Endophilin-As: Their roles in cell membrane behavior and disease

At the C-terminus, endophilins have an SH3 domain (60 residues) that forms a β-barrel with a hydrophobic groove for binding of proline-rich domain (PRD)-containing proteins (Fig. 1A, B) [16, 17]. Key endophilin-SH3 binding partners include the large GTPase dynamin, the phosphoinositide phosphatase synaptojanin [18 –20], and the cytoskeleton regulator N-WASP [21, 22]. A flexible linker (37–58 amino acids), less conserved and less characterized, connects the N-BAR and SH3 domains (Fig. 1A-C). Chimeric EndoAs with swapped linkers reveal that this region influences EndoA endocytic functions [23]. The linker also harbors several post-translational modification (PTM) sites (Fig. 1C), as well as putative Ca²⁺-binding sites, further suggesting implications in regulation. In humans, EndoA1 is predominantly expressed in the brain, EndoA3 in the brain and testes, and EndoA2 is ubiquitously expressed [19, 24, 25]. Adding complexity, each EndoA has multiple splice variants (https://www.uniprot.org/↗).

Clathrin-mediated synaptic membrane retrieval occurs within 15–20 s [47, 48]. Importantly, neuronal EndoAs also mediate clathrin-independent ultrafast endocytosis (UFE), where synaptic membrane retrieval takes 50–100 ms only (Fig. 2, Table 2) [49 –51]. UFE shares mechanistic similarities with CM-SVE: it is primarily driven by EndoA1 and -A2 [52], and involves interactions with dynamin and synaptojanin. Specifically, EndoA1-2 in complex with synaptojanin and the dynamin splice variant Dyn1xA are required to constrict the neck of UFE pits, facilitating fission (Fig. 2) [52, 53]. Accordingly, EndoA1-2 depletion, as well as disruption of their interaction with Dyn1xA, results in stalled UFE pits with wider necks [52, 53]. Although UFE is clathrin-independent, internalized vesicles subsequently fuse with synaptic endosomes, from which clathrin-coated vesicles bud to regenerate synaptic vesicles [50, 54]. Like in CM-SVE, EndoA1-2, together with synaptojanin, are involved in clathrin uncoating of such newly-budded synaptic vesicles (Fig. 2) [52, 55]. In line with this, C. elegans EndoA mutants accumulate synaptic endosomes, reflecting altered breakdown into new synaptic vesicles [55, 56].

Interestingly, EndoAs in pre-synaptic neurons also operate in synaptic vesicle exocytosis (Fig. 2): docking of EndoA1 dimers at the surface of exocytic vesicles stimulates their fusion with the plasma membrane, possibly by inducing membrane remodeling. In contrast, EndoA1 binding to the vesicular glutamate transporter 1 (VGLUT1) on synaptic vesicles limits their exocytosis, likely by restricting EndoA1 availability for its stimulating docking [57]. Moreover, EndoA1 and VGLUT1 form complexes with the adaptor protein intersectin-1, which induces synaptic vesicle clustering and reduces spontaneous exocytosis in the absence of stimulation [58]. EndoA1 and intersectin-1 also cluster vesicles near neurotransmitter release sites to enable fast activity-induced, EndoA1-dependent replenishment of those sites for sustained neurotransmission [59]. In neurosecretory cells, EndoA1 and EndoA2 interactions with intersectin-1 further promote priming and fusion of exocytic vesicles [60]. Finally, EndoAs also support exocytosis-endocytosis coupling in auditory hair cells by stimulating pre-synaptic Ca²⁺ influx by the Cav1.3 channel and interacting with the Ca²⁺ sensor otoferlin [61]. Importantly, EndoA functions in pre-synaptic exocytosis remain to be further validated: contradictory findings in hippocampal cells indicate that the EndoA1/2/3:VGLUT1 interaction does not influence synaptic vesicle exocytosis, but rather stimulates activity-driven endocytosis, thereby contributing to VGLUT1 recycling during synaptic stimulation [62].

EndoAs also function in post-synaptic neurons, where EndoA2 and -A3 mediate AMPA receptor (AMPAR) endocytosis through interaction with the cytoskeleton-associated protein Arc (Fig. 2) [63, 64]. EndoA3 further stimulates AMPAR endocytosis by binding and activating the Arf6 guanine nucleotide exchange factors (GEF) BRAG2a [65]. Together, these findings underscore the versatility of EndoAs in modulating synaptic vesicle and AMPA receptor trafficking at synapses.

Non-neuronal EndoAs also play a role in CME, particularly during dynamin-mediated membrane fission. In 3T3 fibroblasts, exogenously-expressed EndoA2 and dynamin are co-recruited to CCPs for vesicle scission [29, 66]. In the same cell line, overexpression of the EndoA1 SH3 domain inhibits membrane fission during CME of the transferrin receptor, likely by disrupting SH3-mediated interactions [67]. Similarly, EndoA1-3 triple depletion or EndoA2 overexpression in human melanoma cells reduces or enhances dynamin recruitment to the plasma membrane, respectively [68]. In kidney fibroblasts, overexpressed EndoA3 further facilitates membrane fission by interacting with N-WASP and promoting actin polymerization [21]. Interestingly, EndoA1-3 also accumulate at CCPs prior to fission to support the formation of elongated tubular necks [28], suggesting additional roles in CCP maturation. EndoAs in CME are documented to coordinate internalization of cell surface receptors. For example, EndoA1-3 bind to the adaptor CIN85, forming a complex recruited by the ubiquitin ligase Cbl to activated EGFR and c-MET receptor tyrosine kinases (RTKs), initiating their endocytosis [69, 70]. In HEK293T cells, EndoA2 overexpression and its binding to the BPGAP1 RhoGAP also stimulates EGF-stimulated EGFR endocytosis [71]. The three EndoAs further promote ligand-stimulated EGFR internalization in 3T3 fibroblasts by recruiting the cytoskeletal regulator lamellipodin to CCPs to stimulate actin polymerization [72]. EndoA1 and -A2 similarly bind the Arf6 GEF EFA6, promoting Arf6 activation and, subsequently, CME of the transferrin receptor [73]. In conclusion, despite the lack of a consensus mechanism, EndoAs appear to form endocytic complexes with various proteins to stimulate CME of multiple receptors across various cell types. It is however important to note that some of these studies, though referring to CME, did not clearly demonstrate the clathrin dependency of their modalities. Nevertheless, EndoA functions in CME appear to be partially conserved throughout evolution: the Arabidopsis thaliana homologs SH3P1, −2, and −3 promote clathrin uncoating through interactions with the auxilin-like vesicle uncoating factor and the SAC9 phosphoinositide phosphatase [74 –77], and further regulate the trafficking of clathrin-coated vesicles via actin interactions [78].

While studies report EndoA functions at various stages of CME (including SVE), the dynamics and molecular drivers of their recruitment are still debated. EndoA1 has been proposed to be recruited to CCPs through direct interactions with intersectin-1 [79]. Additionally, dynamin depletion reduces EndoA recruitment to the plasma membrane, suggesting potential cooperative recruitment [68]. Adding controversy, several studies report that EndoAs may inhibit CME. For instance, EndoA3 overexpression in kidney fibroblasts impairs CME of transferrin and dopamine D2 receptors [23], while EndoA1 and -A3, by recruiting ataxin-2 to the plasma membrane, reduces EGFR CME [80]. Moreover, high levels of EndoA1, either on artificial membranes or in SK-MEL-2 cells, inhibits fission by inserting into dynamin helices [81, 82]. Other studies even suggest peripheral, dispensable roles for EndoAs in CME, as it was detected only in a subset of CCPs in NIH-3T3 cells [29] and EndoA1-3 triple knockdown in human melanoma cells does not affect transferrin uptake [68]. Collectively, these observations indicate complex, context-dependent functions of EndoAs in CME. Importantly, with most studies to date being carried out with ectopically expressed, tagged EndoAs or in in vitro systems, EndoA contributions to CME warrants further investigation in endogenous contexts.

Mechanistically, FEME is characterized by a priming step, in which EndoA2 is constantly pre-clustered in discrete plasma membrane patches, likely facilitating rapid endocytosis upon cargo detection [92]. Priming involves GTP-Cdc42 recruiting the F-BAR proteins FBP17 and CIP4, which then recruit SHIP1/2 phosphatases. SHIP1/2 convert PI(3,4,5)P₃ into PI(3,4)P₂, triggering lamellipodin recruitment, which ultimately recruits EndoA2. The formation of EndoA2 priming patches may additionally involve liquid–liquid phase separation [93]. In the absence of nearby cargo, Cdc42 is deactivated by RICH1, SH3BP1, and Oligophrenin GTPase-activating proteins (GAPs), leading to patch disassembly [92, 94]. In contrast, upon cargo recognition, EndoA2 patches transition into tubular FEME carriers. Activated receptors are recognized by EndoA2 predominantly at the leading edge, either via direct binding to PRD-containing cytosolic tails or indirectly through adaptor proteins. Initial pit bending is driven by increased local EndoA2 concentration upon receptor capture, and/or ligand-induced receptor crowding [89, 94]. In the case of pathogens, binding to and clustering of their lipid or protein receptor likely induces local membrane curvature [95, 96], further recognized by cytosolic EndoA2 [89]. Subsequently, the N-BAR domain protein Bin1 recruits dynein molecular motors onto nascent carriers which, by moving along microtubules, promotes elongation and fission [87, 89, 97, 98]. The application of such pulling forces on tubular membranes where EndoA2 scaffolding generates frictional resistance to lipid diffusion favors tubule neck squeezing and scission, this mechanism being called friction-driven scission (FDS) [89, 98]. In cells, optimal scission additionally requires actin polymerization, supported by EndoA2 interactions with the actin regulators VASP, Mena, and NHSL1/2, as well as dynamin [89, 99, 100]. Interestingly, EndoA2 foci colocalize with the scaffold protein Alix [101], but also with other BAR domain proteins (ASAP1, SNX9, Pacsin2, and srGAP1), further suggesting cooperative actions that remain to be investigated [92].

More recently, EndoA3 was identified to mediate endocytosis of ALCAM (Activated Leukocyte Cell Adhesion Molecule, also called CD166) [84], L1CAM (L1 Cell Adhesion Molecule, or CD171) [102], and ICAM1 (Intercellular Cell Adhesion Molecule 1, or CD54) (Fig. 3B, Table 2) [103]. EndoA3 colocalizes with these CAMs at the plasma membrane and is pulled down by the cytosolic tails of ALCAM and L1CAM [84, 102, 103], supporting its direct involvement in their endocytic uptake. All cargoes identified to date are immunoglobulin-like cell adhesion molecules (Ig-like CAMs) with heavily glycosylated extracellular portions, suggesting common EndoA3-dependent recognition patterns that remain to be elucidated. Interestingly, EndoA3-mediated CIE of ALCAM is stimulated by extracellular galectin-8, which likely clusters glycosylated cargoes and glycosphingolipids to initiate plasma membrane invagination [84], consistent with the glycolipid–lectin hypothesis [104]. In contrast, although galectin-1, −3, and −8 colocalize with L1CAM-positive endocytic structures, extracellular galectins inhibit L1CAM internalization, possibly due to the formation of galectin lattices trapping it at the membrane [102]. This suggests that galectins may differentially regulate EndoA3-mediated CIE, depending for example on cargo glycosylation patterns that influence galectin interaction affinities, or on local extracellular galectin concentrations. Inside the cell, EndoA3- and ALCAM-positive endocytic structures dynamically associate with the actin cytoskeleton and microtubules [105]. The current mechanistic model suggests that (i) Rac1-stimulated activity of the molecular motor myosin 2 A (and possibly of other Rac1 effectors) promotes actin remodeling and increases membrane tension to favor membrane deformations, and that (ii) kinesin motors pull nascent EndoA3-positive carriers along microtubules to favor tubule elongation and FDS [105]. L1CAM endocytosis also requires the F-BAR protein PSTPIP1 which, consistently with its larger intrinsic curvature radius, is recruited prior to EndoA3 [102], again supporting cooperative, sequential BAR domain protein actions during endocytosis. Following internalization, both ALCAM and ICAM1 undergo retromer-dependent retrograde transport and subsequent polarized redistribution to the plasma membrane [103], implying possible shared post-endocytic fates for EndoA3 cargoes.

Recent studies have expanded the functional repertoire of EndoA proteins beyond endocytosis, including in post-endocytic trafficking, as supported by their detection on endosomes [106]. For instance, EndoA2 binds to the microtubule-associated protein CRMP2, which facilitates the movement of AMPAR-positive endosomes along the cytoskeleton to enable receptor recycling [107]. Moreover, while EndoA1 is not required for endocytosis of the RTK TrkB, it is recruited by the endosomal protein retrolinkin to TrkB-positive early endosomes and directs sorting to late signaling endosomes, promoting downstream ERK signaling [106, 108]. In line with this, and as mostly observed in cancer cell lines (see below), EndoA-mediated trafficking of ligand-stimulated receptors influences intracellular signaling, decreasing signal transduction from the plasma membrane or increasing signal propagation from endosomes. Unfortunately, the molecular determinants governing distinct signaling outcomes (e.g., reduced or enhanced signaling from the plasma membrane or endosomes, activation of selected downstream pathways) following endocytosis remain poorly understood. Curiously, EndoAs may also have peripheral, endocytosis-independent functions in signaling, as EndoA1 was reported to interact with the Germinal Center Kinase-like Kinase, leading to JNK kinases activation [109].

EndoAs also participate in protein turnover and autophagy. In Drosophila neurons, EndoA promotes the formation of highly curved cytosolic membranes to which the autophagy factor Atg3 can dock to initiate autophagosome formation [110]. In mammalian neurons, all three EndoAs also interact with the E3 ubiquitin ligase FBXO32 to cooperatively promote autophagosome biogenesis and regulate protein turnover (Fig. 2) [111]. Similarly, in bovine aortic smooth muscle cells, EndoA2 stimulates ubiquitination and autophagic degradation of the Ca²⁺-activated Cl⁻ channel TMEM16A [112]. More indirectly, EndoA2 from rat cardiomyocytes promotes autophagy by augmenting EndoB1 interaction with the autophagy regulator Beclin-1 [113]. Importantly, and as discussed later, EndoA functions in autophagy are regulated by phosphorylation [110] and Ca²⁺ signaling [114]. In addition, the three EndoAs contribute to apoptosis by binding Alix, promoting cytoplasmic vacuolization [115]. Moreover, EndoAs localize to actin-rich protrusions such as dendritic spines [116] and podosomes [117], and may facilitate their morphogenesis. Indeed, EndoA1 interacts with the adaptor protein p140Cap in dendritic spines to promote actin remodeling essential for spine formation [116, 118]. Intriguingly, EndoA2 is also detected in the nucleus of hematopoietic, fibroblast, and epithelial cells, where it is hypothesized to undergo nucleo-cytoplasmic shuttling during cell cycle. In A. thaliana, during cytokinesis, SH3P2 in complex with dynamin-related protein 1 A induces tubulation of trans-Golgi-derived vesicles to form planar cell plates, further suggesting EndoA functions in cell cycle [119]. In summary, while EndoA functions are clearly not restricted to endocytosis, future studies are required to further delineate their mechanisms of action in such diverse cellular processes.

Being implicated in endocytosis and other key cellular processes, EndoAs must be tightly regulated to modulate their dynamic recruitment to the plasma membrane or other cellular structures upon given signals. PTMs, including phosphorylation, acetylation, and ubiquitination, have been identified on all three EndoAs in high-throughput mass spectrometry screens (Fig. 1C) (https://www.phosphosite.org/↗). EndoA1 is phosphorylated by leucine-rich repeat kinase 2 (LRRK2) at Thr⁷³ and Ser⁷⁵ within the H1I amphipathic helix. This phosphorylation event, likely by reducing membrane association and insertion, favors vesiculation. In contrast, non-phosphorylated form, through deeper membrane insertion, supports tubulation [12, 120, 121]. Accordingly, EndoA1 with phosphorylated Ser⁷⁵ supports synaptic macroautophagy by promoting the formation of highly curved membranes, rather than participating in endocytosis (Fig. 2) [110, 122]. Phosphorylation by Rho-associated kinase ROCKII of EndoA1-Thr¹⁴ also inhibits endocytosis: it disrupts EndoA1 binding to the adaptor protein CIN85, causing reduced EGFR internalization [123, 124]. Localized within the H0 helix, Thr¹⁴ phosphorylation may also reduce insertion within the plasma membrane. Phosphorylation of EndoA2 by Src kinase on Tyr³¹⁵, within its SH3 domain, alters dynamin binding and impairs endocytosis of the membrane-bound metalloproteinase MT1-MMP [125]. EndoA phosphorylation also occurs in other organisms, for example in ascidian Ciona, where phosphorylation of EndoA-Ser²⁶³ by dual specificity Tyr-phosphorylation-regulated kinase 1 (DYRK1) is required for optimal CME [126]. Finally, estrogen receptor-α (ERα), upon activation by 17β-estradiol, may also phosphorylate EndoA2, potentially affecting its endocytic functions [127].

EndoAs can also be regulated by the phosphorylation status of binding partners. For example, Cdk5- and GSK3β-mediated phosphorylation of dynamin and CRMP4 decreases their binding to EndoA2 and inhibits FEME [87]. Aside from phosphorylation, the E3 ubiquitin ligases Itch and Parkin bind to and ubiquitinate the SH3 domain of EndoA1 [128, 129], potentially regulating its stability and turnover. Moreover, EndoA2-Lys²⁹⁴, within the linker, was found conjugated to the ubiquitin-like protein MNSFβ, which regulates phagocytosis induced by the lectin receptor dectin-1 in macrophages [130, 131].

In addition to PTMs, calcium influx through voltage-gated Ca²⁺ channels also regulates EndoA functions in CM-SVE. This regulation appears to be mediated by Ca²⁺/calmodulin, which binds to both EndoA1 and EndoA2. Calmodulin bound to Ca²⁺ enhances EndoA1 association with the plasma membrane and the cytoskeletal regulator p140Cap [132], and promotes EndoA2-mediated membrane tubulation [133]. Alternatively, EndoAs may act as direct Ca²⁺ sensors, since Ca²⁺ initiates an EndoA1 conformational shift from a rigid, membrane-associated form to a more flexible cytosolic state that aids autophagosome formation (Fig. 2) [114]. Ca²⁺ is also suggested to bind directly to EndoAs, and its binding within the EndoA2 linker region induces an auto-inhibited conformation which blocks interactions with endocytic partners [45, 134]. The crystal structure of EndoA1 further suggests potential Ca²⁺ coordination at its N-BAR domain [8]. However, direct Ca²⁺ binding to EndoAs remains debated, as microcalorimetry failed to confirm such interactions [11]. Auto-inhibition via intramolecular interaction between the H0 helix and SH3 domain has also been proposed: such interaction would stabilize H0 helical folding in the cytosol and would be released upon membrane association [135]. An alternative mechanism rather suggests intradimer, intermonomer EndoA1 auto-inhibitory H0:SH3 interactions, relieved by SH3 ligand binding [136]. Finally, EndoAs are also likely transcriptionally regulated via mechanisms that remain mostly unexplored but which, based on dysregulated expression in various cancer types and other diseases (see below), may be implicated in pathogenesis.

The critical contribution of endocytosis to cancer, by regulating the cell surface turnover of signaling, adhesion, and immune proteins, as well as by influencing therapeutic drug delivery, no longer needs to be demonstrated [168]. It is therefore not surprising that endocytic proteins are frequently dysregulated in cancer, and EndoAs are no exception. Of course, EndoA implications in various other cellular processes may further contribute to cancer pathogenesis. Large systematic screens have revealed aberrant EndoA expression, post-translational modifications, and mutations in multiple cancer types (https://bioportal.bioontology.org/↗). More targeted investigations have confirmed the contribution of dysregulated EndoAs to cancer progression, primarily through altered receptor-mediated signal transduction, ultimately impacting cancer cell stemness, migration, invasion, proliferation, and drug resistance. While, to the best of our knowledge, EndoA2 is primarily reported to be pro-tumoral, both pro- and anti-tumoral contributions have been reported for EndoA1 and EndoA3 (Table 1).

The involvement of EndoAs in diverse pathologies highlights their potential for clinical applications. First, the aberrant EndoA expression observed in neurodegenerative, cardiovascular, and cancer contexts suggests possible uses as biomarkers. Since EndoA2 overexpression elicits high production of anti-EndoA2 autoantibodies in patients with low-grade glioma, breast, liver, gastric, and colon cancer [217, 218], serum autoantibody dosage could also be used for minimally-invasive early diagnostic. Given their pro-tumoral and neurodegenerative actions, as well as their implications in pathogen infection, EndoAs also represent promising therapeutic targets. The absence of overt phenotype in single EndoA knockout mice further supports the clinical relevance of selective therapeutic inhibition. Strategies may include trapping with therapeutic antibodies, interference using therapeutic peptides, or inhibition with small molecule inhibitors. Alternatively, the identification of miRNAs targeting the SH3GL2/1/3 genes, as well as the use of therapeutic siRNAs, also provides possibilities to artificially mitigate pathological EndoA overexpression. Conversely, enhancing SH3GL2/1/3 gene expression may be beneficial in diseases where they exert protective effects, for example as tumor suppressor in some cancer types or in cardiovascular diseases. Oligonucleotide therapies using small-activating RNAs or synthetic mRNAs could for example be explored. Therapeutic recombinant proteins are another alternative: for example, EndoA2 fused to the cell-penetrating peptide Tat penetrates the blood–brain barrier in gerbils, preventing ischemia-induced hyperactivity and oxidative stress [219]. Despite these prospects, no EndoA-targeted therapy currently exists, reminding us that their design and development are far from trivial. Among others, EndoA high structural similarity complicates selective targeting, as required to minimize possible side-effects. Less direct therapeutic approaches may also target upstream regulators, such as kinases, or even EndoA cargoes, but again with the possibility of deleterious off-targets. Finally, EndoA-mediated endocytosis could be exploited for targeted drug delivery. For example, immunoliposomes coupled with anti-ALCAM single-chain antibody fragments (scFv) have been used to deliver anti-tumor agents to prostate cancer cells [220], opening avenues for the future design of functionalized nanotherapeutics targeting EndoA-mediated endocytic axes.

Human EndoAs share approximately 65% amino acid sequence identity, along with conserved domain organization (Fig. 1). While EndoA1-3 triple knockout mice undergo perinatal mortality, single knockouts show no overt phenotype, implying partial redundancy [26]. However, it has become clear over the last decade that EndoAs also have distinct, paralog-specific functions. For instance, although EndoA3 is expressed in the brain, it does not appear to participate in synaptic vesicle endocytosis like EndoA1 and -A2 [46]. Additionally, EndoAs at the plasma membrane show non-overlapping localization profiles [84], suggesting affinity for different local membrane domains and a low probability of EndoA heteromerization in cells. Importantly, the EndoA proteins mediate distinct CIE modalities: EndoA2 facilitates FEME of ligand-activated receptors [83], while EndoA3 mediates the internalization of Ig-like CAMs [84]. However, the extent to which EndoA functions are conserved or divergent remains poorly defined, due to several limitations: (i) many studies focus on a single EndoA homolog, without assessing functional overlap with the others, (ii) some EndoA functions were characterized in model organisms such as D. melanogaster and C. elegans which express only one EndoA protein, and (iii) some studies refer generically to EndoAs without specifying the homolog being examined.

A key open question is how the three highly similar EndoAs may have partially distinct functions. One possibility could be EndoA-specific regulation by PTMs, as supported by the identification of phosphorylation sites on residues non conserved across the three human EndoAs in high-throughput phosphoproteomic screens (Fig. 1C). Functional differences may also arise from distinct protein:protein interactions. For example, EBP binds to EndoA2 and -A3, but not to EndoA1 [203], and the protein:protein interaction database BioGRID reports both shared and unique EndoA interactors. Specific interactions with cargoes and adaptor proteins, but also with proteins commonly involved in endocytosis like GTPases and molecular motors, are of particular interest to understand their involvement in different CIE modalities. Importantly, such differences in PTMs and binding partners may also influence EndoA subcellular localization, in particular their shuttling from the cytosol to cellular membranes, thereby possibly resulting in different kinetics at the cell surface. Variations in preferential lipid binding could also induce membrane partitioning, resulting in different functions. Interestingly, the EndoA H0 amphipathic helices show different net charge and hydrophobicity, possibly influencing lipid specificity and binding to specific local membrane domains [221]. However, the lipid-binding preferences of the different EndoA proteins remain to be further explored. Small variations in EndoA 3D structure may also result in membrane selectivity, for example towards specific curvature. Together, these open questions highlight the importance of structure–function studies comparing the three human EndoAs to understand the scope of individual functions at the cellular and pathophysiological levels.

This review illustrates pivotal functions of the EndoA proteins, primarily in endocytosis. In particular, recent advances highlight that EndoAs behave as a versatile module interacting with various other proteins to form functional endocytic machineries that are selectively recruited in several modalities. Interestingly, EndoAs are involved both in rapid (UFE, FEME) and slower (CME) endocytic mechanisms. In vitro kinetic studies indicate that EndoAs bind to biological membranes as dimers, then oligomerize into helical lattices on the membrane to induce tubulation [222]. These studies further reveal that EndoAs first act as curvature sensors, and require 30 ms to transition to membrane benders. While EndoAs induce local curvature within approximately 100 ms, larger membrane deformations like tubulation occur on a sec to min timescale. These studies highlight that the duration of endocytic events likely constrains the mode of action of EndoAs [222, 223].

Importantly, key questions remain unanswered in the EndoA field. First, the molecular mechanisms regulating such dynamic recruitment in selective endocytic routes are still poorly understood. Systematic identification of EndoA cargoes is also needed to fully appreciate the scope of their cellular functions. Complicating the study of EndoA-mediated endocytosis, cargoes are often internalized via multiple modalities, depending on the cellular context. Similarly, most endocytic players are shared among various modalities, with regulated spatial and temporal recruitment allowing more cellular plasticity. While complex and mostly overlooked, such crosstalk, compensation, and competition between EndoA-mediated and other endocytic modalities must be addressed. Beyond cargo internalization, downstream consequences – such as impact on signaling, adhesion, or immunity – although sometimes investigated, also warrants future attention. Emerging evidence for EndoA involvement in additional cellular processes significantly expands their functional repertoire. Yet, the molecular determinants governing context-specific EndoA functions at given subcellular localizations (for example Ca²⁺ signaling and phosphorylation regulating EndoA functions in endocytosis versus autophagy) should be further studied. Unfortunately, fully answering those open questions is hindered by technical challenges. First, their high similarity limits the availability of EndoA-specific molecular tools, particularly antibodies. Consequently, many studies still rely on ectopic expression of tagged proteins prone to artifacts. Similarly, the development of isoform-specific small molecule inhibitors would greatly facilitate functional and mechanistic studies, complementing siRNA-based approaches. Adding more complexity, inconsistent nomenclature, combined with the lack of isoform specificity in some earlier studies, complicate the integration of past findings and needs to be addressed in future studies.

Many studies link EndoAs to diseases, reminding us of the intricate connection between membrane trafficking and pathologies. However, limitations persist: causal relationships are often unconfirmed, most underlying mechanisms remain incompletely characterized, and available data are frequently restricted to in cellulo systems. More generally, despite established single- and triple-knockout mouse models [26], in vivo analyses remain limited, restricting our understanding of EndoA functions in health and disease. Nevertheless, and as discussed above, existing studies collectively support the potential for EndoA-based diagnostic and therapeutic strategies. In conclusion, we anticipate that future research on the EndoA proteins will continue to shape their mechanistic understanding and functional landscape, while suggesting promising translational applications.