Regulation of Membrane-Shape Transitions Induced by I-BAR Domains

Abstract

I-BAR proteins are well-known actin-cytoskeleton adaptors and have been observed to be involved in the formation of plasma membrane protrusions (filopodia). I-BAR proteins contain an all-helical, crescent-shaped IRSp53-MIM domain (IMD) dimer that is believed to be able to couple with a membrane shape. This coupling could involve the sensing and even the generation of negative plasma membrane curvature. Indeed, the in vitro studies have shown that IMDs can induce inward tubulation of liposomes. While N-BAR domains, which generate positive membrane curvature, have received a considerable amount of attention from both theory and experiments, the mechanisms of curvature coupling through IMDs are comparatively less studied and understood. Here we used a membrane-shape stability assay developed recently in our lab to quantitatively characterize IMD-induced membrane-shape transitions. We determined a membrane-shape stability diagram for IMDs that reveals how membrane tension and protein density can comodulate the generation of IMD-induced membrane protrusions. From comparison to analytical theory, we determine three key parameters that characterize the curvature coupling of IMD. We find that the curvature generation capacity of IMDs is significantly stronger compared to that of endophilin, an N-BAR protein known to be involved in plasma membrane shape transitions. Contrary to N-BAR domains, where amphipathic helix insertion is known to promote its membrane curvature generation, for IMDs we find that amphipathic helices inhibit membrane shape transitions, consistent with the inverse curvature that IMDs generate. Importantly, in both of these types of BAR domains, electrostatic interactions affect membrane-binding capacity, but do not appear to affect the curvature generation capacity of the protein. These two types of BAR domain proteins show qualitatively similar membrane shape stability diagrams, suggesting an underlying ubiquitous mechanism by which peripheral proteins regulate membrane curvature.

Introduction

Membrane curvature generation by peripheral proteins has been an area of considerable interest in cell biology (1). Such curvature-sensitive proteins include Bin/amphiphysin/Rvs (N-BAR) domains with a concave membrane-binding interface and an amphipathic membrane-inserting helix at the N-terminus, and IRSp53-MIM domains (IMDs) with a convex membrane-binding interface. N-BAR domains can bind to and reshape membranes both in vitro and in vivo through scaffolding involving their crescent-shaped dimeric structure and wedging through membrane insertion of the N-terminal helix (2–4). IMDs are distantly related to the classical BAR domains. Similar to N-BAR proteins (such as endophilin and amphiphysin), IMDs exist as stable zeppelin-shaped homodimers in buffer solution (5–7) and are able to reshape plasma membranes (8). Differences were observed comparing the effect of IMDs and N-BARs in inducing membrane curvature. First, IMDs induce tubules with larger diameters than N-BAR domains, which is likely due to the lower curvature of the bent dimeric structure of IMDs (8,9). Second, IMDs induce the opposite membrane curvature compared to N-BAR domains. Consistently, in the in vitro experiments, IMDs were shown to bind and induce inward tubulation when coincubated with PI(4,5)P₂-rich liposomes (8,10). Contrarily, N-BAR domains such as endophilin, induce the formation of outward tubules from liposomes (3).

Several proteins are known to contain IMDs. Missing-in-metastasis (MIM) protein (along with actin-bundling protein with BAIAP2 homology (ABBA) and insulin receptor tyrosine kinase substrate (p53IRSp53)) belongs to a family of actin-binding scaffold proteins that can regulate actin polymerization (11,12). MIM was originally reported as a potential metastasis suppressor because it is missing in metastatic bladder-cancer cells (13). MIM, ABBA, and IRSp53 contain homologous IRSp53-MIM domains (IMDs, also known as I-BAR domains) at the N-terminus (5,14). N-terminal IMDs can bundle actin filaments (14) and induce membrane deformations (15). In addition to the IMDs, I-BAR proteins contain protein interaction domains such as SH3 and WH2 domains (16,17).

In vivo, outward protrusions of the plasma membrane could be observed when IMDs were overexpressed in cells (10,12,14,15,18–21), while N-BAR-domain-containing proteins are involved in invagination events, such as clathrin-mediated endocytosis (22). While these peripheral membrane proteins are well known as membrane curvature inducers, as of this writing there is a lack of mechanistic insight into the curvature initiation process induced by these proteins.

Plasma membrane curvature changes involving both protrusions and invaginations, are accompanied by plasma membrane surface area changes. Maintaining an appropriate cell surface area is a task required for the continuous function of a cell. Therefore, both the positive curvatures induced by BAR, F-BAR, and ENTH domains, and the negative curvatures induced by IMDs, are likely to be essential in regulating the total surface area of the plasma membrane (8,21).

Membrane tension is also believed to be a central factor in plasma membrane area homeostasis (23). Plasma membranes have lateral tensions ranging from several μN/m to several hundred μN/m (23–25). In living cells, membrane tension is mainly regulated through intracellular osmotic gradients and membrane-cytoskeleton interactions (23,26). Recently, an increasing number of studies have focused on the role of membrane tension in regulating various cellular processes, such as endocytosis and exocytosis (27–30), mechanochemical and biochemical signaling (31,32), and cytoskeletal remodeling (31,33,34).

In this study, we employed a membrane-shape-stability assay based on giant unilamellar vesicles (GUVs), which uses a decrease in GUV membrane area as an indicator for membrane-shape transitions (35). Our goal is to quantitatively describe the ability of IMDs to induce membrane curvature by determining the protein-number density required for initiating membrane tubulation on GUVs. We correlate this transition density of the protein with the membrane tension of GUVs, and investigate lipid composition and ionic-strength effects on transition densities. From our measurements, we obtain a membrane-stability diagram, which separates stable and unstable regions, depending on membrane tension and protein density. The stability diagram can be well fitted with a theoretical model (36,37). We further compared the membrane-curvature induction abilities of different IMDs and confirmed a significant effect of N-terminal membrane insertion on membrane-shape stability.

This study provides quantitative insights into the biophysics of membrane-protrusion processes and improves the understanding of how the formation of plasma-membrane protrusions might be regulated.

Materials and Methods

GUVs and protein preparation

The lipids DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine), DOPS (1,2-dioleoyl-sn-glycero-3-phospho-L-serine), DOPE (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine), and PI(4,5)P₂ (L-α-phosphatidylinositol-4,5-bisphosphate (Brain, Porcine) (ammonium salt)) were obtained from Avanti Polar Lipids (Alabaster, AL). Texas Red DHPE (Texas Red-1,2-dihexadecanoyl-sn-glycero-3 phosphoethanolamine, triethylammonium salt) was obtained from Invitrogen/Life Technologies (Grand Island, NY). GUVs were prepared by the electroformation method (38–42) in 300 mM sucrose with two alternative lipid compositions: 1), 45% DOPS, 30% DOPE, 24.5% DOPC, and 0.5% Texas Red DHPE; and 2), 30% DOPS, 30% DOPE, 34.5% DOPC, 5% PI(4,5)P₂, and 0.5% Texas Red DHPE. GUV compositions were chosen to mimic the innerleaflet-headgroup composition of plasma membranes (43). Plasmids encoding the N-terminal IRSp53/MIM domains (IMDs) of mouse missing-in-metastasis (MIM/IMD, residues 1–254); mouse actin-bundling protein with BAIAP2 homology (ABBA/IMD, residues 1–249); human-insulin-receptor tyrosine kinase substrate p53 (IRSp53/IMD, residues 1–250); and variants of MIM/IMD, specifically MIM/IMD with residues 1–11 deleted (MIM/IMD D1–11) as well as an EGFP-fused MIM/IMD (MIM/IMD-EGFP) were kindly provided by P. Lappalainen (University of Helsinki, Helsinki, Finland). These IMDs were expressed as His-tag fusion proteins in BL21(DE3) RIL CodonPlus bacteria (Stratagene, La Jolla, CA), and purified with a Q-Sepharose high-performance anion exchange column (GE Healthcare, Mickleton, New Jersey) (8). The protein buffer contained 20 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid; Sigma-Aldrich, Allentown, PA), 150 mM NaCl (sodium chloride; Fisher Scientific, Philadelphia, PA), and 1 mM TCEP (Tris (2-carboxyethyl) phosphine; Pierce/Thermo Fisher Scientific, Philadelphia, PA). The non-EGFP-fused proteins used for fluorescence-microscopy imaging were labeled with Alexa Fluor 488 (AF488) C5-maleimide (Invitrogen, Carlsbad, CA). The labeling efficiencies for MIM/IMD, ABBA/IMD, IRSp53/IMD, and MIM/IMD D1-11 were determined to be 52, 42, 33, and 5%, respectively. We verified absence of any effects of labeling on protein function (see Fig. S4 in the Supporting Material). The protein concentrations used throughout the article refer to the total-subunit concentration. The protein densities on GUVs refer to dimer-number-versus-area density.

GUV stability assay

To investigate the membrane-deformation ability of the I-BAR domains, we employed a GUV-based membrane-shape-stability assay based on micropipette aspiration (see Fig. 1 A) to study the geometry changes of a single GUV when exposed to IMD protein solution. The techniques involved in this assay are described below.

Figure 1.

MIM/IMD induces shape transition of GUV membrane. (A) Pipette-aspirated GUV and parameters used for calculation of membrane tension. (B) Cartoon showing the sequence of GUV transfer steps: (red), GUV dispersion; (green), protein containing solution. (C) Time-lapse confocal images of a transferred GUV: (top) lipid channel; (lower) protein channel. Membrane tension = 0.076 mN/m. Scale bar = 10 μm. As protein density reached a critical point on the GUV, inward tubulation (indicated by arrows) was initiated, associated with a decreasing projection length. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE.

GUV transfer equipment and operations

As shown in Fig. 1 B, separate GUV dispersion (membrane labeled with red dye) and protein solution (labeled with green dye) were prepared as described in Shi and Baumgart (35). Both solutions were 375 μL in volume and were diluted from stock GUV and protein solutions to designated concentrations by using a buffer containing sucrose (400 mM)/glucose (400 mM)/protein buffer (20 mM HEPES, 150 mM NaCl), 1:1:1. This dilution buffer had an osmolarity ∼20% higher than the GUV stock solution, which ensured that GUVs were sufficiently flaccid to allow pipette aspiration. The solution conditions ensured that vesicle transfer occurred between two solutions of identical composition, except for the presence of protein in the receiving solution. Therefore, any observed changes in GUV geometry can be ascribed solely to protein binding, as opposed to any other changes in solution conditions. The preparation of GUV aspiration micropipettes and transfer capillary tubes were previously described in Capraro et al. (4) and Tian and Baumgart (42). The micropipettes were casein-coated before use to avoid membrane adhesion to the inner micropipette walls. All experiments were carried out at room temperature.

The procedure of transferring a single GUV from the GUV dispersion into protein solutions includes the following steps (35): the zero pressure of the system is carefully adjusted before aspirating GUVs. Next, the aspiration pressure is reduced to a negative value to aspirate a single GUV, and then the membrane tension of the GUV is adjusted to a desired value by adjusting the pipette aspiration pressure. Afterwards, the transfer capillary is manually moved forward to cap the aspirated GUV. The capped GUV is then removed from the GUV solution (red in Fig. 1 B) and inserted into the protein solution (green in Fig. 1 B), upon which the transfer capillary is manually moved backward to expose the GUV to protein solution. Finally, the protein binding and GUV shape transition process is monitored via confocal microscopy imaging as soon as the transfer capillary is removed and the GUV is exposed to protein solution (which is defined as t = 0). Confocal fluorescence imaging (Objective: 60× W 1.1 NA; Olympus, Center Valley, PA) was used to continuously capture protein-density increase on the membrane and to follow GUV geometry changes induced by protein binding. Imaging was continued until the protein density on the GUV reached thermodynamic equilibrium (as defined by the absence of additional changes). The fluorescence intensities thus obtained were converted into an equilibrium protein density.

Data processing

The geometry of the aspirated GUV and the parameters used for the calculation of membrane tension and geometry changes (GUV radius, R_v; micropipette radius, R_p; length of pipette-aspirated vesicle fraction, i.e., the projection length, L_p; and pressure, ΔP) are shown in Fig. 1 A. The software IMAGEJ (National Institutes of Health, Bethesda, MD) was used to measure micropipette radius and projection length, while code written in the software MATLAB (The MathWorks, Natick, MA) was used to determine GUV radius and average fluorescence intensity on the GUV contour (35).

Membrane tension, σ, is determined by the following equation:

$$\sigma =\frac{\Delta P}{2\left(\frac{1}{R_p}-\frac{1}{R_v}\right)}.$$ (1)

The measured protein fluorescence intensity on the membrane was converted into a protein-number-area density via a calibration procedure (44). Briefly, GUVs containing varied amounts of BODIPY-labeled lipids were imaged under identical conditions, yielding a linear relationship between fluorescence intensity and number density of BODIPY dyes. The quantum-yield difference between AF-488 and BODIPY then allowed us to determine the conversion factor between protein fluorescence and density on the membrane. The vesicle membrane area was calculated as

$$\text{Area}\left(t\right)=4\pi R_v{\left(\mathit{t}\right)}^2+2\pi R_p{L}_p\left(t\right).$$ (2)

This area was used as an indicator of the GUV geometry change, and the GUV-shape-instability transition point was determined by the following procedure: to determine the transition density, we first chose several (minimally three) measurement points where the membrane area was observed to be constant, and determined standard deviation (SD) and average value for this set of pretransition points. To rigorously define a threshold for the shape-transition instability, we subtracted 2 × SD from that average value. We then determined the transition area (and time) from linear interpolation using the two area data points immediately above and below, respectively, of the threshold value. Likewise, the transition density was found from the transition time defined above and linear interpolation of the protein-area-density measurements.

It is important to note that the GUV volume remained constant over the course of each experiment (see Fig. S1 for a representative example; also see Shi and Baumgart (35)). This condition is essential for the interpretation of our experimental results because it ensures that protein binding does not lead to membrane-pore formation, because volume changes induced by bulk flow through pores would result in concomitant projection-length changes, which would interfere with our method to observe the onset of tubulation transitions.

To monitor inward tubulation, we acquired time-lapsed mean fluorescence intensities in the GUV interior by defining a disk-shaped region of interest within the GUV contour and measuring the mean fluorescence intensity inside of the circle defining this region of interest.

Results and Discussion

IMDs induce membrane invaginations on GUVs

Fig. 1 C provides an example of time-lapsed confocal images of a transferred GUV under a membrane tension of 0.076 ± 0.004 mN/m. For this GUV, as the protein density on the GUV increased, the projection length (L_p) began to decrease at a well-defined transition point and finally disappeared completely. Concomitant with the onset of projection-length decrease, internal fluorescence built up (indicated by the open arrows). We consider this internal fluorescence to result from tubules induced by IMD proteins. This interpretation is in agreement with previous observations showing that this I-BAR protein can induce inward tubulation on liposomes (8). The tubules cannot be clearly identified through confocal microscopy because of their dynamic nature compared to the confocal-laser-scanning-image acquisition speed, and the fact that the diameter of the tubules (∼80 nm) induced by IMDs is smaller than the microscope-resolution limit (8).

To further prove that the observed membrane-shape instability is induced by IMD binding to membranes, we carried out two control experiments. In the first control, GUVs consisting of pure DOPC membranes were transferred into IMD protein solutions. For this lipid composition, no protein binding (and therefore no shape transition) was observed (Fig. S2). In the second control, we transferred GUVs containing biotinylated lipids into a streptavidin solution (no curvature sensitivity is expected for streptavidin). In this experiment, streptavidin binds to the GUV but no membrane instability transition was observed (Fig. S3). These two experiments imply that the projection-length decay observed in IMD experiments is not an artifact due to solution transfer or unspecific protein binding, but is a consequence of the curvature sensitivity of the IMD.

Although the onset of tubule formation induced by IMD cannot be directly visualized, the GUV surface area is a reliable indicator for the onset of the tubulation process (35). To trace the GUV surface area changes, GUV radius (R_v) and projection length (L_p) values were monitored during the protein-membrane-binding process, and the GUV surface area was obtained from these quantities (see the Materials and Methods).

Membrane tension and protein density coregulate membrane-curvature transition

Fig. 2 shows two representative time-lapsed traces of protein density (black vertical axis) and surface area (gray vertical axis) under different membrane tensions (Fig. 2 A, 0.076 ± 0.004 mN/m; Fig. 2 C, 0.255 ± 0.007 mN/m). These traces reveal a transition point (marked by the dashed lines) where the visible membrane area begins to decrease. We defined the onset of the area decrease as the value where the GUV area decreased to 2 SDs below the average value of the pretransition area. The onset of membrane-area decrease corresponds to a protein density defined as the shape-transition density. From the comparison of Fig. 2, A and C, it is observed that the instability-transition density is correlated with membrane tension: higher membrane tension requires higher protein density to induce the instability transition (transition densities for Fig. 2, A and C, are 1249 ± 77/μm² and 2591 ± 21/μm², respectively). Fig. 2, B and D, displays the time dependence of the mean-lipid-dye fluorescence intensities inside of the GUVs shown in Fig. 2, A and C, respectively. The membrane-area decrease observed in Fig. 2, A and C, is clearly correlated with an increase in the mean fluorescence intensity inside GUVs. The time point where membrane area starts to decrease is close to the time point where the mean intensity inside the GUV begins to increase. The increase in the fluorescence intensity inside the GUV is in agreement with the formation of inward membrane tubules, which will result in higher lipid concentration in the GUV interior.

Figure 2.

GUV shape instability depends on IMD protein density and membrane tension. Representative trials at (A) low membrane tension, 0.076 ± 0.004 mN/m and (C) high membrane tension, 0.255 ± 0.007 mN/m. (Dashed lines) Instability-transition protein density. The GUV area is the sum of the spherical part of the GUV and the cylindrical, pipette-aspirated part. The instability-transition density at high membrane tension is larger than at low membrane tension. (B and D) Recorded mean fluorescence intensity of lipid dye inside the GUVs shown in (A) and (C), respectively. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE.

To gain a closer understanding of the coupling of IMD protein density and membrane tension in determining the membrane-curvature instability, we examined a range of membrane tensions varying from 0.03 to 0.36 mN/m guided by the range of membrane tensions in cells (23–25). From a set of numerous measurements, we obtained a GUV-stability diagram. In this diagram, we correlate protein-transition density with the square root of membrane tension. The measurements define a stability boundary where the membrane-curvature transition occurs. If a GUV is located on the left and on top of this boundary the planar membrane state is mechanically stable, whereas toward the right and below of the stability boundary the GUV will have undergone a membrane-shape transition. Outside of the instable region of the membrane-shape-stability diagram, GUVs without tubulations were observed (Fig. 3, open squares), which is consistent with the location of the instability boundary.

Figure 3.

MIM/IMD-induced-GUV-stability diagram correlating membrane tension and density of protein on membranes. Each data point represents a measurement taken on an individual GUV. (Solid squares) Instability-transition protein densities on GUVs where tubulation was observed. These points were fitted by a linear-curvature-stability theory (43). (Open squares) Equilibrium-protein density on GUVs where protein binding had reached equilibrium but tubulation was not observed. (Dashed lines) 95% confidence interval. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE.

We note that IMD proteins were labeled with the synthetic fluorophore AF488 (Invitrogen). In principle, fluorophore labeling might interfere with protein function. To eliminate the possibility that the synthetic fluorophore interferes with membrane binding and curvature generation, we compared AF488-labeled MIM/IMD with unlabeled MIM/IMD as well as MIM/IMD-EGFP proteins in a cosedimentation assay and GUV-instability assay, respectively. Both assays confirmed absence of measurable effects due to fluorophores (Fig. S4).

Furthermore, as shown in Fig. 3, the experimental shape-stability boundary can be fitted with a thermodynamic curvature-instability model resulting from a linear-stability analysis, which predicts a membrane instability induced by proteins that couple with membrane curvature (36,37,43). Assuming protein-protein interactions to follow a two-dimensional van der Waals model, the instability boundary can be expressed as follows (35):

$$\sqrt{\sigma }\le \frac{\kappa \left|C_0\right|}{\sqrt{b}}-\sqrt{\frac{\kappa k_{\text{B}}T}{b\beta }\frac{{\rho }_0^3}{\rho {\left({\rho }_0-\rho \right)}^2}+\left(\frac{{\kappa }^2{C}_0^2}{b}-\frac{2\kappa \alpha }{b{\beta }^2}\right)}.$$ (3)

Here, σ is membrane tension; κ is the membrane-bending rigidity measured in the absence of protein, which we previously determined to be 23 ± 3 k_BT (35); C₀ indicates the spontaneous membrane curvature induced by IMD dimer (C₀ is negative based on the concave curvature preference of IBAR proteins); β is the excluded area of the IMD dimer on the membrane surface, predicted to be 54.9 nm² (cross-sectional area from crystal structure: 18.3 × 0.3 nm) (5); ρ is the protein-transition density; ρ₀ is the full-coverage protein-number density on the membrane, which we assumed to be 18,215/μm² (1/ 54.9 nm²); α is a protein-protein interaction strength; and b is a constant (35,45).

For fitting the experimental data, Eq. 3 can be expressed as

$$\sqrt{\sigma }=a_1-\sqrt{a_2{\rho }_0^3{\rho }^{-1}{\left({\rho }_0-\rho \right)}^{-2}+a_3},$$ (4)

where a₁, a₂, and a₃ are fitting parameters. These three primary fit parameters directly relate to a set of three physically intuitive parameters that describe the coupling of the protein with membrane curvature (35). These parameters are: the transition density at zero membrane tension, a maximal tension at which tubulation is possible, and the spontaneous curvature induced by the protein.

The fitting curve yields a positive x intercept, which corresponds to a protein-transition density (ρ_I0) of 188 ± 59/μm² at zero membrane tension. This positive x intercept is consistent with the fact that the membrane is stable in the planar state in the absence of MIM/IMD proteins. Furthermore, there is a tension limit (σ_max) of 0.35 ± 0.03 mN/m, beyond which GUVs can no longer be tubulated (see the Appendices for formulas that allow calculation of zero-tension protein-transition density and instability-transition tension limit from fit parameters).

From Eq. 5, the fitted constants yield the reciprocal of spontaneous curvature |C₀|⁻¹ = 3.70 ± 0.62 nm (uncertainty from error propagation) for this MIM/IMD homodimer:

$$a_1^2/a_2=\mathit{\kappa }{C}_0^2\beta /k_{\text{B}}T.$$ (5)

The measurements described so far all kept membrane tension constant before the shape instability. As can be deduced from Fig. 3, increasing protein density or reducing membrane tension both are expected to induce membrane tubulation if GUVs contain a sufficient amount of bound protein. To show that reducing membrane tension can induce a membrane-shape-instability transition induced by IMD, we aspirated a GUV with a relatively high membrane tension (0.238 mN/m) and equilibrated it in protein solution (180 nM). These conditions ensure constant protein density on the GUV while avoiding tubulation. We then reduced the membrane tension fivefold within 2 s. As expected, inward tubulations were observed after lowering membrane tension (Fig. 4 A). As in the case of transitions induced by protein density changes, tubulation was accompanied by a decrease in surface area (Fig. 4 B) and an increase in the mean fluorescence intensity of lipid dye in the GUV interior (Fig. 4 C).

Figure 4.

Reduction of membrane tension of a stable GUV equilibrated with IMD can also induce a membrane-shape transition. (A) Time-lapsed confocal images. (Top) Lipid channel and (lower) protein channel of a GUV. This GUV was stable after protein binding reached the equilibrium value. After equilibration at high tension, membrane tension was reduced to 0.033 mN/m within 2 s (starting point indicated by dashed line in B), and GUV area started to decrease (tubulation). (B) Protein-density (solid squares) and GUV-area (shaded squares) traces of images shown in (A). (C) Time-lapsed trace of the mean fluorescence intensity of lipid dye inside the GUV shown in (A). [MIM/IMD] = 180 nM and membrane tension = 0.238 mN/m. Buffer is 50 mM HEPES and 50 mM NaCl, pH = 7.4. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE.

As we have done before for endophilin (35), we asked the question if the protein transition density is affected by the protein bulk concentration. To answer this question for IMD, we carried out GUV instability assays under roughly the same membrane tension but with significantly different protein concentrations. The results (Fig. 5 A) show that protein-transition density does not significantly change with protein bulk concentration, which is consistent with our previous conclusion that protein-binding kinetics (which increases with increasing protein concentration) does not affect the transition density (which the linear instability analysis assumes to be an equilibrium property).

Figure 5.

Equilibrium protein density on GUV is not affected by membrane tension, and bulk protein concentration does not affect the instability-transition protein density. (A) Under the same membrane tension (∼0.2 mN/m), instability-transition density is not affected by protein concentration, and the equilibrium protein density is increasing with protein concentration. A quantity of 5 GUVs for each protein concentration was chosen for this comparison. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE. Error bars are mean ± SE. (B) At the same protein concentration (390 nM), equilibrium protein density is not changing with membrane tension.

Furthermore, we asked if the equilibrium density of the protein on the membrane depends on membrane tension. Such a dependency would indicate that the binding mode of the protein on the membrane might be influenced by the degree of membrane tension. For these experiments we used a broad range of tension, but the same protein bulk concentration as used earlier. Consistent with our earlier findings for endophilin N-BAR domains, we observed that the protein-binding affinity is not affected by membrane tension (Fig. 5 B).

Effects of lipid content

We also examined PI(4,5)P₂-containing GUVs because IMDs were reported to have higher binding affinity to PI(4,5)P₂-rich liposomes and to be able to induce PI(4,5)P₂ clustering (8,10). To carry out this set of experiments, we chose a mole fraction of 5% PI(4,5)P₂, which might be considered close to the PIP₂ content of a plasma membrane. To eliminate the effect of charge difference on protein-membrane binding, we aimed to keep the total charge of the GUV constant by reducing the phosphoserine percentage (from 45% used in the previous measurements) to 30%. In Fig. 6 A, no significant difference in protein membrane-binding equilibrium density can be observed. However, a 35% decrease in the average of transition densities was observed for 5% PI(4,5)P₂ GUVs (Fig. 6 B). To conclude, our observations suggest that 5% PI(4,5)P₂ can effect GUV membrane-curvature transitions at a lower protein density compared to membranes that do not contain PI(4,5)P₂. Interestingly, compared to N-BAR proteins, this PI(4,5)P₂ effect is the opposite for IMD proteins (35).

Figure 6.

PI(4,5)P₂ content in GUVs does not affect the equilibrium density and only weakly decreases the instability-transition density of the MIM/IMD. (A) At the same protein concentration (180 nM), equilibrium protein density is not significantly different (P = 0.76), comparing GUVs in the absence and in the presence of 5% PI(4,5)P₂. Non-PI(4,5)P₂ GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE; 5% PI(4,5)P₂ GUV composition is 30% DOPS + 30% DOPE + 34.5% DOPC + 5% PI(4,5)P₂ + 0.5% Texas Red-DHPE. The total charges of the GUVs are similar. (B) Quantitative comparison of the instability-transition protein density of GUVs with and without PI(4,5)P₂. The GUVs (six GUVs for each lipid composition) chosen for this comparison were under similar membrane tensions of ∼0.12 ± 0.03 mN/m. Student’s t-test yielded a P value of 0.016. A 35% decrease in the average of transition densities was observed for 5% PI(4,5)P₂ GUVs. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. Error bars are mean ± SE.

Effect of ionic strength

Because the measurements so far were all carried out in 50 mM NaCl solution, which is considerably less than the physiological ionic strength of 150 mM, we carried out control experiments at physiological salt concentrations. We observed a roughly 10-fold lower membrane binding for this MIM/IMD in 150-mM NaCl solution when compared to that in 50-mM NaCl solution (Fig. 7 A), which is expected given the known electrostatic contributions to the protein/membrane interaction (46). Interestingly, even though the membrane-binding capacity is lower in 150 mM NaCl solution, no significant difference in the instability transition density was observed (Fig. 7 B). This observation is consistent with our previous finding that variation of the fraction of negatively charged lipids in the membrane does affect equilibrium-binding densities, but not protein density at the membrane-shape transition (35). Taken together, these two findings suggest that electrostatic attraction of the BAR protein to the membrane is important for membrane binding, but not important for the curvature-generation capacity of the protein. Other mechanisms, such as wedging through hydrophobic insertion (10), or via oligomerization, may be responsible for the curvature-generation capacity of the IMD (47).

Figure 7.

Ionic strength does not significantly influence MIM/IMD’s ability to induce membrane-shape instability. (A) Equilibrium protein density on GUVs ([MIM/IMD] = 390 nM). At 150 mM NaCl, the equilibrium protein densities on GUVs are roughly 10-fold lower than at 50 mM NaCl. (B) (Bar graph) Comparison of the transition densities for IMD in GUV-instability assay at 150 and 50 mM NaCl. The GUVs (4 GUVs for each salt condition) used for comparison are under similar membrane tensions. Student’s t-test reveals a P value of 0.20, which suggests that the transition densities at 150 and 50 mM NaCl, respectively, are not significantly different. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE. Error bars are mean ± SE.

IMD N-terminus insertion inhibits membrane invaginations

N-terminus insertion was reported as a factor that influences the capacity of I-BAR proteins to induce filopodia formation (10). For MIM and ABBA, but not for IRSp53, N-terminal insertion was previously reported in Saarikangas et al. (10). To understand the role of N-terminus insertion in membrane-curvature induction through I-BAR proteins, we compared the membrane-curvature generation abilities of ABBA/IMD, IRSp53/IMD, and MIM/IMD D1–11 (i.e., MIM/IMD with residues 1–11 deleted).

Fig. 8 A shows that no significant difference in transition densities was observed for MIM/IMD and ABBA/IMD. On the other hand, compared to these two proteins, significantly lower membrane-curvature-instability transition densities were observed for IRSp53/IMD, as well as the N-terminus-deleted MIM/IMD isoform (MIM/IMD D1–11). This observation indicates that N-terminus insertion for I-BAR domains inhibits membrane invaginations in our GUV system, which makes sense for inward-tubulating membranes.

Figure 8.

N-terminal helix insertion reduces membrane-curvature-generation capacity of IMDs. Comparison of the instability-transition density (A) and equilibrium density (B) on transferred GUVs of MIM/IMD, ABBA/IMD, IRSp53/IMD, and MIM/IMD D1–11. Data points (GUV numbers used here for MIM/IMD, ABBA/IMD, IRSp53/IMD, and MIM/IMD D1–11 were 13, 9, 7, and 9, respectively) chosen for the transition-density comparison are in the same membrane-tension range of ∼0.15 mN/m. GUV composition is 45% DOPS + 30% DOPE + 24.5% DOPC + 0.5% Texas Red-DHPE. [Protein] = 400 nM. Buffer is 7 mM HEPES and 50 mM NaCl, pH = 7.4. Error bars are mean ± SE.

The equilibrium-binding densities of the four I-BARs at the same bulk protein concentration were also quantified and compared in Fig. 8 B. A Student’s t-test supports no significant difference in equilibrium-binding density for the four I-BARs. We note that IMDs being able to distinguish between effects that exclusively modulate protein binding (such as ionic strength) and those that exclusively affect shape transitions (such as N-terminal insertion), are unique strengths of our experimental approach.

Conclusions

The effect of I-BAR proteins on membrane-shape stability

In this contribution, by means of a membrane-shape-stability assay developed recently in our lab (35), we were able to assess the capacity of IMDs to generate membrane curvature. We found that, as for N-BAR domains, IMD-mediated membrane-curvature generation can be described with a theoretical model that allows for extracting three parameters individual to the protein of interest (36,37). The majority of previous contributions has considered only the spontaneous curvature of the protein (parameter 1), but we find that additionally, the protein-protein interaction strength needs to be considered (parameter 2). In our theory, this parameter is related to the density of proteins on the membrane required at negligible membrane tension to tubulate the membrane (35). Parameter 3, the maximal tension at which tubulation is possible, also quantifies the curvature generation capacity of the protein, and depends in a more complicated fashion on features of the protein (35). Given that endophilin has recently been shown to be the key curvature generator in specific internalization processes at the plasma membrane (48,49), a comparison between endophilin N-BARs and IMDs is warranted.

From fitting our experimental data, we obtained a zero-tension instability-transition density for MIM/IMD of 188 ± 59/μm². This positive x intercept indicates that GUVs are stable (i.e., no tubulation) at zero tension unless >188/μm² MIM/IMD proteins are bound to the outer GUV membrane leaflet. Interestingly, this zero-tension shape-transition density is more than threefold lower than that of endophilin N-BAR (∼650/μm² (35)). This shows that the MIM proteins have an even stronger ability in effecting a curvature instability compared to endophilin. This finding is also supported by comparing the maximal tubulation tensions: 0.35 ± 0.03 mN/m for IMD vs. 0.19 ± 0.03 mN/m for endophilin (35), indicating that IMD can tubulate at almost double the membrane tension than endophilin. The spontaneous curvature values of these two proteins are more comparable (the value of the spontaneous curvature radius for IMD is |C₀|⁻¹ = 3.70 ± 0.62 nm—well in the range found for N-BAR domains, |C₀|⁻¹ ≈ 1–6 nm (35,44,50). This underlines the fact that the spontaneous curvature of a protein is not a sufficient parameter to describe the membrane curvature generation capacity or the protein. We note in passing that it might have been expected that the less bent IMD would have a smaller spontaneous curvature than endophilin N-BAR. This observation indicates that the crystal structure of a BAR domain protein does not suffice to yield estimates of the spontaneous membrane curvature generated by the protein. Effects such as membrane insertion of amphipathic protein components such as demonstrated above, and further atomistic details of the protein/membrane interaction (51), strongly affect the degree of spontaneous curvature generated by a specific protein.

Taken together, our data imply that IMD is a significantly stronger membrane-curvature generator compared to endophilin N-BAR domains. This finding is consistent with previous literature that has reported that the IMD alone is sufficient to induce filopodia or filopodia-like membrane protrusions (52,53). As for endophilin, this study shows that the membrane instability induced by I-BAR proteins is regulated by the coupling of protein density and membrane tension. Because membrane tension universally exists in plasma membranes and is a key regulator of membrane-surface area (23), this protein-density versus membrane-tension coregulation mechanism is likely a universal mechanism of regulating membrane-shape transitions.

Effect of lipid content

Saarikangas et al. (10) reported that IMDs can induce PI(4,5)P₂ clustering in PI(4,5)P₂-rich (30% PI(4,5)P₂) GUVs, which precedes the formation of inward tubules. Our GUV instability assay with 5% PI(4,5)P₂ revealed an ∼30% decrease in membrane-instability-transition protein density. This observation suggests that the presence of 5% PI(4,5)P₂ can facilitate the formation of tubules at a lower protein density. However, the effect of PI(4,5)P₂ we observed here is smaller compared to the previous report that used 30% PI(4,5)P₂ (10).

Comparison of different I-BARs

ABBA is known as the I-BAR protein structurally closest to MIM (20). Consistent with this notion, we did not find statistically significant differences in membrane-curvature induction and membrane-binding properties of ABBA/IMD and MIM/IMD. This observation correlates their structure similarity with their functions in membrane-curvature induction.

Moreover, we observed stronger curvature-inducing abilities for IRSp53/IMD and MIM/IMD D1-11 as compared to MIM/IMD and ABBA/IMD. An important difference for these two groups of I-BARs is the N-terminal insertion proposed for MIM/IMD and ABBA/IMD (10). The absence of N-terminus insertion lowers the protein density required for membrane-curvature induction, which suggests that N-terminus insertion inhibits the abilities of I-BAR domains to induce membrane invaginations on GUVs. While N-terminus insertion was also reported to promote the formation of filopodia in cellular experiments (10), we hypothesize this phenomenon to be due to effects not captured in our in vitro experiments. Amphipathic helix insertion of N-BAR proteins (51,54), as well as globular proteins such as the N-terminal homology (i.e., ENTH) domain, promotes the formation of outward tubules from liposomes (55,56). Therefore, from a biophysical perspective, the N-terminal amphipathic-helix insertion of I-BARs is expected to inhibit membrane invaginations.

Author Contributions

Z.C. and T.B. designed research, Z.C. carried out experiments, Z.C. and T.B. analyzed data, and Z.C. and T.B. wrote the article.

Editor: Arne Gericke.

Appendix I: Zero-Tension-Transition Protein Density

When σ = 0, Eq. 4 results in

$${\left(\frac{\rho }{{\rho }_0}\right)}^3-2\ast {\left(\frac{\rho }{{\rho }_0}\right)}^2+\frac{\rho }{{\rho }_0}-\frac{a_2}{a_1^2-a_3}=0.$$

Setting $a_0=a_2/a_1^2-a_3$, and defining the protein cover-fraction as $\mathit{\varphi }=\rho /{\rho }_0$, we obtain the following equation:

$${\mathit{\varphi }}^3-2\ast {\mathit{\varphi }}^2+\mathit{\varphi -}{\mathit{\alpha }}_0=0.$$

Solving this equation yields zero-tension-transition cover-fraction

$${\mathit{\varphi }}_{\mathit{t}0}=\frac{2-2\text{cos}\frac{\theta }{3}}{3},$$

where $\theta =\arccos \left(2-27a_0/2\right)$. With this, the zero-tension-transition protein density is found to be ρ_t0 = ρ₀^∗ϕ_t0.

The error of ϕ is estimated by taking the derivative on both sides of the above equation, yielding

$$3\ast {\mathit{\varphi }}^2\Delta \mathit{\varphi }-4\ast \mathit{\varphi }\Delta \mathit{\varphi }+\Delta \mathit{\varphi }=\Delta a_0,$$

where $\mathit{\varphi }\mathrm{=}{\rho }_{t0}/{\rho }_0$ and Δa₀ is calculated by error propagation of a₁, a₂, and a₃.

Appendix II: Tension Limit for Instability Transition

When $\rho =\left(1/3\right){\rho }_0$, $\sqrt{\sigma }$ reaches a maximum of $a_1-\sqrt{\left(27/4\right)a_2+a_3}$.

The error of $\sqrt{{\sigma }_{\text{max}}}$ is calculated by error propagation of a₁, a₂, and a₃.