Every membrane measurement made outside a cell is made in a substitute. The substitute has a lipid composition you chose, a curvature you only partly chose, an edge that does not exist in a real membrane, and in most cases no enclosed interior. The useful skill is not picking the best mimetic but knowing which property of a real membrane each construct discards, then checking whether your readout depends on it.

The four constructs

Liposomes are closed lipid vesicles, and the only construct here with a sealed aqueous interior, which means the only one that can hold a transmembrane gradient. Size is set by preparation, most commonly by extruding a hydrated multilamellar suspension repeatedly through polycarbonate filters of defined pore size.

Bicelles are mixtures of a long-chain phospholipid and a short-chain one. The long-chain lipid forms a small planar bilayer patch and the short-chain species covers the high-curvature rim, keeping the patch soluble and small enough to tumble fast in solution. Composition is described by q, the molar ratio of long-chain to short-chain lipid.

Nanodiscs replace the short-chain rim with a protein belt: two copies of an engineered amphipathic helical membrane scaffold protein (MSP) wrap the perimeter of a bilayer disc in the conventional double-belt arrangement, a standard structural description of the construct rather than a finding of the work cited here, and so stated uncited. Assembly is conventionally driven by mixing lipid, scaffold and any target protein in a detergent such as cholate and then removing the detergent with adsorbent beads, with the lipid-to-scaffold ratio optimized empirically for each lipid. That protocol is bench practice, not a finding, and is stated uncited.

Polymer-bounded discs replace the belt with an amphipathic copolymer that inserts into a membrane and cuts out a disc of it. Styrene-maleic acid (SMA) copolymers assemble into small bilayer discs encircled by polymer, termed SMA lipid particles or SMALPs [1]. The distinguishing claim is procedural: the protein is purified without ever being removed from a lipid bilayer [2].

Lipid composition and curvature

Composition is the part you control most directly and reproduce most poorly. Two preparations with the same headline mole percentages can differ in what actually assembled, because the species distribution in the final particle need not match the distribution you weighed out. Bicelles supply a measured case, below.

Curvature is the property most often ignored and most often consequential. A small vesicle and a large one made from identical lipid differ in the strain on the outer leaflet and in the packing defects available to a binding partner. In a review of curvature effects on amyloid aggregation, Terakawa and colleagues describe small unilamellar vesicles, with high curvature, and large unilamellar vesicles, with low curvature, as having been demonstrated to show different capabilities to induce nucleation and amyloid formation, and to inhibit it, for amyloid-beta peptides and alpha-synuclein [3]. That is curvature-dependent behavior in vitro, and licenses no statement about what happens in an organism.

A diameter is therefore not a nuisance parameter. If a measurement changes when you switch from extruded 100 nm vesicles to sonicated small ones, the mimetic has entered your result.

Liposomes: the only geometry with an inside

Extrusion is the workhorse: fast, no organic solvent or detergent in the final material, and a reasonably narrow size distribution. Hope and colleagues described repeated extrusion of multilamellar vesicles under moderate pressures, at or below 500 pounds per square inch, through polycarbonate filters of 100 nm pore size, combined with freeze-thaw protocols where required. The product was unilamellar vesicles with diameters in the range 60 to 100 nm and trapped volumes in the region of 1 to 3 microliters per micromole of phospholipid, at lipid concentrations up to 300 micromoles per milliliter and trapping efficiencies up to 30 percent [4].

Note the direction: a 100 nm pore gave vesicles no larger than the pore and often appreciably smaller. Nominal pore size is an upper bound, not a target you hit.

The trapped-volume figure is worth converting, because it sets how much of your sample is encapsulated. Take a 1 mL preparation at 10 mM phospholipid, which is 10 micromoles of lipid. Applying their reported 1 to 3 microliters per micromole gives 10 to 30 microliters of interior volume. The denominator is the total sample volume, 1000 microliters, so roughly 1 to 3 percent of the sample sits inside vesicles. That applies their reported trapped volumes at a concentration of our choosing rather than one the abstract specifies, so read it as an order of magnitude: encapsulation is a small minority of the sample.

What liposomes uniquely reproduce is vectorial behavior. In the same work, extruded soya phosphatidylcholine vesicles loaded with potassium developed, in the presence of valinomycin and a transmembrane sodium-potassium gradient, a potassium diffusion potential exceeding 100 millivolts in magnitude and negative inside, whose decay tracked sodium influx [4]. No disc-shaped mimetic can do this. If your assay is transport, permeability, or anything requiring two distinguishable aqueous compartments, vesicles are the only option in this list.

Bicelles: q is a property of the sample, not of the recipe

Bicelles are attractive for solution NMR because at low q they tumble fast enough to give narrow lines. The difficulty is that q, temperature and total lipid concentration are not independent, and the q you weighed out is not necessarily the q in the particles.

Glover and colleagues studied DMPC/DHPC mixtures at q of 0.5 or below and reported light scattering and electron microscopy data supporting the hypothesis that the aggregates are disc-shaped, with the radius of the planar domain directly proportional to the DMPC/DHPC ratio and inversely proportional to total phospholipid concentration when that ratio was held at 0.5. Only below 130 mM phospholipid did the morphology appear to change [5]. The inverse concentration dependence is the counterintuitive part: diluting a bicelle sample makes the particles larger, not smaller.

Ye and colleagues examined the same system by dynamic light scattering and cryo-transmission electron microscopy across temperature, concentration and q. They reported that stable DMPC/DHPC bicelles form only at low q, specifically 0.5, and that mixtures with q above 0.5 appeared multidisperse, with one component of hydrodynamic radius around 2.5 nm and another of 6 to 8 nm, interpreted as coexistence of small, possibly mixed micelles with much larger bicelles. Incubation at 37 degrees Celsius increased the phase separation, and stability and size at q of 1 and 1.5 depended strongly on temperature and total lipid concentration [6].

Beaugrand and colleagues addressed composition directly by phosphorus-31 NMR over 2 to 400 mM total lipid and q from 0.15 to 2, showing that dilution modifies the actual DMPC/DHPC molar ratio in the bicelles. Their caution is explicit: care must be taken at total lipid concentrations at or below 250 mM, especially at q around 1.5 to 2, because moderate dilution there could produce large, slow-tumbling structures that hinder solution NMR, circular dichroism or dynamic light scattering [7].

On lipid miscibility the literature does not agree. Glover and colleagues reported the DMPC and DHPC components highly segregated across q of 0.05 to 0.5 at 15 and 37 degrees Celsius [5]; Beaugrand and colleagues, supported by infrared spectroscopy and molecular dynamics, reported that the phospholipids are largely segregated only when q exceeds 1 [7]. The two groups used different probes and concentration ranges. Rim-to-core segregation in a fast-tumbling bicelle is method-dependent and not settled; presenting either result as the consensus would overreach.

Nanodiscs: size is set by the belt

The nanodisc's design virtue is that diameter is engineered rather than emergent. Denisov and colleagues customized particle size by changing the length of the amphipathic helical region of the scaffold protein, obtaining, with different scaffolds and DPPC, average particle sizes from 9.5 to 12.8 nm with a very narrow size distribution of about plus or minus 3 percent by HPLC and small-angle X-ray scattering. Their scattering analysis confirmed a lipid bilayer of 5.5 nm thickness across the different disc sizes, and N-terminal functionalization of the scaffold did not prevent homogeneous disc formation [8].

A caution on comparing numbers across reports: a later study from the same group described scaffolds defining the diameter of the nanodisc lipid bilayer from 76 to 108 angstroms, with 164 to 335 DPPC molecules per disc [9]. That is not the same quantity as the 9.5 to 12.8 nm particle size above — one describes the lipid core, the other the particle — so check which was measured before treating two reported diameters as comparable.

Those lipid counts are internally consistent. The diameters differ by a factor of 108 divided by 76, about 1.42, so the bilayer areas differ by about 2.0; the reported lipid counts differ by 335 divided by 164, about 2.04. The lipid census tracks the disc area, as a flat bilayer patch should.

Other routes to size control

Scaffold length is not the only lever. Johansen and colleagues designed a solubility-enhanced scaffold based on MSP1E3D1, carrying a sortase recognition motif and additional negative charges, and circularized it by covalently linking the termini. Circularized and non-circularized discs were structurally similar by small-angle X-ray scattering and size-exclusion chromatography, while the solubility-enhanced discs were more stable than the MSP1E3D1 template and circularization improved stability further; a membrane protein was efficiently incorporated into the circularized discs [10].

A different strategy abandons a fixed belt length. Lyons and colleagues describe saposin A based scaffolds, in which the saposin-lipid scaffold adapts to various sizes of transmembrane region during self-assembly to form a minimal lipid nanoparticle [11]. Adaptive sizing trades the monodispersity that makes MSP discs attractive for the ability to accommodate targets of unknown footprint.

The rim is part of your sample

Whatever bounds the disc contacts the lipid, and that contact is measurable. Denisov and colleagues studied thermotropic phase transitions of DPPC and DMPC in nanodiscs by small-angle X-ray scattering, differential scanning calorimetry and Laurdan generalized polarization. In nanodiscs of all sizes they examined, the transitions were broader and shifted to higher temperatures than in vesicle preparations, and the size dependence of the transition enthalpies indicated a boundary lipid layer in contact with the scaffold protein, estimated at approximately 15 angstroms thick, or two lipid molecules [9].

A boundary layer of fixed thickness occupies a larger share of a small disc than a large one. Treating the disc as a circle and the annulus as uniform — geometry applied here to their dimensions, not a calculation the authors presented — a 15 angstrom annulus on a 76 angstrom disc leaves an inner core of radius 23 angstroms inside an outer radius of 38, making the annulus about 63 percent of the area. On a 108 angstrom disc it is about 48 percent. Even at the large end, roughly half the lipid is perimeter lipid. A nanodisc is not a small piece of bulk membrane; it is mostly edge.

Polymer rims interact with lipid too, and differently depending on chemistry. Colbasevici and colleagues compared discs encased by aromatic styrene-maleic acid with those encased by aliphatic diisobutylene-maleic acid, using EPR on phosphatidylcholine spin labels nitroxide-labeled at carbon 5, 12 or 16 together with coarse-grained molecular dynamics. Lipid dynamics were less constrained in the DIBMA particles, and the affinity of spin-labeled lipids for the polymeric rim was more pronounced with SMA [12].

Detergent-free reconstitution and its limits

The appeal of polymer extraction is that the protein is never stripped of its bilayer. SMA can be applied to proteins from any host source and permits purification without the protein ever being removed from a lipid bilayer [2]. Unger and colleagues review what this has yielded: cryo-EM of SMA-extracted proteins has resolved annular lipids, lipids within central cavities and binding pockets, regions stabilizing multimers, and terminal residues that would otherwise remain unresolved; functionally, it has allowed analysis of proteins unstable in detergents [13].

The limitations are chemical and specific. Both that review and a direct comparison identify sensitivity to low pH and to divalent cations as the notable constraints on SMA [1, 13]. Gulamhussein and colleagues compared SMA with DIBMA and found DIBMA able to solubilize several different membrane proteins from different expression systems, though with lower yield and lower purity than SMA for some; DIBMA-extracted G protein-coupled receptors retained ligand and G protein binding; DIBMA particles were larger than SMALPs and showed decreased sensitivity to magnesium, but appeared less stable, which the authors linked to the larger particle size, while noting the same size may offer a less rigid lipid environment more amenable to protein dynamics. They conclude that the optimal choice of polymer depends on which features of a protein are to be investigated [1]. That is a trade, not a ranking.

What each model does not reproduce

  • No disc has an inside. Bicelles, nanodiscs and polymer particles cannot sustain a transmembrane gradient or support a vectorial transport assay. Only vesicles can [4].
  • Every disc has an edge in contact with the lipid — short-chain lipid [7], scaffold protein [9] or polymer [12] — and in nanodiscs the reported 15 angstrom boundary layer [9] works out, on the annulus geometry applied above, to a large share of the total lipid.
  • Bicelle composition drifts with dilution and temperature, so q is a sample property to be verified, not a recipe constant [6, 7].
  • Nanodisc lipid phase behavior is not vesicle lipid phase behavior: transitions are broader and shifted upward [9].
  • None of these constructs reproduces the leaflet asymmetry, protein crowding or continuous lateral extent of a cell membrane — a structural fact about the constructs rather than a published finding, so stated uncited.
  • Polymer particles inherit chemical sensitivities the protein does not have: SMA is constrained by low pH and divalent cations [1, 13].

What to report, and what remains unsettled

Report the mimetic as fully as you report the analyte: lipid identity and mole ratios, the scaffold or polymer and its variant, q or the lipid-to-scaffold ratio, total lipid concentration, temperature, and the sizing method with its result. Size-exclusion chromatography and dynamic light scattering are the routine checks, and which one produced your diameter matters, because they do not report the same quantity. Those conventions are practice, not findings, and are given here uncited.

Several things remain open. Rim-to-core lipid segregation in fast-tumbling bicelles is reported differently by different methods [5, 7]. How far boundary lipid differs from bulk lipid, and how far that difference propagates to an embedded protein, is characterized for a few systems and assumed for the rest. And whether a protein behaves in any of these particles as it would in the membrane it came from is not answered by the work cited here, which characterizes the mimetics rather than validating them against the native state. Choosing among them is, for now, choosing which known distortion you can tolerate.