Dynamin triple knockout cells reveal off target effects of commonly used dynamin inhibitors

Clathrin-mediated endocytosis is a very well-characterized process that functions in eukaryotes for the selective internalization of cell surface molecules and extracellular materials (Conner and Schmid, 2003; Doherty and McMahon, 2009; Kirchhausen, 2000; Robinson, 2004). In neurons, clathrin-mediated endocytosis is additionally implicated in neurotransmission because of its role in the endocytic recycling of synaptic vesicles at nerve terminals (Brodin et al., 2000; Dittman and Ryan, 2009; Heuser and Reese, 1973; Saheki and De Camilli, 2012).

Clathrin-mediated endocytosis is a highly coordinated process that begins with the assembly of clathrin coat components at the plasma membrane through interactions with plasma membrane lipids and with membrane proteins destined for internalization by this endocytic pathway. The nascent bud grows and invaginates with the assistance of multiple accessory factors and the budding vesicle is finally released into the cytoplasm via a dynamin-mediated membrane fission reaction (Faelber et al., 2012; Ferguson and De Camilli, 2012; Schmid and Frolov, 2011). Previous genetic investigations in many model organisms, including studies of temperature-sensitive alleles in worms and flies (Clark et al., 1997; Koenig and Ikeda, 1989; van der Bliek and Meyerowitz, 1991), as well as knockout and conditional knockout studies in mammalian cells and mice (Ferguson et al., 2007; Ferguson et al., 2009; Liu et al., 2008; Raimondi et al., 2011), strongly supported a critical role of dynamin in endocytic membrane fission. Such a role is additionally supported by quantitative live-cell imaging, which revealed that the peak of dynamin accumulation at clathrin-coated pits coincides with the membrane fission reaction (Taylor et al., 2012). Likewise, the ability of dynamin to cause fission of membrane tubules has been well established in vitro (Bashkirov et al., 2008; Morlot et al., 2012; Pucadyil and Schmid, 2008; Roux et al., 2006). Dynamin assembles into polymers on membrane tubules (Pucadyil and Schmid, 2008; Roux et al., 2006; Zhang and Hinshaw, 2001) and recent structural studies (Chappie et al., 2010; Faelber et al., 2011; Ford et al., 2011) have made progress towards unraveling the detailed molecular mechanism through which dynamin oligomerization and GTP hydrolysis may be coordinated to induce membrane scission.

Mammalian genomes contain three dynamin genes (DNM1, DNM2 and DNM3) whose protein products, dynamin 1, 2 and 3, share ∼80% overall homology and play at least partially redundant roles during the membrane fission reaction of clathrin-mediated endocytosis (Cao et al., 1998; Cook et al., 1996; Ferguson et al., 2007; Raimondi et al., 2011). However, their expression patterns are very different. Dynamin 1 is expressed selectively and at very high levels in neurons, where it is crucially required for synapses to efficiently recycle synaptic vesicles during intense activity (Ferguson et al., 2007; Hayashi et al., 2008; Lou et al., 2008; Armbruster et al., 2013). Indeed, synaptic transmission defects limit the average lifespan of dynamin 1 KO mice to less than 2 weeks (Ferguson et al., 2007). Dynamin 2 is expressed ubiquitously (Cao et al., 1998; Cook et al., 1996), and the knockout of dynamin 2 in mice results in early embryonic lethality in agreement with its house-keeping functions (Ferguson et al., 2009). Dynamin 3 is found most prominently in the brain (but at much lower levels than dynamin 1) and testis (Cao et al., 1998; Cook et al., 1996; Ferguson et al., 2007). Although dynamin 3 KO mice do not exhibit obvious neurological or male fertility defects (Raimondi et al., 2011), dynamin 1, 3 double KO mice are more severely affected than dynamin 1 single KOs as revealed by their short lifespan (only several hours), synaptic transmission dysfunction and membrane trafficking defects at synapses (Lou et al., 2012; Raimondi et al., 2011). The additive effect of dynamin 1 and 3 knockout alleles highlights a redundant role of different dynamin isoforms in supporting endocytosis. Endocytosis appears to be controlled, at least to some extent, by the overall abundance of each isoform rather than by major functional differences between them.

Dynamin 2 KO fibroblasts have been used to investigate the contributions of dynamin isoforms to cellular processes common to all cell types (Ferguson et al., 2009; Liu et al., 2008). These cells exhibit defects in clathrin-mediated endocytosis, but such defects are partially compensated by the unexpected expression of dynamin 1 in these cells. Thus, dynamin 1 and 2 double KO mouse fibroblasts (henceforth described as DKO) were generated from mice with floxed dynamin alleles and transgenic for 4-hydroxytamoxifen (OHT)-inducible Cre recombinase (Cre-ER) (Ferguson et al., 2009). Conditional DKO cells can be grown in vitro and gene recombination to produce DKO cells can be induced by addition of 4-hydroxytamoxifen. DKO fibroblasts have a much more severe defect in clathrin-mediated endocytosis than cells lacking dynamin 2 alone, although fluid-phase endocytosis is not impaired (Ferguson et al., 2009). Endocytic intermediates that accumulate in these cells are deeply invaginated clathrin-coated pits connected to the plasma membrane by long, narrow tubules. Such tubules are surrounded by BAR-domain-containing proteins, F-actin and actin regulatory proteins (Ferguson et al., 2009). Although DKO cells survived for at least several weeks in culture, they failed to proliferate (Ferguson et al., 2009) and exhibited multiple signaling defects (Shen et al., 2011; Sousa et al., 2012). Given the potential overlapping role of the three dynamin isoforms, we considered the possibility that residual dynamin activity provided by dynamin 3 could support the viability of DKO cells, even if this protein is undetectable by available antibodies in these cells. A definitive assessment of the cellular function of dynamin requires the deletion of all 3 dynamin isoforms. Dynamin triple KO cells would also represent the optimal model to test the dynamin dependence of biological processes and to assess potential off-target action of dynamin inhibitors.

To address these issues, we generated fibroblasts from mice harboring floxed alleles of all three dynamin genes and also expressing Cre-ER. Triple KO (TKO) cells obtained from these conditional KO cells upon tamoxifen-induced gene recombination had the same phenotype as dynamin 1 and 2 DKO cells. Surprisingly, dynasore (Macia et al., 2006) and Dyngo-4a (Harper et al., 2011; Howes et al., 2010; McCluskey et al., 2013), two widely used and structurally related small molecule inhibitors of dynamin, still produced a robust impairment of fluid-phase endocytosis and peripheral membrane ruffles in TKO cells. Given the property of these drugs to cause these very strong effects even in cells where dynamin is absent, caution is required in the interpretation of their cellular action.

In our previous characterization of DKO fibroblasts, we found that these cells remain viable over several weeks in culture but exhibit a severe defect in proliferation (Ferguson et al., 2009). Although immunoblotting experiments with an anti-dynamin-3 antibody that yielded a very strong signal on blot of brain lysates (Ferguson et al., 2007; Raimondi et al., 2011) did not indicate the presence of dynamin 3 in either WT or DKO mouse fibroblasts (Fig. 1A and supplementary material Fig. S1), it remained possible that levels of this dynamin isoform (below our threshold for detection) contributed to the survival of DKO cells. To assess for the potential expression of low levels of the DNM3 gene in mouse fibroblasts, lysates from wild-type cells were affinity purified with immobilized GST fused to SH3 domains 1–4 of Tuba, a high-avidity ligand for all three dynamin isoforms (Ferguson et al., 2007; Salazar et al., 2003), and the bound material was analyzed by mass spectrometry. This strategy detected four peptides that uniquely correspond to the mouse dynamin 3 sequence: K.DFINSELLAQLYSSEDQNTLMEESAEQAQR.R; K.HVFALFNTEQR.N; R.IEGSGDQVDTLELSGGAK.I; and R.FLELACDSQEDVDSWK.A, thus demonstrating at least low level expression of this protein in our fibroblast cultures.

To directly determine whether the very low levels of dynamin 3 may perform functions that are essential for viability in DKO cells, we capitalized on the dynamin 1 and 2 conditional double KO (Ferguson et al., 2009) and the dynamin 3 conditional KO (Raimondi et al., 2011) mouse lines that we have previously described. These mutant mice were interbred with one another and with mice transgenic for Cre-ER (Badea et al., 2003; Feil et al., 1996) to generate tamoxifen-inducible triple conditional KO mice (Dnm1^loxP/loxP; Dnm2^loxP/loxP; Dnm3^loxP/loxP; Cre-ER^+/0). These mice were healthy with no obvious abnormalities or fertility defects. Dynamin 1, 2 and 3 TKO cells were generated in vitro by tamoxifen treatment of fibroblasts isolated from these mice. Immunoblotting with isoform-specific antibodies did not detect any dynamin signal in these cells (Fig. 1A).

TKO cells ceased to proliferate around day 4 after exposure to tamoxifen, and PCR genotyping of their genomic DNA confirmed the efficient recombination of all three dynamin genes (Fig. 1B). They could re-attach to the substrate and spread following trypsinization (see supplementary material Fig. S3), and their morphology was not grossly different from cells with floxed alleles but untreated with tamoxifen. Thus, the very low expression of dynamin 3 in cultured mouse fibroblast is not essential for the surprising viability of DKO cells, leading to the conclusion that although the function of dynamin is crucial for normal proliferation, dynamin-dependent cellular processes are dispensable for cell viability.

The same results were obtained using another line of dynamin TKO cells that were derived from mutant mice with the following genotype: Dnm1^−/−; Dnm^loxP/loxP; Dnm3^−/−; Cre-ER^+/0. Fibroblasts derived from Cre-ER^+/0 mice with floxed alleles of all three dynamins were used for all subsequent studies. In all experiments, pools of cells not treated with tamoxifen were used as a control.

As reported for DKO cells (Ferguson et al., 2009), uptake of fluorescent transferrin, a standard assay of clathrin-mediated endocytosis, was drastically impaired in dynamin TKO fibroblasts (Fig. 2A). In contrast, no inhibition (in fact a slight increase) in the uptake of dextran, a marker of fluid-phase endocytosis, was observed in TKO cells (see Fig. 4).

As expected, in TKO cells, the block in clathrin-dependent internalization of transferrin correlated with an increase in signal for endocytic clathrin coats, as revealed by anti-AP2 (α-adaptin) immunofluorescence (Fig. 2B). Furthermore, clathrin-coated pits had the typical properties and appearance of the deeply invaginated pits arrested at the pre-fission stage that had been described in DKO cells (Ferguson et al., 2009). First, they had long narrow tubular connections to the plasma membrane, as demonstrated by standard transmission electron microscopy (Fig. 3A–D) and by a positive signal for a cytochemical reaction that selectively labels invaginations connected to the cell surface (Fig. 3E–G). Second, they were enriched in endophilin 2, a BAR-domain-containing protein involved in growth and stabilization of the tubular membrane neck of clathrin-coated pits (Fig. 2C) and for the Arp2/3 (p34 subunit) actin nucleating complex (Fig. 2D) (Ferguson et al., 2009).

Other changes observed in DKO cells relative to controls, such as a major reduction of caveolin levels (Ferguson et al., 2009) and a robust increase in acetylated tubulin (Ferguson et al., 2009; Tanabe and Takei, 2009) were observed in TKO cells (supplementary material Fig. S2). These effects were not investigated further and it remains unknown whether or not they reflect direct actions of dynamin in these processes.

In summary, the similar viability and endocytic phenotypes observed in DKO and in TKO cells do not support a major relevant contribution of the very low levels of dynamin 3 to these features in fibroblasts.

TKO cells provide the opportunity to investigate the specificity of dynamin inhibitors. We focused on two commonly used dynamin inhibitors, dynasore and Dyngo-4a. Dynasore, the first reported dynamin inhibitor (Macia et al., 2006), is widely used to investigate the role of dynamin in cellular events. Dyngo-4a, a close structural analog of dynasore, has been increasingly used in recent studies because of its higher potency in dynamin inhibition (Harper et al., 2011; Howes et al., 2010; McCluskey et al., 2013). We reasoned that if the effects of these dynamin inhibitors on cells are mediated through dynamin binding and inhibition, then cells that lack dynamin (i.e. TKO cells) should be insensitive to these drugs.

As mentioned above, fluid-phase endocytosis is not impaired in TKO cells relative to the control cells. Dynasore has been reported to inhibit this endocytic pathway (Macia et al., 2006) and we confirmed this effect in WT cells, and found that Dyngo-4a inhibits this pathway to a similar degree and does so at a lower concentration (Harper et al., 2011; Howes et al., 2010; McCluskey et al., 2013). Surprisingly, both dynasore and Dyngo-4a were still able to inhibit dextran uptake even in the TKO cells (Fig. 4A,B).

Although dynamin TKO cells did not exhibit obvious defects, relative to controls, in morphology or in attachment and spreading after replating, dynasore was shown to inhibit cell spreading (Macia et al., 2006) and to suppress the formation of lamellipodia (Yamada et al., 2009). We have now observed that dynasore triggers these effects even in TKO cells, as staining of F-actin with phalloidin revealed the shrinkage of TKO cells after dynasore treatment (supplementary material Fig. S3).

To further investigate the effect of dynasore and Dyngo-4a on the actin cytoskeleton, TKO cells transiently expressing a fluorescent F-actin reporter (the calponin homology domain of utrophin fused to GFP) (Burkel et al., 2007) were imaged by time-lapse spinning disk confocal microscopy before and during treatment with dynasore or Dyngo-4a. Inspection of cortical cell regions showed membrane ruffles that extended and retracted over a distance of several micrometers over a period of a few seconds in both control or TKO fibroblasts, revealing the dynamin independence of this process (supplementary material Fig. S4). However, the intense F-actin signal in lamellipodia was lost in both control and TKO cells following either dynasore or Dyngo-4a treatment (Fig. 5A,B; supplementary material Movies 2, 3, 5, 6, 8, 9, 11, 12), whereas no effect was observed after addition of DMSO, the solvent used to solubilize dynasore and Dyngo-4a (supplementary material Fig. S4,and Movies 1, 4, 7, 10.). The drastic inhibition (Fig. 5D) occurred within minutes of drug addition, as shown by galleries of time-lapse images (Fig. 5C) and quantitative data of membrane ruffling (Fig. 5E). We conclude that dynamin does not play an essential role in peripheral membrane ruffling and that the cessation of membrane ruffling in TKO cells upon dynasore or Dyngo-4a treatment arises from an off-target effect of these inhibitors.

In this study, we have reported the generation and characterization of dynamin 1, 2 and 3 triple conditional KO mouse fibroblasts. Following Cre-recombinase-induced inactivation of all three genes, these cells lose expression of all three dynamins. TKO cells display the major defects previously observed in DKO cells, without additional obvious changes, supporting conclusions on the overall cellular functions of dynamin reported in studies of DKO cells (Ferguson et al., 2009). Thus, although we have now found that very low levels of dynamin 3 can be detected in mouse fibroblasts in culture, as detected by affinity purification and mass spectrometry, this small pool of dynamin 3 does not contribute significantly to fibroblast physiology. This result contrasts with observations in neurons where the absence of the major neuronal dynamin, dynamin 1, unmasks an important function of dynamin 3 in synaptic physiology, as dynamin 1 and 3 double knockout neurons exhibit more severe defects in synaptic function than dynamin 1 KO neurons (Lou et al., 2012; Raimondi et al., 2011).

Our results corroborate our hypothesis that a primary function of dynamin is to support clathrin-mediated endocytosis (Ferguson et al., 2009; Ferguson and De Camilli, 2012), as fluid-phase endocytosis appeared to be unperturbed in TKO fibroblasts. Similar results were observed in dynamin 1 KO neurons and also in dynamin 1 and 3 DKO neurons (i.e. nerve terminals nearly devoid of dynamin). In these neurons, clathrin-mediated endocytosis of synaptic vesicles was strongly impaired, while robust bulk endocytosis, a form of fluid-phase endocytosis, persisted [(Hayashi et al., 2008) and our unpublished observations]. Partial impairment of dynamin function by non-pharmacological methods was shown to inhibit fluid-phase endocytosis in some studies (Cao et al., 2007; Damke et al., 1995), but not in others (Liu et al., 2008; Schlunck et al., 2004). The disruption of all three dynamin genes, as we have done here, provides an unambiguous system to test the cellular processes that do and do not require this extensively studied GTPase.

The generation of dynamin TKO cells allowed us to study the specificity of the effects of two structurally related, widely used dynamin inhibitors, dynasore and Dyngo-4a. Dynasore was shown to inhibit not only clathrin-mediated endocytosis, but also fluid-phase endocytosis (Macia et al., 2006), a finding that we have confirmed. As we have demonstrated here, Dyngo-4a also inhibits this endocytic pathway. However, we have observed that both dynasore and Dyngo-4a still produce the inhibitory effect on fluid-phase endocytosis in cells where its intended target, dynamin, has been eliminated, indicating that this action represents an off-target effect. Additionally, we have found that both dynasore and Dyngo-4a have a powerful blocking effect on membrane ruffling and that this action is also independent of dynamin as it still occurs in TKO cells. This result does not exclude the possibility that dynamin may have some regulatory role in ruffle dynamics given its localization in ruffles and its many interactions with actin regulatory proteins (McNiven et al., 2000; Orth and McNiven, 2003; Schafer, 2004; Yamada et al., 2013), but strongly cautions against the use of these dynamin inhibitors as tools to investigate the role of dynamin in this process.

The mechanisms whereby dynasore, a noncompetitive inhibitor of the GTPase activity of dynamin (Macia et al., 2006), and its close structural analog Dyngo-4a (Harper et al., 2011; Howes et al., 2010; McCluskey et al., 2013) produce their off-target effects remain unknown. Dynasore was originally shown to also inhibit Drp1, a dynamin-like GTPase involved in mitochondrial division, although less efficiently (Macia et al., 2006). Effects of this drug on other members (Faelber et al., 2013; Ferguson and De Camilli, 2012) of the dynamin superfamily of GTPases are a possibility. However, off-target effects of dynasore and Dyngo-4a may also be due to GTPase-independent actions. In addition to dynasore and Dyngo-4a, a variety of other dynamin inhibitors have recently been developed, including Iminodyn-22 (Hill et al., 2010), Dynole 34-2 (Hill et al., 2009), RTIL-13 (Zhang et al., 2008), MiTMAB and OcTMAB (Quan et al., 2007), and indole 24 and 25 (Gordon et al., 2013). Given the findings reported here concerning the non-specific effects on dynasore and Dyngo-4a on fluid-phase endocytosis and membrane ruffling, results arising from the use of such chemicals should be interpreted with caution and be corroborated by independent methods. For example, inhibition of VEGFR2 signaling in endothelial cells by dynasore was taken as an indication that signaling by this receptor is dependent on endocytosis (Sawamiphak et al., 2010; Wang et al., 2010). However, the opposite result, increased signaling of VEGFR2, was observed in endothelial cells where internalization of VEGFR2 was inhibited by the genetic ablation of the endocytic adaptors epsin 1 and 2 (Pasula et al., 2012), or by RNAi-mediated suppression of dynamin 2 (Satish Pasula, Hong Chen, William Sessa and P. D. C., unpublished observations). Genetic models, such as the dynamin TKO cells presented in this study, should serve as an important tool in the conclusive assessment of the dynamin dependence of results obtained with dynamin inhibitors.

Abnormal dynamin function has been linked to diseases in humans and animals. Mutations in the ubiquitously expressed dynamin 2 isoform cause specific, dominantly inherited forms of Charcot–Marie–Tooth disease and centronuclear myopathy in humans (Bitoun et al., 2005; Züchner et al., 2005). Based on disorders arising from dynamin mutations in other mammals, such as exercise-induced collapse in dogs (Patterson et al., 2008) and seizures in the fitful mouse (Boumil et al., 2010), additional dynamin-dependent conditions probably remain undiscovered in humans. Dynamin TKO cells represent a powerful system for testing the function of disease-related mutations of dynamin.

The conditional KO (floxed) and KO alleles of dynamin 1, 2 and 3 used in this study were previously described (Ferguson et al., 2007; Ferguson et al., 2009; Raimondi et al., 2011). Animal care and use were in accordance with our institutional guidelines.

Fibroblast cultures were derived from mice and maintained as previously described (Ferguson et al., 2009). The homologous recombination of the conditional KO alleles was carried out by the activation of an estrogen-receptor–Cre-recombinase fusion protein (Cre–ER) upon 4-hydroxytamoxifen (Sigma) treatment according to our previously established protocol (Ferguson et al., 2009). Briefly, cells were incubated with 3 µM tamoxifen for 2 days, resulting in dynamin depletion at 5–6 days from the start of the treatment period. TKO cells were generally used for experiments between 7 and 9 days. Control cells were the triple conditional KO cells without 4-hydroxytamoxifen treatment. Cell culture, immunoblotting and immunofluorescence were performed as described previously (Ferguson et al., 2009).

Cells for imaging experiments were electroporated with the Amaxa Nucleofector method (protocol A-24) and grown for 16–24 hours in culture medium on 12 mm glass coverslips or on 35 mm glass bottom dishes (Mat-Tek, Ashland, MA, USA) at sub-confluent densities.

DNeasy Blood and Tissue Kit (Qiagen) was used to extract genomic DNA from cultured fibroblasts. The following oligonucleotide primers were used for PCR-based genotyping of the respective alleles: floxed allele of dynamin 1: 5′-TTGTGTATGTGAGTGCACCCATGC-3′ and 5′-CAGCTGGGTATAATGAGGCCTCATC-3′; floxed allele of dynamin 2: 5′-GCAGGAAGACACACAACTGAAC-3′ and 5′-CCTGCTAGTGACCTTTCTTGAG-3′; floxed allele of dynamin 3: 5′-GACATGTTAACATAGGCTAAACC-3′ and 5′-CAGTGCCTTCCAAGTTCAATTCC-3′; Esr-Cre transgene: 5′-CTTGCATGATCTCCGGTATTGA-3′ and 5′-ACATTTGGGCCAGCTAAACATG-3′.

The following antibodies were obtained from commercial sources: mouse anti-clathrin heavy chain (clone TD1, Affinity Bioreagents), mouse anti-α-adaptin (AP2 subunit; clone AP6, Affinity Bioreagents), rabbit anti-dynamin 1 (Epitomics, Burlingame, CA), mouse anti-acetylated α-tubulin (Sigma), mouse anti-caveolin 1 (BD Biosciences, San Jose, CA), rabbit anti-ARPC2 (Arp2/3) (Millipore, Billerica, MA), rabbit anti-GAPDH (Abcam, ab9485). Anti-rabbit and anti-mouse IgG (H⁺L) HRP-conjugated secondary antibodies were obtained from BioRad (Hercules, CA). Alexa-Fluor-594–phalloidin and Alexa-Fluor-488- or -594-conjugated secondary antibodies were obtained from Invitrogen. The following antibodies were previously generated in our laboratory: rabbit anti-endophilin 2 (Milosevic et al., 2011), rabbit anti-dynamin 2 and mouse anti-dynamin 3 (Ferguson et al., 2007). A plasmid expressing the GFP-tagged calponin homology domain of utrophin (Utr-CH) was kindly provided by William Bement (University of Wisconsin–Madison) (Burkel et al., 2007).

For epifluorescence imaging, samples were imaged with a Zeiss Axioplan2 microscope using a Plan-Apochromatic 40× objective and a Hamamatsu ORCA II digital camera under the control of MetaMorph v7.1.2 software (Molecular Devices). For confocal imaging, cells were imaged on a Zeiss LSM 710 laser scanning confocal microscope equipped with a 63× objective using ZEN (Carl Zeiss, Inc.) software or by spinning disk confocal microscopy: Improvision UltraVIEW VoX system including a Nikon Ti-E Eclipse inverted microscope (equipped with a 603 CFI PlanApo VC, NA 1.4 objective) and a spinning disk confocal scan head (CSU-X1, Yokogawa) driven by Volocity (Improvision) software.

For fixed samples, a 2 µm thick section was imaged with optical sections acquired at 100 nm intervals in the z-axis with exposure times that ranged from 150 mseconds to 300 mseconds. The slices were iteratively deconvolved with Volocity software, and collapsed to a single image by maximal intensity projection with ImageJ software (version 1.43u, NIH). For living cells, imaging was performed at 37°C on a heated stage in Phenol-Red-free Dulbecco's Modified Eagle's Medium (DMEM) supplemented with high glucose and L-glutamine (Gibco). Time-lapse images were taken at a rate of 5 frames per minute and exposure times ranged from 50 to 100 mseconds per frame.

Dynamin TKO cells were grown overnight on 35 mm glass bottomed (thickness = 0.15 mm) dishes (Mat-Tek), washed with PBS and then fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) for 1 hour at room temperature. Following washing with 0.1 M sodium cacodylate buffer, cells were post-fixed in 1% OsO₄ in 0.1 M cacodylate buffer for 1 hour at room temperature, washed in 50 mM sodium maleate buffer (pH 5.2), and en bloc stained with 2% uranyl acetate in 50 mM sodium maleate buffer (pH 5.2) for 1 hour in the dark. Cells were then dehydrated and embedded as previously described (Ferguson et al., 2007).

For the cytochemical labeling of surface-exposed membrane, glutaraldehyde-fixed cells were rinsed, incubated in 0.25% OsO₄ + 0.25% potassium ferrocyanide solution for 15 minutes, rinsed three times with cacodylate buffer, incubated for 15 minutes in 1% tannic acid, washed with cacodylate buffer (three times, 5 minutes each) and then with acetate buffer (three times, 5 minutes each) and finally stained en bloc with uranyl acetate as described above. All the solutions were made up in cacodylate buffer except for the uranyl acetate solution. The rapid washes allow some remaining tannic acid to react with uranyl salts to form a colloid, resulting in electron-dense staining on the plasma membrane and its invaginations in particular. This method was developed and communicated to us by John Heuser (Washington University and Kyoto University).

For dextran uptake, cells were incubated for 30 minutes at 37°C in the presence of fixable Alexa-Fluor-488-conjugated dextran (10,000 MW; Invitrogen) at a concentration of 0.5 mg/ml in Phenol-Red-free DMEM. For transferrin uptake, 5 µg/ml Alexa-Fluor-488–transferrin was bound to cells for 30 minutes at 4°C and further incubated at 4°C for 10 minutes. Cells were washed in PBS, fixed, mounted and visualized by confocal microscopy.

Dynasore with >99% purity (Tocris) and Dyngo-4a (Abcam) were dissolved in DMSO to make stock solutions, aliquoted and stored at −20°C (Kirchhausen et al., 2008). For immunofluorescence, cells were rinsed twice in DMEM, incubated with dynasore (80 µM) or Dyngo-4a (30 µM) in DMEM at 37°C for 30 minutes, then fixed and further processed as described above. For live-cell imaging, cells were washed twice in imaging buffer (Phenol-Red-free DMEM), and dynasore (80 µM final) or Dyngo-4a (30 µM) was added to the dish during image acquisition. For control experiments, an equivalent concentration of DMSO (∼0.15%) was added.

Fluorescent spots in fixed control and TKO cells were analyzed using ImageJ after background subtraction. For the quantification of fluorescent transferrin and of immunoreactivity for AP2, endophilin 2 and Arp2/3, all fluorescent puncta in a cell were selected and normalized against the footprint of the cell. For the quantification of dextran uptake, the fluorescent intensity of dextran in a cell (n = 10 cells) was measured and normalized against the cell footprint. For cell motility measurement in living cells, images were acquired at 12 second interval for 20 minutes by spinning disk confocal microscopy. Image time series were registered using the StackReg plug-in of ImageJ (Biomedical Imaging Group, École Polytechnique Fédérale de Lausanne, Switzerland). For the quantification of cell motility (Fig. 5D), images of each time point were subtracted from the image taken 48 seconds later. The resultant set of subtracted images was z-projected onto a single image using an average intensity method (ImageJ) and the change in fluorescent signal at the cell edge was quantified and normalized to the length of the cell perimeter. For the analysis of membrane ruffling (Fig. 5E), the change in fluorescence intensity of three randomly selected regions of interest (2×2 pixels) at the cell periphery in control or TKO cells upon DMSO, dynasore or Dyngo-4a treatment was measured. All data except those of membrane ruffling are presented as means ± s.e.m. (GraphPad Prism Software, La Jolla California USA). Statistical significance was determined using Student's t-tests; ***P<0.001, **P<0.01 and *P<0.02.

We thank Frank Wilson, Louise Lucast and Stacy Wilson for outstanding lab support, John Heuser for the communication of unpublished methods, Olof Idevall-Hagren, Mirko Messa and Jeremy Baskin for discussions.