et al. 1997; Presley et al. 1997; Itin et al. 1999). In cells lacking cytoplasmic dynein, however, mislocalized organelles are still attached to microtubules (Harada et al. 1998), suggesting the existence of additional linker proteins.

Such a function in organelle positioning has been suggested for the family of cytoplasmic linker proteins (CLIPs; for review see Schroer 2000). CLIPs share a microtubule-binding domain, and, as these proteins lack a motor domain, their function may be comparable to the molecular tethers identified in vesicle fusion events (Pfeffer 1999). The founding member of this family, CLIP-170, binds to endosomes and to microtubules (Pierre et al. 1992). Similarly, CLIP-115 is a brain-specific protein that links a unique membrane compartment, the dendritic lamellar body, to microtubules (De Zeeuw et al. 1997). Another CLIP-like protein, GMAP210, exhibits Golgi specificity (Infante et al. 1999). The number of CLIPs assigned to specific compartments is very limited, however, suggest that other proteins with similar function exist (Schroer 2000).

In this report, we show that Hook proteins constitute a novel family of microtubule-binding proteins that may link membrane compartments to microtubules. The Drosophila Hook protein (dHK) is a cytosolic endosome-associated protein necessary for normal trafficking of endocytosed ligands (Krämer and Phistry 1996, Krämer and Phistry 1999). Our analysis of Drosophila mutants lacking Hook protein revealed that dHK functions to assemble or stabilize mature multivesicular bodies (MVBs), an intermediate compartment in the endocytic pathway (Sunio et al. 1999). Here, we introduce three human Hook proteins, which exhibit different compartmental specificities. Although the Hook proteins localize to distinct organelles, they share a novel microtubule-binding domain. Through its interaction with microtubules, the human Hook3 protein (hHK3) may serve in the positioning of the mammalian Golgi complex.

The human EST database (dbEST, GenBank) was searched for Hook homologues using the BLAST program (Altschul et al. 1990). Among the I.M.A.G.E. Consortium cDNA clones (Lennon et al. 1996), this search revealed several overlapping cDNA clones (2662038, 1662556, 379556, and hp0259) representing the same ORF encoding human Hook3. The full-length human Hook3 sequence was obtained by 5′ and 3′ RACE reactions from a human placental cDNA library (CLONTECH Laboratories, Inc.). The sequence of hHK3 has been deposited in Genbank (sequence data available from GenBank/EMBL/DDBJ under accession number AF241830).

For expression in mammalian cells, PCR-generated cDNAs encoding full-length hHK3 or the COOH-terminal truncations of ΔC-hHK1_1–555, ΔC-hHK2_1–548, and ΔC-hHK3_1-555, which were all tagged with a COOH-terminal Myc epitope, were inserted between the Asp718 and XhoI sites of pCDNA3.1 (Invitrogen). Other hHk3 truncations and fusions with dHk (either amino acids [aa] 6–562 or 287–679) were generated in pCDNA3.1 with a NH₂-terminal hemagglutinin epitope tag. HEK293, Cos7, or Vero cells were transfected using Lipofectamine (GIBCO BRL).

HEK293, HeLa, Hep2, normal rat kidney, and Vero cells were obtained from the American Type Culture Collection and cultured using standard techniques. Antibodies used were directed against FTCD (previously 58K; Bashour and Bloom 1998), β-COP (Pepperkok et al. 1993), GM130 (Transduction Laboratories), LAMP-1 (Chen et al. 1985), transferrin receptor (Boehringer), Drosophila Hook (Krämer and Phistry 1996), ERD2/KDEL receptor (Majoul et al. 1998), clathrin (Brodsky 1985), LBPA (Kobayashi et al. 1998), M6PR (Boker et al. 1997), calnexin (Affinity BioReagents, Inc.), TGN46 (Serotec), ERGIC-53 (Schindler et al. 1993), α- or γ-tubulin (Sigma-Aldrich), syntaxin5 (Rowe et al. 1998), and Cox1 (Molecular Probes).

For the generation of antibodies against human Hook proteins, glutathione S-transferase fusion proteins of aa 438–728 from hHK1, 427–719 from hHK2, and 423–630 from hHK3 were purified from bacteria and injected into rabbits or mice. The corresponding purified His₆-tagged fusion proteins were used for preabsorbing specific antibodies in control experiments and for affinity purification (Harlow and Lane 1988). For multiple tissue Western blots, tissues were harvested from mice, homogenized in SDS loading buffer at 25 μg wet tissue/μl, separated by SDS-PAGE, and probed by Western analysis as described (Sevrioukov et al. 1999).

Immunofluorescence staining was essentially as described (Krämer and Phistry 1996), after fixing cells using methanol at −20°C or 4% paraformaldehyde. Endogenous human Hook proteins were detected using affinity-purified antibodies raised in rabbits. Secondary antibodies were coupled to Alexa488 or Alexa568 (Molecular Probes). When double labeling hHK3-overexpressing cells with rabbit anti-KDEL receptor antibodies (Majoul et al. 1998), we used anti-hHK3 sera raised in mice, anti-Myc, or antihemagglutinin antibodies.

For double labeling with two primary antibodies raised in rabbits (hHK3 and syntaxin5), hHK3 antibodies were labeled using a digoxigenin labeling kit (Molecular Probes). Cells were first stained with anti-syntaxin5 antibodies and goat anti–rabbit Fab fragments (Alexa488), followed by digoxigenin-labeled anti-hHK3 antibodies, mouse anti-digoxigenin antibodies (The Jackson Laboratory), and, finally, goat anti–mouse antibodies (Alexa568).

Images were captured using a MRC1024 confocal laser-scanning microscope (Bio-Rad Laboratories) or a cooled ORCA charge-coupled device camera (Hamamatsu); images were prepared for publication using Adobe Photoshop. For nocodazole (NZ) or Brefeldin A (BfA) treatments, cells were incubated for 1 h in normal growth media supplemented with 10 μg/ml NZ or 10 μg/ml BfA. To bundle microtubules in vivo, cells were treated for 1 h with 10 μg/ml NZ, which was washed out and replaced for 2–4 h with media supplemented with 40 μM taxol, after which cells were processed for immunofluorescence imaging.

Three 10-cm dishes of HEK293 cells (∼75% confluent) were harvested and dounce homogenized. Nuclei and cell debris were sedimented at 1,000 g at 4°C. The postnuclear supernatant collected was adjusted to 1% Triton X-100. For cross-linking, BS³ (Pierce Chemical Co.) was added to 100 μM and incubated for 30 min at room temperature (RT) with gentle mixing. To quench reactive groups, ethanolamine was added to 100 mM and incubated 30 min at RT. Samples were immunoprecipitated by incubation for 1 h at 4°C with the appropriate antibody and protein–A agarose and were analyzed by Western blotting (Sevrioukov et al. 1999).

For binding assays, cytosol was prepared from HEK293, Drosophila S2 cells, and the transfected cell lines HEK293:ΔC-hHK1_1-555, HEK293:ΔC-hHK2_1-548, or HEK293:ΔC-hHK3_1-555, as described (Ktistakis et al. 1996). Protein concentrations were 7–13 mg/ml for all preparations; 50-μl aliquots were stored at –80°C.

Microtubule spin-down assays were performed using the microtubule-associated protein spin-down kit (Cytoskeleton) according to the manufacturer's instructions. In brief, microtubules were assembled from purified bovine brain tubulin for 20 min at 35°C in the presence of GTP and stabilized with taxol. Assembled microtubules (10 μg) were incubated with 30 μg of cytosolic proteins in total volume of 50 μl for 20 min at RT. Microtubules and associated proteins were pelleted at 100,000 g through a 40% glycerol cushion. Pellets were dissolved in 10 μl SDS-loading buffer and compared with 10 μl from the supernatant by Western analysis.

To test for direct binding to microtubules, the His₆-tagged fusion proteins His₆–hHK3N1_1–164, hHK3N2_1–224, and hHK3CC_423–630 were bound to Ni²+–agarose beads at a concentration of 6–8 mg/ml. For the binding assay, stabilized microtubules were prepared as follows: 2.5 μl microtubule cushion buffer (PEM [80 mM Pipes, pH 7.0, 1 mM EGTA, 1 mM Mg²⁺, 1 mM GTP] plus 40% glycerol) was added to 20 μl tubulin (5 mg/ml) and incubated 20 min at 35°C. Taxol was added to 180 μl G-PEM (PEM plus 1 mM GTP) to a final concentration of 40 μM. After 20 min of microtubule assembly, 180 μl prewarmed taxol–G-PEM (35°C) was added, and the reaction was stored at RT. 20-μl beads coupled to recombinant proteins were washed once with 100 μl G-PEM and mixed with 50-μl stabilized microtubules. After 2 h at RT, beads were pelleted (1,000 g for 5 min at RT), and a 10-μl aliquot was collected from the supernatant for analysis. The beads were washed three times with 500 μl G-PEM, and bound tubulin was eluted with 10 μl SDS loading buffer. Relative amounts of tubulin were compared by Western analysis.

For testing for the binding of dHK to endogenous microtubules, 100 μg Drosophila cytosolic proteins were adjusted to 20 μl with G-PEM (+) or PEM (−). 32.5 μl PEM was added to the sample (−) and applied to 100 μl cushion buffer without taxol. 5 μl of cushion buffer was added to the sample (+) and incubated 20 min at 35°C. 30 μl G-PEM plus 80 μM taxol was added to the (+) sample and incubated at RT for 20 min, and the sample was applied to 100 μl of cushion buffer (20 μM taxol). Both samples were spun 40 min at 100,000 g at 25°C. 10 μl were collected from the supernatants remaining above the cushion. 50 μl were aspirated, and the cushions were rinsed with 50 μl G-PEM. 110 μl were aspirated from the remaining cushion, and the pellets were harvested in SDS-loading buffer and analyzed by Western blotting. Blots were probed for both dHK and tubulin after cutting the blot at ∼67 kD, and were then reassembled for chemiluminescent detection.

Highly enriched Golgi membranes were a gift from Michael G. Roth (University of Texas Southwestern, Dallas, TX; Ktistakis et al. 1995). This Golgi membrane fraction had been purified from CHO cells. By electron microscopy, half the membranes were identifiable Golgi stacks and they were 20-fold–enriched in mannosidase II activity compared with whole cell membranes (Ktistakis et al. 1996). 120 μl Golgi membranes (200 μg protein/ml) were diluted with 800 μl buffer B (25 mM Hepes, pH 7.3, 25 mM KCl, 2.5 mM MgSO₄, 1 mM DTT, 0.2 M sucrose, protease inhibitors), incubated 5 min on ice, and then pelleted (15,000 g for 9 min at 4°C). Membranes were resuspended in 120 μl buffer B plus 0.1 mg/ml BSA (to a final concentration of 0.2 mg/ml Golgi protein). 10 μl membranes were incubated with ∼30 μg cytosolic proteins in a total volume of 50 μl for 20 min at 37°C. The reaction was cooled on ice for 5 min, and then spun at 15,000 g for 9 min at 4°C. A 10-μl aliquot was taken from the supernatant and combined with 5 μl SDS-loading buffer. The pellet was resuspended in 20 μl buffer B and spun through 100 μl 20% sucrose cushion for 30 min at 15,000 g at 4°C in a swinging bucket rotor. 30 μl the supernatant was aspirated, and the cushion washed with 50 μl buffer B. The pellet containing membranes was collected with 10 μl SDS-loading buffer (5% SDS, 50 mM Tris/HCl, pH6.8) for Western analysis.

Vero cells were fixed in 3% paraformaldehyde, 0.2% glutaraldehyde in PBS (pH 7.4) for 20 min at RT, permeabilized in 0.1% saponin/PBS for 20 min, blocked in 1% normal goat serum, 1% BSA in PBS for 30 min, and then incubated with anti-hHK3 antibodies (purified IgG fraction, dilution 1:200) in blocking buffer for 24 h at 4°C. Primary antibodies were detected with Nanogold–Fab conjugates (Nanoprobes) for 48 h at 4°C; silver enhancement was performed with IntenSE M kit (Amersham Pharmacia Biotech) for 10 min at RT. The cells were postfixed with 0.5% OsO₄ for 30 min, dehydrated with an ascending series of ethanol ≤100%, embedded in Poly/Bed 812 media (Polysciences, Inc.), and polymerized at 60°C for 48 h. Ultrathin sections (60 nm) were cut with Leica Ultracut UCT microtome and stained with 5% uranyl acetate before observation.

The unique role of dHK in endocytic trafficking (Sunio et al. 1999) prompted us to explore the functions of mammalian Hook proteins. The cloning of hHK1 and hHK2 has been previously described (Krämer and Phistry 1999). Further screening of the human EST database revealed a cDNA encoding a fragment of hHK3. A clone encoding the full-length hHK3 protein of 718 aa was obtained by 5′ and 3′ RACE reactions (Materials and Methods). Comparison of homology across the Hook family defined several domains of similarity (Fig. 1 A). The highest degree of sequence identity is found in the acidic NH₂-terminal globular domain. An extended central coiled coil motif, which mediates homodimerization in dHK (Krämer and Phistry 1996; Sevrioukov et al. 1999), is also conserved. The basic COOH terminus, which is predicted to be globular in structure, is the most divergent segment in these proteins.

To begin characterization of these proteins, we affinity purified specific antisera against each of the three human Hook proteins. When tested on crude lysates from HEK293 cells, the antisera against hHK1, hHK2, and hHK3 detected endogenous proteins that migrated slightly slower than predicted from their molecular masses of 85 kD for hHk1, 83 kD for hHK2, and 83 kD for hHK3 (Fig. 1 B, lanes 1, 3, and 5). The signals were abolished by preincubation with an excess of the respective His₆–Hook fusion proteins (Fig. 1 B, lanes 2, 4, and 6), confirming the specificity of the antibodies.

Because of the similarity of coiled coil domains between the three human Hook proteins, we used a combination of cross-linking and immunoprecipitation experiments to determine whether Hook proteins form heterodimers. Postnuclear supernatants from HEK293 cells were immunoprecipitated with antibodies specific for each Hook protein, either directly or after cross-linking with 100 μM BS³. Western analysis of the immunoprecipitates revealed that the endogenous Hook proteins exist as members of three separate protein complexes (Fig. 2). All three human Hook proteins were predominantly found in complexes larger than expected for Hook dimers, but putative additional components of these complexes have not yet been identified. When complexes immunoprecipitated with one antibody were probed for the presence of the two other human Hook proteins, none was detected (Fig. 2, lanes 4, 6, 8, 12, 14, and 16). These results demonstrate that endogenous Hook proteins do not heterodimerize and suggest separate functions for the three human Hooks.

A distinct role for each of the Hook proteins was also supported by their different subcellular localizations. Endogenous Hook proteins were localized using confocal microscopy and double immunofluorescence labeling with well-characterized markers in Hep2 and Vero cells (Fig. 3; and data not shown). hHK1 (Fig. 3 A) and hHK2 (Fig. 3 D) localize to discrete punctate subcellular structures. Especially in the case of hHK2, the structures appeared often closely associated with microtubules. This was most easily visualized in the thin layer of cytosol just above the nucleus (Fig. 3, G–I). The identity of these structures has been elusive thus far: neither hHK1 nor hHK2 colocalized with antibodies labeling early endosomes (anti-EEA1 or antitransferrin receptors), MVBs (6C4), late endosomes (anti-M6PR), lysosomes (Lamp1), the ER (Calnexin), Golgi complexes (FTDC), or mitochondria (Cox1).

By contrast to hHK1 and hHK2, endogenous hHK3 was primarily detected in a perinuclear pattern reminiscent of the Golgi complex. When cells were double labeled for hHK3 and markers specific to the cis- and medial-Golgi compartments (Pepperkok et al. 1993; Bashour and Bloom 1998), a striking coincidence was observed (Fig. 3, J–L, Fig. 4; and data not shown). A fraction of the hHK3 protein is found outside the Golgi complex; this staining does not colocalize with markers for ER-to-Golgi intermediates (Fig. 3, M–O, ERGIC-53). A function of hHK3 in a ubiquitous cellular compartment such as the Golgi complex is consistent with its expression in many tissues and cell lines, including HEK293, Vero, Cos7, and CHO cells (Fig. 1 C; and data not shown).

The close colocalization between hHK3 and FTCD (Fig. 3, J–L), a peripheral membrane protein associated with the cis-Golgi and earlier compartments of the secretory pathway (Bashour and Bloom 1998; Gao et al. 1998), suggested that hHK3 localizes to the cis-face of the Golgi complex. Immunoelectronmicroscopy was used to further analyze the location of hHK3 within the Golgi complex of Vero cells. Immunogold labeling of hHK3 revealed a polarized distribution of hHK3 in the Golgi complex, consistent with labeling of the cis-face of the Golgi (Fig. 4, A–C). Gold particles were detected on the cis-most Golgi stacks (Fig. 4 A) but also the pleiomorphic vesicular–tubular clusters cis to the Golgi stacks (Fig. 4 B). We cannot rule out that a lack of antibody accessibility prevents the detection of hHK3 between Golgi stacks, but it is important to note that the more easily accessible area of the TGN closer to the cell surface was devoid of staining in the vast majority of cells examined.

Because of the absence of double labeling in these immunoelectron micrographs, we wanted to further address hHK3 localization in the Golgi complex by comparison to well-defined Golgi markers using immunofluorescence microscopy. Consistent with its colocalization with FTCD (Fig. 3, J–L), hHK3 also exhibited close colocalization with the cis-Golgi matrix protein GM130 (Nakamura et al. 1995). The close colocalization was most obvious at high magnifications (Fig. 4, D–F, insets). At low magnification, hHK3 colocalized with several Golgi markers in the general area of the Golgi complex, such as the intermediate compartment marker syntaxin5 (Fig. 4, G–I; Rowe et al. 1998), and the TGN markers TGN46 (Fig. 4, J–L; Prescott et al. 1997), and clathrin (Fig. 4, M–O; Brodsky 1985). At higher magnification, however, it was apparent that hHK3 staining was distinct from each of these three markers (Fig. 4, G–O, insets). The combination of these light and electron microscopy localization data suggested that hHK3 is enriched on the cis-face of the Golgi complex.

The drugs BfA and NZ were used to further probe the properties of hHK3. Treatment with BfA results in the release of β-COP proteins into the cytosol (Fig. 5 B) and the disruption of the Golgi complex (for review see Chardin and McCormick 1999). The strongest staining for hHK3, however, remained in a juxtanuclear accumulation even after a 1-h treatment with 10 μg/ml BfA (Fig. 5, A–I). Under these conditions, the cis-Golgi matrix protein GM130, localizes to dispersed punctate structures (Fig. 5 H; Nakamura et al. 1995; Seemann et al. 2000). A significant fraction of hHK3 labeling in BfA-treated cells colocalized with these GM130-positive punctae (Fig. 5, G–I, arrows).

Similar to TGN46 under these conditions (Fig. 5, D–F), hHK3 accumulated close to the MTOC (Fig. 6, A–F). Microtubules were necessary for this juxtanuclear localization of hHK3. When such BfA-treated cells were subsequently treated with 10 μg/ml NZ to disrupt microtubules, in the continued presence of BfA, hHK3 protein was found to redistribute to the peripheral GM130-positive punctate structures (Fig. 5, J–L). In cells treated with NZ only, much of hHK3 was observed associated with the scattered Golgi fragments, similar to other peripheral Golgi proteins (Fig. 5, M–O; Ho et al. 1989).

The requirement for microtubules in localizing hHK3 to its juxtanuclear position prompted us to further examine their association in vivo. During mitosis, hHK3 was localized to the microtubule asters in prophase (Fig. 6, G–I, arrowheads). When the mitotic spindle is evident in metaphase, hHK3 is found in a more diffuse cytoplasmic distribution, indicating that hHK3 function may be mitotically regulated (Fig. 6, J–L). A close association between hHK3 and microtubules in vivo was also evident in cells in which microtubules first had been depolymerized with NZ (10 μg/ml) for 1 h, and then bundled by the addition of taxol (40 μM) for 2–4 h. In such cells, hHK3 was enriched near the end of microtubule bundles (Fig. 6, M–O).

To determine whether hHK3 associates with microtubules in vitro, we used an in vitro microtubule spin-down assay (Fig. 7). Microtubules were assembled from purified tubulin. Taxol-stabilized microtubules were incubated with cytosol from HEK293 cells, or HEK293 cells expressing COOH-terminally truncated forms of the different human Hook proteins. Microtubules and associated proteins were then sedimented by high speed centrifugation. The pellets were probed for each of the human Hooks by Western blotting. All three human Hook proteins associated with microtubules (Fig. 7 A, lane 4). Furthermore, COOH-terminal truncations of each of the Hook proteins were also found to associate with microtubules. Although these experiments demonstrated the association of Hook proteins with microtubules, this assay did not exclude a requirement of other cytosolic partners in the microtubule-binding reaction.

To assess whether hHK3 binds directly to microtubules, we purified a set of His₆-tagged recombinant portions of hHK3. These polypeptides were coupled to Ni²⁺–agarose beads that were then incubated with in vitro–assembled taxol-stabilized microtubules. After the incubation, a low speed spin was used to pellet the hHK3-coupled beads and any bound microtubules. The amount of tubulin in supernatants and pellets was assessed by Western blotting. An NH₂-terminal domain of 164 aa from hHK3 was sufficient to bind directly to microtubules (Fig. 7 B, lane 4), whereas an His₆ fusion protein derived from the hHK3 coiled coil region did not bind (Fig. 7 B, lane 2). These experiments indicate direct binding of hHK3 to microtubules and define a novel microtubule-binding domain.

Because this NH₂-terminal domain exhibits the highest similarity to the dHK (Fig. 1 A), we tested the ability of dHK to bind to microtubules. Standard microtubule spin-down assays demonstrated that dHK from Drosophila cytosol bound exogenous bovine microtubules (Fig. 8 A). To test whether dHK can also bind to Drosophila microtubules, GTP and taxol were used to stabilize endogenous microtubules in Drosophila cytosol. When microtubules were sedimented by high speed centrifugation, dHK was found in the microtubule pellet (Fig. 8 B, lane 4). In the absence of stabilized microtubules, all of the dHK protein remained in the supernatant (Fig. 8 B, lanes 1 and 2). These results indicated that Hook proteins constitute a novel family of microtubule-binding proteins.

To further test the binding of hHK3 to Golgi membranes, we used an in vitro binding assay, using a highly enriched Golgi membrane fraction purified from CHO cells (Ktistakis et al. 1995). These Golgi membranes were incubated with cytosol, and then sedimented at low speed (15,000 g) through a 20% sucrose cushion. Proteins that remained in the supernatant and proteins that pelleted with membranes were analyzed by Western blotting (Fig. 9). No Hook3 was detected in samples containing Golgi membranes with no added cytosol (lanes 1 and 2). This indicated that added cytosol was the only source of hHK3 detected in the assay. In the absence of added membranes, hHK3 remained in the supernatant (Fig. 9, lanes 7 and 8). Endogenous hHK3 from HEK293 cytosol bound to the pelleted membranes (Fig. 9, lanes 4 and 6). However, membrane binding was eliminated by deletion of the COOH-terminal 166 aa (ΔC-hHK3; lane 6). Unlike hHK3, the hHK1 and hHK2 proteins did not pellet in the presence of the Golgi-enriched membranes.

Consistent with these in vitro results, the COOH-terminal 130 aa of hHK3 were sufficient to cause an enrichment of a dHK–hHK3C_589–718 chimera in the Golgi complex (Fig. 10, J–L). By contrast, the corresponding dHK protein was evenly distributed in the cytosol (Fig. 10, M–O). These experiments corroborated the specific binding of hHK3 to Golgi membranes, which we had observed in vivo (Fig. 4). Furthermore, they suggest that the domain-conferring membrane specificity to Hook proteins maps to their most divergent part, the COOH-terminal domain.

The binding of hHK3 to both microtubules and Golgi membranes suggested a role for hHK3 in the microtubule-dependent positioning of the Golgi complex. To test this possibility, we interfered with endogenous hHK3 function in transient overexpression experiments. In Drosophila, the expression of a COOH-terminal truncation of dHK mimics the hook loss-of-function phenotype (Krämer and Phistry 1999). Overexpression of a similarly truncated hHK3 construct (ΔC-hHK3_1–555) dramatically disrupted Golgi morphology in Vero cells (Fig. 10). All Golgi markers that we characterized, including antibodies against the KDEL receptor, FTCD, β-COP, and TGN46, displayed the same scattered fragmented Golgi appearance (Fig. 10, A–C; and data not shown). The absence of the COOH-terminal domain of overexpressed ΔC-hHK3_1–555 was not critical for its Golgi-disrupting effect. Overexpression of the full-length hHK3 caused a similar scattered Golgi morphology (Fig. 10, D–I).

In addition to the changes in Golgi localization, hHK3 overexpression also disrupted the microtubule network. Microtubules appeared less well organized; for example, they often crossed each other in the cell periphery (Fig. 10 R; and data not shown), an occurrence rarely seen in untransfected cells (Fig. 10 P). Microtubular disorganization correlated with a significantly increased fraction of cells with two or more nuclei in the populations of cells expressing full-length hHK3 or several hHK3 truncations and chimeras (Fig. 10 S). This disruption of the microtubular network was not sufficient, however, to cause dispersal of the Golgi complex. Cells expressing several hHK3 and dHK truncations and chimeras retained the well-defined and -positioned Golgi complex, although their effects on microtubules were similar to that of full-length hHK3 or ΔC-hHK3_1–555 (Fig. 10, J–O and S). Together, these experiments are consistent with a direct role of hHK3 in the positioning of the Golgi complex.

Hook proteins may constitute a novel family of microtubule linker proteins. Although motor proteins of the dynein and kinesin families catalyze the directed movement of membranous organelles along microtubules, there is growing evidence that nonmotor linker proteins are necessary to properly position and organize these cellular compartments (for review see Allan and Schroer 1999). The results presented here, together with the genetic analysis of the Drosophila hook gene (Krämer and Phistry 1996, Krämer and Phistry 1999; Sunio et al. 1999), support a model in which Hook proteins are microtubule-binding proteins that interact with specific compartments of the secretory and endocytic pathways.

A comparison of the subcellular localization of different Hook proteins suggests roles in separate membrane-trafficking pathways. Previously, we demonstrated that the dHK is localized to endosomes (Krämer and Phistry 1996). Here, we document the localization of a large fraction of hHK3, but not hHK1 or hHK2, to Golgi membranes (Fig. 3,Fig. 4,Fig. 5). The COOH-terminal domain of hHK3 is required (Fig. 9) for binding to Golgi membranes and is sufficient to redirect dHK to the mammalian Golgi complex (Fig. 10, J–L). This domain is the least conserved between different Hook proteins; its diversity may therefore be responsible for the distinct subcellular localizations of Hook proteins.

Although Hook proteins interact with different membranes, they all share the ability to bind to microtubules. This interaction was demonstrated by several in vivo and in vitro experiments. In vivo, we showed the taxol-induced enrichment of hHK3 at the ends of bundled microtubules (Fig. 6, M–O). In vitro, spin-down assays using exogenous purified brain microtubules (Fig. 7 and Fig. 8 A) or endogenous microtubules (Fig. 8 B) confirmed the interaction of Hook proteins with microtubules.

Our in vitro assays with purified Hook proteins revealed the direct binding of hHK3 to microtubules and mapped its microtubule-binding site to its NH₂-terminal domain (Fig. 7 B), which is highly conserved among Hook proteins (Fig. 1). It is unlikely that this interaction reflected nonspecific binding to negatively charged brain microtubules. The microtubule-binding domain encompassing the NH₂-terminal 164 aa of hHK3 has an isoelectric point of 4.3, arguing against nonspecific electrostatic interactions. Furthermore, Hook proteins also bound to endogenous nonbrain-derived microtubules (Fig. 8 B; Walenta, J.H., and H. Krämer, unpublished data). Although this NH₂-terminal domain is well-conserved among Hook proteins, it is not similar to any previously described microtubule-binding domain for other linker (De Zeeuw et al. 1997; Infante et al. 1999; Scheel et al. 1999) or motor proteins (Gee et al. 1997; Kozielski et al. 1997; Mandelkow and Johnson 1998; Kikkawa et al. 2000). The changing extent to which hHk3 colocalizes with microtubules in different stages of the cell cycle (Fig. 6) indicates its binding to microtubules is dynamically regulated in vivo.

Microtubule binding may also explain the localization of hHK3 in BfA-treated cells. BfA interferes with the formation of coat protein I coats and eventually causes the redistribution of resident cis- and medial-Golgi enzymes to the ER (for review see Chardin and McCormick 1999). Under these conditions, a small fraction of hHK3 localizes to punctate structures, which are marked by the Golgi matrix protein GM130 (Fig. 5, G–I; Seemann et al. 2000). The majority of hHk3, however, maintains a localization near, but distinct from, the MTOC in the absence of the Golgi complex (Fig. 5 and Fig. 6, A–F). This localization is dependent on microtubules, because, when NZ is added to BfA-treated cells, the vast majority of hHk3 relocates to the GM130-positive structures. This behavior is distinct from that of other cis- or trans-Golgi proteins after BfA treatment (Fig. 5; Lippincott-Schwartz et al. 1991; Nakamura et al. 1995; Füllekrug et al. 1999), but is not surprising for a protein that binds to Golgi membranes and microtubules and suggests the possibility that hHK3 may play an active role in the localization of the Golgi complex close to the MTOC.

Such a model is consistent with the results of hHK3 overexpression experiments. Overexpression of hHK3 or a COOH terminally truncated version dramatically dispersed the Golgi complex (Fig. 10), consistent with a proposed role of hHK3 in specifying the position of the Golgi complex. Dispersal of the Golgi complex can be achieved, however, by different manipulations based on very different mechanisms. One class of pharmacological or overexpression experiments interferes with membrane fission or fusion events, such as protein exit from the ER (e.g., Shima et al. 1998; Zaal et al. 1999) or interference with Golgi membrane fusion (Hatsuzawa et al. 2000). Golgi morphology can also be disrupted by a second class of manipulations that interferes with the transport of ER-to-Golgi intermediates along microtubules (Bloom and Goldstein 1998). This can be achieved by the NZ-induced dissociation of microtubules, which results in the appearance of distinct Golgi units at the location of ER exit sites (Storrie et al. 1998). Similar changes in Golgi morphology are caused by the inactivation of cytoplasmic dynein, either genetically (Harada et al. 1998) or by the overexpression of the dynamitin subunit (Burkhardt et al. 1997; Presley et al. 1997).

The effect of hHK3 overexpression on the Golgi complex appears to be more similar to that second class of microtubule-related manipulations. In hHK3-overexpressing cells, all tested Golgi markers (including GM130, TGN46, β-COP, and KDEL receptor) were found in discrete punctate structures rather than distributed in the ER membrane (Fig. 10; data not shown). As in NZ-treated cells (Bloom and Goldstein 1998), hHK3 overexpression did not appear to inactivate Golgi function. HEK293 cells overexpressing hHK3 and displaying the dispersed Golgi morphology could be grown for several weeks under G-418 selection. Furthermore, green fluorescent protein–tagged VSV-G protein was secreted to the plasma membrane in Vero cells overexpressing hHK3 (Walenta, J.H., and H. Krämer, unpublished observation). These consequences of overexpressing hHK3 protein are remarkably different from that of another Golgi-associated microtubule-binding protein, GMAP-210 (Infante et al. 1999). Unlike hHK3, overexpression of GMAP-210 resulted in an enlargement of the juxtanuclear Golgi complex, indicating distinct roles in Golgi morphology for these two proteins.

Similar to the effect caused by inactivating dynein (Quintyne et al. 1999), overexpression of hHK3 not only resulted in the dispersal of Golgi membranes but also a disruption of the microtubule network (Fig. 10). This raises the possibility that overexpression of human Hook3 proteins disrupts the Golgi complex by indirectly interfering with dynein-mediated transport of ER-to-Golgi intermediates. Two observations argue against such an indirect effect: (a) overexpression of several other hHK3 truncations or dHK-hHK3 chimeras caused very similar defects in the microtubule network, but did not disperse the Golgi complex (Fig. 10 S), and (b) hHK3 proteins in vivo and in vitro bind to Golgi membranes consistent with a direct effect. The results on hand, however, do not permit us to unequivocally resolve this issue. Future hHk3 loss-of-function studies will be necessary to further test the proposed role of hHK3 as a Golgi-specific linker protein.

A function of Hook proteins as microtubule linker proteins is also consistent with the analysis of hook loss-of-function mutations in endocytic trafficking in Drosophila (Krämer and Phistry 1996, Krämer and Phistry 1999). In the endocytic pathway of Drosophila and mammalian cells, mature MVBs are intermediates in the normal delivery of cargo to late endosomes (Gruenberg and Maxfield 1995; Futter et al. 1996; Gu et al. 1997; Sunio et al. 1999) and serve as an important compartment in the sorting of endocytic and biosynthetic cargo (Felder et al. 1990; Hirst et al. 1998). The assembly of such mature MVBs is dramatically reduced in hook mutant cells (Sunio et al. 1999). This specific destabilization of an endocytic organelle is distinct from the effects described for interfering with microtubules or dynein function (Gruenberg et al. 1989; Aniento et al. 1993). In those studies, the pharmacological disruption of microtubules or depletion of dynein resulted in the accumulation of cargo in an intermediate compartment before delivery to late endosomes. However, delivery of endocytic cargo to late endosomes is not blocked in cells lacking Hook. In the absence of mature MVBs, delivery to late endosomes might proceed through the direct fusion of the small, immature MVB fragments observed in hook mutants (Sunio et al. 1999). From these studies, it appears that Hook function in Drosophila is not necessary for internalization of cargo or its delivery to late endosomes. Rather, dHK appears to be responsible for the assembly or stability of mature MVBs, analogous to the proposed role of hHK3 in the mammalian Golgi complex.

Finally, it is interesting to consider Hook proteins in their evolutionary context. Yeast, which have fragmented Golgi and do not use microtubules for secretion (Jacobs et al. 1988), do not have Hook family members. With increasing cellular and organismal complexity, we find an increasing number of Hook proteins: Drosophila has a single Hook family member and here we have identified three human Hook proteins. Our results suggest that Hook proteins are an evolutionary response to the increased cellular complexity of higher organisms. This increased complexity necessitates precise organization of membrane compartments and assembly of intermediates for efficient trafficking over long distances. Our knowledge of microtubule-dependent motor proteins provides a solid framework for understanding the transport of organelles along microtubules. However, the model of microtubule linkers necessary for the anchoring of different organelles in their characteristic positions is still rapidly evolving (Pierre et al. 1992; Echard et al. 1998; Nielsen et al. 1999; Schroer 2000). The four Hook proteins discussed here expand on this theme. The identification of additional Hook proteins and their membrane receptors will further enhance our understanding of the interactions between microtubules and specific organelles.

We thank Richard Anderson for advice. We thank Hans-Peter Hauri (University of Basel, Basel, Switzerland), Hans-Dieter Söling (Universität Göttingen, Göttingen, Germany), Joachim Füllekrug (EMBL, Heidelberg, Germany), Frances Brodsky (UCSF, San Francisco, CA), Jean Gruenberg (University of Geneva, Geneva, Switzerland), Bill Balch (Scripps Research Institute, La Jolla, CA), and Annette Hille-Rehfeld (Universität Göttingen, Göttingen, Germany) for antibodies. Michael G. Roth provided invaluable advice regarding our biochemical experiments. We are grateful to Michael S. Brown, Michael G. Roth, George S. Bloom, Ellen A. Lumpkin, and the Krämer laboratory members for comments on the manuscript.

This work was supported by National Institutes of Health grant EY10199 and grant I-1300 from the Welch Foundation.