The functions of anionic phospholipids during clathrin-mediated endocytosis site initiation and vesicle formation

PI(4,5)P₂ and phosphatidylserine (PS) mainly localize to the cytosolic leaflet of the plasma membrane. The head groups of both PI(4,5)P₂ and PS are negatively charged, so their predominance in the cytosolic leaflet results in a net negative charge on the cytosolic face of plasma membrane, where the endocytic machinery localizes. These negatively charged lipids are well positioned to play important roles in endocytosis.

A variety of observations indicate that PI(4,5)P₂ plays key roles in CME. First, previous studies have suggested that PI(4,5)P₂ is enriched at endocytic sites (Fujita et al., 2009; Huang et al., 2004). Second, many endocytic proteins contain PI(4,5)P₂ binding domains and physically interact with PI(4,5)P₂ (Haucke, 2005). Third, alteration of PI(4,5)P₂ levels by genetic manipulation of phosphoinositide kinases or phosphoinositide phosphatases causes endocytic deficiencies (Abe et al., 2008; Desrivières et al., 2002; Stefan et al., 2002; Sun et al., 2007; Zoncu et al., 2007).

How PI(4,5)P₂ functions during distinct stages of the CME pathway is not well understood. Endocytic internalization involves endocytic site initiation (endocytic coat assembly), membrane invagination, vesicle scission, and endocytic coat disassembly. Zoncu et al. reported that PI(4,5)P₂ depletion in response to ionomycin, a strong activator of phospholipase C, or by translocation of the 5′-phosphatase domain of type IV 5-phosphatase to the plasma membrane results in a nearly complete block of clathrin assembly (Zoncu et al., 2007). In contrast, Abe et al. subsequently reported that reduction of plasma membrane PI(4,5)P₂ levels had only modest effects on clathrin assembly (Abe et al., 2008). Both groups found that reduction of PI(4,5)P₂ significantly blocks transferrin internalization, indicating a failure of CME. However, whether PI(4,5)P₂ is required for endocytic site initiation, and at what stages of internalization it might be required, remain to be addressed.

In budding yeast, PI(4,5)P₂ is generated by the sole phosphatidylinositol-4-phosphate 5-kinase, Mss4p, a protein that is essential for cell growth (Desrivières et al., 1998; Homma et al., 1998). PI(4,5)P₂ levels are greatly decreased in mss4^ts mutants (Desrivières et al., 1998; Homma et al., 1998; Stefan et al., 2002). Previous studies of mss4^ts mutants established that Mss4p is essential for proper organization of the actin cytoskeleton (Desrivières et al., 1998; Homma et al., 1998), and for both fluid-phase and receptor-mediated endocytosis (Desrivières et al., 2002). Because cortical actin patches serve as endocytic sites in yeast (Kaksonen et al., 2003), and because recent studies suggest that yeast and mammalian cells share many common components in the CME internalization process (Ferguson et al., 2009; Weinberg and Drubin, 2012), analysis of endocytic patch protein dynamics in the mss4^ts mutant at the nonpermissive temperature may provide novel insights into PI(4,5)P₂ function during CME.

The most abundant anionic phospholipid in the cytosolic leaflet of the plasma membrane, PS, has been proposed to play critical roles in specifying net membrane surface charge and in controlling peripheral membrane protein association with the plasma membrane (Yeung et al., 2008). Despite its likely importance for events on the inner leaflet of the plasma membrane, only a few studies have implicated PS in endocytic internalization. The PS translocase, known as ATP-binding cassette protein A1 (ABCA1), specifically transports PS from the inner leaflet of the plasma membrane to the exofacial leaflet. Loss of functional ABCA1 in Tangier fibroblasts leads to increased PS levels in the inner leaflet of the plasma membrane and enhances endocytosis (Zha et al., 2001). Also, endocytic internalization defects were observed in a triple deletion of the yeast genes DNF1, DNF2 and DRS2, which are involved in maintaining aminophospholipids including PS, in the inner leaflet of the plasma membrane (Pomorski et al., 2003). These results suggest that further investigation of how PS participates in endocytic internalization is needed.

In this study, we performed live cell imaging of endocytic actin patch dynamics in various yeast mutants in which plasma membrane and internal membrane lipid composition were altered. Our results provide insights into how two anionic phospholipids enriched on the cytosolic leaflet contribute to endocytic internalization.

To investigate PI(4,5)P₂ function in endocytic site initiation we first examined the dynamics of the endocytic coat protein Sla1p in both wild-type cells and in an mss4^ts mutant (Stefan et al., 2002) at the nonpermissive temperature. PI(4,5)P₂ levels are greatly decreased in this mss4^ts mutant at the nonpermissive temperature (supplementary material Fig. S1) (Stefan et al., 2002). To compare wild-type and mss4^ts mutant cells under identical conditions we mixed wild-type cells expressing Sla1-GFP with mss4^ts cells expressing both Sla1-GFP and Sac6-RFP (yeast fimbrin, representing actin patch dynamics) and used RFP fluorescence (red channel) to distinguish mss4^ts cells from wild-type cells (Fig. 1A). Sensitive tests for genetic interactions and endocytic assays indicated that the GFP and RFP tags do not affect Sla1 or Sac6 function (data not shown). Sla1-GFP dynamics were examined in the green channel (supplementary material Movie 1). In wild-type cells, Sla1-GFP patches first form at the plasma membrane and then move off the cortex toward the cell center (Fig. 1B; supplementary material Fig. S2, Movie 1), which represents membrane invagination, with a lifetime of 18.8±3.4 sec. However, in mss4^ts mutants, Sla1-GFP patches form at the plasma membrane, but they persist for twice as long (39.5±6.6 sec) and do not move off the cortex before they disassemble (Fig. 1C; supplementary material Fig. S2, Movie 1), indicating a failure in membrane invagination. We also observed similar Sla1-GFP dynamics in mss4^ts cells in which Sac6 is not tagged with RFP (data not shown). Thus, endocytic patches still form in mss4^ts cells at the nonpermissive temperature, indicating that PI(4,5)P₂ plays a non-essential role in endocytic site initiation. However, the reduction of PI(4,5)P₂ levels abolished endocytic patch inward movement, suggesting that PI(4,5)P₂ is required for endocytic membrane invagination.

In yeast, numerous studies have demonstrated that actin assembly is required for endocytic internalization (Kaksonen et al., 2003; Kübler and Riezman, 1993). Because mss4^ts mutant cells failed to undergo endocytic internalization, we sought to visualize actin dynamics by imaging Sac6-RFP in wild-type cells expressing Sla1-GFP and Sac6-RFP or in mss4^ts cells expressing only Sac6-RFP at the nonpermissive temperature (Fig. 1D; supplementary material Movie 2). Sac6p is budding yeast fimbrin, an actin filament crosslinking protein. In wild-type cells, Sac6-RFP patches formed at the cell cortex and then rapidly moved off (Fig. 1E; supplementary material Movie 2), representing membrane invagination and vesicle release upon scission. However, in the same field, ∼60% of cortical actin puncta in mss4^ts mutants displayed comet tail-like actin structures anchored to the cortex that waved in the cytoplasm (Fig. 1F; supplementary material Movie 2). Two-color live cell imaging of mss4^ts cells revealed that Sla1-GFP appears prior to Sac6-RFP (Fig. 1H,I), similar to wild-type cells (Fig. 1G), but no longer displays inward movement. Furthermore, consistent with single color imaging analyses, Sac6-RFP appears to emanate from Sla1-GFP foci on the cell cortex in tail-like structures that eventually disappear (Fig. 1H,I).

Tail-like actin structures have been observed previously in SLA2 (yeast Hip1R) deletion mutants (Kaksonen et al., 2003), similar to mammalian cells depleted for Hip1R (Engqvist-Goldstein et al., 2004). Moreover, previous studies indicated that the ANTH domain of Sla2p can bind to PI(4,5)P₂ in vitro and plays an essential role in endocytic invagination (Sun et al., 2005). Because reduction of PI(4,5)P₂ levels in mss4^ts cells may affect Sla2p recruitment to endocytic sites and therefore indirectly cause the actin comet tail phenotype, we examined Sla2-GFP dynamics in the mss4^ts mutant at the nonpermissive temperature (Fig. 1J; supplementary material Movie 3). The lifetime of Sla2-GFP in mss4^ts cells became about twice as long as it is in wild-type cells (38.1±16.1 sec vs 21.6±10.6 sec). However, as shown in Fig. 1L, Sla2-GFP patches still assemble and disassemble in mss4^ts cells (supplementary material Movie 3) despite a reduction in PI(4,5)P₂ levels. These results indicate that Sla2p absence or PI(4,5)P₂ reduction at endocytic sites causes similar defects in actin organization, consistent with the idea that the PI(4,5)P₂-Sla2p interaction is required for proper actin regulation during endocytic membrane invagination.

The above experiments were conducted using mss4^ts mutants at the nonpermissive temperature (38°C). One complication of performing experiments at this temperature is that the Sla1-GFP lifetime of wild-type cells is shorter at 38°C than at 25°C (18.8±3.4 sec vs 26.8±5.6 sec). Additionally, while PI(4,5)P₂ levels are greatly reduced (supplementary material Fig. S1) (Stefan et al., 2002), they may not be completely depleted by shifting mss4^ts cells to the nonpermissive temperature. To further investigate lipid requirements during endocytic internalization, we examined endocytic patch formation in an rcy1Δ mutant at room temperature.

Rcy1p (recycling 1) is an F-box protein that is involved in an early post-internalization step of endocytosis, before the intersection of the endocytic and VPS (Vacuolar Protein Sorting) pathways (Wiederkehr et al., 2000). Rcy1p was shown to mediate endosome-to-Golgi trafficking (Chen et al., 2005). In rcy1Δ mutants grown at room temperature, endosome-to-Golgi transport is blocked, resulting in formation of abnormal membrane compartments that are likely derived from early endosomes (Chen et al., 2005; Wiederkehr et al., 2000). Interestingly, rhodamine phalloidin staining revealed localization of actin patches in the cytoplasm of rcy1Δ cells at the nonpermissive temperature (18°C) (Kishimoto et al., 2005). In our strain background, we observed this intracellular actin patch phenotype at room temperature. Presumably, the intracellular actin patches form on certain intracellular membrane compartments. Thus, the rcy1Δ mutant may provide an opportunity to examine the lipid requirements for actin patch formation.

We generated rcy1Δ strains, which expressed Sla1-GFP and Sac6-RFP from the native genomic locus for each protein. When we imaged the medial focal plane of wild-type cells using wide-field fluorescence microscopy, we only observed Sla1-GFP and Sac6-RFP patches at the cell cortex (Fig. 2A; supplementary material Movie 4) as shown in a previous study (Kaksonen et al., 2003). However, in rcy1Δ cells, Sla1-GFP and Sac6-RFP patches were not only present at the cell cortex, but also in the cytoplasm (Fig. 2B, left panel; supplementary material Movie 4). Furthermore, two-color live-cell imaging analysis of rcy1Δ cells revealed that Sla1-GFP patches were always joined by Sac6-RFP near the time of Sla1-GFP disappearance, regardless of the patch's localization (Fig. 2B, right panel). This result suggests that the intracellular actin patches assemble and change protein composition in a predictable, dynamic manner, similar to normal cortical endocytic sites. The cortical Sla1-GFP/Sac6-RFP patches in rcy1Δ cells make normal inward movements (Fig. 2B, right panel), consistent with previous observations that deletion of RCY1 does not affect endocytic internalization but rather does affect a post-internalization step (Wiederkehr et al., 2000). By contrast, the intracellular Sac6-RFP patches often persisted much longer (15.3±3.2 sec in rcy1Δ vs 8.8±2.2 sec in wild-type cells) after the Sla1-GFP patch disappeared and exhibited no obvious movements as they disappeared (Fig. 2B, right panel), suggesting that membrane invagination and scission may not occur at these sites.

Previous reports showed that Rcy1p binds to active, GTP-bound Ypt31/32 GTPases and functions as a downstream effector of Ypt31p/32p in the regulation of protein recycling from the endosome to the Golgi (Chen et al., 2005). Neither Ypt31p nor Ypt32p is required for cell growth, but absence of both proteins causes lethality (Benli et al., 1996; Jedd et al., 1997). Given that Ypt31p/32p functions in the same pathway with Rcy1p, we examined whether deletion of YPT31/32 might also induce intracellular actin patch formation, as seen in rcy1Δ cells. The GAL1 promoter was chromosomally integrated to control YPT31 expression in a ypt32Δ strain. In galactose-containing media, this strain grew well and formed actin patches only at the cell cortex (Fig. 2C; supplementary material Movie 5). However, repression of Ypt31p expression by shifting to glucose media for 8 hours resulted in the appearance of intracellular actin patches (Fig. 2D; supplementary material Movie 5), with Sla1-GFP appearing first followed by Sac6-RFP.

A recent study proposed that putative phospholipid translocases (PLTs), such as Drs2p and Dnf1p/2p/3p, are also components of the Ypt31p/32p-Rcy1p pathway (Furuta et al., 2007). In addition, Cdc50p was shown to be a noncatalytic subunit of Drs2p (Saito et al., 2004). Interestingly, intracellular actin patches have been observed in an erg3Δ cdc50Δ double deletion mutant, but not in the single knockouts (Kishimoto et al., 2005). However, in our background, CDC50 single deletion mutants exhibit intracellular actin patches, similar to those seen in rcy1Δ cells (Fig. 2E; supplementary material Movie 6).

Together, our results thus far show that the appearance of intracellular actin patches is a common phenotype among cells lacking Ypt31p/Ypt32p, or Rcy1p, or Cdc50p, providing further support for the notion that all of these proteins function in same pathway. We next sought to determine whether the intracellular actin patches undergo similar dynamic changes in protein composition to normal plasma membrane-associated endocytic actin patches. We previously classified many endocytic proteins as each belonging to one of five modules (Kaksonen et al., 2005; Stimpson et al., 2009): the early coat module, coat module, Wasp/Myo module, amphiphysin module, and the actin module. We co-expressed GFP- and mCherry- (or RFP) tagged pairs of endocytic proteins belonging to different modules in an rcy1Δ strain and performed live-cell imaging analyses on the intracellular actin patches (Fig. 3). Ede1 represents the early coat module (Stimpson et al., 2009). As shown in Fig. 3A, intracellular Sla1-mCherry patches always assemble on Ede1-GFP patches. In some cases, Sla1-mCherry and Ede1-GFP patches disappeared with similar timing. In other cases, Sla1-mCherry patches assembled and disassembled repeatedly on Ede1-GFP patches. We obtained similar results when we examined another early coat protein, Apl1 (a subunit of AP-2) (Fig. 3B). The coat module protein Sla2-GFP joins intracellular patches with similar timing to Sla1-mCherry (Fig. 3C). Las17-GFP and Myo5-RFP, which belong to the Wasp/Myo module, assemble sequentially at the intracellular patches (Fig. 3D), as they do at plasma membrane patches (Sun et al., 2006). Finally, Rvs167-GFP (Amphiphysin module) arrives at intracellular patches with similar timing to the actin module protein Sac6-RFP (Fig. 3E). Together, these results establish that intracellular actin patches in various mutants of the Ypt31p/32p-Rcy1p-PLTs pathway display similar proteins and temporal recruitment order to cortical endocytic actin patches. Thus, intracellular actin patches may be considered as endocytic patch-like structures.

Lack of Rcy1p (or cdc50p, or Ypt31p/32p) not only induces internal actin patches (Fig. 2), but also induces formation of enlarged early endosome membrane compartments, which are likely derived from early endosomes (Chen et al., 2005; Kishimoto et al., 2005; Wiederkehr et al., 2000). We therefore examined whether the intracellular actin patches form on the enlarged early endosome membrane compartments in an rcy1Δ mutant. The exocytic v-SNARE Snc1p is known to be recycled from the plasma membrane to the TGN via early endosomes. Most (78%) of the intracellular actin patches formed on Snc1-GFP positive structures in rcy1Δ mutants (supplementary material Fig. S3A), suggesting that the intracellular endocytic patch-like structures assemble on the previously-described, enlarged early endosome membrane compartments.

Since endocytic patch-like structures form on the enlarged early endosome membrane compartments in rcy1Δ mutants, we next assessed the presence of PI(4,5)P₂ on these membrane compartments. GFP-2XPH (PLCδ) is widely used to visualize PI(4,5)P₂ localization in live cells because the PLCδ PH domain binds specifically to PI(4,5)P₂ (Stefan et al., 2002). Previous studies indicated that PI(4,5)P₂ is restricted largely to the plasma membrane in wild-type cells (Stefan et al., 2002) (supplementary material Fig. S1). In rcy1Δ cells, we found that GFP-2XPH is restricted to the plasma membrane, while Sla1-mCherry patches assemble and disassemble on both the plasma membrane and on the intracellular compartments (Fig. 4A,B). Consistent with this result, in rcy1Δ cells GFP-tagged Mss4p, which is the sole yeast phosphatidylinositol-4-phosphate 5-kinase, was detected at the plasma membrane but not on any intracellular compartments (Fig. 4C). Thus, endocytic actin-like structures form on intracellular membrane compartments that do not contain detectable PI(4,5)P₂. These results demonstrate that actin patch initiation and sequential endocytic protein recruitment can occur on membranes lacking detectable PI(4,5)P₂, suggesting that PI(4,5)P₂ does not play an obligatory role in these processes, in agreement with our findings from mss4^ts experiments (Fig. 1).

Cdc50p is a critical cofactor for Drs2p, which is a translocase specific for phosphatidylserine (PS) (Natarajan et al., 2004). Since intracellular endocytic patch-like structures localize on the enlarged early endosome membrane compartments in cdc50Δ, or rcy1Δ, or ypt31/2Δ mutants (supplementary material Fig. S3A and data not shown), we sought to assess the presence of PS on these membrane compartments. Previously, Yeung et al. developed a probe, GFP-Lact-C2, to monitor the endogenous PS distribution and found that PS is most abundant at the plasma membrane in both mammalian and yeast cells (Yeung et al., 2008). We also observed that GFP-Lact-C2 predominantly localizes to the plasma membrane of wild-type yeast (Fig. 5A). Strikingly, GFP-Lact-C2 is detected on both the plasma membrane and on intracellular membrane compartments in rcy1Δ cells (Fig. 5B, left panel). Kymograph analysis revealed that the GFP-Lact-C2 labeled intracellular structures are relatively stable and do not move dramatically over time (Fig. 5B, right panel).

To determine whether the intracellular actin patches form on the GFP-Lact-C2 labeled structures, we co-expressed GFP-Lact-C2 and Sla1-mCherry in wild-type cells and in rcy1Δ cells. In wild-type cells, both GFP and mCherry signals were exclusively associated with the cell cortex, and Sla1-mCherry patches underwent normal inward movement (Fig. 5C,D; supplementary material Movie 7), showing that GFP-lact-C2 expression does not affect endocytic internalization. In rcy1Δ cells, most of the intracellular patches (∼80%) formed on the GFP-lact-C2 positive internal structures (Fig. 5C,D). Similar results were also obtained when we co-expressed GFP-Lact-C2 and Ede1-RFP in rcy1Δ cells (supplementary material Fig. S3B). In addition to our observation that the intracellular endocytic patch-like structures formed on Snc1-GFP positive structures (supplementary material Fig. S3A), GFP-Lact-C2 was shown to co-localize with mRFP-Snc1 on the internal structures in rcy1Δ cells (Fairn and Grinstein, 2010), further confirming that the intracellular actin patches form on the enlarged early endosome membrane compartments, which appear enriched for PS.

In wild-type yeast cells, PS mainly localizes on the plasma membrane, where it can amount to more than 30% of the total phospholipid (Zinser and Daum, 1995). One possibility is that PS serves as an indicator of plasma membrane identity, restricting certain events to the plasma membrane. In rcy1Δ mutants, PS appears at both the plasma membrane and the enlarged early endosome membrane compartments, and endocytic proteins assemble on both membrane structures. A recent study suggested that in budding yeast PS concentrates at incipient bud sites and small buds (Fairn et al., 2011), which are the same regions of the yeast cell surface where endocytic actin patches are most highly concentrated (Adams and Pringle, 1984; Kilmartin and Adams, 1984). To test whether PS deficiency affects endocytic actin patch formation, we examined cells lacking Cho1p, a key enzyme that synthesizes PS (Hikiji et al., 1988). In cho1Δ cells, GFP-Lac-C2 no longer localizes to the plasma membrane, but instead is cytosolic (Yeung et al., 2008) (Fig. 6B), suggesting that plasma membrane levels of PS are significantly reduced. We imaged in the same field cho1Δ cells endogenously expressing both Sla1-GFP and Sac6-RFP, and wild-type cells endogenously expressing Sla1-GFP (Fig. 6C), distinguishing cho1Δ cells from wild-type cells by using the red channel (data not shown). Control studies showed that Sac6-RFP did not affect endocytic protein behavior (data not shown). Maximum intensity projections from a two-minute video (supplementary material Movie 8) show that in cho1Δ cells Sla1-GFP no longer localizes preferentially in the small bud (Fig. 6D), the site where endocytic actin patches normally concentrate. Among 50 small budded cho1Δ cells we have examined, 85% of the cells lost polarized patch localization. Two-color live-cell imaging revealed that when Sla1-GFP patches formed on the cell cortex, they still underwent inward movement after Sac6-RFP recruitment (Fig. 6E). However, the lifetime of Sla1-GFP is longer in cho1Δ cells than in wild-type cells (36.6±8.7 vs 27.5±5.7 sec, P<0.001). These results suggest that PS may contribute to the recruitment of endocytic proteins to the plasma membrane and affect the efficiency of endocytic patch establishment, especially in the daughter cells. However, once an endocytic actin patch assembles, depletion of PS from the plasma membrane does not prevent endocytic membrane invagination.

In this study, we combined live-cell imaging and yeast genetics to examine how two anionic phospholipids enriched on the cytosolic leaflet of the plasma membrane, PI(4,5)P₂ and PS, function to control endocytosis. Our data suggest that while PI(4,5)P₂ is essential for endocytic membrane invagination, it does not appear to play obligatory role in endocytic site initiation. Furthermore, our results suggest that PS is important for the efficient recruitment and spatial restriction of endocytic proteins to the plasma membrane.

Using chemical enzyme-translocation strategies to rapidly hydrolyze PI(4,5)P₂ at the plasma membrane, Zoncu et al. and Abe et al. provided strong evidence that PI(4,5)P₂ plays essential roles during endocytosis in mammalian cells (Abe et al., 2008; Zoncu et al., 2007). However, whether PI(4,5)P₂ is required for endocytic site assembly in mammalian cells remains controversial (Abe et al., 2008; Zoncu et al., 2007). We performed two sets of in vivo experiments to address this question in yeast, where most of the endocytic proteins have mammalian counterparts. First, endocytic patch proteins, including Sla1p, Sac6p, and Sla2p, still form patches when PI(4,5)P₂ is severely depleted in mss4^ts cells at the nonpermissive temperature (Fig. 1). Second, intracellular endocytic actin patch-like structures, which we showed are similar in many respects to cortical endocytic patches, assemble in rcy1Δ cells on enlarged early endosome membrane compartments lacking PI(4,5)P₂ detectable by GFP-2XPH. We cannot completely rule out the possibility that the absence of GFP-2XPH on intracellular membranes is caused by the absence of some plasma membrane associated co-receptor for GFP-2XPH binding. However, in the absence of Rcy1p, Sjl2p (inositol-polyphosphate 5-phosphatase) is still recruited to cortical endocytic patches (data not shown), which disassemble normally after internalization. Thus, Sjl2p likely hydrolyzes PI(4,5)P₂ on endocytic vesicles released from the plasma membrane in rcy1Δ cells. Since the enlarged early endosome membrane compartments observed in rcy1Δ cells presumably evolved from cortical endocytic vesicles, it is likely that PI(4,5)P₂ is depleted on the these membrane compartments. Furthermore, we did not observe Mss4p on the enlarged early endosome membrane compartments, suggesting that no PI(4,5)P₂ is made on these membranes. Based on these results, we conclude that PI(4,5)P₂ does not play obligatory roles at the endocytic site initiation step in yeast. Our results are consistent with the findings of Abe et al. in mammalian cells, wherein reduction of PI(4,5)P₂ levels resulted in only a minor clathrin assembly defect (Abe et al., 2008).

In mammalian cells, PI(4,5)P₂ depletion resulted in near absence of AP-2 from the plasma membrane (Abe et al., 2008; Zoncu et al., 2007). However, we found that Apl1p, which is a subunit of yeast adaptor protein 2 (AP-2), still localizes to the plasma membrane in mss4^ts cells cultured at the nonpermissive temperature (data not shown). Additionally, Apl1p also formed intracellular patches on the enlarged early endosome membrane compartments lacking PI(4,5)P₂ (Fig. 3B). Thus, AP-2 recruitment in yeast cells does not appear to require PI(4,5)P₂.

Mss4p localizes at the plasma membrane (Desrivières et al., 1998; Homma et al., 1998; Sun et al., 2007), but Mss4p patches do not colocalize with endocytic actin patches (Sun et al., 2007), suggesting that endocytic sites do not form where PI(4,5)P₂ is synthesized. We speculate that PI(4,5)P₂ may accumulate at endocytic sites concomitantly with endocytic proteins. Although PI(4,5)P₂ is not required for endocytic coat recruitment, the lifetime of Sla1 patches becomes prolonged when PI(4,5)P₂ levels are reduced (Fig. 1), suggesting that PI(4,5)P₂ plays a critical role in endocytic protein activity. Recruitment of PI(4,5)P₂ and endocytic coat proteins may be synergistic. PI(4,5)P₂ may be enriched to maximum levels during endocytic site assembly and membrane invagination, similar to endocytic coat proteins (Sun et al., 2007). In agreement with this hypothesis, Sla1p inward movement, which likely represents membrane invagination, is abolished in PI(4,5)P₂-deficient cells (Fig. 1C), suggesting that PI(4,5)P₂ is required for endocytic membrane invagination. Furthermore, depletion of either Sla2p or PI(4,5)P₂ causes a similar “actin tail” phenotype (Fig. 1F) (Kaksonen et al., 2003), raising the possibility that the interaction between PI(4,5)P₂ and Sla2p at endocytic sites is required for endocytic membrane invagination.

Previous studies in which PS translocases were genetically manipulated suggested that the asymmetric distribution of PS between plasma membrane leaflets plays an important endocytic role in yeast and mammals (Farge, 1995; Farge et al., 1999; Pomorski et al., 2003; Zha et al., 2001). The fluorescent biosensor GFP-Lact-C2 makes visualization of PS distribution possible in live cells (Yeung et al., 2008). A recent yeast study using this probe showed that PS is confined to the plasma membrane, preferentially accumulating at the cortex of small buds and at the neck of growing buds (Fairn et al., 2011). We observed a similar localization in wild-type cells (Fig. 5A; Fig. 6A). Interestingly, PS levels have been reported to peak at the time of bud emergence during the yeast cell cycle (Cottrell et al., 1981). Because endocytic actin patch formation is polarized to buds, it is reasonable to ask if endocytic actin patch assembly is linked functionally to PS accumulation. To examine PS function in vivo, one useful tool is the PS-deficient mutant cho1Δ. We found a loss of patch polarization to the small bud and a substantial increase of patch lifetime in cho1Δ cells. A recent study revealed that the absence of Cho1p not only causes reduced PS, but also reduced phosphatidylethanolamine (PE) in yeast (Fairn et al., 2011). However, cells lacking Psd1p, which is a main phosphatidylserine decarboxylase, show a marked decrease in PE and a slight increase in PS (Fairn et al., 2011). In contrast to cho1Δ cells (Fig. 6), psd1Δ cells show normal actin patch establishment (data not shown), suggesting that PS plays a more critical role than PE in endocytic site establishment in the small buds of yeast. We propose that PS plays a role in guiding/restricting endocytic proteins to the plasma membrane. Higher concentrations of PS in the small bud and bud neck may efficiently recruit endocytic proteins to these locations. At least two other lines of evidence support this idea. First, the loss of functional ABCA1 in Tangier fibroblasts, which leads to more PS in the inner leaflet to the membrane, enhances endocytosis (Zha et al., 2001). Second, Itoh et al. showed that endocytic F-BAR protein binding to phospholipid bilayers in vitro requires the presence of PS and is enhanced by phosphoinositides (Itoh et al., 2005).

Further studies are required to determine how PS affects endocytic protein targeting. The negative charge associated with PS may contribute to the membrane targeting of endocytic proteins through polybasic clusters or cationic domains on the endocytic proteins. It is also worth noting that a recent study showed that proper Cdc42 localization is lost in cho1Δ cells, leading the authors to suggest that PS may affect this localization (Fairn et al., 2011). Thus, whether PS regulates endocytic patch initiation through the Cdc42 pathway must also be investigated.

Whether PS and PI(4,5)P₂ act redundantly during endocytic site formation is not clear. A cho1Δ mss4^ts double mutant shows a severe growth defect at semi-restrictive temperature (34°C), while both cho1Δ cells and mss4^ts cells grow relatively normally at the same temperature (supplementary material Fig. S4A). This result supports the conclusion that PS and PI(4,5)P₂ may share functions in some cellular processes. However, actin patches were still observed in cho1Δ mss4^ts cells at the restrictive temperature (38°C) (data not shown), raising the possibility that PS and PI(4,5)P₂ may not be the only lipids capable of supporting endocytic site initiation.

Previous evidence suggested that Rcy1p and Ypt31/32 GTPases work together to mediate endosome-to-Golgi transport (Chen et al., 2005). In addition, a recent study suggested that Cdc50p, which is a noncatalytic subunit of the phospholipid translocase Drs2p, also functions in the Rcy1-Ypt31/32 pathway (Furuta et al., 2007). Here, we showed that depletion of Ypt31/32 GTPases or Cdc50p shares a common phenotype with rcy1Δ cells, the appearance of dynamic, intracellular endocytic actin patch-like structures. Strikingly, these intracellular actin patches formed on internal membranes labeled by the PS probe in each of these various deletion mutants (Fig. 5 and data not shown). Furthermore, the PS probe co-localized with Snc1p on the internal membranes (Fairn and Grinstein, 2010), supporting the conclusion that these internal membrane compartments are related to endosomes. Thus, our results provide strong evidence that Rcy1p, Ypt31p/32p and Cdc50p contribute to the same biological process regulating endosome-Golgi trafficking. In addition, these results imply that PS asymmetry plays important roles in endosome-Golgi transport. The cho1Δ rcy1Δ double mutant in principle would be a good tool for further investigations of this possibility. However, knocking out both CHO1 and RCY1 causes an extremely severe growth defect (supplementary material Fig. S4B), making this approach unfeasible.

Why would “endocytic patches” form dynamically on the internal membranes in the mutants of the Rcy1-Ypt31/32-PLTs pathway? Normally, both PI(4,5)P₂ and PS are only enriched on the cytosolic leaflet of plasma membrane. If there are no lipid composition changes during endocytic internalization, PI(4,5)P₂ and PS would be on the cytosolic leaflet of pinched off endocytic vesicles. We speculate that PI(4,5)P₂ and PS need to be removed from the cytosolic leaflet of pinched off endocytic vesicles, presumably by inositol-polyphosphate 5-phosphatase and PS translocase, respectively, while the vesicles evolve into early endosomes, and then other membrane compartments. In this way, endocytic proteins are restricted to the plasma membrane. PS may function to distinguish the plasma membrane from other organelle membranes. We speculate that in the absence of a functional Rcy1p-Ypt31/2 pathway, PS cannot be removed from the cytosolic leaflet of early endosomes, and the resulting presence of PS on the intracellular membrane compartment recruits endocytic proteins to this compartment.

The GFP-2xPH PLCδ plasmid and mss4^ts plasmid were kindly provided by S. Emr (Cornell University). The GFP-Lact-C2-p416 plasmid was obtained from Haematologic Technologies. Yeast strains used in this study are listed in supplementary material Table S1. C-terminal GFP, RFP or mCherry tags were integrated by homologous recombination, as described previously (Sun et al., 2007).

Yeast strains were grown in standard rich media (YPD) or synthetic media supplemented with the appropriate amino acids. For induction of the GAL1 promoter, 3% galactose and 0.5% sucrose were used as carbon sources. For experiments involving cho1Δ, synthetic media was supplemented with 1 µM choline.

Cells were attached to concanavalin A-coated coverslips, which were sealed to slides with vacuum grease (Dow Corning). All imaging studies were performed at 25°C or 38°C using an Olympus IX81-OMAC microscope equipped with a temperature-controlled enclosure (Precision Control Weather Station), 100×NA 1.4 objective and Orca ER camera (Hamamatsu).

Two-color images were either obtained by sequential acquisition using RFP and GFP filter sets or by simultaneous acquisition using an image splitter as previously described (Kaksonen et al., 2005). Image analysis was performed with ImageJ (http://rsbweb.nih.gov/ij/) as described previously (Kaksonen et al., 2003). Particle tracking analysis was performed by Imaris software.

We thank Scott Emr for providing plasmids. We also thank Betsy Wong, Takuma Kishimoto, and Susheela Carroll for providing yeast strains. We are grateful to Alex Grassart for advice on the particle tracking analysis. We thank the members of the Drubin lab for helpful discussions, and Aaron Cheng, Ann Marie Faust, Connie Peng, Christa Cortesio, and Eric Lewellyn for critical reading of the manuscript.