ESCRT-III activation by parallel action of ESCRT-I/II and ESCRT-0/Bro1 during MVB biogenesis

The endosomal sorting complexes required for transport (ESCRT) pathway facilitates multiple fundamental membrane remodeling events. Previously, we determined X-ray crystal structures of ESCRT-III subunit Snf7, the yeast CHMP4 ortholog, in its active and polymeric state (Tang et al., 2015). However, how ESCRT-III activation is coordinated by the upstream ESCRT components at endosomes remains unclear. Here, we provide a molecular explanation for the functional divergence of structurally similar ESCRT-III subunits. We characterize novel mutations in ESCRT-III Snf7 that trigger activation, and identify a novel role of Bro1, the yeast ALIX ortholog, in Snf7 assembly. We show that upstream ESCRTs regulate Snf7 activation at both its N-terminal core domain and the C-terminus α6 helix through two parallel ubiquitin-dependent pathways: the ESCRT-I-ESCRT-II-Vps20 pathway and the ESCRT-0-Bro1 pathway. We therefore provide an enhanced understanding for the activation of the spatially unique ESCRT-III-mediated membrane remodeling. DOI: http://dx.doi.org/10.7554/eLife.15507.001


Introduction
The endosomal sorting complex required for transport (ESCRT) pathway mediates topologically unique membrane budding events. In multivesicular body (MVB) biogenesis, ESCRT-0, I and II sort ubiquitinated cargo by binding ubiquitin and endosomal lipids. ESCRT-III assembles into spiraling polymers for cargo sequestration, and together with the AAA-ATPase Vps4, remodels the membranes to generate cargo-laden intralumenal vesicles (ILVs).
How is Snf7 activated to promote ESCRT-III assembly and cargo sequestration? Previous studies have shown that ESCRT-II and Vps20 modulate Snf7 protofilaments, emphasizing a role of the upstream ESCRTs in defining the assembly and architecture of the ESCRT-III complex (Henne et al., 2012;Teis et al., 2010). Recently, we have determined X-ray crystal structures of Snf7 protofilaments in the active conformation (Tang et al., 2015). Here, using genetics and biochemistry, we identify two parallel ubiquitin-dependent pathways that regulate Snf7 activation through both the Snf7 N-terminal core domain and the C-terminal a6 helix, providing an enhanced understanding of the activation of ESCRT-III-mediated membrane remodeling at endosomes.

Results
The a1/2 hairpin confers Vps20 with a unique identity Although Vps20 and Snf7 display a high degree of homology, they cannot complement each other. In order to identify regions of Vps20 essential for its function, we designed a series of Vps20-Snf7 chimeras and analyzed them by an established quantitative Mup1-pHluorin MVB sorting assay (Henne et al., 2012). Although a full-length Vps20 is required for function, retaining only the a1/2 hairpin of Vps20 while replacing the remainder of Vps20 with Snf7 (Vps20 1-105 -Snf7 107-240 ) is sufficient for sorting, albeit at~70% efficiency ( Figure 1A, Figure 1-figure supplements 1-2), suggesting that a1/2 is the minimal region unique to Vps20. This is consistent with the role of a1 of Vps20 in binding to the ESCRT-II subunit Vps25 (Im et al., 2009).

Screening for Vps20-independent Snf7 activation mutants
To investigate the role of Vps20 in nucleating Snf7 in vivo, we next applied an unbiased random mutagenic approach. We performed error-prone polymerase chain reaction on SNF7 and selected mutants that suppress the vps20D phenotype by growth on L-canavanine ( Figure 1B). Two snf7 point mutations in conserved residues, snf7 Q90L and snf7 N100I , showed a partial rescue of the canavanine sensitivity of vps20D ( Figures 1C-1D). Remarkably, in 'closed' Snf7, Gln90 of a2 is proximal to a4 (Tang et al., 2015), and Asn100 is an asparagine cap of the a2 helix ( Figure 1E). We propose that these mutations destabilize closed Snf7 by displacing a4 from a2 and extending the a2/3 helix.
Since conformationally active Snf7 resides on membranes, we performed liposome sedimentation assays. As predicted, Q90L enhances Snf7 membrane association from 41% to 78% ( Figure 1F). To further identify whether these substitutions trigger 'opening' in the core domain, we applied circular dichroism (CD) spectroscopy (Greenfield, 2006;Peter et al., 2004) on Snf7 a1-a4 , a truncated Snf7 construct with reduced membrane binding compared to the full-length proteins (Buchkovich et al., 2013). In the presence of liposomes, we observed a decrease of the negative absorption band at 208 nm and an increase at 222 nm in Q90L and N100I mutants, indicating an increase of a-helicity ( Figure 1G). These data agree with the hypothesis that Snf7 Q90L and Snf7 N100I trigger structural rearrangements, where the a2/3 loop becomes a-helical and extends into one elongated a-helix  Tang et al., 2015). Notably, this structural rearrangement still occurs only upon membrane binding. Moreover, snf7 Q90L and snf7 N100I complement snf7D in vivo, and Snf7 Q90L assembles into protofilaments in vitro (Figure 1-figure supplement 4), confirming a functional role of the mutants in activating Snf7.

Auto-activated Snf7 bypasses Vps20
Given that snf7 Q90L and snf7 N100I only modestly suppress vps20D, we hypothesized that a more stabilized 'open' Snf7 on endosomal membranes would improve the suppression. We combined the activation mutations with R52E (Henne et al., 2012) to further trigger 'opening', and swapped a0 of Snf7 with the N-terminal myristoylation motif of Vps20 to enhance its membrane-binding affinity (Buchkovich et al., 2013). This yielded myr-snf7 R52E Q90L and myr-snf7 R52E Q90L N100I , hereafter denoted as snf7** and snf7***, which sorted cargo with increased efficiencies, albeit not completely restoring wild-type levels ( Consistent with these observations, the ESCRT-dependent cargo GFP-Cps1 partially localized to the vacuolar lumen in vps20D with snf7** or snf7*** ( Figure 2B), indicating a substantial level of MVB sorting. Moreover, snf7** and snf7*** were also able to rescue the canavanine sensitivity of vps20D ( Figure 2C). Thus, these snf7 suppressors exhibit the ability to sort cargo at MVB.

Auto-activated Snf7 bypasses ESCRT-I and ESCRT-II
Intrigued by the vps20D suppression, we next wanted to test if these auto-activated Snf7 mutants could also bypass the loss of other ESCRT components ( Figure 3A). Among them, the downstream ESCRT-III subunits Vps24 and Vps2 are known to modulate Snf7 architecture (Henne et al., 2012;Teis et al., 2008) and recruit the AAA-ATPase Vps4 via their C-terminal MIM motifs for ESCRT-III disassembly (Obita et al., 2007). We found that auto-activated Snf7 does not suppress vps24D, vps2D or vps4D ( Figure 3B, Figure 3-figure supplement 1). This is consistent with the role of the suppressors in activating but not modulating or disassembling Snf7 filaments, reinforcing the division of labor among ESCRT-III subunits.
We next tested ESCRT-I mutants. ESCRT-I is a heterotetramer of Vps23, Vps28, Vps37 and Mvb12. Vps23 UEV domain recognizes ubiquitinated cargo, Vps37 N-terminal helix binds to membranes, and Vps28 CTD engages Vps36 GLUE domain of ESCRT-II. We expressed the suppressors in ESCRT-I single and ESCRT-I/II double deletion mutants and we observed near wild-type sorting efficiencies ( Figure 3D) with enlarged ILV sizes (Figure 3-figure supplements 5-6). Our data suggest that ESCRT-I and ESCRT-II set up the ESCRT-III architecture to program vesicle dimension.

Discussion
The ancient and conserved ESCRT-III membrane-remodeling machinery plays a critical role in numerous fundamental cellular processes, including MVB biogenesis, viral budding and cytokinesis. Building on our previous study (Tang et al., 2015), we focused on the predominant ESCRT-III subunit, Snf7, to understand the molecular mechanisms governing ESCRT-III for its dynamic conversion from an auto-inhibited soluble monomer to a membrane-bending polymer. Remarkably, a recent cryo-EM study on ESCRT-III IST1/CHMP1B co-polymer suggested that CHMP1B (Did2/Vps46) undergoes a similar structural rearrangement for assembly (McCullough et al., 2015), implying that the core domain extension is a common theme of ESCRT-III activation.
Here, using a mutagenic approach, we identified novel Snf7 point mutations that release the auto-inhibition of a3 and a4 as observed in the conformationally open structures. Surprisingly, this leads to Snf7 activation that functionally bypasses the ESCRT-III nucleator Vps20, as well as the ESCRT-II and ESCRT-I complexes. This suggests that Snf7, along with its downstream ESCRT components, Vps24, Vps2 and Vps4, but not ESCRT-I/II, are among the minimal machinery required for membrane remodeling.
Consistent with our observation, a very recent study suggested that ALIX and ESCRT-I/II function as parallel CHMP4B (Snf7 ortholog in human) recruiters in cytokinetic abscission (Christ et al., 2016).
While biochemical data suggest that Snf7 can be activated by specific point mutations in the core domain or truncation at the C-terminus in vitro, our genetic evidence indicat that the conformational equilibrium of Snf7 is tightly regulated by two pathways in vivo to achieve ubiquitin-dependent cargo sorting at endosomes: 1) ESCRT-I/ESCRT-II/Vps20 activates the N-terminal core domain of Snf7; 2) ESCRT-0/Bro1 activates the C-terminal a6 of Snf7 ( Figures 4E-F). Our results provide novel insights into a two-stage activation pathway for ESCRT-III-mediated membrane remodeling.
For Bro1 purification, Saccharomyces cerevisiae BRO1 was cloned into the pET23d vector (Novagen, Billerica, MA, USA) with an N-terminal His 6 -tag, induced by 1 mM IPTG at 18 o C overnight from BL21 E. coli cells, and purified by TALON metal affinity resin (Clontech). Protein-bound TALON resins were washed in 500 mM NaCl, 20 mM HEPES pH 7.4, 20 mM imidazole, and eluted in 150 mM NaCl, 20 mM HEPES pH 7.4, 400 mM imidazole.

Yeast strain and plasmids
See Supplementary file 1 for a list of plasmids and yeast strains used in this study.

SNF7 random mutagenesis for vps20D suppressor screening
The DNA sequence of Saccharomyces cerevisiae SNF7 with 500bp of 5'UTR and 500bp of 3'UTR was amplified by Taq DNA polymerase with 20 mM MnCl 2 and manipulated dNTP (N=A, T, G, or C) concentrations of 250 mM for three dNTPs and 25 mM for the other dNTP. Four individual 50 mL PCR reactions with different dNTP ratios were mixed, purified and transformed in vps20D yeast, along with a restriction enzyme digested vector of 3'UTR-pRS416-5'UTR. Yeast cells were plated and grown on YNB-uracil for 3 days at 26 o C, and replica plated on YNB-uracil with 4.0 mg/mL of L-canavanine. Canavanine-resistant yeast colonies were selected, and gap-repaired pRS416 snf7 mutant were prepped, amplified and sequenced.
The degrees of ellipticity were measured at 4 o C and scanned from 260 nm to 200 nm. Molar ellipticity, , was then normalized using the following equation and plotted versus wavelength, where n=142 is the number of peptide bonds.

Negative stain transmission electron microscopy
Visualization of ESCRT-III assembly using purified recombinant ESCRT components was performed as previously described (Henne et al., 2012). Visualization of MVB in vam7 tsf yeast cells was performed as previously described (Buchkovich et al., 2013). Briefly, 30 ODV of mid-log vam7 tsf yeast cells were grown at 38 o C for 3 hr, and then fixed with 2.5% (v/v) glutaraldehyde for 1 hr and spheroplasted with zymolyase and gluculase before embedding in 2% ultra-low temperature agarose. Cells were incubated in 1% osmium tetroxide/1% potassium ferrocyanide for 30 min, 1% thiocarbohydrazide for 5 min, and 1% osmium tetroxide/1% postassium ferrocyanide for 5 min. After dehydration through an ethanol series, samples were transitioned into 100% propylene oxide and embedded in Spurr's resin. Note that osmotic gradients during fixation or dehydration might account for the MVB morphological defects and the larger mean ILV diameter compared to samples prepared by high-pressure freezing and automated freeze-substitution. However, all yeast cells used in these experiments were treated equally. All TEM was performed on a Morgnani 268 transmission electron microscope (FEI) with an AMT digital camera.