Paul A.
Beales
*a,
Barbara
Ciani
*bc and
Alexa J.
Cleasby
bc
aSchool of Chemistry and Astbury Centre for Structural Molecular Biology, University of Leeds, Leeds LS2 9JT, UK. E-mail: p.a.beales@leeds.ac.uk
bCentre for Membrane Interaction and Dynamics, Department of Chemistry, University of Sheffield, Sheffield S3 7HF, UK. E-mail: b.ciani@sheffield.ac.uk
cCentre for Chemical Biology, Department of Chemistry, University of Sheffield, Sheffield S3 7HF, UK
First published on 16th March 2015
Compartmentalisation of cellular processes is fundamental to regulation of metabolism in Eukaryotic organisms and is primarily provided by membrane-bound organelles. These organelles are dynamic structures whose membrane barriers are continually shaped, remodelled and scaffolded by a rich variety of highly sophisticated protein complexes. Towards the goal of bottom-up assembly of compartmentalised protocells in synthetic biology, we believe it will be important to harness and reconstitute the membrane shaping and sculpting characteristics of natural cells. We review different in vitro membrane models and how biophysical investigations of minimal systems combined with appropriate theoretical modelling have been used to gain new insights into the intricate mechanisms of these membrane nanomachines, paying particular attention to proteins involved in membrane fusion, fission and cytoskeletal scaffolding processes. We argue that minimal machineries need to be developed and optimised for employment in artificial protocell systems rather than the complex environs of a living organism. Thus, well-characterised minimal components might be predictably combined into functional, compartmentalised protocellular materials that can be engineered for wide-ranging applications.
This statement couldn't be more evident than in the stunningly complex and vastly interlinked systems that make up each and every living cell. Cells are some of the most incredibly adept devices we know and understanding them is the first step towards manipulating and harnessing cells as machines and applying them in novel ways.
One of the cell's most ingenious skills is also one of its most simple: compartmentalisation. Cell biologists have been intrigued for decades by the organisational abilities of cells and their clever methods of compartmentalisation. The ability to keeps things separate sounds simple, and yet, behind the scenes lies hundreds of interacting proteins and precise pathways that orchestrate and permit a multitude of mini-reactions to occur independently, repeatedly and simultaneously in the space of one cell. Eduard Buchner commented in his 1907 Nobel Lecture that ‘We are seeing cells more and more clearly as chemical factories, where the various products are manufactured in separate workshops, the enzymes acting as the overseers.’
This same complexity is one reason designing experiments to exclusively probe one aspect of a cell's function can be near impossible in vivo, and yield results difficult to interpret. Cell biology relies heavily on the use of model systems, which have considerably furthered the current understanding of how cells work, but lack the finer mechanistic detail.
Synthetic biology† or reconstitution studies, addresses this issue by offering a powerful way to study proteins in a stripped down system that can be tightly controlled. Liu & Fletcher1 outline the benefits succinctly; in vitro experiments can refine and confirm molecular processes outside of a complicating cellular environment, whilst providing insight into how these biological functions could be engineered towards new functions. It is this idea that lies at the heart of synthetic biology; creating an interface between biology and engineering and the understanding to turn cell biology into a physical and engineering discipline.
It's important to appreciate there is so much more to synthetic biology than uncovering the secrets of how cells work. In vitro reconstitution experiments are slowly revealing a whole toolkit of proteins with enormous potential – the full scope of which is yet to be realised and with only man's imagination as the limit to the possibilities. Occurring throughout cells, dynamic membrane remodelling is gained by the precise interplay of proteins and lipids that are able to generate, sense and stabilise membrane curvature.
Here, we introduce a handful of these proteins and the experimental setups that enabled the study of the mechanisms through which they shape membranes, together with examples of how this knowledge has been used to produce minimal systems of membrane remodelling tools. The scope of this review is to present advances in recent in vitro studies of membrane deformation mechanisms by natural systems as an opportunity for developing new instruments for the generation of ad hoc compartments leading to the engineering of functional protocells. First we give an overview of different model membrane systems and how they are best applied in studying these protein–membrane interactions and construction of protocellular systems. We then review recent mechanistic insights into the function of proteins involved in the sculpting and scaffolding of membranes gained from biophysical studies on reconstituted minimal systems. This leads us into a brief overview of the theory of membrane elasticity and the important additional insights of combining appropriate theoretical models with biophysical experiments in understanding biological mechanism. These biophysical insights form the basis of a discussion of how these lessons from nature can be best applied to the design of protocellular materials where we argue the merits of designing minimal machineries optimised for in vitro function and give recent examples of such reengineered or biomimetic nanomachines. We close with an outlook for the future development of nanomachines for dynamic control of compartmentalisation in engineered protocells, proposing how the field might proceed from biophysical insight of natural machineries to biological components that can be engineered into predictable protocellular machines.
A variety of methods exist for creating substrate-supported model membranes. The most commonly used techniques are the vesicle fusion method, where small or large unilamellar vesicles, which are dispersed in an aqueous solution above the substrate, fuse with the supporting surface to form a single lipid bilayer, and the Langmuir–Blodgett (LB) method, where the substrate is passed twice through a lipid monolayer formed at an air–water interface by dipping the cleaned solid-surface into and out of the aqueous phase.17 Several methods have also been developed to create asymmetric supported lipid bilayers where each lipid monolayer has a different lipid composition.18–20 This mimics the lipid asymmetry known to occur in natural membranes and allows characterisation of how such asymmetry modulates the physical properties and behaviour of the membrane.
Due to solid membrane substrates not generally being relevant as natural biological interfaces and their hindrance for incorporation of transmembrane proteins, softer supports have been developed. These include supported membranes on polymer cushions or hydrogels and membranes extended from the surface by “molecular stilts”: functionalised lipids such as oligoethylene glycols with a thiol functionality at one end for covalent surface attachment and a sterol at the other to insert into the hydrophobic core of the membrane.21–25 Furthermore, techniques have been developed to create regions of unsupported membrane within a supported lipid membrane by forming the bilayer atop a microfabricated substrate that contains small holes. The supported membrane does not enter these holes but rather spans the gap, providing regions of membrane that do not directly interact with the substrate.26–28 Finally it should be noted that supported membranes have been formed on the surface of colloidal particles; this allows the dispersion of supported membranes into bulk solution and introduces curvature to the membranes comparable to those found in biological cells and organelles.29,30
(a) Lipid nanodiscs are lipid bilayer micelles whose edge tension is stabilised by α-helical scaffold proteins.31,32 Depending on the scaffold protein that is employed, these disc-like membrane fragments vary in diameter between approximately 10–18 nm with a narrow size distribution.33 These model membranes have primarily been employed as a more biologically relevant alternative to detergent solubilisation of integral membrane proteins for structural and functional assays.34–36 Most relevant to the study of membrane remodelling protein complexes is the use of lipid nanodiscs for single molecule fluorescence spectroscopy studies.37 Due to the small surface area of membrane available on a lipid nanodisc, experimental protocols can be designed such that only a single protein or protein complex will interact with a nanodisc. This scenario is much harder to achieve using model membrane systems based upon lipid vesicles, which have a much larger surface area per particle for interactions to take place and the number of interacting proteins per lipidic particle would follow a Poisson distribution.
(b) Droplet Interface Bilayers (DIBs) are formed using water-in-oil emulsions.38 Water droplets dispersed in an oil containing dissolved lipids recruit a lipid monolayer to the oil–water interface. When two of these droplets are brought into contact, oil is excluded from the interfacial region between the droplets such that a lipid bilayer membrane forms between the two aqueous compartments. Electrophysiological measurements have been used to study the ion permeability of these membranes in the presence of integral and peripheral membrane proteins.39–41 Experimental geometries have also been developed that allow simultaneous electrophysiological and fluorescence microscopy studies.42 As we will see, DIBs are also a starting point for some of the more advanced methods for fabrication of GUV model membranes. Notably, DIBs and vesicles, which will be discussed below, have the capacity to compartmentalise chemical entities and processes within an encapsulated volume and therefore are prime for a stepwise increase in complexity of reconstituted biological systems towards the bottom-up development of artificial cells and tissues.
(c) Small or Large Unilamellar Vesicles (SUVs or LUVs) are nanoscale or sub-micrometer unsupported membrane shells with an encapsulated aqueous compartment segregated from the bulk water phase. SUVs (<100 nm diameter) are usually formed by high energy sonication of multilamellar vesicles (MLVs; formed by the dispersion of dried lipid films into an aqueous buffer by vortexing).43 LUVs (100–1000 nm diameter) can be formed by several different methods; most commonly they are formed by multiple extrusions of MLVs through polycarbonate filters with monodisperse pore sizes of between 100–400 nm. Alternatively they can be formed by reverse phase evaporation or detergent extraction methods. Aqueous vesicle dispersions are often studied by spectroscopic or calorimetry techniques that report the statistical ensemble averaged behaviour of a large population of vesicles; example techniques include isothermal titration calorimetry, fluorescence spectroscopy, dynamic light scattering, electrophoretic light scattering and electron spin resonance spectroscopy. More recently techniques have been developed that allow single vesicle analysis of sub-micrometer vesicles, e.g. by anchoring these vesicles to a surface and using spatially-resolved surface analytical techniques.44,45 One of the advantages of SUV and LUV systems is the ability to create membranes of high curvature, thus allowing systematic study of curvature-dependent effects that influence protein–membrane interactions.46–48
(d) Giant Unilamellar Vesicles (GUVs) are vesicles that are >1 μm in diameter (most typically in the range of 10–30 μm) and so are observable by optical microscopy techniques. This allows experimental observations to be conducted at the single vesicle level. Beyond the average behaviour of a sample, single vesicle analysis can reveal the full distribution of vesicle behaviours, transient intermediate states and is sensitive to rare events. Due to their large size (similar to that of biological cells), GUV membranes have low curvature and are usually considered to be effectively flat for the purposes of interactions with nanoscale proteins and protein complexes. However methods have been developed to study curvature effects in GUVs by extruding cylindrical lipid tubules from the GUV surface that are a few tens of nanometers in diameter: the cylindrical curvature of these structures implies high curvature for only one of the two principal curvatures of the membrane, the second principal curvature being equal to zero.49–51 For high spherical membrane curvature (both principal curvatures are high), then SUVs or LUVs are more appropriate model systems.
While optical microscopy is often used to study GUVs, GUV experimental analysis can go well beyond the recording of qualitative pictures, as summarised in Table 1. Dense (e.g. sugar) solutions are often encapsulated within GUVs such that the encapsulated phase is more dense than the surrounding solution, causing the vesicles to sediment to the bottom of the sample dish, making them easier to find and observe.52 Methods have also been developed that allow reconstitution of integral membrane proteins into GUV model membranes.53
Measurement | Technique | Brief description and comments | Key and example references |
---|---|---|---|
Membrane morphology | Optical microscopy | GUV morphology changes directly observed by optical microscopy (phase contrast, epifluorescence, fluorescence confocal) can give qualitative clues to the nature of material interactions with the membrane. | 52, 166 |
Membrane fluidity | Fluorescence recovery after photobleaching (FRAP) | A region of the membrane is bleached of fluorescence with a high energy laser of a confocal microscope. The rate of fluorescence recovery by dyes from outside of the region of interest can be quantified into a lateral diffusion constant of lipids. This methods assumes homogeneous diffusion across a larger area of membrane (i.e. the photobleached region) | 164, 170, 171 |
Fluorescence correlation spectroscopy (FCS) | The autocorrelation function of fluorescence fluctuations of a pixel on the GUV membrane is used to quantify the diffusion constant of fluorophore-labelled lipids or membrane-bound proteins. This is a more localised measurement than FRAP (above) and can be confined to the nanoscale using stimulated emission depletion microscopy FCS (STED-FCS172,173). | 172–180 | |
Mechanical moduli | Micropipette aspiration | GUVs are aspirated into a micropipette; stress-strain plots are used to quantify κb, KA and the lysis stress and strain. κb is difficult to measure as this requires a lot of data points in the low tension regime. Membrane contour analysis (below) can be better for measuring κb. | 123–125, 181, 182 |
Membrane contour analysis | The radial localisation of the GUV's membrane is determined with sub-pixel resolution. Fitting the power spectrum of membrane fluctuations allows the bending modulus to be quantified. | 183–188 | |
Membrane permeability | Dye translocation kinetics | The rate of influx of water soluble fluorescent probes into GUVs can be used to quantify the membrane's permeability and the fractional permeable area of membrane. | 189–191 |
Patch-clamp electrophysiology | Electrophysiological patch-clamp recordings have been achieved in GUVs to quantify ion permeability of membranes. | 192 | |
Membrane phase behaviour | Fluorescent lipid partitioning | Imaging the partitioning of dyes between coexisting lipid phases in phase separated membranes allow the study of membrane phase behaviour and the construction of phase diagrams of lipid mixtures. | 193 |
Lipid packing and hydration | Laurdan fluorescence | The generalised polarisation function of Laurdan obtained from ratiometric imaging of two environment-sensitive emission maxima reveals information about changes in the local structure and hydration of the membrane. | 165, 194, 195 |
Membrane binding | Fluorescence localisation imaging | Imaging of fluorescent membrane-active compounds (e.g. proteins) can be used to semi-quantitatively investigate their binding at GUV membranes, including relative binding to different coexisting membrane phases or curvatures. | 49, 52, 99 |
Flow cytometry | The colocalisation of fluorescent proteins or particles with GUVs in flow cytometry experiments can be used to quantify the kinetics and extent of adsorption. | 196 |
Methods for preparation of GUVs are summarised in Fig. 1. The most popular method for GUV production is the electroformation method.54,55 Lipid films are dried from chloroform stock solutions onto Pt wire or ITO-coated glass electrodes. After rehydration with an aqueous solution (often a sucrose solution), an a.c. electric field protocol is applied to the electrodes for 1–2 h. The initial sine wave a.c. frequency usually ranges between 10–20 Hz and many protocols gradually reduce this frequency in a stepwise fashion during the electroformation procedure. The amplitude of the a.c. field that is used usually varies between 1.5–5 V, with the lower end of this range more appropriate when charged lipids form part of the lipid mixture. While electroformation usually produces a high yield of high quality GUVs, disadvantages of this method are that high salt or high (∼>20 mol%) anionic lipid content significantly suppress GUV yield or prevent their formation altogether. However it should be noted that this technique works very well for phosphocholine-rich membranes, the most predominant lipid type in Eukaryotic membranes. Physiological salt buffer can be added to the extravesicular medium after electroformation is complete. Protocols using a higher frequency a.c. field have been reported that facilitate GUV formation in physiological salt buffers.56 Alternatively, the gentle hydration method can be used to form GUVs, but this method gives a lower yield of high quality GUVs.57 Recently formation of GUVs assisted by a polyvinyl alcohol gel support has been reported that permits a wide range of lipid compositions and buffer conditions to be used.58
Fig. 1 Overview of some of the GUV formation methods. (a) electroformation; (b) emulsion-transfer method; (c) microfluidic double emulsions; (d) microfluidic jetting. |
A further disadvantage of these GUV preparation methods is the challenge of encapsulation of different chemical entities inside GUVs compared to those in the extra-vesicular aqueous solution. Upon forming the GUVs in the presence of compounds required for encapsulation, similar concentrations will be present inside and outside the vesicles. Separation of unencapsulated compounds from the sample is challenging, with tradition approaches used for LUVs, such as size exclusion chromatography or dialysis, proving to be inefficient. More recent developments in GUV fabrication methods use emulsion-based assembly protocols and allow more efficient specific encapsulation of compounds. Emulsion phase transfer methods are based on the transfer of lipid stabilised water-in-oil droplets through an oil–water interface, where it picks up a second lipid monolayer, resulting in formation of a completed GUV in water with encapsulated contents.59 This method can also be adapted to produce asymmetric membranes.60 Microfluidic lab-on-a-chip methods are being developed for GUV fabrication, in particular using flow-focussing droplet microfluidics geometries.61,62 Recently droplet microfluidics and emulsion phase transfer strategies have been combined into a single on-chip GUV fabrication protocol.63 Microfluidic jetting has also been reported using DIBs as a starting point for GUV formation; jets of fluid cause GUVs to break away from the DIB.64 This method can be used for high encapsulation efficiency and formation of asymmetric membranes.65 DIBs in conjunction with the emulsion phase transfer method have been used to create multicompartment GUVs.66,67
Fig. 3 Cellular fission and fusion events in trafficking of cargo via membrane compartments. Endocytosed cargo is enveloped in clathrin coated vesicles (red), with the aid of fission proteins (BAR and ENTH domains). Clathrin-coated vesicles are incorporated into early endosomal structures. The tubules of endosome membranes are shaped by molecular motors (kinesin, dynein), while moving along microtubule (blue arrows) towards the centre of the cell. Some vesicular cargo is also recycled to the cell surface. During movement, the endosomes ‘mature’ into multivesicular bodies (MVB), changing their shape and protein composition (yellow area). MVBs are generated by accumulation of cargo destined for degradation by the lysosome (pink area) via membrane budding and scission events performed by the ESCRT complexes. The trans-Golgi network (TGN) is another sorting station for vesicular cargo that utilises COPI protein coats for membrane scaffolding. This figure is a derivative of Fig. 1 from ref. 68, used under CC BY 3.0 licence. |
Fig. 4 Mechanism of membrane fusion and fission by SNARE complexes. (a) Tetrameric coiled coil structure of an assembled neuronal SNARE complex involved in synaptic exocytosis (pdb code 1SFC). (b) Complementary v- and t-SNARE coiled coils sitting on donor vesicle and target membranes ‘zipper’ to bring the bilayers closer and initiate ‘fusion pore’ formation. The individual α-helical coiled coil chains in (a) and the respective cartoons representations in (b) are identically colour coded. |
The first observation of content mixing was seen using an in vitro fluorescence-based assay using liposomes.72 This work suggested SNARE proteins form the minimum machinery required for fusion. SNARE complexes are specific, as v-SNAREs do not interact with other v-SNARE containing liposomes or protein free vesicles. The interaction occurs via a unique hairpin-like structure formed by the SNARE complex, named SNAREpin, and is similar to structures formed by some viral fusion proteins.73 The Weber paper has been cited ~2000 times since its publication in 1998 and received much attention, not only due to the cell biology questions it answered, but also the prospect of studying membrane fusion in a fully reconstituted system.
Assembly of a stable SNARE complex occurs spontaneously in solution when using free SNAREs and is energetically favourable. In cells this is controlled by a series of regulatory proteins, but can formation of the SNARE complex and fusion be controlled in synthetic cell systems? SNARE complexes can assemble before fusion and these unfused intermediates accumulate at lower temperatures, therefore something other than high concentrations is required to induce fusion. It was noticed that prefusion ‘priming’ steps occur more slowly in a cell free system, possibly as the proteins need to topically rearrange themselves into a circular array to give rise to the ‘fusion pore’ (Fig. 4b) that initiates fusion.74,75 Some viral proteins behave in a similar fashion, forming pre-fusion intermediates at low temperatures that only fuse once the temperature has increased or been triggered by a low pH (such as within a late endosome).76
Lipid bilayer nanodiscs were used recently to demonstrate that one SNARE only is required for maximum rates of membrane fusion, but release of membrane contents on physiological timescales requires zippering of multiple SNAREs into SNAREpin structures, suggesting that SNARE complexes prevent the fusion pore from resealing during cargo release.77 SNARE-driven fusion in liposomes occurs at rates comparable to physiological fusion following removal of an inhibitory N-terminal domain in t-SNAREs, without affecting the rate of the v- t-SNARE complex formation itself. Full length v-/t-SNARE reconstituted liposomes in the presence of Ca2+ also enables membrane fusion at near physiological rate, confirming the role of Ca2+ as a regulatory factor in fusion.78
The size of the lipid vesicles and the lipid composition itself may also affect the efficiency of fusion. For instance, recently we have learned that for physiological concentrations of SNARE proteins to perform efficient membrane fusion there is a specific requirement for small neutral lipids that do not form bilayer structures (i.e., phosphatidylethanolamine). In the absence of these types of lipids, vesicles can still dock and form trans-SNARE complexes but cannot remodel membranes for fusion.79
It might be expected that in a synthetic system with minimal machinery that fusion would occur more slowly than in vivo due to the absence of other components that could facilitate fusion, for example SNAP and NSF. NSF, stimulated by SNAPs, use ATP hydrolysis to dissociate the stable SNARE complex and seem to allow further rounds of fusion to occur. The possibility remains that they also act to regulate fusion, potentially through the inhibitory N-terminal domain of t-SNAREs. It is generally agreed that fusion occurs through a fusion pore, more recently termed a ‘porosome’, whereby several SNARE complexes form at the site where fusion is to take place and arrange themselves into a ring structure giving the membrane a dimpled appearance.80 The detailed structure of porosomes has been revealed using reconstituted fusion pores in large unilamellar lipid vesicles (LUVs) and examined by transmission electron microscopy (TEM). This work revealed a basket-shaped morphology with a SNARE complex ring at the base of the fusion pore and also that the porosome was still functional.81
Measurement of content mixing of large biomolecules within vesicles can be elegantly performed by incorporation of fluorescently labelled DNA hairpins inside vesicles. Using FRET to quantify real fusion events, a decrease in FRET signal would occur when a labelled DNA strand encounters a complementary strand from another vesicle.82 This approach discounted instances of vesicle aggregation or vesicle rupture, which can obscure actual membrane fusion, and allows a highly reliable method for detecting fusion. Adopting this method showed that SNARE complex alone is the minimal machinery required for fusion pore formation and expansion (large enough to allow cargo transfer up to 11 kDa).
An interesting characteristic of SNARE proteins is their preference for phase-separated lipid bilayers, or lipid ‘rafts’, as it was suggested from in vivo work showing SNAREs' preference for cholesterol-rich regions. Lipid rafts represent liquid-ordered regions of the membrane, rich in saturated lipids and cholesterol, which act as platforms for an array of interactions concerned with protein intracellular trafficking. These micro-regions are more ordered and tightly packed than the surrounding bilayer, but are largely inaccessible to direct observation in vivo. However, it was found that SNAREs in fact have a higher affinity for liquid-disordered regions of membrane83 as demonstrated by monitoring the localisation of fluorescently-labeled SNAREs reconstituted in phase separated GUVs.
Intriguingly, enveloped viruses also use membrane fusion machineries to enter a cell. These machineries are functionally similar to, but mechanistically distinct from the SNAREs. In analogy to SNAREs, coiled-coil structures are pivotal in the fusion mechanism of many viral proteins, and the formation of a low energy membrane structure via the fusion pore. Typical examples are gp41, a protein that adopts a functional six helix bundle allowing HIV-1 fusion with the cell membrane,84 and influenza hemagglutinin, which undergo pH-driven conformational change to form a fusion-competent coiled coil trimer.85 It has been common to study the membrane interaction of viral fusion peptide fragments in reconstituted systems to demonstrate the mechanistic diversity between different viral fusion machineries.86,87 However, an example of functional reconstitution of a full fusion protein has been reported with the assembly of the full GPC envelope glycoprotein of Hemorrhagic-Fever Arenavirus into proteoliposomes. In this study, pH-dependent fusion was demonstrated using fluorescence lipid-mixing and content-mixing assays.88
Fig. 6 Mechanism of membrane fusion and fission by BAR domains (a) structure of the N-BAR domain of hBRAP1/B1N2 showing the curved shape of the dimer (pdb code 4I1Q). (b) BAR domain sensing membrane curvature through its structure and shaping further bending via oligomerisation. The colour coding of protein chains represented by the crystallographic structures is identical to colour coding used for the protein subunits in the cartoon representations. |
During BAR domain interactions with the membrane, lipid clustering can occur.92 The idea of proteins preferentially binding to liquid ordered phases, namely ‘lipid rafts’, was previously explored in experiments conducted with SNAREs. Phosphoinositide-clustering is an effect observed for mammalian I-BAR domains96 but is also displayed by BAR and F-BAR domains, which effectively ‘freeze’ the dynamics of phosphoinositides and, in turn, whole regions of the membrane rich in these acidic lipids. The link between sensing curvature and inducing bending in membranes was an active role proposed for protein density.97 It is possible to create an environment where membrane tension, curvature and protein density can be systematically controlled, using GUVs as a membrane model system. Optical tweezers can be used to create nanotubes with different shape and sizes by pulling of membranes from GUVs. Protein density at the GUV surface and tubules can be measured by fluorescence, whereas the pulling force and tube radius gives a measure of curvature and tension. The radius of the tubules is directly proportional to membrane tension in the GUV. When the interaction of amphiphysin (an N-BAR protein enriched at clathrin coated necks during endocytosis) with membranes is analysed vs. degree of curvature and tension, simultaneous roles of BAR proteins as curvature sensors and inducers are revealed.50 For instance, low densities of amphiphysin sense but do not significantly induce curvature. At high densities, amphiphysin strongly deforms membranes via the formation of a scaffold (Fig. 6b). Similar behaviour was also observed previously for dynamin, a GTPase involved in membrane fission and for endophilin, an endocytic protein enriched at neural synapses.98,99
Some BAR domain containing proteins can be regulated by autoinhibition such as in the case of pacsin. Pacsin is a F-BAR protein found primarily in neurons that can be studied in vitro using liposomes enriched with Folch lipids (a type of lipid found primarily in brain tissue for signalling). The interaction between pacsin and dynamin (an F-BAR containing protein) demonstrated how different BAR domain-containing proteins can efficiently regulate each other. It has been shown that pacsin exists in an autoinhibited state in vitro and in vivo.100 Full length pacsin binds with less affinity to the membrane than its F-BAR domain alone and also prefers less curved surfaces. In comparison, endophilin, an N-BAR protein does not show any significant difference in membrane deformation activity between full-length endophilin and its N-BAR domain alone. This evidence suggests that N-BAR proteins may induce curvature by a different mechanism to F-BAR proteins, possibly such as protein insertion and that relieving F-BAR proteins autoinhibition is a mechanism to regulate membrane deformation.
In contrast to BAR domains, ENTH domain structures are not naturally curved but utilise an N-terminal amphipathic helix to ‘wedge’ into membranes promoting deformation and even tubulation at high concentrations. Epsin NH2-terminal homology (ENTH) domain is a membrane-binding motif first identified in the endocytotic protein, epsin. The ENTH domain can bind to certain acidic lipids, particularly phospholipids, including PI(4,5)P2 and PI(1,4,5)P3. ENTH-domain proteins are involved in clathrin-mediated membrane budding, whereby a clathrin protein coat forms around budding vesicles. In particular, ENTH proteins can recruit clathrin adaptors to the site of membrane budding enabling regulation and organisation of clathrin on membranes. In vitro, isolated ENTH domains promote fission of vesicles when added to lipid bilayers but cannot act in concert like BAR domains and facilitate membrane fission generated by dynamin.101
The ESCRT machinery consists of four distinct complexes generally acting in sequential mechanism. ESCRT-0 and ESCRT-I perform specific cargo recognition and concentration at sites of MVB intraluminal vesicle formation, whereas ESCRT-II interacts with and anchors to acidic lipids in the endosomal membrane allowing ESCRT-III to form an organised structure on the membrane surface at the site of fission and initiate budding102 (Fig. 7). ESCRT-III is the core membrane scission machinery that performs membrane fission. It creates filamentous circular arrays from polymerisation of its subunits, delineating spirals narrowing in size and ultimately generating variable size necks for nascent vesicles.103
Much is known about the structure of the filament array of ESCRTs and how they sequester cargo into intraluminal vesicles, however it is only recently that reconstitution of MVB formation has been successfully reproduced. Elegant work using GUVs demonstrated a coordinated assembly of ESCRT complexes on phospholipid capable of inducing a region of negative curvature that initiates budding and stabilises the neck of a nascent vesicle. Ultimately, narrowing of the vesicle neck by gradual disassembly of the complex drives intraluminal vesicle fission. In analogy to BAR domains, ESCRT-III proteins exist in an autoinhibited state, which has to be released for initiating complex assembly. The enzyme VPS4 provides the mechano-chemical force necessary to activate ESCRT-III for assembly on the endosomal membrane, disassembly of the complex and ultimately recycling ESCRT subunits, returning them in the monomeric inhibited state. This assembly-disassembly cycle allows for further rounds of fission to occur thereby generating continuous cargo encapsulation.104–106
So far, we can only predict via computational modelling how ESCRT-III polymers generate inward budding in membranes. In the ‘dome’ model, ESCRT-III polymers assembled into a three-dimensional spiral define the necks of a nascent vesicle and provide the physical force to push the membrane.107 In the ‘coatless’ model, ESCRT complexes localise only at the neck region of nascent membrane buds and bend the membrane aided by inducing lateral lipid phase separation.108–110 Intriguingly, the high redundancy of ESCRT-III in Nature suggests that varying the content of different ESCRT-III subunits may give origins to different polymeric structures,111 possibly generating vesicles of different sizes.
Experimental approaches employing supported lipid bilayers and liposomes have helped to shed light on the mechanism of membrane shaping through actin polymerisation. An increasing number of in vitro studies have demonstrated how to successfully reconstitute actin networks inside and outside liposomes112 or on lipid bilayers.113 A reconstituted actin cytoskeleton can be coupled to a vesicle's interior to provide mechanical shape and stability to the membrane compartment. For instance, spectrin, a cytoskeletal protein component of erythroid cells, can be reconstituted as a shell at the outer membrane of GUVs by addition of ATP, providing a rigid framework for vesicle shape.114 Co-encapsulation of myosin motors with actin filaments induces contraction of the meshwork, ultimately reproducing membrane contractility.115In vitro reconstitution studies of actin shells anchored outside or inside liposomes have revealed how the strength of anchoring to membranes controls cortical actin remodelling,116 providing a simple model for polarity of actin shells and mechanistic basis for a cell's movement. When actin layers are assembled and anchored outside liposomes, the action of molecular motors generates cortical tension, which results in opposite effects on the membranes. High tension causes cortical actin to retract to one side of the liposome, whereas low tension coupled with strong anchoring between actin and membranes, results in vesicles shrinkage. In a model where an actin shell is assembled and anchored inside liposome vesicles, actomyosin filaments retract away from the membrane if the strength of anchoring is weak, whereas they are pulled towards the membrane when the attachment of the cortex is strong.
The cytoskeleton is also crucial for facilitating membrane scission in cell division events. For instance, a reconstituted minimal complex of a bacterial homologue of tubulin (FtsZ), and a single accessory protein (FtsA) has been shown to be competent for cell division when incorporated into liposome vesicles.117 The beautiful simplicity of the bacterial division machinery demonstrates that the main requirement for membrane severing is an organised assembly of proteins, without the need for excess membrane.
However, cytoskeletal cortices do not only play an active role in moulding a cell's plasma membrane but they define shape changes of intracellular membrane compartments. Membrane tubulation is an event initiated either by proteins that can scaffold positive membrane curvature (such as COPI or BAR proteins), or generated by the pulling forces of molecular motors attached to localised regions of the organelle's membrane. Growth and retraction of membrane tubules from larger organelles is another mean of cargo transport in the membrane trafficking system of a cell, in addition to vesicles. This dynamic motion of tubular structures is due to the presence of molecular motors that can walk along the fast growing tip of cytoskeletal microtubules (plus end motors, e.g. kinesins) or in the opposite direction (minus end motors, e.g. dyneins). Reconstitution of membrane tubule ‘pulling’ processes has been reproduced using both cell extracts and minimal components.118 An alternative means to generate tubulation is to ‘push’ from the inside of a vesicle. Encapsulated microtubule components can create long protrusions when polymerisation is induced.119
felastic = fbend + fstretch | (1) |
The stretching term fstretch = τ, where τ is the surface tension of the membrane. This term is often neglected as the energy cost of stretching a lipid membrane is much greater than that of bending the membrane. Therefore, in many cases the stretching of the membrane is not considered to be significant. Membrane stretching can also be described by an elastic area expansion modulus (KA), where the change in free energy per unit area upon stretching the membrane Δfstretch = (KA/2)[ΔA/A0]2, where ΔA is the change in membrane area and A0 is the initial membrane area.
Typical values for KA for lipid membranes have been found to be in the range 0.1–0.2 J m−2, or 25–50kBT nm−2.122 It should be noted that lipid membranes are not particularly stable under stretching deformations; they rupture upon fractional area increases (ΔA/A0) between 2–5%.123–125
The bending contribution to the free energy of a membrane is therefore the most significant. This can be described by two mechanical moduli:
(2) |
It has been shown that values for the Gaussian modulus κG are constrained by the bending modulus κb such that −1 ≤ κG/κb ≤ 0.7κG is challenging to determine directly from experiments, but the few experimental studies of κG in amphiphile systems have found that κG ≈ −0.8κb.126,127 Conveniently, due to the Gauss–Bonnet theorem, the Gaussian curvature term in eqn 2 only contributes to the change in free energy of the system if the membrane undergoes a topological transition (e.g. fusion, fission, poration). The Gaussian elasticity term can therefore be ignored for bending of a membrane of fixed topology, i.e. when only considering the membrane bending capabilities of these proteins, prior to scission, merger or poration events that may occur later in the membrane remodelling process.
Eqn 2 can also often be further simplified due to the fact that the lipid composition of the membrane is symmetric across the membrane for almost all experimental model membrane systems. When topological changes in the membrane do occur, it is often useful to treat the elastic energies of the two lipid monolayers of the membrane separately with a finite spontaneous curvature, C0. However for bending at fixed topology, the elastic energy of the whole lipid bilayer is considered together and due to the symmetry of the lipid composition in each monolayer, C0 = 0. Therefore the total free energy of model membranes for bending at fixed topology can be simplified to
Felastic = ∫felasticdA = 2κb∫H2dA | (3) |
An example of the application of theoretical models is the modelling of the interaction of F-BAR domains with lipid membranes.134 Combining simulations and membrane elasticity theory predicts an initial binding energy of an F-BAR dimer to the membrane of 4kBT, which reduces to 2.3kBT at equilibrium during membrane tubulation. Therefore this shows that single F-BAR dimers at low density on a membrane would not individually be able to generate enough energy to bend a membrane with a bending modulus of 20kBT. Multiple F-BAR dimers would be required to act in concert at high density on a membrane in order to manipulate its curvature, in agreement with experimental observations reviewed above.
Ultimately, it should be noted that although the Helfrich model considers the membrane to be a continuous sheet with no molecular details. However, the membrane mechanical moduli in this model can be directly related to the intermolecular interactions between lipids by considering the lateral stress profile in the membrane.135,136 This can allow consideration of protein-induced changes in the mechanical moduli of membranes due to the ways in which proteins modulate the interactions between lipids, for example by direct insertion of part of the protein complex into the membrane.137
Finally, it is worth highlighting that computational studies have also proved an extremely practical tool to provide suggestive insights into the molecular-scale mechanisms of membrane remodelling processes, and interactions between membranes and protein complexes. While the focus of this perspective is the in vitro reconstitution of membrane remodelling machineries, there is a wealth of literature in which computer simulation approaches have provided new mechanistic ideas impossible to gather by traditional experimental work. In addition to the work on the F-BAR domain discussed above, important contributions to the possible molecular mechanisms underlying membrane remodelling include simulations of the membrane interactions of N-BAR domains138,139 and SNARE proteins.140,141
Engineering of a minimal machinery that performs membrane fission in a controllable manner is however challenging. Any curvature-inducing protein system can eventually promote membrane fission at high densities. However, the creation of new vesicles in a controlled manner requires a multistep mechanism including factors to induce initial membrane bending (Clathrin and COPI coats and even actin filaments), entities to stabilise and promote further budding (BAR and ENTH domain-containing proteins) and regulatory factors that promote membrane fission mechanically-driven by nucleotide hydrolysis (dynamin). Since the development of the PURE system for in vitro protein expression, assembling a minimal machinery to recreate cell division using model membranes has moved a step closer. In situ expression of membrane deforming proteins using the PURE system has been shown to produce extensive membrane deformation in liposomes preparations upon the activity of expressed N-BAR domains.150
Initiation of membrane scission in a model vesicle, for mimicking a dividing cell, is a process that must begin from the lumen of the vesicle and requires proteins that can generate narrow necks and induce negative curvature, such as I-BAR proteins or ESCRT complexes. In situ expression of minimal ESCRT machineries inside GUVs or liposomes could be the first step towards recreating membrane-bound compartments competent for cell division. The drawback of using complexes such as ESCRTs is the high redundancy of the functional assembly, making reconstitution of such a system for large scale production (e.g. for biotechnology applications) challenging. Monocellular ancient organisms such as archea bacteria, for instance, utilise ESCRT complexes with a reduced number of components in comparison to yeast or humans, given their simpler metabolism. These ESCRTs modules can be regarded as a source or an inspiration for assembly of ‘divisomes’, or remodelling of membranes with appropriate lipid composition.151
Interesting developments in the design of minimal vesicular transport systems include coupling of liposomes to molecular motors capable of walking on molecular tracks. Liposomes coated with bundled actin have been used to build simple transport systems based on myosin motors.152,153 Complementary DNA strands are an efficient manner to anchor membrane vesicles to a kinesin molecule gliding on microtubules.154 In such system, loading, transport and unloading of fluorescent cargo-loaded liposomes can be reproduced using a flow-cell – tuning of the strength of binding between the DNA strands allows for vesicle unloading. Alternatively, coupling of cargo vesicles can be achieved via streptavidin–biotin linkage and quantum dots coupled to the trailing ends of actomyosin filaments via capping proteins.155
Inspired by the function of natural proteins, several synthetic alternatives have been shown to mimic properties of these proteins at lipid biomembranes. DNA nanotechnology is proving to be one useful tool in this approach.156 DNA–lipid conjugates have been successfully applied to mimic SNARE fusion proteins, membrane adhesion receptors, tetherins and transmembrane pores.157–160,161 Supramolecular chemistry has also provided synthetic organic macromolecules that can, for example, trigger membrane adhesion or poration.162,163 Synthetic nanoparticles may also prove useful mimics of membrane remodeling proteins in minimal systems. While much of the existing work on nanoparticle–membrane interactions was not motivated by biomimicry of natural membrane-active proteins, nanoparticles have been shown to be able to bend and scaffold the membranes of GUVs.164–167
In order to develop advanced minimal cells by bottom-up assembly techniques that incorporate scaffolding, shaping and membrane vesicle trafficking capabilities, we suggest that a process of development described in Fig. 8 should be followed. Firstly, the components need to be characterised using biophysical investigative approaches. These components then need to be stripped back to minimal functional entities to provide maximum simplicity in their design and function. This process will require multiple iterations between protein complex redesign and biophysical characterisation. Thirdly, two or more independent membrane-active components need to be successfully reconstituted into model systems to demonstrate mutual compatibility and minimise unwanted interference between individual components. This will require the simplified, minimal components to behave in a predictable manner in how they interface with other components of the synthetic biology toolbox when combined together in systems of increasing complexity. Finally, incorporating metabolic processes within the compartments of synthetic cells can further increase complexity and functionality.
Nature shows us how control of dynamic chemical compartmentalisation is required to regulate a large number of parallel processes within composite cellular systems; an important feature that synthetic biologists will wish to harness and exploit. The pathway to realising functional cell-like materials has become much clearer over the past 5–10 years with many significant advances in the understanding, design and application of in vitro model membrane–protein interaction systems. Now the next steps need to be taken in establishing minimal standardised parts and tools to generate compartmentalisation, and show that they can be successfully incorporated into organised functional multicomponent systems of increasing complexity.
Footnote |
† Synthetic biology is defined as “(a) the design and construction of new biological parts, devices and systems and (b) the re-design of existing natural biological systems for useful purposes.” |
This journal is © the Owner Societies 2015 |