Phosphatidylinositol 3,4-bisphosphate synthesis and turnover are spatially segregated in the endocytic pathway

Phosphoinositides play crucial roles in intracellular membrane dynamics and cell signaling, with phosphatidylinositol (PI) 3-phosphates being the predominant phosphoinositide lipids at endosomes and lysosomes, whereas PI 4-phosphates, such as phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2), are enriched at the cell surface including sites of endocytosis. How PI 4-phosphates and PI 3-phosphates are dynamically interconverted within the endocytic pathway and how this is controlled in space and time remains poorly understood. Here, combining live imaging, genome engineering, and acute chemical and genetic manipulations, we found that local synthesis of PI(3,4)P2 by phosphatidylinositol 3-kinase C2α at plasma membrane clathrin-coated pits is spatially segregated from its hydrolysis by the PI(3,4)P2-specific inositol polyphosphate 4-phosphatase 4A (INPP4A). We observed that INPP4A is dispensable for clathrin-mediated endocytosis and is undetectable in endocytic clathrin-coated pits. Instead, we found that INPP4A partially localizes to endosomes and that loss of INPP4A in HAP1 cancer cells perturbs signaling via AKT kinase and mTOR complex 1. These results reveal a function for INPP4-mediated PI(3,4)P2 hydrolysis in local regulation of growth factor and nutrient signals at endosomes in cancer cells. They further suggest a model whereby synthesis and turnover of PI(3,4)P2 are spatially segregated within the endocytic pathway to couple endocytic membrane traffic to growth factor and nutrient signaling.

Phosphoinositides are a minor class of comparably shortlived membrane phospholipids that mediate crucial cellular and organismal functions, including cell signaling and regulation of intracellular membrane traffic (1,2). The different PI 3 species differ regarding their subcellular localization, with PI 4-phosphates being concentrated in the exocytic pathway and at the plasma membrane and PI 3-phosphates, such as phosphatidylinositol 3-phosphate (PI(3)P), predominantly found at endosomes and lysosomes (1,2). Hence, as membranes are internalized from the plasma membrane, they have to undergo conversion of their phosphoinositide content from PI 4-phosphates, in particular PI(4,5)P 2 , to PI 3-phosphates, e.g. PI(3)P, the major signature lipid of endosomes and lysosomes (3).
How this phosphoinositide conversion occurs and how it is spatiotemporally controlled is unclear. Prior studies have established that PI(4,5)P 2 , a lipid required for endocytic clathrin-coated pit (CCP) formation (4), is hydrolyzed during late stages of CME by PI(4,5)P 2 5-phosphatases (5)(6)(7). Work by several laboratories has shown that loss of PI(4,5)P 2 as CCPs mature is accompanied by clathrin-mediated recruitment of phosphatidylinositol 3-kinase C2␣ (PI3KC2␣) (8,9), a class II PI 3-kinase family member implicated in human disease (10) that may synthesize PI(3,4)P 2 (9) and/or PI(3)P, possibly depending on cell type and its nanoscale localization (11,12). Depletion of PI3KC2␣ stalls CCP dynamics and results in cell surface accumulation of nonconstricted U-shaped intermediates of CME (i.e. prior to membrane scission by dynamin), a phenotype rescued by active WT but not kinase-inactive mutant PI3KC2␣ (9), indicating that PI 3-phosphate synthesis by PI3KC2␣ is required for CCP maturation and membrane constriction. Conflicting models have been postulated regarding the exact nature of the lipid product synthesized by PI3KC2␣ during CME. Based on the activity of native PI3KC2␣ immunoprecipitated from mammalian cells in combination with knockdown/ rescue experiments and analysis of the cellular PI 3-phosphate content, it has been postulated that PI3KC2␣ specifically synthesizes PI (3,4)P 2 at endocytic CCPs (9) to facilitate CCP membrane constriction via lipid-mediated recruitment of the dynamin-associated Phox homology-Bin-amphiphysin-rvs domain protein sorting nexin 9 (SNX9 and its close homolog SNX18) to the invagination neck (13). Further support for this model (Fig. 1A, model I) derives from the fact that loss of SNX9/18 or mutation of its PI 3-phosphate-binding PX domain phenocopies loss of PI3KC2␣ with respect to CME (13,14). Moreover, re-expression of hyperactive PI3KC2␣ in cells depleted of the endogenous enzyme results in elevated plasma membrane PI(3,4)P 2 levels and facilitates CME in living cells (15), suggesting a tight relationship between local PI(3,4)P 2 synthesis by PI3KC2␣ and the efficiency of CME.
A recent imaging study based on composite protein/lipidbinding probes derived from the clathrin uncoating factor auxi-

Spatial segregation of PI(3,4)P 2 synthesis and turnover
lin, a protein that acts post-scission in CME, proposed a different model for endocytic phosphoinositide conversion. According to this model (Fig. 1A, model II), short bursts of PI(3)P and PI(3,4)P 2 are observed only after vesicle scission, with PI3KC2␣ responsible for the majority of the late PI(3)P signal (16), whereas PI(3,4)P 2 may be synthesized by another enzyme, conceivably a class I PI 3-kinase. Although PI3KC2␣, in principle, is capable of synthesizing PI(3)P, e.g. on endosomes (17,18), this model conflicts with the accumulation of nonconstricted CCPs, i.e. a step prior to scission, in PI3KC2␣-depleted cells or cells from PI3KC2␣ knockout mice (9). Finally, based on in vitro studies using liposomes, it has been proposed that PI3KC2␣ indeed produces PI(3,4)P 2 at CCPs, but this lipid may be rapidly converted to PI(3)P by the PI(3,4)P 2 -specific inositolpolyphosphate 4-phosphatase INPP4A, an enzyme previously localized to endosomes (19). This PI(3)P pool then serves to recruit SNX9 to late-stage CCPs (Fig. 1A, model III), where it may facilitate assembly of F-actin (20).
In this study, we combine live imaging, genome engineering, and acute chemical and genetic manipulations to reveal that local synthesis of PI(3,4)P 2 by PI3KC2␣ at CCPs is spatially segregated from its hydrolysis at endosomes by INPP4A to locally regulate cell signaling via Akt and mTORC1.

Spatial segregation of PI(3,4)P 2 synthesis and turnover
resides. Similar observations were made previously upon constitutive targeting of INPP4B or MTM1 to the plasma membrane (9). These experiments thus suggest that CME is facilitated by a plasma membrane-localized endocytic pool of PI(3,4)P 2 .

CCP dynamics are controlled by synthesis of PI(3,4)P 2 mediated by PI3KC2␣
To challenge these data using an independent approach, we analyzed the ability of WT and different mutants forms of PI3KC2␣ to rescue defective CME in cells lacking the endogenous enzyme. PI3KC2␣-depleted COS7 cells were transfected with plasmids encoding eGFP-tagged WT PI3KC2␣, a kinaseinactive mutant of PI3KC2␣, or a re-engineered class III-like variant carrying the substrate-binding loop of the PI(3)P-selective PI 3-kinase Vps34 that synthesizes PI(3)P but is unable to generate PI(3,4)P 2 , as demonstrated earlier (9). As PI3KC2␣, in addition to its role in endocytosis, also regulates endosomal recycling of transferrin (17), the effects of PI3KC2␣ manipulation on CME cannot simply be monitored by measuring the efficiency of transferrin uptake in genetic knockdown/rescue experiments. We therefore analyzed the effect of the different PI3KC2␣ variants on CCP dynamics as a direct readout of its role in regulating CME. The WT and kinase-inactive and class III-like mutants were correctly targeted to endocytic CCPs (Fig. 2, A-C) and expressed at similar levels (Fig. 2D). The accumulation of long-lived static endocytic CCPs in COS7 cells depleted of endogenous PI3KC2␣ was effectively rescued by re-expression of siRNA-resistant WT PI3KC2␣ but not by kinase-inactive or class III-like mutant PI3KC2␣ (Fig. 2, E and F). Moreover, WT but not kinase-inactive or class III-like mutant PI3KC2␣ restored the partial depletion of plasma membrane PI(3,4)P 2 in PI3KC2␣ knockdown COS7 cells (Fig.  2G), suggesting a tight correlation between PI(3,4)P 2 synthesis by PI3KC2␣ and CCP dynamics. Importantly, class III-like mutant PI3KC2␣ was perfectly able to restore defects in transferrin recycling (17) or sonic hedgehog signaling from Rab8positive endosomes (18) via endosomal synthesis of PI(3)P. This pool of PI(3)P is likely small and/or spatiotemporally restricted (at least in COS7 cells), as we failed to detect PI3KC2␣-mediated PI(3)P synthesis in living COS7 cells either under control conditions or upon selective inhibition of Vps34 ( Fig. 2H and Fig. S2), the major PI(3)P-synthesizing enzyme in eukaryotic cells. Collectively, these data indicate that CCP dynamics and, therefore, CME are regulated by local synthesis of PI(3,4)P 2 by PI3KC2␣.

The PI(3,4)P 2 -specific 4-phosphatase INPP4 is absent from CCPs and dispensable for CME
Our data are incompatible with PI3KC2␣-mediated PI(3)P synthesis (Fig. 1A, model II) at CCPs, in agreement with a recent study (33), indicating that the phosphoinositide sensors generated by He et al. (16) may be nonspecific. However, they do not distinguish between a direct role of PI(3,4)P 2 in recruitment of endocytic effector proteins such as SNX9/18 (Fig. 1A, model I) and an indirect role of PI(3,4)P 2 as a substrate for the PI(3,4)P 2specific 4-phosphatase INPP4A to locally produce PI(3)P at CCPs, which then recruits SNX9/18 (Fig. 1A, model III). The latter mechanism requires INPP4A to act locally at CCPs to generate PI(3)P from PI(3,4)P 2 . It further predicts that loss of INPP4A should phenocopy endocytic defects seen in cells lacking either PI3KC2␣ or SNX9/18. To test these predictions, we monitored the localization and dynamics of eGFP-INPP4A or eGFP-PI3KC2␣ in living COS7 cells coexpressing monomeric red fluorescent protein-clathrin light chain by two-color TIRF imaging with automated particle tracking. PI3KC2␣ displayed striking colocalization with mRFP-clathrin light chain across all lifetime cohorts, in agreement with its key regulatory role in CME (Fig. 3, A-C). In striking contrast, INPP4A was largely cytoplasmic and absent from CCPs, with no correlation to clathrin recruitment (Fig. 3, D-F). eGFP-PI3KC2␣ also strongly colocalized with endogenous clathrin at endocytic CCPs in fixed cells (Fig. 3, G and I), whereas eGFP-INPP4A was absent from endocytic clathrin puncta (Fig. 3, H and I). Instead, we found that, in COS7 cells, a fraction of fluorescent protein-tagged INPP4A colocalized with markers of early endosomes, such as endogenous Rab5 or EEA1 (Fig. 4, A and E), whereas colocalization with late endosomal LAMP1 or endogenous Rab7 (Fig. S3) was less overt. Overexpression of eGFP-Rab5 facilitated recruitment of INPP4A to early endosomes in transfected COS7 (Fig. 4, B and C) or HAP1 cancer cells (Fig. 4, F and G). In contrast, overexpression of eGFP-Rab7 did not affect INPP4A localization in either COS7 (Fig. 4D) or HAP1 cells (Fig. 4H). These data indicate that INPP4A is absent from CCPs and instead localizes, at least in part, to early endosomes, consistent with earlier data (19).
It is possible that INPP4A functionally regulates CME without being enriched at CCPs. Hence, we analyzed the consequences of INPP4A depletion on CCP dynamics and ultrastructure in COS7 cells, which lack expression of INPP4B (Fig. S4E), the only other INPP4 family member in mammals. Depletion of INPP4A by either of two distinct siRNAs (Fig. 5A) had no effect on the dynamics of CCPs (Fig. 5, B and C), in contrast to the accumulation of static AP-2-enriched CCPs observed in COS7 cells depleted of either PI3KC2␣ or SNX9/18 (Fig. 5, D and E, and Fig. S4A). Double loss of INPP4A and either PI3KC2␣ or SNX9/18 impaired CCP dynamics (Fig. 5, F and G, and S4A), akin to single loss of PI3KC2␣ or SNX9/18. These data suggest that PI3KC2␣ and SNX9/18 act upstream of INPP4A within the endocytic pathway. Consistent with our results regarding CCP dynamics, depletion of INPP4A also did not affect the ultrastructure or number of endocytic intermediates (Fig. 5H) or the efficiency of transferrin-CME (Fig. 5I). To corroborate these data from siRNA knockdown experiments, we generated INPP4A KO clones using CRISPR/Cas9 (Fig. 5J). INPP4A KO COS7 cells were indistinguishable from WT cells with respect to CCP dynamics (Fig. 5, J and K) or plasma membrane AP-2 levels (Fig. S4B).
In summary, our data show that INPP4A is probably absent from CCPs and dispensable for CME. Our combined results are therefore most compatible with a model according to which PI(3,4)P 2 is synthesized locally at endocytic CCPs by PI3KC2␣ and is hydrolyzed downstream in the endosomal pathway by INPP4 (Fig. 1A, model I) and possibly other phosphatases, e.g. PTEN (43).

Loss of INPP4A in HAP1 cells perturbs signaling via Akt and mTORC1
Apart from its function in endocytosis, PI(3,4)P 2 has wellknown roles in cell signaling. Plasma membrane PI(3,4)P 2 , gen-erated by the concerted action of class I PI 3-kinase and the 5-phosphatase SHIP2/INPP5 (33,34), activates Akt, a gatekeeper for cell survival and an inducer of nutrient signaling via the late endosomal/lysosomal mTORC1 complex. Conversely,

Spatial segregation of PI(3,4)P 2 synthesis and turnover
local synthesis of a late endosomal/lysosomal pool of PI(3,4)P 2 by PI3KC2␤, an enzyme activated when growth factor signaling ceases (25), represses mTORC1 activity via an Akt-independent mechanism in serum-deprived cells (24). Hence, PI(3,4)P 2 differentially regulates cell signaling via Akt and mTORC1, depending on its subcellular location and cellular nutrient conditions.
As a fraction of INPP4A is localized to endosomes (19) (Fig.  4), where it may hydrolyze PI(3,4)P 2 , we hypothesized that INPP4A might control cell signaling via Akt and/or mTORC1. The levels of INPP4A and its close relative INPP4B vary greatly between different cell types and tissues. For example, loss of INPP4B-mediated PI(3,4)P 2 hydrolysis is associated with overactive Akt signaling in breast cancer cells (32,35,36). Among the cell types studied, we found INPP4A to be highly expressed in HAP1 cells, a near-haploid cancer cell line characterized by elevated PI(3,4,5)P 3 /PI(3,4)P 2 production via hyperactive class I PI 3-kinase signaling (37). COS7 cells, used for most of our studies described above, expressed very low levels of INPP4A, whereas INPP4B was not detectable in either of these cell lines (Fig. S4E). To test whether loss of INPP4A might alter PI(3,4)P 2 -dependent signaling cascades, we analyzed INPP4A KO HAP1 cells. These cells showed normal CME ( Fig. S5F; akin to INPP4A KO in COS7 cells (Fig. 5)) but displayed elevated levels of PI(3,4)P 2 (Fig. 6A) that correlated with overactive Akt signaling (Fig. 6, B and C). Additional knockdown of INPP4B in INPP4A KO cells did not result in any further elevation of pAkt signaling (Fig. 6, B and C), indicating that INPP4A is the major, if not the only, physiologically relevant INPP4 family member in HAP1 cells. Interestingly, INPP4A KO cells also displayed delayed recovery of mTORC1 activity upon refeeding poststarvation, monitored by the levels of the mTORC1-specific substrate pS6 kinase 1 (Fig. 6, B and D). This is consistent with a local role of endolysosomal PI(3,4)P 2 in repressing mTORC1 signaling in serum-starved cells (24). These phenotypes were not observed in INPP4A KO COS7 cells, which express the enzyme at very low levels (Fig. S4, C and D, and S5, A-D). Collectively, these results show that INPP4A, by controlling the levels of endosomal PI(3,4)P 2 , differentially regulates growth and nutrient signaling via Akt and mTORC1 in HAP1 cancer cells.

Discussion
In this study, we provide evidence that synthesis and turnover of PI(3,4)P 2 are spatially segregated within the endocytic pathway to couple endocytic membrane traffic to growth factor and nutrient signaling. We show that, although an exogenous supply of either PI(3)P or PI(3,4)P 2 can, in principle, rescue functional endocytic defects elicited by loss of PI3KC2␣, CCP dynamics in living cells are mainly controlled by PI(3,4)P 2 under physiological conditions, in contrast to the model suggested by He et al. (16) on the basis of auxilin-derived lipid/ protein sensors; see also the study by Hammond and co-workers (33) regarding their specificity. Although PI3KC2␣ is able to synthesize PI(3)P in vitro and at endosomes in vivo (17,18), the high PI(4)P-to-PI substrate ratio at the plasma membrane (38) would render PI(3)P synthesis at cell surface CCPs inefficient. Instead, we propose that PI3KC2␣ uses the abundant plasma membrane lipid PI(4)P, whose local levels may be further elevated by clathrin-associated PI(4,5)P 2 5-phosphatases, to synthesize an endocytic pool of PI(3,4)P 2 that serves to recruit effector proteins such as SNX9/18 (Fig. 6E) and the flat BAR domain protein FCHSD2 (39) to promote CCP maturation prior to dynamin-mediated membrane fission. Unlike SNX9/ 18, FCHSD2 binds to PI(3,4)P 2 but not to PI(3)P and has been suggested to be responsible for the majority of local actin polymerization at CCPs (39).
Our results also do not support a function of the PI(3,4)P 2specific 4-phosphatase INPP4A/B in CME, proposed from in vitro experiments (20). We find INPP4A to be absent from CCPs, and loss of INPP4A in COS7 or HAP1 cells neither perturbs CCP dynamics nor causes defects in CME at the ultrastructural or functional levels (Figs. [3][4][5]. Our findings, together with earlier data by Shin et al. (19), indicate instead that INPP4A operates at the level of early endosomes to convert PI(3,4)P 2 , either derived from endocytic pools (e.g. PI3KC2␣) or from class I PI 3-kinase/SHIP2 signaling downstream of receptor activation, to PI(3)P (Fig. 6E). INPP4-mediated generation of a local pool of PI(3)P after endocytic vesicle fission has occurred (i.e. at the level of early endosomes) is consistent with the late timing of PI(3)P appearance post-scission, monitored by auxillin-2ϫ FYVE (16) and confirmed by us (data not shown), and may contribute to the multiple roles of PI(3)P in the endolysosomal system and in autophagy (1,40). Conceivably, it could also aid endosomal recruitment of SNX9 (20) or other PI(3)P-binding actin-polymerizing factors to promote endosomal F-actin polymerization.
Further support for a role of INPP4A at endosomes derives from our analysis of INPP4A function in signaling-active HAP1 cells. We show that loss of INPP4A results in hyperactivity of the PI(3,4)P 2 effector Akt and local repression of mTORC1 signaling by accumulated endolysosomal PI(3,4)P 2 pools, consistent with the idea that PI(3,4)P 2 differentially regulates cell signaling, depending on its subcellular location and cellular nutrient conditions (24,25,34). How such differential regulation, e.g. of Akt and mTORC1, is accomplished at the molecular mechanistic level is unclear. PI(3,4)P 2 can directly bind to Akt and has been shown to be required for sustained Akt activation (32,36,41,42), possibly in conjunction with additional factors, e.g. other proteins activated downstream of class I PI 3-kinase signaling cascades. The mechanism of PI(3,4)P 2 -mediated repression of mTORC1 activity at late endosomes/lysosomes remains incompletely understood but may involve inhibitory association of 14-3-3 proteins with the Raptor subunit of mTORC1 (24), although additional mechanisms may contribute. Future studies will be needed to examine the cross-talk and regulation of different PI(3,4)P 2 effectors in cell signaling.
A somewhat surprising finding is the observation that loss of INPP4A in COS7 cells, which do not endogenously express INPP4B, did not result in grossly elevated PI(3,4)P 2 levels. Hence, PI(3,4)P 2 must be hydrolyzed by other phosphatases in these cells. A promising candidate for this activity is the PI 3-phosphatase PTEN, an enzyme localized to endosomes (23), in addition to its established function in removal of PI(3,4,5)P 3 at the plasma membrane (41). Recent work by several laboratories has uncovered a direct role of PTEN in PI(3,4)P 2 hydrolysis

Spatial segregation of PI(3,4)P 2 synthesis and turnover
in several cell types (33,43). Moreover, consistent with a role in PI(3,4)P 2 hydrolysis, PTEN has been found to reduce signaling at nascent CCPs and to slow their maturation (44). It therefore seems likely that loss of function of INPP4A/B is compensated by the 3-phosphatase activity of PTEN, resulting in production of PI(4)P, a lipid found in secretory organelles, including recycling endosomes (31), and at the plasma membrane (45). How PI(3,4)P 2 is funneled to these enzymatically distinct pathways of turnover remains enigmatic and will need to be addressed in future studies.

Experimental procedures
Cell culture COS7 cells obtained from the ATCC were cultured in DMEM (Gibco) with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). HAP1 WT cells and INPP4A KO cells were obtained from Horizon TM and cultured in Iscove's modified Dulbecco's medium (Gibco) supplied with 10% FBS and 1% penicillin/streptomycin. Cells were not used beyond passage 30. All cell lines were routinely tested for mycoplasma contaminations on a monthly basis.

siRNA and plasmid transfection
Cells were seeded and transfected with siRNA on day 1 using jetPRIME (Polyplus Transfection) according to the manufacturer's instructions by reverse transfection, expanded on day 2 before a second round of knockdown was performed on day 3 using X-tremeGENE (Roche) according to the manufacturer's instructions, and processed on day 4. Rescue transfections were performed on day 3 using X-tremeGENE according to the manufacturer's instructions. For experiments not involving siRNA, plasmid transfections were done using jetPRIME.

Transferrin uptake and surface labeling
Cells treated with siRNA or transfected with different plasmids were seeded on coverslips coated with Matrigel (BD Biosciences). Before the experiment, cells were treated in serumstarved medium for 1 h and used for transferrin uptake or transferrin receptor surface labeling. For the rapalog-induced system (Takara, AP21967), rapalog or DMSO was added during 1-h starvation. For transferrin uptake, cells were incubated with 25 mg/ml transferrin-Alexa 647 (Molecular Probes, Invitrogen) for 10 min at 37°C. Cells were first washed twice with ice-cold PBS containing 10 mM MgCl 2 and then acid-washed at pH 5.3 (0.2 M sodium acetate and 200 mM sodium chloride) on ice for 1 min to remove surface-bound transferrin. Cells were then washed twice more with ice-cold PBS containing 10 mM MgCl 2 and fixed with 4% paraformaldehyde (PFA) for 45 min at room temperature. For surface labeling, cells were incubated with 25 mg/ml transferrin-Alexa 647 at 4°C for 45 min, washed three times with ice-cold PBS (10 mM MgCl 2 ) on ice for 1 min, and fixed with 4% PFA for 45 min at room temperature. Transferrin labeling was analyzed using a Nikon Eclipse Ti micro-

Spatial segregation of PI(3,4)P 2 synthesis and turnover
scope and ImageJ software. Internalized transferrin per cell was quantified and normalized to the amount of surface-bound transferrin determined in the same experiment as a measure of the efficiency of internalization.

TIRF microscopy
TIRF imaging was performed using a Zeiss Axiovert200M microscope equipped with an incubation chamber (37°C and 5% CO 2 ), a 3100 TIRF objective, and a dual-color TIRF setup from Visitron Systems using Slidebook imaging software (3i Inc.). For analysis of CCP dynamics, a time-lapse series of 3 min with a frame rate of 2 s/frame was recorded.

Automated CCPs tracking
CCPs were tracked automatically (46) and statistically analyzed as follows. An instantaneous density was used as a metric for CCP lifetime density. Conceptually, the instantaneous density is the average density of CCPs (of a specific lifetime or lifetime range) in a given frame of the movie, i.e. the number of CCPs that are visible per area, in units of [number of events/ square microns], with no time dependence. 150 s was set as a threshold to distinguish dynamic from persistent CCPs (i.e. defined as CCPs with lifetimes of more than 150 s). The instantaneous density is calculated by counting the number of visible CCPs per area in each frame. Then the number of dynamic CCPs of the known lifetime (i.e. visible CCPs with uncensored trajectories below 150 s) is corrected for the corresponding observation bias, where the correction factor in this case depends on both CCP lifetime and the frame number, because more censored trajectories are visible at the beginning and the end of the movie than in the middle. The number of persistent CCPs is the sum of both uncensored and censored CCPs of a known lifetime of 150 s or more (which may include structures that are visible continuously for the entire length of the movie) without bias correction. Following this approach, the fraction of dynamic CCPs of all CCPs represents the unitless ratio of dynamic CCP instantaneous density over total CCP instantaneous density, where total ϭ dynamic ϩ persistent.

Colocalization probability
Colocalization probability is calculated in analogy to intensity plots of lifetime cohorts (46,47), except that, for the purposes of this analysis, the fluorescence intensity is replaced by the colocalization probability. The colocalization probability here is defined as the probability that a spot is detected in the second (dependent) color channel within a threshold distance of the first (reference) color channel, where that threshold distance is defined by the resolution limit (Rayleigh criterion for diffraction-based resolution).

EM
For thin-section EM analysis of clathrin-coated structures, cells were grown in 6-cm Petri dishes, fixed with 2% glutaraldehyde in PBS, mechanically detached, post-fixed with 1% osmium tetroxide in cacodylate buffer, and processed for epoxy resin embedding. Embedded cell pellets were ultrathin-sectioned and imaged by transmission EM using a Zeiss 900 transmission electron microscope. Clathrin-coated structures where quantified along the plasma membrane in cell profile montages and normalized to the length of the plasma membrane perimeter.

CRISPR/Cas9-mediated genome engineering
Primers for guide RNA generation in COS7 cells for INPP4A were designed as follows: forward, 5Ј-GGAAGCCCGCTGCA-CCGCAC-3Ј; reverse, 5Ј-GTGCGGTGCAGCGGGCTTCC-3Ј. The primers were annealed and cloned into the pX330-U6-Chimeric_BB-CBh-hSpCas9 plasmid (Addgene, 42230) using the BbsI site. COS7 cells were transfected with the px330-plasmid containing the guide RNA and a fluorescent marker (GFP) for indication of successfully transfected cells for 72 h. GFPexpressing cells were selected by FACS and plated on 96-well plates to obtain single-cell clones. Colonies were expanded and screened by immunoblotting using specific anti-INPP4A antibodies to identify KO clones.

Immunofluorescence
Cells were seeded on coverslips coated with Matrigel (BD Biosciences). For PI(3,4)P 2 staining, cells were washed once with PBS containing 10 mM MgCl 2 and fixed in 2% PFA either with (COS7 cells) or without (Hap1 cells) 0.5% glutaraldehyde for 20 min at room temperature. Cells were washed three times with PBS containing 50 mM NH 4 Cl and processed for permeabilization with PBS containing 0.5% saponin, 1% BSA, and 10% normal goat serum for 30 min. Primary and secondary antibodies were incubated for 1-2 h in PBS buffer containing 1% BSA and 10% normal goat serum. Cells were washed three times with PBS after primary and secondary antibody treatment and preceding mounting with 4Ј,6-diamidino-2-phenylindole. For PI(3)P staining, COS7 cells were washed once with PBS containing 10 mM MgCl 2 and fixed in 2% PFA for 15 min at room temperature. Cells were washed three times with PBS containing 50 mM NH 4 Cl and permeabilized with 20 M digitonin in buffer A (20 mM Pipes (pH 6.8), 137 mM NaCl, and 2.7 mM KCl) for 5 min. After washing with buffer A three times, cells were incubated for 30 min in blocking buffer (buffer A with 5% normal goat serum and 50 mM NH 4 Cl) containing 0.055 mg/ml GFP-FYVE. Primary and secondary antibodies were incubated for 1 h in buffer A containing 5% normal goat serum. Cells were post-fixed with 2% PFA for 5 min and washed three times with PBS containing 50 mM NH 4 Cl and twice with PBS.
For antibody staining, COS7 cells were washed once with PBS containing 10 mM MgCl 2 and fixed in ice-cold methanol at Ϫ20°C for 10 min. Cells were then blocked for 30 min in blocking buffer (10% normal goat serum in PBS with 0.02% Triton X-100). Primary and secondary antibodies were incubated for 1-2 h in the same buffer. For fixation in 4% PFA, 0.05% saponin was used instead of Triton X-100 in blocking buffer.

Spatial segregation of PI(3,4)P 2 synthesis and turnover
Rescue of defective CME by membrane-permeant PI 3-phosphates COS7 cells were washed twice in DMEM with 4.5 g/liter glucose without FBS prior to addition of PIP/AMs. PI(3)P/AM and PI(3,4)P 2 /AM were dissolved in dry DMSO to a stock concentration of 50 mM and mixed with an equal volume of 10% pluronic F127 to enhance solubility. After thorough mixing, the sample was diluted in DMEM to a final concentration of 100 M and incubated with COS7 cells for 20 min. An equal volume of DMEM supplemented with 20% FBS was added to the cells prior to TIFR imaging.

Data analysis and statistics
Data are expressed as a means Ϯ S.E. of at least three experiments. The statistical significance between groups was evaluated with one-way ANOVA or unpaired t test. Significant differences were accepted at p Ͻ 0.05.