Phosphoinositides alter lipid bilayer properties

Phosphoinositides regulate the function of structurally and functionally distinct membrane proteins (Hilgemann et al., 2001; McLaughlin et al., 2002; Suh and Hille, 2008; Logothetis et al., 2010). This regulation may result from phosphoinositide binding to the target protein, and several specific phosphoinositide-binding domains have been identified. Specific binding mainly (but not exclusively; Rosenhouse-Dantsker and Logothetis, 2007) involves the interaction of positively charged residues with the negatively charged phosphatidylinositol-4,5-bisphosphate (PIP₂) phosphates (McLaughlin and Murray, 2005; Hurley, 2006; Suh and Hille, 2008; Hansen et al., 2011; Leonard and Hurley, 2011; Whorton and MacKinnon, 2011). In many cases, the regulation seems to be more general, however, in the sense that other anionic phospholipids can substitute for the phosphoinositides (Fan and Makielski, 1997; Kim and Bang, 1999; Hilgemann et al., 2001; Zhang et al., 2003; Vaithianathan et al., 2008), suggesting that the regulation involves less specific electrostatic interactions with a positively charged region—but leaving open the question of how the phosphoinositides alter function.

The unifying feature among the diverse class of membrane proteins that are regulated by PIP₂ is that many are embedded in a lipid bilayer, and changes in bilayer properties are known to alter membrane protein function (Bienvenüe and Marie, 1994; Lee, 2004; Andersen and Koeppe, 2007; Lundbæk et al., 2010a). In this context, it is important that PIP₂ has ∼50% polyunsaturated fatty acids (Marsh, 2012b), which would tend to reduce the bilayer bending modulus (Rawicz et al., 2000; Bruno et al., 2013). PIP₂ constitutes ∼1% of total cell lipid, varying among different cell types (Ferrell and Huestis, 1984; Nasuhoglu et al., 2002; Wenk et al., 2003; Xu et al., 2003; Horowitz et al., 2005; Li et al., 2005; Hilgemann, 2007). Because the total cell lipid profile includes intracellular membranes, and PIP₂ is localized predominantly to the inner leaflet of the plasma membrane (Balla and Várnai, 2002), the area-averaged mole percentage of PIP₂ in the plasma membrane is likely to be >1% (Hilgemann, 2007)—and local PIP₂ concentrations may be even higher (Wang and Richards, 2012). Given the lack of a clear mechanism for many of the PIP₂-mediated effects, it thus becomes important to explore whether changes in extant, or exogenously applied, PIP₂s alter bilayer properties at relevant concentrations, in which case some of PIP₂’s effects may be caused by nonspecific bilayer-mediated mechanisms, as recently was proposed in the case of invertebrate phototransduction (Hardie and Franze, 2012).

Bilayer-spanning proteins are energetically coupled to the host lipid bilayer through hydrophobic interactions between the transmembrane domains and the lipid bilayer hydrophobic core, which can influence the lipid packing around the protein (Killian et al., 1989, 1996) or the protein conformation (Keller et al., 1993; Brown, 1994; Greathouse et al., 1994). This lipid reorganization (bilayer deformation) has an associated energetic cost, the bilayer deformation energy (ΔG_def). If protein conformational changes (e.g., from state I to state II) alter the local lipid bilayer deformations, the resulting difference in the bilayer deformation energy, $ΔGbilayerI→II$⁠, becomes the bilayer contribution to the total free energy of the conformational transition $(ΔGtotalI→II)$ (Nielsen and Andersen, 2000; Andersen and Koeppe, 2007; Lundbæk et al., 2010a):

where n_I and n_II denote the surface densities of the protein in states I and II, $ΔGproteinI→II$ and $ΔGbilayerI→II$ denote the protein and bilayer contributions to $ΔGtotalI→II$⁠, and k_BT consists of Boltzmann’s constant and the temperature in Kelvin. Changes in lipid bilayer physical properties, which are expressed as changes in $ΔGbilayerI→II$⁠, thus will alter the protein conformational equilibrium and function.

Amphiphiles modulate lipid bilayer properties by adsorbing to the bilayer–solution interface (Lundbæk et al., 2010a; Rusinova et al., 2011; Bruno et al., 2013), thereby altering the bilayer response to a deformation (changing ΔG_def and thus ΔG_bilayer). Key bilayer properties that may be affected by amphiphile adsorption include the intrinsic curvature, as a result of the altered profile of intermolecular interactions along the bilayer normal (Seddon, 1990; Cantor, 1997; Andersen and Koeppe, 2007; Marsh, 2007), bilayer elasticity (Evans et al., 1995), lipid disorder, and possibly lipid demixing (Zhelev, 1998; Heerklotz, 2002). These amphiphile-mediated changes in bilayer properties in turn may shift the conformational equilibrium of membrane proteins. (It is given that ΔG_def will change, but the changes may or may not be sufficient to cause measurable changes in protein function.) We therefore set out to determine whether PIP₂ and its structural analogues alter bilayer properties and, if so, to parse out which bilayer properties are affected.

Many different probes have been used to characterize the changes in bilayer properties that are caused by drugs and other amphiphiles (Seddon et al., 2009). For the present studies, we used gramicidin A (gA) channels as probes that serve as prototypical bilayer-spanning proteins, which experience minimal interference from specific lipid–protein interactions. gA channels form by the transmembrane dimerization of nonconducting monomeric subunits residing in apposing bilayer leaflets (O’Connell et al., 1990). Channel formation is associated with a local bilayer compression (Elliott et al., 1983; Huang, 1986; Harroun et al., 1999), which makes gA channels sensitive to changes in bilayer properties (Andersen et al., 2007; Lundbæk et al., 2010a,b).

Using gA channels as probes, we have explored the effects of phosphoinositides on lipid bilayers while varying the method by which membrane PIP₂ was manipulated, the acyl chain composition, and the number and positions of phosphate groups on the inositol ring (Fig. 1). For comparison, we also examined the bilayer-perturbing effects of other related short-chain phospholipids.

Our results show that increasing or decreasing the mole percentage of long-chain PIP₂ in the plasma membrane results in altered lipid bilayer properties. Both long- and short-chain phosphoinositides, as well as nonphosphoinositide short-chain phospholipids, are bilayer-modifying agents.

All phospholipids were purchased from Avanti Polar Lipids, Inc., except for diC₈PI (1,2-dioctanoyl-sn-glycero-3-phosphoinositol), which was purchased from Echelon Biosciences (Fig. 1). The diC₈ phospholipid stock solutions were 5 mM in water, except for the diC₈PE stock solution, which was 5 mM in ethanol (Pharmco-AAPER). The long-chain PIP₂s were 1,2-dioleoyl-sn-glycero-3-phospho-(1′-myo-inositol-4′,5′-bisphosphate) and the porcine brain extract of PIP₂, which is ∼80% C_18/20:4PIP₂. To make aqueous “solutions” of diC_18:1 and brain PIP₂, the powder was suspended in H₂O. The suspension was sonicated in an ultrasonic homogenizer (model 300 V/T; Biologics Inc.) for 30 min on ice at 90% power and 90% duty cycle using 1-s pulses. The resuspended long-chain phosphoinositides were flash frozen and stored at −40°C in a glass vial. To make DC_18:1PC (1,2-dioleoyl-sn-glycero-3-phosphocholine)/PIP₂ lipid mixtures, the long-chain PIP₂ was dissolved in chloroform (Honeywell Burdick and Jackson) and mixed with DC_18:1PC in chloroform at the appropriate mole percentage. Chloroform was evaporated under N₂, and the mixture was suspended in n-decane (Sigma-Aldrich). 8-Aminonaphthalene-1,3,6-trisulfonate (ANTS) was purchased from Invitrogen. For the single-channel experiments, we used a 15–amino acid analogue, [Ala¹]gA (AgA(15)), a chain-shortened 13–amino acid gA analogue of opposite helix sense, des-d-Val-Gly-gA⁻ (gA⁻(13)), and a 15–amino acid analogue in which the NH groups on the four Trp residues are methylated, [Val¹,N-Me-Trp^9,11,13,15]gA (N-Me-gA(15)); they were synthesized and purified as described previously (Greathouse et al., 1999). For the fluorescence assay, we used the natural mixture of gramicidins from Bacillus brevis (Sigma-Aldrich), which is ∼80% [Val¹]gA (Abo-Riziq et al., 2006); the mixture is denoted as gD (for R. Dubos, who discovered the gramicidins; Dubos, 1939).

Planar lipid bilayers were formed across a 1.5-mm–diameter aperture in a Teflon partition using DC_18:1PC or DC_18:1PC–long-chain PIP₂ suspensions in n-decane. The total phospholipid concentration was the same (15 mg/ml) in the pure and mixed lipid suspensions. The aqueous solution was a 1 M NaCl and 10 mM HEPES, pH 7.0, solution; the temperature was 25 ± 1°C. Single-channel measurements were performed at 200 mV using a patch clamp amplifier (model 3900A; Dagan Corp.) and the bilayer punch method (Andersen, 1983). The current signal was filtered at 5 kHz, sampled at 20 kHz, and digitally filtered at 500 Hz. After the bilayer was formed, gA analogues of different lengths and helix senses were added to both sides of the bilayer, and the solutions were stirred for 5 min (in a few experiments, only N-Me-gA(15) was added). After ∼30 min, to allow for gA incorporation, control current traces were recorded. In the experiments in which the lipids were added through the aqueous phase, short- or long-chain phospholipids were added to both sides of the bilayer. The solutions were stirred for 1 min and equilibrated for 10 min before recording.

Single-channel current transitions were identified using the algorithm described previously (Andersen, 1983), implemented in software written in Visual Basic (Microsoft Corp.). Single-channel lifetimes (τ) were determined by fitting single exponential distributions to survivor histograms of the lifetime distributions using Origin 6.1 (OriginLab). At least 300 channels were collected in each experiment. Changes in appearance frequency (f), the number of channels (opening followed by closing events) per unit of time, were determined only in experiments in which the bilayer remained intact after adding the phospholipid. The results were reported as means ± SD; in case there were only two observations, the results were reported as means ± range. Relative changes in appearance frequency were determined by normalizing the appearance frequency in the first two recordings right after the 10-min incubation with the compound to the appearance frequency from a recording right before compound addition. Only recordings lasting ≥300 s were used to evaluate changes in relative channel appearance frequencies.

gA in lipid bilayers exists in equilibrium between nonconducting monomers and conducting dimeric channels:

where k₁ and k₋₁ are the association and dissociation rate constants, respectively. The dimer dissociation rate constant can be estimated as 1/τ, and changes in the association rate constant can be estimated from changes in the channel appearance frequency because k₁ = f/[M]².

The channel’s hydrophobic length (l), ∼2.2 nm for a 15–amino acid analogue (Elliott et al., 1983; Huang, 1986), is less than the lipid bilayer’s unperturbed hydrophobic thickness (d₀), 3–4 nm (Benz et al., 1975; Lewis and Engelman, 1983; Lundbæk et al., 2010b), and gA channel formation is associated with a local bilayer compression with an associated deformation energy, ΔG_def. Using the theory of elastic bilayer deformations (Huang, 1986; Helfrich and Jakobsson, 1990; Dan et al., 1994; Nielsen et al., 1998; Nielsen and Andersen, 2000; Partenskii and Jordan, 2002; Rusinova et al., 2011), ΔG_def can be expressed as

where c₀ is the intrinsic monolayer curvature and H_B, H_X, and H_C are phenomenological elastic coefficients determined by the bilayer elastic moduli, the channel radius, and d₀. The bilayer responds to the compression by imposing a disjoining force (F_dis) on the channel,

acting to dissociate the dimer. Changes in bilayer physical properties will alter the phenomenological elastic coefficients and thus ΔG_def, f, and τ (Lundbæk et al., 2010b; Greisen et al., 2011), which make the gA channels suitable as probes of changes in bilayer properties.

The energetics of the gA monomer↔dimer equilibrium is described by

where [M] and [D] denote the surface concentrations of gA monomers and dimers, $ΔGbilM→D$ is the change in ΔG_def upon dimer formation, and $ΔGtotM→D$ is the free energy contribution from protein interactions. We are interested in the amphiphile-induced changes in bilayer properties, and the associated $ΔΔGbilM→D$ is determined by the changes in the time-averaged gA activity in the presence relative to that in the absence of amphiphiles. From Eq. 4:

where the subscripts denote the presence or absence of amphiphile and $ΔΔGtotM→D$ can be estimated as

The first approximation holds as long as [M] is in excess and does not vary after the amphiphiles have been added—as long as [M]_amph − [M]_cntrl/[M]_cntrl ≈ 0. The second approximation holds when the subunit–subunit interactions do not vary—as long as $ΔΔGprotM→D≈0.$

Additional information can be obtained through simultaneous measurements on two gA channels of different lengths (with different hydrophobic mismatch and helix sense, to prevent the formation of heterodimers), which allows for distinguishing between effects on bilayer curvature and elasticity. This is possible because, according to Eqs. 2 and 3, if the bilayer elasticity (expressed as changes in H_B) is altered, the effects will be hydrophobic mismatch dependent, with greater changes in the shorter channel’s appearance frequency and lifetime (Lundbæk et al., 2005; Rusinova et al., 2011). If the relative changes in lifetime are similar for the longer and shorter channels, curvature effects dominate (Artigas et al., 2006; Ashrafuzzaman et al., 2006).

To assay for the phosphoinositides’ bilayer-modifying effects, we used a gramicidin-based fluorescence quench assay (Ingólfsson and Andersen, 2010; Ingólfsson et al., 2010; Rusinova et al., 2011) in which fluorophore-loaded lipid vesicles were mixed with a gramicidin-permeable fluorescence quencher (Tl⁺) such that changes in gramicidin channel activity (the number of channels in the large unilamellar vesicle [LUV] membrane) were monitored as changes in the rate of fluorescence quenching. In brief, after solvent evaporation, DC_22:1PC (1,2-dierucoyl-sn-glycero-3-phosphocholine) was rehydrated in 100 mM NaNO₃, 25 mM ANTS, and 10 mM HEPES, pH 7. After vesicle extrusion and desalting to remove excess ANTS, the ANTS-loaded LUVs were prepared for fluorescent experiments by diluting 1:20 in 140 mM NaNO₃ and 10 mM HEPES, pH 7. ANTS-loaded LUVs were doped with gD; after incubation for 24 h at 12.5°C to allow for bilayer incorporation of gD, the LUVs were incubated at 25°C with or without (control) the test phospholipid for 10 min (unless otherwise noted). The LUVs and the Tl⁺-containing quench buffer were mixed in a stopped-flow spectrometer (SX.20; Applied Photophysics) with a dead time of <2 ms, and the time course of the ANTS fluorescence signal was recorded (see Figs. 8 A and 9 A). The resulting fluorescence quench curves were fit with a stretched exponential (Berberan-Santos et al., 2005) in Figs. 8 B and 9 B to determine the fluorescence quench rate at 2 ms, which reports on the time-averaged number of gramicidin channels in the LUV membrane.

All data plots and statistical analyses were performed in Origin 6.1. One population Student’s t test was used to establish whether means were significantly different from 1 as indicated by asterisks (P ≤ 0.05). Two population, two-tailed, independent Student’s t tests were used to establish whether two means were significantly different (P ≤ 0.05). A one-way ANOVA test performed on all compared means indicated significant differences (P ≤ 0.05).

Fig. S1 shows the diC₈PIP₂ effect on the single-channel lifetimes, appearance frequencies, as well as the channel $ΔΔGbilM→D$ for N-Me-gA(15) channels.

We first examine the effects of endogenous long-chain PIP₂, in which the PIP₂ was added to the lipid suspensions used to form mixed PIP₂/DC_18:1PC bilayers; we then examine the changes in bilayer properties produced by the addition of exogenous phosphoinositides—long-chain PIP₂, as well as short-chain phosphoinositides with varying numbers (and positions) of phosphate groups in the inositol ring—through the aqueous solution; finally, we compare the effects of short-chain (diC₈) phosphoinositides with those of other diC₈ phospholipids.

To explore the mechanisms by which phosphoinositides alter lipid bilayer properties and to estimate the changes in energetic cost of protein-imposed bilayer deformations, we used single-channel electrophysiology to monitor the changes in gA channel lifetimes and appearance frequencies. These experiments were performed using gA analogues of different chirality and length, gA⁻(13) and AgA(15), which allows for detailed information about the phosphoinositide effects on bilayer properties (Lundbæk et al., 2005).

Figs. 2 and 3 show results obtained in mixed brain PIP₂/DC_18:1PC or diC_18:1PIP₂/DC_18:1PC bilayers (Fig. 2); in each case, the PIP₂ was added at up to 2 mol %.

Using either mixture, the lifetimes of both gA⁻(13) and AgA(15) channels were longer than in pure DC_18:1PC (control) bilayers (Fig. 3, A and C). Both long-chain PIP₂s thus alter lipid bilayer properties at physiologically relevant mole percentages.

In addition to altering the single-channel lifetimes, the long-chain PIP₂s caused flickering channel appearances (Fig. 2), with no change in the current through the fully conducting state. The flickers appear as rapid transitions into and out of one or more subconductance states, meaning that they reflect partial pore block rather than gA dimer dissociation. We do not understand how the long-chain PIP₂s cause these conductance transitions, but they are unlikely to result from interactions between the indole NH groups and phosphate oxygens in the inositol bisphosphate head groups (Kim et al., 2012), as 2 mol % brain PIP₂ produced similar effects in N-Me-gA(15) channels (Fig. 4, A and B), where the indole residues cannot function as hydrogen bond acceptors (Sun et al., 2008). These transitions were not examined further, but the overall behavior is similar for gA⁻(13) and AgA(15) channels and is reminiscent of the behavior induced by the green tea catechins (Ingólfsson et al., 2011).

Because experiments with phosphoinositides often are done using exogenous addition, we examined the effect of brain PIP₂ and diC_18:1PIP₂ as well as diC₈PIP₂ when added through the aqueous solution. None of the PIP₂s altered the single-channel current amplitudes of gA⁻(13) or AgA(15) channels, indicating that they do not interact directly with, or accumulate in the vicinity of, the channels (Figs. 5 and 6, A and B). (The small differences in the gA^–(13) and gA(15) current amplitudes in the presence of the PIP₂s relative to controls are variable and not statistically significant.) Nor do the long-chain PIP₂s produce the flickering behavior that was observed in the mixed PIP₂/DC_18:1PC bilayers (Fig. 5). Surprisingly, given the results in Figs. 2 and 3, diC_18:1PIP₂ had no effect on f or τ for either channel type (Figs. 5; 6, A and C; and 7, A and B).

Both diC₈ and brain PIP₂ increased the single-channel lifetimes, and, when added at the same nominal concentrations, the lifetime increases relative to control were similar for diC₈ and brain PIP₂ (Fig. 7 A). diC₈PIP₂ produced no changes in the appearance frequencies of gA⁻(13) channels and only small changes in AgA(15) channels, whereas brain PIP₂ increased f for both channel types (Fig. 7 B), with the larger changes in f being observed for the longer AgA(15) channels. It is surprising that both brain PIP₂ and diC₈PIP₂ produce larger changes in f for gA(15) as compared with gA⁻(13) channels, but the greater effect of brain PIP₂ on f than on τ can be understood by noting that the bilayer needs to thin by >1 nm to reach the transition state for channel formation, whereas the two subunits need to move apart by only a small distance to reach the transition state for channel dissociation (Greisen et al., 2011).

The changes in $ΔGbilayerM→D$ are modest (Fig. 7 C) but of sufficient magnitude to alter gA channel function. We also evaluated the changes in $ΔGbilayerM→D$ for N-Me-gA(15) channels, where $ΔΔGbilayerM→D$ was −2.3 kJ/mol at 50 µM diC₈PI(4,5)P₂ (1,2-dioctanoyl-sn-glycero-3-phospho-(1′-myo-inositol-4′,5′-bisphosphate; Fig. S1).

As was observed previously with other amphiphiles (Lundbæk et al., 2010b), the changes in the gA^–(13) single-channel lifetimes (expressed as $ln{τ13/τ13,cntrl}$⁠) vary as a linear function of the changes in the AgA(15) single-channel lifetimes (expressed as $ln{τ15/τ15,cntrl}$⁠; Fig. 7 D). The slope is 1.30 ± 0.08 (R = 0.996) and not significantly different from a slope of 1. Thus, though the relative changes in single-channel appearance frequencies, f, and lifetimes, τ, elicited by either PIP₂ species were not significantly different for the longer AgA(15) and the shorter gA⁻(13) channels, which would suggest that they primarily alter intrinsic curvature, the aggregate results suggest that they also increase bilayer elasticity (Lundbæk et al., 2010b; Rusinova et al., 2011).

We also explored the long-chain PIP₂s’ and diC₈PIP₂’s bilayer-modifying effects using a gramicidin-based fluorescence assay that takes advantage of the gramicidin channels’ permeability to Tl⁺, a heavy-ion quencher of the water-soluble ANTS fluorophore. In this assay, changes in the fluorescence quench rate reflect changes in the Tl⁺ influx and channel activity (Ingólfsson and Andersen, 2010; Ingólfsson et al., 2010). The gA dimerization constant varies with changes in lipid bilayer properties (Andersen and Koeppe, 2007; Lundbæk et al., 2010a), meaning that one can attribute the changes in quench rate after amphiphile addition to modulation of bilayer properties, which in turn govern the gramicidin monomer↔dimer equilibrium and thus the number of channels in the LUV membrane (Ingólfsson and Andersen, 2010; Rusinova et al., 2011).

Brain PIP₂ and diC₈PIP₂ increased the fluorescence quench rates, whereas diC_18:1PIP₂ did not alter the quench rates at the tested concentrations (Fig. 8). The results in Fig. 8 are consistent with the electrophysiological results (Figs. 5–7), yet the different bilayer-modifying potencies of diC_18:1PIP₂ and brain PIP₂ are surprising because bilayer partition coefficients of fatty acids increase with increasing acyl chain saturation (Yokoyama and Nakagaki, 1993; Richieri et al., 1999). One would expect that the same would apply for diC_18:1PIP₂, compared with its polyunsaturated counterpart brain PIP₂ that mostly is C_18:0/20:4PIP₂. It is not clear, however, whether this reasoning should apply to situations in which the PIP₂s are micellar because the free energy of transfer between micelles and bilayers is likely to be modest (Bullock and Cohen, 1986; Heerklotz and Seelig, 2000). The observation that brain PIP₂ was more effective than diC_18:1PIP₂ thus suggests that the exchange between micelles and the bilayer is slower for diC_18:1PIP₂ than for brain PIP₂. Nevertheless, longer incubation times of vesicles with either long-chain PIP₂ did not alter the quench rates. (The relative changes in quench rates observed with 50 µM brain PIP₂ were 2.7 ± 1 vs. 3.1 ± 1.0; with 50 µM diC_18:1 PIP₂, they were 1.0 ± 0.1 vs. 1.05 ± 0.03 after 10 and 30 min, respectively). Moreover, although the longer acyl chain lengths would suggest greater partitioning of diC_18:1 or brain PIP₂ into the bilayer, as compared with the shorter diC₈PIP₂, the dose–response curves for diC_18:1 or brain PIP₂ were right shifted compared with those of their diC₈ counterpart.

The short-chain (diC₈) phosphoinositides are relatively water soluble, with critical micelle concentrations (CMCs) in the high micromolar range (Table 1), which makes them easy to handle in functional assays. We therefore examined the bilayer-modifying effects of diC₈ phosphoinositides with different numbers and positions of the phosphate groups on the inositol ring using the fluorescence quench method.

All of the diC₈ phosphoinositides increased gramicidin channel activity, as reflected by the increased quench rate, with PI being the most active and PI(3,4,5)P₃ and PI(3,5)P₂ the least bilayer active: PI > PI(4,5)P₂ = PI(3,4)P₂ > PI(3,4,5)P₃ ≥ PI(3,5)P₂ (Fig. 9).

The concentrations at which we observed changes in bilayer properties varied from 10 µM for PI, PI(3,4)P₂, and PI(4,5)P₂ to 150 µM for PI(3,5)P₂ and PI(3,4,5)P₂. Overall, the potency decreased with increasing head group charge (compare with Fig. 1), as would be expected if the short-chain phosphoinositides’ bilayer-modifying potencies depended on their ability to partition into the bilayer–solution interface. Though the formal charge on PI(3,5)P₂ is the same as that on PI(4,5)P₂ and PI(3,4)P₂, its bilayer-modifying potency was less than that of PI(3,4,5)P₃, suggesting that the phosphate configuration around the inositol ring modulates the interaction with the bilayer. Unlike the adjacent phosphates in PI(3,4)P₂ or PI(4,5)P₂, the distant phosphates in PI(3,5)P₂ may prevent the phosphoinositol ring from inserting as far into the bilayer (Redfern and Gericke, 2005), resulting in weaker bilayer effects.

For comparison with the short-chain phosphoinositides, we examined other short-chain phospholipids—diC₈PS (1,2-dioctanoyl-sn-glycero-3-phospho-l-serine), -PE, and -PC. All three alter lipid bilayer properties with similar dose–response curves (Fig. 10); they have similar bilayer-modifying potencies even though their head groups vary in size and charge.

The bilayer partition coefficients of surfactants can be estimated from their CMCs (Bullock and Cohen, 1986; Heerklotz and Seelig, 2000). (In case CMCs are not available, the partition coefficient of a neutral amphiphile can be estimated from the logarithms of the calculated water/octanol partitioning ratio (clogP; Escher and Schwarzenbach, 1995; Avdeef, 2001; Seydel and Wiese, 2002). Based on the estimated mole percentage of each phospholipid in the LUV membranes (Table 1) and the dose–response curves (Fig. 10), the zwitterionic diC₈PC and -PE have similar bilayer-modifying potencies, whereas the anionic diC₈PI and -PS are more potent.

The short-chain diC₈PC (1,2-dioctanoyl-sn-glycero-3-phosphocholine), -PE, -PS, and -PI vary also in their effects on planar bilayers, with diC₈PS and diC₈PE eliciting the largest channel lifetime increases, and diC₈PC causing the least change (Fig. 11 A).

Although the changes are small, the linear relationship between the natural logarithm of the AgA(15) and gA⁻(13) single-channel lifetimes, in the presence of diC₈ phospholipids, has a slope of 1.16 ± 0.05, R = 0.994 (Fig. 11 D), suggesting that they do increase bilayer elasticity (Lundbæk et al., 2010b), in addition to changing the curvature (Rusinova et al., 2011). Except for the effects of diC₈PS on AgA(15) channels, the relative changes in f (Fig. 11 B) were larger than the changes in τ. At the highest concentration tested, none of the phospholipids (whether anionic or zwitterionic) altered the gA single-channel current transition amplitudes. That the anionic phospholipids do not alter the single-channel current transition amplitudes is in striking contrast to the results obtained with free fatty acids (Bruno et al., 2007, 2013).

The mechanisms by which phosphoinositides alter membrane protein function remain enigmatic in the sense that, for many proteins, it is unclear to what extent the phosphoinositides act through direct binding to their membrane protein effectors, as opposed to altering lipid bilayer properties or some other indirect mechanism that in turn will alter the energetics of protein conformational transitions. We used gramicidin-based fluorescence quenching and single-channel assays to determine whether various polyphosphoinositide isoforms alter lipid bilayer properties and found that they do so at physiologically relevant concentrations (mole fractions in the membrane).

First, we highlight that bilayer modulation by PIP₂ occurs at “relevant” concentrations and consider what bilayer properties are affected. Next, we discuss implications of our findings in view of what is known about PIP₂ and its mechanism of action. Finally, we discuss and compare the bilayer-modifying effects of PIP₂ to those of isoforms with different phosphate group organizations as well as acyl chain variants and other phospholipids.

The PIP₂ concentrations tested here are comparable with those used in cell physiological experiments. Following Rusinova et al. (2011), we analyzed the bilayer contribution to the free energy of gA channel formation (Fig. 7 C) when the PIP₂ analogues are applied through the aqueous phase and found that the changes are modest: $ΔΔGbilM→D ≤ kBT$ (= 2.5 kJ mol⁻¹ at 25°C). (This estimate pertains to gA channels in a relatively soft bilayer; $ΔΔGbilI→II$ for an integral membrane protein may be larger or smaller, depending on the protein radius, shape, and the host bilayer material properties.) Both long- and short-chain polyphosphoinositides have little effect on bilayer properties at 3 µM (nominal concentration; Figs. 7 and 9), meaning that functional effects of long-chain polyphosphoinositides that occur at or below the micromolar range are unlikely to be bilayer mediated. In contrast, effects that are observed only at concentrations ≥10 µM may, at least in part, be caused by altered bilayer properties. Many studies have used much higher [diC₈PIP₂] (Runnels et al., 2002; Zhang et al., 2003; Lopes et al., 2005; Rohács et al., 2005; Hernandez et al., 2008), in which case it becomes important to consider the possibility of membrane effects.

PIP₂-sensitive channels vary considerably in their phosphate isomer specificity (Fan and Makielski, 1997; Rohács et al., 2003; Zhang et al., 2003; Rohács, 2007), and the binding pockets in PIP₂-binding membrane-spanning proteins vary greatly, meaning that the definition of binding sites for PIP₂ becomes ambiguous (Suh and Hille, 2008). Other than requirements for positive and aromatic residues, the binding sites are not clearly defined (Rosenhouse-Dantsker and Logothetis, 2007; Suh and Hille, 2008), however, and mutations of specific residues, or even entire polybasic regions, may modulate but not eliminate the PIP₂ effects (Rohács et al., 2003; Zhang et al., 2003; Rohács, 2007; Vaithianathan et al., 2008). These ambiguities suggest that some PIP₂ effects may be the result of less specific interactions, such as altered energetic coupling between the target protein and the host bilayer. Indeed, nonphosphoinositide phospholipids can alter channel function if applied at high concentrations from 0.01 to 1 mM (Fan and Makielski, 1997; Kim and Bang, 1999; Hilgemann et al., 2001; Zhang et al., 2003; Vaithianathan et al., 2008). Importantly, the PIP₂ abundance percentage in the inner leaflet of the plasma membrane is on the order of 1%, perhaps higher (Ferrell and Huestis, 1984; Balla and Várnai, 2002; Nasuhoglu et al., 2002; Wenk et al., 2003; Xu et al., 2003; Horowitz et al., 2005; Li et al., 2005; Hilgemann, 2007), and PIP₂ alters gramicidin channel stability at 2 mol % (Fig. 2). Thus, physiological depletion or enrichment of PIP₂ in the inner leaflet of the plasma membrane (Nasuhoglu et al., 2002; Xu et al., 2003; Horowitz et al., 2005; Li et al., 2005; Suh and Hille, 2005), whether global or local, is likely to be associated with changes in lipid bilayer properties.

Local PIP₂ enrichment may be considerable, as there are 70–80 phospholipids in the first two lipid shells around a bilayer-spanning protein of 3-nm radius. If one PIP₂ molecule is required to alter protein function, the local PIP₂ mole percentage would be ∼1; if four PIP₂s were needed, the mole percentage would be ≥5! PIP₂ binding to its target proteins thus may alter the local bilayer environment (the bilayer material properties), and the question becomes, what are the relative energetic contributions of specific binding and nonspecific bilayer-mediated effects on protein function?

Whether bilayer-mediated effects become important will depend on the affinity of PIP₂ (or other ligands) for the protein. Some ion channels are regulated at very low PIP₂ concentrations, whereas other PIP₂ targets require higher concentrations, as reflected in their regulation by physiological stimulation of PIP₂ hydrolysis (Huang et al., 1998; Kobrinsky et al., 2000; Runnels et al., 2002; Lopes et al., 2002; Zhang et al., 2003; Rohács et al., 2005; Rohács, 2007). In the case of Kir channels, the PIP₂ affinity is inversely correlated with the channels’ susceptibility to regulation by PIP₂ isomers and other molecules (Zhang et al., 1999; Schulze et al., 2003; Lee et al., 2005; Rosenhouse-Dantsker and Logothetis, 2007). Thus, channels with lower affinity for PIP₂ are likely to be more sensitive to their environment, in which case dynamic changes in the environment (including PIP₂-induced changes in bilayer properties) may be important for function. These features may be tested by exposing the channels to short-chain nonphosphoinositide phospholipids, which modify the bilayer (Fig. 11 C) and thus allow for a positive control of sensitivity to changes in lipid bilayer properties.

Results obtained using the gramicidin-based fluorescence assay show that all compounds tested increase the ANTS fluorescence quench rates because they shift the gA monomer↔dimer equilibrium toward the conducting dimers, indicating that each of them alters lipid bilayer properties, albeit with different potencies.

A compound’s apparent bilayer-modifying potency reflects several factors: its affinity for the bilayer (its partition coefficient), which relates the concentration of mono-disperse molecules in solution to the surface density or volume concentration in the bilayer; the state of the compound in solution, where micelle or other aggregate formation will influence the number of monomeric molecules available to partition into the bilayer (see last paragraph of Discussion); and the ability of a molecule in the bilayer to alter bilayer properties. The apparent potencies of the short-chain phosphoinositides decrease with increasing charge, consistent with the expected influence of charge to lower the partition coefficient into lipid bilayers (Peitzsch and McLaughlin, 1993). In the LUV experiments, the short-chain nonphosphoinositide phospholipids and diC₈PI have similar potencies. The similar effects of the charged and zwitterionic phospholipids could be caused by the combined effect of a decreased partition coefficient (Marsh and King, 1986; Peitzsch and McLaughlin, 1993; Marsh, 2012a) and a larger effective head group size (Israelachvili et al., 1980; Kleinschmidt and Tamm, 2002) for the charged phospholipids. Thus, molecule per molecule, the charged phospholipids would be expected to have increased bilayer-modifying potency (Lundbæk et al., 1997; see also Lundbæk and Andersen, 1994; Bruno et al., 2013). In the case of the polyphosphoinositides, the decreased potencies, relative to diC₈PI (compare Fig. 9 with Fig. 10), may be the result of a dominant effect of the charge-dependent decrease in the partition coefficient.

Using isothermal titration calorimetry, we explored the partitioning of diC₈PI, diC₈PIP₂, and diC₈PI(3,4,5)P₃ into lipid vesicles but did not observe heats of partitioning (unpublished data), either because the partition coefficients are too low to be measured, as suggested from the CMCs (Table 1 legend), or because the partitioning is primarily entropy driven. In any case, although the head group charges make it challenging to estimate the mole fractions of PI, PIP₂, and PIP₃ in the bilayer (they preclude using clogP for the estimate), the decreasing bilayer-modifying potency with increasing charge on the inositol ring suggests that the differences in potency are caused mainly by differences in bilayer partitioning.

The bilayer-modifying potencies of the long-chain PIP₂ and the short-chain phosphoinositides and other phospholipids varied with the experimental system, but there was overall concordance between the results obtained in planar bilayers and LUVs, indicating that bilayer-modifying effects are little affected by changes in ionic strength (and the associated changes in ionization) or the presence of decane. The key difference we observed was with the long-chain PIP₂s. When added through the aqueous solution, brain PIP₂ altered bilayer properties, whereas diC_18:1PIP₂ did not—and diC_18:1PIP₂ remained inactive after prolonged (30 min) incubation. This difference may be important for the design and interpretation of experiments in biological membranes. To gain further insight into these observations, we attempted to determine the partition coefficients of the long-chain PIP₂s by separating LUV-partitioned PIP₂ from micellar PIP₂ using ultracentrifugation and assessing the PIP₂ concentrations by phosphorus assay (Chen et al., 1956). The density of PIP₂ micelles, however, allows them to sediment with LUVs (unpublished data), precluding separation of LUVs from PIP₂ micelles by centrifugation.

To determine how the short- and long-chain phospholipids alter lipid bilayer properties, we sought to use gA single-channel recordings to explore whether the effects could be ascribed to changes in bilayer elasticity or curvature. (Long-chain PIP₂s also may alter the bilayer lateral organization [Wang and Richards, 2012], but we have discovered no evidence for lipid de-mixing.) Both long- and short-chain PIP₂s produce modest relative changes in f and τ. Even when comparing the lifetime changes observed with the gA^–(13) and AgA(15) channels (Figs. 7 D and 11 D), we cannot unambiguously distinguish between PIP₂ effects on curvature and elasticity. Furthermore, it remains unclear to what extent the electrostatic interactions among the PIP₂ head groups, which would tend to increase the effective head group size (Kleinschmidt and Tamm, 2002), would influence the intrinsic curvature (Lundbæk et al., 1997) and maybe also the head group hydration. Importantly, the PIP₂s had no effect on the gA single-channel current amplitudes, indicating that their mole percentage in the first lipid shell around the channel was low (compare the contrasting results obtained with other anionic compounds; Lundbæk et al., 1997; Bruno et al., 2007). Though not conclusive evidence, this suggests that electrostatic effects are likely to be of minor importance at physiological mole percentages.

In contrast, diC₈PC, -PI, -PS, and -PE produce greater increases in f and τ for the short, as compared with the long, gA channels, indicating that short-chain non-polyphosphoinositide phospholipids increase bilayer elasticity. The results, nevertheless, do not rule out further effects of lipid curvature (Lundbæk et al., 2010a,b). In contrast to the LUV experiments, in which the short-chain phospholipids had similar potencies, the bilayer-modifying rank-order potency in the planar bilayer experiments was PS ≥ PE > PI ≥ PC, which cannot be accounted for simply by different partitioning (Table 1). The different rank-order potencies in the LUVs and planar bilayers, as well as the differences observed for short-chain phospholipids described here and long-chain lysophospholipids from a previous study (Lundbæk and Andersen, 1994), reflect the compounds’ detailed molecular properties, including the effective cross-sectional areas of the acyl chains relative to the head groups—as well as the lengths of their acyl chains relative to those of the bilayer-forming lipids in the host bilayer (Aagaard et al., 2006; Ingólfsson and Andersen, 2011). Ingólfsson and Andersen (2011), in particular, found that the sign of the effect of an n-alkyl alcohol on lipid bilayer properties varied with the n-alkyl length (short-chain alcohols increased gA channel dimerization; long-chain alcohols decreased dimerization). The chain length at which the transition occurred furthermore varied with the host bilayer phospholipid acyl chain length. The relative differences in bilayer-modifying potency observed in the stopped-flow and single-channel experiments thus may differ (see also Ingólfsson and Andersen, 2010).

The different rank-order potencies of the short- and long-chain PIP₂s in single-channel versus fluorescence assays again are likely to reflect different perturbations of the membrane acyl chains. The hydrophobic and electrostatic contributions to the free energy of adsorption at the bilayer–solution interface tend to be additive (Zaslavsky et al., 1978; Israelachvili et al., 1980; Tanford, 1980; King and Marsh, 1987; Buser et al., 1994), meaning that the difference in ionic strength between the fluorescence and single-channel experiments should have similar effects on the adsorption of the long- and short-chain PIP₂s.

The lack of effect of diC_18:1PIP₂ in the fluorescence and electrophysiological assays, when added through the aqueous phase, is in striking contrast to the effects observed when the compound is added through the lipid phase (compare Figs. 3, 6, and 7). The key difference between these experiments is the elimination of the micelle→water and water→bilayer partitioning steps, which also may have implications for the design and interpretation of cell physiological experiments. For example, the negative correlation between the PIP₂ concentration needed to elicit an effect and the length of the PIP₂ acyl chain (Rohács et al., 1999; Jin et al., 2002; Zhang et al., 2003; Vaithianathan et al., 2008) is often attributed to higher partitioning of PIP₂s with longer acyl chains. Yet, the modest differences between the dose–response curves for diC₈- and brain PIP₂, which cannot be accounted for by the difference in hydrophobicity of their acyl chains, show that the correlation between CMC and bilayer partitioning is complicated—at least in systems containing both micelles and bilayers (see beginning of this section). The apparent differences in bilayer partitioning of the diC_18:1 and brain PIP₂ analogues in our assays also may manifest themselves in cell physiological studies, as apparent differences in potency that may not, in fact, have a functional basis.

Both long- and short-chain phosphoinositides are bilayer-modifying molecules in that they can alter their target protein function by a combination of direct (protein binding) and indirect (bilayer mediated) effects. The changes in bilayer properties occur at physiologically relevant concentrations (mole percentage of membrane lipids), in the case of the long-chain PIP₂s, and at pharmacologically relevant concentrations, in the case of short-chain compounds. It thus becomes important to consider the changes in membrane properties resulting from local PIP₂ depletion or enrichment, which may be a contributing mechanism through which PIP₂ alters protein function. Our results suggest a strategy for identifying, or excluding, such nonspecific bilayer effects. Effects at low micromolar concentrations of (long or short chain) PIP₂s are likely to reflect specific interactions with the target protein, whereas effects observed only at 10 µM or higher concentrations are likely to reflect, at least in part, altered bilayer properties. It is furthermore possible to test for nonspecific bilayer effects using not only short-chain phosphoinositides but also other short-chain phospholipids.

This work was supported by National Institutes of Health (NIH) grant GM021342 and American Recovery and Reinvestment Act of 2009 Supplement GM021342-35S1. The peptide facility is supported by NIH grants GM103429 and GM103450.

We thank Denise V. Greathouse for the purified gA analogues and Donald W. Hilgemann, Helgi I. Ingólfsson, Ruchi Kapoor, Jens A. Lundbæk, R. Lea Sanford, and Sidney A. Simon for helpful discussions.

Donald W. Hilgemann served as guest editor.