Characterization of micron-scale protein-depleted plasma membrane domains in phosphatidylserine-deficient yeast cells Free

The fluid mosaic model describing the dynamic distribution of proteins at the plasma membrane (PM) is the model that has largely been used to date, with modifications and developments (Singer and Nicolson, 1972; Nicolson, 2014; Kusumi et al., 2012). Lateral diffusion of proteins is not free and is influenced by protein interaction with other PM proteins and cytoskeletal elements. In cholesterol-rich domains, such as lipid rafts, certain proteins can accumulate due to protein–protein or protein–lipid interactions (Lingwood and Simons, 2010). In mammalian cells, protein assemblies are stabilized by cholesterol and actin filaments, which defines a sub-micron heterogeneous distribution pattern of membrane proteins (Saka et al., 2014). In addition, in yeast, high-resolution analysis has shown that different types of plasma membrane proteins form heterogeneous domains (Spira et al., 2012). They are mostly dynamic, and the domains are spread over the entire plasma membrane without significant bias. In giant unilamellar vesicles (GUVs) and giant plasma membrane vesicles (GPMVs), membrane phase separation leads to the formation of even larger domains of proteins and lipids (Veatch and Keller, 2003; Baumgart et al., 2007; Elson et al., 2010; Carquin et al., 2016). Phase separation in GUVs and GPMVs has been well studied and is often compared to the nanoscale membrane domains found in the cells; however, large-scale phase separation has not been observed in the PM of the living cells due to unknown reasons.

The PM is composed of diverse lipid species, and the role of phosphatidylserine (PS) and phosphatidylinositol phosphates (PIPs) in various cellular functions has been studied (Uchida et al., 2011; Cho et al., 2015; Middel et al., 2016; Tsuchiya et al., 2018; Michell, 2008; Balla, 2013). To explore a new role for PS, we have here analysed PS-deficient cho1Δ yeast cells. In this study, we show that stable large protein-depleted membrane domains are detected in the PM of PS-deficient cho1Δ cells, which we named the ‘void zone’. Transmembrane proteins, certain peripheral membrane proteins and certain phospholipids are excluded from the void zone. This property is very similar to the cholesterol-enriched membrane domain in phase-separated GUVs and GPMVs. However, the void zone is maintained in an energy-dependent (ATP and glucose) manner, suggesting that the void zone is a novel membrane domain unique to living cells. Furthermore, we found that vacuoles, the lysosomal organelle of yeast, contact with the void zone on the PM.

GFP–Snc1-pm, a mutant of v-SNARE Snc1, is distributed throughout the PM due to a defect in its endocytosis (Lewis et al., 2000). In wild-type cells grown at 37°C, GFP–Snc1-pm was localized to the entire surface of the PM, although small-scale heterogeneous distribution was observed (Fig. 1A, surface). On the other hand, in PS-deficient cho1Δ cells grown at 37°C, GFP–Snc1-pm was heterogeneously distributed in micron-scale domains on the PM (Fig. 1A). A GFP–Snc1-pm-less region was barely detectable at 30°C, but was frequently present during incubation at 37°C for over 6 h, and was not detected after heat shock at 42°C for 20 min (Fig. 1B). This Snc1-pm-less region of the PM is referred to as ‘void zone’ in the present study. The size and shape of the void zone was very diverse among cells; many of them were irregular and not smoothly circular (Fig. S1). When cho1Δ cells were observed immediately after staining with FM4-64 lipophilic dye, FM4-64 was distributed throughout the PM, including the void zone (Fig. 1C; Fig. S2), suggesting that the PM is not lost or significantly damaged in cells harbouring the void zone. To examine whether the void zone influences the distribution of other transmembrane proteins, the PM location of four different transmembrane proteins, Pma1, Pdr5, Pdr12 and Sfk1 (Serrano et al., 1986; Bauer et al., 1999; Audhya and Emr, 2002; Mioka et al., 2018), was compared with that of Snc1-pm (Fig. 1D; Fig. S2). The results indicate that all these proteins showed void zones in the same region as void zones detected by Snc1-pm (Fig. 1D). These results suggest that the void zone is a membrane protein-depleted region, with a lack of all other membrane proteins (see below), and that the void zone represents an abnormal lipid domain that inhibits the lateral movement of the transmembrane proteins into the domain.

We also investigated the distribution of eisosomes, large immobile protein complexes that form furrow-like invaginations in the fungal PM (Walther et al., 2006; Douglas and Konopka, 2014). The eisosome components, Pil1 and Sur7, were not distributed in the void zone (Fig. S3). The void zone was also devoid of the eisosome structure.

To further examine whether transmembrane proteins are completely absent in the void zone, electron microscopy combined with the freeze-fracture replica method was applied (Fujita et al., 2010; Tsuji et al., 2017). In this method, transmembrane proteins are detected as the granular structures called intramembrane particles (IMPs) (Fig. 1E). In the protoplasmic face of the PM in cho1Δ cells, most of the regions were IMP-rich although a submicron to micron-sized smooth region without IMPs was detectable (Fig. 1F; Fig. S1). These IMP-deficient regions were found in 30% of electron microscopy images (n=50), consistent with the results of fluorescence microscopy. To examine whether the IMP-deficient area is the void zone, we labelled Pma1–GFP with an anti-GFP antibody and colloidal gold-conjugated protein A (Fig. 1E). As expected, only the IMP-rich area stained for Pma1–GFP and there was no labelling in the IMP-deficient area (Fig. 1G). This result is consistent with the fluorescence microscopy results (Fig. 1D), and indicates that the void zone corresponds to the IMP-deficient area. Thus, there are very few transmembrane proteins in the void zone.

We next examined the stability of the void zone by time-lapse imaging. Most of the void zones were detected over tens of minutes, and some void zones were detectable for longer periods of time (over 90 min). Bud growth and cytokinesis appeared to proceed normally in the void zone-containing cells, suggesting that the void zone does not influence cell cycle progression (Fig. 2A). Consistent with this, cho1Δ cells showed no significant growth defects at 37°C compared to growth at 30°C (Fig. 2B). The void zone rarely formed at 30°C (Fig. 1B), but once formed, the void zone was largely maintained, even after 90 min of incubation at 30°C (Fig. 2C). These results suggest that the void zone is a very stable lipid domain. On the other hand, the void zone was essentially undetectable in a saturated culture with high cell density and very low cell growth. Thus, we investigated whether the frequency of the void zone formation is associated with the growth phase (Fig. 2D). Yeast cells show rapid growth in the presence of abundant carbon sources, such as glucose. When glucose is depleted, the cells switch to a slower growth rate using ethanol as a metabolic carbon source via a diauxic shift, and consequently enter the stationary phase (Gray et al., 2004). As shown in Fig. 2D, the proportion of cells with void zones was increased early in the logarithmic growth phase and began to decrease starting from the middle phase. Remarkably, the void zone rapidly disappeared when the cells entered the diauxic shift, and no changes were observed after 12 h, suggesting that the void zone is formed and maintained only in the presence of abundant carbon sources. To investigate the disappearance of the void zone, we examined whether certain stresses induce the disappearance. cho1Δ cells grown at 37°C were incubated for an additional 3 h at 37°C under three stress conditions, including ATP depletion, glucose starvation and translation inhibition; then, the cell numbers and the frequency of the void zone were measured. Cell growth was almost completely blocked by all these stresses, and the proportion of cells with void zones was significantly reduced by ATP depletion and glucose starvation, but not by translation inhibition (Fig. 2E). These results suggest that the maintenance of the void zone is energy dependent and does not necessarily require cell growth.

We also noticed that the void zone tends to occur in mother cells rather than daughter cells (Fig. 2F). To confirm this, the distribution of the void zone was categorized into the following groups; cells with no bud or small bud (type 1), cells with mid-large bud in which the void zone occurs only in the mother cell (type 2), and cells in which the void zone occurs in the mother and daughter cells (type 3). We found that almost all void zones appeared only in the mother cells before or during budding (type 1 and 2); very few type 3 cells were detected, and no cells with void zones formed only in the daughter cells were observed (Fig. 2F). This biased distribution was observed even under the condition of OD₆₀₀=4.0–6.0, where there is a high proportion of cells of void zones. Exocytosis and endocytosis frequently occur in a bud, which is a polarized growth site, compared to a mother cell (Bi and Park, 2012). These results suggest that the void zone is generated in a more static membrane.

To examine whether the void zone can be detected in lipid mutants other than cho1Δ, we investigated cells with mutations of the genes involved in the synthesis of phosphatidylethanolamine (PE) and phosphatidylcholine (PC). In PE- and PC-depleted cells, GFP–Snc1-pm was evenly distributed in the PM, and the void zone was not detected (Fig. 3A). We next examined whether the recovery of PS levels dissipates the void zone by adding lyso-PS to the culture medium. Incorporated lyso-PS is rapidly converted into PS in the endoplasmic reticulum (ER) via a CHO1-independent pathway (Fairn et al., 2011; Maeda et al., 2013). The recovery of PS was verified by expressing the PS-specific biosensor Lact-C2 (Yeung et al., 2008). In cho1Δ cells without lyso-PS, the proportion of cells with a void zone did not change over 60 min and mRFP–Lact-C2 remained diffuse in the cytosol (Fig. 3B). On the other hand, the addition of lyso-PS, but not lyso-PC, significantly reduced the incidence of the void zone and resulted in the localization of mRFP–Lact-C2 to the PM (Fig. 3B). These results indicate that void zone formation is a reversible process. After the addition of lyso-PS, void zones did not disappear uniformly, but rather appeared to be gradually repaired from the boundaries (Fig. 3C). In some cho1Δ cells, mRFP–Lact-C2 was distributed outside of the void zone (Fig. 3D). This void zone was gradually repaired by PS around the void zone. Our results suggest that the void zone is a lipid domain that consists of a lipid (or lipids), and lateral movement of these lipids in the PM is facilitated by interaction with PS.

We next examined whether the PM-localized peripheral membrane proteins are distributed into the void zone. Two types of proteins were tested (Fig. 4A). Ras2 is a small GTPase with a C-terminal lipid moiety inserted into the cytoplasmic leaflet of the PM. Ras2 undergoes farnesylation and palmitoylation, and the latter is required for membrane localization (Bhattacharya et al., 1995). Gap1C is the C-terminal cytosolic region of the amino acid permease Gap1. Gap1 localizes to the PM, and Gap1C without the transmembrane domain is also localized at the PM due to its amphipathic helix structure (Popov-Čeleketić et al., 2016). These proteins were mainly localized to the PM in the wild-type cells, but were absent from the void zone, similar to Snc1-pm, in cho1Δ cells (Fig. 4B; Fig. S2). Because different types of peripheral membrane proteins are excluded from the void zone, it suggests that the void zone is a specialized lipid environment. Normally, PS is abundantly distributed in the cytoplasmic leaflet of the PM. The formation of the void zone may be due to lipid changes in the cytoplasmic leaflet.

Molecular dynamics computer simulation, giant unilamellar vesicle (GUV) assays and giant plasma membrane vesicle (GPMV) assays have shown that transmembrane helices and transmembrane proteins are excluded from cholesterol-enriched lipid domains (Sengupta et al., 2008; Schäfer et al., 2011). Transmembrane proteins are also excluded from ergosterol-enriched raft-like domains that are generated in the vacuolar membrane of yeast in the stationary phase (Toulmay and Prinz, 2013; Wang et al., 2014; Tsuji et al., 2017). Therefore, we examined whether the void zone is a sterol-enriched lipid domain. To assess the distribution of ergosterol, a major sterol in yeast, cells were stained with the sterol-binding dye filipin. Filipin showed a punctate distribution in the PM of the wild-type cells as reported previously (Grossmann et al., 2007), whereas in cho1Δ cells, filipin mainly accumulated in the void zones (Fig. 5A), indicating that the void zone is rich in ergosterol.

Similar to sterols, sphingolipids are important for the formation of specialized membrane domains (Dietrich et al., 2001; Baumgart et al., 2003; de Almeida et al., 2003; Veatch and Keller, 2003). We therefore examined whether the biosynthesis of ergosterol and sphingolipid is required for the formation of the void zone. Erg2, Erg3 and Erg4 are enzymes required for ergosterol synthesis (Silve et al., 1996; Arthington et al., 1991; Lai et al., 1994). Ipt1 is an inositolphosphotransferase required for the synthesis of mannosyl-diinositolphosphorylceramide [M(IP)2C], the most abundant sphingolipid in yeast (Dickson et al., 1997). Elo2 and Elo3 are fatty acid elongases, and participate in the long chain fatty acid biosynthesis of sphingolipids (Oh et al., 1997). Sur2 and Scs7 are hydroxylases involved in the hydroxylation of sphingolipids (Haak et al., 1997). Disruption of any of these genes essentially abolished the formation of the void zone (Fig. 5B). The suppression of void zone formation by erg mutations may be due to defects in endocytosis (Pichler and Riezman, 2004). However, erg3Δ and erg4Δ have little effect on endocytosis, whereas ergosterol intermediates accumulate in these mutants and cause the loss of membrane integrity (Abe and Hiraki, 2009). The suppression of void zone development may be due to lower packing ability of ergosterol intermediates. An inhibitor of ergosterol synthesis, fluconazole, also blocked the formation of the void zone (Fig. 5C). These results indicate that ergosterol and sphingolipids are necessary for the formation of the void zone and suggest that the void zone may be a novel lipid domain composed of abundant ergosterol and sphingolipids.

We further investigated whether the formation of the void zone results from an increase in the ergosterol level. Total ergosterol levels were found to be increased by incubation at 37°C and by deletion of CHO1; cho1Δ cells grown at 37°C had slightly higher ergosterol levels than those grown in other conditions (Fig. 5D). Thus, there may be a threshold of the ergosterol level required for the void zone formation.

The finding that the void zone is an ergosterol-rich lipid domain prompted us to investigate the distribution of other lipids that can be visualized using the corresponding probes. To examine the distribution of phosphatidylinositol 4-phosphate [PI(4)P], phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] and phosphatidic acid (PA), the PH domains of yeast Osh2 and human PLCδ and the PA-binding domain of yeast Spo20 were used as probes, respectively (Roy and Levine, 2004; Lemmon et al., 1995; Watt et al., 2002; Nakanishi et al., 2004). These probes were uniformly distributed in the PM in the wild-type cells; however, none of the probes were detected in the void zone (Fig. 5E; Fig. S2). This result suggests that PI(4)P, PI(4,5)P2 and PA cannot enter the void zone by lateral diffusion. The majority of yeast phospholipids have mono- or di-unsaturated fatty acids (Schneiter et al., 1999; Ejsing et al., 2009; Klose et al., 2012). In GUVs and GPMVs, these unsaturated lipids have also been found to be separated from cholesterol-enriched membrane domains (Baumgart et al., 2003, 2007; Risselada and Marrink, 2008; Lingwood and Simons, 2010).

We consider that the void zone is a lipid domain with an unusual assembly of ergosterol that cooperates with sphingolipids, and thereby has novel properties with limited lateral diffusion of the peripheral membrane proteins and certain types of glycerophospholipids.

Observations using the lipid probes suggest phase separation in the PM, and that the lipid composition of the void zone is different from that in the other PM regions. The PM and the ER form membrane contact sites (MCSs) via ER-resident tethering proteins that interact with phospholipids of the PM (Saheki and De Camilli, 2017). To test whether the void zone can influence this ER–PM contact, we examined two ER marker proteins, Hmg1 and Rtn1 (Koning et al., 1996; De Craene et al., 2006). The cortical ER (cER) was clearly absent at the void zone in the cho1Δ cells (Fig. 6A; 89.2% of Hmg1, 92.9% of Rtn1, n>100 cells, respectively). The ER–PM-tethering proteins Tcb1, Tcb2 and Tcb3, and Ist2 have been reported to bind to PS and PI(4,5)P2, respectively (Schulz and Creutz, 2004; Fischer et al., 2009). Since PS is not synthesized in the cho1Δ cells and PI(4,5)P2 is not distributed in the void zone (Fig. 5E), the association of cER with the PM may be lost in the void zone.

To investigate whether the void zone influences the distribution or morphology of other organelles, several organelle markers were examined in the cho1Δ cells. Surprisingly, we found that vacuoles were in contact with the void zones (Fig. 6B). The observations using Snc1-pm and FM4-64 indicate that the contact between the void zone and the vacuole was detected in 45.1% of the cho1Δ cells with the void zone (45.1±7.0%, five independent experiments, n>50 cells each). The proximity of the non-void zone regions and vacuoles was observed only in 7.8% of the cho1Δ cells (7.8±1.7%, three independent experiments, n>100 cells each) compared to that in the wild-type cells (10.1±2.1%, three independent experiments, n>100 cells each). These data suggest that frequent contact of the vacuoles with the PM is specific to the void zone. The fact that not all void zones are in contact with the vacuoles indicates that the exclusion of the transmembrane proteins from the void zones was not caused by the vacuole contact (Fig. 6B). On the other hand, the trans-Golgi network (TGN), lipid droplets, and mitochondria did not contact the void zone (Fig. S4), suggesting that only vacuoles interact with the void zone. This type of contact between the PM and the vacuoles or lysosomes has not been reported, and we refer to the contact between the void zone and vacuoles as the ‘V–V contact’.

In stationary phase yeast cells, a raft-like domain is formed in the vacuolar membrane, where lipophagy, the uptake of the lipid droplets, occurs (Toulmay and Prinz, 2013; Wang et al., 2014; Tsuji et al., 2017). The properties of this vacuolar microdomain are very similar to the properties of the void zone, including absence of the transmembrane proteins and ergosterol enrichment. To test whether this vacuolar microdomain is present in the void zone contact area, we examined three vacuolar transmembrane proteins, Vph1, Cot1 and Zrc1. These proteins were uniformly distributed in the vacuolar membrane of the wild-type cells; however, in some vacuoles in contact with the void zone of the cho1Δ cells, these proteins were excluded from the contact area (Fig. 6C). One of these proteins, Vph1, had the highest frequency of segregation on vacuoles in contact with the void zone (73.5% of Vph1, 41.5% of Cot1 and 30.0% of Zrc1, n>50). In contrast, the exclusion of FM4-64 dye in vacuoles of the cells with the V–V contacts was not detected (n>50; Fig. 6B). These data suggest that certain vacuoles form a membrane domain at the V–V contact site.

To understand the dynamics of the V–V contact, we used time-lapse imaging with GFP–Snc1-pm and Vph1–mRFP (Movie 1) and detected two patterns. First, the movement of a vacuole into the void zone was observed (Fig. 6D, shown as ‘a’). The speed of the vacuole migration to the void zone was highly variable between individual cells. However, some void zones had no contact with the vacuoles during the 1 h observation. Second, the V–V contact lasted over half an hour (Fig. 6D, shown as ‘b’). In one hour of observation of the cells with the void zone, 60% of the cells maintained the V–V contact for over 30 min (Fig. 6E). The disappearance of the void zone after its formation was rarely observed, and this phenomenon appeared to be unrelated to the presence or absence of the V–V contacts. In addition, no vacuoles left the void zone while the void zone was maintained. The molecular basis of the V–V contact is unknown although this contact appears to be stable. Vacuoles in contact with the void zone underwent fission and fusion (Fig. S4).

To further understand the mechanism of the void zone formation, we created a series of deletion mutants in the background of the cho1Δ or glucose-repressible P_GAL1-3HA-CHO1 mutations, and examined their effect on the generation of the void zone (Fig. 7A). Various genes involved in sterol trafficking were tested, and the results indicate that kes1Δ (osh4Δ) mildly and arv1Δ significantly reduced the void zone formation. Kes1 is one of the yeast oxysterol-binding proteins that exchanges sterols for PI(4)P between the lipid membranes (Jiang et al., 1994; de Saint-Jean et al., 2011). Arv1 was implicated in the GPI-anchor biosynthesis and transport and in intracellular sterol distribution (Kajiwara et al., 2008; Beh and Rine, 2004). Both Kes1 and Arv1 are involved in the sterol transport; however, their contribution to the sterol transport to the PM is very low (Georgiev et al., 2011, 2013). These proteins regulate sterol organization in the PM; the mutations influence the sensitivity to the sterol-binding drugs and sterol-extraction efficiency of methyl-β-cyclodextrin (MβCD) (Georgiev et al., 2011, 2013). These differences in sterol organization may influence the formation of the void zone. Npc2 is an orthologue of Niemann–Pick type C protein and plays an essential role, together with Ncr1, in sterol insertion into the vacuolar membrane from the inside of the vacuole, which is required for the formation of the raft-like vacuolar domain during lipophagy in the stationary phase (Tsuji et al., 2017). However, deletion of Npc2 did not influence generation of the void zone or formation of the V–V contacts associated with the protein-depleted vacuolar domain (Fig. S5). This result suggests that the void zone and vacuolar domains at the V–V contact are formed independently of Ncr1- and Npc2-mediated sterol transport.

The void zone formation was strongly suppressed by deletion of vacuolar proteins, Vma2 and Fab1. Vma2 is a subunit of the V-ATPase that regulates pH homeostasis (Marshansky et al., 2014). Consistent with this observation, other genes involved in pH homeostasis (NHA1, NHX1, RIM21 and RIM101) were required for the void zone formation (Fig. 7A, pH regulation) (Sychrová et al., 1999; Brett et al., 2005; Obara et al., 2012). Thus, we examined the effect of pH of the medium on the formation of the void zone. The formation of the void zone was strongly suppressed by increasing pH in the medium from 6.6 to 7.5 (Fig. 7B). On the other hand, the low pH medium (pH 4.0) slightly increased the formation of the void zone. Interestingly, the proportion of cells with V–V contacts was significantly reduced in the low pH medium (Fig. 7C). The mechanism of these pH-dependent phenomena is unclear although they may be important in assessment of the molecular basis of the formation of the void zones and V–V contacts.

Fab1 is a phosphatidylinositol 3-phosphate [PI(3)P] 5-kinase that generates phosphatidylinositol 3,5-bisphosphate [PI(3,5)P2] (Cooke et al., 1998). PI(3,5)P2 functions as a signal lipid in intracellular homeostasis, adaptation and retrograde membrane trafficking (Jin et al., 2016). We speculated that defects in retrograde transport may indirectly influence the void zone formation in the PM, and thus examined various genes involved in membrane transport. Strikingly, conserved protein complexes, retromer, CORVET, HOPS and ESCRT, were required for the void zone formation (Fig. 7A,D). As the name implies, the Vps proteins belonging to these complexes were identified from mutants defective in vacuolar protein sorting (VPS) (Robinson et al., 1988; Rothman et al., 1989). Dysfunction of these complexes perturbs intracellular vesicle trafficking (Schmidt and Teis, 2012; Balderhaar and Ungermann, 2013; Burd and Cullen, 2014), which may influence the PM recycling of cargo and lipids involved in the void zone formation. Similarly, proteins involved in the retrograde transport, such as the SNAREs Pep12 and Tlg2, the epsin-like adaptors Ent3 and Ent5, and the clathrin adaptors Gga1 and Gga2, were required for the void zone formation. The dynamin-like GTPase Vps1, Arf-like GTPase Arl1 and Rab6 GTPase homologue Ypt6 are known to be closely related to the membrane trafficking (Vater et al., 1992; Li and Warner, 1996; Rosenwald et al., 2002). Consistent with this notion, vps1Δ, arl1Δ, and ypt6Δ inhibited the void zone formation. Impaired membrane transport in the inner membrane system can be manifested as changes in the PM lipid organization and/or defects in the pH control. However, Apl2 and Apl1, subunits of the adaptor complexes AP-1 and AP-2, respectively, had little contribution to the void zone formation, presumably because these mutants had insignificant disruption of the membrane trafficking compared to the effects of ent3Δ, ent5Δ and gga1Δ gga2Δ (Yeung et al., 1999; Sakane et al., 2006; Morvan et al., 2015). Deletion of APL5, which encodes the subunit of AP-3 responsible for the transport from the Golgi to the vacuole (Dell'Angelica, 2009), slightly reduced void zone formation, suggesting the importance of the vacuolar functions for void zone formation.

Deletion of the autophagy-related genes (atg1Δ, atg10Δ, atg12Δ and atg15Δ) did not influence void zone formation. ATG1 is one of the core ATG genes (Mizushima et al., 2011). The void zone was generated and the V–V contact with the vacuolar microdomain was observed in the absence of Atg1 (Fig. 7A; Fig. S5). This result suggests that void zone formation is independent of autophagy, consistent with our notion that direct or indirect effects on lipid organization in the PM (e.g. via the Vps pathway) are influencing the void zone formation.

We also examined the effect of mutations in the flippase-related proteins. The deficiency of Cdc50 or Any1 and Cfs1, localized in endosomes and the TGN, had little effect on the void zone formation (Saito et al., 2004; van Leeuwen et al., 2016; Yamamoto et al., 2017); however, disruption of Lem3 localized in the PM completely suppressed the generation of the void zone (Kato et al., 2002). The flippase complexes between Lem3 and Dnf1 or Dnf2 translocate glycerophospholipids to the cytoplasmic leaflet of the lipid bilayer (Pomorski et al., 2003; Saito et al., 2004; Furuta et al., 2007). We assumed that disruption of the phospholipid asymmetry by lem3Δ may influence ergosterol behaviour in the PM. To test this hypothesis, we examined the sensitivity to the antifungal ergosterol-binding drug amphotericin B (AmB) (Kamiński, 2014). The results indicate that lem3Δ is highly sensitive to AmB, which is detected in the cho1Δ background cells, and this effect was reversed by addition of erg6Δ, which causes defects in ergosterol biosynthesis (Fig. 7E). Thus, the disruption of phospholipid asymmetry alters the ergosterol distribution in the PM, thereby suppressing void zone formation (see Discussion).

Although it is not yet clear through which mechanisms some of these additive mutations inhibit the void zone formation, these findings may help to elucidate the nature of the void zone in the future.

In artificial membranes and GPMVs, phase separation causes micron-scale separation of proteins and lipids, but there were no known examples of such large-scale separation in the living cell PM. We found that in PS-deficient yeast cells grown at high temperatures, the protein-depleted membrane domain ‘void zone’ develops in the PM (Fig. 8). The void zone is a novel membrane domain that has properties similar to those of phase-separated membranes, but its formation depends on a variety of physiological cellular functions. Although void zones occur under very limited conditions, the formation of void zones indicates that the PM in living cells has the potential to generate phase separation in certain lipid compositions.

PS is a phospholipid mainly distributed in the inner leaflet of the PM and which has a relatively high affinity for cholesterol (Maekawa and Fairn, 2015; Nyholm et al., 2019). Therefore, loss of PS may alter the relative affinity of the phospholipids for ergosterol in the inner leaflet and may also influence the transbilayer distribution of ergosterol. Almost all yeast phospholipids, including PS, have at least one unsaturated fatty acid; however, a small fraction of phosphatidylinositol (PI) has only saturated fatty acids (Schneiter et al., 1999; Ejsing et al., 2009; Klose et al., 2012). Molecular species of PS, PE, PC and PI were characterized in the isolated PM; interestingly, 29.2% of PIs have two saturated fatty acids (Schneiter et al., 1999). The relative affinity of ergosterol to various phospholipid species is unclear. One hypothesis is that ergosterol is not clustered due to the interactions with unsaturated PS in the wild-type cells; however, in cho1Δ cells, saturated PI becomes the most dominant interaction partner of ergosterol in the inner leaflet of the PM, thus creating a driving force for phase separation. When PS is resynthesized in the cho1Δ cells after the addition of lyso-PS, the void zone may disappear because PS becomes the predominant interaction partner of ergosterol (Fig. 3B,C).

Generation of the void zone requires PS deficiency and high temperature conditions. In artificial membranes and GPMVs, phase separation occurs at a temperature below the miscibility transition temperature, which is dependent on the lipid composition of the membrane (Veatch and Keller, 2002; Baumgart et al., 2007; Levental et al., 2009). Contrary to these phenomena, the void zone is rarely formed at 30°C and occurs by incubation at 37°C (Fig. 1B); thus, we speculate that a change in miscibility transition temperature due to remodelling of lipid composition at 37°C may be one of the triggers for void zone formation in cho1Δ cells. In yeast, high temperature reduces the degree of unsaturation and increases the acyl chain length of glycerophospholipids (Klose et al., 2012), and these events promote phase separation in liposomes (Engberg et al., 2016). In addition, our results showing that several hours of incubation at 37°C was required to generate the void zone (Fig. 1B), and that the developed void zone was maintained for 90 min after the temperature was returned to 30°C (Fig. 2C) are consistent with this idea, and suggest that remodelling of the lipid composition takes a long time. A combination of PS deficiency and high temperature-induced lipid remodelling may create a specific membrane environment that results in void zone development.

The void zones rapidly disappeared when the cho1Δ cells undergo a diauxic shift (Fig. 2D) or when glucose or ATP is depleted from the medium (Fig. 2E). During glucose starvation, activities of the PM H⁺-ATPase Pma1 and V-ATPase are reduced (Young et al., 2010; Dechant et al., 2010). Therefore, the disappearance of the void zone may be caused by a disturbance of pH homeostasis consistent with the suppression of the void zone formation by mutations in the genes involved in pH homeostasis, including VMA2 (Fig. 7A). Reduced activities of Pma1 and V-ATPases lead to acidification of the cytosol, and void zone formation is also inhibited in the medium with elevated pH (Fig. 7B). The relationship between pH and phase separation is largely unknown; however, it has been reported that phase separation occurs at low pH and not at high pH in artificial membranes that mimic human stratum corneum (Plasencia et al., 2007).

Our results indicate that long chain fatty acids and hydroxylation of sphingolipids are necessary for void zone formation (Fig. 5B). It has been reported that GUVs prepared from the yeast total lipids have extensive phase separation, which depends on long fatty acid elongation and hydroxylation of sphingolipids (Klose et al., 2010). M(IP)2C with a C26 fatty acid is the most common yeast sphingolipid species and its levels are significantly reduced in the elo2Δ and elo3Δ mutants (Oh et al., 1997; Ejsing et al., 2009). A recent study using asymmetric GUVs suggested that C24 SM and not C16 SM has a role in cholesterol retention in the inner leaflet of the lipid bilayer (Courtney et al., 2018). Consistent with this, dehydroergosterol, a closely related fluorescent analogue of ergosterol, is mainly located in the inner leaflet of the PM, and this asymmetry is maintained by sphingolipids in yeast (Solanko et al., 2018). This type of interaction of ergosterol with sphingolipids may be necessary for void zone formation. The distribution of sterols and sphingolipids is also partly regulated by vesicular transport-related complexes such as GARP and retromer (Fröhlich et al., 2015; Marquer et al., 2016). This may be one reason why retrograde vesicular transport is required for the formation of the void zone (Fig. 7A).

Analysis using several mutant strains on the cho1 background revealed that retrograde intracellular trafficking has a significant effect on void zone formation (Fig. 7A). Perturbed membrane transport may alter the distribution of the proteins and the key lipids as described above. Moreover, ARV1 and LEM3 are important for void zone formation. In arv1Δ cells and Δtether cells, which lack the ER-PM contacts, transport of ergosterol to the PM is essentially unchanged, whereas there were significant changes in ergosterol accessibility to the compounds, such as increased efficiency of ergosterol extraction with MβCD and increased sensitivity to the sterol-binding drug nystatin (Georgiev et al., 2013; Quon et al., 2018). Similarly, lem3Δ does not change the total ergosterol level (Mioka et al., 2018), and Dnf1 and Dnf2 do not contribute to the asymmetric distribution of the ergosterol analogues across the PM bilayer (Solanko et al., 2018). However, lem3Δ cells are highly sensitive to AmB (Fig. 7E). Thus, similar to the arv1Δ and Δtether mutations, lem3Δ may alter the ergosterol distribution in the PM, thus preventing the formation of the void zone.

Before this report, there were no known examples of micron-scale phase separation in the PM of living cells, and its reason is not clear (Lee et al., 2015; Levental et al., 2020). Thus, the void zone seems to be the first report of a micron-scale phase separation in the PM in vivo. The following features of the void zone are similar to those of the liquid-ordered domain in phase-separated GUVs and GPMVs: (1) absence of the transmembrane proteins, (2) absence of certain peripheral membrane proteins, (3) exclusion of certain types of glycerophospholipids, and (4) enrichment in sterols and the contribution of sphingolipids (Baumgart et al., 2007; Sengupta et al., 2008; Kaiser et al., 2009; Schäfer et al., 2011). On the other hand, the formation of void zones requires phospholipid asymmetry in the PM, regulators of pH homeostasis and normal retrograde vesicular transport (Fig. 7A), and the maintenance of the void zone is energy dependent (Fig. 2E). These results suggest that the void zone is a unique and delicate domain that requires a variety of biological processes for its formation. We consider that this property of the void zone is a crucial difference from the liquid-ordered domain of artificial membranes and would make it difficult to reproduce the void zone in vitro. Differences between living and artificial membranes regarding phase separation have been reported previously; macroscopic phase separation does not occur in living wild-type yeast cells (Fig. 1A,D), but occurs in GUVs formed with lipids extracted from the wild-type strains (Klose et al., 2010). Similarly, whereas no phase separation is observed in living mammalian cells over a wide range of temperatures, GPMVs prepared from the same cells show macroscopic phase separation (Lee et al., 2015). Further studies of the void zone would help to bridge the gap between the artificial and biological membranes. Besides the PS–sterol interaction, there may be other underlying mechanisms by which living cells avoid macroscopic phase separation.

We detected a contact between the void zone and the vacuoles (Figs 6B,D, 8A). Since there are also void zones without vacuolar contact (Fig. 6B), and these increase at low pH (Fig. 7C), we consider that the occurrence of void zones causes vacuolar contact, rather than vacuolar contact creating void zones. In addition, the significant decrease in transmembrane proteins in the void zone suggests that V–V contacts may be formed by lipid–protein interactions similar to those at ER–PM contacts. One possibility is that certain vacuole-localized proteins may be tethering the vacuole to the PM via binding to enriched ergosterol in the void zone. Our results also suggest that membrane domains may be formed on the vacuolar membranes in contact with the void zone (Fig. 6C). A contact between the membrane domains accompanied by protein separation has been detected in the nucleus–vacuole junction (NVJ), one of the MCSs (Pan et al., 2000; Dawaliby and Mayer, 2010; Toulmay and Prinz, 2013). This vacuolar domain formation may be involved in the formation of V–V contacts.

Time-lapse imaging revealed that the V–V contact lasted for at least 30 min in most cases (Fig. 6D,E). These results suggest that the contact vacuoles do not contribute to the rapid degradation and disappearance of the void zone. The contact vacuoles are more likely to play a role in sealing the void zone to protect the cells. Recent studies have reported that, in phase-separated membranes with the liquid-disordered and liquid-ordered domains, membrane permeability is increased at the interface between the two domains (Cordeiro, 2018). A similar result indicated that the ER–PM contact sites are important for the maintenance of the integrity of the PM (Omnus et al., 2016; Collado et al., 2019). In this study, we found that the void zone has liquid-ordered phase-like properties and that cortical ER is dissociated from the void zone. Therefore, it is possible that the void zone induces high local permeability on its borders and low local integrity; thus, the void zone may be a fragile region of the PM. The V–V contact may represent a protective cellular response to these PM abnormalities.

Chemicals were purchased from Wako Pure Chemicals Industries (Osaka, Japan) unless indicated otherwise. Standard genetic manipulations and plasmid transformation of yeast were performed as described previously (Elble, 1992; Guthrie and Fink, 2002). Yeast strains were cultured in rich YPDA medium containing 1% yeast extract (Difco Laboratories, MI, USA), 2% Bacto peptone (Difco), 2% glucose and 0.01% adenine, or synthetic dextrose (SD) medium containing 0.17% yeast nitrogen base without amino acids and ammonium sulfate (Difco), 0.5% ammonium sulfate, 2% glucose, and the required amino acids or nucleic acid bases. Unless otherwise indicated, cells were cultured in YPDA medium. To induce expression from the GAL1 promoter, 3% galactose and 0.2% sucrose were used as carbon sources instead of glucose (YPGA medium). To deplete PS in the P_GAL1-3HA-CHO1 background cells, the cells were grown on YPDA plates for more than 1 day before inoculation into YPDA medium. Strains carrying URA3-harbouring plasmids were cultured in SD medium containing 0.5% casamino acids (Difco), 0.03% tryptophan and 0.01% adenine (SDA-U). When cho1Δ or Cho1-depleted strains were cultured in SD or SDA−U medium, ethanolamine was added to the final concentration of 1 mM. For serial dilution spot assays, cells were grown to early log phase in an appropriate medium and adjusted to a concentration of 10⁷ cells/ml. After serial tenfold dilution, 4-μl drops were spotted onto appropriate plates.

Yeast strains and plasmids used in this study are listed in Tables S1 and S2, respectively. Standard molecular biological techniques were used for the construction of the plasmids, PCR amplification and DNA sequencing (Sambrook and Russell, 2001). PCR-based procedures were used to construct gene deletions and gene fusions with GFP, mRFP, 3HA and the GAL1 promoter (Longtine et al., 1998). All constructs produced by the PCR-based procedure were verified by colony PCR to confirm that the replacement occurred at the expected locus. Sequences of the PCR primers are available upon request.

To construct pRS416-GFP-GAP1C (pKT2205), the C-terminal cytoplasmic region of Gap1 corresponding to amino acids 552–602 (Popov-Čeleketić et al., 2016) and 257 bp downstream of the GAP1 coding region were amplified by PCR, and the PEP12 region of pRS416-GFP-PEP12 (pKT1487) (Furuta et al., 2007) was replaced with the PCR product. To construct pRS316-mCherry-SPO20PABD (pKT2206), the DNA fragments of mCherry and the PA-binding region of Spo20 corresponding to amino acids 51–91 (Nakanishi et al., 2004) were inserted into pRS316 along with P_TPI1 and T_ADH1.

Cells were observed using a Nikon ECLIPSE E800 microscope (Nikon Instec, Tokyo, Japan) equipped with an HB-10103AF super-high-pressure mercury lamp and a 1.4 NA 100× Plan Apo oil immersion objective lens (Nikon Instec) with appropriate fluorescence filter sets (Nikon Instec) or differential interference contrast optics. Images were acquired using a cooled digital charge-coupled device (CCD) camera (C4742-95-12NR; Hamamatsu Photonics, Hamamatsu, Japan) and the AQUACOSMOS software (Hamamatsu Photonics).

GFP-, mRFP- or mCherry-tagged proteins were observed in the living cells grown in early to mid-logarithmic phase, harvested and resuspended in SD medium. Cells were immediately observed using a GFP bandpass (for GFP) or a G2-A (for mRFP and mCherry) filter set. Observations were compiled based on the examination of at least 100 cells. For supplementation of 18:1 lyso-PS and 18:1 lyso-PC (Sigma-Aldrich, St Louis, MO, USA), a stock solution (10 mg/ml in 0.1% Nonidet P-40) was added to the culture medium to the final concentration of 20 μM. For sterol staining, cells were fixed with 5% formaldehyde, washed with PBS, and labelled for 10 min in 0.5 mg/ml filipin (Sigma-Aldrich) in PBS. For staining of the PM by FM4-64 (Thermo Fisher Scientific, MA, USA), cell suspensions were mixed with an equal volume of 100 μM FM4-64 on a glass slide and observed immediately. For staining of the vacuolar membranes, cells were labelled for 15 min in 5 μM FM4-64, washed with SD medium, and immediately observed. Fluorescence of filipin and FM4-64 was observed using a UV and a G2-A filter set, respectively. For the time-lapse imaging, a cell suspension was spotted onto a thin layer of SD medium containing 1 mM ethanolamine and 2% agarose on a glass slide, which was quickly covered with a coverslip. During the time-lapse imaging, the sample was maintained at 37°C by a Thermo Plate (Tokai Hit, Fujinomiya, Japan). The incidence of the void zone was assessed based on three or more images of different focal planes, and based on single focal plane in the time-lapse imaging.

Fluorescence intensity in the cell periphery (void zone and non-void zone regions) was measured by ImageJ.

Yeast cells sandwiched between a 20-μm-thick copper foil and a flat aluminium disc (Engineering Office M. Wohlwend, Sennwald, Switzerland) were quick-frozen by high-pressure freezing using an HPM 010 high-pressure freezing machine according to the manufacturer’s instruction (Leica Microsystems, Wetzlar, Germany). The frozen specimens were transferred to the cold stage of a Balzers BAF 400 apparatus and fractured at −115°C to −105°C under vacuum at ∼1×10⁻⁶ mbar (1×10⁻⁴ Pa).

For genuine morphological observation, samples were exposed to electron-beam evaporation of platinum-carbon (Pt/C) (1–2 nm thickness) at an angle of 45° to the specimen surface followed by carbon (C) (10–20 nm) at an angle of 90°. After thawing, the replicas were treated with household bleach to digest biological materials before mounting on the EM grids for observation.

For labelling, freeze-fractured samples were subjected to a three-step electron-beam evaporation: C (2–5 nm), Pt/C (1–2 nm), and C (10–20 nm), as described previously (Fujita et al., 2010). Thawed replicas were treated with 2.5% SDS in 0.1 M Tris-HCl (pH 8.0) at 60°C overnight; with 0.1% Westase (Takara Bio) in McIlvain buffer (37 mM citrate and 126 mM disodium hydrogen phosphate, pH 6.0) containing 10 mM EDTA, 30% fetal calf serum, and a protease inhibitor cocktail for 30 min at 30°C; and with 2.5% SDS in 0.1 M Tris-HCl (pH 8.0) at 60°C overnight. For labelling of GFP–Pma1, replicas were incubated at 4°C overnight with a rabbit anti-GFP antibody (kindly provided by Dr Masahiko Watanabe, Hokkaido University, Sapporo, Japan) in PBS containing 1% BSA followed by colloidal gold (10 nm)-conjugated protein A (PAG10; University of Utrecht, Utrecht, The Netherlands) for 60 min at 37°C in 1% BSA in PBS. Replicas were observed and imaged with a JEOL JEM-1011 EM (Tokyo, Japan) using a CCD camera (Gatan, Pleasanton, CA, USA).

Total lipids were extracted basically by the Bligh and Dyer method (Bligh and Dyer, 1959). Cells were grown in 100–200 ml of YPDA medium to 0.8–1.0 optical density at 600 nm (OD₆₀₀)/ml at 30°C or 37°C. The cells were collected and resuspended in 3.8 ml of chloroform/methanol/0.1 M HCl/0.1M KCl (1:2:0.8) and lysed by vortexing with glass beads for 1 min. Then, 1.0 ml each of chloroform and 0.1 M HCl/0.1 M KCl were added followed by centrifugation (2000 g for 5 min), isolation of the lipid-containing phase, and evaporation of the solvent. The extracted lipids were dissolved in an appropriate volume of chloroform. Total phospholipids were determined by performing a phosphorus assay (Rouser et al., 1970).

For phospholipid analysis, samples were subjected to thin-layer chromatography (TLC) on a TLC plate (Silica gel 60; Merck Millipore, MA, USA), with the solvent system chloroform/methanol/acetic acid/formic acid (50:30:4.5:6.5). After migration, plates were dried and sprayed with primuline solution (Sigma-Aldrich). Fluorescent spots of PE, PS, PI and PC were scraped into glass tubes and subjected to lipid extraction and phosphorus assay (Rouser et al., 1970). Phospholipid content was determined by three independent experiments. For ergosterol analysis, samples containing 20 nmol phosphate were subjected to TLC on an HPTLC plate (Silica gel 60; Merck Millipore, MA, USA) in the solvent system hexane/diethyl ether/acetic acid (80:20:1) (Dodge and Phillips, 1967). After migration, the plates were dried and sprayed with a 10% (w/v) cupric sulfate solution in 8% (w/v) orthophosphoric acid. Plates were heated in an oven at 180°C for 20 min. Plates were scanned with a CanoScan 8800F image scanner (Canon, Tokyo, Japan) and the acquired images were quantified using the ImageJ software.

Unless otherwise stated, more than 100 cells were used to measure cell counts, and the mean of the data was calculated based on three independent experiments.

Significant differences in Figs 2C, 5C and 7C were determined using a two-sided unpaired Student's t-test. Significant differences for all other figures were determined by the Tukey–Kramer test.

We thank Masahiko Watanabe (Hokkaido University, Sapporo, Japan) for providing the rabbit anti-GFP antibody for freeze-fracture replica labelling. We thank our colleagues in the Tanaka laboratory for valuable discussions and Eriko Itoh for technical assistance.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.256529