CALM Regulates Clathrin-Coated Vesicle Size and Maturation by Directly Sensing and Driving Membrane Curvature

1During clathrin-mediated endocytosis (CME), clathrin-coated pits (CCPs) are most frequently initiated by the formation of a small patch of the heterotetrameric AP2 clathrin adaptor complex and clathrin at sites of high PtdIns4,5P2 concentration. If not aborted due to a lack of sufficient PtdIns4,5P2 and/or cargo, local membrane curvature steadily increases as more clathrin and a variety of additional clathrin adaptors are recruited until a bulbous membrane structure of ∼80–100 nm diameter is formed on top of a membrane stalk that can undergo scission to generate a clathrin-coated vesicle (CCV). The most abundant clathrin adaptors in endocytic CCVs isolated from tissue culture cells are CALM (clathrin assembly lymphoid myeloid leukemia protein) and AP2, each of which account for 30%–35% of the adaptors in a CCV. AP2 binds to and sorts general, often large, transmembrane cargo such as the transferrin receptor (TfR) into CCVs. CALM binds to and sorts the small R-SNAREs VAMPs 2, 3, and 8. CALM and its neuronal specific homolog AP180, possess large, natively unstructured, C-terminal tails containing clathrin, AP2 α- and β2-appendage binding sites, and a membrane-proximal, PtdIns4,5P2-binding N-terminal ANTH stacked-helical domain, defined as residues 19–289.

2Several studies have shown that depletion of CALM in tissue culture cells and vertebrate neurons or of its orthologs in D. melanogaster and C. elegans causes an increase in CCP/CCV size and probably also a decrease in their uniformity of shape. These morphological alterations most likely result from changes in membrane curvature/sculpting. However, structural investigations had shown that CALM possessed neither a BAR domain nor an amphipathic helix that could influence membrane curvature, which was consistent with CALM apparently not affecting membrane curvature in biophysical assays. In an attempt to reconcile these contradictory observations of a key cellular process, we set out to determine whether, and if so, how CALM could directly affect CCP/CCV size and thus the cargo-carrying capacity because these are two key features in understanding the mechanism of endocytic CCP/CCV formation. Here we show that CALM possesses an N-terminal amphipathic helix that regulates CCP/CCV size and maturation and hence endocytic rate.

3Endogenous CALM was depleted by siRNA and subsequent quantitated ultrastructural studies by electron microscopy (EM) and stimulated-emission depletion (STED) microscopy showed that CALM depletion results in substantially enlarged CCPs and CCVs compared to that of control cells (Figures 1, S1A, S1C, and S1E; see also ), but no discernable change in the number of clathrin-coated structures (Figures 1B and S1A). The CCPs/CCVs in control cells have an average diameter of ∼90 nm. This value increased by ∼2-fold upon depletion of CALM producing an ∼4-fold increase in surface area, hence an ∼8-fold increase in volume (Figures 1, S1C, and S1E). Importantly, our quantitation of CCP/CCV profiles of EM micrographs showed that “necked” CCPs/“closed” CCVs were the predominant fraction in control cells (∼70%), whereas the shallow “open” CCPs accounted for only ∼30% (Figures 1A and S1D). This was reversed in CALM-depleted cells, strongly suggestive of an alteration in the efficiency of CCP/CCV maturation, likely resulting from a reduced ability to curve/sculpt the plasma membrane (Figures 1A and S1D). In agreement with this, total internal reflection fluorescence (TIRF) microscopy revealed that the time taken to proceed from initiation of a CCP to its scission approximately doubles when CALM is depleted (Figures 2A and 2B).

4We then compared the effects of expressing endogenous levels of full-length, siRNA-resistant CALM-wild-type (WT), CALM(PIP−), a mutant which has no PtdIns4,5P2 binding due to the mutations Lys28Glu, Lys38Glu, and Lys40Glu and is therefore cytosolic or CALM(VAMP−), which has no R-SNARE binding due to the mutation Met244Lys in conjunction with endogenous CALM depletion (Figures S1A and S1B). Immunofluorescence indicated no obvious reduction in the number of punctate endocytic CCPs/CCVs (Figure S1A), whereas EM analysis of the CCP/CCV profiles showed that the expression of CALM-WT or CALM(VAMP−) rescued the relative ratios of open versus closed and sizes of CCPs/CCVs, whereas re-expression of CALM(PIP−) did not (Figures 1A and S1D). These data therefore indicate that CALM is not a major factor in CCP initiation, but rather that its presence strongly affects aspects of early to middle stages of CCP/CCV formation, and consequently influences the final CCV size. Because both CALM-WT and CALM(VAMP−) rescue phenotypes similarly, we further suggest that CALM’s SNARE binding is not a checkpoint during CCP/CCV formation.

5To substantiate our findings, we assessed the relative dynamics of AP2 and CALM during CCP formation in living cells with TIRF microscopy. These data showed that AP2 levels begin to decrease 15–20 s prior to the sharp decrease in the CALM levels that corresponds to the point when CCV scission occurs (Figures 2C–2F; see ). Within these 20 s prior scission, net membrane curvature must still increase to produce a necked and bulbous structure; hence, it is the level of CALM and not that of AP2 that correlates best with the increasing amount of positive curvature during CCP/CCV formation. CALM’s binding partners FCHO, NECAP, and clathrin also bind AP2 and so should be present in CCPs even when CALM is absent. This suggests that CALM’s effects on CCP/CCV formation are not indirect through the action of these binding partners, but are consistent with a direct effect of CALM itself.

6The impact of CALM on the size and shape of CCPs/CCVs led us to re-examine the structure of CALM to determine how it could directly affect membrane curvature. Inspection of the N-terminal residues 1–18 of CALM with the Heliquest server (http://heliquest.ipmc.cnrs.fr) suggested that this region could, in fact, assume an amphipathic helix (AH) (Figure 2G). Membrane-interacting AHs adopt their helical conformations when they attach to phospholipid membranes; and indeed, liposome-based surface plasmon resonance (SPR) measurements showed that this potential N-terminal AH of CALM (we now term residues 1–18 as AH0, to be consistent with previous literature on other clathrin adaptors) contributed to the membrane binding of CALM. CALM ANTH bound to PtdIns4,5P2 liposomes at physiological salt concentration (170 mM) with a KD of 1.5 μM, whereas deletion of AH0 (CALM ANTH[ΔH0]), or the mutation of the key hydrophobic AH0 residues Leu6, Ile10, and Val17 to serines (CALM ANTH[H0mut]) both led to a decrease in affinity by ∼8-fold (Figures 2H, 2I, and S2A). Furthermore, in the presence of PtdIns4,5P2-containing liposomes, WT CALM ANTH showed a small but reproducible increase in α-helical content by circular dichroism (CD) (Figure S2B).

7Comparison of our CALM ANTH:VAMP8 complex structure with previous structures of the human and Drosophila CALM ANTH domains showed that residues 5–14 can indeed form a short N-terminal helix that is connected to helix1 by a linker with the central axes of the helices diverging by ∼40° at residues 14–17 (Figures 3A, 3E, S3A, and S3B). The structure of a C-terminally truncated form of the ANTH domain of CALM (CALM ANTH(1–264)) (Table S1; Figures 3C, S3A, and S3C) although lacking the SNARE binding site, again showed residues 5–18 form a helix but this time extending helix1 by an additional four turns (Figures 3C, 3E, S3A, and S3C). The increased stabilizing energy that would arise from the continued α-helical H-bonding pattern seen in the structure of CALM ANTH(1–264) suggests that this extension of helix1 is the most likely conformation for residues 5–18 when the helix formation is induced by either membrane insertion or by fortuitous crystal packing. The failure to detect the existence and effects of this helix, in previous in vitro studies, likely results from this region’s high proteolytic sensitivity.

8In other peripheral membrane proteins such as ArfGAP1, membrane-interacting amphipathic helices show membrane curvature sensing (MCS) and are referred to as ALPS (amphipathic lipid packing sensing) helices. MCS arises from a combination of physicochemical effects including the insertion of the hydrophobic residue side chains of the helix into curvature-induced defects in membrane lipid packing and the interactions between the hydrophilic amino acid side chains of the helix and membrane phospholipid headgroups.

9When used in a sensitive microscope-based MCS assay on individual fluorescent liposomes, CALM ANTH showed a preference for binding to small liposomes <100 nm diameter (around the size of closing CCPs), demonstrated by the enlarged population of smaller liposomes with higher protein densities (Figures 4A and S4A–S4G). CALM ANTH(ΔH0) showed a significantly reduced ability to sense curvature (Figures 4A and S4A–S4G), with a tendency to populate markedly lower protein densities. Taken together, our in vitro data indicate that CALM indeed possesses MCS capability, which is largely mediated by its AH0.

10The deformation/tubulation of liposomes is often used as an indicator of membrane sculpting activity in vitro. CALM ANTH-WT at 0.5 μM was extremely efficient at extensively tubulating 200 nm PtdIns4,5P2-containing liposomes within 1 minute (Figures 4B, 4C, S4H, and S4I). CALM ANTH(ΔH0) and CALM ANTH(H0mut) at 0.5 μM were largely unable to deform the liposomes despite binding to them (Figure 4C). At concentrations of <0.1 μM, CALM ANTH-WT was unable to tubulate liposomes. The observed tubulation was not, however, reminiscent of the straight-sided tubules produced by BAR domains. It was instead “pearlized”, with the pearls having approximately the same diameter (40–50 nm) (Figures 4B, S4H, and S4I) as the smallest endocytic CCVs found in neuronal synapses, which are enriched for CALM and AP180. In previous liposome tubulation assays, CALM ANTH was used at concentrations of 4–20 μM. At the concentration of CALM used here, we calculate a resulting surface coverage of ∼16% (see Supplemental Experimental Procedures), a value close to that calculated for epsin, which forms smaller (17 nm diameter) tubules from liposomes. The value for CALM surface coverage in vitro resembles our estimate of its coverage in vivo of an endocytic CCV’s surface area of ∼17%. Our calculation (see Supplemental Experimental Procedures) shows the value for CALM will be much greater than that for epsin because it has been estimated that epsin constitutes less than 1% of clathrin adaptor content in endocytic CCVs. We conclude that the ANTH domain of CALM can drive pearlized liposome tubulation using its AH0.

11We propose, therefore, that in vivo the targeting of CALM into a forming CCP through its preferential association with curved membranes should have two major effects: CALM will stabilize existing membrane curvature (by effectively shifting the thermodynamic equilibrium in favor of curved structures and will also actively drive further membrane deformation/curvature toward that of endocytic CCVs (a diameter of ∼90 nm, which corresponds to CALM’s preferred binding curvature). Although we cannot definitively prove insertion of an unmodified AH0 of CALM into a membrane, the biochemical and biophysical properties of CALM’s AH0 are consistent with it undergoing membrane insertion, as is accepted for similar AHs in other proteins. This role for CALM is also consistent with its recruitment profile (Figures 2C–2F and ) and as such would also help to concentrate CALM-binding endocytic R-SNAREs in CCVs.

12In addition to helix insertion or wedging the other main contributor to protein driven membrane curvature comes from molecular crowding i.e., the lateral pressure exerted by colliding proteins that are attached on one side of a membrane. The degree of membrane curvature driven by molecular crowding depends on protein size, concentration, and membrane affinity. CALM, due to its ∼70 kDa molecular weight in combination with its >300 residue C-terminal unstructured region, should therefore contribute significantly to driving membrane curvature. The AH0 of CALM would also affect molecular crowding effects because AH0 not only increases the general membrane affinity of CALM, but also preferentially targets CALM to and thus concentrates it on the forming curved CCPs. CALM can thus initiate a positive amplification cascade of increasing membrane curvature starting from its initially low level in very early i.e., recently initiated CCPs to the final value in a CCV (corresponding to ∼90 nm diameter) through multiple functions of its AH0. Taken together with the observation that CALM is highly abundant in CCVs (30%–35% of the clathrin adaptors), these data allow us to propose that the AH0 of CALM controls the efficiency of endocytic CCP/CCV maturation and size by both sensing and driving membrane deformation.

13To confirm this hypothesis, we examined the effect of deleting AH0 or mutating AH0 on the ability of CALM to influence CCP/CCV properties in vivo by expressing endogenous levels of full-length, siRNA-resistant CALM(ΔH0) or CALM(H0mut) following endogenous CALM depletion (Figure S1B). Immunofluorescence indicated no obvious reduction in the number of punctate endocytic CCPs/CCVs (Figure S5A). However, despite CALM(ΔH0) or CALM(H0mut) being present in CCPs/CCVs and able to mediate normal trafficking of VAMP8 (Figure S5A), EM analysis showed that CCP/CCV morphology and the ratio of open to necked/closed structures resembled that of cells depleted of CALM (Figures 1A, 5, S1C, and S1D). That is, expression of CALM(ΔH0) or CALM(H0mut) failed to rescue the enlarged size phenotype and the normal ratio of open versus necked/closed clathrin-coated structures.

14In light of the major changes in CCP:CCV ratio, the time taken for CCP maturation and CCP/CCV size, we investigated the effect of CALM depletion on receptor/ligand internalization rates by measuring the steady state uptake rates at 37°C of radio-iodinated Tf and EGF. As shown in Figures 6A and 6B, CALM depletion caused ∼50% reduction of both Tf and EGF uptake. This endocytosis defect was restored to normal in cells expressing siRNA-resistant CALM-WT, but not rescued in cells expressing CALM(ΔH0). Immunofluorescence correspondingly showed intracellular accumulation of Tf and EGF in control cells after 6 min endocytosis, whereas both ligands remain in punctae associated with the plasma membrane in cells depleted of endogenous CALM (Figures 6C and 6D). We thus conclude that the absence of CALM that possesses a functional AH0 delays CCV maturation (Figures 2A and 2B), resulting in a significant endocytic defect.

15CALM and AP2 depletion both caused a major reduction in Tf and in EGF internalization rates (Figures 6A and 6B; see also ). EM analysis however, showed in marked contrast to CALM depletion, AP2 depletion (of μ2) resulted in an ∼12-fold decrease in the number of endocytic CCPs/CCVs (see also ) and a striking ∼2-fold reduction in the diameter of the remaining clathrin-coated endocytic structures (note, this reduction in diameter is equivalent to a reduction in volume by ∼90%) (Figures 1B, 7A, S1C–S1E). These AP2 depletion phenotypes were completely rescued by restoring the AP2 levels with re-expression of siRNA-resistant μ2 (Figure 7A). The clathrin adaptor complement of AP2-depleted endocytic structures should be dominated by CALM (the major PtdIns4,5P2-binding endocytic adaptor remaining), suggesting that increasing amounts of CALM-WT in CCPs/CCVs results in them being very small (Figures 1B, 7A, and S1C). Unfortunately, we have found it impossible to accurately quantify the relative amount of CALM and clathrin in AP2-depleted cells because the number and size of endocytic CCPs/CCVs was so drastically reduced. In support of our proposal that the absence one of the two major clathrin adaptors causes the other one to dominate, quantification of STED images showed that depletion of CALM leads to a 2- to 3-fold increase in the AP2:clathrin ratio in clathrin-coated endocytic structures (data not shown).

16In agreement with the proposal that CALM drives increasing membrane curvature, we also find that increased CALM expression (that is, endogenous CALM levels combined with similar levels of expression of siRNA-resistant CALM-WT) reduced CCP/CCV size and caused an increase in the number of necked/closed structures (Figures 7B, S1C, and S1D), whereas a combination of endogenous CALM and expression of CALM(ΔH0) expression did not (Figures 7B, S1C, and S1D). Unfortunately, it was not possible to create clonal cell lines stably expressing any higher levels of CALM (probably due to toxicity through inhibition of CME; ) or to overexpress AP2 (which would require simultaneous overexpression of all four subunits). In conclusion, we propose that in vivo despite both being important for clathrin-mediated endocytosis, AP2 and CALM have antagonistic effects on CCP/CCV size and the ratio of open versus necked/closed structures, hence their relative levels are a major determinant in defining various CCP/CCV properties.

17The initiation of CCP formation is generally believed to most often involve a pioneer complex of AP2/FCHO/Eps15 being recruited to a region of elevated PtdIns4,5P2, which is then stabilized on the membrane by AP2 cargo being drawn in and clathrin recruited and polymerized. Addition of more clathrin adaptors, including CALM, and of clathrin itself drives maturation of the CCP. Because CALM preferentially binds to curved membranes and CCP shape is a dynamic/equilibrium system, an increased amount of CALM in a CCP would tend to shift the net steady state of the structure toward increased membrane curvature; that is, CALM effectively drives membrane curvature. The increasing CCP curvature will result in CALM having an increased affinity for that region of the membrane and thus more CALM will be recruited to it i.e., a positive amplification cascade is established. Positive membrane curvature will increase until the CCP can be readily necked and then undergoes scission. The necked structure would likely resemble a single pearl such as those we see spontaneously produced on PtdIns4,5P2-containing liposomes. The increase in the number of open CCPs and concomitant enlargement both resulting from the absence of fully functional CALM suggests a defect in the formation of late stage necked, bulbous CCPs. This defect would most likely be caused by the increased difficulty in necking a larger open CCP. Hence, it will take longer to neck/close the resulting larger CCP under these, suggesting that in the absence of CALM, CCP necking/scission tends toward being rate limiting. This proposal is supported by the observation that depletion of CALM caused a doubling of the time taken to reach a point at which a neck is formed that is small enough to allow vesicle scission to occur. That the AH0 of CALM plays a principle role in controlling of CCP/CCV size, curvature, and maturation can be inferred from the observation that neither expression of CALM(ΔH0) nor of CALM(H0mut) rescued the enlarged CCP/CCV phenotype, the CCV formation efficiency and the reduced endocytic rate. Intriguingly a similar increase in CCP size is seen when AP2 is replaced by AP2 lacking its α-appendage, the domain responsible for concentrating many CME proteins, including CALM, in CCPs. These data suggest that a failure to recruit enough CALM into these AP2 positive CCPs could be the explanation for this phenotype.

18Why should the depletion of CALM or AP2 have opposite effects on CCP/CCV morphology? Why should the presence of larger CCPs/CCVs in CALM depleted cells with little reduction in the number of CCPs/CCVs reduce overall endocytic rate? We believe that both issues are linked to CALM’s and AP2′s different structural/biophysical properties and the type of cargo they recognize. First, CALM possesses a curvature sensing/driving AH0 whereas AP2, as far as we know, does not. Second, the transmembrane cargoes bound by AP2 often have large extracellular portions (including TfR, EGFR) that, upon collision, will generate lateral pressure causing membrane curvature in the opposite direction to that of inward budding favored by adaptor-mediated molecular crowding on the membrane’s cytoplasmic face. This is analogous to a model put forward to explain the effect of GPI-linked cargo in COPII-coated vesicle formation. In a normal CCV, CALM will oppose this effect and so drive CCV closure. However, in endocytic CCPs where CALM has been depleted and therefore AP2 is predominant, there will be a need for a reduction in large cargo molecule density to reduce cargo collisions and so allow CCP closure. This can be achieved if no more cargo packed into these larger CCPs than is packed into ordinary sized CCPs. That this occurs is suggested by the similar and even slightly reduced intensity of TfR staining of the diffraction limited spots at the surface of control and CALM-depleted cells shown in Figure 6. However, it should also be noted that if the situation is such that in control cells most TfR is in internal compartments and any TfR delivered to the plasma membrane is immediately incorporated into CCPs as suggested, then larger CCPs resulting from CALM depletion cannot take up any more TfR than CCPs in control cells because no more TfR can be available to be taken up. In this case, endocytic rate will be determined by the rate at which the CCPs mature.

19In contrast, the VAMP cargo of CALM has a negligible extracellular/luminal portion that does not oppose the positive membrane curvature driven by adaptor molecular crowding and the effect of CALM AH0 insertion. Thus, when CALM is the predominant adaptor (i.e., AP2 is depleted), the result is greatly reduced CCV size exacerbated by the fact that there is no large luminal domain cargo (AP2 binding) actively sorted into these structures to oppose CCP positive membrane curvature.

20As CCP curvature approaches its final value and the bud neck forms, to avoid steric clashing/molecular crowding on the luminal side of the membrane, large YxxΦ and ExxxLL motif containing cargoes and their bound AP2 will tend to be excluded from this area and concentrate in the crown of the bulb. This would explain why AP2 levels appear to decline prior to scission when analyzed by TIRF microscopy. As a result of its sensing and driving positive membrane curvature and it not binding cargo that is excluded from the neck region of a CCP, CALM should continue to be recruited to the curved CCP membrane and should also not be partitioned into the crown. Hence, CALM’s levels appear to continue to increase until scission. CALM may actually also aid in removing AP2 from near the neck as in the absence of AP2 cargo that has been excluded for steric reasons but with the positive curvature preferred by CALM, CALM should be able to more effectively compete with AP2 for binding of PtdIns4,5P2 in this region.

21Because an alteration in the clathrin adaptor complement alters the size and curvature of a CCP/CCV, we conclude that the presence of clathrin adaptors with the ability to drive membrane curvature by both insertion/wedging and molecular crowding plays a major role in defining CCP/CCV morphology. However, we cannot say from our studies what the contributions of membrane composition and the clathrin lattice are in driving and/or stabilizing CCP/CCV membrane curvature.

22The coats of other transport vesicles contain and are attached to the membrane via small GTPases: Sar1:GTP (in COPII vesicles) or Arf1:GTP (in other CCVs and COPI vesicles), both of which possess AHs that display MCS, drive membrane curvature and can reduce membrane rigidity, thus facilitating membrane deformation. The only possible GTPase or Arf candidate in endocytic CCPs/CCVs is Arf6, but this is not found in significant amounts and depletion of Arf6 only marginally affects endocytosis. Hence, we propose that in endocytic CCPs, the AH0 of the abundant adaptor CALM assumes the role of a curvature sensor/driver similar to the AHs of small GTPases in other transport vesicles.

23If one considers that nonsense mutations in CALM cause severe anemia and reduced life-span in mice, that complete deletion of the gene additionally causes other severe phenotypes resulting in >90% pre-weaning mortality, that CALM is implicated in alterations in cognitive function with increasing age, and finally that mutations in CALM are a risk factor for Alzheimer disease, the apparent inability of CALM depletion to affect endocytic rate in cultured cells previously reported in several studies was difficult to understand. However, by measuring the more physiological steady-state endocytic uptake rate at 37°C rather than using a 4°C block, we show that in fact CALM depletion does cause a marked reduction in endocytic rate of both EGF and transferrin. This is in at least partial agreement with a very recent study showing that a conditional CALM knockout specifically caused reduced transferrin uptake in mouse erythroblasts, although the authors suggest that CALM may be involved in a specialized endocytic mechanism in these cells.

24In conclusion, we have demonstrated that the N-terminal 18 residues of CALM form a previously unsuspected amphipathic helix (AH0) and that this structural element plays a central role in the ability of CALM to strongly influence CCP/CCV size, formation, rate of maturation, and thus endocytic uptake. We propose that it is the disruption of these properties in combination with effects on CALM’s ability to drive the internalization of small R-SNAREs that explains the pathophysiological effects caused by mutations in the CALM gene.

25For additional information, including protein chemistry, reagent preparation, and constructs used, see the Supplemental Experimental Procedures.

26HeLaM-VAMP8-HA cells were selected under 0.5 mg/ml G418 and additionally transduced with myc-tagged, siRNA-resistant WT or mutant versions of CALM for the add-back experiments, using the retrovirus Phoenix system. The derived cells were selected with 0.2 mg/ml hygromycin B and 0.5 mg/ml G418. The highest level of myc-CALM expression was similar to the levels of endogenous CALM. Endogenous CALM was depleted with siRNA ACAGTTGGCAGACAGTTTA at 20 nM for 72 hr. Endogenous AP2-μ2 was depleted with siRNA AGTGGATGCCTTTCGGGGTCA at 20 nM for 96 hr.

27Indirect immunofluorescence was performed after paraformaldehyde fixation and permeabilization with 0.1% Triton X-100 or 0.5% saponin. For imaging of transferrin endocytosis, the living cells were first serum-starved for 30 min, followed by exchange for medium supplemented with Alexa Fluor-conjugated transferrin (Invitrogen) for the indicated times. For TIRF microscopy, HEK293 cells were cultured and transfected using CALM-GFP and AP2 μ2-mCherry as detailed previously. Cells were plated on clean coverslips on the same day as transfection and imaged 48 hr later to allow several cell divisions and to give transfected cells time to adapt moderate protein overexpression. To determine CCP maturation time in CALM knockdown cells, HT1080 cells were transfected with either CALM or control siRNA followed by transfection with CALM-GFP and AP2 μ2-mCherry. The maturation time of CCPs was determined as described previously. For STED super resolution microscopy, samples were fixed and processed as described above. For microscopy, a Leica TCS SP8 gSTED equipped with a 100×/1.4 Oil STED Orange lens and a 592 nm depletion laser was used. Further details on STED imaging and quantification are summarized in the Supplemental Experimental Procedures.

28Binding of recombinant CALM ANTH to membranes was determined with a BIAcore 3000 SPR biosensor (GE Healthcare). Details of the method are described elsewhere and in the Supplemental Experimental Procedures.

29Liposomes for single liposome curvature assay were prepared using a previously described lipid hydration method. The size of liposomes was calibrated by combining confocal fluorescence microscopy and dynamic light-scattering measurement. Single isolated liposomes of different diameters labeled by DOPE-Atto 655 were immobilized on a Neutravidin functionalized microscope glass surface. A total of 500 nM CALM ANTH-WT and CALM ANTH(ΔH0) labeled with Alexa 488 were allowed to bind the surface attached liposomes from solution. The fluorescently labeled protein and lipid were sequentially imaged, and the integrated intensities extracted and used to quantify the density of bound protein and the size of the individual liposomes.

30All experiments were performed in a 10 mM Tris (pH 7.4), 170 mM NaCl, 0.05 mM TCEP buffer. For all samples, liposomes were imaged before the addition of protein to ensure that the binding molecule did not affect liposome morphology or fluorescence. Images were analyzed as described elsewhere. We have previously shown that recruitment by membrane curvature follows a power law and quantified the recruitment potency as the slope of a straight-line fit to a log-log representation of protein density versus liposome diameter. In this representation of the data, a more negative slope represents a stronger recruitment by membrane curvature. Technical details are explained in full in the Supplemental Experimental Procedures.

31Brain polar lipid extract (BPLE) and PtdIns4,5P2 (Avanti Polar Lipids) were hydrated in chloroform:methanol (4:1) supplemented with HCl; 0.5% PtdIns4,5P2 BPLE was dried down with argon and rehydrated in 20 mM Tris (pH 7.4), 170 mM NaCl. The solution was extruded through a 200 nm polycarbonate membrane (Nuclepore) 21 times; 10 μl of 1 mg/ml 200 nm liposomes was used in a 100 μl reaction. Control samples contained no protein. CALM proteins (0.5 μM) were incubated for 1 min at 20°C. A formvar carbon-coated/glow-discharged EM grid was immediately placed on the surface of a 50:50 mix of the liposome protein solution and 9:1 2% methyl cellulose:2% uranyl acetate (to prevent air-drying artifacts of delicate structures as routinely used in the preparation of ultrathin frozen sections for immunolabelling; ), incubated for 1 min, removed, wicked, and dried overnight. The grids were viewed (TEM, Phillips) and images were captured (Megaview III; Olympus). A new batch of liposomes and protein was used for six repeated experiments, resulting in final counts of several hundred liposomes for each condition. Liposome tubulation was categorized, counted, and plotted (Excel, Microsoft).

32Cells were fixed with 2.5% glutaraldehyde/2% paraformaldehyde, post-fixed with 1% osmium tetroxide, dehydrated, and embedded in epoxy resin. Randomly cut and orientated sections were stained and CCPs were recorded with a charge-coupled device (CCD) camera (Megaview III; Olympus) on a CM100 transmission electron microscope (TEM; Philips). One hundred profiles were traced with a Wacom Bamboo tablet and overlaid using Adobe Illustrator CS6.

33Cells were fixed with 0.1% glutaraldehyde / 4% paraformaldehyde, infused with 1.7 M sucrose/15% polyvinyl pyrolidone, and frozen in liquid nitrogen. Sections were cut in a cryo-ultramicrotome (Ultracut UCT/EM FCS; Leica), collected with 1% methyl cellulose/1.15 M sucrose, and immunolabelled with mouse anti-myc (Upstate) followed by rabbit anti-mouse (DAKO) and protein A: 15 nm colloidal gold (Utrecht University, The Netherlands). Images were captured using a CCD camera (Eagle 4K) on a Tecnai G2 Spirit BioTWIN TEM (FEI).

34The internalization of iodinated transferrin (Tf) and EGF was determined according to and. In brief, triplicates of cells were grown on 12-well dishes and incubated in either 2 μg/ml 125I-labeled Tf or 1 ng/ml 125I-labeled EGF at 37°C for 2, 4, and 6 minutes. Subsequently, the cells were placed on ice and washed. Thereafter, the cells were incubated twice in 0.2 M acetic acid (pH 2.6), 0.5 M NaCl for 5 min to remove the surface-bound ligand. Finally, the cells were solubilized and collected in 1 M NaOH. The acid-releasable radioactivity represented surface-bound ligands, whereas the acid-resistant fraction solubilized in NaOH represented the internalized fraction. The latter was plotted against time. The resulting linear regression coefficients represent the endocytic rate constants ke.