The notion that lipids are not merely passive fillers of the space between membrane proteins, but actively interacting with them, controlling protein conformational transitions and affecting their function, is now well-accepted (Maer et al., 2022; Levental and Lyman, 2023). Proteins define lipids’ structural and compositional distributions in cell membranes, and, conversely, lipids influence localization and properties of proteins. While some protein-lipid interactions involve chemical interactions between lipids and specific protein residues, others involve indirect interaction through perturbation of the lipid bilayer by proteins, thereby affecting membrane properties such as curvature. This has been shown for different families of integral membrane proteins (IMPs) (Sprong et al., 2001; Phillips et al., 2009; Corradi et al., 2019; Thakur et al., 2023), especially ion channels and transporters (Sunshine and McNamee, 1994; Nielsen et al., 1998; Kloda et al., 2007; Cordero-Morales and Vasquez, 2018). In most cases, membrane lipid composition influences ion channel gating properties and ligand binding (Gruner, 1985; Keller et al., 1993; Bezrukov, 2000; van den Brink-van der Laan et al., 2004; Rostovtseva et al., 2006; Sachs, 2010; Elinder and Liin, 2017; Mlayeh et al., 2017; Maer et al., 2022; Beverley and Levitan, 2024), by affecting channels’ protein-lipid interface.

The functions of conventional ion channels and transporters in the plasma membrane are regulated by ligands, or small molecules directly interacting with the receptor binding site of channel protein [see e.g., Table 1 in a recent review (Cordero-Morales and Vasquez, 2018)]. Less studied are amphiphilic small molecules that modulate ion channel behavior, not by directly interacting with channel-forming proteins, but indirectly by modifying the lipid membrane properties (Cordero-Morales and Vasquez, 2018). Not surprisingly, the effect of changes in the bilayer pressure on membrane proteins, or the so-called “force-from-lipid” effect, has been mostly studied on mechanosensitive or stretch-activated ion channels (Sukharev et al., 2001; Gullingsrud and Schulten, 2004; Sachs, 2010; Teng et al., 2015; Gerhold and Schwartz, 2016). A purified bacterial mechanosensitive channel of large conductance (MscL), for example, remained mechanosensitive even after reconstitution into model planar membranes (Martinac et al., 1987). However, determination of the precise molecular mechanism of how lipids modulate ion channel gating—through direct or indirect interaction, activating or deactivating—is a challenging task due to the structural complexity of the plasma membrane ion channels, such as mechanotransduction channels, glutamate receptor channels, transient receptor potential (TRP) channels, or Ca²⁺-activated large-conductance K⁺ (BK) channels. To overcome these apparent difficulties, the small ideally cation-selective channel gramicidin A (grA) has been extensively studied since the early 1980s (Elliott et al., 1983; Lundbaek and Andersen, 1999; Suchyna et al., 2004; Andersen et al., 2005; Lundbaek et al., 2010; Weinrich et al., 2017). As will be discussed later, the conductance and characteristic lifetime of this small dimeric channel respond exceptionally well to the changes in the bilayer charge, thickness, and lateral pressure distribution and are thus valuable for reporting on the channel’s lipid environment.

Amphitropic proteins (AP) are a subfamily of peripherally membrane-bound proteins that interact directly with the lipid membrane rather than with intrinsic membrane proteins and are therefore strongly influenced by lipid composition (Anderluh and Lakey, 2008; Prinz, 2010; Holthuis and Menon, 2014). APs are involved in various cell signaling pathways and lipid trafficking (Cho and Stahelin, 2005; Monkhouse and Deane, 2024). However, due to the transient nature of AP-membrane binding, the molecular mechanisms of these interactions are generally poorly understood. The effects of structurally and functionally diverse amphiphiles on the properties of bilayer membranes were described by Olaf Andersen and colleagues as changes in lipid intrinsic curvature, which reflects the attractive and repulsive forces between lipid molecules (Lundbaek and Andersen, 1999; Andersen and Koeppe, 2007; Lundbaek et al., 2010). These forces include electrostatic interactions between lipid headgroups and hydrophobic interactions between the acyl chains and depend on the “shape” of the lipid molecule and the lipid headgroup dipole charge. For instance, the bilayers formed from lamellar dioleoyl-phosphatidylcholine (DOPC) and nonlamellar dioleoyl-phosphatidylethanolamine (DOPE) present a striking difference in the lipid packing stress profile: the transition from DOPC to DOPE was estimated as the change of ΔP ∼ 100 atm or 10⁷ pascals in the lateral pressure in the hydrocarbon area of a planar membrane (Bezrukov, 2000; Gullingsrud and Schulten, 2004). This could be a result of the decrease of the cross-sectional area per lipid molecule from 0.72 nm² in lamellar DOPC (Kucerka et al., 2005) to 0.645 nm² in nonlamellar DOPE/DOPC (3:1 mol/mol) mixture (Rand et al., 1990). This, in turn, causes some difference of ≈0.16 nm in the hydrophobic thickness between DOPC and DOPE bilayers (Rand et al., 1990; Kucerka et al., 2005) (Table 1).

When APs interact with a membrane containing IMPs, a ternary protein-lipid-protein system is created (Figure 1). Even in the absence of direct interactions between APs and IMPs, the two proteins can affect each other by modifying lipid membrane properties, either at the global (i.e., whole-membrane) or local (i.e., confined to a small area around the bound or integrated protein) scale. These lipid-mediated protein-protein interactions are indirect and, therefore, difficult to study; multiple experimental approaches are required to clarify individual interactions contributing to the whole system’s behavior.

In this review, we describe how these difficulties could be surmounted by using a system of an IMP, the grA channel, reconstituted in a planar lipid membrane (PLM) in the presence of an AP, the water-soluble dimeric tubulin. Importantly, though there are no known direct interactions between tubulin and grA, tubulin absorption to the membrane surface modifies the channel behavior, thus revealing the role of the lipid membrane. We leverage our detailed understanding of the tubulin-lipid interaction mechanism, as well as the self-reporting properties of the grA channel, to understand how tubulin affects the properties of the lipid membrane. We also discuss here the converse relationship, where the presence of grA conducting dimer alters the local curvature and creates binding sites for tubulin in an otherwise inert membrane composition. We also compare the effect of the globular protein tubulin on the grA channel with that of another well-studied AP, the intrinsically disordered α-synuclein (αSyn).

The general structure of this review is shown in Figure 1. Global and local effects are differentiated by their vertical position in the figure, while the three components (AP, lipids, IMP) are distributed horizontally. Arrows indicate the direction of effects (not chemical equilibrium). In Section 2, we discuss the role of lipid composition on tubulin membrane binding. Section 3 delves into the effect of tubulin on the lipid membrane properties, as reported by the grA lifetime. Section 4 explores what can be learned about global and local membrane properties from the grA conductance. The use of flickering in the grA conductance to detect the presence of tubulin is discussed, with implications for the creation of local binding sites for APs around IMPs, in case the latter significantly distorts the lipid membrane structure. Finally, in Section 5, we compare the effects of αSyn and tubulin on grA channel properties.

A characteristic effect of tubulin on grA channel is a generation of fast flickering channel conductance in DOPE (Figure 6A) and DPhPC membranes (Rostovtseva et al., 2024). Considering that the grA channel does not “gate” in the conventional sense used for ion channels gating, the origin of this puzzling phenomenon was investigated following a model proposed earlier by Armstrong and Cukierman (2002) and Ring (1986), where the intensity of grA channel flickering was related to the bilayer thickness. This hypothesis was tested recently (Rostovtseva et al., 2024) in experiments using the membranes formed from phosphatidylcholine with a C(22:1) unsaturated acyl chain (diC(22:1)PC), whose hydrophobic thickness is 0.76 nm larger than that of DOPC (Table 1). As shown in Figure 7A, in diC(22:1)PC membranes, the fluctuations between grA open and zero-conductance levels are well-pronounced and time-resolved only in the presence of tubulin. The corresponding power spectral densities (Figure 7B) show a noticeable asymmetry in current blockages with respect to the sign of the applied voltage. Such asymmetry is a result of one-sided tubulin addition (30 nM tubulin in the cis side) in these experiments. Similar asymmetry was also observed in DPhPC and PC/PE membranes (Rostovtseva et al., 2024). A natural interpretation of this asymmetry is that the positive potential applied to the side of tubulin addition (cis side) pushes a positively charged α-tubulin membrane binding domain (Figure 5) towards the membrane plane, leading to a higher power spectral density of tubulin-induced current fluctuations at positive voltage than at negative (+200 vs. −200 mV in Figure 7B). Notably, the highest spectral density, which corresponded to the highest frequency of blockage events, was obtained in the thickest diC(22:1)PC bilayers. Following the interpretation of grA conductance flickering phenomena proposed by Armstrong and Cukierman (2002), the current blockages could be understood as the result of tubulin-induced modulations of the lipid funnel that forms the channel entrance (cartoon in Figure 7C). Alternatively, the tubulin dimer diffusing along the membrane surface could transiently block the channel conductance while approaching the lipid funnel. Interestingly, when instead of full-length tubulin, a synthetic peptide comprising the helix H10 of α-tubulin–the identified membrane-binding tubulin domain (Figure 5, see Section 2) – was used in grA experiments, no effect on channel conductance or induction of current flickering was observed (Rostovtseva et al., 2024). At the same time, similar to the full-length tubulin, the synthetic peptide also increased grA lifetime, but less effectively (≈4 times increase) than tubulin, and in a much higher (μM) concentration range.

These results suggest that tubulin’s effect on channel conductance and the generation of rapid conductance blockages might have a different origin than the effect of tubulin on grA lifetime. In Section 3.2, we explored how tubulin-induced changes in global membrane properties affect grA lifetime: smaller PE headgroups, as compared with PC headgroups, appeared to be more prone to adjusting to the tubulin-induced redistribution of the packing forces towards lipid hydrocarbon chains. This flexibility of the PE headgroups provides conditions for stronger tubulin α-helix anchoring (Figures 2, 3, 5) causing the increase of grA lifetime (Figure 6). Notably, both full-length tubulin and the synthetic peptide reduce the bilayer deformation energy contribution of grA channel formation reflected in an increased lifetime.

In the case of diC(22:1)PC membranes, binding of tubulin dimers is limited to the regions of membranes where headgroup packing is distorted (less dense, analogous to a smaller headgroup) in the lipid funnel formed by the grA channel. This local change in the membrane properties leads to stable binding of tubulin only where grA channels have formed (pathways §4 in Figure 1). The limitation to the region of the lipid funnel leads to the virtually unchanged global properties of the membrane and thus unchanged grA lifetime; however, the localized binding is clearly manifested via transient channel blockages by the bulky body of the tubulin dimer. The absence of conductance flickering in the presence of the α-tubulin membrane binding peptide is not surprising considering its drastically smaller size of 14 kDa compared with a 100 kDa tubulin globule (Figure 7C) and, therefore, its inability to induce local modulations near the channel entrances. In relatively thin DOPC membranes, there is no measurable tubulin-induced flickering; by contrast, in thicker diC(22:1)PC membranes (Table 1), the flickering is most pronounced. These results show that the deeper grA is embedded into the bilayer, the larger the effect of tubulin on channel conductance and flickering. This makes the possibility of direct tubulin-grA interactions in our experiments extremely unlikely and points towards local grA-induced lipid packing defects as the sites of tubulin binding and the source of current fluctuations.

We can conclude that the tubulin-grA interaction studied here is an example of a complex phenomenon in which protein binding and protein-protein interactions are regulated by lipids. Both the binding of APs and the incorporation of integral proteins into the membrane alter its properties and, via this alteration, protein function. Moreover, it is natural to expect that the effect of tubulin binding on membrane mechanics is reciprocal.

To further test our model of separation between global and local effects of AP on lipid membranes and embedded IMPs, we used another well-studied AP, αSyn. αSyn is a small, 14 kDa intrinsically disordered neuronal protein highly expressed in the central nervous system and constituting up to 1% of total cytosolic proteins in normal brain cells (Kruger et al., 2000). It is mostly known as a major component of the Lewy bodies found in the brains of Parkinson’s disease (PD) patients (Spillantini et al., 1997), the inclusions that are a pathological hallmark of this neurodegenerative disorder (Strohaker et al., 2019). Despite the obvious structural, functional, and genetic differences between αSyn and tubulin, they both belong to the AP family and are transiently associated with the cellular membrane. Similar to tubulin, αSyn was also found associated with mitochondrial membranes in vivo (Li et al., 2007; Nakamura et al., 2008; Nakamura, 2013; Robotta et al., 2014; Rovini et al., 2020) and with model membranes in vitro (Bodner et al., 2009; Pfefferkorn et al., 2012). Despite the obvious differences between tubulin—a compactly folded 110-kDa globular protein with a well-defined crystal structure (Nogales et al., 1998)—and αSyn—an unstructured 14-kD polypeptide—both proteins have one similarity in their structures: a disordered, highly negatively charged CTT. When tubulin or αSyn is added to the planar membrane bathing solutions with reconstituted VDAC, they both induce characteristic blockages of channel conductance (Rostovtseva et al., 2017; Rostovtseva et al., 2021). The CTTs of either tubulin or αSyn are the domains responsible for the reversible blockage of the VDAC pore. It was shown that when CTT of either protein is captured by VDAC pore, its selectivity is reversed to cationic, which is opposite to the anion-selective open state (Gurnev et al., 2011; Hoogerheide et al., 2017; Hoogerheide et al., 2018). Notably, the first step of both proteins’ interaction with VDAC is their transient binding to the membrane, followed by a partial and reversible block of the VDAC pore by their anionic CTTs driven by the applied potential. Therefore, the membrane binding of αSyn, as that of tubulin, has an array of physiological implications.

Due to the apparent importance of αSyn in neurodegeneration, an impressive amount of biophysical studies were devoted to αSyn interaction with membranes (Eliezer et al., 2001; Ulmer et al., 2005; Fusco et al., 2014; Fusco et al., 2017). One of the fascinating features of this small protein is that while being disordered in bulk solution, upon binding to the lipid membrane, αSyn’s N-terminal domain adopts a helical structure (Eliezer et al., 2001; Ulmer et al., 2005; Pfefferkorn et al., 2012; Fusco et al., 2014) with preferential binding to anionic lipids. αSyn also has a pronounced binding affinity to zwitterionic lipids (O'Leary et al., 2018; Jacobs et al., 2019) with a preference non-lamellar lipids, such as PE (Jo et al., 2000). Its membrane binding has a strong sensitivity to the membrane curvature and induces membranes remodeling, such as tubulation (Jiang et al., 2013). In summary, lipids with small and/or negatively charged headgroups enhance both binding affinity and helix formation of αSyn’s N-terminal membrane-binding domain. Therefore, αSyn seems to be an excellent AP candidate for testing its ability to modify lipid bilayer using a grA sensor probe.

The results of the effect of αSyn on grA parameters are shown in Figure 8. Similar to α-tubulin membrane-binding peptide (Rostovtseva et al., 2024), αSyn increases channel lifetime in a dose-dependent manner up to ≈5 times at 500 nM but does not measurably change grA conductance (Figure 8B). The concentration range at which αSyn affects grA is about 10 times higher than that of tubulin (Figure 6), but much lower than the micromolar range of α-tubulin synthetic peptide (Rostovtseva et al., 2024). We can conclude that αSyn affects the global properties of lipid membrane but, similarly to α-tubulin peptide, either does not sense the local membrane curvature near grA entrance, or, more likely, does not have a sufficiently bulky body to detectably disrupt ion flow through the channel. These findings naturally raise the question of whether large globular APs more effectively sense the local membrane curvature, which could be a subject of further studies.

Here we have established the reciprocal lipid-mediated interactions of an AP tubulin and an integral membrane protein grA in the absence of direct protein-protein interactions. The implications of these solely lipid-mediated interactions are broad, impinging on protein-membrane binding assays, the action of membrane proteins, and the mechanisms of action of membrane-altering small molecules.

First, these results suggest that membrane deformation can play a significant role in the binding of APs. In Section 2, we showed that tubulin binding to a lipid membrane, as observed by three different biophysical techniques, requires a multisite binding model to unify the individual observations. The presence of PE lipids, which alter the membrane properties to allow a greater degree of hydrophobic interactions, is important. Then, in Section 3, we showed that tubulin modifies the global properties of the membrane, which is expected in turn to alter tubulin’s binding propensity. These observations imply that, when describing the mechanics of binding, not only must the energetics of the multiplicity of binding sites be accounted for, but also the energy of membrane alteration. The latter is a collective, presumably protein sequence-dependent, effect.

Second, the lipid-mediated interactions between IMPs and their AP partners analyzed here provide a pathway by which the interactions between these proteins can be modulated in vivo. For example, the voltage-gating properties of VDAC are not significantly affected by the presence of tubulin with truncated CTTs (Rostovtseva et al., 2008). However, considering that the lipid composition of mitochondrial membranes is dynamic, especially under conditions of oxidative stress (Crimi and Esposti, 2011), or apoptosis (Kagan et al., 2004; Paradies et al., 2011), we can suggest that mitochondrial membrane remodeling modulates tubulin binding and, therefore, its regulatory interaction with VDAC. Because mitochondrial membranes are known to undergo dynamic remodeling in many cellular processes such as cell proliferation in disease and healthy development, we speculate that this type of regulation could be particularly relevant to other mitochondrial membrane proteins.

Finally, some membrane-altering small molecules, such as anesthetics, tricyclic antidepressants, and psychedelics, have properties similar to APs (Kapoor et al., 2019; Castellanos et al., 2020). The recognition of lipid-mediated protein-protein interactions could be instrumental in understanding the off-target action of some of these drugs on cell and organelle membranes and, therefore, on membrane proteins residing in, or peripherally associated with, those membranes.