|Home | About | Journals | Submit | Contact Us | Français|
The endosomal sorting complex required for transport (ESCRT) machinery is responsible for membrane remodeling in a number of biological processes including multivesicular body biogenesis, cytokinesis, and enveloped virus budding. In mammalian cells, efficient abscission during cytokinesis requires proper function of the ESCRT-III protein IST1, which binds to the microtubule interacting and trafficking (MIT) domains of VPS4, LIP5, and Spartin via its C-terminal MIT-interacting motif (MIM). Here, we studied the molecular interactions between IST1 and the three MIT domain-containing proteins to understand the structural basis that governs pairwise MIT-MIM interaction. Crystal structures of the three molecular complexes revealed that IST1 binds to the MIT domains of VPS4, LIP5, and Spartin using two different mechanisms (MIM1 mode versus MIM3 mode). Structural comparison revealed that structural features in both MIT and MIM contribute to determine the specific binding mechanism. Within the IST1 MIM sequence, two phenylalanine residues were shown to be important in discriminating MIM1 versus MIM3 binding. These observations enabled us to deduce a preliminary binding code, which we applied to provide CHMP2A, a protein that normally only binds the MIT domain in the MIM1 mode, the additional ability to bind the MIT domain of Spartin in the MIM3 mode.
The endosomal sorting complex required for transport (ESCRT)2 is an evolutionarily conserved molecular machine that functions to sculpt biological membranes to produce curvature away from the cytoplasm (1,–4). It is required for a number of important biological processes, including intraluminal vesicle formation during multivesicular body biogenesis (5), budding of enveloped viruses (6), abscission during cytokinesis (7), and repair of small plasma membrane wounds (8). The ESCRT machinery consists of a series of protein complexes, including ESCRT-0, -I, -II, -III, and VPS4 complexes (9, 10). They are assembled, activated and eventually disassembled in an orderly and timely fashion during ESCRT activity. Specific protein-protein interactions involving modular protein binding domains play important roles in these processes. Some of these interactions are mediated by microtubule interacting and trafficking (MIT) domains (11). The MIT domain consists of a trihelical structural fold and binds short peptide sequences collectively known as the “MIT-interacting motifs” (MIMs).
In the ESCRT machinery, MIMs have been identified at the C-terminal ends of proteins that belong to the ESCRT-III complex (12, 13). There are four major ESCRT-III proteins: CHMP2 (yeast ortholog: Vps2), CHMP3 (Vps24), CHMP4 (Snf7), and CHMP6 (Vps20) (12). They share a common structural fold that includes an N-terminal core helical domain and a C-terminal region that includes the MIM sequence (14). In their inactive state, the ESCRT-III proteins are kept as monomeric, soluble proteins in the cytosol by engaging the C-terminal regions in intramolecular interactions with their N-terminal core domains (13, 15, 16). Upon activation, they oligomerize to form the ESCRT-III complex (17, 18), which is directly responsible for generating membrane curvature and likely catalyzing the vesicle scission reaction as well (19, 20). In addition to the four major ESCRT-III proteins, there are at least three ESCRT-III related proteins: CHMP1 (Did2), CHMP5 (Vps60), and IST1 (Ist1). Although their structures are similar to that of an ESCRT-III protein, they are generally not considered to be major structural components of the membrane-bound ESCRT-III complex but instead play important regulatory roles for the ESCRT machinery function (21,–28).
Membrane-bound ESCRT-III proteins are removed and recycled into the cytosol at the end of the vesicle budding reaction cycle, a process catalyzed by the AAA-ATPase VPS4 together with its cofactor LIP5 (Vta1) (12, 18, 29,–32). Both VPS4 and LIP5 contain the MIT domain at their respective N termini, with VPS4 having one and LIP5 having two (23, 33,–35). They are recruited to the membrane by binding to the exposed C-terminal MIMs of the membrane-bound ESCRT-III proteins. Binding of VPS4 and LIP5 by the ESCRT-III proteins also serves to regulate the VPS4 ATPase activity (22, 28). Hence, the MIT-MIM interaction plays an important role in determining the direction of information flow in the ESCRT machinery function. Here, we studied the structural basis that determines the specific binding mechanism in pairwise MIT-MIM interactions using the ESCRT-III protein IST1 as an example.
IST1 was initially identified as a yeast gene whose mutation affected multivesicular body biogenesis in vacuolar protein sorting (36, 37). Its mammalian ortholog was later found to be required for efficient abscission during cytokinesis (38). Biochemical analysis showed that IST1 binds to a number of MIT domain containing proteins, including VPS4, LIP5, and Spartin (38,–40). Spartin is a multifunctional protein that is involved in localization or shaping of membrane compartments (41,–44). The MIT-MIM-mediated interactions between IST1 and its binding partners are important for IST1 function as truncation of the C-terminal MIM sequence of IST1 led to defects in cytokinesis (24, 38). Interestingly, both Spartin and VPS4 are localized to the mid-body of dividing cells during cytokinesis, and this localization is dependent on their interaction with IST1 (24, 38, 39).
The MIT-MIM interaction has been shown to occur in one of the five binding modes, MIM1-MIM5, depending on where a MIM binds to on a MIT domain and the structure a MIM adopts when it interacts with a MIT domain (35, 45,–51). For example, a peptide binds as a short α-helix to a groove formed between the second and third α-helices of the MIT domain in the MIM1 mode (45, 46), whereas it binds as an extended loop structure to a groove formed between the first and third α-helices of the MIT domain in the MIM2 mode (47). Peptides are often identified as potential MIMs based on their unique consensus binding sequences. They usually bind to the MIT domain via one of the five identified mechanisms.
IST1 displays some unique MIT domain binding features that have not been observed in other ESCRT-III proteins. Previous mutagenesis studies have suggested that IST1 can bind to the MIT domains of VPS4, LIP5, and Spartin utilizing different MIT-MIM binding mechanisms (24, 39). Given the importance of IST1 in recruiting VPS4 and Spartin to the mid-body of dividing cells during cytokinesis, we sought to understand the structural basis of the interaction between IST1 and these different MIT domain-containing proteins. To this end, we have determined the high resolution crystal structures of the IST1 C-terminal MIM in complex with the MIT domains of VPS4, LIP5, and Spartin, which showed that IST1 binds to VPS4 and LIP5 in the MIM1 mode but to Spartin in the MIM3 mode. Structural comparison showed that both MIT and MIM contribute to determine the specific binding mechanism in pairwise MIT-MIM interaction. Using a preliminary binding code derived from this analysis, we were able to convert CHMP2A, a protein that normally binds the MIT domain only in the MIM1 mode, into one that can also bind to the MIT domain of Spartin in the MIM3 mode.
Human IST1312–335, IST1290–335, Spartin8–101, VPS4B1–89, and LIP51–162 were expressed in Escherichia coli Rosetta (DE3) cells using a modified pET-28b vector with a SUMO protein tag inserted between a His6 tag and the respective protein coding region. A tryptophan residue was appended to the C termini of IST1 fragments to facilitate spectral detection at 280 nm during purification. The addition of the residue did not affect IST1 interaction with various MIT domains as evidenced by the crystal structures as well as binding affinity measurement. Site-directed mutagenesis was performed using a standard PCR mutagenesis protocol (Stratagene). The codons used to introduce the point mutations for IST1290–335 were F321A (TTT → GCG), L324A (CTT→ GCG), R327A (AGG → GCG), F328E (TTT → GAG), and L331A (CTG → GCG). The codons used to introduce the point mutations for CHMP2A201–222 were D209F (GAT → CAG) and L216F (CTT → TTT).
All proteins were purified using a protocol as previously described (52). Briefly, bacterial cells were grown to mid-log phase in LB media at 37 °C and induced with 0.2 mm isopropyl β-d-1-thiogalactopyranoside for an additional 16–20 h at 16 °C. Harvested cells were lysed with sonication in buffer A (25 mm Tris (pH 8.0), 300 mm NaCl, 5 mm 2-mercaptoethanol, and 10 μg/ml (w/v) phenylmethylsulfonyl fluoride). Cell lysate was cleared by centrifugation, and supernatant was loaded onto a Ni2+-nitrilotriacetic acid affinity column. Bound protein was washed with buffer A and eluted with buffer A supplemented with 250 mm imidazole. Fractions were pooled, and ULP1 protease was added to cleave off the His6-SUMO tag. After dialysis against a buffer containing 50 mm Tris (pH 8.0) and 25 mm NaCl, the digested protein mixture was passed over a second Ni2+-nitrilotriacetic acid column to remove undigested protein and the tag. For protein complexes used in structural studies, IST1312–335 was mixed with Spartin8–101, VPS4B1–89, or LIP51–162 individually, and each complex was further purified by gel filtration chromatography on a Superdex75 column (GE Healthcare) equilibrated with a buffer containing 25 mm Tris (pH 7.3) and 100 mm NaCl.
All crystals were grown using the sitting drop vapor diffusion method. The Spartin8–101-IST1312–335 complex (20 mg/ml) was crystallized in 2.9 m sodium malonate (pH 6.0) at 4 °C, and crystals were cryo-protected in 3.0 m sodium malonate (pH 6.0). The VPS4B1–89-IST1312–335 complex (40 mg/ml) was crystallized in 2.58 m sodium malonate (pH 5.0) at 25 °C, and crystals were cryo-protected in 3.0 m sodium malonate (pH 5.0). The LIP51–162-IST1312–335 complex (10 mg/ml) was crystallized in 30% (w/v) polyethylene glycol monomethyl ether 5000 (PEG5000MME), 0.1 m (NH4)2SO4, and 0.1 m MES (pH 7.0) at 4 °C, and crystals were cryo-protected in 25% (w/v) PEG5000MME, 0.1 m (NH4)2SO4, 0.1 m MES (pH 7.0), and 30% (v/v) ethylene glycol.
All diffraction data were collected at Advanced Photon Source LS-CAT 21-ID-F (Argonne National Laboratory). Crystals were cryo-cooled in liquid nitrogen during data collection. The Spartin8–101-IST1312–335 complex crystal diffracted to 1.8 Å and belonged to the space group of P43212 with one complex in the asymmetric unit. The VPS4B1–89-IST1312–335 complex crystal diffracted to 2.5 Å and belonged to the space group of P622 with one complex in the asymmetric unit. The LIP51–162-IST1312–335 complex crystal diffracted to 1.6 Å and belonged to the space group of P212121 with one protein complex in the asymmetric unit. All data were processed using HKL2000 (HKL Research). The structures were solved by molecular replacement using the solution structures of the Spartin MIT domain (PDB entry 2DL1) and the VPS4B MIT domain (PDB entry 1WR0) and the crystal structure of LIP5 N-terminal domain (PDB entry 4TXP) as initial search models. The PHENIX software suite was used for structure determination and refinement (53). Model building was done with COOT (54). Data collection and structure refinement statistics are shown in Table 1.
To measure the binding affinity between IST1 C-terminal fragments and various MIT domains, IST1 fragments were titrated against the MIT domains using a VP-ITC microcalorimeter (MicroCal) at 25 °C. Data were analyzed using the Origin 7 Software. All proteins were dialyzed against a buffer containing 25 mm HEPES (pH 7.5) and 50 mm NaCl, centrifuged to remove any particulate, and degassed before analysis. IST1 fragments were injected at 0.5 mm against 0.05 mm Spartin8–101, 0.5 mm against 0.05 mm LIP51–162, and 0.85 mm against 0.09 mm VPS4B1–89.
GST pulldown analysis was performed following standard protocols in a phosphate-buffered saline solution supplemented with 1 mm DTT and 0.1% (v/v) Tween 20 (52). GST alone, GST-tagged IST1290–335 and its mutants, and GST-tagged CHMP2A201–222 and its mutants were first immobilized on glutathione-agarose beads. Purified MIT domains were incubated with the protein-immobilized beads for 1 h at 4 °C. The beads were extensively washed with the buffer before bound proteins were analyzed on SDS-PAGE and visualized by either Coomassie staining or Western blotting.
Proteins were separated by 15% SDS-PAGE and transferred onto nitrocellulose membrane. The membrane was first blocked by 5% (w/v) nonfat milk dissolved in the TBST buffer (50 mm Tris (pH 7.6), 150 mm NaCl, 0.05% (v/v) Tween 20) followed by incubation with mouse monoclonal antibody against a FLAG tag (1:5000 dilution, Sigma, catalogue number F3165) for 1 h. After three washes with the TBST buffer, the membrane was incubated with horseradish peroxidase-conjugated goat anti-mouse IgG (H+L) (1:5000 dilution, Anaspec, catalogue number 28173). After another three washes with the TBST buffer, the chemiluminescence reagent (Pierce ECL Plus Western blotting Substrate, Thermo) was added to the membrane. X-ray films (Premium Autoradiography Film, Denville Scientific) were exposed and developed to visualize the FLAG-tagged proteins.
Mutational and NMR spectroscopic studies have previously suggested that IST1 contains two potential MIM sequence motifs at its C terminus (24). Residues 312–335 resemble a peptide sequence that could bind in the MIM1 mode, and residues 290–308 resemble a peptide sequence that could bind in the MIM2 mode. These two sequence segments were hypothesized to wrap around the MIT domain, each binding to a different surface groove (24). With this model in mind, we initially attempted to crystallize various MIT-IST1 complexes using the IST1290–335 fragment but were not met with success.
To improve our chance of obtaining diffraction-quality crystals, we set to reevaluate the energetic contributions of the two predicted MIMs of IST1 to MIT domain binding. Isothermal titration calorimetry (ITC) was used to determine the dissociation constants (Kd) for individual binding reactions between IST1 fragments and the MIT domains. We compared the binding behaviors of IST1290–335 and IST1312–335 to LIP5 MIT domain (LIP5MIT, residues 1–162), VPS4B MIT domain (VPS4BMIT, residues 1–89), and Spartin MIT domain (SpartinMIT, residues 8–101). The results showed that there was no significant difference in binding affinity between the two IST1 fragments when they bound to any of the three MIT domains (Fig. 1). Furthermore, in the double MIM binding model of IST1, Leu-297 was predicted to be important for binding (24). However, L297D mutation of IST1 did not have a significant impact on the binding affinities between IST1290–335 and any of the three MIT domains, suggesting that the residue is not required for MIT-IST1 interaction (Fig. 1). We, therefore, concluded that residues 290–311 of IST1 are dispensable for its interaction with the three MIT domains and that IST1312–335 is a minimal fragment that is suitable for use in complex formation with all three MIT domains in the following structural studies.
We first crystallized IST1312–335 in complex with either VPS4BMIT or LIP5MIT. The VPS4B-IST1 complex structure was refined to 2.5 Å resolution (Table 1) with an R-factor/Rfree of 22.5%/27.1%. The LIP5-IST1 complex structure was refined to a 1.6 Å resolution (Table 1) with an R-factor/Rfree of 17.8%/20.1%. Some of the N-terminal residues (312–319 for both the VPS4B complex and the LIP5 complex) of IST1 were not visible in the electron density map and assumed to be disordered.
LIP5MIT contains two tandem-linked MIT domains. Binding of IST1 occurs exclusively on the first domain. In both structures the binding site for the IST1 peptide is located on the surface groove formed by the second and third α-helices of the MIT domain (Fig. 2A). Most of the residues that form the peptide binding sites in the two structures occupy equivalent positions on the binding surface. For example, the binding site of VPS4BMIT consists of side chains of Leu-31, Val-38, Leu-42, Cyc-62, Leu-66, Ala-69, Gln-35, Gln-39 and Thr-63 (Fig. 2B). The equivalent positions in the LIP5MIT structure are Tyr-36, Met-43, Met-47, Leu-60, Met-64, Leu-67, Leu-40, Gln-44, and Ser-61. Despite the differences in identity for some of the residues, they all engage in similar interactions with the IST1 residues. The bound IST1 peptide forms a three-turn α-helix. Four residues on the helix are responsible for nearly all of the interactions with the two MIT domains. Three hydrophobic residues, Leu-324, Phe-328, and Leu-331, bind to an overall hydrophobic surface (Fig. 2C). The fourth residue of IST1, Arg-327, forms salt bridges/hydrogen bonds with two acidic residues on the MIT domain, Asp-67 and Glu-70 (Asp-65 and Glu-68 for LIP5) (Fig. 2D). The overall buried surface area at the interface is ~543 Å2 and 545 Å2 for the VPS4B-IST1 complex and the LIP5-IST1 complex, respectively. The mode of IST binding in these two structures is similar to what has been previously described as the MIM1 binding mode, as observed in the structures of the Vps4-Vps2 complex and the VPS4A-CHMP1A complex (45, 46). Therefore, IST1 binds to VPS4MIT and LIP5NTD in the MIM1 mode.
We next crystallized IST1312–335 in complex with SpartinMIT. The structure was refined to 1.8 Å resolution (Table 1) with an R-factor/Rfree of 19.0%/22.1%. As in the other two IST1 complex structures, residues 312–317 of the bound IST1 peptide were disordered.
The most significant difference in IST1 binding to Spartin as compared with VPS4B and LIP5 is that IST1 binds to a different surface groove of the MIT domain. The binding site, instead of being located on the surface groove between the second and third α-helices, is located on the surface groove between the first and third α-helices of the MIT domain (Fig. 3A). A similar mode of interaction has been previously observed in the structure of the microtubule severing AAA-ATPase Spastin MIT domain in complex with ESCRT-III protein CHMP1B (48). This type of interaction where a peptide binds to a MIT domain as a helix on the surface groove between the first and third α-helices has been referred to as the MIM3 binding mode. Therefore, the C-terminal peptide sequence of IST1 can bind to different MIT domains either in the MIM1 or in the MIM3 mode.
Residues that line the binding groove of SpartinMIT are mostly hydrophobic, but a few polar side chains are also present. These include Tyr-20, Phe-24, Val-27, Asn-28, Leu-31, Asp-34, Asn-32, and Glu-35 of helix 1 and Lys-80, Thr-84, Asn-87, Val-88, and Arg-91 of helix 3 (Fig. 3B). Interestingly, there are also two residues from helix 2, Leu-53 and Tyr-46, lining the floor of the binding groove. The bound IST1 peptide forms a three-turn α-helix with a total of seven residues making direct contacts with the Spartin MIT domain. At the core of the interface, hydrophobic residues Ile-319, Phe-321, Leu-324, Phe-328, and Leu-331 of IST1 make van der Waals contacts with the hydrophobic surface of the MIT domain (Fig. 3C). In particular, two bulky aromatic side chains Phe-321 and Phe-328 fit snugly into two deep surface pockets. At the periphery, Asp-322 and Arg-327 of IST1 participate in specific hydrogen bond and salt bridge interactions with the Spartin residues. Arg-327 forms a total of four hydrogen bonds with Glu-35, Asn-28, and Asn-32 from Spartin (Fig. 3D). Compared with the MIM1 binding surface on VPS4B or LIP5, the MIM3 binding surface on Spartin is much more extensive with some deep binding pockets. The overall buried surface area at the interface is ~740 Å2.
As noted above, the MIM3 binding mode has been previously observed in the crystal structure of the Spastin-CHMP1B complex (48). Comparison between the two structures revealed interesting features unique to the MIM3 binding mode. Although CHMP1B binds to Spastin as a much longer six-turn α-helix, the C-terminal three turns fits into a hydrophobic groove on the surface of the Spastin MIT domain in a nearly identical fashion as IST1 to Spartin (Fig. 4A). Six residues on these three helical turns make critical contribution to Spastin binding. Asp-186, Leu-188, Arg-191, Leu-192, and Leu-195 of CHMP1B are structurally analogous to Asp-322, Leu-324, Arg-327, Phe-328, and Leu-331 of IST1. The binding pockets for these side chains are formed mostly with conserved residues on the MIT domain surface (Fig. 4C). Hence, the binding modes for the C-terminal three helical turns of IST1 and CHMP1B are similar in the two cases. Based on the structural alignment, Gln-185 of CHMP1B is at a similar position as Phe-321 of IST1. Although they are both critical for high affinity MIM3 binding, their modes of interaction are quite different. Phe-321 of IST1 inserts itself into a deep hydrophobic pocket on Spartin, whereas Gln-185 of CHMP1B is involved in an extensive hydrogen bond network with His-120 and Asn-184 of Spastin (Fig. 4B).
Although the IST1 peptide used in our study is long enough to form a six-turn α-helix, only a three-turn α-helix was observed in the structure. A close examination of the Spartin-IST1 structure offered a plausible explanation. N-terminal to Phe-321, the bulky side chain of Tyr-20 of Spartin sticks out on the protein surface (Fig. 4D). If the IST1 peptide were to form a longer helix, it would clash directly with Tyr-20. Instead, the N-terminal portion of the IST1 peptide breaks away from the helical trajectory. This conformation appears to be further stabilized by the interaction of Ile-319 with Spartin. Consequently, the N terminus of the IST1 peptide extends away from the MIT domain and adopts a flexible conformation not observable in the current crystal structure. The structural equivalent of Tyr-20 in Spastin is His-120, which has moved out of the helical path due to its interaction with Gln-185 of CHMP1B. Therefore, CHMP1B is able to bind to Spastin as a much longer helix.
Many ESCRT-III proteins contain C-terminal sequences that can interact the MIT domain in the MIM1 binding mode. Yet, only two of them (IST1 and CHMP1B) have been observed to also interact the MIT domain in the MIM3 binding mode (45, 47, 48). To gain further insight into the mechanism of selectivity for an ESCRT-III protein to bind specific MIT domains either in the MIM1 or in the MIM3 mode, we compared the MIT-MIM interfaces in the three determined MIT-IST1 complex structures.
Five residues of IST1 appear to be important for the complex formation in the three structures. Common to all three are Leu-324, Phe-328, Arg-327, and Leu-331. The three hydrophobic residues bind to hydrophobic pockets on the surface of the respective MIT domain, and Arg-327 is involved in multiple hydrogen bond interactions. Among them, the binding pocket for Phe-328 appears to be deeper in the Spartin MIT domain than in the other two MIT domains. On the other hand, there is a significant difference in the way Phe-321 interacts with the three MIT domains due to the distinct conformation the IST1 peptide adopts in the three structures (Fig. 5A). In the LIP5-IST1 complex structure, Phe-321 is located in a loop region of the peptide and points away from the interface (Fig. 5B). It does not make any contact with LIP5. Phe-321 interacts with the other two MIT domains in different manners. In the Spartin-IST1 complex structure, Phe-321 is deeply buried in a hydrophobic pocket and interacts with Phe24, Tyr-20, and Thr-84 (Fig. 5D). In the VPS4B-IST1 complex structure, the phenylalanine side chain interacts with a much shallower binding pocket formed by Val-45 and the aliphatic portion of the Arg-59 side chain (Fig. 5C).
To evaluate the energetic contribution of individual IST1 residues to all three binding interactions, point mutations were introduced at selective sites on IST1, and mutant proteins were examined for their binding interactions with the three MIT domains. We chose to use IST1290–335 in our binding analysis because full-length IST1 was unstable and degraded rapidly. Wild-type IST290–335 displayed strong binding to all three MIT domains. F321A had little impact on binding to LIP5 but displayed dramatically reduced binding to Spartin in both ITC and GST pulldown analyses (Fig. 6, A and B). GST pulldown analysis also showed that F321A retained partial binding to VPS4B, but an accurate determination of Kd could not be made by ITC. These results were consistent with the structural observation where Phe-321 displayed different degrees of engagement with the three MIT domains. Leu-324, Phe-328, Leu-331, and Arg-327, on the other hand, are engaged in similar degrees of binding in all three MIT domain complex structures. Not surprisingly, L324A, F328E, and L331A abolished or dramatically reduced binding to all three MIT domains (Fig. 6, C, D, and E). R327A also displayed a significant decrease in binding to all three MIT domains (data not shown).
Given the importance of Phe-321 in supporting IST1 binding to the Spartin MIT domain in the MIM3 binding mode, we wondered whether adding a phenylalanine residue to an ESCRT-III protein that normally only binds in the MIM1 mode would provide the protein the additional ability to bind in the MIM3 mode. The ESCRT-III protein CHMP2A contains the MIM1 consensus sequence of (D/E)XXLXXRLXXL(K/R) (X indicates any residue) (Fig. 7A). The crystal structure of Vps2 (the yeast ortholog of human CHMP2A) in complex with the MIT domain of Vps4 showed that the protein binds to Vps4 in the MIM1 mode (46). Furthermore, CHMP2A does not bind to the MIT domain of Spartin (39).
We first introduced a D209F mutation (the equivalent position of Phe-321) to the CHMP2A sequence. However, GST pulldown analysis followed by Western blotting detection showed that CHMP2A201–222, D209F did not bind to the Spartin MIT domain. We recalled that the second hydrophobic residue of the MIM1 consensus sequence in IST1 is a bulky phenylalanine that binds to a deep surface pocket on the Spartin MIT domain. In other MIM1 containing ESCRT-III proteins including CHMP2A, this position is usually occupied by a leucine that may not provide sufficient binding energy for a stable MIM3 interaction (Fig. 7A). Hence, we introduced a second mutation to the CHMP2A sequence. Indeed, the double mutant, CHMP2A201–222, D209F/L216F, showed binding to the Spartin MIT domain (Fig. 7B). To determine where the mutant CHMP2A fragment binds on Spartin, we tested its binding to SpartinMIT, F24E. Phe-24 is located on the surface groove between helices α1 and α3 of the Spartin MIT domain and is critical for a peptide to bind in the MIM3 mode. No binding was observed between CHMP2A201–222, D209F/L216F and SpartinMIT, F24E. This result strongly suggested that the mutant CHMP2A bound to the Spartin MIT domain in the MIM3 mode. Interestingly, it retained its ability to bind to the VPS4B MIT domain in the MIM1 mode (data not shown). Therefore, the sequence of FXXLXXRFXXL appears to be sufficient for binding to the VPS4B MIT domain in the MIM1 mode as well as to the Spartin MIT domain in the MIM3 mode.
The ESCRT-III proteins play critical roles in the ESCRT machinery. Not only has their polymerization and subsequent membrane association been demonstrated to be directly linked to producing membrane curvature and generating force for vesicle scission, but they are also responsible for recruiting proteins, including ATPases and deubiquitinases, to the site of action. In addition, there are at least three ESCRT-III like proteins whose main function involves regulation of the ESCRT machinery. Central to the function of the ESCRT-III proteins is their binding to a unique protein-protein interaction domain called the MIT domain. Interactions between the MIT domain and the ESCRT-III proteins have been shown to be important for the proper progression of the cellular processes that require the ESCRT machinery. Therefore, understanding the structural basis of MIT domain interaction is key to understanding the molecular mechanism underlying the ESCRT-III function.
The MIT domain contains an up-and-down triple helix bundle fold and recognizes MIM peptide sequences located at or near the C termini of the ESCRT-III proteins (11). Despite the simple structural fold of the MIT domain, its interaction with MIM is by no means simple. A single MIT domain can utilize different surface grooves to bind different MIMs, with bound peptides adopting different structural conformations. Conversely, a single MIM can also bind to different MIT domains using different binding mechanisms. To date, at least five different modes of MIT-MIM interaction have been described structurally. Although these structures offered a glimpse into the complicated nature of MIT-MIM interaction, we are still far from having a code of binding that can be used to reliably predict how a specific pair of MIT-MIM might interact. In the current study we have determined the crystal structures of three new MIT-MIM complexes with the same MIM peptide sequence from IST1. This enabled us for the first time to directly compare how one MIM can engage in different types of MIT-MIM interactions.
Based on the results from previous structural studies, the consensus sequence for a peptide to bind in the MIM1 mode has been defined as (D/E)XXLXXRLXXL(K/R). The ESCRT-III proteins that are predicted to contain a MIM1 binding sequence include CHMP1, CHMP2, CHMP3, and IST1. The crystal structures of the VPS4B-IST1 complex and the LIP5-IST1 complex showed that IST1 binds to these MIT domains in the MIM1 mode. Only four residues of IST1, Leu-324, Arg-327, Phe-328, and Leu-331, are involved in direct binding in both structures. Therefore, it seems that the core sequence requirement for a peptide to bind in the MIM1 mode can be reduced to LXXRΦXXL (Φ stands for hydrophobic residues). The second hydrophobic residue in this sequence is often a leucine but in IST1 is occupied by a phenylalanine residue that binds to a deep surface pocket. It is possible that the more extensive interface provided by the bulky side chain offsets the requirement of the two polar interactions at the end of the original MIM1 consensus sequence.
IST1 was originally proposed to bind to the VPS4 MIT domain using two sequence segments, one in the MIM1 mode and the other in the MIM2 mode (24). During our study, we discovered that the predicted MIM2 segment was dispensable for IST1 to form stable complexes with the three tested MIT domains. Several lines of evidence supported our conclusion. First, there is no significant difference in binding affinity between a longer IST1 fragment with both MIM segments and a shorter one with only the MIM1 segment. Second, mutation of a consensus leucine residue that is required in the canonical MIM2 binding mode has no impact on MIT-IST1 interaction. Finally, the crystal structure of the LIP5-IST1 complex showed that the potential MIM2 binding site on the first MIT domain of LIP5 (which binds IST1) is engaged in an intramolecular interaction and thus not available for binding the putative MIM2 segment of IST1. Taken together, although we cannot rule out the possibility of minor contacts involving the sequence upstream of residues 312–335, it is likely that the C-terminal 24 residues of IST1 provide nearly all of the binding energy in engaging the MIT domains.
The MIM3 binding mode has only been previously observed in the Spastin-CHMP1B complex structure where CHMP1B binds as a six-turn α-helix to the surface groove formed by the first and third helices of the MIT domain. This interaction was observed to be of higher affinity than the MIM1 binding mode. The difference was attributed to the fact that the MIM peptide binds as a longer helix and its associated larger binding interface. In the current study we showed that a longer helix is not a structural requirement for IST1 to engage in the high affinity MIM3 interaction with Spartin. In fact, although the IST1 fragment used in crystallization is long enough to form a six turn α-helix, the surface feature of Spartin is not conducible for extended IST1 binding. How does the shorter IST1 helix achieve high affinity binding? Structural comparison between the two MIT-MIM3 complex structures offered some clues. Six MIT binding residues on the C-terminal portion of CHMP1B have corresponding residues in IST1. Two of them have different sequence identities. Both Phe-321 and Phe-328 of IST1 are buried deeply in their respective binding pockets. They likely provide significant binding energy to allow a shorter IST1 helix to bind Spartin tightly. In CHMP1B, these two residues are replaced by glutamine and leucine, respectively. Although both are still involved in MIT domain interaction, they are unlikely to provide a similar level of binding energy. Therefore, the C-terminal half of CHMP1B alone is not sufficient and requires additional contacts at the N-terminal half for high affinity binding.
The above observation suggests that structural features in both MIT and MIM are important in determining the specific binding mechanism in pairwise interaction. As we have seen in the two MIT-MIM3 complex structures, the surfaces where the IST1 helix and the C-terminal half of the CHMP1B helix bind are very similar. The deep binding pockets at these locations enable the ESCRT-III proteins to bind both Spartin and Spastin in the MIM3 mode. Yet, for a specific ESCRT-III protein to bind, certain sequence requirements must be satisfied as we have seen in the case of IST1 and CHMP1B. In the presence of the two bulky phenylalanine residues, it is likely that a three-turn helix is sufficient for IST1 to bind to both Spartin and Spastin in the MIM3 mode. On the other hand, binding of CHMP1B requires the peptide to adopt a longer six-turn helix due to the lack of sufficient binding energy from the C-terminal half alone. This binding is possible with Spastin but not with Spartin due to steric clashes with surface features on the Spartin MIT domain. This explains why IST1 can bind to both Spartin and Spastin, whereas CHMP1B only binds to Spastin.
Most of the ESCRT-III proteins that contain a C-terminal consensus MIM1 sequence fragment do not contain phenylalanine in the two corresponding positions and, therefore, cannot bind the Spartin MIT domain in the MIM3 mode. However, by introducing two phenylalanine residues into the C-terminal sequence of CHMP2A, we enabled the ESCRT-III protein to bind the Spartin MIT domain in the MIM3 mode. The role of the two phenylalanine residues was further hinted by examining evolutionary data. Although IST1 orthologs exist in all eukaryotic genomes, there is no Spartin or Spastin ortholog in fungal genomes. Phylogenetic analysis showed that there is a strong correlation between the presence of Spartin or Spastin gene in a given genome and the conservation of the two phenylalanine residues in the corresponding IST1 sequence (Fig. 8). These data suggest that the early function of IST1 was likely to bind the MIT domain only in the MIM1 mode. As Spartin and Spastin genes were later acquired in the genomes of high eukaryotes, IST1 also evolved to have the additional ability to bind in the MIM3 mode.
IST1 is required for recruiting both VPS4 and Spartin to the mid-body of dividing cells. The ability of IST1 to bind different MIT domains either in the MIM1 or in the MIM3 mode is thus biologically important. How IST1 partitions binding among various MIT domains in the cell is an interesting question in addressing the molecular mechanism of the ESCRT machinery. Given that IST1 is the only ESCRT-III protein recognized by Spartin and the binding affinity between the two proteins is much higher than that between IST1 and VPS4, it is tempting to speculate that IST1-Spartin interaction predominates in the cell. Our results have so far provided a preliminary code that can be used to selectively study specific pair of MIT-IST1 interaction through site-directed mutagenesis, which will provide new mechanistic insights with regard to the molecular interaction between IST1, VPS4, and Spartin in cytokinesis and other biological processes.
We thank the staff at the Advanced Photon Source Sector 21 (21-ID-F) for access and help with data collection. Use of the Advanced Photon Source, an Office of Science User Facility operated for the United States Department of Energy (DOE) Office of Science by Argonne National Laboratory, was supported by the United States DOE under Contract DE-AC02-06CH11357. Use of the LS-CAT Sector 21 was supported by the Michigan Economic Development Corp. and the Michigan Technology Tri-Corridor (Grant 085P1000817). We thank C. Vild and Y. Li for helpful discussion and K. Chinnaswamy and B. Wan for access and help for ITC experiments.
2The abbreviations used are: