A Flat BAR Protein Promotes Actin Polymerization at the Base of Clathrin-Coated Pits

FCHSD2 Is a Bona Fide CME Protein

The interaction of FCHSD2 and Nwk with CME components is the primary indication of their participation in this process (O’Connor-Giles et al., 2008, Rodal et al., 2008). To investigate this matter further, we started by looking at the subcellular localization and the dynamics of FCHSD2 during CME. Endogenous FCHSD2 showed clear colocalization to clathrin punctae on the plasma membrane (Figure 1B). Likewise, exogenously expressed FCHSD2-Venus formed dynamic punctae colocalizing with the majority of mCherry-clathrin punctae (mCherry-clathrin light chain A) (80% ± 7% measured over the period of 10 s, n = 11 cells and Figure 1C). Live-cell imaging using a series of CME components revealed that FCHSD2 recruitment to CCPs occurred at a mid-to-late stage of pit maturation. FCHSD2 signal appeared on CCPs after clathrin and intersectin and peaked before the recruitment of dynamin and ARP3 (Figures 1D and S1A; Video S1). Further confirming the participation of FCHSD2 in CME, FCHSD2 knockout (KO) cells displayed reduced transferrin uptake when compared with control cells (Figures 1E and S1B). Interestingly, FCHSD2 KO cells compensate for slowed endocytosis by increasing the amount of total and surface transferrin receptor (Figure S1C). This compensation phenomenon has also been reported for cells depleted for PI3KC2α, epsin, or dynamin (Ferguson et al., 2009, Messa et al., 2014, Posor et al., 2013). Re-expression of FCHSD2 could partially rescue the transferrin uptake defect (Figures 1E and S1B). We attribute this partial rescue to variations in expression level within the population of rescued cells.

In several of our live cell experiments, we noticed the accumulation of FCHSD2 punctae on the peripheral regions of moving cells resembling adhesion sites. This led us to look at the relationship of FCHSD2 and integrins, a CME cargo associated with focal adhesions (Ezratty et al., 2009). In cells simultaneously expressing integrin β3, clathrinLC, and FCHSD2, we observed the presence of dynamic FCHSD2 and clathrin punctae on the majority of disassembling focal adhesions (Figure S2A; Video S2). Accordingly, FCHSD2 KO and knockdown (KD) cells showed significant impairment of active integrinß1 internalization using a microscopy-based antibody-feeding assay (Figure S2B) and by flow cytometry (Figure S2C). In agreement with the integrin uptake defect, FCHSD2 knockout led to a significant reduction of cell migration as measured with a wound-healing assay (Figure S2D).

To further investigate the role of FCHSD2 on CME, we tested the effect of FCHSD2 depletion on CME dynamics. For that, we used BSC1 cells stably expressing GFP-tagged σ2-adaptin (AP2σ2-GFP) (Ehrlich et al., 2004). As shown in Figure 1F, cells depleted for FCHSD2 (Figure S1D) displayed longer CCP lifetimes than control cells (44.9 s ± 21.7 s for FCHSD2 small hairpin RNA [shRNA] cells versus 32.7 s ± 16.3 s for control cells, mean ± SD). Morphological analysis of pits by electron microscopy (EM) revealed that FCHSD2 KD and KO resulted in an increased frequency of intermediate stage pits (U-shaped) at the expense of terminal pits (Ω-shaped) (Figure 1G). In agreement with the mid-to-late recruitment of FCHSD2 to CCPs (Figure 1D), we found no change on the frequency of early stage pits (shallow pits) in FCHSD2-depleted cells. Taken together, our results show that FCHSD2 is a bona fide CME component whose depletion leads to a kinetic defect in CME.

FCHSD2 Accounts for a Large Part of the Actin Contribution to CME

The interaction of FCHSD2 with N-WASP (Figure 1A) suggests a participation of this protein controlling actin polymerization during CME. To understand what the relative importance of FCHSD2 to the actin component of CME is, we decided to evaluate the CME phenotype in cells on which knockdown had been performed for the essential subunit of the ARP2/3 complex, ARP3 (Gournier et al., 2001). This strategy allowed us to look specifically at the effect of branched actin, the main type of actin structure formed during CME (Collins et al., 2011), without affecting linear actin structures and circumventing typical actin poisons that are very toxic and were shown to only partially affect the actin around CCPs (Collins et al., 2011). Cells with ARP3 KD (Figure S1E) showed a substantial reduction (∼43%) in transferrin uptake (Figure 1H) and displayed a large increase in CCP lifetimes with around a third of events lasting at least 180 s (the entire duration of the movies) (Figure 1F). Similar to the effect of FCHSD2 KO, ARP3 KD also led to an increase in surface transferrin receptor (increase of 37% ± 15% compared to wild-type cells, n = 5 independent experiments with >5,000 cells).

Next, we compared the transferrin uptake of cells with single and combined ARP3 and FCHSD2 depletion. In accordance with FCHSD2 and ARP2/3 acting on CME via the same pathway with ARP2/3 as a downstream factor, cells with combined ARP3 KD and FCHSD2 KO showed a similar reduction in transferrin uptake as cells with ARP3 KD alone (Figure 1H). An incomplete KD may contribute to the small difference in transferrin uptake between ARP3 KD in wild-type and in FCHSD2 KO cells (p value = 0.04). Importantly, with ARP2/3 as a downstream factor of FCHSD2, we could use the ARP2/3 contribution to CME as a baseline to infer that FCHSD2 is a major contributor to the ARP2/3 participation in CME. (Figure 1H).

FCHSD2 Is Recruited to CCPs by Intersectin via an SH3-SH3 Interaction

CCP maturation is characterized by a cascade of phosphoinositide conversions that work as timekeepers of the process by regulating the recruitment of individual CME components (He et al., 2017, Posor et al., 2013). To identify if FCHSD2 has a preference for any specific phosphoinositide, and could explain its recruitment to CCPs, we used a liposome-pelleting assay with liposomes of specific compositions. As shown in Figure 2A, FCHSD2 binds preferentially to PI(3,4)P₂ or PI(3,4,5)P₃ containing liposomes. PI(3,4)P₂ was recently shown to be a late CME lipid generated by PI3KC2α (Posor et al., 2013). In agreement with a potential role of PI(3,4)P₂ on FCHSD2 function, KD of PI3KC2α showed similar decrease in transferrin uptake as the combined KD of PI3KC2α in FCHSD2 KO cells (Figure S3A). However, KD of PI3KC2α had no significant effect on FCHSD2 recruitment to CCPs (Figure S3B), suggesting that phosphoinositides are not primarily responsible for the recruitment of FCHSD2 to CCPs but rather modulate its function after recruitment.

To investigate which FCHSD2 domains are required for its recruitment to CCPs, we expressed a series of truncation constructs in cells and evaluated their colocalization with clathrin punctae. Only constructs that included the SH3-2 (FCHSD2 full length and F2B12) localized to CCPs (Figure 2B). Expression of the SH3-2 alone also formed punctae that colocalized with clathrin, albeit to a lesser extent when compared to constructs that also contained the F-BAR domain (Figure 2B). In addition, expression of these truncation constructs revealed that, similarly to what has been described for Nwk (Kelley et al., 2015), both SH3 domains cooperate to autoinhibit FCHSD2. This autoinhibition avoids promiscuous membrane binding (Figure S3C, as judged by formation of membrane protrusions and membrane binding in cells) and is predicted to be relieved when the SH3 domains are occupied by their cognate binding proteins. Thus, our data show that FCHSD2 is a bona fide CME component whose recruitment to CCPs is dependent on its second SH3 domain.

Given the key role of FCHSD2 SH3-2 in the localization of FCHSD2 to clathrin punctae, we tested if intersectin, a binding partner of FCHSD2 SH3-2 (Figure 1A), is required for FCHSD2 recruitment to CCPs. We performed knockdown for both intersectin isoforms (ITSN1/ITSN2) and quantified the colocalization between clathrin punctae and FCHSD2. Accordingly, cells silenced for ITSN1/2 showed a significant reduction of clathrin punctae colocalizing with FCHSD2 when compared to control cells (Figures 2C, 2D, and S3D). To further understand the relationship between intersectin and FCHSD2, we fine-mapped this interaction. Previous work mapped the interaction between the second SH3 of Nwk to a C-terminal fragment of dap160 containing multiple SH3s (O’Connor-Giles et al., 2008, Rodal et al., 2008). Taking this as a starting point, we tested the capacity of each of the five SH3 domains of ITSN1 (SH3a-e) to interact with FCHSD2 SH3-2 using GST pull downs. As shown in Figure 3A, the fourth SH3 domain of ITSN1 (SH3d) could pull down FCHSD2 SH3-2. Isothermal titration calorimetry (ITC) revealed a 248 nM (±75 nM) affinity (Kd) with a ∼1:1 stoichiometry (Figure S4A). Thus, FCHSD2 is recruited to CCPs by intersectin via an SH3-SH3 interaction.

SH3 domains generally bind short proline-rich sequences rather than mediate domain-domain interactions (Kurochkina and Guha, 2013). To understand how this unusual SH3-SH3 interaction occurs, we solved the crystal structure of the FCHSD2 SH3-2/ITSN1 SH3d complex at 3.4 Å (Table S1). The structure revealed tandem binding between the SH3 domains with the canonical proline-binding interface of FCHSD2 SH3-2 interacting with the surface opposite to the proline-binding interface of ITSN1 SH3d (Figures 3B, S4B, and S4C). The domains form an extensive contact surface composed largely of hydrophobic residues and a few charge interactions (Figures 3C and 3D). To confirm the SH3-SH3 contacts as the bona fide biological interface, we performed GST pull downs using a series of mutants for both SH3 domains. Double mutants on FCHSD2 SH3-2 (Y576S+F607S or V622K+L623K) were required to abolish the interaction to ITSN1 SH3d (Figure 3E). On the other hand, single residue mutations on ITSN1 SH3d (I1078S, I1078K, R1119A, or R1119E) could abolish the interaction with FCHSD2 SH3-2 (Figure 3F). Additionally, mutation of residues located on crystal contacts (FCHSD2 SH3-2 N613 and D583; ITSN1 SH3d T1086) did not show any effect on binding (Figures 3E, 3F, and S4D).

Next, we set out to validate in cells the biological significance of the FCHSD2 SH3-2/ITSN1 SH3d structure. In contrast to its wild-type counterpart, expression of the FCHSD2 SH3-2 interface mutant (Y576S+F607S) failed to localize to clathrin punctae (Figure S4E). On the other hand, expression of wild-type ITSN1 SH3d, but not the interface mutant counterpart R1119E, resulted in a substantial reduction of FCHSD2 punctae (Figure S4F). To further confirm the role of the FCHSD2 SH3-2/ITSN1 SH3d crystal interface on the correct localization of FCHSD2, we used the transmembrane domain of the mitochondrial translocase complex subunit TOM20 to mistarget the ITSN1 SH3d to the mitochondria. In agreement with the interface from the crystal structure, wild-type ITSN1 SH3d could efficiently recruit FCHSD2 to the mitochondria while the interface mutant ITSN1 SH3d R1119E could not (Figures 3G and S4G). Finally, we tested the functional significance of the FCHSD2 SH3-2/ITSN1 SH3d interaction. Expression of wild-type ITSN1 SH3d acted as a dominant negative and could inhibit transferrin uptake while the interface mutant counterpart ITSN1 SH3d R1119E had a much smaller effect (Figure 3H). Altogether, our results establish FCHSD2 as a CME protein recruited by intersectin via an SH3-SH3 interaction, and the ITSN1 SH3d overexpression can be used to interfere with FCHSD2 function.

FCHSD2 Is a Strong Activator of Actin Polymerization in the Presence of Membranes

Another important feature of FCHSD2 is its relationship with the actin cytoskeleton via interaction of its first SH3 domain with the ARP2/3 complex activator N-WASP (O’Connor-Giles et al., 2008, Rodal et al., 2008) (Figure 1A). Activation of N-WASP via interaction with SH3 domains has been described for many proteins, including Nwk (Padrick and Rosen, 2010, Rodal et al., 2008). To test the capacity of FCHSD2 to activate N-WASP, we used a well-established in vitro assay (Doolittle et al., 2013) in which the kinetics of actin polymerization can be followed by monitoring the increase in the fluorescence of pyrene-labeled actin. For all reactions, we used a minimal set of components (actin, Arp2/3, and full-length N-WASP) where we could test the effect of various purified FCHSD2 fragments. Without any activator, this minimal set of components represents a baseline of actin polymerization where N-WASP is largely inactive and the majority of the signal comes from spontaneous actin nucleation (Figure S5A).

In reactions containing only the minimal components, none of the tested FCHSD2 fragments showed a significant increase of actin polymerization when compared to the control reactions (Figure 4A). In contrast, when liposomes where added to the reaction, FCHSD2 fragments containing both the BAR domain and the SH3-1 (F2B1 and F2B12) activated actin polymerization to the same level as the positive control N-WASP-VCA (the uninhibited fragment of N-WASP (Rohatgi et al., 1999) (Figure 4A). We further confirmed the requirement of the FCHSD2 SH3-1 domain for the activation of N-WASP by mutating the canonical proline-binding site of the SH3-1 domain. These mutations (Y478A/Y480A or P521A/Y524A) rendered F2B12 incapable of increasing actin polymerization while mutants for the proline-binding site of FCHSD2 SH3-2 (Y576S/F607S or V622K/L623K) still maintained full activation capacity (Figures 4B and S5B). Titration of FCHSD2 demonstrated that it is capable of activating actin polymerization even at substoichiometric concentrations (half-maximal activity ∼6.3 nM in reactions containing 50 nM N-WASP) (Figure S5C). FCHSD2 SH3-1 alone could not significantly increase actin polymerization even at high concentrations (e.g., 1 μM) (Figure S5D), suggesting an active role of the interaction of FCHSD2 with the membrane on FCHSD2-dependent activation of N-WASP.

There are two levels of N-WASP regulation: one level is where the intramolecular autoinhibition of N-WASP is released by the action of allosteric activators such as PI(4,5)P₂ and cdc42 (Rohatgi et al., 1999) and another level where oligomerization of N-WASP by multivalent interactors greatly increases its affinity for ARP2/3 complex (Padrick et al., 2008). The requirement of liposomes for efficient activation of N-WASP by FCHSD2 together with the fact that FCHSD2 is a dimer (Figure S5E), and therefore can bind two N-WASP molecules, suggests that FCHSD2 may act on these two regulatory levels of N-WASP activation. To test this possibility we used conditions that allowed us to differentiate the contribution of FCHSD2 on each of these levels. As PI(3,4)P₂ is not an allosteric activator of N-WASP (Ma et al., 1998), we could test the effect of FCHSD2-mediated N-WASP oligomerization on membranes (using liposomes containing PI(3,4)P₂) and compare it to conditions where oligomerization and recruitment to an allosteric activator would occur (using liposomes containing PI(3,4)P₂ and PI(4,5)P₂). As shown in Figure 4C, reactions with liposomes containing both PI(3,4)P₂ and PI(4,5)P₂ lipids showed maximum actin polymerization while reactions with PI(3,4)P₂ liposomes showed an intermediate level of activation (Figures 4C and S5F). On the other hand, reactions containing PI(4,5)P₂ liposomes showed no increase in actin polymerization (Figures 4C and S5F). These results show a contribution of both oligomerization and allosteric activation in FCHSD2-mediated N-WASP activation.

Our results in cells revealed FCHSD2 as a major activator of ARP2/3-dependent actin polymerization during CME (Figure 1H). To test this result in vitro we compared the actin polymerization activity of FCHSD2 with that of SNX9 and CIP4. These two F-BAR domain proteins are proposed to activate N-WASP during CME via direct activation of N-WASP (SNX9) (Yarar et al., 2007) or indirectly via cdc42 (CIP4) (Fricke et al., 2009). As shown in Figure 4D, FCHSD2 displayed much higher actin activation activity than SNX9 with 90nM FCHSD2 still showing higher activation capacity than 560 nM SNX9. As expected, in the absence of cdc42, CIP4 failed to activate actin polymerization (Figure S5G). In the presence of GTPγS-loaded cdc42, CIP4 showed strong actin activation activity, however, still to a lower degree than FCHSD2 (Figure 4E).

Next, we decided to test the connection between both of the features of FCHSD2: recruitment to CCPs and activation of actin polymerization. As N-WASP activation leads to the recruitment of the ARP2/3 complex to an existing actin filament, we used the accumulation of ARP3 at dynamin spots as a reporter for N-WASP activation. Confirming that FCHSD2 activates N-WASP at CCPs, FCHSD2 KO cells showed a significant reduction of ARP3 punctae colocalizing with dynamin in a live cell imaging experiment (Figure S5H). Taken together, these results show that FCHSD2 is a major activator of actin polymerization via N-WASP at CME sites.

FCHSD2 Localizes to the Planar Membrane around CCPs

The actin cytoskeleton forms a distinctive structure at CME sites (Collins et al., 2011). To understand how FCHSD2 contributes to the formation of this structure, we examined where on CCPs FCHSD2 was localized. We started by probing the relative Z-displacement of FCHSD2 during CCP lifetime. For this, we compared the fluorescence profiles of FCHSD2, AP2 (AP2σ2), and dynamin1 captured sequentially in total internal reflection fluorescence (TIRF) and widefield (WF). Using this imaging paradigm, proteins located in budding clathrin vesicles lose their signal on the TIRF channel (as the evanescent field is only ∼100 nm in depth) while maintaining it for a few more seconds on the WF channel. In contrast, if a protein stays on the plasma membrane during the whole lifetime of the CCP, the fluorescence decays at the same time for both TIRF and WF channels (Figure 5A). As expected for their established roles in CME, AP2 profiles showed a delayed fluorescence decay on the WF channel (i.e., AP2 buds off with the clathrin-coated vesicle), while dynamin fluorescence decayed at the same time on both channels (i.e., dynamin stays on the plasma membrane) (Figure 5B). FCHSD2 signal decayed at the same time in both TIRF and WF channels (Figure 5B), suggesting FCHSD2 is not located on or around the clathrin cage, but rather stays close to the plasma membrane during CME.

To further address the localization of FCHSD2 on CCPs, we imaged FCHSD2 and AP2 with stimulated emission depletion (STED) super-resolution microscopy. While AP2 showed a complete colocalization with clathrin, FCHSD2 formed multiple punctae arranged in a semi-circle (sometimes a circle) around CCPs (Figures 5C and 5D). To further confirm the position of FCHSD2 on CCPs, we used 3D STED, which allows increased resolution imaging at the Z plane (albeit at the expense of lower resolution at XY). The orthogonal (XZ) view of FCHSD2 and clathrin by 3D STED confirmed that FCHSD2 is localized around and at the plasma membrane side of CCPs (Figure 5E). In agreement with FCHSD2 being localized out of the boundaries of the budding CCP, FCHSD2 was not present in purified CCVs (Figure S6A).

F-BAR domains are elongated, slightly curved, dimeric membrane-binding domains. The intrinsic shape of these domains determines the curvature of the membrane they are bound to (Qualmann et al., 2011). As such, expression of most F-BAR domains generates inward tubules in cells (Boucrot et al., 2012, Frost et al., 2008). In contrast, expression of FCHSD2 (and Nwk) F-BAR domain generates large numbers of filopodia-like cellular protrusions (Figure S6B) (Kelley et al., 2015). Interestingly, we observed that overexpressed FCHSD1/2 F-BAR domains coat the whole cellular membrane and not only the protrusions it induces (Figure S6B, inset). The data we present here shows that FCHSD2 is concentrated on a region of the membrane that is primarily flat, suggesting that FCHSD2 F-BAR is either distinctly shaped or has a distinct mode of binding to membranes. To investigate this atypical behavior, we started by checking the curvature preference of FCHSD2 in vitro. For that, we adapted the nanoparticle tracking analysis (NTA) system (Dragovic et al., 2011) to determine curvature preferences of protein domains to liposomes using low protein concentrations and avoiding curvature generation artifacts (A.C., L.A.-S., and H.T.M., unpublished data). This method is able to distinguish to which liposome sizes a particular protein binds within a population of heterogeneously sized liposomes. In agreement with its cellular localization, NTA shows that FCHSD2 binds preferentially to larger liposomes (Figure 5F).

To understand the molecular basis of FCHSD2 curvature preference, we used single particle CryoEM to determine the structure of the FCHSD2 F-BAR domain + SH3-1 domain (F2B1) (Figures S6C and S6D). We obtained a 9.2 Å reconstruction (Figure S6E) that reveals a largely flat F-BAR with densities on its tips compatible with the SH3-1 domains (Figures 5G and 5H). Furthermore, our FCHSD2 F-BAR map shows similarities with the crystal structure of the FES F-BAR (PDB: 4DYL, no associated publication), which is also a flat F-BAR domain (Figure S6F). Altogether, our results show that FCHSD2 stays on the surrounding plasma membrane during CME and binds to membranes via its atypical, flat F-BAR domain.

CME and Its Multiple Actin Regulators

The functional connection between the actin cytoskeleton and CME has been a matter of controversy for more than a decade. While this relationship is well established in yeast (Merrifield and Kaksonen, 2014), its importance in mammalian systems seems to be context- or cell-dependent (Boulant et al., 2011, Fujimoto et al., 2000). Nonetheless, a growing body of evidence supports the idea that actin is an intrinsic component of mammalian CCPs (Ferguson et al., 2009, Grassart et al., 2014) and modulates the effectiveness of CCP budding. Moreover, the intimate connection between CCPs and the actin cytoskeleton is further reinforced by a large number of CME proteins showing actin-related activities such as HIP1/HIP1R, the long isoforms of ITSN1/2 and notably, multiple F-BAR proteins (Daumke et al., 2014, McMahon and Boucrot, 2011, Qualmann et al., 2011). In which conditions these proteins are required, what their relative contribution is, and how they are organized to optimize the mechanical action of the actin cytoskeleton during CME remains unclear.

The unique position of FCHSD2 at the planar membrane region surrounding CCPs suggests a possible explanation for the apparent redundancy of actin regulators during CME. By coordinating the position and the timing of each actin regulator, cells would be able to change the strength and direction of actin mechanical forces during multiple stages of vesicle maturation. It is reasonable to assume that the multiple geometries acquired by the clathrin coat and the underlying budding membrane as CCPs mature would benefit from distinct force vectors. This possibility is supported by a recent work showing that, at least in yeast, different types of actin network are required during different stages of endocytosis (Picco et al., 2018). In addition to the distribution of force vectors during CME, the contribution of each individual actin regulator may also be tissue- and cargo-dependent or be involved in the transport of vesicles post scission. Future efforts to access the spatiotemporal organization of actin regulators in multiple cells during CME would be necessary to test these hypotheses.

A Model for FCHSD2 Regulation and Function

In this study, we moved away from using actin polymerization inhibitors because of their toxic effects in cells. Using the inactivation of the ARP2/3 complex, we could obtain a consistent baseline for the contribution of branched actin polymerization during CME and determine that FCHSD2 is responsible for a large part of this contribution (Figure 1H). We cannot rule out that ARP3 KD can have indirect effects on CME by affecting, for example, membrane tension. Nonetheless, if these side effects occured, they would imply an even higher importance of FCHSD2 to the ARP2/3 contribution to CME. The relative importance of FCHSD2 to CME was further reinforced by the reduced recruitment of ARP3 to CCPs in FCHSD2 KO cells (Figure S5H), and a similar contribution measured by transferrin uptake. In addition, FCHSD2 activated actin polymerization via N-WASP in vitro more efficiently than the F-BAR proteins we tested (Figures 4 and S5).

The timing of FCHSD2 recruitment combined with the phenotypes of FCHSD2 KO cells (increased CCP lifetimes and the accumulation of U-shaped pits) show that FCHSD2 has a kinetic role at the intermediate stage of CCP maturation, when the transition from a shallow invagination to a constricted pit ready for scission occurs. During this transition, the actin cytoskeleton at CCPs evolves into a cone-shaped, densely branched network with its origin at the flat region around CCPs (Collins et al., 2011). The position and function of FCHSD2 strongly suggests that this protein is involved in the generation of this actin structure.

Our data suggest that recruitment and activation of FCHSD2 are stepwise events. While the presence of FCHSD2 in CCPs is dependent exclusively on intersectin, FCHSD2 activation seems to be dependent on the phosphoinositide PI(3,4)P₂. Further supporting a stepwise cascade of events, both SH3 domains participate in FCHSD2 autoinhibition (Figure S3C). We believe that this uncoupling allows cells to spatiotemporally control FCHSD2 activation. A recent work elegantly showed that intersectin migrates to the edge of the clathrin coat as CCPs mature (Sochacki et al., 2017). Therefore, we hypothesize that FCHSD2 is recruited by intersectin and is kept in an inhibited/low activity state until it reaches the edge of the CCP, where the accumulation of PI(3,4)P₂ recruits FCHSD2 to the flat membrane triggering actin polymerization.

Taking all this information into account, we propose the following model for FCHSD2 function (Figure 6). After CME initiation, FCHSD2 is recruited to CCPs by intersectin via an SH3-SH3 interaction and intersectin-FCHSD2 complexes accumulate at the edge of CCPs. The flat FCHSD2 F-BAR domain and presence of the late phosphoinositide PI(3,4)P₂ drives the binding of FCHSD2 to the planar region apposed to CCPs and allows it to activate actin polymerization. The FCHSD2-dependent actin structure around CCPs helps the efficient invagination of the maturing pit. Finally, other CME players (both actin and actin-independent) take over and ensure the completion of the process.

The Rise of Non-canonical F-BARs

The BAR superfamily is characterized by the presence of an elongated, dimeric membrane-binding domain (Qualmann et al., 2011). The superfamily is further subdivided into N-BAR, F-BAR, and I-BAR families. While this division is primarily based on the homology between these domains, the initial crystal structures described suggested that this classification also implied a shape of the molecule and by consequence its curvature preference (Frost et al., 2008, Henne et al., 2007, Qualmann et al., 2011, Rao and Haucke, 2011). While for N-BAR and I-BAR proteins this correlation still holds true, the F-BAR family has shown to be far more flexible regarding its curvature preference. The first non-canonical F-BAR characterized were the SRGAP proteins that were shown to generate large filopodia when expressed in cells (Coutinho-Budd et al., 2012). The crystal structure of SRGAP F-BAR domain revealed an inverted F-BAR, with its lipid binding sites situated in the convex rather than the concave surface (Sporny et al., 2017). Similarly, the structure of the FES F-BAR domain revealed a flat BAR domain (PDB: 4DYL, no associated publication). The CryoEM map we present here adds to the list of unusual F-BARs and supports the idea of the F-BAR domain as a module used by evolution to recognize multiple curvatures.

Why would cells need a BAR domain to recognize a flat membrane? The FCHSD2 F-BAR is a large domain (467 aa), and one could argue that a small globular domain, such as a PH domain, could do the same job. We reason that a flat BAR domain with multiple binding sites scattered over its length provides a structural framework to ensure that FCHSD2 binds only to the planar membrane around the CCPs rather than competing for other phosphoinositol lipids on the curved regions of the budding membrane. Small globular membrane binding domains are exquisite at recognizing specific lipids. However, with a footprint of only a few lipid headgroups, small domains on their own are possibly unable to read the long-range differences that distinguish a flat membrane from a curved one.

Beyond CME

Besides its CME role, intersectin has been described to participate in a series of other processes including other endocytic routes, exocytosis, and viral entry (Herrero-Garcia and O’Bryan, 2017). The intimate connection between FCHSD2 and intersectin and the affinity of FCHSD2 to the pivotal signaling phosphoinositol PI(3,4,5)P₂ opens the possibility that the combined actin remodeling capacity of these proteins are used in processes other than CME.

Conclusion

In conclusion, our work places FCHSD2 as a pivotal player regulating actin polymerization during CME. The unique position of FCHSD2 on the planar membrane around CCPs is, to our knowledge, the first description of a CME-specific mammalian protein localized on a region other than the clathrin coat or the neck of the CCPs. The mechanism we describe for FCHSD2 illustrates how cells link in space and time the formation of CCPs and the reorganization of the cortical actin cytoskeleton during mammalian endocytosis. These findings also provide a mechanistic framework that may be applicable to other membrane bending events involving the actin cytoskeleton.

Key Resources Table

Contact for Reagent and Resource Sharing

Further information and requests for reagents should be directed to and will be fulfilled by the Lead Contact, Leonardo Almeida-Souza (lalmeida@mrc-lmb.cam.ac.uk).

Experimental Model and Subject Details

Methods Details

Quantification and Statistical Analysis

Statistical analysis was performed with Prism 6.0. Student’s t test was used for pairwise comparisons. Tukey’s multiple comparisons test was used to determine significant differences between multiple samples. Data are presented as mean ± SD or SEM as indicated in the figure legends. Significance levels used were as follows: ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001.

Data and Software Availability

The accession number for the FCHSD2 SH3-2/ITSN1 SH3d crystal structure reported in this paper is PDB: 6GBU. The accession number for the cryo-EM map for the FCHSD2 F-BAR reported in this paper is EMDB: EMD-4371.