Selective microautophagy of proteasomes is initiated by ESCRT-0 and is promoted by proteasome ubiquitylation

The ubiquitin-proteasome system (UPS) is a highly conserved proteolytic system that selectively degrades cellular proteins (Lecker et al., 2006). This pathway involves a cascade of reversible enzymatic reactions that covalently conjugates ubiquitin to substrates and targets them to the proteasome for degradation. Deubiquitylating enzymes (DUBs) remove ubiquitin from the substrates for recycling before the substrates are degraded into small peptides by the proteasome (Tomko and Hochstrasser, 2013). The 26S proteasome is composed of the 20S core particle (CP) and the 19S regulatory particle (RP), and the RP in turn comprises base and lid subcomplexes (Budenholzer et al., 2017). Dysfunction of the UPS has been implicated in the pathogenesis of many human diseases, including various cancers and neurodegenerative disorders (Thibaudeau and Smith, 2019).

Proteasomes are dynamic structures that change when cells are in distinct nutrient environments. Proteasomes are concentrated in the nucleus in yeast and most mammalian cells under rich nutrient conditions (Enenkel, 2014) but relocate to the cytoplasm and form membraneless proteasome storage granules (PSGs) under carbon starvation (Laporte et al., 2008). PSG formation occurs along with reorganization of many other cellular components into granules and foci in response to stress conditions; as with many other cellular granules, PSGs dissipate rapidly once favorable conditions are resumed, with proteasomes reentering the nucleus (Narayanaswamy et al., 2009; Sagot and Laporte, 2019).

Certain types of starvation can also lead to proteasomes being degraded by macroautophagy, likely in a nonselective manner (Li and Hochstrasser, 2020; Marshall et al., 2016; Marshall and Vierstra, 2018; Nemec et al., 2017; Waite et al., 2016). In macroautophagy, portions of the cytoplasm are engulfed by a double-bilayer membrane structure that matures into a closed vesicle called the autophagosome. The autophagosome fuses with the lysosome (or yeast vacuole) membrane, releasing a single large membrane vesicle, the autophagic body (AB), to the lysosome interior. The AB membrane and contents are then broken down by lysosomal hydrolases.

Recently, we have identified an AMP-activated protein kinase (AMPK)-regulated endosomal sorting complex required for transport (ESCRT)-dependent microautophagy pathway of proteasome degradation that occurs concomitantly with PSG formation under low-glucose conditions (Li et al., 2019). The ESCRT pathway is a widely deployed mechanism for topologically similar membrane remodeling events, such as multivesicular body (MVB) formation, enveloped virus budding and cell scission (Hurley, 2015). Interestingly, aberrant proteasomes are more likely to be removed by ESCRT-dependent microautophagy than to be stored reversibly in PSGs under these conditions (Li et al., 2019). These findings suggest an important role for AMPK signaling and ESCRT-mediated membrane remodeling in proteasome quality control when glucose levels are low (Li and Hochstrasser, 2020; Segev, 2020).

AMPK and ESCRTs are highly conserved protein complexes participating in diverse cellular processes (Coccetti et al., 2018; Vietri et al., 2020). AMPK is a heterotrimer, composed of one catalytic α-subunit and two regulatory subunits, β and γ (Ghillebert et al., 2011). The kinase is a master regulator of cellular energy homeostasis and is activated by glucose limitation and other stress conditions. In cells, AMPK is regulated by phosphatases and upstream kinases, as well as by autoinhibition of the α-subunit (Coccetti et al., 2018). Its activity is altered in many disease states, such as inflammation, diabetes and cancer, and a variety of compounds have been developed to regulate AMPK activity in human disease treatment (Carling, 2017).

The ESCRT factors are part of at least five distinct complexes, ESCRT-0, I, II, III and Vps4 (Schmidt and Teis, 2012). These complexes mediate a series of membrane binding and remodeling steps, starting with target protein recognition and enrichment by ESCRT-0, I and II and leading to membrane remodeling by ESCRT-III and Vps4 (Schmidt and Teis, 2012; Vietri et al., 2020). These steps can occur at different cellular membranes. To understand the roles of ESCRTs in determining the selectivity of proteasome microautophagy, we focused on Vps27, which along with Hse1, forms the Saccharomyces cerevisiae ESCRT-0 complex (Bilodeau et al., 2002). Vps27 is regarded as a ‘gateway protein’ for cargo protein recognition and downstream ESCRT complex assembly (Schmidt and Teis, 2012).

In the current study, we identify a mechanistic role for ESCRT-0 in selective proteasome microautophagy. Specifically, proteasomes in cells subjected to low-glucose stress are preferentially modified by low molecular mass or weight (LMW) ubiquitylation of proteasome subunits. Multiple ubiquitylating enzymes contribute to this pathway; we could show that the essential Rsp5 ubiquitin ligase is involved and that the N-lobe noncovalent ubiquitin-binding site in Rsp5 catalytic domain contributes as well. Proteasomes ubiquitylated in this way can be recognized by the ubiquitin-binding domains (UBDs) in the Vps27 subunit of ESCRT-0 and are subsequently cleared through ESCRT-dependent microautophagy. Our data suggest a model in which AMPK functions upstream in the ESCRT-dependent microautophagy pathway under low-glucose conditions by regulating Vps27 subcellular trafficking and vacuolar degradation.

Vps27 is known to relocalize from endosomes to the vacuole membrane when cultured yeast cells deplete the glucose supplied in the medium (diauxic shift), and this ESCRT-0 subunit is necessary for microautophagy (Oku et al., 2017). To understand how AMPK is involved in ESCRT-dependent microautophagy, we examined whether AMPK mutations affect Vps27 subcellular trafficking. We fused a yeast codon-optimized enhanced green fluorescent protein (yEGFP) to the N-terminus of Vps27 in the context of the normal VPS27 promoter and terminator, and chromosomally integrated the yEGFP-VPS27 fragment into the URA3 locus of vsp27Δ cells. When wild-type (WT) cells were grown in low-glucose for 1 or 4 days, the yEGFP–Vps27 fusion protein trafficked to the vacuole where it was degraded but left an intact yEGFP cleavage fragment (Fig. S1A); this fragment could also be visualized in the vacuole by yEGFP fluorescence (Fig. 1A). GFP is known to be resistant to vacuolar protease digestion (Klionsky et al., 2021). By contrast, in AMPK mutant cells under the same conditions, yEGFP–Vps27 failed to reach the vacuole and instead concentrated in cytoplasmic granules (Fig. 1A,B); these granules underwent what appeared to be rapid docking and separation events (highlighted with dashed square in Movie 1, which depicts a snf1Δ cell after 1 day in low glucose). Therefore, under glucose-limiting conditions, AMPK is required for the trafficking of Vps27 to the vacuole where it would normally be degraded.

To confirm that AMPK is required for vacuolar degradation of Vps27 during this process, we inactivated vacuolar proteases by deleting the PEP4 and PRB1 genes (Hecht et al., 2014). Disappearance of Vps27 was greatly decreased in this double mutant (Fig. S1B, lane 8 versus lane 1), and deletion of either the Snf1 or Snf4 subunit also led to strong Vps27 accumulation (Fig. S1B, lanes 2, 3). Interestingly, Atg8 and Vps4 made quantitatively distinct contributions to Vps27 degradation in low-glucose medium (top), but their loss had similar effects under glucose-free conditions (bottom) (Fig. S1B, lanes 4, 5). Specifically, Vps27 depletion was only weakly impaired in the atg8Δ strain under the former conditions, whereas vps4Δ cells had a greater defect. Atg8 is essential for macroautophagy but is dispensable for microautophagy; under low-glucose conditions, defects in microautophagy due to loss of the Vps4 ESCRT factor are much more pronounced (Li et al., 2019). These results suggest that Vps27 degradation occurs primarily via the ESCRT-dependent microautophagy pathway under low-glucose conditions.

Given the punctate cytoplasmic localization of Vps27 in AMPK mutants (Fig. 1A), we examined whether Vps27 cytoplasmic granules colocalized with proteasomal foci under glucose starvation. Interestingly, Vps27 puncta frequently colocalized with proteasomal foci in AMPK mutant cells after 4 days in low glucose but not under glucose-free conditions (Fig. 1C,D). Hence, Vps27 might function directly in proteasome quality control by selecting proteasomes for ESCRT-dependent microautophagy in low glucose.

Vps27 contains multiple well-studied functional domains (Fig. 2A), including the Vps27, Hrs and STAM (VHS) ubiquitin-binding domain (UBD), and two ubiquitin-interacting motifs (UIMs) (Bilodeau et al., 2002). It also has several proline-threonine-alanine-proline-like (PTAPL) motifs for interaction with Vps23 and ESCRT-I complex recruitment (Bilodeau et al., 2003; Katzmann et al., 2003), a Fab-1, YOTB, Vps27 and EEA1 (FYVE) domain that binds phosphatidylinositol-3′-monophosphate (PI3P)-rich membranes (Gaullier et al., 1998), and a C-terminal clathrin-binding domain (Bilodeau et al., 2003). We tested whether simultaneous mutation of the Vps27 UBDs, vps27-VHS-UIM (Ren and Hurley, 2010), affected membrane association and vacuolar trafficking of yEGFP–vps27-VHS-UIM. In cells grown in rich medium, WT yEGFP–Vps27 concentrated in small foci or granules (Fig. 2B), which are assumed to be endosomes (Hatakeyama et al., 2019; Katzmann et al., 2003), whereas the mutant derivative displayed a more diffuse pattern (Fig. 2B), which correlates with cargo protein-sorting defects (Ren and Hurley, 2010). In low glucose, WT yEGFP–Vps27 did not accumulate appreciably within cytoplasmic granules even in the presence of MG132 (a proteasome inhibitor; Lee and Goldberg, 1998); proteasome inactivation slightly reduced vps27-VHS-UIM granule staining (Fig. 2B,C). Interestingly, few apparent yEGFP cleavage products were detected, yet yEGFP–vps27-VHS-UIM loss was still apparent (Fig. S1C). Disappearance of the mutant protein, unlike the WT version, was inhibited in MG132-treated pdr5Δ cells, and this occurred regardless of whether Vps27 was fused to yEGFP (Fig. 2D; Fig. S1C). These results confirmed that the Vps27 UBDs, and presumably ubiquitin binding, are needed for efficient Vps27 recruitment to endosomes and thus proper cargo sorting, as previously reported (Ren and Hurley, 2010). They also suggest that these domains modulate the interaction of Vps27 with proteasomes during carbon limitation.

Vps27, as well as components of all the other ESCRT complexes, are essential for proteasome microautophagy during low-glucose stress (Li et al., 2019). This can be followed by the fragmentation of proteasome subunits into GFP-sized and larger fragments; these more complex fragmentation patterns are specifically seen when microautophagy is active, although the basis of this is unclear (Li et al., 2019). They nevertheless provide a useful signature for microautophagy-mediated proteasome degradation and likely other substrates such as Vps27 itself (e.g. see Fig. S1C).

It is striking that despite the sharp drop in overall Vps27 levels during carbon limitation, Vps27-dependent proteasome microautophagy, as measured by subunit fragmentation, continued (Fig. 2D). In the vps27-VHS-UIM mutant, proteasome fragmentation was modestly reduced under these conditions, while as noted above, the vps27-VHS-UIM protein itself was stabilized. Inhibition of proteasome activity generally enhances detection of fragmented proteasome subunits (Li et al., 2019), so we determined whether MG132 still affected the degree of subunit fragmentation in the vps27-VHS-UIM mutant. As shown in Fig. 2D, in glucose-starved proteasome-inhibited mutant cells, a moderate reduction in proteasome subunit fragmentation was seen compared to similarly inhibited WT cells. At the same time, PSG formation, in particular that of RP-containing PSGs, was significantly reduced in MG132-treated vps27-VHS-UIM cells accompanied by enhanced nuclear proteasomes (Fig. 2E; Fig. S2). From these results, we hypothesize that proteasomes, particularly aberrant or inactive ones, are recognized by Vps27 for microautophagy during glucose starvation, and that this requires the Vps27 UBDs for full efficacy.

To begin testing the above hypothesis, we checked whether ubiquitin affected proteasome subunit degradation, given that ubiquitylation plays an important role in cargo protein recognition by ESCRTs (Shields and Piper, 2011). Doa4 is a DUB enzyme that participates in the ESCRT pathway by removing ubiquitin from ubiquitylated cargoes (Amerik et al., 2000), while deleting DOA4 disrupts ubiquitin homeostasis and causes ubiquitin depletion from cells (Swaminathan et al., 1999). The microautophagy-specific proteasome subunit fragmentation pattern was lost in the doa4Δ mutant, whereas generation of free GFP from GFP-tagged subunits increased (Fig. 3A), presumably due to concomitant macroautophagy (Li et al., 2019). The WT fragmentation pattern was restored by ectopically expressing ubiquitin (Fig. 3A), suggesting microautophagy of proteasomes is ubiquitin dependent. As expected, Doa4 DUB activity is required for subunit fragmentation as the Doa4 catalytically inactive mutant doa4-C571S (Papa and Hochstrasser, 1993) failed to rescue the defect (Fig. 3B).

Mutations of ESCRT components interact genetically with doa4Δ and suppress the ubiquitin-depletion phenotype in cells lacking active Doa4 (Swaminathan et al., 1999). The microautophagy-specific proteasome fragmentation pattern was restored in doa4Δ vps27Δ and doa4Δ vps4Δ cells to the levels of the vps27Δ and vps4Δ single deletions (Fig. 3C); this partial suppression was expected because these ESCRT components are themselves necessary for normal proteasome fragmentation in low-glucose conditions (Li et al., 2019). Loss of AMPK, unlike loss of the ESCRT factors, did not suppress the fragmentation defects of doa4Δ cells, although it did suppress the hyperaccumulation of free GFP (Fig. S3). This could be due to the stronger defect in proteasome fragmentation of AMPK mutants (Li et al., 2019). Overall, these results suggest that ubiquitin is important for ESCRT-mediated proteasome sorting and degradation by microautophagy.

Using a nanobody to GFP, we directly examined proteasome ubiquitylation by co-immunoprecipitation (co-IP) from extracts of cells expressing GFP-tagged proteasomes. Interestingly, we observed a distinct LMW ubiquitylation pattern of proteasome subunits in ESCRT mutants along with a reduction in high molecular mass or weight (HMW) conjugates; the latter are likely to be noncovalently bound ubiquitylated proteasome substrates (Fig. 4; Fig. S4A). A broad range of LMW ubiquitylated species also accumulated in total cell lysates of ESCRT mutants; however, we attributed this to the known accumulation of Lys63-linked ubiquitin chains conjugated to cargo proteins that is caused by the block in the MVB pathway in these ESCRT mutants (Lauwers et al., 2009). Importantly, no ubiquitin-K63 chain conjugates were detected in the proteasome co-IPs (Fig. S4B). Proteasomes that were salt washed (Marshall et al., 2016) retained the LMW ubiquitylated species (Fig. S4C), and specific LMW ubiquitylated RP and CP subunits were detected in 6×His–ubiquitin pulldown experiments performed under denaturing protein extraction conditions (Fig. S4D). Consistent with this, in the IP samples, we detected a small number of ubiquitin-ubiquitin linkages by mass spectrometry (MS) and all involved K48. Proteasome ubiquitylation was also found to be higher following large-scale affinity purifications from ESCRT mutants expressing a 3×FLAG-tagged Rpn11 RP subunit (Fig. S5). Finally, besides ubiquitin, proteasome subunits and proteasome-associated proteins, such as RP assembly chaperones, were the dominant proteins identified by MS analysis from the immunoprecipitated proteasomes from WT and ESCRT mutants (Fig. S6; Table S1), also arguing against ubiquitylated substrates bound to proteasomes being major contributors to the LMW ubiquitylation species in ESCRT mutants. Overall, these results support the idea that the LMW ubiquitylated species are primarily ubiquitin-modified proteasome subunits that accumulate in ESCRT mutants under low-glucose conditions.

To test the hypothesis that aberrant or incompletely assembled proteasomes are preferred substrates for ubiquitylation, we performed proteasome ubiquitylation analyses in proteasome mutants (pre9Δ and sem1Δ) that bear high levels of aberrant proteasomes (Tomko and Hochstrasser, 2014; Velichutina et al., 2004) as well as in MG132-treated pdr5Δ cells containing catalytically inhibited proteasomes. As predicted, LMW ubiquitylated proteasome species were enriched in pre9Δ and MG132-treated pdr5Δ cells (Fig. 5A). This was not the case for sem1Δ cells; notably, this mutation limits 26S proteasome egress from the nucleus and prevents PSG accumulation in low glucose (Li et al., 2019), suggesting proteasome subunit ubiquitylation occurs after nuclear export. Conversely, ubiquitylation of proteasomes was almost abolished in doa4Δ cells, which have very little free ubiquitin (Fig. 5A), in line with the role of ubiquitin in proteasome fragmentation (Fig. 3). These results are consistent with aberrant proteasomes being more strongly modified by LMW ubiquitylation.

Next, we tested whether ESCRT-0 subunit Vps27 interacted with proteasomes. Given that endogenous Vps27 was also significantly degraded by microautophagy in low glucose (Fig. S1B) and that ESCRT-0-proteasome interaction is likely transient, we made constructs for constitutive overexpression of HA-tagged Vps27 or a mutant vsp27-VHS-UIM-PTAPL construct expressed from the strong GPD promoter to facilitate detection of Vps27-proteasome interaction. The vps27-VHS-UIM-PTAPL mutant protein (Fig. 2A) was designed to reduce or eliminate the ubiquitin-binding ability by mutating the Vps27 UBDs and simultaneously disturbing its association with downstream ubiquitin-binding ESCRT complexes (Bilodeau et al., 2002, 2003; Katzmann et al., 2003). Multiple UBDs have been identified in ESCRT components, including in Hse1 and Vps27 in ESCRT-0, Vps23 and Mvb12 in ESCRT-I, and Vps36 in ESCRT-II (Schmidt and Teis, 2012). The Vps27 PTAPL motifs interact with Vps23 for ESCRT-I recruitment, thereby increasing sorting efficiency of ubiquitylated cargo proteins (Bilodeau et al., 2003; Katzmann et al., 2003). With GPD-driven expression, HA–Vps27 and HA–vps27-VHS-UIM-PTAPL could be detected in cell lysates from vps27Δ cells grown under low-glucose conditions (Fig. 5B). No ATP was added to the IP lysis buffer, which favored CP–RP dissociation so that any preference of Vps27 for binding CP or RP could be determined.

Vps27 interacted with the proteasome RP complex, as demonstrated by its co-IP by anti-GFP antibodies when RP subunits Rpn5 or Rpn2 were tagged with GFP; binding was not detected with the CP, using CP subunit Pre10–GFP for co-IP (Fig. 5B). The CP is likely subjected to microautophagy through its association with RP and the RP–Vps27 interaction, given that full 26S proteasomes are dominant forms under low-glucose conditions (Fig. S7A). Surprisingly, the vps27-VHS-UIM-PTAPL mutant co-precipitated more RP complex than did WT Vps27, but there was also more of the mutant ESCRT subunit in the total lysates (Fig. 5B). This suggests that Vps27 can likely interact with the RP in ways that are independent of RP ubiquitylation and also that vps27-VHS-UIM-PTAPL may be less susceptible to degradation, possibly because it acts to stall microautophagy. Consistent with this interpretation, the difference between WT and mutant Vps27 binding to the RP complex was similar in vps4Δ cells (Fig. 5C), which cannot carry out microautophagy (Li et al., 2019). Interestingly, an enhanced interaction between proteasomes and both Vps27 and the vps27-VHS-UIM UBD mutant was observed in MG132-treated cells, and the binding of CP to Vps27 or vps27-VHS-UIM was now detectable as well (Fig. 5D), presumably because proteasome catalytic inhibition prevents 26S proteasome dissociation into separate CP and RP complexes (Kleijnen et al., 2007).

Vps27 interacts with Hse1 to form the ESCRT-0 complex (Bilodeau et al., 2003; Ren and Hurley, 2010). Hse1 is unable to bind ubiquitin in vitro in the absence of Vps27 (Bilodeau et al., 2002). We checked whether overproduced Hse1 was sufficient to interact with proteasomes in vivo. Using GPD-driven Hse1–HA expression in hse1Δ cells, we did not detect any interactions with proteasomes, nor did levels of LMW ubiquitylated proteasome species change (Fig. S7B). However, when both Vps27 and Hse1 were overproduced, Hse1 co-precipitated with the RP along with Vps27 (Fig. 5E), suggesting that the ESCRT-0 complex binds the RP via its Vps27 subunit.

We would predict a Vps27 mutant, such as vps27-VHS-UIM-PTAPL, which disturbs its ubiquitin-binding ability, should accumulate (aberrant) proteasomes that are modified by LMW ubiquitylation under glucose-limited conditions. To reconstitute endogenous levels of Vps27, we created a set of strains with either WT VPS27 or different vps27 mutant alleles (vps27-VHS-UIM, -PTAPL, -VHS-UIM-PTAPL) integrated at the chromosomal URA3 locus; the alleles were under the control of the natural VPS27 promoter and terminator in vps27Δ cells that also expressed GFP-tagged proteasome subunits (Pre10 or Rpn5). Integration of the empty vector served as a negative control. Based on proteasome ubiquitylation analysis, LMW ubiquitylated proteasome species accumulated in the vps27-VHS-UIM-PTAPL and vps27-VHS-UIM strains but not in VPS27 or vps27-PTAPL cells (Fig. 5F). This suggests that proteasomes modified by LMW ubiquitylation accumulate when these ubiquitylated forms cannot be recognized by ESCRT-0, or alternatively, that the interaction between Vps27 and unidentified ubiquitylated proteins other than proteasome is essential for the accumulation of proteasomes modified by LMW ubiquitylation.

It should be noted that these mutants will also block other ESCRT-0-dependent mechanisms such as the endosomal-MVB pathway (Bilodeau et al., 2003; Katzmann et al., 2003; Ren and Hurley, 2010). A distinctive ladder of ubiquitylated species accumulated in the mutants, as seen in the total lysates in Fig. 5F. These may be K63-linked ubiquitin oligomers that are important in the MVB pathway but appear to play little or no role in ubiquitylation of proteasomes (Fig. S4B). Because of their abundance, a fraction may nevertheless bind to proteasomes under the conditions used for the experiment in Fig. 5F.

We next examined whether the above Vps27 mutants also affected proteasome trafficking and degradation in the vacuole. We used the same set of yeast strains (Fig. 5F) and checked proteasome fragmentation (as a biochemical signature of microautophagy) and trafficking. Agreeing well with the data from vps27-VHS-UIM pdr5Δ cells (Fig. 2D), a minor proteasome fragmentation defect was observed in vps27-VHS-UIM cells (Fig. 6A). As predicted from proteasome ubiquitylation (Fig. 5F), vps27-PTAPL cells did not have any defects in proteasome fragmentation; by contrast, a much stronger defect in proteasome fragmentation was detected in vps27-VHS-UIM-PTAPL cells, particularly after four days of glucose deprivation (Fig. 6A), indicating the vps27-VHS-UIM-PTAPL mutations disrupted ESCRT-dependent microautophagy of proteasomes to the same extent as vps27Δ cells.

As noted above, Vps23 (ESCRT-I) directly associates with Vps27 by binding its PTAPL motifs (Bilodeau et al., 2003; Katzmann et al., 2003). We created a vps27-VHS-UIM vps23Δ strain, which retains the PTAPL elements in Vps27, and determined whether this mutant mimicked the proteasome microautophagy defect of vps27-VHS-UIM-PTAPL VPS23 cells. An additive defect in proteasome fragmentation was observed in vps27-VHS-UIM vps23Δ as compared to the individual vps23Δ and vps27-VHS-UIM-PTAPL mutants, especially in low-glucose conditions, at day 1 (Fig. 6A, left panel), suggesting that these ESCRT components function collaboratively to recognize proteasomes for microautophagy.

Cytological analyses by fluorescence and immunogold electron microscopy further demonstrated equivalent defects in proteasome trafficking and degradation in vps27-VHS-UIM-PTAPL and vps27Δ cells. Rather than forming spherical and bright PSGs as in VPS27 (WT) cells, irregular and faint proteasome puncta were observed in the cytoplasm, particularly at the perivacuolar region of vps27-VHS-UIM-PTAPL and EV (vps27Δ) cells (Fig. 6B), suggesting that accumulation of aberrant proteasomes interferes with normal PSG formation in glucose-limited cells. To preserve autophagic body (AB) and intralumenal vesicle (ILV) structures generated by macroautophagy and microautophagy, respectively, under low-glucose stress, the chromosomally integrated VPS27 and vps27-VHS-UIM-PTAPL alleles (at the URA3 locus) were placed in a vps27Δ pep4Δ prb1Δ background. Although ILV formation in the vacuole was previously observed with a Vps27–Hse1 complex lacking any intact UIMs (Bilodeau et al., 2002), ILV formation was blocked and only ABs were observed in vps27-VHS-UIM-PTAPL and vps27Δ (EV) cells, while both ILVs and ABs were observed in VPS27 cells (Fig. 6C,D). This phenotype is reminiscent of the microautophagy defect in vps4Δ cells (Li et al., 2019). Proteasome distribution in the vacuole was followed by anti-CP antibodies labeled with gold beads (black arrows in Fig. 6C). A comparable proteasome distribution pattern was found in vps27-VHS-UIM-PTAPL and vps27Δ (EV) cells (Fig. 6E,F). The results suggest that the ubiquitin-binding ability of ESCRTs is essential for ILV formation and thus proteasome microautophagy.

Given the intriguing ubiquitylation pattern of proteasomes in ESCRT mutants, we sought to find the ubiquitin ligase(s) responsible for their ubiquitylation. Rsp5 is the principal ligase involved in cargo protein ubiquitylation in the ESCRT-dependent endosomal-MVB pathway of protein targeting to the vacuole and is also known to interact with ESCRT-0 (MacDonald et al., 2020). Rsp5 belongs to the NEDD4 family of homologous to E6AP C-terminus (HECT) ligases (Shah and Kumar, 2021). NEDD4 ligases share an N-terminal lipid-interacting C2 domain, one to four central WW protein interaction motifs (Chen and Sudol, 1995), and a large C-terminal catalytic HECT domain; the HECT domain bears an N-lobe with a noncovalent ubiquitin-binding site and a C-lobe with a covalent (catalytic) ubiquitin-binding site (Shah and Kumar, 2021).

We tested whether Rsp5 contributed to proteasome ubiquitylation under low-glucose conditions. Rsp5 is essential for cell viability, so we performed the analysis in a temperature sensitive (ts) mutant, rsp5-1, which has an L733S mutation that causes dysfunction of Rsp5 at non-permissive temperatures (Wang et al., 1999). Proteasome ubiquitylation was inhibited in rsp5-1 at elevated (semi-permissive) temperature (30°C) but was comparable to WT at room temperature (RT, ∼24°C) (Fig. 7A). The defects in proteasome ubiquitylation correlated with reduced proteasome subunit fragmentation in rsp5-1 (Fig. S8A). Fragmentation of RP subunits Rpn5 and Rpn2 was more sensitive to defective Rsp5 compared to the Pre10 CP subunit, suggesting the RP is the main ubiquitylation target for initiating proteasome microautophagy.

To verify the importance of Rsp5 for proteasome ubiquitylation, we made a set of Rsp5-expressing constructs on CEN plasmids carrying RSP5, rsp5-L733S, or rsp5-F618A – the last to disable the HECT domain noncovalent ubiquitin-binding site (French et al., 2009) – and placed them in an rsp5Δ background. As expected, plasmid-borne rsp5-L733S (rsp5-1) caused defects in proteasome ubiquitylation comparable to those seen in rsp5-1 cells at 30°C (Fig. 7). The rsp5-F618A mutant also displayed defects in proteasome ubiquitylation (Fig. 7B), indicating that noncovalent ubiquitin binding by the Rsp5 HECT domain is important for proteasome ubiquitylation. Furthermore, the interaction between proteasome RP and Vps27 was reduced in rsp5-1 cells at 30°C (Fig. S8B). Together, these results implicate the Rsp5 ligase in proteasome ubiquitylation and proteasome–Vps27 interaction in cells subject to low-glucose stress.

Previously, we had identified a novel proteasome degradation pathway in low-glucose conditions involving AMPK-regulated ESCRT-dependent microautophagy (Li et al., 2019). In this study, we sought to understand more about AMPK regulation and the mechanistic role of ESCRTs in proteasome microautophagy. In low-carbon conditions, we find that AMPK promotes Vps27 protein turnover and its trafficking to the vacuole (Fig. 1; Fig. S1). Vps27 is essential for ILV generation in vacuoles during microautophagy (Fig. 6C,D). Genetic or pharmacological induction of aberrant or inactive proteasomes leads to increased LMW ubiquitylation of these particles in cells under low-glucose stress. Vps27 (ESCRT-0) appears to play an essential role in recognizing proteasome ubiquitylation and eliminating aberrant proteasomes through microautophagy (Figs 4–6), and thus also facilitating proper PSG assembly. PSG formation and proteasome microautophagy are concomitant and potentially physically linked processes, so it might be hard to cleanly separate the role of Vps27 in PSG formation from its function in microautophagy. The Rsp5 ubiquitin ligase contributes to proteasome ubiquitylation in this proteasome quality control pathway (Fig. 7). Our results are summarized in the working model in Fig. 8.

Exactly how AMPK regulates ESCRT-dependent microautophagy has yet to be determined. Previous studies have shown that target of rapamycin complex 1 (TORC1) phosphorylates and antagonizes the function of Vps27 in microautophagy (Hatakeyama et al., 2019), while AMPK negatively regulates TORC1 activity through phosphorylation and aggregation of its Kog1 subunit, which results in disassembly of the TORC1 complex during glucose starvation (Hughes Hallett et al., 2015). The AMPK and TORC1 signaling pathways have interlinked and generally antagonistic effects (González et al., 2020). For ESCRT-dependent microautophagy in low-glucose conditions, AMPK may inhibit TORC1 and thus relieve the suppressive effects of TORC1 on this process. It is also possible that AMPK competes with TORC1 and functions directly on ESCRT components for microautophagy initiation or activation, which would be reminiscent of the role of AMPK in macroautophagy (Kim et al., 2011).

Ubiquitylation of proteasomes could fine-tune their function in various ways, such as protein–proteasome interaction, proteasome activity, and cellular localization (Kors et al., 2019). At least 12 yeast proteasome subunits are known to be ubiquitylated in vivo and over 19 ubiquitylation sites have been detected (Hirano et al., 2015; Starita et al., 2012; Swaney et al., 2013). Only a small set of growth conditions and methods for proteasome purification have been reported in these studies, so the real numbers are likely to be substantially higher. Previous studies on possible functions of proteasome ubiquitylation are limited as well. In human prostate cancer cells, ubiquitylation of the CP subunit α2 favors the binding of δ-aminolevulinic acid dehydratase (ALAD); ALAD binds in place of 19S RP and functions as an endogenous proteasome inhibitor (Schmitt et al., 2016). RP ubiquitylation can also regulate proteasome activity. For example, polyubiquitylation of Rpn13 interferes with proteasome binding and degradation of ubiquitin-conjugated substrates in human cell lines (Besche et al., 2014), while monoubiquitylation of Rpn10 can limit the capacity of the UIM of Rpn10 to interact with substrates (Isasa et al., 2010). In our study, it is intriguing that aberrant proteasomes are preferably subject to LMW ubiquitylation (Fig. 5A), which appears to serve as a sorting signal for ESCRT-dependent clearance by microautophagy. It is unknown whether aberrant proteasomes are targeted by ESCRT-0 after their incorporation into PSGs or by bypassing PSGs altogether.

Ubiquitylation also plays a key role in proteasome macroautophagy (Marshall et al., 2016). The exact nutrient stress conditions help determine the distinct routes of proteasome autophagy in cells, and different proteasome ubiquitylation patterns may allow the diversification of proteasome autophagy routes, including the selective removal of aberrant or inactive proteasomes. Nitrogen starvation normally triggers nonselective proteasome macroautophagy that does not appear to involve extensive proteasome ubiquitylation (Marshall et al., 2016). By contrast, low glucose levels can simultaneously induce proteasome macroautophagy and microautophagy, and the latter appears to favor the removal of aberrant or inactive proteasomes that are either directly modified by LMW ubiquitylation or tightly bound to proteins that are modified in this way.

Chemical inactivation of proteasomes in yeast cells grown under non-starvation conditions triggers a selective proteasome macroautophagy pathway accompanied by extensive ubiquitylation of the inactivated proteasomes (Marshall et al., 2016). Although proteasome ubiquitylation is a common feature in determining the selectivity of proteasome microautophagy and macroautophagy, the molecular components involved in these two selective proteasome autophagy pathways are different. For instance, the Cue5 autophagy receptor and the Hsp42 chaperone are engaged in the recognition of inactivated proteasomes for selective macroautophagy (Marshall et al., 2016) but are dispensable for selective proteasome microautophagy (J.L. and M.H., unpublished data). On the other hand, the ESCRT-0 complex is involved in the recognition of aberrant proteasomes accumulated under low-glucose conditions; its role in selective proteasome macroautophagy has not been examined. Insofar as ESCRTs are required in nonselective proteasome macroautophagy upon nitrogen starvation, we would predict ESCRTs are also involved in selective proteasome macroautophagy but likely in a different way than in microautophagy, such as autophagosome closure (Zhou et al., 2019).

Multiple components of the ESCRT-0, -I and -II complexes contain UBDs, which provides structural plasticity for a wide variety of ubiquitylated cargo recognition and sorting mechanisms (Schmidt and Teis, 2012; Shields et al., 2009; Shields and Piper, 2011). In addition, having multiple UBDs can boost the avidity for ubiquitylated cargoes as individual ESCRT UBDs have a low affinity for ubiquitin; this might also allow recognition of very large cargo protein complexes with complicated ubiquitylation patterns (Piper et al., 2014), such as the LMW ubiquitylation of proteasomes we report here (Figs 4 and 5).

We still detected Vps27–proteasome RP interaction when ubiquitin binding by Vps27 or ESCRT-0 should have been eliminated, as in the vps27-VHS-UIM-PTAPL mutant (Fig. 5B–F). This suggests that ESCRT-0 can also interact with proteasomes in ways that are independent of RP ubiquitylation. It is possible that alternative binding partners may participate in these ESCRT-0–proteasome interactions. Nevertheless, the vps27-VHS-UIM-PTAPL mutation blocked ILV formation in the vacuole lumen to a degree comparable to the deletion mutant of VPS27 (Fig. 6). This implies that the ubiquitin-binding activity of Vps27 has a redundant role with ESCRT-I and/or ESCRT-II components in capturing ubiquitylated proteasomes. We note that the vps27-PTAPL mutant, which is defective for ESCRT-I and ESCRT-II binding, did not cause any major defects in proteasome degradation (Figs 5F and 6A), suggesting ESCRT-0 has additional sites for recruiting these downstream complexes, in line with previous findings (Shields and Piper, 2011).

In conclusion, we provide evidence here that ESCRT-0 functions in recognizing aberrant proteasomes through a proteasome ubiquitylation signature that makes these proteasomes favored targets for microautophagic elimination under low-glucose stress. This ensures that normal proteasomes, sequestered in PSGs, are readily available once a new source of glucose has been secured to allow a return to growth.

Yeast genetic manipulations were conducted using standard protocols (Dunham et al., 2015). Yeast strains used in this study are listed in Table S2. Unless specified in the figure legends, yeast cells were grown overnight at 30°C with vigorous agitation in synthetic complete (SC) medium (0.67% yeast nitrogen base without amino acids, 0.5% casamino acids, 0.002% adenine, 0.004% tryptophan, 0.002% uracil, and 2% glucose; labeled 2% C in figures); tryptophan and uracil single or double dropout media were used for plasmid selection. Cells were then back-diluted in fresh SC medium and grown to mid-exponential phase. The cells were pelleted and rinsed once with sterile water, followed by different treatments. For glucose starvation, cells were resuspended in SC medium containing 0.025% glucose (labeled 0.025% C in figures) or lacking glucose and cultured for ∼1 day and/or ∼4 days at 30°C, as specified in the figures. For experiments with the proteasome inhibitor MG132, all strains included the pdr5Δ allele to allow efficient intracellular accumulation of the drug (Collins et al., 2010). Cells were grown in SC medium as above, and mid-log cells were resuspended in SC medium containing 0.025% glucose and DMSO or 50 μM MG132 (Santa Cruz Biotechnology, catalog #sc-201270, lot #H1219) dissolved in DMSO and cultured for ∼1 day at 30°C.

Genes of interest were amplified from yeast genomic DNA and cloned into the following plasmids: pRS316 (Sikorski and Hieter, 1989), p424GPD or p426GPD (Mumberg et al., 1995) through restriction enzyme digestion and T4 DNA ligase-mediated ligation (NEB). The resulting plasmids were verified by enzymatic digestion and DNA sequencing. QuikChange site-directed mutagenesis (Agilent) was performed to generate the doa4-C571S and rsp5-L733S and rsp5-F618A mutants. Plasmids used in this study are listed in Table S3.

For epifluorescence microscopy, yeast cells were visualized on an Axioskop microscope (Carl Zeiss) equipped with a Plan-Apochromat 100×/1.40 NA oil DIC objective lens, a CCD camera (AxioCam MRm; Carl Zeiss), and a HBO100W/2 light source. Images were taken using AxioVision software with an auto exposure setup and processed using Adobe Photoshop CC software.

For time-lapse video recordings, yeast cells were viewed on an LSM 880 Airyscan NLO/FCS confocal microscope with an Alpha Plan-Apochromat 100×/1.46 NA oil objective lens. Excitation was performed with an argon laser at 488 nm, and emission was collected in the range of 493–556 nm for GFP imaging. Time-lapse images were acquired using Airyscan 32-channel ultra-sensitive area detector and the video was processed with ZEN software.

Yeast cells were fixed with 4% paraformaldehyde (PFA) and 0.2% glutaraldehyde in PBS for 30 min at 30°C followed by further fixation in 4% PFA for 1 h at 4°C. The fixed cells were rinsed once with PBS and resuspended in 10% gelatin. The solidified blocks were trimmed and placed in 2.3 M sucrose on a rotor overnight at 4°C, and then transferred to aluminum pins and frozen rapidly in liquid nitrogen. The frozen blocks were cut into 60 nm thick sections on a Leica Cryo-EM UC6 UltraCut using the Tokuyasu method (Tokuyasu, 1973). The cell sections were placed on carbon/Formvar-coated grids and floated in a dish of PBS for immunolabeling. Grids were placed section side down on drops of 0.1 M ammonium chloride to quench untreated aldehyde groups, then blocked for nonspecific binding on 1% fish skin gelatin in PBS. Single labeled grids were incubated with a primary rabbit anti-20S antibody (Enzo Life Sciences, catalog #BML-PW9355, lot #-05101719) at 1:200 dilution, and Protein A–gold beads (10 nm; Utrecht Medical Center) were used for secondary staining. The grids were rinsed in PBS, fixed with 1% glutaraldehyde for 5 min, rinsed again and stained with a mixture of 0.5% uranyl acetate to methylcellulose. Grids were viewed under a transmission electron microscope (FEI Tecnai G2 Spirit BioTWIN) at 80 kV. Images were taken using a SIS Morada 11-megapixel CCD camera and iTEM (Olympus) software. Acquired images were processed using Adobe Photoshop CC software.

Total proteins were extracted using the method of Kushnirov (2000), and western blotting was performed as described previously with minor modifications (Li et al., 2016). The equivalent of one optical density unit at 600 nm (OD₆₀₀) of cells were pelleted (6000 g for 1 min) and washed once with sterile water. Cells were incubated in 400 μl 0.1 M NaOH for 5 min at room temperature (RT) and were then pelleted (9400 g for 2 min) and resuspended in 100 μl SDS sample buffer (10% glycerol, 2% SDS, 0.1 M DTT, 62.5 mM Tris-HCl pH 6.8, 4% 2-mercaptoethanol, 0.008% Bromophenol Blue) and heated at 100°C for 5 min. Cell debris was removed by centrifugation (21,130 g for 1 min). Equal volumes of the supernatants were loaded onto 10% (v/v) SDS-PAGE gels, followed by the transfer of proteins to PVDF membranes (EMD Millipore, catalog #IPVH00010, lot #R0HB88840).

The membranes were incubated with the following primary antibodies: rabbit anti-Vps27 (Hatakeyama and De Virgilio, 2019) at 1:1000 dilution; JL-8 anti-GFP monoclonal antibody (TaKaRa, catalog #632381, lot #A8034133) at 1:2000 dilution; or an anti-Pgk1 monoclonal antibody (Abcam, catalog #ab113687, lot #GR3373682-5) at 1:10,000 dilution. Primary antibody binding was followed by anti-mouse-IgG (GE Healthcare, catalog #NXA931V, lot #17041890) or anti-rabbit-IgG (GE Healthcare, catalog #NA934V, lot #17136627) secondary antibody conjugated to horseradish peroxidase at 1:10,000 dilution. The membranes were incubated in ECL detection reagent (Mruk and Cheng, 2011), and the ECL signals were detected using film (Thomas Scientific, catalog #1141J52).

Cells equivalent to 35 OD₆₀₀ units were pelleted and washed once with ice-cold sterile water. Cells were resuspended in 450 μl lysis buffer (20 mM Tris pH 7.5, 0.5 mM EDTA pH 8, 200 mM NaCl, 10% glycerol, 1 mM PMSF, 10 mM NEM, Roche cOmplete protease inhibitor catalog #11836153001). Total cell lysates were prepared by beating with acid-washed glass beads using a FastPrep-24 device at 4°C for five cycles of 30 s agitation with 1 min breaks. After addition of 600 μl lysis buffer and Triton X-100 to 0.1% final concentration in cell lysates, cellular membranes were solubilized by incubating for 30 min on ice. Crude cell lysates were cleared at 16,000 g for 10 min at 4°C. Cleared cell lysates were incubated with 25 μl GFP-Trap agarose resin slurry (ChromoTek) for 1 h to immunoprecipitate proteasomes. The resin was washed three times with lysis buffer containing 0.02% Triton X-100 (except that NaCl concentration was increased to 250 mM in Fig. S4C) and resuspended in 50 μl 2× SDS sample buffer. Bound proteins were eluted by incubating the beads for 10 min at 42°C. The elutes were analyzed by western blotting as above with the following primary antibodies: polyclonal rabbit anti-ubiquitin (Dako, catalog #Z0458, lot #20016967) at 1:2000 dilution; monoclonal rabbit anti-ubiquitin-K63, clone Apu3 (EMD Millipore Corp, catalog #05-1308, lot #3137755) at 1:1000 dilution; or monoclonal anti-HA (Sigma, catalog #H9658, lot #089M4796V) at 1:2000 dilution.

Yeast cells were transformed with the plasmid pUB221, expressing 6xHis-myc-ubiquitin. Cells equivalent to 40 OD₆₀₀ units were pelleted and washed once with ice-cold sterile water. Cells were resuspended in 450 μl buffer A (6 M guanidine-HCl, 0.1 M Na₂HPO₄, 10 mM imidazole, pH 8). Total cell lysates were prepared by beating with acid-washed glass beads using a FastPrep-24 device at 4°C for six cycles of 30 s agitation with 1 min breaks. After addition of 600 μl buffer A, crude cell lysates were cleared at 1690 g for 15 min at 4°C. Cleared cell lysates were incubated with 80 μl Ni-NTA-agarose bead slurry (Qiagen, catalog #30230, lot #163032536) for 1.5 h to bind His-tagged ubiquitin and ubiquitylated proteins. The beads were washed three times with buffer A and another three times with a mixture of buffer A and TI buffer (25 mM Tris-HCl pH 6.8, 20 mM imidazole) at 1 to 3 volume ratio, followed by a final one time wash with TI buffer. The beads were resuspended in 50 μl 2× SDS sample buffer. Bound proteins were eluted by incubating the beads for 5 min at 100°C. The elutes were analyzed by western blotting as above with the following primary antibodies: monoclonal anti-Myc antibody (Sigma, catalog #M5546, clone 9E10) at 1:2000 dilution and the JL-8 anti-GFP monoclonal antibody as above.

Proteasomes were affinity purified from yeast cells as described previously (Li et al., 2015), with minor modifications. Briefly, cells were harvested after growth under indicated conditions and ground to fine powder with liquid nitrogen. About 6–7 ml of cell powders were thawed on ice, resuspended in 10 ml buffer A (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, 5 mM MgCl₂, 5 mM ATP, Roche cOmplete EDTA-free protease inhibitor catalog #1872580001), and incubated for 15 min on ice. Cell debris were pelleted at 30,000 g for 20 min at 4°C. Total protein concentrations of the supernatants were determined using a Pierce BCA protein assay kit (Thermo Scientific, catalog #23225, lot #SJ256254) according to the manufacturer's protocol. Supernatant equivalent to ∼100 mg total protein was incubated with 200 µl (packed) resin of anti-FLAG M2 affinity gel (Sigma, catalog #A2220, lot #SLCH0130) for 2 h on a rotator at 4°C. The proteasome-bound resin was washed twice with 12 ml buffer A for 10 min, and then incubated with 3 resin volumes of 200 µg ml⁻¹ 3×FLAG peptide (Sigma, catalog #F4799, lot #SLCJ4916) for 45 min to elute proteasome complexes. Proteasomes were concentrated with 100 K MWCO centrifugal filters (Merk Millipore, catalog #UFC510024, lot #R9HA55989) and quantified with a BSA standard using a G:Box Chemi HR16 imager (Syngene).

Yeast cell extracts were prepared as described previously with minor modifications (Tomko and Hochstrasser, 2014). Low glucose starved yeast cells were washed twice with ice-cold sterile water and frozen in liquid nitrogen. The frozen cells were ground with mortar and pestle, and the resulting cell powder was thawed in extraction buffer (50 mM Tris-HCl, pH 7.5, 5 mM MgCl₂, 10% glycerol, 5 mM ATP). Extracts were centrifuged for 10 min at 21,000 g to remove cell debris. Protein concentrations were determined using the BCA assay. Equal amounts of total protein (50 µg) were loaded onto 4% native PAGE gels and resolved for 3 h at 100 V at 4°C. Native PAGE-separated proteins were then transferred to PVDF membranes and subjected to western blotting analysis as above with primary rabbit polyclonal antibodies against Pre6 (Jäger et al., 2001) at 1:5000 dilution and ECL with anti-rabbit IgG secondary antibody conjugated to horseradish peroxidase (GE Healthcare, catalog #NA934V, lot #16889289) at the same dilution.

Protein samples were prepared as described above in the immunoprecipitation section and resolved on 10% SDS-PAGE gels. Gel bands were excised and fixed with two incubations in 45% methanol and 10% acetic acid for 30 min at RT, and rinsed once with MilliQ water before the in-gel digestion procedure.

Gel bands were cut into small pieces and rinsed with 800 µl water on a tilt-table for 10 min and then washed for 20 min with 800 µl 50% acetonitrile (ACN) in 100 mM NH₄HCO₃ (ammonium bicarbonate, ABC). The samples were reduced by incubating with 200 µl 4.5 mM DTT in 100 mM ABC for 20 min at 37°C, then alkylated by incubating with 200 µl 10 mM iodoacetamide in 100 mM ABC for 20 min at RT in the dark. The gels were washed for 20 min with 800 µl 50% ACN in 100 mM ABC, then washed for 20 min with 800 µl 50% ACN in 25 mM ABC. The gel pieces were briefly dried by SpeedVac, then resuspended in 200 µl of 25 mM ABC containing 500 ng of sequencing grade trypsin (Promega, catalog #V5111) and incubated for 16 h at 37°C. Peptides in the supernatant were moved to a new Eppendorf tube, and residual peptides were extracted from the gel by the addition of 500 µl 80% ACN and 0.1% trifluoroacetic acid (TFA) and combined with the first supernatant. Peptides were dried in a SpeedVac and dissolved in 25 µl MS loading buffer (2% ACN, 1% TFA), with 5 µl injected for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis.

LC-MS/MS analysis was performed on a Thermo Scientific Q Exactive Plus equipped with a Waters nanoAcquity UPLC system utilizing a binary solvent system (A, 100% water, 0.1% formic acid; B, 100% ACN, 0.1% formic acid). Trapping was performed at 5 µl min⁻¹, 99.5% buffer A for 3 min using a Waters ACQUITY UPLC M-Class Symmetry C18 Trap Column (100 Å, 5 µm, 180 µm×20 mm, 2G, V/M). Peptides were separated at 37°C using a Waters ACQUITY UPLC M-Class Peptide BEH C18 Column (130 Å, 1.7 µm, 75 µm×250 mm) and eluted at 300 nl min⁻¹ with the following gradient: 3% buffer B at initial conditions; 5% B at 1 min; 25% B at 90 min; 50% B at 110 min; 90% B at 115 min; 90% B at 120 min; return to initial conditions at 125 min. Mass spectra was acquired in profile mode over the 300–1700 m/z range using 1 microscan, 70,000 resolution, AGC target of 3E6, and a maximum injection time of 45 ms. Data-dependent MS/MS were acquired in centroid mode on the top 20 precursors per MS scan using 1 microscan, 17,500 resolution, AGC target of 1E5, maximum injection time of 100 ms, and an isolation window of 1.7 m/z. Precursors were fragmented by HCD activation with a collision energy of 28%. MS/MS were collected on species with an intensity threshold of 10⁴, charge states 2–6, and peptide match preferred. Dynamic exclusion was set to 20 s.

Tandem mass spectra were extracted by Proteome Discoverer software (version 2.2.0.388, Thermo Scientific) and searched in-house using the Mascot algorithm (version 2.7.0, Matrix Science). The data were searched against the Swissprotein database with taxonomy restricted to Saccharomyces cerevisiae (7905 sequences). Search parameters included trypsin digestion with up to 2 missed cleavages, peptide mass tolerance of 10 ppm, and MS/MS fragment tolerance of 0.02 Da. Cysteine carbamidomethylation was configured as a fixed modification. Methionine oxidation; phosphorylation of serine, threonine and tyrosine; propionamide adduct to cysteine; and GG dipeptide (ubiquitin residue) on lysine were configured as variable modifications. Normal and decoy database searches were run, with the confidence level set to 95% (P<0.05). Scaffold Q+S (version 4.11.0, Proteome Software Inc.) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability by the Scaffold Local FDR algorithm. Protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least 2 identified peptides. Gene component term analysis with P<0.01 was conducted using the Generic Gene Ontology Term Finder online software (Princeton University).

GraphPad Prism 9 was used to generate the heat map in Fig. S6B and most bar graphs. Microsoft Excel 2013 was used to perform ANOVA single factor analysis and generate bar graphs in Fig. 6D–F. ImageJ was used to quantify protein band intensities in Fig. 5B,C,E. The number of cells counted for image quantification are described in the figure legends. Each experiment was repeated at least three times and the percentages shown in the figures represent the average of all the experiments. Error bars represent standard deviations.

We thank Carolyn Breckel and Mengwen Zhang from our lab and Xiaofeng Wang at Virginia Tech for critical reading of the manuscript. We are grateful to Masahide Oku and Yasuyoshi Sakai at Kyoto University, and Riko Hatakeyama and Claudio De Virgilio at University of Fribourg for sharing antibodies, yeast strains, or plasmids; Morven Graham for electron microscopy assistance at the Yale University Center for Cellular and Molecular Imaging; and Jean Kanyo for LC-MS/MS assistance at the Yale University Keck MS & Proteomics Resource. We thank the MS & Proteomics Resource of the W.M. Keck Foundation Biotechnology Resource Laboratory at Yale University for providing the necessary mass spectrometers and the accompanying biotechnology tools, funded in part by the Yale School of Medicine and by the Office of The Director, National Institutes of Health (S10OD02365101A1, S10OD019967 and S10OD018034).

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259393.