Membrane reshaping by protein condensates

Abstract

Proteins can be organized into dynamic, functionally important assemblies on a fluid membrane surface. Phase separation has emerged as an important mechanism for forming such protein assemblies on the membrane during cell signaling, endocytosis, and cytoskeleton regulation. Protein-protein phase separation thus adds novel fluid mosaics to the classical Singer and Nicolson model. While phase separation can modulate the biochemical activities of proteins, the condensates formed in this process can also generate localized mechanical forces to change membrane morphologies. This is evident from the recent reports of protein condensate-driven membrane reshaping in processes such as endocytosis, autophagosome formation, and protein storage vacuole morphogenesis in plants. How do these condensate-driven curvature generation mechanisms contrast with the classically recognized scaffolding and amphipathic helix insertion activities of specific membrane remodeling proteins? A salient feature of these condensate-driven membrane activities is the integration of surface and bulk properties of the condensates and surroundings with the molecular-level protein-lipid interactions. This review highlights the current understanding of the physical mechanisms underlying curvature generation by protein condensates in various biological pathways.

1. Introduction.

Phase separation of biomolecules into liquid-like condensates has been in the spotlight recently for its diverse roles in cell physiology. On the one hand, phase separation drives the formation of membraneless organelles such as nucleoli, stress granules, and P-bodies [1]. On the other hand, association of the condensates with membranes has been found to be functionally important in cell signaling, endocytosis, and autophagy [2–5]. In biological systems, phase separation appears to serve two major purposes – i) compartmentalization of biomolecular activities and ii) regulation of concentration and composition of biomolecules by a switch-like action [6]. Most recently, membrane-interactions of protein condensates have gained significant attention through the key discoveries of their abilities to initiate and regulate membrane curvature in processes such as endocytosis [4, 7], generation of autophagosomes [5], and formation of protein storage vacuoles in plants [8]. In addition to molecular protein-protein and protein-lipid interactions, surface and material properties of the condensates are implicated in membrane remodeling activities [9].

Biomolecular condensate formation is mediated by multivalent interactions between mutually attractive molecular motifs, called ‘stickers’, that are generally present as repetitive units spatially separated by flexible or non-flexible linkers, also called ‘spacers’ [10]. Often sticker pairs are parts of the same molecular species, leading to phase separation mediated by homotypic interactions. If the sticker pairs are situated on separate molecular species, phase separation involves heterotypic interactions [11]. Under conditions mapped by a thermodynamic phase boundary, an aqueous solution containing biopolymers exhibiting homotypic or heterotypic interactions can demix into a dilute phase and a condensed phase, the latter having concentrations typically hundreds to thousands fold higher compared to the former. The threshold concentration for phase separation depends on solution parameters such as temperature, pH, ionic strength, and molecular crowding [12, 13]. In biological systems, additional factors such as ligand binding and post-translational modifications can also modulate the threshold concentration [12]. Finally, the membrane surface has been found to be a major factor that regulates the morphology and functionality of protein condensates in a growing number of biological pathways [14, 15].

Due to their liquid-like behavior and positive interfacial tension at the condensed/dilute interface, biomolecular condensates typically form spherical droplets in solution. When the droplets come into contact with biological interfaces such as membranes, or artificial interfaces such as glass surfaces, they can spread on the interface through adhesion [9]. Droplet adhesion can be characterized by a contact angle formed by the interface between the two fluid bulk phases (dense phase and dilute phase of the biomolecular condensates) against a third phase surface (for example, membrane or cytoskeletal filaments) [9, 16]. Depending on the surface interactions, a droplet adhered (i.e. partial wetting, the contact angle is < 180° but > 0°) to a surface can completely spread on or ‘wet’ the surface (contact angle is 0°) by a phase transition process known as wetting transition [17]. Partial membrane wetting by protein condensates has been recently found to drive membrane remodeling and protein reorganization on membrane surface in several biological processes, such as actin-independent endocytosis in yeast [7] and the generation of protein storage vacuoles in plant embryos [8].

Interesting surface phenomena can occur in the presence of binding interactions between the membrane and the proteins that form condensates in the bulk. A recent report by Zhao et al. [17] theoretically predicts the effects of membrane binding on phase separating proteins inside and outside the bulk two-phase regime. Inside the two phase regime of the bulk phase diagram where proteins can spontaneously form three-dimensional droplets, membrane binding of the droplet can be characterized by either partial wetting (finite contact angle) to complete wetting (zero contact angle), depending on the strength of the droplet/membrane interactions (and distance from the critical point bulk phase coexistence). This behavior was observed in recent experimental work by Lu et al., where the increase in electrostatic interactions between charged polymer condensates and membrane promotes complete membrane wetting by the condensates [18]. Even outside the bulk two-phase regime, phase separating proteins anchored to a membrane surface can assemble by a surface phase transition through lateral assembly while the bulk solution still remains in a single phase. This has been characterized with a wide range of peripheral proteins that form sub-micron size, two-dimensional membrane-clusters well below their corresponding bulk threshold concentrations for phase separation [4, 19]. Protein phase separation on membranes has been reported to be associated with several critical biological pathways, such as T cell receptor signaling [2], actin regulation [20], endocytosis [3, 4], and postsynaptic density assembly [21]. The realization that membrane-associated condensates can generate and regulate membrane curvature has added a new perspective to the developing understanding of biological membrane shaping phenomena.

The generation of membrane curvature is essential in various cellular pathways, including endo and exocytosis, organelle biogenesis, and filopodia formation. Growing evidence suggests that membrane-associated protein phase separation can play an important role in these processes. Various mechanisms such as i) partial wetting and capillary force generation [9], ii) free energy release through adhesion between viscoelastic media [7], and iii) generation of compressive stress by compacted intrinsically disordered regions (IDRs) [22] have been proposed, that can act either individually or in synergy in the curvature generation process. Here, we review the recent discoveries in membrane reorganization by protein phase separation in cellular function and discuss the underlying molecular mechanisms.

2. Three dimensional condensates can remodel membranes by wetting.

Three dimensional condensate droplets form a contact angle (Figure 1A) when adhering to surfaces such as solid glass coverslips or soft membrane surfaces in a process known as ‘wetting’. The wetting properties of a droplet on a vesicle surface and the resulting droplet/membrane geometry can be characterized by an effective contact angle (α) and an intrinsic contact angle (θ) (Figure 1A) [23]. The angle α is directly measurable by optical microscopy and is widely used for describing the droplet/membrane morphologies and it can be expressed as a function of interfacial tensions along the three-phase contact line. The angle θ depends on the droplet and membrane material properties. Although θ can not be directly measured at optical resolution, it can be derived from the effective contact angles [23]. The membrane wetting properties of condensates formed in aqueous solutions of the polymers dextran and polyethylene glycol (PEG) have been well characterized by Dimova, Lipowsky, Keating, and coworkers [24, 25]. Aqueous mixtures of dextran and PEG loaded into giant unilamellar vesicles (GUVs) phase separate into a dextran rich and a PEG rich phase when their concentrations are raised into the two-phase regime by causing efflux of water from the GUVs [24, 25]. Dextran-rich droplets can undergo a wetting transition, as indicated by a sharp decrease in the contact angle between droplet and GUV surface [24]. Mechanical forces generated from this wetting process can result in various membrane morphologies, such as membrane buds and tubules depending on the availability of excess membrane area and the membrane spontaneous curvature [26, 27]. These studies have established strong theoretical and experimental backgrounds for understanding condensate-driven membrane remodeling in vivo.

Figure 1. Protein condensates can regulate membrane curvature by various mechanisms.

A. Elastic and capillary interactions in droplet / membrane adhesion. During wetting of an elastic surface such as a membrane by a liquid like droplet (yellow) a three phase contact line forms. The effective contact angle α is a function of membrane tension, droplet surface tension, and droplet-membrane interfacial tension (γ_mem, γ_drop, and γ_dm respectively) whereas the intrinsic contact angle θ is related to the material properties of the droplet and membrane [23]. For a droplet of radius R capillary forces apply a torque of ~2πγ_dropR² that pulls the membrane against its bending rigidity κ which acts in the opposite direction relative to the capillary forces [9]. These two opposing effects would deform the membrane on a length scale of the elasto-capillary length, L_EC = √(κ/γ_drop), shown as the radius of curvature of the red circle in the right panel. B. Scaled membrane tension σ, defined as σ = γ_mem/γ_drop, determines lens-shaped (top) versus engulfed (completely wrapped, bottom) end-states of droplet-membrane interactions. The effective contact angles α and β can be used to describe the droplet/membrane morphologies. For σ > 1, partial wetting of membrane by the droplet gives rise to a lens shaped stage (α < 180°, β > 90°). For σ < 1, the membrane can wrap around the droplet surface, leading to a partially wrapped stage (α < 180°, β < 90°) followed by complete engulfment of the droplets [18]. C. Viscoelastic model of membrane bending by endocytic condensates in yeast cells [7]. Top panel, the endocytic condensate droplet (yellow) formed between membrane (mem, grey) and cytosol (c, light blue) is assumed [7] to maximize its contact area with the cytosol, which increases the cytosolic indentation depth (h՛ > h). If that droplet area increase occurs while keeping the droplet volume constant, an invagination with depth of δ would be created at the membrane-droplet interface, which induces membrane invagination. In this model, energy is proposed to be gained from the work of adhesion at the droplet-cytosol (dark blue) and the droplet-membrane (cyan) interfaces (U_a|d-c and U_a|d-m respectively, green arrows). Bottom panel, the total energy cost is a sum of elastic penalties (red arrows) to cause cytosolic and membrane deformation (U_e|c and U_e|m respectively), surface energy penalties to form extra droplet/cytosol and droplet/membrane interfacial area (U_i|d-c and U_i|d-m respectively) [29], and viscosity penalty to displace the cytosolic material (U_v|c). The invagination depth (δ) depends on the balance of total energy gain (ψδ) and total energy cost (Φδ^ε+1), where ε depends on the deformation geometry, and ψ and Φ summarize material properties determining gain and cost, respectively. D. During autophagy, protein droplets containing p62 cause wetting of autophagosomal double-membrane sheets. Top, the shape of a double-membrane sheet formed between the droplet (yellow) and cytosol (blue) surfaces can be explained by assuming two different spontaneous curvatures, one at the cytosolic surface (m_c), and the other at the droplet surface (m_d), and a curvature asymmetry can be defined as m_{cd ≡} m_c – m_d. The double-membrane may close either toward the droplet (when m_cd ≥ 0) or toward the cytosol (when m_cd < 0), to choose a specific cargo. Bottom, an enlarged section from the top panel to show protein-protein interactions, such as p62 present in droplet and LC3 present on membrane surface, can determine the direction of autophagosome enclosure [5]. E. Generation of membrane buds and tubules from GUVs enclosing a phase separated aqueous two-phase system. Efflux of water from GUVs causes phase separation of the enclosed polymers and also generates excess membrane area. The excess membrane area release can lead to formation of membrane buds and tubules by an interplay between wetting behavior of the phase separated system and membrane spontaneous curvature. F. Membrane bound BAR-protein, endophilin, can lead to adhesion between two adjacent membranes when present with its multivalent binding partner lamellipodin [4]. Adhesion of formed membrane tubules on the flat membrane surface (top) can increase membrane tension by adhesion and wrapping transitions (bottom), suppressing tubule and bud formation. G. Anchored IDRs (red and blue chains) on a flat membrane show decreasing segment density (shaded area) at increasing distance from the membrane. In such conditions, a phase separating IDR would form less sticker-sticker interactions (right panel) further away from the membrane. If the membrane bends toward the IDR, more sticker-sticker interactions will take place (left panel). The increase in binding interactions would compensate for the compressive stress generated on the membrane. If the membrane and anchored chains exhibit attractive interactions (green highlighted regions) it will provide additional effects to bend the membrane toward the IDRs [22].

In a recent perspective, Gouveia et al. have highlighted the role of capillary forces generated by biomolecular condensates causing partial wetting of elastic biological surfaces, such as membranes or cytoskeletal filaments, involving structural remodeling of both condensates and the surface [9]. Capillary force (F) generated on the membrane surface by a droplet of radius R can apply a torque (F×R) of ~2πγ_dropR², γ_drop being the condensate surface tension, to bend the membrane toward the direction of the droplet adhesion (Figure 1A). This will be resisted by the bending stiffness of the membrane κ. As a result of these elastic and capillary forces, a membrane deformation of a length scale of the elasto-capillary length L_EC = √(κ/γ) takes place.

Various examples have been found where membrane wetting by three-dimensional protein condensates plays important roles in driving cellular pathways. Membrane wetting by either synthetic polymer condensates or protein condensates leads to various membrane shape transitions such as budding, tubulation, vesiculation, and reticulation [26, 28]. Factors that would regulate the resulting membrane morphologies include droplet-membrane adhesion force, bending rigidity of the membrane, material properties of the condensate, membrane spontaneous curvature, and availability of excess membrane area. In the following subsections, we discuss how interactions with three dimensional condensates can drive membrane remodeling in various cellular pathways.

2.1. Three-dimensional condensates exhibiting electrostatic membrane interactions generate membrane curvature

The interaction of liquid condensates and lipid membranes can lead to interesting membrane shape changes. Lu et al. found that spontaneous condensate-engulfment resembling endocytosis could be driven by attractive electrostatic interactions when condensates composed of charged polymeric substances were introduced to oppositely charged liposomes [18]. The condensates formed different morphological states on the GUV surface that could be classified by their effective contact angles, α and β (Figure 1B). A relation between cos α and the interfacial tension parameters can be used to determine the scaled membrane tension σ, defined as σ = γ_mem/γ_drop. When Lu et al. increased the interaction strength between the droplet and the membrane by increasing the proportions of oppositely charged components, a transition from membrane adhered spherical droplet state to complete membrane wetting state occurred. Depending on σ, the wetting transition was found to occur via two separate pathways, which could be characterized by different droplet/membrane morphologies of the partial wetting and the complete wetting stages. Formation of membrane-engulfed spherical droplets, or ‘endosomes’, was possible under the condition σ < 1, which seemed to take place through a ‘lens shape’(α < 180°, β > 90°) followed by a ‘partial wrapping’ (α < 180°, β < 90°) intermediate stage (Figure 1B). For σ > 1, only the lens-shaped partial wetting stage occurred, and the complete wetting gave rise to liposomes covered by a spherical layer of condensates. Evidence of spontaneous engulfment of condensates suggests the possibility that they can be utilized as delivery agents for therapeutic purposes in the future [18, 30].

In the case of protein condensates undergoing membrane wetting, electrostatic interactions can modulate membrane/condensate morphology. Mangiarotti et al. showed that when condensates of the plant protein glycinin adhere to the membranes containing zwitterionic lipids, dewetting, partial wetting, and complete wetting states appear in a salt concentration-dependent manner. A lower salt concentration favors dewetting whereas a higher salt concentration favors a complete wetting [28]. Interestingly, in the presence of either cationic or anionic lipids, the transition boundaries between dewetting to partial wetting and partial wetting to complete wetting transitions are shifted to higher salt concentrations. Ionic strength was also shown to modulate the bulk phase behavior of glycinin. However, by what mechanism increasing membrane charge leads to membrane dewetting by the condensates at lower salt concentrations has remained elusive thus far.

2.2. Viscoelastic and adhesive properties of protein condensates modulate membrane curvature during endocytosis in yeast.

In yeast, endocytic proteins containing prion-like domains (PLDs) form hemispherical, membrane-adhered condensates that cause membrane invagination and enable clathrin and actin-independent endocytosis [6]. To explain the mechanism, Bergeron-Sandoval et al. provide a viscoelastic model emphasizing the tendencies of the condensate to expand its cytoplasm-condensate interfacial area by creating an indentation into the viscoelastic cytoplasm. Coupled with the increase in indentation depth, an invagination of depth δ at the membrane-condensate interface forms to ensure that the volume of the viscoelastic condensate remains constant (Figure 1C, top panel).

The authors use a modified version of the Johnson-Kendall-Roberts (JKR) theory, which allows estimation of the energy gain from the work of adhesion and the energy penalties when a sphere adheres to a deformable substrate [29]. The membrane bending by PLD condensates could be deconvoluted into two adhesion processes – on one side, adhesion of the condensate to cytosolic indentation takes place, and on the other side adhesion of membrane into the droplet invagination occurs. According to a modified JKR theory as implemented by Bergeron-Sandoval, the total energy cost for each of these two adhesion processes was expressed as the sum of two penalty terms - i) elastic energies to bend the membrane or the cytosolic surface, and ii) energy to create extra contact area at the droplet/cytosol or droplet/membrane interfaces. In addition, Bergeron-Sandoval et al. considered an additional energy penalty due to the displacement of the viscous cytosol. The energy gained from the work of adhesion was derived from the JKR model terms that consider changes in surface energy when two objects form a new interface by replacing two existing surfaces [29]. When adhesion is favorable, the total change in surface energy becomes negative when the surface energy of the newly formed interface is lower than the energy of the other two disappearing surfaces. However, in the absence of such disappearing surface area, it is not entirely clear what drives the expansion of the droplet/cytosol interface.

The Bergeron-Sandoval model predicts the membrane invagination depth occurring from the balance of the energy penalty and the energy released through the adhesion of the condensates with the membrane and cytosol (Figure 1C, bottom panel). Computing experimentally derived elastic and mechanical parameters of condensates, cytosol, and membrane within this model, the authors showed that the energy released by adhesion could favor invagination depths between 15–80 nm which agrees with the experimentally determined endocytic membrane invaginations in yeast cells.

2.3. Membrane wetting drives autophagosome generation.

Autophagy is an essential recycling process in cells that involves sequestration of a portion of cytoplasm by double-membrane autophagosomes. Agudo-Canalejo et al. showed that membrane wetting by condensates of p62, a crucial protein in autophagy, plays a central role in the formation of the autophagosome and isolation of cytoplasmic portions by the autophagosome [5]. Membrane wetting can generate autophagosome-like double-membrane sheets at the droplet-cytosol interface. The shape of such double membrane sheets can be represented by two different bilayer spontaneous curvatures, one of the bilayer facing the cytosol (m_c), and the other of the bilayer facing the droplet (m_d) (Figure 1D, top panel). Differences in these two spontaneous curvatures can be defined as m_{cd ≡} m_c – m_d. The double-membrane may enclose either toward the droplet (m_cd ≥ 0) or toward the cytosol (when m_cd < 0), to choose a specific cargo. Interestingly, the membrane spontaneous curvature could be regulated by specific interactions between p62 from the droplet with LC3, another protein present on the autophagosomal membrane (Figure 1D, bottom panel). Therefore p62-LC3 interactions determine whether the autophagosomal membrane encloses toward the droplet or toward the cytosol. This study opens up the scope for further research to explore how an interplay between droplet material properties, non-specific droplet and membrane interactions, and specific protein–protein interactions drives sequestration of various protein condensates into autophagosomes under physiological and pathological conditions.

2.4. Storage protein droplets drive vacuolar membrane remodeling during plant embryo maturation.

Lipowsky, Dimova, and coworkers using GUV-encapsulated aqueous two-phase system (PEG / dextran) demonstrated that formation of membrane tubule networks could result from droplet-membrane interactions [26]. In such GUVs, hyperosmotic stress induces demixing of the two polymer phases (PEG-rich and dextran-rich) due to the increase in the polymer concentrations inside the GUV (Figure 1E). The loss of the GUV interior volume also generates excess membrane area that allows changes in the vesicle geometry such as bud and tubule formation. The membrane morphology is governed by the wetting abilities of the two-aqueous phases and the membrane spontaneous curvature. Stabilization of tubules along with buds requires higher values of membrane spontaneous curvature compared to bud formation only.

Recently, membrane remodeling by phase separated protein droplets has been found to play a crucial role in plant physiology [8]. During seed maturation in plant embryos, the membrane-bound large storage organelles known as vacuoles undergo morphogenesis into multiple protein storage vacuoles, which involve remodeling of the vacuolar membranes. Kusumaatmaja et al. reported that this transition could be facilitated by the phase separation of storage proteins into micron-sized condensates leading to membrane remodeling driven by wetting of the vacuolar membrane by the condensates. Depending on the droplet wettability and membrane spontaneous curvature, the vacuolar membrane could be remodeled by the storage protein droplets into either bud without tubulation, or an ensemble of membrane buds with tubules, or a network of droplet-wet nanotubes without budding. These observations suggest that in biological systems, a regulatory mechanism of droplet and membrane mechanical properties could pre-exist to direct the transitions between different membrane morphologies under various physiological conditions.

3. Two-dimensional protein phase separation on the membrane surface and effects on membrane curvature.

In the previous section, we discussed the wetting transition of three-dimensional droplets, the formation of which requires conditions found inside the two-phase coexistence (dilute phase and droplet phase) regime of the bulk phase diagram. In the presence of membrane binding, proteins can laterally phase separate on the membrane surface even at bulk concentrations below the two-phase bulk coexistence regime [19]. This surface-specific phase transition can organize proteins laterally into coexisting protein-enriched and protein-depleted phases. Surface phase transitions appear to be involved in several cellular signaling and membrane trafficking processes [2, 3, 19]. Protein domains, primarily known for their bulk phase separation behavior within biological cells and in vitro, can also exhibit a surface phase transition when tethered to model membranes using suitable synthetic anchoring groups [22]. Conversely, proteins known to be involved with membrane remodeling have recently been found to phase separate on the membrane [4]. In this section, we will discuss the current experimental evidence where protein phase separation on the membrane has been associated with membrane reshaping phenomena.

In addition to the two-dimensional surface phase transition, another surface-specific phenomenon, the ‘prewetting’ transition, has been proposed to take place in the presence of membrane binding when the bulk concentration is close to the two-phase coexistence boundary but still below the bulk threshold concentration. Unlike a surface phase transition, which results in two-dimensional demixing, a prewetting transition separates between - an apparent two-dimensional ‘thin’ layer of molecular thickness and a three-dimensional, ‘thick’ layer [17]. Protein condensation on the membrane can also be coupled with lipid phase separation, as experimentally demonstrated with membrane-reconstituted T-cell receptor signaling clusters [31, 32]. Rouches et al. have provided a theoretical model that considers coupling between bulk phase separation, phase separation of membrane-tethered components, and membrane phase transition resulting in the formation of membrane-associated condensates [33]. More research is required to understand if membrane-bending effects can result from surface phase transitions and prewetting transitions.

3.1. Phase separation can modulate membrane curvature generation by BAR-proteins

Proteins containing BAR domains have been proposed to function as curvature generators in both clathrin-mediated and clathrin independent endocytosis (CME and CIE, respectively) [34, 35]. Among the mechanisms for membrane bending that have been proposed are: i) scaffolding by the crescent shaped BAR domain, ii) wedging by insertion of N-terminal amphipathic sequences that form a helix upon membrane interaction, and iii) generation of steric pressure by protein crowding. Several members of BAR superfamily proteins have recently been shown to undergo phase separation either upon binding to multivalent partners or via self-association [3, 4]. BAR protein-mediated phase separation has been demonstrated to likely be significant for both CME [3] and a CIE pathway known as FEME [4]. The question therefore arises, how does phase separation affect the curvature generation activities of BAR proteins?

BAR-proteins, apart from their membrane-active ‘BAR-domain’, contain auxiliary domains known as Src homology 3 (SH3) domains that specifically bind to proline-rich-motif (PRM) containing proteins such as lamellipodin (LPD), dynamin, and synaptojanin. We showed that the BAR protein endophilin phase separates upon binding to its multivalent partners, such as the C-terminal tail of the adapter protein LPD that recruits endophilin to the leading edge of cells and the intracellular loop of a G-protein coupled receptor (GPCR) that is internalized by the endophilin-dependent FEME pathway [4]. The study suggests that phase separation can promote formation and maturation of FEME priming sites by causing protein clustering on the membrane and by facilitating partitioning of other endocytic proteins into FEME sites. Apart from forming protein assemblies, reconstitution of endophilin and its multivalent partner LPD was observed to have two major impacts on liposomes. Membrane surfaces coated with endophilin and LPD underwent extensive membrane-membrane adhesion (Figure 1F) and, secondly, tubules towards the vesicle exterior, generated by endophilin, underwent an apparent shortening in length. We also observed an increase in the membrane tension with giant vesicles, which can be caused by adhesion of a large number of membrane tubes or smaller vesicles on the GUV surface. The endophilin/LPD system might also form three-dimensional droplets of sub-optical resolution on the membrane surface that could undergo wrapping and thus increase the membrane tension. These observations indicate that when endophilin is clustered into the FEME priming patches prior to receptor activation, its curvature generation abilities could be modulated, such as through inhibition, by LPD via multivalent interactions. It remains to be determined how curvature generation activity is triggered following receptor activation and whether this requires disengagement of endophilin-LPD interactions.

Why do cells need to regulate curvature generation activities of BAR proteins? BAR proteins due to their inherent abilities to bind anionic lipids, are known to generate membrane curvature spontaneously when introduced to model membranes containing such lipids [36]. However, in a specific cellular pathway such as receptor endocytosis, recruitment of BAR proteins via membrane lipids is not always sufficient for curvature generation. For example, membrane curvature generation in FEME is initiated by stimulation of specific GPCRs [37, 38]. This is evident from cryo-EM images of FEME sites before receptor activation that showed membranes to remain flat underneath endophilin rich priming patches [39]. These patches on the flat membrane could provide a local source of curvature generators whose activity could be triggered by receptor activation.

Another situation where SH3-PRM interactions could play regulatory roles is the dynamin and BAR-protein mediated membrane scission process. Dynamin is a GTPase that drives membrane scission to produce endocytic vesicles during CME and several CIE pathways. The C-terminal tail of dynamin contains multiple PRMs that interact with the SH3 domains of endophilin, amphiphysin, and BIN1. BAR-proteins are known to recruit dynamin to the neck of the membrane buds to facilitate membrane scission. While BAR-proteins and dynamin are individually known to drive membrane remodeling, they were found to inhibit each other’s membrane fission activities [40]. Interactions between dynamin-2 PRMs and the BAR proteins, endophilin and amphiphysin, have been found to inhibit the curvature generation abilities of both dynamin-2 and the BAR proteins in vitro [41]. How do the binding interactions mediated by SH3 domains influence the curvature generation by BAR-proteins? Answering this might unravel a new curvature regulation mechanism that the binding partners of the BAR-proteins control.

Apart from N-BAR proteins, members of F-BAR and I-BAR families were also found to undergo phase separation. However, the underlying mechanisms for phase separation might not be the same. Unlike N-BAR proteins, the F-BAR protein Fcho1/2 does not form liquid-like droplets in a crowded environment. However, Fcho1/2 phase separates upon multivalent interactions with its binding partner Eps15 and their phase separation was shown to be important for the initiation of CME [3]. Since the F-BAR domain can generate membrane curvature in the presence of anionic lipids [42], a relevant question would be whether multivalent SH3-PRM interactions also play a role by regulating the curvature generation behavior of Fcho1/2 in CME.

In contrast to N-BAR and F-BAR proteins, I-BAR family proteins sense and generate negative membrane curvature [43]. The I-BAR protein IRSp53 was shown to undergo phase separation inside membrane tubules upon sensing negative membrane curvature. The SH3 domain of IRSp53 allows its binding with VASP, an actin regulatory protein. IRSp53 and VASP were found to form clusters both in solution and on the plasma membrane but it is not known whether these proteins undergo phase separation by heterotypic interactions [44]. The clustering of VASP and IRSp53 facilitates actin polymerization during filopodia formation [45]. Notably, VASP also can interact with LPD PRMs and their interactions have recently been found to drive phase separation in solution [4]. We speculate that phase separation of IRSp53, VASP, and LPD can provide a new perspective on cytoskeletal regulation and curvature generation during the formation of membrane protrusions.

3.2. Phase separation of intrinsically disordered proteins on the membrane can generate curvature

Proteins with intrinsically disordered regions (IDRs) are known to generate membrane curvature by ‘steric pressure’ that is released by bending the membrane ‘outward’ or toward the direction of protein binding [46]. In a recent report, Yuan et al. [22] showed that membrane-tethered intrinsically disordered low complexity domains of proteins such as FUS, and hnRNPA2, can lead to ‘inward’ curvature upon phase separation of the proteins on the membrane. Phase separating IDRs can be described as polymer chains with a combination of ‘stickers’ and ‘spacers’ where stickers represent the polymer segments showing attractive interactions. To understand the curvature generation mechanism, a model can be considered with self-interacting polymer chains anchored to the membrane at one end of the chain. As theoretically shown by DiMarzio and McCrackin, the segment density of a tethered polymer decreases steadily with the distance from a surface when the polymer exhibits a certain level of attractive interactions with the surface [47]. Yuan et al. proposed that decreased segment density of IDRs would allow less sticker-sticker interactions to take place as the distance increases from the membrane (Figure 1G, left panel). Through bending toward the direction of polymer attachment (Figure 1G, right panel), therefore, the system can increase the segment density and allow more sticker-sticker interactions to form. The energy required to bend the membrane can be compensated by increased binding interactions between the IDR segments (stickers). In addition to the protein-protein interactions, interactions between membrane and the protein chain could also modulate membrane spontaneous curvature. Theoretical work from Lipowsky and coworkers suggested that when a membrane anchored polymer chain experiences substantial attractive interaction toward the membrane, maximizing this adsorption would tend to bend the membrane toward the side of polymer binding [48]. However, if the polymer-membrane interaction is repulsive, the membrane could bend in the opposite direction (away from the polymer). Therefore, both protein-protein and protein-membrane interactions play regulatory roles when membrane curvature is induced by membrane-attached IDRs.

An important direction to explore is how membrane phase separation of proteins that contain both ordered and disordered domains affect the membrane curvature. BAR-proteins, an important class of membrane curvature generators, have folded membrane scaffolding domains and disordered linkers. In the previous section, we discussed the effect of LPD-IDR binding on the curvature generation behavior of the BAR-protein endophilin. According to the current membrane-anchored associative polymer model, endophilin-LPD assembl might also have tendencies to bend the membrane inward. This effect could oppose endophilin’s intrinsic outward curvature generation behavior and might provide an alternative mechanism behind the observed shortening of outward tubules and increase in membrane tension. Whether the IDRs from the BAR-proteins and their binding partners play any roles in regulating curvature generation activities in vivo is not yet known.

4. Outlook

Membrane/protein-condensate interactions provide biologically relevant functionalities. Observations of membrane-reshaping behaviors of protein condensates have added new modalities to understanding membrane curvature generation in cells. Size, shape, and directions of curvature generated by condensates rely on physical properties, such as interfacial tension and viscosity of the membrane, condensate, and the surrounding microenvironment. Therefore, what cellular mechanism regulates these physical parameters and how their critical values are maintained under various physiological conditions, is yet to be discovered. An exciting question to explore further would be how the local distribution of membrane lipids plays role in facilitating condensate-driven membrane curvature generation. In addition, while various classes of protein condensates have been shown to induce membrane curvature, how phase separation of the classical curvature generator proteins can regulate their functionality in cells remains to be elucidated.

Over the past 50 years, insights into protein organization on membranes have added new layers of complexities to the original considerations of the Fluid-Mosaic membrane model [49]. Peripheral and integral proteins are already known to form heterogeneously distributed protein complexes regulated by other proteins on the membrane, specific lipid headgroups, and cytoskeletal components. In this context, protein phase separation on the membrane offers a new mechanism for organizing proteins to form submicron to micron-sized mosaics mediated by protein-protein and protein-lipid interactions. Adhesion of three-dimensional condensates to the membrane surface and the resulting membrane shape change can be understood by considering interactions between two different fluid surfaces, which adds a new perspective to the classical Fluid-Mosaic model.