PI3KC2α, a class II PI3K, is required for dynamin-independent internalization pathways

Endocytosis of molecules from the extracellular milieu can occur through distinct internalization routes. The clathrin- and caveolin-mediated pathways, which require dynamin, a large GTPase responsible for the scission of newly formed vesicles at the plasma membrane, are by far the best characterized (Conner and Schmid, 2003; Nabi and Le, 2003). Endocytic mechanisms that occur independently of clathrin and caveolin but also require functional dynamin, (Mayor and Pagano, 2007), for example the internalization of interleukin 2 (Lamaze et al., 2001), high-affinity immunoglobulin E receptor (Fattakhova et al., 2006) and γc cytokine receptor (Sauvonnet et al., 2005), have also been identified. However, growing numbers of molecules are reported to be internalized in a dynamin-independent manner, such as GPI-anchored proteins (GPI-APs) (Mayor and Pagano, 2007), major histocompatibility complex I (MHCI) (Radhakrishna and Donaldson, 1997) and various bacterial toxins and viruses (Damm et al., 2005; Sandvig and van Deurs, 2008; Sandvig and van Deurs, 2002). Dynamin-independent internalization occurs through specific compartments, whose formation and dynamics are regulated by small GTPases. Two main routes of cargo internalization have been identified: in HeLa cells, the Arf6-dependent pathway regulates the entry of MHCI and CD59 (Donaldson et al., 2009), and the CLIC–GEEC (clathrin-independent carrier–GPI-AP-enriched early endosomal compartments) pathway (Kirkham and Parton, 2005; Mayor and Riezman, 2004; Sabharanjak et al., 2002), which is regulated by Cdc42, Arf1 and ARHGAP10 (Chadda et al., 2007; Kumari and Mayor, 2008). GEECs are required for the internalization of a variety of molecules, including GPI-anchored proteins (Sabharanjak et al., 2002), cholera toxin (at least in part) (Kirkham and Parton, 2005) and VacA toxin (Gauthier et al., 2005), and, recently, GRAF1 (GTPase regulator associated with focal adhesion kinase-1) has been identified as a non-cargo marker (Lundmark et al., 2008) and proposed to be a regulator of this dynamic tubular compartment. The CLIC–GEEC pathway also internalizes a large fraction of the fluid phase into the cells (Lundmark et al., 2008; Sabharanjak et al., 2002).

With the aim of identifying new proteins acting in the dynamin-independent pathways, we performed a survival-based sh-RNA screen in a cell line expressing a GPI-anchored diphtheria toxin (DT) receptor that guides DT entry in a dynamin-independent fashion (Skretting et al., 1999). PI3KC2α (phosphatidylinositol 3-kinase, class 2, alpha polypeptide) was identified as a positive regulator of DT internalization and further validated by survival assays and immunofluorescence analyses. PI3KC2α belongs to the class II phosphoinositide 3-kinases (PI3Ks), comprising three proteins (α, β, γ), which are resistant to the classic PI3K inhibitors (Domin et al., 1997; Falasca and Maffucci, 2007). PI3KC2α is a high molecular weight monomer with a C2 domain at the C-terminus and a clathrin-binding domain at the N-terminus. PI3KC2α has been implicated in several cellular processes, including clathrin-mediated endocytosis, (Gaidarov et al., 2005), ATP-dependent priming of neurosecretory vesicles (Meunier et al., 2005; Wen et al., 2008), response to insulin stimulation (Falasca et al., 2007) and vascular smooth-muscle contraction (Wang et al., 2006). Studies in vivo identified PI3P as the only product of PI3KC2α, and it has been reported that insulin and Ca²⁺ can induce PI3P production in a PI3KC2α-dependent fashion at the plasma membrane and secretory vesicles, respectively (Falasca et al., 2007; Maffucci et al., 2005; Wen et al., 2008). Our data indicate that PI3KC2α also plays an essential role in the dynamin-independent physiological internalization processes by promoting recruitment of early endosome antigen 1 (EEA1) to internalizing vesicles, a step required for the intracellular trafficking of vesicles generated by dynamin-independent processes.

On the basis of previous results that has shown that several bacterial proteins can be successfully used to study the mechanisms of internalization and trafficking (Sandvig and van Deurs, 2002), we utilized the cellular toxicity of diphtheria toxin (DT, also known as DTX) to identify molecules required for its internalization. DT enters the cell through a heparin-binding EGF-like growth factor (HB-EGF) precursor (Moya et al., 1985; Naglich et al., 1992; Simpson et al., 1998) named the diphtheria toxin receptor (DTR, officially known as HBEGF). After receptor binding, the toxin is delivered to early endosomes before it translocates to the cytosol, causing cell death (Falnes and Sandvig, 2000). It was reported previously that the tetracycline-regulated expression of a mutant form of dynamin I (K44A) in HeLa cells prevented cell intoxication, suggesting that DT enters the cell using a dynamin-dependent pathway (Fig. 1) (Lanzrein et al., 1996; Skretting et al., 1999). However, when the transmembrane and cytoplasmic domains of the DT receptor were replaced with a GPI anchor (DTR–GPI), the induced expression of the DynK44A mutant did not prevent cell intoxication, suggesting that the modified DTR allowed DT internalization independently of dynamin (Lanzrein et al., 1996; Skretting et al., 1999). To visualize the internalization of DT in the presence or absence of mutant dynamin, we monitored the uptake of a nontoxic mutant of DT, CRM197 (Uchida et al., 1972), coupled to a Cy3 fluorophore (DT–Cy3). In HeLa DynK44A cells, which express only the wild-type DT receptor, the induction of mutant dynamin prevented DT–Cy3 entry, and no intracellular fluorescence was detected at any of the time-points tested (Fig. 1A, upper panel; and data not shown). In agreement with previous results (Skretting et al., 1999), it appears that DynK44A expression prevented cell intoxication, and protein synthesis was not inhibited by toxin treatment (Fig. 1B). By contrast, in HeLa DynK44A DTR–GPI cells, which express a GPI-linked version of the DT receptor, DynK44A induction did not prevent DT–Cy3 internalization (Fig. 1A, lower panel), as indicated by the intracellular fluorescence observed after DT–Cy3 treatment and by the sensitivity to intoxication with DT (Fig. 1B). Similarly, the overexpression of a dominant-negative construct for Eps15, which inhibits clathrin-mediated endocytosis (Carbone et al., 1997), also failed to prevent DT internalization in this cell line (supplementary material Fig. S1A). Antibodies against DTR were internalized in a dynamin-independent manner (supplementary material Fig. S1B) and colocalized with DT–Cy3 both at the plasma membrane and in vesicular compartments (supplementary material Fig. S1C). These data provide further evidence for dynamin-independent internalization of the toxin and its modified receptor.

In HeLa cells expressing DynK44A and DTR–GPI, DT–Cy3 appeared in elongated structures emanating from the plasma membrane after 10 minutes of loading in the media (Fig. 1C, left) and accumulated in large vesicles at the perinuclear region of the cell by 20–30 minutes of continuous internalization (Fig. 1C, right). At 10 minutes, in the cells expressing DynK44A, DT–Cy3 colocalized with molecules reported to internalize in a dynamin-independent manner, such as CD59 (Naslavsky et al., 2004), CtxB, (Glebov et al., 2006), MHCI (Radhakrishna and Donaldson, 1997) and GPI–GFP (Sabharanjak et al., 2002) but not with transferrin, a marker for clathrin-mediated (and dynamin-mediated) endocytosis (Fig. 2). These experiments indicate that DT internalization through a GPI-anchored receptor can be used to investigate internalization pathways occurring independently of dynamin.

In an attempt to identify the molecular components involved in dynamin-independent pathways, we designed a screen exploiting the cellular toxicity of DT and its ability to be internalized in a dynamin-independent way in HeLa DynK44A DTR–GPI cells. The screen, briefly described in the Material and Methods section, was based on cell survival after transfection of HeLa DynK44A DTR–GPI cells with a library of sh-RNAs, followed by induction of DynK44A expression and intoxication with DT. As cell death occurs when the DT is internalized and transported to early endosomal compartments, genes that prevented cell intoxication when downregulated would potentially encode proteins involved in dynamin-independent DT internalization and trafficking from the plasma membrane to early endosomes.

Among several proteins (listed in supplementary material Table S1), PI3KC2α, a member of class II of the phosphoinositide 3-kinases (PI3Ks), was identified in the screen. To confirm the results of the screen, HeLa DynK44A DTR–GPI cells were transfected with a pRETROSUPER–PI3KC2α silencing construct (pRS–PI3KC2α), and cell survival was compared with that of cells transfected with an empty control vector as well as with that of cells not expressing mutant dynamin. Cells that downregulated PI3KC2α while simultaneously expressing DynK44A showed improved survival after intoxication with DT to a much higher extent than control cells (Fig. 3A, left). Similar results were obtained after transfecting the cells with a silencing construct, pGIPZ–PI3KC2α, from an Open Biosystems library, which targets a different sequence of the gene encoding PI3KC2α, possibly ruling out off-target effects (Fig. 3A, right). The silencing efficiency of the pRETROSUPER–PI3KC2α and pGIPZ–PI3KC2α constructs was tested by quantitative PCR and western blotting (supplementary material Fig. S2A–D).

To validate further PI3KC2α as a component in dynamin-independent pathways, we performed an internalization assay at a single-cell level. DT uptake and trafficking were followed by fluorescence microscopy using DT–Cy3 in cells transfected with the pGIPZ–PI3KC2α or nonsilencing constructs, identified as GFP-positive cells. As shown in Fig. 3B, silencing of PI3KC2α strongly reduced the DT–Cy3 fluorescence in cells that also expressed DynK44A. Quantitative analysis showed that the level of DT–Cy3 in sh-PI3KC2α-transfected cells, analyzed by measuring the internalized fluorescence at 30 minutes, was reduced by ~70% in comparison with nonsilencing transfected cells (Fig. 3B,D). DT internalization was almost completely restored to wild-type levels when a silencing-resistant form of PI3KC2α (PI3KC2α shRes) was co-transfected with the shRNA construct (Fig. 3C,D), thus confirming the role for PI3KC2α in DT uptake. The silencing effect of the pGIPZ–PI3KC2α construct on wild-type or the RNAi-resistant form of PI3KC2α was tested by western blot (supplementary material Fig. S2F). Importantly, the knockdown of PI3KC2α did not interfere with the total DT–Cy3 fluorescence (supplementary material Fig. S3A), measured after loading the DT–Cy3 at 4°C for 30 minutes. This indicates that the DT receptors are correctly transported to the plasma membrane in cells expressing the sh-PI3KC2α construct and strongly suggests that the decreased DT–Cy3 fluorescence observed is due to a reduced DT–Cy3 internalization and/or trafficking. Of note, transferrin (Tfn) internalization was unaffected after downregulation of PI3KC2α (supplementary material Fig. S3B), suggesting a specific role for this protein in dynamin-independent endocytosis.

To establish whether PI3KC2α has a role in the internalization and/or trafficking of physiological proteins, we tested whether the internalization of CD59, an endogenous GPI–AP found at the plasma membrane of several cell types, was dependent on PI3KC2α.

CD59 is internalized independently of dynamin (Naslavsky et al., 2004; Sabharanjak et al., 2002), and co-internalization of CD59-specific antibodies with DT–Cy3 in HeLa DynK44A DTR–GPI indeed showed that CD59 and DT–Cy3 display similar trafficking patterns (Fig. 2). Internalization of CD59 in HeLa and HeLa DynK44A DTR–GPI cells expressing sh-PI3KC2α or a nonsilencing control was tested both by immunofluorescence and FACS analysis. Immunofluorescence analysis of cells probed with an antibody against CD59 for 20 minutes at 37°C showed that a major fraction of CD59 was retained at the plasma membrane in cells where PI3KC2α was downregulated (Fig. 4A). To quantify this effect, FACS analysis was performed in cells incubated with the antibody to CD59 for 10 or 20 minutes at 37°C (Fig. 4B), and internalized CD59 was calculated as described in the Materials and Methods section. Downregulation of PI3KC2α resulted in a small, but significant, decrease of CD59 internalization that could be partially rescued after introduction of the shRNA-resistant PI3KC2a construct (Fig. 4C). Importantly, silencing of PI3KC2a did not affect the steady-state levels of CD59 at the plasma membrane, measured by loading CD59 antibodies at 4°C (supplementary material Fig. S4A), indicating that the decreased CD59 internalization in cells transfected with sh-PI3KC2α is not due to decreased CD59 at the plasma membrane. Furthermore, the early time-points (10 and 20 minutes) used in this experiment rule out the possibility that PI3KC2α interferes with CD59 recycling, further supporting a role for PI3KC2α in CD59 internalization.

We also tested whether PI3KC2α participates in the uptake of the extracellular fluid that primarily occurs through clathrin- and dynamin-independent pathways (Kumari and Mayor, 2008; Mayor and Pagano, 2007). Using direct fluorescence and a quantitative fluorimetric assay to measure internalized dextran, a marker commonly used to assay fluid-phase endocytosis, we observed that downregulation of PI3KC2α resulted in a decrease of dextran internalization (Fig. 4D,E).

These results indicate that PI3KC2α can regulate the dynamin-independent internalization of endogenously expressed proteins as well as fluid-phase uptake, suggesting an important role for this protein in physiologically relevant pathways.

PI3Ks are lipid kinases that catalyze the phosphorylation of the 3′ position of the inositol ring. Eight PI3Ks are present in human and they are grouped in three classes (I–III), depending on their sequence homology and substrate specificity. PI3Ks, generating a variety of phosphorylated phosphoinositides in different membrane compartments, are important regulators of endocytosis and intracellular vesicle trafficking. In particular, hVPS34, the unique member of class III PI3K responsible for the constitutive pool of PtdIns(3)P (PI3P), is required for efficient binding and tethering functions of the FYVE-containing protein EEA1 in endosomal membranes (for reviews, see Gillooly et al., 2001; Overduin et al., 2001). Wortmannin, a fungal metabolite that potently inhibits the activity of class I (PI3Kα and β) and class III (hVps34) but not class II PI3Ks (and particularly PI3KC2α) (Domin et al., 1997), neither prevented the internalization of DT–Cy3 (Fig. 5A) nor the intoxication of the cells, which died at the same rate and to the same extent as control cells (data not shown), suggesting that hVPS34, or other wortmannin-sensitive PI3Ks, do not play a role in the DT internalization pathway in HeLa DynK44A DTR–GPI cells. Furthermore, in wortmannin-treated cells, DT-containing vesicles were still positive for EEA1, indicating that EEA1 was still efficiently recruited to these endocytic structures (Fig. 5A). Similar results were obtained using LY294002, another PI3K inhibitor specific for the class I and III enzymes (data not shown), suggesting that EEA1 recruitment to DT-labelled vesicles occurs independently of the classical PI3Ks. Wortmannin treatment, however, reduced the size of DT-vesicles (Fig. 5A), suggesting a possible role for wortmannin-sensitive kinases in homo- and/or heterofusion of DT-vesicles. Of note, PI3KC2α was the unique member of class II able to interfere with the DT internalization process. Downregulation of PI3KC2β had no effect on DT–Cy3 cellular entry (Fig. 5B), and PI3KC2γ is not expressed at detectable levels in HeLa cells (data not shown).

Recent reports suggest that PI3KC2α is the kinase responsible for the production of a dynamic pool of PI3P under stimulated conditions in specific membranes (Falasca and Maffucci, 2007; Wen et al., 2008). Because of the well-documented role for PI3P on EEA1 recruitment to endocytic membrane compartments (Lindmo and Stenmark, 2006; Patki et al., 1997), we tested whether PI3KC2α was responsible for the wortmannin-insensitive EEA1 recruitment to DT-vesicles. The EEA1-positive compartment was analyzed by immunofluorescence in cells treated with sh-PI3KC2α or a nonsilencing control and with unlabeled DT. Strikingly, downregulation of PI3KC2α both in the presence and absence of wortmannin, resulted in a significant reduction of EEA1 recruitment to vesicles (Fig. 6A,B). The small difference on EEA1 loading observed in the presence or absence of wortmannin also indicates that wortmannin-sensitive PI3Ks play a minimal role in EEA1 loading on DT-containing vesicles and further supports the finding that wortmannin does not affect DT internalization. To address directly the possible role for PI3KC2α in the EEA1 recruitment to DT-vesicles internalized in a dynamin-independent manner (Fig. 5A), we analyzed the level of EEA1 in PI3KC2α-silenced cells that also expressed DynK44A. As shown in Fig. 6C, in cells coexpressing DynK44A and sh-PI3KC2α, the number of EEA1-positive structures was strongly decreased compared with the number in the nonsilencing control. This effect was specific for PI3KC2α, as demonstrated by the fact that downregulation of PI3KC2β had no effect on EEA1 loading on endosomes (supplementary material Fig. S5A,B). Together, these data suggest a specific role for PI3KC2α in favouring EEA1 loading on vesicles derived from dynamin-independent pathways.

To address whether EEA1 loss is the consequence of lacking the internalizing vesicles or whether EEA1 is required for their formation and trafficking, we functionally silenced the EEA1 protein. Transfection of the pGIPZ–EEA1 construct efficiently downregulated the EEA1 protein (supplementary material Fig. S5C,D) and resulted in a significant reduction of DT–Cy3-positive vesicles in cells that overexpress DynK44A (Fig. 6D), without reducing the total DT–Cy3 fluorescence (supplementary material Fig. S5E) measured after loading the DT–Cy3 at 4°C for 30 minutes in EEA1-silenced cells. In agreement with other reports (Huang et al., 2004), downregulation of EEA1 had no effect on the internalization of transferrin (data not shown), suggesting a functional role for EEA1 specifically in dynamin-independent internalization processes.

To identify the site of action of PI3KC2α during DT uptake, we analyzed its cellular localization during DT internalization. Similar to other cell lines (Didichenko and Thelen, 2001; Domin et al., 2000), in HeLa DynK44A DTR–GPI cells, PI3KC2α was localized across the entire cell with a punctate staining (Fig. 7) and no evident plasma membrane or endosomal enrichment. However, DT internalization led to a relocalization of PI3KC2α from the cytoplasm to plasma membrane and vesicular compartments at the periphery of the cells. Furthermore, during DT–Cy3 internalization, PI3KC2α showed a high degree of colocalization with internalized DT–Cy3 (Fig. 7). Relocalization of PI3KC2α was also observed in cells internalizing MHCI and GPI–GFP (Fig. 7), and PI3KC2α colocalized with the internalizing proteins. Notably, transferrin internalization did not lead to any movement of PI3KC2α across the cell. Furthermore, the internalized PI3KC2α-positive structures generated by DT treatment were not enriched at other membrane proteins, such as the transferrin receptor (supplementary material Fig. S6A), suggesting a role for PI3KC2α specifically in dynamin-independent internalization pathways.

As mentioned previously, two main pathways are known to act in dynamin-independent internalization: the GEEC pathway and the Arf6-dependent pathway (Kumari et al., 2010). In HeLa cells, the Arf6 pathway mediates internalization of MHCI and CD59, whereas the GEEC pathway is reported to assist the uptake of cholera toxin, dextran and GPI–GFP. The observation that DT colocalizes with molecules that enter the cells using both pathways (Fig. 2) suggests that the DTR–GPI can guide DT entry by multiple portals. To test this hypothesis, overexpression experiments with Arf6 and GRAF1 were used to interfere with the two pathways. Expression of constitutively active Arf6(Q67L), unable to hydrolyze GTP to GDP, is reported to impair the intracellular trafficking of the cargo molecules that remain trapped in Arf6-positive vacuolar compartments (Brown et al., 2001). When Arf6(Q67L) was expressed in HeLa DynK44A DTR–GPI cells, we observed DT–Cy3 accumulation in the typical Arf-6 enlarged vesicles (Fig. 8A), indicating that DT can use this internalization pathway. Similarly, overexpression experiments using a truncated form of GRAF1 (BAR+PH domains), which is able to inhibit specifically the dynamics of the GEECs (Lundmark et al., 2008), resulted in the trapping of DT in the GEEC tubules stabilized and visualised by the BAR+PH–GFP construct, suggesting that DT can also enter through this pathway (Fig. 8B). We concluded that, in our system, DT internalizes using both pathways, and, in agreement, we found that downregulation of Cdc42, a key molecule for GEEC formation (Sabharanjak et al., 2002) but not required for the Arf6 pathway, resulted only in a partial decrease of DT–Cy3 internalization (data not shown).

In the past, the realization that several molecules, including GPI-APs and glycosphingolipids, can still be internalized when dynamin-dependent pathways are impaired has attracted much attention in the endocytic field (Doherty and McMahon, 2009; Hansen and Nichols, 2009; Howes et al., 2010; Kumari et al., 2010; Mayor and Pagano, 2007). The abundance of lipid-linked molecules at the plasma membrane suggests that these ‘alternative’ portals could have important roles in physiological processes such as growth and differentiation, cell adhesion and regulation of signalling receptors. The most convincing evidence for the role for dynamin-independent pathways in physiological processes comes from in vivo analysis of model organisms. In the nematode Caenorhabditis elegans, in which a single dynamin gene has been identified, the analysis of animals expressing mutant dynamin led to the understanding that other mechanisms must exist to substitute partially for dynamin activity in vital endocytic processes, such as the synaptic vesicle cycle (Clark et al., 1997). Similar evidence was recently obtained from the analysis of dynamin I knockout mice (Ferguson et al., 2007). Furthermore, a number of toxins and viruses are reported to hijack these portals to enter specific cell types, as in the case of HIV1 (Vidricaire and Tremblay, 2007), polyomavirus (Gilbert and Benjamin, 2000), VacA (Gauthier et al., 2005), cholera toxin (at least partially) (Kirkham et al., 2005), cytotoxic necrotizing factor 1 (CNF1) (Contamin et al., 2000) and ricin (Simpson et al., 1998), indicating that these pathways could be relevant in pathogenic conditions. However, the mechanisms underlying these processes are still unclear, and very little information is available about the molecular components involved.

With a novel cell-survival screening strategy using a library of shRNAs in a cell line in which DT internalizes in a dynamin-independent manner (Skretting et al., 1999), we identified several molecules, which, when reduced, inhibit the cellular uptake and/or trafficking of DT. We further validated PI3KC2α as a key molecule for dynamin-independent pathways. Depletion of PI3KC2α can efficiently inhibit DT internalization and cell intoxication in our system. Importantly, the internalizations of CD59, known to traffic independently of dynamin in the cell, and dextran, a fluid-phase marker, are also impaired by PI3KC2α silencing, reinforcing the role for PI3KC2α in dynamin-independent internalization of physiological substrates. Furthermore, PI3KC2α appears to be the unique PI3K involved in dynamin-independent internalization. DT internalization occurs in the presence of wortmannin, indicating that class I or III PI3Ks, which are sensitive to this drug, are not implicated in the internalization of DTR–GPI. This is in agreement with the previously published result for the folate receptor (Kalia et al., 2006), another endogenous GPI-AP. Moreover, silencing of PI3KC2β, the other member of class II PI3Ks present in HeLa cells, does not have an effect on dynamin-independent internalization.

A further indication that PI3KC2α has a role in dynamin-independent internalization events is suggested by immunofluorescence analyses showing PI3KC2α redistribution to endomembranes during dynamin-independent internalization events. Similarly, PI3KC2α is temporarily recruited to the plasma membrane and secretory vesicles after stimulation with insulin and Ca²⁺, respectively (Falasca and Maffucci, 2007; Wen et al., 2008), suggesting that recruitment of PI3KC2α to specific membranes could be a common mechanism of activation leading to a local and temporary production of PI3P. The signal that dictates the relocalization of PI3KC2α to specific endomembranes is not known at the moment, but the presence of a PX domain in the C-terminal part of the protein with high affinity for phosphatidylinositol (4,5)-bisphosphate (Stahelin et al., 2006), abundant at the plasma membrane and in some internalizing vesicles, such as the Arf6 compartments (Porat-Shliom et al., 2008), suggests that its intracellular localization could be mediated, at least in some conditions, by protein–lipid interactions.

In recent years, two main distinct portals for dynamin-independent entry have been identified: the Arf6-dependent pathway, which shares some characteristics and cargo with stimulated macropinocytosis (Donaldson et al., 2009), involved in MHCI and CD59 (in HeLa cells) internalization, and the constitutive CLIC–GEEC pathway, specific for some GPI-APs and, at least in part, for fluid-phase internalization (Kumari et al., 2010). The small GTPases Arf1 and Cdc42, together with ARHGAP10 and GRAF1, have been recognized as regulators of the dynamic tubular structures characteristic of this latter pathway (Doherty and Lundmark, 2009; Kumari and Mayor, 2008). Fluorescence analyses indicate that DT colocalizes with markers of both pathways. DT is detected in Arf6-vacuoles, arising from aberrant trafficking from the plasma membrane after transfection of the Arf6 mutant, and in the GEEC tubules, visualised by the GRAF1–BAR+PH–GFP construct. These data suggest that the two pathways cooperate in DT internalization in our system and that the DT, differentially from VacA toxin internalization that is specifically GEEC-mediated (Gauthier et al., 2007), can also exploit the inducible Arf6 pathway. The fact that the silencing of PI3KC2α efficiently prevented cell intoxication led us to hypothesize that PI3KC2α could act in both pathways or in a step shared by the GEEC/Arf6 pathways. It is still unclear whether PI3KC2α also acts in other dynamin-independent pathways, such as macropinocytosis, that could be stimulated by DT treatment. The fact that DT internalization is not prevented by wortmannin (this study) and staurosporine (Skretting et al., 1999), both inhibitors of macropinocytosis (Araki et al., 1996; Dharmawardhane et al., 2000; Mercer and Helenius, 2009; Mercer et al., 2010), suggests that this is probably not the case. Moreover, DT treatment does not induce alteration in the CD59 internalization (supplementary material Fig. S4B) or nonspecific internalization of the transferrin receptor. These results suggest that the toxin does not trigger the prominent membrane reorganization and the nonselective internalization that is coupled with macropinocytosis.

While other studies are required to determine the positioning of PI3KC2α in the dynamin-independent pathways, our analysis suggests that PI3KC2α function is regulated by EEA1 recruitment to vesicular compartments generated by dynamin-independent pathways. Downregulation of PI3KC2α results in a strong reduction of EEA1 loading on endosomes in cells contemporaneously treated, or not, with wortmannin that inhibits the other PI3Ks or expressing DynK44A, which blocks clathrin-mediated endocytosis. As PI3KC2α is recognized as the kinase responsible for the production of a dynamic pool of PI3P under stimulated conditions, it is likely that its activity directly mediates EEA1 recruitment to specific membranes, a possibility that is reinforced by other studies reporting that vesicles derived from dynamin-independent pathways can acquire EEA1 independently of fusion with the classical endosomal compartments generated by clathrin-mediated endocytosis (Kalia et al., 2006; Naslavsky et al., 2003). Our data suggest that EEA1 recruitment to the generating endomembranes plays a key function in dynamin-independent processes. The fact that silencing of EEA1 impairs DT internalization indicates that loss of EEA1 recruitment is not just a consequence of the lack of internalizing endosomes but, rather, suggests a functional role for EEA1 in DT internalization and/or trafficking. The identification in our original screen of p38β/MAPK11 (C.K., E.K.M. and A.E.S., unpublished), a stress-response kinase known to phosphorylate and regulate EEA1 localization to endosomes (Cavalli et al., 2001; Mace et al., 2005), further reinforces our findings. An open question that remains to be addressed is whether EEA1 is the unique PI3P-binding molecule recruited to endosomes generated by dynamin-independent pathways. There is a possibility that other FYVE/PX-containing proteins with the ability to bind to PI3P, such as rabenosyn-5, Hrs or sorting nexins, could additionally be engaged in the process. However, initial experiments exclude a role for rabenosyn-5 as its localization is unaffected in PI3KC2α-silenced cells treated with wortmannin (supplementary material Fig. S6B).

The finding that EEA1 can interfere with DT internalization and/or trafficking could suggest an unexpected role for EEA1 at the plasma membrane. Further studies are required to test this hypothesis, which is indirectly supported by the fact that, after insulin treatment, PI3KC2α has been identified at the plasma membrane, where it generates PI3P and promotes recruitment of the FYVE probe at this site (Falasca and Maffucci, 2007). However, the fact that downregulation of EEA1 results in a significant reduction of DT–Cy3-positive vesicles in cells analyzed after 30 minutes of continuous uptake of DT might also indicate a role for EEA1 in later steps, for example in fusion of primary tubulovesicular carriers arising from the plasma membrane (Kirkham et al., 2005).

Finally, there is an emerging role for PI3KC2α as a general membrane trafficking regulator. Indeed, PI3KC2α has been reported previously to interact with clathrin and to reduce the internalization of Tfn when overexpressed (Gaidarov et al., 2001; Gaidarov et al., 2005), an effect we did not observe under depletion conditions. Our analysis indicates that PI3KC2α is involved in dynamin-independent endocytosis, acting both in the GEEC and the Arf6 pathway. It is possible that PI3KC2α, similar to other proteins such as some GTPases, can exert key roles in different internalization pathways and, at the same time, regulates other vesicular trafficking processes, such as exocytosis in neurosecretory cells (Wen et al., 2008). Further analyses using different cell types and other cargos that internalize in a dynamin-independent manner, together with the identification of specific inhibitors for class II PI3Ks, will help to understand fully the pleiotropic roles of PI3KC2α.

All chemicals were obtained from Sigma-Aldrich unless otherwise specified. The mAb against EEA1 was purchased from BD Transduction laboratories. HA antibodies, used for the detection of HA-tagged DynK44A, were from Covance (mAb) and from Affinity BioReagents (pAb). Polyclonal antibodies against PI3KC2α and rabenosyn-5 were purchased from Santa Cruz Biotechnologies. HB-EGF mAb was from R&D Systems, MHCI (W6/32) mAb from AbCam, CD59 mAb from BC Pharmingen, GFP mAb from Roche and TfnR mAb from Invitrogen. Secondary Alexa-Fluor-conjugated antibodies: Tfn Texas Red, Tfn Alexa Fluor 488, cholera toxin B subunit (CtxB) Alexa Fluor 488 and dextran Alexa Fluor 594 were purchased from Invitrogen. DT and mutant DT-CRM197 (Uchida et al., 1972) were purchased from Calbiochem. Cy3 (Amersham, GE Healthcare) was conjugated to DT-CRM197 (DT–Cy3) according to the manufacturer's instructions.

Decay-accelerating factor (DAF) fused to GFP (GPI–GFP) has been described previously (Nichols et al., 2001); GRAF1 and Arf6 constructs were kind gifts from Richard Lundmark (University of Umeå, Sweden) and Julie Donaldson (NIH, USA), respectively.

HeLa DynK44A and HeLa DynK44A DTR–GPI cell lines were cultured as described previously (Lanzrein et al., 1996; Llorente et al., 1998; Skretting et al., 1999). DynK44A expression was induced by tetracycline removal for 36–48 hours. Approximately 80% of HeLa DynK44A and 60% of HeLa DynK44A DTR–GPI cells expressed DynK44A. Transfections were performed using either calcium phosphate precipitation (Graham and van der Eb, 1973) or LipofectAMINE 2000 (Invitrogen) reagent according to the manufacturer's instructions. shRNA constructs were expressed for 72 hours. GPI–GFP, pEGFP–GRAF1–BAR+PH and pECFP–Arf6Q67L (Addgene plasmid 11387) constructs were expressed for 16 hours before the analysis.

pGIPZ–PI3KC2α is lentiviral shRNAmir from Open Biosystems (www.openbiosystems.com) that allows the visualization of transfected and/or silenced cells by GFP expression. Wild-type PI3KC2α was cloned into the Gateway-compatible pCMV-Myc vector (generous gift from Jesper Christiansen, BRIC, University of Copenhagen) using the Open Biosystems full-length cDNA (clone ID 8322710) as a template. PI3KC2α shRes was generated by site-direct mutagenesis (Stratagene), introducing two silent mutations in the region of wild-type Myc-tagged PI3KC2α recognized by the pGIPZ–PI3KC2α silencing construct.

PI3KC2α was identified as a hit in a screen using a library of shRNA, kindly provided by the NKI Institute, The Netherlands, targeting more than 8000 human genes (a list of target genes is available at http://www.screeninc.nl) (Brummelkamp and Bernards, 2003). Briefly, the screen was performed using pools of shRNAs targeting 1000 genes that were transfected into HeLa DynK44A DTR–GPI cells. HeLa DynK44A DTR–GPI or HeLa DynK44A cells transfected with the empty pRS vector were used as a negative control. Two days post-transfection, cells were plated in the absence of tetracycline for 48 hours and treated with 25 ng/ml DT for 4 hours and allowed to grow for two weeks. Genomic DNA was purified from surviving cells using a QiaAmp DNA Micro Kit (Qiagen). Transfected constructs were amplified by PCR as described previously (http://www.screeninc.nl) and sequenced. A total of 63 positive hits were identified in the screen (see supplementary material Table S1). For the survival assay, silencing constructs were transfected individually into HeLa DynK44A DTR–GPI cells and, two days post-transfection, the cells were plated in the absence of tetracycline for 48 hours and treated with 25 ng/ml DT for 1 hour and allowed to grow for approximately 10 days. Cells were fixed in 4% PFA and stained with crystal violet for 5 minutes at room temperature. Cells were dissolved in 10% acetic acid, and absorbance was measured in triplicate at 620 nm. Survival assays were performed at least three times for each construct. Empty pRS constructs and nonsilencing shRNA constructs were used as negative controls, respectively.

For internalization assays, cells were incubated with DT–Cy3 (~5 μg/ml) for the indicated times at 37°C. Tfn Alexa Fluor 488 or Tfn Texas Red conjugates (50 μg/ml) and CtxB Alexa Fluor 488 conjugates (1 μg/ml) were added to the cells for 5 or 10 minutes at 37°C.

For the CD59, MHCI and GPI–GFP colocalization analysis with DT–Cy3, the corresponding antibody was added to the cells for 10 minutes together with DT–Cy3 at 37°C, and cells were processed for immunofluorescence as described below.

For the PI3KC2α relocalization studies, DT–Cy3 (~5 μg/ml), CD59 (5 μg/ml) and MHC I (4 μg/ml) antibodies were added to the cells for 20 minutes at 37°C. Tfn Alexa Fluor 488 was internalized for 5 minutes at 37°C. The cells were processed further for immunofluorescence as described below.

Cells were fixed in 4% PFA (BDH ProLABO) and permeabilized in 0.1% Triton X-100 in PBS at room temperature. In some experiments (e.g. GRAF1–BAR+PH expression), cell fixation was performed at 37°C (Lundmark et al., 2008). Cells were blocked in cell culture medium for 30 minutes, and primary antibodies were added in blocking solution for 2 hours at room temperature or at 4°C overnight.

For wortmannin treatment, cells were preincubated for 15 minutes at 37°C with 100 nM wortmannin or with the carrier (0.01% DMSO). DT–Cy3 or unstained DT was added for 30 minutes at 37°C with or without 100 nM wortmannin.

For the analysis of fluid-phase internalization, transfected HeLa cells were incubated with Dextran–Alexa Fluor 594 (10 MW) for 20 minutes at 37°C and either fixed and processed for fluorescence or lysed in 1% Nonidet P-40 lysis buffer (Nonidet P-40, 150 mM NaCl, 50 mM Tris-HCl pH 7.4, 20 mM NaMoO₄ containing a protease inhibitor cocktail). Cell lysates were centrifuged at 20,000 g for 20 minutes at 4°C and the protein concentration measured for normalization. The amount of Dextran Alexa Fluor 594 in the supernatant was measured from the emission at 612 nm after exciting at 584 nm using a FLUOstar OPTIMA.

Upon exposure to low pH, DT translocates unspecifically to the cytosol (data not shown) (Skretting et al., 1999), and acid washes to remove DT bound to the plasma membrane were therefore not possible. Furthermore, loading at 4°C was not performed (all time-points are therefore a measure of continuous uptake after DT addition to the culture medium) as this prevented uptake of DT. This might be related to the fact that dynamin-independent pathways are reported to be sensitive to low temperatures (Lundmark et al., 2008).

Fluorescence micrographs were acquired using an Axiovert 135 (Carl Zeiss) with a ×63 (NA 1.4) Plan Apochrome objective in immersion oil, at room temperature with a CoolSNAP cf2 Photometrics camera using MetaMorph software (MDS Analytical Technologies). Confocal images were acquired using an Axiovert 200M LSM 510 (Carl Zeiss) using a ×63 (NA 1.2) c-Apochromat objective in dH₂O at room temperature using LSM 510 META software and LSM image examiner software (Carl Zeiss). Images were exported in preparation for printing using Photoshop (Adobe).

The mean and s.e.m. of integrated intensities per cell from each coverslip were determined using Metamorph software. At least 40 cells in each condition were analyzed in two or three independent experiments.

Quantification of internalized DT–Cy3 was achieved using images of cells coexpressing DynK44A and different silencing constructs (or the nonsilencing control) treated with DT–Cy3 for 30 minutes at 37°C. To ensure that the decrease in internalized DT–Cy3 was not due to a reduced binding at the plasma membrane, the total fluorescence was quantified in cells loaded with DT–Cy3 for 30 minutes on ice.

Quantitative analysis of the EEA1-positive compartments was performed comparing the total fluorescence of nonsilenced cells versus sh-PI3KC2α-transfected cells, and the images were processed for background correction using Metamorph software, such that all the endosomes with intensity above the background were identified.

Protein synthesis was measured using [³⁵S]methionine, as described previously (Llorente et al., 1998). Total RNA was isolated using the RNeasy Minikit (Qiagen). Quantitative PCR was performed as described previously (Rose et al., 2007). Oligonucleotides used for RT-qPCR are available upon request.

The degree of internalization of CD59 was analyzed using a fluorescence-activated cell sorter (FACS) assay. HeLa DynK44A DTR–GPI cells transfected with the pGIPZ–PI3KC2α or nonsilencing shRNA constructs were incubated with the CD59 antibody (5 μg/ml) for 10 and 20 minutes at 37°C. The cells were washed in PBS and fixed in 4% PFA for 10 minutes. The cells were then stained with anti-mouse Alexa Flour 594 for 1 hour and analyzed on a FACS Calibur (BD Biosciences) using CellQuestsoftware (Becton Dickinson). Total CD59 staining (internalized and plasma membrane CD59) was measured in permeabilized cells (0.1% saponin in 1.5% FBS heat inactivated in PBS for 1 hour at 4°C) and plasma membrane CD59 staining was measured in nonpermeabilized cells. Quantifications of the mean fluorescence in GFP-positive cells were performed using Modfit LT software (Verity Software House). The percentage of internalization was calculated as described previously (Janvier and Bonifacino, 2005). Steady-state surface levels of CD59, determined by loading CD59 antibodies on ice before fixation, indicated that the plasma membrane level of CD59 is unchanged after PI3KC2α silencing (supplementary material Fig. S4).

Permeabilization of the cells had no detectable effect on the levels of CD59 at the plasma membrane at steady state (data not shown).

In the rescue experiment, pGIPZ–PI3KC2α and PI3KC2α shRes were co-transfected in HeLa DynK44A DTR–GPI cells that were analyzed for CD59 internalization. The data reported have been normalized to nonsilencing controls.

The results are expressed as means + s.e.m. Statistical significance was determined using unpaired, two-tail distribution, t-tests. Data indicated with an asterisk have values of *P<0.01. Each experiment was repeated at least three times, unless otherwise specified.

We thank Sandra Schmid (Dept of Cell Biology, The Scripps Research Institute, La Jolla, CA 92037, USA) for the HeLa DynK44A cell line, Kirsten Sandvig (Institute for Cancer Research at the Norwegian Radium Hospital, Montebello, Oslo, Norway) for the HeLa DynK44A DTR–GPI cell line and for discussion of the screen design. Furthermore, we thank Bo van Deurs (Dept of Medical Anatomy, University of Copenhagen, Denmark) for valuable discussions, Julien Vandamme (BRIC, University of Copenhagen) for cloning PI3KC2α and Toshia Myers (BRIC, University of Copenhagen) for help in editing the manuscript. The authors declare that they have no competing financial interests.