Target of rapamycin complex-2 (TORC2), a conserved protein kinase complex, is an indispensable regulator of plasma membrane homeostasis. In budding yeast (Saccharomyces cerevisiae), the essential downstream effector of TORC2 is protein kinase Ypk1 and its paralog Ypk2. Muk1, a Rab5-specific guanine nucleotide exchange factor (GEF), was identified in our prior global screen for candidate Ypk1 targets. We confirm here that Muk1 is a substrate of Ypk1 and demonstrate that Ypk1-mediated phosphorylation stimulates Muk1 function in vivo. Strikingly, yeast lacking its two Rab5 GEFs (Muk1 and Vps9) or its three Rab5 paralogs (Vps21/Ypt51, Ypt52, and Ypt53) or overexpressing Msb3, a Rab5-directed GTPase-activating protein, all exhibit pronounced reduction in TORC2-mediated phosphorylation and activation of Ypk1. Vps21 coimmunoprecipitates with TORC2, and immuno-enriched TORC2 is less active in vitro in the absence of Rab5 GTPases. Thus, TORC2-dependent and Ypk1-mediated activation of Muk1 provides a control circuit for positive (self-reinforcing) up-regulation to sustain TORC2-Ypk1 signaling.
Studies in Saccharomyces cerevisiae (Loewith et al., 2002; Wedaman et al., 2003) were the first to show the presence of two evolutionarily conserved multicomponent protein kinase complexes—Target of rapamycin complex-1 (TORC1) and TORC2—in which a TOR polypeptide is the catalytic subunit, as now found in all eukaryotes (Eltschinger and Loewith, 2016; González and Hall, 2017; Saxton and Sabatini, 2017; Tatebe and Shiozaki, 2017). Whereas the role of TORC1 in both yeast and animal cells is in sensing at the lysosome/vacuole whether supplies of nutrients are sufficient to support protein synthesis and other anabolic processes required for cell growth (Jewell and Guan, 2013; Kim et al., 2013; Lim and Zoncu, 2016; Powis and De Virgilio, 2016), TORC2, located at the cell cortex in both yeast and animal cells, promotes the biosynthesis of glycerolipids and sphingolipids and controls other pathways required for plasma membrane (PM) biogenesis and homeostasis (Gaubitz et al., 2016; Guri et al., 2017; Roelants et al., 2017a, 2018).
The fact that basal signaling emanating from TORC2 is essential for cell viability also was first shown in yeast (Kunz et al., 1993; Helliwell et al., 1994). Subsequent work has demonstrated that TORC2 activity is responsive to various stresses and insults that can perturb the integrity of the cell envelope. Certain challenges (sphingolipid depletion, hypotonic conditions, heat shock, and elevated exogenous acetic acid) markedly stimulate TORC2 function (Roelants et al., 2011; Berchtold et al., 2012; Sun et al., 2012; Guerreiro et al., 2016), whereas others (hypertonic conditions and cell wall damage) markedly reduce TORC2 activity (Lee et al., 2012; Muir et al., 2015; Leskoske et al., 2018). The primary downstream effector of yeast TORC2 is the AGC family protein kinase Ypk1 (mammalian orthologue is SGK1; Casamayor et al., 1999) and its paralog Ypk2 because it has been shown that constitutively active alleles of either Ypk1 or Ypk2 rescue the inviability of tor2 mutations and other mutations (avo3) that result in deficient TORC2 function (Kamada et al., 2005; Aronova et al., 2008; Roelants et al., 2011; Leskoske et al., 2017). TORC2 activates Ypk1 function by phosphorylating multiple sites within its C-terminal regulatory tail (Leskoske et al., 2017). To understand what physiological processes are under TORC2-Ypk1 control, we conducted both genome-wide genetic (Roelants et al., 2002) and global biochemical (Muir et al., 2014) screens to identify candidate substrates and regulators of TORC2-Ypk1 signaling. Validation and characterization of many of these candidates has shed considerable light on those sectors of cellular physiology that are regulated by the TORC2-Ypk1 axis (Roelants et al., 2010, 2011, 2017b, 2018; Lee et al., 2012; Muir et al., 2014, 2015; Alvaro et al., 2016).
Among the presumptive, but as yet uncharacterized, targets of Ypk1 that we identified is Muk1 (Muir et al., 2014). Muk1 and its paralog Vps9 are guanine nucleotide exchange factors (GEFs) that catalyze GDP-GTP exchange in the Rab5 class of GTPases in S. cerevisiae, namely, Vps21/Ypt51 and its paralogs Ypt52 and Ypt53 (Burd et al., 1996; Hama et al., 1999; Paulsel et al., 2013; Bean et al., 2015). As in other eukaryotes, the function of Rab5 GTPases was thought to be confined to promoting vesicle formation, trafficking, and sorting in the early stages of the endocytic pathway (Singer-Krüger et al., 1994, 1995; Cabrera and Ungermann, 2013), and disruption of Rab5 function in yeast leads to accumulation of cargo in endosomes and disruption of their delivery to the vacuole (Horazdovsky et al., 1994; Gerrard et al., 2000). Muk1 and Vps9 have overlapping roles as Rab5 GEFs because a muk1Δ vps9Δ double mutant has a more severe phenotype than either single mutant and, tellingly, exhibits a phenotype closely resembling a vps21Δ ypt52Δ ypt53Δ triple mutant (Cabrera and Ungermann, 2013; Paulsel et al., 2013). Although Muk1 and Vps9 both act on Rab5 GTPases, the presence of two distinct proteins suggests that they may be differentially regulated.
In this study, we sought, first, to determine whether Muk1 is indeed an authentic and physiologically relevant target of Ypk1 and, if so, the consequences of its Ypk1-mediated phosphorylation. In the process, and as documented here, we discovered quite unexpectedly that Rab5 function is necessary to support maximal TORC2 function, revealing a previously unappreciated new connection between the vesicle trafficking machinery and the control of PM homeostasis by TORC2-Ypk1 signaling.
Muk1 is a substrate of protein kinase Ypk1
An ∼260-residue segment of Vps9 (451 residues) is necessary and sufficient for its Rab5 GEF activity (Carney et al., 2006; Barr and Lambright, 2010; Bean et al., 2015). Compared with Vps9 itself (Gough et al., 2001), the catalytic (Vps9 homology) domain of Muk1 (612 residues) is split by an 83-residue insert that contains, in tandem, two matches to the consensus phospho-acceptor site motif of Ypk1 (RxRxxS; 168RSRSSSG174 and 179RPRRSSS185; Mok et al., 2010; Muir et al., 2014; Fig. 1 A). These same sites are completely conserved among S. cerevisiae and all its sensu stricto relatives, as well as in more divergent yeast species (e.g., Candida glabrata, Kluyveromyces lactis, and Ashbya gossypii; see https://portals.broadinstitute.org/cgi-bin/regev/orthogroups/show_orthogroup.cgi?orf=YPL070W), and phosphorylation at these same sites in vivo has been detected in several global phosphoproteomic analyses of S. cerevisiae (Albuquerque et al., 2008; Holt et al., 2009; Swaney et al., 2013).
We confirmed that Muk1 is a phosphoprotein, first by showing that an epitope-tagged derivative (Muk1-myc) migrates even in standard SDS-PAGE as a series of multiple, slower-mobility isoforms that are collapsed to a faster-mobility species by treatment with calf intestinal phosphatase (CIP; Fig. 1 B). To determine whether any isoforms were attributable to phosphorylation by Ypk1, we expressed Muk1-myc in either WT or otherwise isogenic cells lacking Ypk2 and carrying an allele of Ypk1, Ypk1(L424A) (hereafter Ypk1-as), sensitive to acute inhibition by the adenine analogue 1-(tert-butyl)-3-(3-methylbenzyl)-1H-pyrazolo[3,4-d]pyrimidin-4-amine (3MB-PP1; Lopez et al., 2014; Muir et al., 2014). We found that only in the analogue-sensitive cells and only in the presence of inhibitor were the slowest-mobility species greatly reduced and the most hypophosphorylated species markedly increased (Fig. 1 C), showing that Muk1 is phosphorylated in a Ypk1-dependent manner in vivo. We have shown previously that in other Ypk1 substrates containing Ser residues juxtaposed (immediately upstream, immediately downstream, or both) to the primary phospho-acceptor Ser, such residues can also be phosphorylated by Ypk1 both in vivo and in vitro (Roelants et al., 2011; Muir et al., 2014). Hence, to test whether Ypk1-mediated phosphorylation of Muk1 was occurring at the expected sites, we examined the mobility pattern of WT Muk1 tagged with a 3xFLAG epitope or a derivative in which all six potential Ser residues at the Ypk1 consensus motifs were mutated to Ala. For this purpose, we used Phos-tag SDS-PAGE, a technique that enhances retardation of a protein depending on the extent of its phosphorylation (Kinoshita et al., 2009). We found that lack of the six Ser residues resulted in a marked reduction in the slowest-mobility isoforms and a pronounced increase in the fastest-mobility (hypophosphorylated) isoform (Fig. S1), indicating that these sites are phosphorylated in vivo.
To more readily visualize Ypk1-specific phosphorylation of Muk1, we expressed a smaller fragment (residues 1–220) of WT Muk1-myc or an otherwise identical fragment in which the six potential Ser residues at the Ypk1 consensus motifs were mutated to Ala. We also took advantage of the fact that TORC2-dependent activation of Ypk1 is stimulated in response to inhibition of sphingolipid biosynthesis by antibiotics such as myriocin (Myr; which inactivates l-Ser:palmitoyl-CoA acyltransferase; Miyake et al., 1995) or aureobasidin A (AbA; which inactivates inositolphosphate-ceramide synthase; Sugimoto et al., 2004), as we and others have shown (Roelants et al., 2011; Berchtold et al., 2012; Khakhina et al., 2015). Indeed, when cells were treated with AbA, we found that the Muk1(1–220) fragment displayed a readily detectable slower-mobility isoform that was eliminated by the 6A mutations or by absence of just Ypk1 itself (Fig. 1 D), because under normal growth conditions very little Ypk2 is expressed (Gasch et al., 2000; Roelants et al., 2002). Moreover, as expected, this species appeared when cells were treated with AbA, but not in its absence (Fig. S1 B). Finally, purified recombinant Muk1, but not Muk1(6A), was robustly phosphorylated in vitro by purified Ypk1-as in the absence of 3-MB-PP1, but not in its presence (Fig. 1 E). Thus, Muk1 is a bona fide Ypk1 substrate.
Muk1 function is stimulated by Ypk1-mediated phosphorylation
The three yeast Rab5 isoforms (Vps21/Ypt51, Ypt52, and Ypt53) are involved in the early stages of the endocytic pathway, which is responsible for clearing integral PM membrane proteins, including amino acid permeases (Singer-Krüger et al., 1994; Cabrera and Ungermann, 2013; Paulsel et al., 2013). Hence, defects in that process will increase the steady-state level of such permeases, enhancing uptake of cognate amino acid analogues, thus conferring on the cell greater sensitivity to the toxic effect of such compounds (Grenson et al., 1966; Lin et al., 2008; Fig. 2 A). Indeed, we found that compared with otherwise isogenic WT cells, cells lacking both Rab5 GEFs or all three Rab5 isoforms were markedly more sensitive to the growth-inhibitory action of canavanine, an Arg analogue taken up by the Arg permease (Can1), whereas cells lacking Ypt7, the sole yeast orthologue of Rab7, which acts later in the endocytic pathway (Schimmöller and Riezman, 1993; Rink et al., 2005), were not (Fig. 2 B), indicating that the observed defect was specific for Rab5.
This phenotype provided a means to assess the effect of Ypk1-mediated phosphorylation on Muk1 function. Indeed, we found that a muk1Δ vps9Δ double mutant conferred a greater degree of canavanine sensitivity than a vps9Δ mutation alone and that reexpression of Muk1 from its endogenous promoter on a CEN plasmid reproducibly restored readily detectable canavanine resistance to the muk1Δ vps9Δ cells (Fig. 2 C). Moreover, a site-directed Muk1 mutant (D353A) that substituted a residue in its Vps9 homology domain known to be essential for catalytic function (Bean et al., 2015) was unable to confer canavanine resistance to the muk1Δ vps9Δ cells, demonstrating that the GEF function of Muk1 is required (Fig. 2 C). Most tellingly, Muk1(6A), which cannot be phosphorylated on any of its Ypk1 sites, was also unable to confer canavanine resistance to muk1Δ vps9Δ cells, whereas the phosphomimetic mutant Muk1(6E), which simulates full Ypk1-dependent phosphorylation at these same sites, was at least as potent as WT Muk1 at restoring canavanine resistance to the muk1Δ vps9Δ cells (Fig. 2 C). We found exactly the same trends when we examined the sensitivity of the same set of strains to the growth-inhibitory action of ethionine, a Met analogue taken up by the Met permease (Mup1; Isnard et al., 1996; Guiney et al., 2016; Fig. 2 D). In addition, we found exactly the same trends monitoring rescue of an independent phenotype—namely, the poor growth of muk1Δ vps9Δ cells at an elevated temperature (37°C; Paulsel et al., 2013; Shideler et al., 2015; Fig. S1 C). Furthermore, we confirmed that the difference in behavior between Muk1(6A) and Muk1(6E) could not be attributed to any difference in their relative level of expression, compared with each other or WT Muk1 (Fig. 2 E). Taken together, these findings indicated that Ypk1-mediated phosphorylation of Muk1 is required for the function of this Rab5 GEF.
Rab5 GTPases are necessary for efficient TORC2-mediated Ypk1 activation
Based on the precedent that TORC1 function requires its interaction with both the Rheb (Avruch et al., 2009; Yang et al., 2017) and Rag (Sancak and Sabatini, 2009; Nicastro et al., 2017) classes of small GTPases, and on the fact that Ryh1, a Schizosaccharomyces pombe GTPase that most closely resembles human Rab6 and its S. cerevisiae orthologue (Ypt6), was reported to stimulate TORC2 in fission yeast (Tatebe et al., 2010), we examined whether Rab5 function influenced TORC2 activity by monitoring its phosphorylation and activation of Ypk1.
For that purpose, we used, first, a version of Ypk1 (Ypk17A) that contains only TORC2-dependent phosphorylation sites, which when examined by Phos-tag SDS-PAGE allows for convenient and reliable detection of the isoforms produced by its TORC2-mediated phosphorylation, as we have extensively documented elsewhere (Muir et al., 2015; Leskoske et al., 2017). Hence, we analyzed the phosphorylation pattern of the Ypk17A reporter in cells before and after treatment with AbA. TORC2-dependent phosphorylation of Ypk17A could be readily visualized in WT cells as a series of isoforms well resolved by Phos-tag SDS-PAGE, and inhibition of complex sphingolipid biosynthesis with AbA stimulated the appearance of the slowest-mobility (most highly phosphorylated) isoform (Fig. 3 A, left), in accord with our prior observations with Myr (Muir et al., 2015; Leskoske et al., 2017). Strikingly, however, in the absence of both Rab5 GEFs, basal and AbA-induced TORC2 phosphorylation of Ypk17A were both markedly reduced, and loss of Vps9 (the major Rab5 GEF) alone had a stronger effect than loss of Muk1 (Fig. 3 A). Likewise, compared with WT control cells, basal and AbA-induced TORC2 phosphorylation of Ypk17A was dramatically decreased in cells lacking all three Rab5 paralogs (Fig. 3 B). Thus, in cells that cannot generate active Rab5, TORC2-mediated phosphorylation of Ypk1 was very inefficient.
We and others have demonstrated previously that phosphorylation by TORC2-activated Ypk1 of multiple targets up-regulates flux through the sphingolipid biosynthetic pathway and is required for the cell to survive inhibition of ceramide synthesis caused by the presence of Myr (Roelants et al., 2011; Berchtold et al., 2012; Muir et al., 2014; Leskoske et al., 2017). Similarly, as expected, we also found that Ypk1 function is required for the cell to survive the challenge of the depletion of complex sphingolipids caused by AbA treatment (Fig. 3 C). Thus, the degree of resistance that cells display to agents like Myr and AbA provides a convenient phenotypic readout for the efficiency of Ypk1 activation by TORC2. By this criterion and in agreement with our biochemical analysis (Fig. 3 A and B), cells lacking both Rab5 GEFs and cells lacking all three Rab5 paralogs were much more sensitive to the growth-inhibitory action of AbA than otherwise isogenic WT control cells (Fig. 3 D), whereas cells devoid of Ypt7 (as an additional control for specificity) were not. Thus, cells that cannot generate active Rab5 exhibit physiological behavior consistent with inefficient TORC2-mediated phosphorylation of Ypk1. As a further test of this conclusion, we examined cells overexpressing Msb3, one of two apparent Rab5-specific GAPs encoded in the S. cerevisiae genome. Although Msb3 has a paralog (Msb4), only Msb3 had demonstrable Rab5 GAP activity in vivo and in vitro in two independent studies (Lachmann et al., 2012; Nickerson et al., 2012). We reasoned, therefore, that overexpressed Msb3 should lower the steady-state pool of activated (GTP-bound) Rab5 in the cell. Indeed, we found that such cells grew significantly less robustly on AbA than the control cells harboring the empty vector (Fig. 3 E). Collectively, these data indicate that Rab5 in its GTP-bound state is necessary to support TORC2 phosphorylation and activation of Ypk1.
Active Rab5 stimulates TORC2 phosphorylation of Ypk1
As a further test of the conclusion that activated Rab5 promotes TORC2 phosphorylation of Ypk1, we examined the effect of reexpressing in the Rab5-deficient vps21Δ ypt52Δ ypt53Δ triple mutant WT Vps21, Vps21(Q66L) (a GTP hydrolysis-defective mutant [so-called “GTP-locked” variant]), or Vps21(S21L) (a mutant that preferentially binds GDP [so-called “GDP-locked” variant]; Tall et al., 1999; Lo et al., 2011; Plemel et al., 2011). We focused on Vps21 because it is constitutively expressed and the most abundant yeast Rab5 isoform (Schmidt et al., 2017), has been shown to play the major role in endocytic vesicle trafficking (Horazdovsky et al., 1994), and when absent, causes more pronounced phenotypes compared with loss of the other two Rab5 GTPases (Singer-Krüger et al., 1994; Nickerson et al., 2012; Nakatsukasa et al., 2014). We found that, when present as the sole Rab5, WT Vps21 was able to rescue AbA-induced TORC2-phosphorylation of Ypk17A (Fig. 4). This rescue was reproducibly more efficacious in cells expressing GTP-locked Vps21 but undetectable in cells expressing GDP-locked Vps21 (Fig. 4), consistent with the need for a Rab5 GTPase to be in its GTP-bound state to promote TORC2 phosphorylation of Ypk1. The weaker rescue conferred by WT Vps21 in the vps21Δ ypt52Δ ypt53Δ mutant cells may be attributable to its reduced expression relative to GTP-locked Vps21. Alternatively, the high level of GAL promoter-driven expression could overwhelm the capacity of the Rab5 GEFs, thereby lowering the fraction of total WT Vps21 in the GTP-bound state. By contrast, once loaded, GTP-locked Vps21 persists in its active state. In any event, because GTP-locked Vps21 cannot cycle and is defective in endocytic trafficking (Tall et al., 1999; Lachmann et al., 2012), its ability to stimulate TORC2 phosphorylation of Ypk1 appears to be a direct effect of Vps21-GTP and not some secondary consequence of its roles in vesicle trafficking per se. Nonetheless, we explored whether the reduction in TORC2 phosphorylation of Ypk1 in cells deficient in GTP-Rab5 might be attributable to such indirect effects.
Lack of GTP-Rab5 does not cause mislocalization of either TORC2 or Ypk1
Deletion of the C-terminal PH domain in Avo1, an essential subunit of TORC2, reportedly rendered cells inviable, but they could be rescued by expression of an Avo1 construct in which a CaaX box was substituted for its PH domain (Berchtold and Walther, 2009), suggesting that association with the PM is crucial for TORC2 function. Given that Rab5 GTPases are important in endocytic vesicle internalization, and also have been implicated in recycling of endosomes back to the PM (Jovic et al., 2010; Feyder et al., 2015; MacDonald and Piper, 2016), it seemed possible that cells deficient in GTP-bound Rab5 might have decreased TORC2 activity simply owing to altered TORC2 localization. However, when we analyzed the subcellular distribution of TORC2, monitored by fluorescence microscopy in cells marked with Avo3-GFP, another essential TORC2 subunit very tightly bound to the catalytic subunit Tor2 (Wullschleger et al., 2005; Karuppasamy et al., 2017; see also Fig. S4 A), we saw no difference in PM localization of TORC2 between WT and muk1Δ vps9Δ mutant cells (Fig. S2 A). Thus, TORC2 does not appear to be mislocalized in the absence of active Rab5.
Alternatively, it seemed possible that the decrease in TORC2 phosphorylation of Ypk1 observed in cells deficient in GTP-bound Rab5 might be due to diminished accessibility of Ypk1 at the PM. For its proper folding and function, Ypk1 is obligatorily phosphorylated on T504 in its activation loop by Pkh1 (and Pkh2; Roelants et al., 2002, 2004), which are protein kinases very tightly bound to PM structures called eisosomes (Walther et al., 2007; Douglas and Konopka, 2014). However, we saw no difference in Ypk1 activation loop phosphorylation between WT and muk1Δ vps9Δ mutant cells (Fig. S2 B), suggesting that there is no impairment in Ypk1 availability at the PM. Similarly, substantial data support the conclusion that interaction with the so-called CRIM element in the Avo1 subunit of TORC2 is responsible for substrate presentation to the active site in the Tor2 catalytic subunit of TORC2 (Liao and Chen, 2012; Tatebe et al., 2017; Yao et al., 2017), and we saw no significant difference in Ypk1 association with Avo1 between WT and muk1Δ vps9Δ mutant cells in GST pull-down experiments (Fig. S2 C). Thus, lack of efficient phosphorylation by TORC2 in cells deficient in GTP-Rab5 cannot be explained by the inability of Ypk1 to encounter TORC2 at the PM. As a further test, we expressed PHSlm1-Ypk1-3xHA, which others have demonstrated is efficiently targeted to the PM via its PtdIns4,5P2-binding PH domain and serves as an effective substrate of TORC2 (Niles et al., 2012). We found, however, that TORC2-dependent phosphorylation of the C-terminal regulatory tail in PHSlm1-Ypk1-3xHA was still poor in cells deficient in GTP-Rab5 (Fig. S2 D). Therefore, we explored whether there might be a more direct, physical interaction between Vps21 and TORC2, or between Vps21 and Ypk1.
Vps21 associates with Tor2 in TORC2
As one means to test whether Vps21 and TORC2 are able to physically associate, we examined whether they were able to coimmunoprecipitate. For this purpose, we first expressed either FLAG-Vps21 or FLAG-mNeonGreen (mNG; Shaner et al., 2013) in cells coexpressing a 3xHA epitope-tagged version of the Tor2 catalytic subunit of TORC2, either from a CEN plasmid in tor1Δ tor2Δ cells (Fig. S3 A) or integrated at the TOR2 locus on chromosome XI (Fig. S3 B). We chose mNG as an initial control because mNG (237 residues) is quite comparable in size to Vps21 (210 residues), as well as quite similar in overall surface charge (pI = 5.7 for mNG versus pI = 5.0 for Vps21]. In multiple trials, when extracts of the cells were immunoprecipitated with anti-FLAG antibodies, we reproducibly found readily detectable coimmunoprecipitation of 3xHA-tagged Tor2 from the cells expressing FLAG-Vps21, but not from the cells expressing FLAG-mNG (Fig. S3, A and B). We conducted similar immunoprecipitation experiments using cells expressing either FLAG-Vps21 or FLAG-mNG and coexpressing Ypk1-myc, because phosphorylation and activation of certain other AGC family protein kinases requires association of their regulatory domains with other classes of small GTPases (Pearce et al., 2010; Leroux et al., 2018). However, only an equivalent level of nonspecific background interaction was observed for both FLAG-Vps21 and FLAG-mNG with Ypk1-myc (Fig. S3 C). Thus, these initial experiments indicated that Vps21 associates with TORC2, not Ypk1.
As a critical control for specificity, we repeated these same experiments using both FLAG-Vps21 and FLAG-Ypt7, and found again that only Vps21 exhibited a readily detectable association with Tor2 (Fig. 5 A), indicating that the observed interaction does not occur with an unrelated small GTPase. Because both Vps21 and Ypt7 are membrane-anchored proteins, this specificity made it highly unlikely that Vps21–TORC2 interaction resulted from the fact that both Vps21 and TORC2 are membrane associated. To rule out this possibility, we directly documented that the level of detergent present in the cell lysis and immunoprecipitation buffers, as described in Materials and methods, was sufficient to fully solubilize both proteins (Fig. 5 B). Finally, we found that both WT and GTP-locked Vps21 were efficacious in pulling down Tor2, but we also observed detectable association with GDP-locked Vps21 (Fig. S3 D), similar to prior reports showing that both GTP-locked Rheb and GDP-locked Rheb bound to mTORC1 (Long et al., 2005). Collectively, these results demonstrate that Vps21 makes specific contact with TORC2.
Rab5 GTPase stimulates TORC2 activity
Given the interaction between Vps21 and Tor2 that we detected by coimmunoprecipitation, we explored the possibility that this association, whether stable or transient, is able to stimulate the catalytic function of TORC2, analogous to the stimulation of mTORC1 activity by Rheb in mammalian cells (Avruch et al., 2009; Yang et al., 2017). In vegetatively growing and unstressed cells, Ypt53 expression is repressed (Schmidt et al., 2017). Therefore, to obtain TORC2 from cells depleted of Rab5 GTPases, we used immuno-isolation with anti-FLAG antibodies from extracts of vps21Δ ypt52Δ mutant cells expressing Avo3-3xFLAG, which we were able to demonstrate also successfully enriched for the Tor2 catalytic subunit of TORC2, which was marked with a 3xHA tag (Fig. S4 A). We used the identical approach to prepare TORC2 from otherwise isogenic WT cells (Fig. S4 A). Regardless of the strain background, our procedure using FLAG-tagged Avo3 for immuno-isolation recovered TORC2 complexes that contain equivalent amounts of HA-tagged Tor2. As the substrate, we used a purified recombinant protein in which the 78 C-terminal residues of Ypk1 (680 residues), which contain all of its TORC2 phosphorylation sites (Leskoske et al., 2017), was fused to the C terminus of maltose-binding protein (MBP; 370 residues; Duplay et al., 1984), hereafter referred to as MBP-Ypk1CT.
We first incubated TORC2 from WT cells with MBP-Ypk1CT and [γ-32P]ATP and found robust incorporation, as expected. We found that this incorporation was almost completely eliminated when we used the mutant substrate MBP-Ypk1CT(6A), in which all six of the TORC2-specific phospho-acceptor residues were mutated to Ala, confirming that the observed phosphorylation was occurring at the previously defined TORC2-specific sites in Ypk1 (Fig. S4 B). TORC2 itself, and not some other protein kinase present in the preparation, was responsible for the observed phosphorylation because incubation with a compound (QL-IX-55) specifically designed and demonstrated by others to selectively inhibit yeast TORC2 (Liu et al., 2012) markedly reduced the incorporation into MBP-Ypk1CT (Fig. S4 C). Likewise, when we prepared TORC2 in the same way from cells expressing a Tor2 allele (Tor2(L2178A)) designed to be specifically sensitive to an orthogonal inhibitor (NVP-BEZ235; Kliegman et al., 2013), incorporation into MBP-Ypk1CT was almost completely eliminated (Fig. S4 D). These controls established that the observed phosphorylation of MBP-Ypk1CT was occurring in a strictly TORC2-dependent manner. We further established that, under our assay conditions, incorporation was roughly linear with time of incubation (Fig. S4, E and F).
Having thoroughly validated the specificity of this in vitro reaction, we then compared TORC2 isolated from WT cells to TORC2 isolated from the vps21Δ ypt52Δ cells and found, in multiple trials, that the specific activity of the WT TORC2 was reproducibly higher (Fig. 6, A and B). Moreover, addition of GTP-loaded Vps21 to the TORC2 isolated from vps21Δ ypt52Δ mutant cells reproducibly stimulated incorporation into MBP-Ypk1CT (Fig. 6 C), whereas under the exact same assay conditions, GTP-loaded Ypt7 did not (Fig. 6 D). The most parsimonious model to explain these in vitro results is that GTP-bound Rab5 associates with Tor2 and causes a conformational change in TORC2 that allows it to access the C-terminal end of Ypk1 more efficiently, fully consistent with all of the in vivo evidence presented in this study that active (GTP-bound) Rab5 GTPases promote TORC2 activity toward its substrate Ypk1 (Fig. 7).
This study was initiated to determine whether we could extend our understanding of TORC2-Ypk1 signaling by establishing whether a candidate substrate for Ypk1 that emerged in a global screen (Muir et al., 2014) was authentic and physiologically relevant. As we have documented here, the Rab5-specific GEF Muk1 is indeed a target of Ypk1. Even though Muk1 is clearly phosphorylated by other protein kinases, which may influence other aspects of Muk1 function, our genetic evidence demonstrates that Ypk1-mediated phosphorylation is required for the GEF activity of Muk1 in vivo. S. cerevisiae contains just two functional Rab5-specific GEFs that possess the Vps9 homology domain, Vps9 itself and Muk1 (Paulsel et al., 2013; Bean et al., 2015). The human genome encodes nine proteins with readily discernible Vps9 homology domains (Carney et al., 2006; Barr and Lambright, 2010); however, none of them is the obvious orthologue of either Vps9 or Muk1. It has been reported that SGK1, the mammalian homologue of Ypk1 (Casamayor et al., 1999), facilitates PM recycling of KCNQ1 channels in a Rab5-dependent manner, but whether that behavior was dependent on SGK1-mediated phosphorylation of any Rab5 GEF was not explored (Seebohm et al., 2008). Likewise, it has been reported that SGK1 function promotes association of a Rho- and Rac-specific GEF (ARHGEF2/GEF-H1) with a component (Sec5/EXOC2) of the mammalian exocyst complex, but again whether that behavior was dependent on SGK1-mediated phosphorylation of the GEF was not determined (Wang et al., 2015).
Given TORC2-Ypk1–mediated activation of a Rab5-specific GEF, we then explored what the role of that pathway might be. It is now well established that TORC1 localization and activity are influenced by small GTPases of the Rag and Rheb classes, respectively. The Rag GTPases tether TORC1 at the surface of the lysosome/vacuole through interaction with the RAPTOR (yeast counterpart Kog1/Las24) subunit (Sancak and Sabatini, 2009; Nicastro et al., 2017). By contrast, recent structural work (Yang et al., 2017) has shown that TORC2 activation by Rheb (Avruch et al., 2009) is mediated by its direct interaction with a pocket constituted by the M- and N-HEAT repeats and FAT domain of mTOR. Whether small GTPases modulate TORC2 in any similar fashion has been more elusive. It has been reported that mammalian Rab35 affects TORC2 signaling to the AGC kinase AKT, but only indirectly through the effect of Rab35 on intracellular trafficking of PtdIns 3-kinase (Wheeler et al., 2015). The closest S. cerevisiae orthologue of mammalian Rab35 is Ypt1, which clearly functions in the ER-to-Golgi step of the yeast secretory pathway (Novick, 2016), not at the PM. Evidence for a more direct involvement of a small GTPase in TORC2 regulation was provided for fission yeast Ryh1 (Tatebe et al., 2010). However, the closest S. cerevisiae orthologue of Ryh1 is Ypt6 (mammalian counterpart Rab6), which functions in cis- to trans-Golgi vesicle trafficking (Suda et al., 2013), not at the PM. We were quite surprised, therefore, when we observed that deficiencies of active GTP-bound Rab5—due to absence of the Vps9 and Muk1 GEFs, to absence of all three Rab5 isoforms (Vps21, Ypt52, and Ypt53), or to elevation of the Rab5-specific GAP Msb3—all lowered both basal and stress-induced TORC2-mediated phosphorylation of Ypk1, as assessed biochemically, genetically, or both. Thus, in S. cerevisiae, Rab5 GTPases are an important positive effector of TORC2 function.
Yeast TORC2 components are located at the cell cortex in close apposition to the PM but are highly dynamic (Berchtold and Walther, 2009; Leskoske et al., 2018). Muk1 and its target Vps21 have both been localized to endosomes (Cabrera and Ungermann, 2013; Paulsel et al., 2013; Bean et al., 2015; Shideler et al., 2015). Our findings that Muk1 is a substrate of the TORC2-Ypk1 kinase cascade and that Vps21 is required for full TORC2 activity raise the possibilities that the observed dynamic behavior of yeast TORC2 reflects, first, that a fraction of the population of active TORC2 is on endosomes and, second, that a fraction of the GTP-bound Vps21 is at the PM. Similarly, by monitoring a fluorescent mSIN1 reporter (S. cerevisiae counterpart Avo1), it was observed in mammalian cells that active mTORC2 could be found at both the PM and on endosomes (Ebner et al., 2017).
On the other hand, despite the fact that Rab5 function is involved in endosome internalization, trafficking, and recycling, we could find no impairment in TORC2 localization in cells deficient in active Rab5. So, unlike Rag GTPases, Rab5 does not seem to have a role in tethering TORC2 to the PM per se. Rather, our data, especially the ability of Vps21 to coimmunoprecipitate with Tor2, even in the presence of detergent, and the ability of GTP-loaded Vps21 to stimulate TORC2-mediated phosphorylation of Ypk1 in vitro, indicate a more direct role for GTP-bound Rab5 in TORC2 function, as suggested for Ryh1 in fission yeast (Tatebe et al., 2010). Even though S. pombe Ryh1 has greater overall relatedness to S. cerevisiae Ypt6 (63% identity; 76% similarity) than to S. cerevisiae Vps21 (37% identity; 61% similarity), it is the case that mammalian Rab5B, Vps21, and Ryh1 all share with each other greater resemblance in their Switch I, Switch II, and hypervariable loop regions than they do with other classes of Ras superfamily GTPases (Fig. S5). In the case of Rheb, it is residues mainly in its Switch I and Switch II regions that make the main contacts with the M- and N-heat repeat regions and FAT domain in mTOR (Yang et al., 2017). Thus, it is possible that Rab5 binds to the Tor2 subunit of yeast TORC2 in the same binding pocket as that occupied by Rheb in the mTOR subunit in mTORC1. In both yeast and mammals, aside from Lst8, the ancillary subunits in TORC2 are all completely different from those in TORC1 (Eltschinger and Loewith, 2016; Tatebe and Shiozaki, 2017), and interaction of these other components with the catalytic subunit likely modifies the geometry of the binding pocket, so that it accommodates the right GTPase effector and excludes spurious ones. Furthermore, yeast and other fungi encode two highly related, but not identical, TOR isoforms; Tor2 is the primary catalytic subunit in TORC2, whereas Tor1 is the primary catalytic subunit in TORC1 (Helliwell et al., 1994; Roelants et al., 2017a). Hence, subtle sequence differences between Tor2 and Tor1 in their M- and N-heat repeat regions and FAT domain (Hill et al., 2018) may further fine tune the pocket in Tor2 to be selective for Rab5.
Regardless of the detailed molecular mechanism by which Rab5 stimulates the function of TORC2 (Fig. 7), Muk1 is activated in a TORC2-dependent and Ypk1-mediated manner, presumably generating more GTP-bound Rab5. Given that active Rab5 promotes TORC2 function, this control circuit provides self-reinforcing positive feedback that would lead to sustained TORC2-Ypk1 signaling, once it has been initiated. In this same regard, using the split-YFP (Venus) system to tag Vps21 and Msb3 for bimolecular fluorescence complementation, which traps protein–protein complexes (Sung and Huh, 2007; Kerppola, 2009), a substantial fraction of this Rab5-GAP complex was found at the PM, not on endosomes (Lachmann et al., 2012), indicating that Msb3 may be responsible for deactivation of the PM-associated pool of GTP-bound Vps21, thus providing a mechanism for down-regulation of Rab5-mediated stimulation of TORC2.
Materials and methods
Strains and plasmids
S. cerevisiae strains (Table 1) were constructed using standard genetic methods (Amberg et al., 2005). For integrations into the genome, proper insertion was confirmed by PCR amplification and sequencing. Plasmids (Table 2) were constructed using standard molecular biology techniques (Green and Sambrook, 2012) and verified by sequencing.
Cell culture conditions
Yeast strains were propagated in either standard rich medium (YP) containing 2% dextrose/glucose or a defined synthetic medium (SC; Sherman, 2002) with the same carbon source and supplemented with appropriate nutrients to maintain selection for plasmids. For galactose induction of gene expression, cells were pregrown in SC medium containing 2% raffinose-0.2% sucrose and induced by addition of galactose (2% final concentration) for 3 h. To inhibit complex sphingolipid synthesis, a stock of AbA (630466; Takara Bio USA) dissolved in ethanol was added at the indicated final concentrations, and an equivalent volume of the same solvent was added to control cultures. Stocks of the amino acid analogues l-canavanine (C1625; Sigma-Aldrich) and l-ethionine (E1260; Sigma-Aldrich) dissolved in water were added to plates of SD-containing agar medium at the indicated concentrations. For inhibition of analogue-sensitive Ypk1(L424A), a stock of the adenine analogue 3MB-PP1 (17860; Cayman Chemical) dissolved in DMSO was added at the indicated final concentration, and an equivalent volume of the same solvent was added to control cultures.
Expression from the corresponding plasmids (Table 2) of MBP-Muk1-(His)6, MBP-Muk1(6A)-(His)6, Vps21-(His)6, and Ypt7-(His)6 in the Escherichia coli derivative LOBSTR-BL21(DE3) (Andersen et al., 2013) was induced by addition of IPTG (0.75 mM final concentration). After incubation of the cultures at 16°C for 16 h, the cells were harvested by centrifugation and lysed by sonication, and the corresponding proteins were purified by immobilized metal ion affinity chromatography using Ni2+-charged, nitrilotriacetate-derivatized agarose beads (Qiagen) following standard procedures. After elution with 200 mM imidazole, pooled eluates were dialyzed into PBS containing 2 mM DTT. For Vps21-(His)6, the dialyzed protein was further purified by ion exchange chromatography on a MonoQ 10/100GL column (Sigma-Aldrich) using an AKTA Fast Protein Liquid Chromatography system (GE Healthcare). To load purified Vps21-(His)6 or purified Ypt7-(His)6 with the desired guanine nucleotide, samples were incubated for 10 min in 5 mM EDTA to strip any bound Mg2+-nucleotide complexes, then incubated for 30 min with 2 mM GTPγS (Roche Diagnostics), followed by addition of MgCl2 (20 mM final concentration) and incubation for an additional 30 min. MBP-Ypk1(603–680) [denoted MBP-Ypk1CT] and MBP-Ypk1CT(6A) were purified by affinity chromatography on amylose-agarose (New England Biolabs), following the manufacturer’s protocol. Analogue-sensitive Ypk1-as [Ypk1(L424A)-TAP] was purified from yeast by affinity purification using IgG-agarose and elution by cleavage with Pre-Scission protease (GE Healthcare), as described in detail previously (Muir et al., 2014).
Ypk1 protein kinase assays
To monitor the protein kinase activity of Ypk1, samples of purified Ypk1-as (0.15 µg) were mixed with, typically, 1 µg of the purified substrate of interest in kinase assay buffer (125 mM potassium acetate, 12 mM MgCl2, 0.5 mM EDTA, 2 mM DTT, 1% glycerol, 0.2 mg/ml BSA, 12.5 mM β-glycerol phosphate, 1 mM NaVO4, and 50 mM Hepes, pH 7.4, containing cOmplete EDTA-free protease inhibitor cocktail [1 mini-tablet per 10 ml; Roche]) in the absence or presence of 3MB-PP1 (10 µM final concentration), and reaction was initiated by addition of ATP (10 µM final concentration) containing 2 µCi [γ-32P]ATP. After incubation at 30°C for 30 min, reactions were terminated by the addition of an equal volume of 6× SDS gel sample buffer followed by boiling for 5 min, and the resulting products were resolved by SDS-PAGE and analyzed by both protein staining (0.05% Coomassie Blue R-250, 10% acetic acid, and 25% isopropanol) and autoradiography of the same fixed and stained gel performed using a Phosphorimager screen (Molecular Dynamics) and a Typhoon imaging system (GE Healthcare).
Immuno-isolation of TORC2 and other FLAG-tagged proteins
For immunoprecipitation of FLAG-tagged proteins expressed in yeast, including TORC2 tagged with Avo3-3xFLAG, frozen cell pellets were resuspended in an equal volume of ice-cold lysis buffer (200 mM KCl, 10 mM magnesium acetate, 10 mM β-glycerol-phosphate, 10 mM NaF, 10% glycerol, 0.1 mM EDTA, and 50 mM Hepes, pH 7.4, containing 2 mM 4-(2-aminoethyl)benzenesulfonyl fluoride and cOmplete EDTA-free protease inhibitor cocktail [1 mini-tablet per 10 ml; Roche]), mixed with an equivalent volume of chilled glass beads, and lysed by vigorous vortexing for 10 min at 4°C. The resulting lysate was separated from the beads by draining after puncturing the conical tip of the plastic centrifuge tube with a 25-gauge hypodermic needle. The recovered liquid was diluted with 0.5 volume of lysis buffer containing 2% n-dodecyl β-d-maltoside (DDM; D4641; Sigma-Aldrich) and incubated with gentle agitation on a rollerdrum for 20 min at 4°C. The detergent-treated extract was clarified by centrifugation at maximum rpm in a microfuge (Model 5424; Eppendorf), and the resulting supernatant fraction was withdrawn and mixed with 15 µl of a slurry of magnetic affinity resin decorated with anti-FLAG M2 antibodies (M8823; Sigma-Aldrich) that had been preequilibrated in lysis buffer containing 0.5% DDM. After incubation with resin for 2 h with gentle agitation on a rollerdrum at 4°C, the resin was removed from the bulk solution and moved to the side of the tube using a magnet tube rack (Dynal; Thermo Fisher Scientific) and then resuspended in and washed six times with 1 ml lysis buffer containing 0.5% DDM. For TORC2 purification, the remaining bound protein was eluted with 3xFLAG peptide (Sigma-Aldrich) in 1× kinase assay buffer, and samples of the resulting eluate were used immediately thereafter in in vitro kinase assays. For other FLAG-tagged proteins, the remaining bound protein was eluted with urea-SDS gel sample buffer (6 M urea, 6% SDS, 25% glycerol, 0.1% bromophenol blue, and 150 mM Tris-HCl, pH 6.8).
TORC2 protein kinase assays
To monitor the protein kinase activity of TORC2, samples of TORC2 (each derived by immuno-isolation from the equivalent of 50 µg clarified lysate, as described above) were incubated with purified MBP-Ypk1CT or MBP-Ypk1CT(6A), in kinase assay buffer in the absence or presence of the indicated amounts of GTPγS-loaded Vps21 or Ypt7, and reactions were initiated by the addition of ATP (10 µM final concentration) containing 2 µCi [γ-32P]ATP. Reactions were incubated at 30°C for 30 min unless otherwise noted, and the products were analyzed in the same manner as for the Ypk1 protein kinase assays, except SYPRO Ruby (Thermo Fisher Scientific) was used for protein staining. For specific inhibition of WT TORC2, the enzyme was incubated with QL-IX-55 (gift of Nathanael Gray, Harvard Medical School, Cambridge, MA) dissolved in DMSO or with an equivalent volume of the same solvent at 30°C for 5 min before addition of the other reaction components. For specific inhibition of the TORC2 containing Tor2(L2178A), the enzyme was incubated with NVP-BEZ235 (Cayman Chemical) dissolved in dimethyl formamide or with an equivalent volume of the same solvent.
Preparation of cell extracts and immunoblotting
Whole-cell lysates of yeast were prepared by alkaline lysis with 1.85 M NaOH and 7.4% 2-mercaptoethanol followed by TCA precipitation, as described in detail previously (Westfall et al., 2008). For samples treated with CIP, the TCA precipitates of whole-cell lysates were washed with ice-cold acetone, resuspended in 100 µl CIP resuspension buffer (126 mM sorbitol, 42 mM NaCl, 10.5 mM MgCl2, 0.24 mM EDTA, 2% 2-mercaptoethanol, and 180 mM Tris base), diluted with 900 µl of 50 mM Tris-HCl, pH 8.0, and incubated with 45 units of CIP (New England Biolabs) for 2 h at 30°C. Reactions were terminated by addition of TCA (25% final concentration), and the precipitated protein was resuspended in a buffer containing 0.1 M Tris base and 5% SDS before addition of an appropriate volume of 5× SDS sample buffer.
In most cases, protein samples were separated by standard SDS-PAGE in gels containing the appropriate concentration of acrylamide to resolve the species of interest, transferred electrophoretically to nitrocellulose filter paper, and analyzed by immunoblotting. To resolve the spectrum of phospho-isoforms of particular phosphorylated proteins (such as Muk1 and Ypk1), resuspended whole-cell lysates were separated using Phos-tag SDS-PAGE in 8% acrylamide gels containing 35 µM Phos-Tag reagent (Fujifilm Wako Chemicals USA), according to the manufacturer’s instructions, and, after transfer to nitrocellulose, analyzed by immunoblotting.
Primary antibodies used in this study (at the dilution indicated) were mouse anti-FLAG (1:5,000, F1804; Sigma-Aldrich); rabbit anti-DYKDDDDK (1:2,500, 2368; Cell Signaling Technologies); mouse anti-c-myc mAb 9E10 (1:100; Monoclonal Antibody Facility, Cancer Research Laboratory, University of California, Berkeley, Berkeley, CA); mouse anti-HA 6E2 (1:5,000, 2367; Cell Signaling Technologies); rabbit anti-GST (1:10,000, SC-459; Santa Cruz Biotechnology); rabbit anti-Pgk1 (1:200,000; Baum et al., 1978); rabbit anti-SGK1(pT256) (1:1,000, SC-16744; Santa Cruz Biotechnology); mouse anti-Pma1 (1:5,000, ab4645; Abcam); and rabbit anti-Ypk1(pT662) (1:20,000; gift from Ted Powers, University of California, Davis, Davis, CA). Infrared dye–labeled secondary antibodies were CF770-derivatized goat anti–mouse IgG (1:20,000, 20077; Biotium) and CF680-derivatized goat anti–rabbit IgG (1:20,000, 20067; Biotium). The resulting filter-bound immune complexes were visualized using an infrared imaging system (Model Odyssey CLx; LI-COR Biosciences).
GST pull-down analysis
Lysates were prepared in the same manner as for immunoprecipitation of FLAG-tagged proteins (above). The resulting extracts were mixed with a slurry (10 µl) of glutathione-Sepharose 4B beads (GE Healthcare) preequilibrated in lysis buffer containing 0.5% DDM. After incubation for 2 h at 4°C, the beads were separated from the protein solution and washed 6× with 1 ml lysis buffer, and the bound proteins were eluted with urea-SDS gel sample buffer and analyzed.
Yeast cells expressing Avo3-GFP (excitation λmax 489 nm; emission λmax 508) were grown to mid-exponential phase in synthetic medium lacking Trp, mounted onto 1% agarose pads, and viewed immediately at 23°C using a confocal microscope (Model LSM 710; Zeiss) equipped with a 100× oil-immersion objective (1.40 NA), excited with an argon laser at 488 nm at 2.3% power (100 mW), and emission monitored using a bandpass filter (495–550-nm window). Images were acquired with an iXon3 EM-CCD camera (Andor) using Metamorph software (Molecular Devices) and processed using ImageJ (Collins, 2007).
Online supplemental material
Fig. S1 shows that phosphorylation of full-length Muk1 occurs at its Ypk1 consensus phospho-acceptor site motifs and is stimulated by treatment of the cells with AbA; and, using a growth assay for temperature sensitivity, that Ypk1 phosphorylation is necessary for Muk1 function. Fig. S2 shows that, in the absence of the Rab5 GEFs, localization of TORC2 is unaffected, phosphorylation of Ypk1 by Pkh1 is unimpaired, Ypk1 association with the Avo1 subunit of TORC2 is unimpaired, and forced PM association of Ypk1 cannot restore TORC2-dependent phosphorylation. Fig. S3 shows that Vps21 coimmunoprecipitates with Tor2, but not Ypk1, and the effect of guanine nucleotide on Vps21-Tor2 association. Fig. S4 shows that phosphorylation of MBP-Ypk1CT occurs on its TORC2-specific sites, is blocked by specific inhibitors of TORC2, and is stimulated by Vps21-GTPγS. Fig. S5 shows a sequence alignment of the yeast Rab5 isoforms to other classes of small GTPases from both yeast and human cells.
We thank Nathanael Gray for the generous gift of QL-IX-55, Alex Merz (University of Washington, Seattle, WA) for Vps21 mutants, Kirsten Young for technical assistance, Dr. Anita Emmerstorfer-Augustin for help with strain construction, the staff of the University of California Berkeley DNA Sequencing Facility for sequence analysis, Dr. Françoise M. Roelants for her constant guidance, Ben Wharton for proofreading, and all other members of the Thorner Lab for their constructive criticism and enthusiastic support.
This work was supported by National Institutes of Health Predoctoral Traineeship GM07232, by a Summer Research Grant from the University of California Berkeley Graduate Division, and by a Cancer Research Coordinating Committee Fellowship from University of California Systemwide (all to M.N. Locke) and by National Institutes of Health, National Institute of General Medical Sciences Research Grant R01GM21841 (to J. Thorner).
The authors declare no competing financial interests.
Author contributions: M.N. Locke designed, executed, and analyzed experiments and drafted the manuscript. J. Thorner designed and analyzed experiments and revised the manuscript.