Complementary α-arrestin-ubiquitin ligase complexes control nutrient transporter endocytosis in response to amino acids

How cells adjust nutrient transport across their membranes is incompletely understood. Previously, we have shown that S. cerevisiae broadly re-configures the nutrient transporters at the plasma membrane in response to amino acid availability, through endocytosis of sugar- and amino acid transporters (AATs) (Müller et al., 2015). A genome-wide screen now revealed that the selective endocytosis of four AATs during starvation required the α-arrestin family protein Art2/Ecm21, an adaptor for the ubiquitin ligase Rsp5, and its induction through the general amino acid control pathway. Art2 uses a basic patch to recognize C-terminal acidic sorting motifs in AATs and thereby instructs Rsp5 to ubiquitinate proximal lysine residues. When amino acids are in excess, Rsp5 instead uses TORC1-activated Art1 to detect N-terminal acidic sorting motifs within the same AATs, which initiates exclusive substrate-induced endocytosis. Thus, amino acid excess or starvation activate complementary α-arrestin-Rsp5-complexes to control selective endocytosis and adapt nutrient acquisition.


Introduction
Cells regulate the import of amino acids, glucose, and other nutrients to fuel metabolism, sustain growth or maintain homeostasis. For the selective transport of amino acids across the plasma membrane (PM) and other cellular membranes, the human genome encodes more than 60 known amino acid transporters (AATs). Mutations in AATs cause severe defects of amino acid metabolism, and the deregulation of AATs is linked to a range of human pathologies, including neurodegenerative diseases, diabetes and cancer (Kandasamy et al., 2018;Smith, 1990;McCracken and Edinger, 2013;Zhang et al., 2017).
AATs belong to the family of solute carriers (SLCs) and form selective pores that change from an outward-to an inward-facing conformation and thereby transport their amino acid substrates across membranes. Hence, the addition of AATs to the PM or their selective removal by endocytosis determines quantity and quality of amino acid transport. In human cells, the molecular mechanisms leading to the selective endocytosis of AATs are largely unclear.
The mechanisms for the endocytic down-regulation of AATs are beginning to emerge in the budding yeast, S. cerevisiae. Yeast cells frequently experience acute fluctuations in amino acid availability (Broach, 2012) and can rapidly remodel their 21 PM-localized AATs to optimize the import of amino acids with regard to the quantity and quality of the nitrogen source available Grenson, 1966;André, 2018;Van Belle and André, 2001;Bianchi et al., 2019). The selective, ubiquitin-dependent endocytosis of AATs and other integral PM proteins frequently requires the HECT-type ubiquitin ligase Rsp5, the orthologue of human Nedd4 (Hein et al., 1995;Dupré et al., 2004;Belgareh-Touzé et al., 2008). To specifically recognize and ubiquitinate PM proteins, Rsp5 interacts with the a-arrestin proteins, a family of ubiquitin ligase adaptors (Nikko et al., 2008;Lin et al., 2008). The budding yeast genome encodes 14 a-arrestin proteins (arrestin-related trafficking adaptors, ARTs; Art1-10, Bul1-3 and Spo23), which function in a partially redundant manner (Becuwe et al., 2012a;MacGurn et al., 2012;Babst, 2020;O'Donnell and Schmidt, 2019;Nikko and Pelham, 2009). Six a-arrestins have been identified in humans (Alvarez, 2008). a-Arrestin proteins use PPxY (PY) motifs to interact directly with the three WW domains of Rsp5/Nedd4 (Becuwe et al., 2012a;Rauch and Martin-Serrano, 2011). How a-arrestins recognize their cargoes in general, and more specifically AATs, is only partially understood, but appears to involve their arrestin domains. These domains have large, disordered loop-and tail-like insertions that contribute to the function of ARTs (Baile et al., 2019).
The activity of aÀarrestins is controlled by complex post-translational modifications (PTMs) involving (de-)ubiquitination and (de-)phosphorylation. In several cases, the binding of aÀarrestins to Rsp5 results in their ubiquitination, which is required for function (Lin et al., 2008;Becuwe et al., 2012b;Hovsepian et al., 2017). In case of Art1, ubiquitination determines its localization (Lin et al., 2008). Additional ubiquitination of ARTs can target them for proteasomal degradation, which is counteracted by deubiquitinating enzymes (Ho et al., 2017). Several aÀarrestins are activated by dephosphorylation events in response to nutrient availability or other cellular stresses (Becuwe et al., 2012a;O'Donnell et al., 2013;Alvaro et al., 2014;Becuwe et al., 2012b;Becuwe and Léon, 2014;Merhi and André, 2012). One example are high levels of amino acids, which activate the target of rapamycin complex 1 (TORC1). This leads to inhibition of the kinase Npr1 and thereby promotes the activity of Art1 (MacGurn et al., 2011). At the PM, Ppz phosphatases activate Art1 by dephosphorylation of its Npr1-dependent phosphorylation sites (Lee et al., 2019) and thereby stimulate ubiquitination and endocytosis of the Art1 cargoes Mup1 (methionine transporter), Can1 (arginine transporter), Tat2 (tyrosine and tryptophan transporter) and Lyp1 (lysine transporter) in response to substrate excess. Similarly, TORC1 signaling also promotes endocytosis of the general amino acid permease (Gap1) via the ARTs Bul1/2 (Merhi and André, 2012). It seems that nutrientdependent TORC1 activation controls aÀarrestin-mediated ubiquitin-dependent AAT endocytosis to adjust amino acid influx to metabolic needs. Activated Art1 recognizes specific sorting signals in AATs. The flux of methionine through Mup1 or arginine through Can1 requires conformational changes into the inward-facing conformation, which probably dislodge a C-terminal plug that otherwise 'seals' the pore. At the same time, it exposes an N-terminal acidic patch that is recognized by activated Art1. Art1 interacts with this acidic patch and then orients Rsp5 to ubiquitinate nearby lysine residues (Guiney et al., 2016;Gournas et al., 2018;Busto et al., 2018;Ghaddar et al., 2014). Similar results have been obtained for the uracil transporter Fur4 (Keener and Babst, 2013). A further layer of complexity may be added by stimulus-induced phosphorylation of nutrient transporters (Marchal et al., 1998;Paiva et al., 2009), akin to the recognition of G-protein-coupled receptors (GPCRs) by b-arrestins in mammalian cells (Nikko et al., 2008;Nobles et al., 2011). Hence, a molecular basis of aÀarrestindependent nutrient transporter endocytosis in response to excess nutrients is emerging. We refer to this process as substrate-induced endocytosis.
The molecular mechanisms of starvation-induced endocytosis are not clear. Hence, there is a significant knowledge gap in understanding how cells control nutrient transporter abundance upon amino acid and nitrogen scarcity. Here, we have characterized how amino acid abundance alters several nutrient transporters at the PM of the budding yeast S. cerevisiae. We find that TORC1 signaling and the general amino acid control (GAAC) pathway toggle Art1-or Art2-Rsp5 complex activities to induce endocytosis of the same set of four different AATs depending on amino acid availability. In these AATs, activated Art1-or Art2-Rsp5 complexes recognize distinct acidic sorting signals in the N-terminal (Art1 sorting signal) or C-terminal (Art2 sorting signal) regions to initiate Rsp5-dependent ubiquitination and subsequent endocytic down-regulation. Using such complementary aÀarrestin switches in combination with distinct acidic sorting motifs in a single nutrient transporter could represent a more general mechanism to adjust transport across cellular membranes and thereby to meet metabolic demands.

Amino acid availability induces selective endocytosis of nutrient transporters
We employed S. cerevisiae as a model system to address how eukaryotic cells adjust their nutrient transporters at the plasma membrane (PM) to nutrient availability. First, we used live cell fluorescence microscopy to analyze in yeast cells the localization of 149 putative PM proteins that were C-terminally GFP-tagged at their native chromosomal locus (Saier et al., 2016;Babu et al., 2012;Breker et al., 2014;Huh et al., 2003). This collection included 16 different amino acid transporters (AATs) out of the 21 AATs that localize to the PM (Bianchi et al., 2019). In cells growing exponentially under defined (rich) conditions, we detected 50 GFP-tagged proteins at the PM, including eight different AATs and six different carbohydrate transporters ( Figure 1A Others and we had shown earlier that amino acid and nitrogen starvation (hereafter starvation) induced the degradation of membrane proteins via the MVB pathway (Müller et al., 2015;Jones et al., 2012). Consistently, in response to starvation 15 (out of 50) different PM proteins were selectively removed and transported into the vacuole, including four AATs (Mup1, Can1, Lyp1, Tat2) and four carbohydrate transporters (Hxt1, Hxt2, Hxt3, Itr1) ( Figure 1B, Supplementary file 1). The general amino acid permease Gap1 and the ammonium permease Mep2 were up-regulated and now localized to the PM. The localization of 35 GFP-tagged proteins to the PM remained largely unchanged, although in some instances the vacuolar GFP fluorescence was increased ( Figure 1A,B, Figure 1-figure supplement 1A, Supplementary file 1). For five AATs, we did not detect signals under rich or starvation conditions (Supplementary file 1).
Next, we examined how amino acid availability could regulate AAT endocytosis. Therefore, we used the high-affinity methionine transporter Mup1 as a model cargo because its regulation in response to nutrient excess is well characterized (Busto et al., 2018;Lee et al., 2019;Gournas et al., 2018;Guiney et al., 2016;Baile et al., 2019). Moreover, Mup1 is one of the most abundant PM proteins and it is easy to follow its endocytosis and subsequent transport into the vacuole (Busto et al., 2018). In absence of methionine Mup1-GFP localized to the PM (Figure 1-figure supplement 2A). Low levels of methionine in the growth medium did not efficiently trigger its endocytosis ( Figure 1-figure supplement 2A,B). Yet, above a critical methionine concentration the vast majority of Mup1-GFP was removed from the PM by endocytosis and was subsequently transported to the vacuole via the multivesicular body (MVB) pathway ( Figure 1-figure supplement 2A,B). This was confirmed by western blot (WB) analysis from total cell lysates. Without methionine or with low levels of methionine in the growth medium only full-length Mup1-GFP was detected (Figure 1-figure supplement 2B, lanes 1+2). Once the critical methionine concentration was surpassed, Mup1-GFP was delivered into the vacuole. The proteolytic degradation of Mup1-GFP inside the vacuole then released free GFP, which remained stable and could be monitored by western blotting (Figure 1-figure supplement 2B, lane 4). Substrate-induced endocytosis was rapid and efficient: within 60-90 min after methionine addition Mup1 was almost quantitatively delivered to the vacuole. Excess methionine appeared to exclusively induce endocytosis of Mup1 since all other tested PM proteins remained at the cell surface ( Figure 1A, Supplementary file 1). This exclusive substrateinduced endocytosis of Mup1 was also dependent on its methionine transport activity as shown by using the Mup1-G78N Figure 1. Amino acid and nitrogen starvation triggers broad but specific endocytosis and lysosomal degradation of plasma membrane proteins. (A) Left: a library of 149 yeast strains expressing chromosomally GFP-tagged membrane proteins was tested for plasma membrane (PM) localization during nutrient replete exponential growth. Right: verified PM proteins were starved 6-8 hr for amino acids and nitrogen (-N) or treated with 20 mg/ml L-methionine (+Met) after 24 hr of exponential growth. The localization of GFP was assayed by fluorescence microscopy. (B) Summary of the phenotypes of GFP-tagged PM proteins during starvation. Indicated are numbers of PM proteins that are down-regulated, up-regulated or unchanged compared to the exponential growth phase, each exemplified by one representative strain. PM: plasma membrane; V: vacuole. Scale bars = 5 mm. See also    (Busto et al., 2018;Figure 1-figure supplement 2C). These results are consistent with earlier reports showing that the high-affinity AATs Mup1, Can1, Tat2 and Lyp1 undergo exclusive endocytic down-regulation in response to excess of their respective substrates but remained stable at the PM when the substrate of another permease was present in excess (Gournas et al., 2017;Busto et al., 2018;Nikko and Pelham, 2009).
In contrast, the starvation-induced endocytosis of Mup1 was independent of its capability to transport its substrate (Figure 1-figure supplement 2C) and was also observed upon starvation for individual essential amino acids (Figure 1-figure supplement 2D). We re-evaluated the starvationinduced endocytosis of Mup1, Can1, Lyp1, Tat2, Ina1, Fur4, Hxt1, Hxt2 and Hxt3 in a different genetic background (SEY6210; Figure 1-figure supplement 2A, Figure 3C, Hence, in response to starvation, a broad range of nutrient transporters, including four AATs (Mup1, Can1, Lyp1, Tat2) and four carbohydrate transporters (Hxt1, Hxt2, Hxt3, Itr1) as well as other membrane proteins, were selectively removed from the PM by endocytosis and transported into the vacuole within 3-6 hr. In contrast, substrate-induced endocytosis of AATs was quick and exclusive ( Figure 1A). Earlier work showed that substrate-induced endocytosis of the AATs Mup1, Can1 and Lyp1 required TORC1 signaling to allow ubiquitination of the transporters by the HECT-type ubiquitin ligase Rsp5 in complex with the a-arrestin Art1 (Lin et al., 2008;Guiney et al., 2016;Busto et al., 2018;Gournas et al., 2018;MacGurn et al., 2011). Remarkably, the same set of AATs underwent simultaneous starvation-induced endocytosis. The molecular mechanism of their starvation-induced endocytosis was not clear.

A genome-wide screen identifies regulators of starvation-induced endocytosis of AATs
To identify the genes that are required for the starvation-induced endocytosis of Mup1, we conducted a fluorescence-based genome-wide screen. We measured Mup1 sorting into the lumen of vacuoles via the MVB pathway by fusing the pH-sensitive GFP variant pHluorin to the C-terminus of Mup1 (Mup1-pHluorin) (Henne et al., 2015;Prosser et al., 2010). Fluorescence microscopy and flow cytometry (Figure 2A,B) showed a strong signal of Mup1-pHluorin at the PM of cells growing under defined rich conditions. In response to starvation, the fluorescence of Mup1-pHluorin was efficiently quenched in the acidic vacuoles of wild type (WT) cells, but not when the MVB pathway was blocked (e.g. ESCRT-I mutant vps23D) (Figure 2A,B). Hence Mup1-pHluorin is a suitable reporter to identify mutants that block trafficking of Mup1 from the PM into the vacuole.
Next, Mup1-pHluorin was introduced into the yeast non-essential knock-out collection. 4745 mutants expressing Mup1-pHluorin were grown in selection medium to exponential phase before they were subjected to starvation for 18-22 hr ( Figure 2C). Automated flow cytometry was used to measure Mup1-pHluorin fluorescence intensity of at least 15,000 cells before and after starvation. The vast majority of mutants efficiently quenched the fluorescence of Mup1-pHluorin after starvation, but 128 mutants still exhibited Mup1-pHluorin fluorescence after starvation ( Figure 2C, Supplementary file 2). To determine at which step Mup1-pHluorin transport into vacuoles was blocked in these mutants, we used live cell fluorescence microscopy. This allowed us to classify four phenotypes ( Figure 2D, Supplementary file 2). Class one mutants retained Mup1-pHluorin at the PM after starvation (105 mutants, e.g. apm2D mutants). In class two mutants, Mup1-pHluorin was detected on small intracellular objects (seven mutants, e.g. vam3D mutants). Class three mutants accumulated Mup1-pHluorin in larger class E compartment-like objects (15 mutants, e.g. vps4D mutants). One Class four mutant (pep4D) did not efficiently quench Mup1-pHluorin fluorescence inside the vacuole, as it is deficient in the main lysosomal protease. Gene ontology (GO) analysis confirmed that our screen identified major general regulators of endosomal transport (e.g. the ESCRT complexes) and was enriched for proteins that are annotated as components of the PM, endosomes and vacuoles (Figure 2-figure supplement 1, Supplementary file 3) suggesting that it was successful. In addition, we identified many genes and components of protein complexes that would not be predicted to be involved in the regulation of endocytosis (Figure 2-figure supplement 1, Supplementary file 2, 3).
Next, we aimed to narrow down genes that function specifically during starvation-induced endocytosis and to distinguish them from general endocytic regulators (i.e. core components of endocytic trafficking machineries). Therefore, most mutants (124) were exposed to methionine excess for 90 The aÀarrestin Art2 is specifically required for amino acid transporter endocytosis in response to starvation A key step for the endocytic down-regulation of PM proteins is their selective ubiquitination. This selectivity is mediated by aÀarrestin molecules that direct the HECT-type ubiquitin ligase Rsp5 to ubiquitinate specific PM proteins (Babst, 2020). Among the 14 aÀarrestins encoded by the yeast genome, our screen identified only Ecm21/Art2 to be specifically required for starvation-induced endocytosis of Mup1. We compared Art2-dependent starvation-induced endocytosis to Art1-dependent substrateinduced endocytosis of Mup1-GFP. Efficient methionine-induced endocytosis of Mup1 required Art1 but not Art2 ( Figure 3A), consistent with earlier observations (Guiney et al., 2016;MacGurn et al., 2011;Busto et al., 2018). Starvation-induced endocytosis of Mup1 was slower than methionineinduced endocytosis. In WT cells, 240-360 min after onset of starvation the majority of Mup1-GFP was removed from the PM and sorted into vacuoles for degradation ( Figure 3A,B, Figure 3-figure supplement 1A). In art2D mutants, but not in art1D mutants, the vast majority of Mup1-GFP remained at the PM and was no longer delivered to vacuoles ( Figure 3A). This was confirmed by WB analysis. In WT cells, the levels of full length Mup1-GFP decreased, while free GFP accumulated over time ( Figure   . Importantly, Art2 was required for the efficient starvation-induced endocytosis of all four AATs, Mup1, Can1, Lyp1, and Tat2 ( Figure 3A,C). Interestingly, the same four AATs (Mup1, Can1, Lyp1 and Tat2) also depend on Art1 for efficient substrate-induced endocytosis (Gournas et al., 2017;MacGurn et al., 2011;Lin et al., 2008;Nikko and Pelham, 2009). Starvation-induced endocytosis of Ina1 was also dependent on Art2 (Figure 3-figure supplement 1C), while other PM proteins (e.g. the uracil transporter Fur4 or the hexose transporters Hxt1 and Hxt2) were largely independent of Art2 ( Figure 3-figure supplement 1D).
These results suggested a swap between the aÀarrestins Art1 and Art2 by amino acid availability as a rule for the regulation of four AATs. The starvation-induced endocytosis of Mup1, Can1, Lyp1 and Tat2 required Art2. Substrate-induced endocytosis of Mup1, Can1 and Lyp1 required Art1, whereas in case of Tat2 substrate-induced endocytosis was less stringent and could be mediated either by Art1 or Art2 (Nikko and Pelham, 2009). and condition were analyzed by flow cytometry. The exemplified histograms display decrease of fluorescence in wild type (WT)-like strains and fluorescence retention in mutants with defects in the starvation-induced endocytosis of Mup1-pHluorin (e.g. vps23D). (C) Workflow of the flowcytometry-based genome-wide screen for mutants defective in starvation-induced endocytosis of Mup1-pHluorin. (D) Summary of phenotypes of all mutants scored in the starvation-induced endocytosis screen (C) as determined by fluorescence microscopy. Class one mutants retain Mup1-pHluorin fluorescence at the plasma membrane (PM); class two mutants in small cytosolic objects; class three mutants in class E-like objects (E); class four mutants within vacuoles (V). Each phenotype is exemplified by one representative deletion mutant. Indicated are the numbers of strains that share a similar phenotype and the number of hits specific for starvation-induced endocytosis of Mup1-pHluorin (red). Scale bars = 5 mm. See also Our experiments so far demonstrated that Art1 and Art2 mediate the endocytic down-regulation of Mup1, Can1 and Lyp1 in response to changes in amino acid availability in a mutually exclusive manner. Hence, their activity must be strictly regulated. aÀarrestin proteins are subject to complex multi-level regulation and their activation can require transcriptional regulation, (de-)phosphorylation and ubiquitination (Hovsepian et al., 2017;Lin et al., 2008;Becuwe et al., 2012b). WB analysis of total cell lysates indicated that 6xHis-TEV-3xFlag-tagged Art1 protein levels were unchanged in response to starvation ( Figure  Quantitative reverse transcription (RT) PCR analysis indicated that also the ART2 mRNA levels increased during starvation ( Figure 4B). In search of the pathway responsible for induction of Art2, we queried the results of our screen and identified the eIF2 kinase Gcn2 and several more GCN genes encoding key components of the general amino acid control (GAAC) pathway (Supplementary file 2). Gcn2 is activated during amino acid starvation by unloaded tRNAs and subsequently phosphorylates eIF2a to reduce protein synthesis in general, but thereby stimulates specifically the translation of the transcription factor Gcn4. Gcn4 then activates the transcription of genes, many of which are involved in amino acid metabolism (Hinnebusch, 2005). gcn4D was not scored in our genome-wide screen because it failed the quality control due to its slow growth phenotype. Interestingly, the promoter of ART2 contained predicted Gcn4-binding sites (Venters et al., 2011;Schuldiner et al., 1998). Disrupting the GAAC pathway (gcn4D) eliminated the induction of Art2 in response to starvation at mRNA and protein levels ( Figure  The expression of a constitutively translated Gcn4 C construct (Mueller and Hinnebusch, 1986) increased Art2 protein levels already under rich conditions, as revealed by WB analysis ( Figure 4F, compare Art2 protein levels in lanes 1 and 3,  Figure 4G) and bypassed the requirement of Gcn4 during starvation.
These results indicated that during amino acid starvation, Art2 transcription was induced via the GAAC pathway, leading to the up-regulation of Art2 protein levels. Moreover, it seemed that the up-regulation of Art2 protein levels was sufficient to drive Mup1 endocytosis. Hence, under amino acid replete conditions Art2 must be tightly repressed to prevent AAT endocytosis. Art2-HTF Art2 directs Rsp5-dependent ubiquitination of C-terminal lysine residues in Mup1 To determine how Art2 contributed to Mup1 endocytosis, we examined its role in Mup1 ubiquitination. WT cells were harvested and Mup1-GFP was immunoprecipitated in denaturing conditions before and at different time points after starvation. Equal amounts of immunoprecipitated full-length Mup1-GFP were subjected to SDS-PAGE and WB analysis to compare the extent of its ubiquitination at different time points ( Figure 5A). This analysis indicated that a pool of Mup1 was ubiquitinated prior to the onset of starvation ( Figure 5A, lane 1). At the onset of starvation, ubiquitination of Mup1-GFP appeared to decrease for some time ( Figure 5A lanes 2-4), until ubiquitination of Mup1-GFP began to increase again after 2-3 hr during starvation ( Figure 5A lanes 4-6), temporally coinciding with Art2 induction and starvation-induced endocytosis. Mup1-GFP was still ubiquitinated in art2D cells growing under rich conditions ( Figure 5B lane 4) and seemingly de-ubiquitinated at the onset of starvation, but the increase of ubiquitination during starvation was no longer observed ( Figure 5B, lane 6). Hence, Art2 was essential for the starvation-induced ubiquitination of Mup1.
aÀArrestins use PY motifs to bind to at least one of the three WW domains of Rsp5 (Lin et al., 2008). Starvation-induced endocytosis of Mup1 (but not methionine-induced endocytosis) was particularly dependent on the WW3 domain of Rsp5 ( Figure 5C,D, Figure 5-figure supplement 1A). Art2 has four putative PY motifs ( Figure 5C). Mutations in the PY motif that resides within the predicted arrestin fold of Art2 (P748,P749,Y750) reduced starvation-induced endocytosis of Mup1-GFP ( Figure 5E). The Art2 P748A,P749A,Y750A mutant was expressed at similar levels than the WT protein (Figure 5-figure supplement 1B). We suggest that interaction between WW3 in Rsp5 and the PY motif (748-750) of Art2 was required for the efficient starvation-induced endocytosis of Mup1.
Collectively, these results suggested that Art2 directs Rsp5 to the ubiquitination of the C-terminal lysine residues K567 and K572 during starvation, whereas Art1-Rsp5 complexes ubiquitinated Mup1 on the two N-terminal lysine residues K27 and K28 in response to methionine excess ( Figure 5F).

The C-terminus of Mup1 contains an acidic sorting signal for Art2
In addition to ubiquitination, our mass-spectrometry analysis of immunoprecipitated Mup1-GFP from starved cells identified potential phosphorylation sites at the N-terminus (S9,23,31,42 and T34) and at the C-terminus of Mup1 (T552 and S568,573) ( Figure 6A, Figure 5-figure supplement 1C). Additional phosphorylation on T560 was reported in other studies (Swaney et al., 2013;Iesmantavicius et al., 2014). Of note, the C-terminal phosphorylation sites were proximal (T552, T560) or directly adjacent (S568, S573) to lysine residues K567 and K572 that were ubiquitinated in an Art2-dependent manner in response to starvation ( Figure  Mutation of the C-terminal threonine residues to alanine (T552,560A) to prevent phosphorylation specifically blocked starvation-induced endocytosis of Mup1 T552,560A -GFP, but not methionineinduced endocytosis ( . It seemed that the phosphorylation of C-terminal threonine residues (T552, 560) contributed to the starvation-induced Art2-Rsp5 ubiquitination of Mup1 and thus might add an additional layer of regulation, similar to the recognition of GPCRs by b-arrestins in mammalian cells (Shukla et al., 2014;Nikko et al., 2008;Marchal et al., 2000;Hicke et al., 1998).
An extended acidic Art1 sorting motif at the N-terminus of Mup1 is required for ubiquitination of K27,28 and subsequent Mup1 endocytosis in response to methionine excess (Guiney et al., 2016;Busto et al., 2018). Consistent with our model involving two distinct mechanisms for Mup1 endocytosis, mutations in the acidic N-terminal Art1 sorting motif of Mup1 (D43A,G47A,Q49A,T52A,L54A) blocked Art1-Rsp5-dependent methionine-induced endocytosis, but not Art2-Rsp5-dependent starvation-induced endocytosis ( Figure 6-figure supplement 1A). The C-terminus of Mup1 also contains an acidic patch (D549-D555), close to the ubiquitination sites K567 and K572 and the C-terminal threonine phosphorylation sites (T552, 560) involved in starvation-induced endocytosis ( Figure 6A,B). Mutation of the acidic residues in this region to basic amino acids (R) demonstrated that this C-terminal acidic region was specifically required for starvation-induced endocytosis. Live cell fluorescence microscopy revealed that Mup1 D549R,D551R,E554R, D555R -GFP remained at the PM in response to starvation, whereas methionine-induced endocytosis was not impaired ( Figure 6B). Even Art2 overexpression failed to induce endocytosis during exponential growth or starvation when the C-terminal acidic patch in Mup1 was mutated ( Figure 6-figure supplement 1B). Moreover, immunoprecipitation of Mup1 D549R,D551R,E554R,D555R -GFP and subsequent SDS-PAGE and WB analysis revealed that it was no longer efficiently ubiquitinated ( Figure 6C, lanes 4-6), suggesting that the C-terminal acidic patch was essential for the Art2-Rsp5dependent ubiquitination during starvation.
Comparing the amino acid sequences of the C-terminal tails of the four Art2-dependent AATs (Mup1, Can1, Tat2 and Lyp1) and of Ina1 indicated similar acidic patches ( Figure 6D). To analyze if the acidic patch in Can1 also contributed to starvation-induced endocytosis, we mutated D567, E569, E574 and E575 to arginine. Mutant Can1 D567R,E569R,E574R,D575R -GFP mostly localized to the PM under growing conditions. Importantly, the Art2-dependent starvation-induced down-regulation of Can1 D567R,E569R,E574R,D575R -GFP was impaired ( Figure 6-figure supplement 1C). These results imply that Mup1, Can1 and potentially also Lyp1, Tat2 and Ina1 have acidic amino acid sequences at their C-termini that could serve as sorting signal for Art2-mediated starvation-induced endocytosis.
The C-terminal acidic sorting signal of Mup1 is sufficient for Art2dependent starvation-induced endocytosis It seemed that the last 26 amino acid residues (aa 549-574) of Mup1 harbor three features that are collectively required specifically for starvation-induced endocytosis: putative phosphorylation sites, the acidic patch and the ubiquitination sites. Hence, we tested if the C-terminal region of Mup1 was sufficient to convert an Art2-independent cargo into an Art2 cargo. We selected the low-affinity glucose transporter Hxt3, which was efficiently removed from the PM in response to starvation (Figure 6-figure supplement 2A, Figure 1B, Supplementary file 1). Live cell fluorescence microscopy and WB analysis showed that starvation-induced endocytosis of Hxt3-GFP was independent of Art2, but instead required Art4 ( Figure 6E, Figure 6-figure supplement 2A). In art4D mutants, but not in art2D mutants, Hxt3-GFP remained mostly at the PM ( Figure 6-figure supplement 2A) in response to starvation, and its vacuolar degradation was impaired ( Figure 6-figure supplement 2B, lane 4, Figure 6-figure supplement 2C). However, when the C-terminal 30 amino acids of Mup1 (aa 545-574) were fused onto the C-terminus of Hxt3 ( Figure 6E), they restored starvationinduced endocytosis in art4D mutants. This Hxt3-Mup1-C-GFP chimera was now efficiently removed from the PM and transported to the vacuole in the vast majority of art4D cells (86%, n = 76 cells) ( Figure 6E), most likely because it became an Art2 substrate. Indeed, the additional deletion of Figure 6. The C-terminus of Mup1 harbors a transplantable, starvation-responsive acidic degron. (A) Left: scheme of Mup1 topology with N-and C-terminal ubiquitination sites and acidic patches targeted by Art1-Rsp5 and Art2-Rsp5, respectively, and the C-terminal phosphorylation sites of Mup1 that promote its starvation-induced endocytosis. Ubiquitinated lysines (K) shown in green and phosphorylated serines (S) and threonines (T) in red with numbers corresponding to amino acid positions in the Mup1 sequence. Right: Hxt3 as an Art4-Rsp5 dependent cargo during nitrogen starvation. (B) Live-cell fluorescence microscopy analysis of Mup1-GFP endocytosis in cells expressing MUP1-GFP (wild type (WT)) or MUP1 D549R,D551R,E554R, D555R-GFP. Cells were treated with 20 mg/ml L-methionine (+ Met) for 1.5 hr or starved (-N) for 6 hr after 24 hr exponential growth. (C) SDS PAGE and western blot analysis with the indicated antibodies of immunoprecipitated Mup1-GFP from cells expressing MUP1-GFP (WT) or MUP1 D549R,D551R, E554R,D555R-GFP starved for the indicated times after 24 hr of exponential growth. Equal amounts of immunoprecipitated Mup1-GFP were loaded to compare the extent of ubiquitination. (D) Amino acid sequence alignment of the C-terminal acidic patches of Mup1, Can1, Lyp1, Tat2 and Ina1. The boxes indicate acidic residues (red) and phosphorylation sites (blue), which are required for Art2-dependent starvation-induced endocytosis. Red letters illustrate further acidic residues. (E) Live-cell fluorescence microscopy analysis of wild type (WT), art4D and art2D art4D cells expressing HXT3-GFP (top), HXT3-MUP1-C-GFP (second row), HXT3-MUP1-C K567,572R-GFP (third row) or HXT3-MUP1-C D549R,D551R,E554R,D555R-GFP (bottom). Cells were starved (-N) for 6 hr after 24 hr exponential growth. Scale bars = 5 mm. See also Figure 6-figure supplements 1 and 2. The online version of this article includes the following figure supplement(s) for figure 6:  ART2 (art2D art4D) blocked starvation-induced endocytosis and demonstrated that Art2 now mediated the degradation of Hxt3-Mup1-C-GFP in art4D mutants. The degradation of Hxt3-Mup1-C-GFP and the concomitant accumulation of free GFP were further examined by SDS-PAGE and WB analysis from total cell extracts ( Figure 6-figure supplement 2D, E). This analysis showed that the majority of the full-length Hxt3-Mup1-C-GFP chimera was degraded in WT cells, art2D and art4D mutants during starvation ( Figure 6-figure supplement 2D, lanes 2, 4, 8), but only poorly in art2D art4D double mutants ( Figure 6-figure supplement 2D, lane 6, Figure 6-figure supplement 2E).
Taken together, these results demonstrate that the C-terminus of Mup1 (aa 545-574) encodes a portable acidic sorting signal that can be recognized by Art2 and directs Rsp5 to ubiquitinate specifically two proximal lysine residues to promote starvation-induced endocytosis.

A basic patch of Art2 is required for starvation-induced degradation of Mup1
After having defined that the C-terminus of Mup1 (and possibly also the C-terminus of further Art2targets) provides a degron sequence for Art2-Rsp5 complexes, we addressed how it could be recognized. Upon inspection of the predicted arrestin domain in Art2, we noted a stretch of positively charged residues within the arrestin-C domain ( Figure 7A). Converting these basic residues into an acidic patch (Art2 K664D,R665D,R666D,K667D ) abolished starvation-induced endocytosis of Mup1 ( Figure 7B). Western blot analysis of total cell lysates showed that the Art2 basic patch mutant protein was expressed at similar levels as WT Art2 and was also upregulated after 3 hr of starvation (Figure 7-figure supplement 1A, lane 6). In addition, the Art2 basic patch mutant also impaired, at least partially, starvation-induced endocytosis of Can1, Lyp1 and Ina1, while the endocytosis of Tat2 was independent of the basic patch ( Figure 7B, Figure 7-figure supplement 1B,C).
Overall, it seemed that Art2 employed a positively charged region in its arrestin fold to recognize C-terminal acidic patches in at least three different AATs (Mup1, Can1, Lyp1) and thus mediate their endocytosis in starvation conditions.

Discussion
We have made progress toward understanding how yeast cells selectively re-configure their repertoire of transporters at the PM in response to their nutritional status. The model in Figure 7C provides the conceptual framework for the mutually exclusive but complementary action of Art1-Rsp5 and Art2-Rsp5 ubiquitin ligase complexes for mediating the selective endocytosis of the methionine transporter Mup1 in response to changes in amino acid availability. Similar concepts may apply for the arginine transporter Can1, the lysine transporter Lyp1 and the tryptophan and tyrosine transporter Tat2. Based on our results and work from others, the endocytic down-regulation of these AATs can be described by the following set of rules: (1) Reciprocal (de-)activation of the aÀarrestins Art1 and Art2. (2) Art1 or Art2 each recognize different acidic sorting signals on their client AATs that may require additional phosphorylation. (3) To do so, they employ basic patches in their extended arrestin C-domains and defined PY motifs to orient Rsp5 with high specificity toward proximal lysine residues. These rules satisfy the plasticity required for different aÀarrestin and AAT interactions that drive exclusive or relatively broad substrate specificity depending on the metabolic context.
While both Art1 and Art2 lead to the degradation of AATs, they answer to distinct metabolic cues and are thus wired into distinct signaling pathways. Activation of Art1 by amino acid influx requires the coordinated interplay of TORC1 signaling to inactivate Npr1 (a kinase that negatively regulates Art1) and the action of phosphatases (Gournas et al., 2017;Lee et al., 2019;MacGurn et al., 2011;Tumolo et al., 2020). In response to amino acid limitation, TORC1 is no longer active. This will activate Npr1 to phosphorylate Art1, thereby inactivating it. At the same time, the lack of amino acids will activate the eIF2a kinase Gcn2. Gcn2 will phosphorylate eIF2a, which leads to the global down-regulation of translation, but enables specific translation of the transcription factor Gcn4 (Hinnebusch, 2005). Gcn4 then induces transcription of genes required for amino A B 1 8 6 2 3 8 3 1 9 3 8 8 4 4 2 4 7 8 5 5 4 7 0 0 8 2 3 8 7 7 N C acid biosynthesis and of ART2, which causes an increase in Art2 protein levels and thus formation of Art2-Rsp5 complexes. This appears as primary means to activate Art2, since unscheduled increase in Art2 protein levels was sufficient to drive Art2-dependent nutrient transporter endocytosis already in cells growing under rich conditions. When amino acids become available again, the system can efficiently reset. TORC1 is reactivated resulting in Art1 reactivation. Conversely, Gcn4 will become instable and rapidly degraded by the ubiquitin proteasome system (Kornitzer et al., 1994;Meimoun et al., 2000;Irniger and Braus, 2003), and thus, the transcription of ART2 will cease. Interestingly, two de-ubiquitinating enzymes (Ubp2, Ubp15) de-ubiquitinate Art2 to influence its protein stability (Ho et al., 2017;Kee et al., 2006). Inhibiting their activity could provide additional control to repress Art2-dependent endocytosis in cells growing under rich conditions. Our screen identified also two de-ubiquinating enzymes, Doa4 and Ubp6 to be specifically required for starvation-induced endocytosis of Mup1. They could act directly on Art2 or Mup1 or help to maintain homeostasis of the ubiquitin pool during starvation.
Art2 is subject to extensive post-translational modification, including ubiquitination and phosphorylation. Database searches and our own proteomic experiments identified 68 phosphorylation sites and 20 ubiquitination sites in Art2 (data not shown) (Swaney et al., 2013;Albuquerque et al., 2008;Holt et al., 2009). How these modifications help to control the activity of Art2 remains a complex and open questions. Several arrestins were found to be phospho-inhibited in specific conditions (MacGurn et al., 2011;Becuwe et al., 2012b;O'Donnell et al., 2013;Hovsepian et al., 2017;Merhi and André, 2012;Llopis-Torregrosa et al., 2016), the common molecular basis of which is unknown. An exciting hypothesis would be that a-arrestin hyper-phosphorylation would add negative charges, and thereby prevent the recognition of acidic patches on transporters through electrostatic repulsion. Interestingly, our screen identified the pleitropic type 2A-related serine-threonine phosphatase Sit4 as a class one hit. Hence, Sit4 may be linked directly or indirectly to de-phosphorylation of Art2 and controlling its activity as reported recently for the Art2-dependent regulation of vitamin B1 transporters (Savocco et al., 2019).
Through the complementary activation of Art1 and Art2, cells can coordinate amino acid uptake through at least four high-affinity amino acid transporters with amino acid availability. The regulation of hexose transporters by glucose availability appears to be conceptually related, with distinct aÀarrestin-Rsp5 complexes in charge of down-regulating the same transporters at various glucose concentrations with distinct mechanisms and kinetics (Hovsepian et al., 2017;Nikko and Pelham, 2009). In particular, the endocytosis of high-affinity hexose transporters during glucose starvation involves Art8, the closest paralogue of Art2, whose expression is also controlled by nutrient-regulated transcription (Hovsepian et al., 2017). Altogether, a picture emerges in which the transcriptional control of aÀarrestin expression by nutrient-signaling pathways is critical to cope with nutrient depletion. This becomes part of the complex regulation of the ART-Rsp5 ubiquitin ligase network.
Our work also extends on previous findings regarding the determinants of aÀarrestin/transporter interaction, indicating communalities between starvation-and substrate-induced endocytosis. Art1-Rsp5 and Art2-Rsp5 complexes both recognize specific acidic sequences on Mup1 ( Figure 7C). Activated Art1-Rsp5 complexes recognize an acidic stretch in the N-terminus of Mup1 and Can1, close to the first transmembrane domain. The exposure of these N-terminal acidic patches is linked to substrate transport and conformational transitions in the transporters from the outward-open to the inward-open conformation, which in Mup1 also includes the so-called 'C-plug' (aa 520-543) (Busto et al., 2018;Guiney et al., 2016;Gournas et al., 2017). This conformational switch drives lateral re-localization of Mup1 and Can1 into a disperse PM compartment, where they are ubiquitinated by Art1-Rsp5 (Gournas et al., 2018;Busto et al., 2018). Art2 recognizes specifically an acidic patch in the C-terminal tail of Mup1, and thereby directs Rsp5 to ubiquitinate two juxtaposed C-terminal lysine residues. The C-plug is very close to the C-terminal acidic patch but is not part of the C-terminal Mup1 degron. We speculate that in the absence of nutrients AATs will spend more time in the outward open state with the C-Plug in place. In this state, activated Art2-Rsp5 complexes can still engage the C-terminal acidic patches. Hence, toggling Art1/Art2 activation in combination with accessibility of N-or C-terminal acidic sorting signals in AATs, in part regulated by their conformational state, must fall together to allow selective endocytosis.
An additional layer of regulation for endocytosis is provided by phosphorylation of AATs close to the acidic sorting signal. At the moment, we can only speculate about the kinase responsible for the phosphorylation of the C-terminal serine or threonine sites of Mup1. Perhaps, constitutive PM-associated kinases such as the yeast casein kinase one pair (Yck1/2) are involved, which are known to recognize rather acidic target sequences and to regulate endocytosis (Hicke et al., 1998;Paiva et al., 2009;Nikko et al., 2008;Marchal et al., 2002).
a-Arrestins lack the polar core in the arrestin domain that is used for cargo interactions in barrestins (Aubry et al., 2009;Polekhina et al., 2013). Instead Art1-Rsp5 and Art2-Rsp5 complexes each use a basic region in their arrestin C-domain to detect the acidic sorting signal in their client AATs. Studies on the interaction between GPCRs and b-arrestins revealed a multimodal network of flexible interactions: The N-domain of b-arrestin interacts with phosphorylated regions of the GPCR, their finger loop inserts into the transmembrane domain bundle of the GPCR and loops at the C-terminal edge of b-arrestin engage the membrane (Staus et al., 2020;Huang et al., 2020). Perhaps, a similar concept also holds true for a-arrestins. This is not unlikely given that their arrestin fold appears to be interspersed with disordered loops and very long, probably unstructured N-and/or C-terminal tails, some of which participate in cargo recognition or membrane interactions (Baile et al., 2019).
Despite the possible plasticity in substrate interactions, the selectivity of Art1-Rsp5 and Art2-Rsp5 complexes in ubiquitinating lysine residues proximal to the acidic patches of Mup1 is remarkable. Mup1 has 19 lysine residues at the cytoplasmic side: four at the N-terminal tail, six in the C-terminal tail and nine in the intracellular loops of the pore domain. Yet, Art1-Rsp5 complexes only ubiquitinate K27 and K28, whereas Art2-Rsp5 complexes only ubiquitinate K567 and K572. Even in the Hxt3-Mup1-C chimeric protein, Art2-Rsp5 complexes ubiquitinated only the lysine residues close to the acidic patch, despite six further lysine residues in the directly adjacent C-terminal tail of Hxt3. How is this possible? We speculate that these two a-arrestin-Rsp5 complexes orient the HECT domain of Rsp5 with high precision toward the lysine residues that are spatially close to the acidic patches. Once ubiquitinated, the AAT can engage the endocytic machinery to be removed from the PM.
In conclusion, Art1-Rsp5 complexes act rapidly to prevent the accumulation of excess amino acids, whereas the Art2-Rsp5 complexes help to degrade idle high-affinity amino acid transporters over longer periods of starvation to recycle their amino acid content. Starvation-induced endocytosis and the subsequent degradation of membrane proteins is required to maintain intracellular amino acid homeostasis (Müller et al., 2015;Jones et al., 2012). As such, it is well suited that Art2 activity and thus starvation-induced endocytosis of AATs is co-regulated and coordinated with de novo amino acid biosynthesis via the GAAC pathway. The down-regulation of AATs together with glucose transporters and further PM proteins could also free up domains at the PM that are populated by selective nutrient transporters (Spira et al., 2012;Grossmann et al., 2008) for transporters with broader substrate specificity such as the general amino acid permease Gap1 and the ammonium transporter Mep2, which are strongly up-regulated during starvation. Hence starvation-induced endocytosis could prepare cells -anticipatory -for non-selective nutrient acquisition, as soon as nutrients become available again. Consistent with this notion and a role of Art2 during starvation, yeast cells with mutations in ART2 show an altered adaptation to repetitive cycles of starvation (Wloch-Salamon et al., 2017). However, at the moment, it is not clear how these physiological defects are linked to the role of Art2 in starvation-induced endocytosis. Hence, despite the emergence of regulatory mechanisms, we are still far from understanding how each individual ART-Rsp5 complex controls nutrient acquisition and how this process translates into fine-tuning of metabolism, cell growth and survival. Even more unclear is how the activities of the entire ART-Rsp5 network are harmonized to achieve a coordinated cellular response to nutrient availability. Since a more global disruption of the ART-Rsp5 network causes proteotoxic stress with strong growth defects (Nikko and Pelham, 2009;Zhao et al., 2013;Hein et al., 1995), addressing these important questions will require new approaches.
Altogether, our results provide a better understanding of the molecular mechanisms that couple metabolic signaling and nutrient availability to nutrient transporter endocytosis in budding yeast. This might provide basic insights and principles for the vastly more complex regulation of the nutrient transporter repertoire in human cells under normal as well as pathophysiological conditions.

Yeast strains, media and growth conditions
Yeast strains used for the microscopy screen for starvation-responsive endocytosis cargoes were mainly derived from the Yeast C-terminal GFP Collection (Huh et al., 2003) with addition of further C-terminally-tagged transporters in BY4741 (MATa his3D1 leu2D0 met15D0 ura3D0) and SEY6210  strain background (Supplementary file 1). The flow cytometry screen for genes affecting the starvation-induced endocytosis of Mup1-pHluorin was performed using the non-essential gene deletion strain collection purchased from Open Biosystems (BY4742: MATa his3D1 leu2D0 lys2D0 ura3D0) transformed with pRS416 expressing Mup1-pHluorin. For all other experiments, the SEY6210 parental strain was used (see key resources table).
Yeast cells were cultivated in rich medium and kept for at least 24 hr in exponential growth phase by dilution into fresh medium before onset of all experiments at mid-log phase. All cultivations were done at 26˚C. Substrate-induced endocytosis of Mup1 was triggered by the addition of 20 mg/L methionine (#9359.1, Roth, Germany) and, unless otherwise stated, analyzed 1.5 hr after the treatment. For starvation experiments, exponentially growing cells (0.4-0.6 OD 600 /ml) were washed twice with and resuspended in starvation medium (1 OD 600nm /ml) and incubated for the indicated times.

Genetic modifications and cloning
Genetic modifications were performed by PCR and/or homologous recombination using standard techniques. All chromosomal tags were introduced at the C-terminus of the target ORFs to preserve the endogenous five prime regulatory sequences. Chromosomally modified yeast strains were analyzed by genotyping PCR and/or DNA-sequencing. Plasmid-expressed genes including their endogenous or heterologous promoters were amplified from yeast genomic DNA and cloned into centromeric vectors (pRS series). All plasmids were analyzed by DNA-sequencing. Standard techniques were used for yeast transformation, mating and tetrad analysis or haploid selection. Yeast stains and plasmids are listed in the key resources table and primer in Supplementary file 4.

Flow cytometry screen
The yeast non-essential knock out collection (YSC1054; Open Biosystems) was transformed in 96well format using a standard lithium acetate/PEG-4000 yeast transformation protocol with a pRS416 plasmid encoding Mup1-pHluorin. Transformants were selected on agar plates, inoculated into 160 ml of rich medium (YNB medium lacking uracil as specified above containing 1.5x of all amino acids) and incubated for 14 hr shaking (180 rpm) at 26˚in 96-well plates (#83.3924, Sarstedt, Germany). Each plate also contained the WT BY4742 parental strain and art2D as negative and positive controls. 60 ml of the overnight cultures were transferred to 96-deep-well plates (#951033405, 96/2000 ml, white border, Eppendorf, Germany) containing 600 ml rich medium and further incubated for 5 hr. At this point, 120 ml of culture referred to as 'exponentially growing' were transferred to standard 96-well plates and analyzed by flow cytometry. The remaining culture was harvested by centrifugation (1807 g; 3 min), the medium was aspirated, cells were washed twice with 600 ml starvation medium and recovered by centrifugation. Subsequently, cells were resuspended in 600 ml starvation medium and incubated shaking (180 rpm) for 18-22 hr. 200 ml nitrogen-starved culture were transferred to standard 96-well plates and analyzed by flow cytometry. All pipetting steps were performed using the MEA Robotic System (PhyNexus, USA). Flow cytometry was performed in 96-well format using the Guava easyCyte 8HT-System (Sr.No. 6735128143), EMD Millipore, Merck, Germany) with the following settings: Energy GRN: 72 plus (YEL = 8; RED = 8; NIR = 8; RED2 = 8; NIR2 = 8), green channel, forward scatter FSC = 14; sideward scatter SSC = 28; 15.000-20.000 counts/sample, acquire: 40 s, 3 s mix prior to acquisition. GuavaSoft 2.7 software was used for data analysis. The positive/negative cut-off was set for each plate empirically at the intercept of the log/ starvation histograms of the WT and art2D controls (art2D, which emerged as a well-reproducible hit early in the screen, was included as a negative control in all further plates). All potential hits were reexamined by fluorescence microscopy. To this end, at least 100 starved cells were analyzed by fluorescence microscopy after starvation and the percentage of cells showing a degradation-deficient phenotype (Mup1-pHluorin at the plasma membrane, in small cytosolic objects, class E-like objects or small objects within vacuoles) of the total number of cells counted was calculated (Supplementary file 2). Strains with more than 45% cells with retained fluorescence after at least 18 hr of starvation were considered as hits. For a stringent final selection, we compared those hits to the original flow cytometry screen and finally only considered those in which at least once more the 30% Mup1-pHluorin fluorescence was also retained after starvation in the flow cytometry screen. In addition, most hits were also scored for methionine-induced endocytosis of Mup1-pHluorin. Hits were considered starvation-specific if the fluorescence was quenched in more than 67% of cells after 90 min of methionine treatment (20 mg/ml).

Gene ontology analysis
Gene Ontology (GO) enrichment analysis (Harris et al., 2004) was performed using 128 genes listed in Supplementary file 2. They were mapped against the generic GO-Slim: process, generic GO-Slim: component and macromolecular complex terms: component GO sets using the GoSLIM mapper of the Saccharomyces genome database. We calculated the ratio of the observed (cluster frequency) vs. the expected number of genes (genome frequency) associated with the GO term, referred to as enrichment over genome (McClellan et al., 2007;Schmidt et al., 2019). Only GO terms with more than one associated gene were reported. The full analysis is presented in Supplementary file 3.

Fluorescence live cell wide-field microscopy
For microscopy, cells were concentrated by centrifugation and directly mounted onto glass slides. Live cell wide-field fluorescence microscopy was carried out using a Zeiss Axio Imager M1 equipped with a sola light engine LED light source (Lumencore), a 100x oil immersion objective (NA 1.45) standard GFP and mCherry fluorescent filters, a SPOT Xplorer CCD camera, and Visitron VisiView software (version 2.1.4). The brightness and contrast of the images in the figures were linearly adjusted using Photoshop CS5 (Adobe Version 12.0.4 Â 64).

Western blot quantification
WB signals were quantified by densitometry using ImageJ2 (Version 2.0.0-rc49/1.51 hr; RRID:SCR_ 003070) (Rueden et al., 2017), quantifications were exported to Microsoft Excel (Version 16.16.2; RRID:SCR_016137), normalized to the respective Pgk1 loading controls, and presented as mean ± standard deviation from at least three independent experiments. WT in exponential growth phase was set to 1. For the quantification of GFP-tag processing, full-length and clipped GFP WB signals were quantified from the same exposure and added to calculate the total GFP signal for each lane. The signals for full-length and clipped GFP were then reported as the percentage from the total GFP signal.

Mass spectrometry sample preparation and analysis
Coomassie-stained gel bands were excised from SDS-PAGE gels, reduced with dithiothreitol, alkylated with iodoacetamide and digested with trypsin (Promega) as previously described (Faserl et al., 2019). Tryptic digest were analyzed using an UltiMate 3000 RSCLnano-HPLC system coupled to a Q Exactive HF mass spectrometer (both Thermo Scientific, Bremen, Germany) equipped with a Nanospray Flex ionization source. The peptides were separated on a homemade fritless fused-silica micro-capillary column (75 mm i.d. x 280 mm o.d. x 10 cm length) packed with 3.0 mm reversed-phase C18 material. Solvents for HPLC were 0.1% formic acid (solvent A) and 0.1% formic acid in 85% acetonitrile (solvent B). The gradient profile was as follows: 0-4 min, 4% B; 4-57 min, 4-35% B; 57-62 min, 35-100% B, and 62-67 min, 100% B. The flow rate was 250 nL/min. The Q Exactive HF mass spectrometer was operating in the data dependent mode selecting the top 20 most abundant isotope patterns with charge >1 from the survey scan with an isolation window of 1.6 mass-to-charge ratio (m/z). Survey full scan MS spectra were acquired from 300 to 1750 m/z at a resolution of 60,000 with a maximum injection time (IT) of 120 ms, and automatic gain control (AGC) target 1e6. The selected isotope patterns were fragmented by higher-energy collisional dissociation with normalized collision energy of 28 at a resolution of 30,000 with a maximum IT of 120 ms, and AGC target 5e5.
Data analysis was performed using Proteome Discoverer 4.1 (Thermo Scientific) with search engine Sequest. The raw files were searched against yeast database (orf_trans_all) with sequence of Mup1-GFP added. Precursor and fragment mass tolerance was set to 10 ppm and 0.02 Da, respectively, and up to two missed cleavages were allowed. Carbamidomethylation of cysteine was set as static modification. Oxidation of methionine, ubiquitination of lysine, and phosphorylation of serine threonine, and tyrosine were set as variable modifications. Peptide identifications were filtered at 1% false discovery rate.

RNA isolation and quantitative PCR (RT-qPCR)
Exponentially growing or starved cells (40 OD 600nm ) were harvested by centrifugation and immediately frozen in lq. N 2 . Cell pellets were lysed with 1 mm glass beads in a FastPrep-24 homogenizer (MP biosciences) in Qiagen RLT buffer, and RNA was extracted using the RNeasy Mini Kit (#74104, Qiagen, Germany). Yield and purity were determined photometrically. cDNA was prepared from 5 mg DNAse I-treated RNA using the RevertAid First Strand cDNA Synthesis Kit (#K1622, Thermo Fisher, USA) with oligo-dT primer according to the standard protocol. qPCR was performed in 10 ml scale with 4 ml of cDNA, 5 ml TaqMan Gene Expression Master Mix (#4369016, Thermo Fisher, USA) and 0.5 ml TaqMan probe on a PikoReal 96 Real-Time PCR System (Thermo Fisher, USA) with 7 min initial denaturation (95˚C) and 40 cycles of 5 s 95˚C and 30 s 60˚C. TaqMan gene expression assays were from Thermo Fisher (PGK1: Sc04104844_s1; ECM21/ART2: Sc04099967_s1). All probes and primer anneal within coding sequences. Each RT-qPCR analysis was done from two to three independent biological samples in 3-4 technical replicates. Data were analyzed with the PikoReal software (version 2.2; Thermo Fisher) with manual threshold adjustment, and relative mRNA abundance was calculated in Microsoft Excel using the DDC T method. Statistical comparisons were calculated using the Student t-test. . Supplementary file 2. Flow cytometry and fluorescence microscopy-based screen for genes specifically involved in starvation-induced endocytosis.
. Supplementary file 3. Gene ontology term analysis of 128 genes required for starvation-induced endocytosis of Mup1-pHluorin.
. Supplementary file 4. Primer for cloning and PCR-based genetic modifications.
. Transparent reporting form

Data availability
All data generated or analysed during this study are included in the manuscript and supporting files.