Drug uptake, lipid rafts and vesicle trafficking modulate resistance to an anticancer lysophosphatidylcholine analogue in yeast

Background: The antitumor lipid edelfosine proteins. Results: Impairing vesicular trafficking to the plasma membrane alterations without affecting internalization of the

(ATL), kills yeast cells and selectively kills several cancer cell types. To gain insight into its mechanism of action, we performed chemogenomic screens in the Saccharomyces cerevisiae gene-deletion strain collection, identifying edelfosineresistant mutants. LEM3, AGP2 and DOC1 genes were required for drug uptake. Edelfosine displaced the essential proton pump Pma1p from rafts, inducing its internalization into the vacuole. Additional ATLs, including miltefosine and perifosine also displaced Pma1p from rafts to the vacuole, suggesting that this process is a major hallmark of ATL cytotoxicity in yeast. Radioactive and synthetic fluorescent edelfosine analogues accumulated in yeast plasma membrane rafts and subsequently the endoplasmic reticulum. Although both edelfosine and Pma1p were initially located at membrane rafts, internalization of the drug towards endoplasmic reticulum and Pma1p to the vacuole followed different routes. Drug internalization was not dependent on endocytosis and was not critical for yeast cytotoxicity. However, mutants affecting endocytosis, vesicle sorting or trafficking to the vacuole including the retromer and ESCRT complexes, prevented Pma1p internalization and were edelfosineresistant. Our data suggest that edelfosineinduced cytotoxicity involves raft reorganization, and retromer-and ESCRTmediated vesicular transport and degradation of essential raft proteins leading to cell death. Cytotoxicity of ATLs is mainly dependent on the changes they induce in plasma membrane raft-located proteins that lead to their internalization and subsequent degradation. Edelfosine toxicity can be circumvented by inactivating genes that then result in the recycling of internalized cell-surface proteins back to the plasma membrane.
A major problem in the treatment of cancer is the development of drug resistance. Understanding the molecular mechanisms involved in the generation of resistant phenotypes is critical in trying to overcome drug resistance and to improve current therapy. The antitumor activity of another ATL, perifosine (octadecyl (1,1-dimethylpiperidino-4-yl)-phosphate, D-21266; Fig. 1), is being tested in ongoing clinical trials; and miltefosine ( Fig. 1) is used in the clinic as a palliative treatment for cutaneous metastases from breast cancer (16). Gaining insight into the mechanisms of action of ATLs, as well as in the identification of the molecular processes that lead to drug resistant and sensitive phenotypes, is of major importance for the clinical development of these synthetic compounds.
Edelfosine is cytotoxic against the budding yeast Saccharomyces cerevisiae (17), although the underlying mechanism is not fully understood. The yeast flippase subunit Lem3p has been shown to be essential for drug uptake (18), and we have previously found that edelfosine displaces the essential H + -pump Pma1p from plasma membrane microdomains, resulting in its degradation (17). S. cerevisiae is a useful tool for identifying human drug targets and modulators of drug activity (19). Comparison of the yeast and human genomes has revealed that 30% of known genes involved in human disease have yeast orthologs, and a large number of yeast genes exhibit a link to human disease genes (20,21). In this study, and the accompanying paper (Czyz et al.), we performed a chemical-genetic screen versus the deletion strain set and identified genes that led to either increased edelfosine resistance or sensitivity. We also synthesized a set of fluorescent edelfosine analogues that retained the properties of the parent compound and allowed its subcellular localization to be determined. By combining genetic and functional approaches, we unveiled a novel mechanism of action in the killing of S. cerevisiae that may be extrapolative to tumor cells. In addition to the identification of three genes required for edelfosine uptake our data indicate that edelfosine accumulates in lipid rafts and ER in S. cerevisiae, and mediates its cytotoxic activity by affecting raft protein composition which eventually leads to intracellular acidification (see accompanying paper, Czyz et al.). This alteration in raft composition seems to be conserved within the ATL family, as it was also observed for miltefosine and perifosine. The selective displacement of crucial proteins from lipid rafts results in their endosome-mediated transport to the vacuole, where they undergo degradation. The end result is an alteration in intracellular pH that eventually leads to cell death. The yeast genome-wide screen and subsequent experiments performed here revealed that vesicular trafficking is a critical process mediating edelfosine resistance, with mutants that result in increased recycling between endosomal compartments and the plasma membrane being resistant to edelfosine cytoxocity. This is a highly conserved process and thus this mode of resistance may be extrapolatable from yeast to tumor cells.

Yeast strains, plasmids and growth conditions
The collection of nonessential haploid yeast deletion strains derived from parental strain BY4741 (MAT a his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) was obtained from Euroscarf. All other strains used in this study are listed in Table 1. The endoplasmic reticulum marker encoded by ELO3 was subcloned into the pUG35 (URA3) vector (22) using EcoR1 and Sal1 restriction enzyme sites, in frame with green fluorescent protein (GFP) coding sequence at its 3' end. The YCplac111-SEC63-GFP (SEC63-GFP under control of SEC63 promoter into CEN/LEU2 vector) was kindly provided by Symeon Siniossoglou (University of Cambridge, UK) (23). The centromeric plasmid used to express Pma1p-GFP (in pRS316, URA3) was kindly provided by Annick M. Breton (Institut de Biochimie et Génétique Cellulaires, France) (24). The centromeric vector YCplac111 (Stratagene) was used to express the VPS29 and VPS35 genes in their respective vps29Δ and vps35Δ knockout strains. Red fluorescent protein (RFP) tagged marker strains (25) for the indicated organelles and cellular structures (kindly provided by Peter Arvidson, Harvard, USA) were used to examine the subcellular location of fluorescent edelfosine.
Yeast strains were grown on standard rich medium YPD (1% yeast extract, 2% Bacto-peptone, 2% glucose), synthetic minimal medium (SD; 0.17% yeast nitrogen base without amino acids, 2% glucose and supplements according to the requirements of the strains), and synthetic complete medium (SC; SD medium with 0.079% complete supplement mixture (ForMedium, Norwich, UK). Agar (2%) was added for solid plates. Unless otherwise indicated all yeast cultures were incubated at 28°C.
Growth of cells treated with edelfosine was monitored by optical density at a wavelength of 595 nm (OD 595 ). A preculture was grown overnight in YPD to log-phase. A volume of 2 ml of fresh YPD was then inoculated to an OD 595 of 0.3 and immediately treated with edelfosine. Cells were incubated for 24 h and sample aliquots were taken at the designated times to measure absorbance. Edelfosine was routinely used at 15 μM for treatment in liquid medium unless otherwise indicated.

Large-scale edelfosine resistance screens
The haploid deletion collection in the BY4741 background was screened for resistance to edelfosine in SC liquid medium. Strains were pinned from 96-well frozen stock plates by using a stainless steel 96-pin replicator (Nalgene Nunc International) into 96-well plates containing 150 μl of liquid YPD medium supplemented with G418 (150 mg/l; Gibco-BRL). The plates were incubated at 28°C for 3 days and then pin replicated onto liquid SC in 96-well plates containing either no drug or 60 μM edefosine. Plates were incubated at 28°C and growth was quantitatively scored every 24 h over a period of 7 days by optical density readings at OD 595 using a microplate spectrophotometer (model 550; Bio-Rad Laboratories). Edelfosine resistance was scored on the basis of the relative growth of each mutant against that of the wild-type strain in the same plate and with that of each mutant strain in plates where no drug was added. Putative edelfosine-resistant strains identified during the screening of the yeast knockout collection were further retested at least in duplicate under the same conditions described for the screen. Gene deletion strains that consistently grew at or above a 4-fold higher rate than wild-type cells in the presence of 60 μM edelfosine for 72 h in five independent screening assays were considered drug-resistant. These cutoff conditions led to reliable and highly reproducible results. Selected resistant strains were retested again from the original haploid deletion collection stock for drug resistance, using 60 μM edelfosine, to minimize those arising due to second site mutations accumulating in other regions of the genome during drug exposure.
Growth of resistant strains was further confirmed in solid medium. Cultures grown in 96-well plates were resuspended using a 96well plate shaker (Microtec; Infors AG), and four serial 10-fold dilutions in water were done. Five-microliter aliquots of the diluted cultures were spotted using a replica plater for a 96-well plate (Sigma-Aldrich) to OmniTray plates (Nalgene Nunc International) containing SC solid medium, to which edelfosine was added once the medium had cooled to ∼50ºC. The plates were incubated at 28°C for 3 days and analyzed for cell growth every 24 h.
Gene classification was done subjectively with reliance on the Saccharomyces Genome Database (SGD), the Yeast Proteome Database, the Comprehensive Yeast Genome Database at MIPS and the literature.

Edelfosine uptake in yeast mutants
[ 3 H]edelfosine uptake was analyzed in the 91 yeast mutants that showed the highest drug resistance (Supplementary Table S1), as described (7) with slight modifications. [ 3 H]edelfosine (specific activity, 42 Ci/mmol) was synthesized by Amersham Buchler, Braunschweig, Germany. Briefly, the mutants were incubated at 28 o C to late exponential phase in 1 ml of YPD in a 96-well microtiter plate. The OD 595nm of each well was measured to ensure the same amount of cells from each mutant was present per well. [ 3 H]edelfosine (60 µM and 2 x 10 4 cpm per well) was added and the cells were incubated at 28 o C at 800 rpm for 1 h. The non-incorporated [ 3 H]edelfosine was washed-out from the culture by exhaustive washing (5 times) with 1 ml of 3% bovine serum albumin (BSA) in phosphate buffered saline (PBS). Washed cells were resuspended in 500 µl of PBS and mixed with 1.5 ml of scintillation liquid. The radioactivity incorporated into cells was measured in a Beckman liquid scintillation counter. The uptake assay was carried out three times (data for each mutant strain was calculated from three independent biological replicates). Deletion mutants with reduced edelfosine uptake were complemented with the corresponding wild type genes expressed from a centromeric plasmid (pRS416), and drug uptake was assayed as above.

Detergent-resistant membrane isolation and western blots
Yeast detergent-resistant membranes (DRM) were isolated as described (17) from total crude membranes. Yeast cells were grown to mid-log phase, and an amount of cells equivalent to 50 OD 595 units was untreated or incubated in the presence of 15 μM edelfosine for the indicated times. Cells were then washed three times with PBS and once with 1 M sorbitol containing 10 mM NaN 3 and 10 mM NaF. The cell pellet was stored at -20 °C. Cells were resuspended in 350 μl of TNE buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 5 mM EDTA, 20 μg/ml leupeptin, 20 μg/ml aprotinin, 1 mM phenylmethylsulfonyl fluoride, and protease inhibitor mixture (Roche)), and broken with the aid of Fastprep (four 15-s pulses, 5.5 speed) in the presence of glass beads at 4°C. Unbroken cells were removed by a 500 x g spin for 5 min at 4°C, and the resulting supernatant (total cell lysate) was centrifuged for 3 h at 4°C in an SW40 rotor at 35,000 rpm (218,000 x g) to pellet crude cellular membranes. The pellet was resuspended in Mes-buffered saline (MBS; 25 mM Mes, pH 6.5, 150 mM NaCl) including 60% sucrose, and Triton X-100 was added to a final concentration of 1%. After 1 h at 4°C, membranes were further disrupted with several strokes in a Potter-Elvehjem tissue grinder and 350 μl of the extracted membranes were mixed with 2.27 ml of 60% sucrose in MBS. This was placed at the bottom of an ultracentrifuge tube and overlaid with a sucrose step gradient from 5 to 50% sucrose in MBS (0.89 ml at 5%, 0.89 ml at 30%, 1.78 ml at 40%, 1.78 ml at 45%, and 3.54 ml at 50%). Sucrose gradients were centrifuged in an SW40 rotor at 38,000 rpm (257,000 x g) for 18 h at 4°C in a Beckman Optima LE-80K ultracentrifuge and 1-ml fractions were collected from the top of the gradient. For each fraction, 20 μl was subjected to SDS-PAGE, immunoblotting, and enhanced chemiluminescence detection.
For incorporation of edelfosine into DRMs, cells were incubated with 15 µM [ 3 H]edelfosine (5.3 x 10 5 dpm/mol) for the indicated times and subjected to lipid raft isolation as above.
DRM isolation from yeast treated with edelfosine (Inkeysa and Medmark), miltefosine (Cayman Chem) and perifosine (Aeterna Zentaris) was also performed as indicated in the accompanying paper (Czyz et al.) using Optiprep gradients. Densitometry was performed using Image J (Wayne Rasband, National Institutes of Health, USA). Experiments were done at least three times.
For western blot analysis, proteins were transferred to polyvinylidene difluoride membranes (PVDF) (GE Healthcare, Little Chalfont, Buckinghamshire, UK), and blots were incubated with antibodies to Pma1p and Gas1p (kind gifts of Ramón Serrano, Universidad Politécnica de Valencia, Spain, and Howard Riezman, University of Geneva, Switzerland, respectively), and subsequently with horseradish peroxidase-conjugated secondary antibodies followed by detection using enhanced chemiluminescence.

Edelfosine fluorescent analogues
Fluorescent analogues of edelfosine ( Fig. 1A) were synthesized in a way in which the physico-chemical properties and biological activity of the parent drug remained essentially unperturbed. We designed synthetic strategies that allowed precise control of the changes introduced in the structure and properties of the parent drug to produce de novo fluorescent bioactive analogues of the alkyl-lipid. Synthesis of the fluorescent analogues of edelfosine used in this study, namely containing the conjugated all-(E)-phenyltetraene or all-(E)phenyltrienyne blue-emitting chromophores (PTE-ET and PTRI-ET, respectively; Figure  1), and a 2-substituted borondifluorodipirromethene group (BODIPY or BDP) attached to and aligned with the alkyl chain of the ether-lipid (green-emitting Et-BDP-ET and Yn-BDP-ET; Figure 1) has been detailed elsewhere (4,29,30). Spectral features of the distinct fluorescent edelfosine analogues used in this study have been previously described (4,29,30). The filters used to examine the PTE-ET and PTRI-ET analogues were those for DAPI staining (BP 546 FT 580 LP 590). The filters to examine the BDP-ET analogues were those corresponding for GFP (41017 HQ 470/40X Q495LP HQ525/50).

Microscopy
Both Nomarski interference contrast, light, and fluorescence microscopy were used to visualize different subcellular organelles and proteins as well to examine the localization and effect on the cell of edelfosine. Images were obtained using a Carl Zeiss Axioplan 2 microscope. Images were acquired using Openlab software (PerkinElmer).
Drug localization was monitored with fluorescent analogues of edelfosine that were added directly to 1 ml of log-phase cultures in YPD. Cells were incubated in darkness for 30 min, washed, applied directly to the slide and immediately imaged. PTE-ET and PTRI-ET were added at a final concentration of 15 µM, and due to their proneness to undergo photobleaching, pictures were taken as soon as the samples were exposed to ultraviolet light. Et-BDP-ET and Yn-BDP-ET, which have a more intense and resilient fluorescence, were added at a final concentration of 5 µM and could be imaged normally. Images of other fluorescent markers were always taken after the analogues were imaged. For studies of drug internalization the soluble salt Lucifer yellow (LY, Sigma) and the lypophilic dye FM 4-64 (Biotium) were simultaneously used as endocytosis markers. Cells were preincubated for 1 h, at either 4ºC or 30ºC, in the presence of 2.5 mg/ml LY and 10 µM FM 4-64. Cells were washed, resuspended in YPD and subsequently incubated for 15 min (or 1 h for 0ºC) with 15 µM PTE-ET, while keeping the previous temperature, and imaged immediately afterwards.
Pma1p internalization was monitored using strains transformed with a GFP-tagged form. Various strains were transformed with either a vector carrying the fusion protein or an empty one. Cells were grown to early log phase (OD 595 =0.2) and preloaded with 10 μM FM4-64 for 45 min. Cells were washed, resuspended in the same volume of medium and treated with 15 µM edelfosine. After a 4 hr incubation cells were harvested and imaged.

Statistical analysis
Values are expressed as a mean ± SD with the number of experiments indicated.

Cytotoxic activity of edelfosine in S. cerevisiae
In order to set the conditions for chemogenomic screens, we first determined the minimal concentration necessary to inhibit growth of wild-type S. cerevisiae strains in liquid culture. Yeast lacking Lem3p (lem3∆), which have been reported to be resistant to edelfosine (18), were used as a positive control of growth inhibition. We also included yeast mutants lacking Dnf2p and Drs2p, two phospholipid translocases (flippases) involved in lipid and endocytic trafficking (31). Lem3p is an activator of Dnf2p and other members of the phospholipid translocase family. Dnf2p is a plasma membrane phospholipid and lysophospholipid translocase that can use edelfosine as a substrate, while Drs2p is a trans-Golgi localized phospholipid translocase that affects vesicular trafficking. Edelfosine inhibited wild type yeast growth at 15 µM, while the lem3Δ strain was resistant to edelfosine even at 70 µM (Fig. 2a). The dnf2Δ and drs2Δ strains showed an intermediate phenotype with growth inhibition at 35 µM and 17.5 µM, respectively, probably due to partial functional redundancy among them and additional members of this protein family (32).

Identification of edelfosine-resistant mutants in S. cerevisiae
Next, we performed large-scale chemical-genomic screens using a haploid collection of S. cerevisiae deletion mutants, individually deleted for non-essential genes (~4,800 genes) to test for enhanced resistance to edelfosine compared to wild-type cells. Strains that grew in liquid culture at a rate ≥4fold that of wild-type cells in the presence of 60 μM edelfosine for 72 h in five independent screening assays were considered drugresistant (Fig. 2b). These cutoff conditions led to highly reproducible results. From these screens we identified 262 genes whose inactivation conferred resistance to edelfosine in S. cerevisiae (Table 1, and Supplementary  Table S1). Functional analysis of these genes revealed the enrichment of statistically significant clusters of biological processes involved in edelfosine resistance. This analysis identified genes involved in vesicular transport (including ESCRT and retromer complexes), protein biosynthesis, and mitochondrial processes (Fig. 2c, Table 2, and  Supplementary Table S1). Table 2 shows some of the most relevant genes involved in edelfosine resistance regarding vesicular traffic and drug uptake. The resistant mutants were significantly enriched in endosome-and mitochondria-localized proteins. In this work we focused on the mechanism of resistance conferred by mutants in the endosomal trafficking pathways.

Mutant strains involved in edelfosine uptake
Because edelfosine must be incorporated by the target cell to exert its cytotoxic effect (4,5), we examined whether the edelfosine resistance of the selected strains was due to deficient drug uptake.
[ 3 H]edelfosine uptake was analyzed in the 91 yeast mutants that showed the highest drug resistance (Supplementary Table S1) and only lem3∆, agp2∆ and doc1∆ mutants resulted in drug uptake deficiency (Fig. 2d). This phenotype was reverted by transformation of each mutant with the corresponding wild-type gene (Fig. 2d). It has been previously demonstrated that inactivation of LEM3 blocked edelfosine uptake (18). Interestingly, our data indicated that the L-carnitine and bleomycin transporter Agp2p (33,34), and Doc1p, which is involved in ubiquitination of the anaphase promoting complex (APC), were also involved in edelfosine uptake (Fig. 2d).
These results indicate that resistance to edelfosine in the majority of mutants identified in our screen is not due to reduced uptake of the drug.

Fluorescent analogues of edelfosine and drug localization to the plasma membrane and ER
Because the uptake of edelfosine is critical for its cytotoxic action in yeast we examined its localization in S. cerevisiae. To do so we used a first generation of fluorescent edelfosine analogues, tagged with UVabsorbing all-(E)-phenyltetraene or all-(E)phenyltrienyne blue-emitting chromophores (PTE-ET and PTRI-ET, respectively) ( Fig. 1), and a second generation of fluorescent analogues where a 2-substituted borondifluorodipirromethene group (BODIPY or BDP) was attached to and aligned with the alkyl chain of the ether-lipid yielding greenemitting Et-BDP-ET and Yn-BDP-ET analogues ( Fig. 1) (4,29,30). Absorption and emission spectra in the visible wavelength range, enhanced fluorescence yield and higher resistance to photodegradation of the latter analogues improved the lateral resolution of living-cell imaging techniques and extended the recording time (30).
The above fluorescent analogues preserved the anticancer activity of the parent drug, PTE-ET showing the highest killing activity (4,15,30). Likewise, similar actions of the above fluorescent compounds were found in yeast, with PTE-ET being the most effective fluorescent analogue in inhibiting yeast growth (data not shown). In addition, similarly to what was observed with radioactive edelfosine (Fig. 2d), no uptake of any of the fluorescent edelfosine analogues was detected in lem3∆ mutants as compared to wild-type yeast (Fig. 3). These data suggest that the above fluorescent edelfosine analogues behave as bona fide analogues to analyze edelfosine localization in yeast. We found that all of the emitting edelfosine analogues yielded a similar localization pattern in yeast (Fig. 3) corresponding to a bag-like envelope surrounding the nucleus, (visualized by the nuclear marker Sik1p-RFP), which resembled the morphological traits of ER (Fig. 4). On these grounds, PTE-ET was used hereafter.
Using GFP-tagged markers for distinct subcellular organelles, we found that PTE-ET was mainly localized to the ER, as assessed by using Elo3p-GFP as an ER marker (Figs. 4, 5 and 6a). We found good colocalization between PTE-ET and Elo3p-GFP (Figs. 4, 5 and 6a), whereas no co-localization was observed with markers for other subcellular structures and compartments including Golgi (Anp1p, Chc1p), spindle pole (Spc42p), endosomes (Snf7p), cytoskeleton (Sac6p), peroxisomes (Pex3p) and lipid droplets (Erg6p) (data not shown). Further confirmation of the ER location of edelfosine was found by co-localization of PTE-ET fluorescent analogue with the ER marker Sec63p (Fig. 5). These data suggest that fluorescent edelfosine analogues are incorporated into the PM and transit to the ER.

ER localization of edelfosine is not dependent on endocytosis
One of the largest clusters of edelfosine-resistant mutants comprised strains defective in endocytosis. We examined the putative involvement of endocytosis in the ER localization of edelfosine. Inactivation of either END3 or END4 leads to defects in endocytosis (35). Our chemogenomic screen identified end3∆ as edelfosine resistant in addition to endocytosis-related mutants with inactivated MYO5, LDB17 or RHO4 genes (Table 1). However, PTE-ET was still ER localized in the end4 pep4∆ mutant and wildtype cells (incubated at 4ºC to slow down the rate of endocytosis) (Fig. 6b). In both cases, inhibition of endocytosis was confirmed by a decrease or block in the uptake of the soluble endocytosis marker Lucifer yellow and the lipophilic marker FM4-64 (Fig. 6b).
These results show that endocytosis does not play a role in the ER location of edelfosine. Furthermore, because the end4 pep4∆ mutant is edelfosine-resistant (17), drug localization to the ER is not critical for its cytotoxicity.

Pma1p internalization, lipid rafts and endocytosis in ATL-mediated cytotoxic activity
Because the above data suggested that inhibition of endocytosis prevented cell death, but was not crucial for edelfosine intracellular localization, we hypothesized that an essential step for edelfosine toxicity could be the internalization of certain membrane proteins by endocytosis. We have previously shown that edelfosine accumulates in plasma membrane lipid rafts in S. cerevisiae displacing Pma1p from these membrane microdomains resulting in Pma1p transit to the vacuole (17). In this regard, we found that incubation of yeast at 4ºC abrogated internalization of GFP-tagged Pma1p following edelfosine treatment, but not the intracellular location of edelfosine (Figs. 6b  and 6c).
We went on to analyze displacement of proteins from lipid rafts by edelfosine. The 100-kDa yeast lipid raft protein Pma1p was mainly located in fractions 4-6 of the sucrose gradient of untreated S. cerevisisae, thus identifying the raft-enriched fractions (Fig. 7a,  inset). As shown in Fig. 7a, we found that edelfosine was mainly located in non-raft fractions at early time points, but became enriched in rafts at longer incubation times. Interestingly, edelfosine accumulation in rafts was accompanied by Pma1p displacement from these membrane domains (Figs. 7a and 7b), which was followed by its internalization and degradation in the vacuole (17). However, the glycosylphosphatidylinositol (GPI) anchored protein Gas1p, another lipid raft protein, was not displaced from rafts (Fig.  7b). This Pma1p displacement from rafts was blocked in the end4 pep4∆ mutant (Fig. 7b), despite the fact that this mutant accumulated large amounts of edelfosine in lipid rafts (Fig.  7a) and drug internalization was not prevented (Fig. 6b). Altogether, these results suggest that edelfosine and Pma1p are delivered to the interior of the cell through distinct mechanisms. Thus, Pma1p internalization is mediated by endocytosis, whereas edelfosine internalization is endocytosis-independent. In addition, it is interesting to note that most of the edelfosine was visualized intracellularly in the ER after 3 h of incubation, while a significant amount of the drug was also found in lipid rafts (data not shown), suggesting the presence of rafts in the ER. This notion is further supported by the high proportion of edelfosine in rafts in end4 pep4∆ mutant, while being visualized in the ER by fluorescence microscopy (cf. Figs. 6b and 7a). Presumably, resistance in the end4 pep4∆ mutant comes from a downstream effect in the mechanism of drug toxicity; that is, uptake of the drug and its translocation to the ER continues, but Pma1p endocytosis/degradation is impaired.
We also found that, like edelfosine, the other clinically relevant ATLs miltefosine and perifosine were able to induce Pma1p displacement from lipid rafts and subsequent internalization into the vacuole in yeast, as assessed by detergent-resistant membrane isolation (Figs. 8a and 8b) and fluorescence microscopy (Fig. 8c). In contrast, treatment of yeast with non-ATL compounds including cycloheximide, brefeldin A or hygromycin B, did not promote Pma1p displacement from plasma membrane rafts (data not shown). Thus, Pma1p displacement from rafts seems to be a general feature of ATL action, and not a general process occurring under druginduced stress.

Involvement of vesicle-mediated transport in edelfosine resistance
A major cluster of edelfosine-resistant mutants included genes that mediated vesicular trafficking ( Table 2, Supplementary  Table S1, and Fig. 2c). A significant number of these genes are engaged in protein sorting at late endosomes, a process that seems to act as a major determinant for resistance to edelfosine.
Multiple pathways mediate recycling of internalized cell-surface proteins from endosomes back to the PM. Sorting of proteins at endosomes is a critical step for recycling. Sorting of proteins can be carried out by different protein complexes including sorting nexins, retromer complex and ESCRT complex (36). Mutants in all of these pathways were identified in our screen.
We found that deletion of 4 out of the 5 components (VPS35, VPS29, VPS26/PEP8 and VPS17) of the retromer, a highly conserved protein complex that mediates recycling to the PM as well as retrograde transport to the trans-Golgi network (TGN) (37), resulted in edelfosine resistance ( Table  2). Drug resistance was also observed in mutants for the sorting nexins SNX4 and SNX42 (Table 2), which participate in an alternate retrograde transport mechanism (38). Deletion of the proteases KEX1 or KEX2, involved in processing cargo proteins, also caused drug resistance ( Table 2). RAB7 has a role in retromer recruitment to endosomes in human cells (39), and deletion of its functional homolog in yeast YPT7 caused edelfosine resistance ( Table 2). Deletion of two components (VPS52 and VPS54) of the Golgi-Associated Retrograde Protein complex (GARP) (40) and the only nonessential member of the SNARE complex, TLG2, or the palmitoyltransferase SWF1 required for Tlg1p stability involved in vesicle fusion to the TGN, also led to edelfosine resistance (Table 2).
We examined whether retrograde transport was required for the localization of edelfosine to the ER. To this aim we used the two most edelfosine-resistant strains with deletion in retromer genes, namely vps29∆ and vps35∆. The involvement of the retromer in edelfosine cytotoxicity was confirmed by showing that vps29∆ and vps35∆ mutants restored the edelfosine-sensitivity phenotype when they were transformed with a centromeric plasmid containing their cognate VPS29 or VPS35 genes (Fig. 9). Next, we found that ER location of edelfosine was not affected in the vps29∆ and vps35∆ mutants (Fig. 10a). PTE-ET localized to the ER in the above deletion mutants, as assessed by using the ER-specific marker Elo3p-GFP, indicating that drug accumulation in the ER is independent of retrograde transport. However, Pma1p remained largely at the plasma membrane in these retromer mutants following exposure to edelfosine, as shown by fluorescence microscopy with a Pma1p-GFP fusion protein (Fig. 10b). These data further support that a critical step in edelfosine cytotoxicity is displacement of Pma1p from lipid rafts for subsequent degradation by the vacuole, and not accumulation of edelfosine in the ER.
In addition, our chemogenomic screen identified 11 of the 18 known members of ESCRT complexes as genes whose inactivation results in edelfosine resistance (complex 0: VPS27; complex I: VPS23/STP22, VPS37/SRN2; complex III: DID4/VPS2, VPS24, VPS20; ESCRT IIIrelated proteins: DID2 and VPS4; and proteins associated with ESCRT function: DOA4, UBP2 and BRO1) ( Table 2). The ESCRT complexes comprise sequential steps in the internalization of the early endosome membrane to form intraluminal vesicles (ILV). As the endosome matures it becomes a multivesicular body (MVB). Proteins targeted to the endosome membrane can be recycled or sorted to the MVB that then deliver the ILVs to the vacuole (yeast equivalent of the lysosome), where proteins are eventually degraded.
The edelfosine-resistant gene DOA4 is an ubiquitin isopeptidase required for recycling ubiquitin from late endosomes and is essential for the maintenance of the free ubiquitin pool. Ubiquitination of proteins is a major determinant in their recognition by the ESCRT complex and subsequent vacuolemediated degradation (41). In fact, genes coding for the E3 ubiquitin ligase protein adaptors BUL2 and BSD2 were also identified in our screen (Supplementary Table S1). In agreement with these findings, we show in the accompanying paper (Czyz et al.) that edelfosine indeed induces ubiquitination and internalization of not only Pma1p, but also nutrient H + symporters like the arginine transporter Can1p and uracil transporter Fur4p.
Yeast strains defective in retrograde transport still deliver edelfosine to the ER, however, a higher proportion of Pma1p is found at the PM in retromer mutants. This implies that recycling of Pma1p, and possibly other lipid raft associated proteins, confers edelfosine resistance.

DISCUSSION
ATLs such as edelfosine have shown promise for cancer treatment, but it has been difficult to further optimize their efficacy as the mechanism of action of these drugs remains to be fully elucidated. In this study we present the first chemogenomic screen undertaken to uncover the underlying genetic mechanisms that mediate edelfosine resistance in yeast. The most powerful aspect of carrying out genetic screens is that the approach allows identification of unsuspected pathways that modulate cell sensitivity and resistance to the drug. We had previously determined that edelfosine disrupts lateral organization of the plasma membrane as an early event in its mode of action in yeast (17). In this study and the accompanying paper (Czyz et al.), we have found that following interaction of edelfosine with lipid rafts, resistance and sensitivity to this drug is mediated through two major processes, namely: i) disturbance of pH homeostasis (Czyz et al.); and ii) sorting of specific proteins at the endosome via sorting nexins and retromer at early and late endosomes, respectively and by the ESCRT vesicular trafficking pathway.
Based on the results reported here and the accompanying paper (Czyz et al.), we propose a working model for edelfosinemediated cytotoxicity in yeast that involves lipid raft reorganization (Fig. 11, a-d) and loss of pH homeostasis (Fig. 11e), leading to cell death. Loss of lateral organization in the yeast plasma membrane is expected to impact a number of proteins that use lipid microdomains as scaffolds to modulate their activity/function. Despite the fact that several proteins would be affected, our results consistently point to a main role for the essential plasma membrane H + -pump Pma1p in mediating the toxic effect of edelfosine, as well as of the clinically relevant ATLs miltefosine and perifosine. Pma1p is tightly associated with plasma membrane lipid rafts and this interaction is necessary for its ATPase activity (42) (Fig. 11a). Lipid rafts are highly compact microdomains enriched in ergosterol and sphingolipids and treatment with edelfosine disrupts these structures (Fig.  11a). Edelfosine and cholesterol have complementary conical geometries, which allow both molecules to bind tightly and form bilayers in vitro (43). It has been shown that edelfosine alters the organization of lipid rafts (44), and this biophysical change leads to alterations in the raft protein composition (4,12,17). This modification in raft organization is likely more extensive when sphingolipid (FEN1, SUR4) or ergosterol (ERG3) metabolic genes are deleted, these mutants being hypersensitive to edelfosine [Czyz et al. and (17)]. Once edelfosine alters lateral organization of the plasma membrane, Pma1p is displaced from rafts (17) followed by its internalization via endocytosis to late endosomes, and then eventually to the vacuole to be degraded (Fig. 11, b-c).
Through the use of radioactive and fluorescent edelfosine analogues we provide conclusive evidence that rules out the endocytic route for entrance of edelfosine into yeast. Our data suggest that edelfosine is internalized into the ER (Fig. 11d) through a non-endocytic pathway, but this intracellular location of the drug is not per se crucial for edelfosine toxicity in yeast. We interpret the retrieval from our edelfosine-resistance screen of mutants lacking components in several endocytic and trafficking pathways as a way to recycle Pma1p back to the plasma membrane. The model depicted in Fig. 11 implies that edelfosine treatment displaces Pma1p to the late endosome, and then eventually to the vacuole to be degraded. Our data suggest a putative mechanism of Pma1p degradation involving its ubiquitination (Fig.  11, c (1) and c (2) ), followed by recognition through the ESCRT complex (Fig. 11c (3) ), a process highly conserved in eukaryotes, and then incorporation into MVB (Fig. 11c (4) ), and subsequent degradation in the vacuole (Fig.  11c (5) ). In endosome traffic-defective mutants, where the above mechanisms are not wholly functional, Pma1p could escape the endosome, preventing its entering the degradation pathway, and cycle back to the plasma membrane (Fig. 11, c (6) and c (7) ).
In support of this model, it has been described that a temperature-sensitive conditional mutant of PMA1 (pma1-7, Pma1p ts ) undergoes growth arrest because newly synthesized protein is directed to the vacuole via endocytic intermediates at the restrictive temperature (45). Rerouting of this Pma1p ts mutant to the plasma membrane and a concomitant growth recovery could be achieved by inactivation of certain suppressor of pma1 (SOP) genes (44). Eight of these SOP mutants have defects in endosome-vacuole targeting of proteins. Interestingly, half of them [vps29Δ, vps35Δ (retromer), vps27Δ (ESCRT), vps8Δ (CORVET)] were also identified in our resistance screen to edelfosine. In vps27Δ and vps8Δ, traffic of an inducible Pma1p ts fused to GFP was monitored. It was demonstrated that these mutations blocked progress of the protein to the vacuole allowing its transport from the endosome to the plasma membrane (45). In this work we show similar results for vps35∆ and vps29∆ after treatment with edelfosine (Fig. 11, c (6) and c (7) ). Thus, the effect of edelfosine resembles that of the pma1-7 mutant, with the difference in the case of edelfosine being that Pma1p reaches the endocytic compartment from the plasma membrane. A blockage in the trafficking pathway leading to the vacuole results in rerouting of Pma1p back to the plasma membrane overcoming the cytotoxic effect.
While it has been well documented recycling of proteins is enhanced in cells lacking ESCRT components, it is intriguing how lack of endosomal sorting function results in recycling of internalized Pma1p. We propose internalized Pma1p destined for degradation in the vacuole is scrutinized by the sorting machinery, ensuring it does not go back to the plasma membrane. Lack of sorting allows for diversion of internalized Pma1p to return to the plasma membrane. In this regard it is interesting to note that lack of the recently characterized sorting protein Ere1p (but not Ere2p), which is critical for recycling of the raft associated transporter Can1p (36), confers resistance to edelfosine (Supplementary Table  S1). Targeting of membranes by the prototypic ATL edelfosine is unveiling a fascinating network of communication between plasma membrane and intracellular membranes, that rely on intracellular pH, to control cellular growth or death decisions.   lem3Δ cells were incubated and imaged as above. Since little to no intracellular fluorescence was observed in lem3Δ cells, higher exposition times had to be used, which accounts for the increased noise. Images shown are from representative experiments repeated at least three times.   (b) Influence of decreased endocytosis on drug uptake was assayed by comparing end4 pep4Δ cells (kept at the semi-permissive temperature of 25ºC) and wild-type cells switched to 4ºC with wild-type cells at 30ºC. Cells were incubated in the presence of the endocytic markers lucifer yellow (LY) and FM4-64, and afterwards with PTE-ET. (c) GFP-tagged Pma1p is internalized to the vacuole after edelfosine treatment (EDLF) at 30ºC; whereas at a temperature blocking endocytosis (4ºC), this process, unlike drug uptake (see panel b), is impaired. Images shown are from representative experiments repeated three times.      (2), recognition of ubiquitin moiety by the ESCRT complex (3), leading to recycling of the ubiquitin and enclosement of Pma1p in lumenal vesicles of the MVB (4). Fusion of the MVB with the vacuole would lead to degradation of these Pma1p-containing vesicles (5). Vacuolar hydrolases could also degrade Pma1p. The retromer complex is essential for the effect of the drug. We postulate this complex is either withdrawing Pma1p from the endosome to the Golgi apparatus (GA) via retrograde transport (6), or allowing the delivery of lysosomal hydrolases essential for the degradation of Pma1p (not shown). If any of these degradation pathways are impaired, a greater quantity of Pma1p could be available for recycling to the plasma membrane (7). (d) Both ergosterol and edelfosine are internalized. The drug accumulates in the endoplasmic reticulum (ER), and in some hitherto uncharacterized compartment in the cytoplasm. (e) pH homeostasis in physiological conditions is tightly controlled by Pma1p-mediated proton extrusion, V-ATPase proton sequestering, and mitochondria-mediated proton pumping. Edelfosine decreases the availability of the first buffering mechanism, causing acidification. When edelfosine is added to cells lacking a functional V-ATPase,