Quantification of protein–lipid selectivity using FRET
- 729 Downloads
Membrane proteins exhibit different affinities for different lipid species, and protein–lipid selectivity regulates the membrane composition in close proximity to the protein, playing an important role in the formation of nanoscale membrane heterogeneities. The sensitivity of Förster resonance energy transfer (FRET) for distances of 10 Å up to 100 Å is particularly useful to retrieve information on the relative distribution of proteins and lipids in the range over which protein–lipid selectivity is expected to influence membrane composition. Several FRET-based methods applied to the quantification of protein–lipid selectivity are described herein, and different formalisms applied to the analysis of FRET data for particular geometries of donor–acceptor distribution are critically assessed.
KeywordsFRET Protein–lipid interactions Membrane proteins Fluorescence
Preferential interactions of proteins with selected lipids are able to drive enrichment of the bilayer around the protein in these lipids and impoverishment in others, creating local heterogeneities that can potentially extend to several lipid shells around the protein. Some superficial membrane proteins demonstrate specific binding to some lipid classes, a phenomenon that can control protein recruitment to the membrane and activate signaling cascades (Czech 2000). In addition, transmembrane proteins display differential interactions with lipids of different acyl-chain lengths due to packing constraints in the lipid–protein hydrophobic interface, which have a significant effect on the activity of several proteins. Membrane proteins also have been shown to present binding sites for lipids in hydrophobic pockets away from the protein–lipid interface, and binding of specific lipids to such sites is essential for activity in several cases (Lee 2003).
Interactions between membrane proteins and lipids have been generally addressed using electron spin resonance (ESR) (Marsh and Horváth 1998). This technique can discriminate between immobilized lipids (near the protein interface) and mobile lipids in the bulk, and has been the main approach in these type of studies. Fluorescence static or collisional quenching methods (London and Feigenson 1981; Everett et al. 1986; O’Keeffe et al. 2000; Williamson et al. 2002) can also provide a similar structural type of information. These techniques are able to probe the lipid environment in direct contact with the protein, but are insensitive to the presence of lipids displaced from the protein–lipid interface. On the other hand, Förster resonance energy transfer (FRET) is sensitive to distances up to 100 Å, and like all fluorescence methodologies offers the maximum sensitivity among spectroscopic techniques. In this way, and as will be shown in this review, it is an excellent tool for the study of protein–lipid interactions.
FRET between two molecules (donor and acceptor) exhibits a sixth-power dependence on the distance between them. The propensity for FRET to occur for a given donor–acceptor molecular pair is generally described by the Förster radius (R0), the distance from the acceptor at which the probability of donor deactivation by energy transfer is 50%. Typical values for R0 lie between 20 and 60 Å. Due to the high sensitivity of FRET to changes in distances around R0, it is possible, through the use of chemical labeling or intrinsic fluorophores, to directly measure the binding of lipids to proteins, or protein-mediated enrichment of particular lipids in the membrane.
In this review we focus on the problem of quantifying the interaction of membrane proteins with biologically relevant lipid molecules making use of FRET methodologies. In the field of protein–lipid interactions, FRET has also been extensively used to characterize and quantify binding of lipids to soluble proteins (Petrescu et al. 2001; Gadella and Wirtz 1991; Dansen et al. 1999) and partition of membrane proteins to lipid membranes (Romoser et al. 1996; Corbin et al. 2004) and particular lipid phases (Wang et al. 1988), most notably to liquid-ordered (raft-like) phases (Zacharias et al. 2002). However, we will not discuss these particular FRET applications.
Apart from the use of FRET as a molecular ruler (Stryer and Haugland 1967), the information retrieved from its application in biological sciences is often qualitative, neglecting the full potential of FRET in the study of molecular interactions. This is particularly true for studies of interactions in biomembranes, due to the additional level of complexity introduced by the possibility of energy transfer to multiple acceptors, including nonbound molecules. In fact, as a consequence of partition of donors and acceptors to the lipid bilayer, the concentration of acceptors around each donor increases dramatically, resulting in a nonzero probability of energy transfer to many nonbound, nearby acceptors. However, the availability of a large amount of precise information on the properties of biomembranes and fluorescent lipid probes, as well as accurate numerical and analytical solutions to the problem of FRET in planar distributions of acceptors, surmounts these problems. We will critically present and discuss some of these formalisms, as well as the simplifications required for their application and the respective usefulness to the problem of quantifying protein–lipid selectivity.
FRET donors and acceptors
List of commonly used FRET pairs for study of protein–lipid interaction
Förster radius (Å)
Rhodopsin retinal group
50 (for EGFP)o
Several fluorescently tagged lipids have been shown to mimic properties of their natural analogues and were extremely helpful in elucidating several problems in lipid trafficking and sorting (Maier et al. 2002). Nevertheless, for each different lipid probe, studies must be carried out to characterize its properties and confirm this mimicry. Problems observed with some cholesterol fluorescent derivatives are a good illustration of the need for careful selection of lipid probes (Loura et al. 2001). One important concern to bear in mind when studying protein–lipid selectivity is to avoid changing the particular lipid property we are interested in through the incorporation of the fluorescent tag, i.e., in case of studies on protein selectivity for particular acyl-chain characteristics (length or unsaturation), the fluorescent label should be inserted in the hydrophilic section of the molecule, and vice versa (Fernandes et al. 2004).
Measuring FRET efficiencies
FRET efficiencies are generally measured through quantification of the extent of donor fluorescence quenching introduced by the presence of the FRET acceptor. Therefore, two measurements are required (with and without acceptor). The acceptor species is not necessarily fluorescent. If it is fluorescent, FRET efficiencies can also be calculated from the sensitized fluorescence emission, through measurement of acceptor fluorescence after donor excitation. However, this approach is slightly more complex, as several corrections must be introduced into the analysis of the fluorescence signal, namely direct acceptor fluorescence upon donor excitation, and the differences in extinction coefficients between donors and acceptors must be accounted for. Generally, FRET efficiencies obtained through the monitoring of donor fluorescence quenching are more accurate (Lakowicz 2006).
Transient-state fluorescent data presents several advantages relative to steady-state FRET measurements. Apart from the additional kinetic information that can be related to particular donor–acceptor distributions (Fernandes et al. 2006b), time-dependent data is much less prone to artifacts such as light scattering, inner-filter effects, bleaching or contamination with autofluorescence, and errors are significantly minimized. Minimization of errors might be crucial when selective enrichment of lipids around the protein is not able to cause dramatic differences in FRET efficiencies due to the use of donor–acceptor pairs with high Förster radius (R0) (Fernandes et al. 2004). Fluorescent decays can then be integrated and the data analyzed as intensities if necessary.
Energy migration (or homotransfer) between donors is also a matter of concern when attempting to measure FRET efficiencies in lipid bilayers. When donor energy migration takes place, the concentration of this species should be kept to a minimum in the experiment as this phenomenon induces an increase in heterotransfer efficiencies (Subramaniam et al. 2003), i.e., the excitation travels from an initially excited donor that is too far away from the acceptor, and when a shorter distance is reached heterotransfer takes place.
Another problem that must be avoided when choosing fluorescent probes for quantitative FRET measurements is diffusion of the donor molecule during its excited-state lifetime. Fluorophores presenting very long lifetimes will not be static after excitation, and this excited-state diffusion will lead to a bias towards higher FRET efficiencies as the probability of the donor residing at a closer distance from the acceptor during its lifetime is increased relative to the static situation (Thomas et al. 1978). Diffusion coefficients of lipids in lipid bilayers are generally on the order of 10–8 cm2/s, while protein diffusion coefficients can be even lower; in that case, for donor lifetimes higher than a few hundred nanoseconds, FRET can no longer be described by the static regime (Thomas et al. 1978). Most commonly used fluorophores present lifetimes lower than this limit and FRET data obtained from donors and acceptors in lipid bilayers are associated with a static system.
Qualitative FRET studies
Qualitative FRET studies for protein–lipid binding proved to be useful in the characterization and discrimination of lipid binding sites of the acetylcholine receptor (AChR). Antollini and Barrantes (1998) measured the effect of adding different lipids on the FRET efficiencies from the tryptophan residues of this protein to the fatty-acid derivative, 6-lauroyl-2-dimethylaminonaphthalene (laurdan). When the 18:1 fatty acid was added, a maximum decrease in FRET of 60% was observed. Addition of cholesterol led to a 35% decrease in FRET, while 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) was slightly less effective (25%). Adding cholesterol and DOPC together led to a decrease in FRET identical to that observed with the 18:1 fatty acid, suggesting the presence of two types of lipid binding sites in the protein, inaccessible to either cholesterol or phospholipids, while always accessible to fatty acids (Antollini and Barrantes 1998)
Also for AChR, Narayanaswami and McNamee (1993) showed through FRET measurements from the Trp residues to cis- and trans-parinaric acid that the receptor accommodated preferentially the cis-parinaric acid in its vicinity, suggesting preferential interaction of the protein with less rigid lipids. In another study using the fluorescent sterol cholestatrienol as the FRET acceptor from tryptophan residues in rhodopsin, it was shown that the protein exhibited greater affinity for cholesterol than for ergosterol, since the former was much more effective in decreasing the efficiency of FRET to cholestatrienol, likely through competition for a common binding site in the protein (Albert et al. 1996).
Hughes et al. (2002), using fluorescence lifetime imaging (FLIM), showed that an isoform of phospholipase D (PLD1b) tagged with GFP in HeLa cells, which is responsible for the conversion of phosphocholines (PC) to phosphatidic acid (PA), was susceptible to FRET from BODIPY-labeled PC molecules, but BODIPY-phosphoethanolamine (PE) was not able to induce a change in GFP lifetimes, suggesting preferential interaction of PLD1b with PC lipids. Catalytic mutants of PLD1b did not experience FRET from BODIPY-PC, reinforcing this conclusion.
Qualitative FRET studies can also be the only correct option when highly complex protein–lipid structures are formed such as in protein-mediated membrane fusion. In these cases, unless highly organized structures are formed, such as the lamellar structures induced by lysozyme (Coutinho et al. 2008), no analytical solution for FRET is adequate and recovery of very detailed information from FRET is impossible. That was the case for the interaction of the N-terminal amphipathic alpha-helix of a Bin–Amphiphysin–Rvs (BAR) domain with lipid bilayers (Fernandes et al. 2008).
Absence of noninteracting species in FRET analysis
This strategy was applied to the study of phosphatidyl-d-myo-inositol 4,5-bisphosphate (PIP2) sequestration by the basic effector domain of myristoylated alanine-rich C kinase substrate (MARCKS) (Gambhir et al. 2004) and by a basic peptide from phospholipase C-ξ (Nomikos et al. 2007). Controls were performed to ensure that the concentration of BODIPY-labeled PIP2 was low enough that the probability of occurrence of donor–acceptor pairs in the absence of specific interactions was negligible. Using 1% N-(6-tetramethylrhodaminethiocarbamoyl)-1,2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine (TRITC-PE) (from total lipid content), a lipid species that is not sequestrated by the protein (and therefore whose distribution around the protein is random), energy transfer to the Texas Red labeled MARCKS peptide was still substantial (Gambhir et al. 2004). The Förster radii (R0) for FRET to Texas Red from rhodamine or BODIPY are very similar, and comparable energy transfer efficiencies are expected for random distribution of the two labeled lipids. A concentration of 0.1% was then chosen for the BODIPY-labeled lipid, as at this concentration no FRET was detected using the TRITC-PE probe. In these conditions, all FRET observed from BODIPY-PIP2 to the Texas Red labeled peptide could be assigned to sequestration of PIP2 lipids by the peptide. In order to calculate a dissociation constant for this interaction, the value for the energy transfer efficiency in the protein–lipid complex was assumed to be 100%, i.e., complete quenching of the donor-labeled lipid after interaction of the protein. This approximation will result in underestimation of binding, and the dissociation constant retrieved can be considered as an upper limit for its real value (Gambhir et al. 2004; Nomikos et al. 2007).
Protein dimensions are often comparable to or larger than typical R0 values (20–60 Å). The presence of a transmembrane protein in the membrane leads to an exclusion of lipids from a significant fraction of the area around the protein axis. In that case, FRET efficiencies between protein and lipid species are smaller than would be expected from donors and acceptors with negligible size, especially for donor/acceptor pairs presenting low R0 (Fig. 1b). If the protein is modeled as a cylinder, this lipid exclusion effect can be described by a protein exclusion radius (Re). In several protein–lipid selectivity studies it is assumed that, due to large Re values, FRET efficiencies to or from unbound lipid species can be neglected when R0 < Re (Pap et al. 1993; Antollini et al. 1996; Levi et al. 2000, 2003; Bonini et al. 2002). Using this rationale, the properties of protein–lipid interaction can be extracted from energy transfer data even at high concentrations of the interacting species, allowing for quantification of binding affinities in the absence of strong and specific lipid binding sites, which is the case of lipid binding to annular sites in the protein. However, as is clearly visible from Fig. 1b, significant energy transfer will still occur to acceptors located past the first shell of lipids around the protein, unless R0 ≪ Re and acceptor concentration is kept low. Moreover, if a fraction of the donor or acceptor groups in the protein are located in the periphery of the molecule, this lipid exclusion effect becomes less significant.
In addition to the geometrical parameters, one crucial introduction to the model was that of an interaction parameter Kr, which represents the apparent dissociation constant of laurdan for the lipid belt region, that is, the ratio of the dissociation constant of laurdan over that of the unlabeled lipid. Kr < 1 implies preferential location of laurdan in the lipid belt region, whereas Kr > 1 denotes laurdan’s exclusion from this region. From spectral data, the Förster radius for the tryptophan–laurdan pair was found to be R0 = 2.9 nm. Fixing this value, H was allowed to vary between 0 and 1 nm based on previous results, and it was found that within this range of H values an exclusion distance of r = 1.4 ± 0.1 nm provided an adequate fit to the data, together with Kr ≅ 1. Because the data are not analyzed globally (i.e., the parameters are not optimized simultaneously), it could of course be argued whether a different set of (H, r, Kr) values might fit the data equally well. Naturally, as with all formalisms for lipid selectivity, this method works best if good estimates are known for H and r, and the sole optimizing parameter is Kr. The same methodology was applied to a fluorescent derivative of sphingomyelin (N-[10-(1-pyrenyl)decanoyl]sphingomyelin), which was found to exhibit moderate selectivity for the annular region, with Kr ≅ 0.55 (Bonini et al. 2002). Upon sphingomyelinase digestion of the membrane, FRET efficiency increased by about 50%, indicating that the resulting pyrenyl-ceramide species has higher affinity for the protein than the parental sphingomyelin derivative. On the other hand, a FRET analysis using the described methodology was instrumental in a multitechnique (infrared and fluorescence spectroscopies, differential scanning calorimetry) study which showed that AChR causes formation of specific phosphatidic-acid-rich lipid domains, which include the protein (Poveda et al. 2002). Egg phosphatidylcholine (egg-PC)/1,2-dimirystoyl-sn-glycero-phosphatidic acid (DMPA)/cholesterol (2:1:1) vesicles were prepared in absence of protein, and no evidence of domain formation was verified in these conditions. However, reconstitution of AChR in this lipid mixture leads to enhanced selectivity of the protein for gel-phase probe trans-parinaric acid (Kr = 0.6 ± 0.1), denoting enrichment of gel-phase lipid DMPA in the vicinity of AChR. This is a specific effect for phosphatidic acid, which is not observed for other phospholipid classes (phosphatidylcholine, phosphatidylserine, and phosphatidylglycerol). In particular, when the FRET experiment was carried out replacing DMPA with 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC), uniform acceptor distribution was inferred (Kr = 1.0 ± 0.1, negative control). On the other hand, in the absence of protein, all the phospholipid classes, including phosphatidic acid, exhibit ideal mixing behavior. Because PA and cholesterol have been implicated in functional modulation of the reconstituted AChR, the authors raised the hypothesis that such a specific modulatory role could be mediated by domain segregation of these lipid classes.
FRET analysis including contributions from bulk acceptors
The model introduced by Fernandes et al. (2004) is not subject to the limitations mentioned for the formalisms described above. Through the use of detailed information concerning the position of donor and acceptor labels, the authors showed that it was possible to recover accurate selectivity constants for lipid binding in the annular shell of lipids around the M13 major coat protein (MCP). The problem of lipid selectivity by the M13 MCP has been extensively addressed through electron spin resonance (ESR) techniques (Wolfs et al. 1989; Peelen et al. 1992), and selectivity constants were recovered for the interaction of the protein with different phospholipids. M13 MCP is composed of a single transmembrane segment and as such is expected to immobilize around 10–12 phospholipids in its immediate vicinity or annular binding sites (Marsh and Horváth 1998; Cornea et al. 1997). However, due to MCP aggregation, only five annular binding sites were identified by protein molecule when using ESR (Wolfs et al. 1989). Optimization of the protein purification procedure allowed monomeric MCP to be obtained after membrane incorporation (Spruijt et al. 1989) but failed to produce long-lived immobilization of the annular shell of lipids (Sanders et al. 1992). This type of immobilization is essential in order to distinguish the ESR spectra of immobilized spin-labeled lipids from the labeled lipids probing the bulk lipid (unaffected by the protein).
Labeled phospholipids relative association constants toward M13 major coat protein (adapted from Fernandes et al. 2004)
One important difference between the ESR and FRET techniques is that the latter is not dependent on lipid immobilization and therefore is not restricted to lipids adjacent to a given protein molecule. Not only labeled lipids in the first shell of lipids will be potential acceptors to a donor-labeled integral protein; acceptors in the other lipid shells surrounding the protein will also contribute to the final result. For that reason, this study apparently confirms the hypothesis that single transmembrane domains are only able to influence lipid composition in the first shell of lipids around it. It is likely, however, that larger proteins are able to induce the formation of lipid enrichment at larger distances from the protein.
This formalism presents several advantages and disadvantages. An advantage, relative to the formalism of Fernandes et al. (2004) as originally derived, is that it is directly applicable to proteins of any given size (as long as the protein can be approximated by a cylinder, with the donor located on its axis). The most important disadvantage probably resides in the complex analytical theory underlying the model, which forces most users to use the user-friendly empirical fitting functions given in the paper. Inevitably, the fitting functions’ parameters can only be presented for discrete values of the model parameters. Therefore, approximations have to be made in most cases (the only alternative would be to avoid the fitting functions and to calculate the FRET efficiency by numerical computation of the exact solution).
The model described in the previous section is reminiscent of an older treatment derived by Polozova and Litman (2000) for analysis of FRET between phospholipids labeled with pyrene and the rhodopsin retinal group in model membranes with different lipid compositions. To account for selectivity of rhodopsin for pyrene-labeled lipids, the authors developed a cluster model in which the protein (acceptor) is surrounded by a homogeneous cluster of lipids presenting a different donor lipid concentration than the bulk membrane. In the model, these clusters are circular and characterized by a cluster radius (Rc). An important difference from the FRET viewpoint, as will be commented below, is that the donor is now the lipid species whereas the acceptor is the protein.
While there is no doubt that the results fully support the authors’ conclusions, the quantitative validity of the formalism seems doubtful. Equations 8–10 are based on the derivation by Fung and Stryer (1978) of the kinetics of FRET in planar geometry. The latter authors originally derived the expression of the FRET S term (used here in Eqs. 9–10) by accounting for the interactions between one given donor molecule (all donors being equivalent) and all acceptors surrounding it. This was done by introducing the distribution function of donor–acceptor distances for uniform two-dimensional geometry. Now consider a circular cluster with an acceptor molecule in the center and donors surrounding it. It is clear that the donors within the cluster are not equivalent (unless they are all located at the same distance to the acceptor, which is generally not the case, and obviously not valid for the multiple annular lipid layers reported), and there is no circular symmetry or uniform distribution of acceptors surrounding each of them. Therefore, the FRET term for donors belonging to a cluster (Eq. 9) is not correct, and the effect of this inaccuracy in the results presented for Rc and the ratio of concentration of lipid species in the clusters is not clear.
The sensitivity of FRET over distances of typically 10–100 Å is particularly suitable for studies of protein interactions, as protein dimensions are often in this range. The extraordinary sensitivity of fluorescent techniques was responsible for the widespread use of FRET for detection of protein interactions, a trend that was highly accentuated with the use of fluorescent proteins and genetic engineering (Tsien 1998). However, the application of FRET methodologies to the study of association between membrane components has been generally limited to high-affinity interactions, such as protein–protein interactions or those typically observed for membrane proteins and specifically bound lipids. For FRET studies of lower-affinity interactions between membrane proteins and lipids it is very likely that noninteracting species contribute to the FRET process, increasing the complexity of FRET data analysis. Quantitative studies in these conditions will either rely on meeting specific conditions that allow for disregard of energy transfer to noninteracting acceptors or on an accurate description of this FRET contribution.
The different examples described in this review illustrate the usefulness of FRET in the quantification of protein–lipid selectivity. FRET’s lack of dependence on direct contact also allows for probing of specific lipid enrichment away from the first shell of lipids around the protein, a property absent from other techniques used for protein–lipid selectivity studies which depend on protein–lipid contact. The extension of lipid enrichment around the protein to several lipid shells is likely to occur for large proteins and has the potential to induce significant heterogeneity in membrane composition and to originate membrane compartmentalization, a phenomenon that is likely to be of high biological relevance. These factors, together with the general advantages of fluorescence measurements, reinforce the value of FRET studies in the characterization of protein–lipid interactions.
Financial support for this work was provided by Fundação para a Ciência e Tecnologia (Portugal).
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- Antollini SS, Soto MA, Bonini de Romanelli I, Gutiérrez-Merino C, Sotomayor P, Barrantes FJ (1996) Physical state of bulk and protein-associated lipid in nicotinic acetylcholine receptor-rich membrane studied by laurdan generalized polarization and fluorescence energy transfer. Biophys J 70:1275–1284CrossRefPubMedGoogle Scholar
- Fernandes F, Prieto M, Loura LMS (2009) Membrane microheterogeneity: Förster resonance energy transfer characterization of lateral membrane domains (companion paper)Google Scholar
- Narayanaswami V, McNamee MG (1993) Protein–lipid interactions and Torpedo californica nicotinic acetylcholine receptor function. 2. Membrane fluidity and ligand-mediated alteration in the accessibility of gamma subunit cysteine residues to cholesterol. Biochemistry 32:12420–12427CrossRefPubMedGoogle Scholar
- Pap EHW, Bastiaens PIH, Borst JW, van den Berg PAW, van Hoek A, Snoek GT, Wirtz KWA, Visser AJWG (1993) Quantitation of the interaction of protein kinase C with diacylglycerol and phosphoinositides by time-resolved detection of resonance energy transfer. Biochemistry 32:13310–13317CrossRefPubMedGoogle Scholar
- Peelen SJCJ, Sanders JC, Hemminga MA, Marsh D (1992) Stoichiometry, selectivity and exchange dynamics of lipid–protein interaction with bacteriophage M13 coat protein studied by spin label electron spin resonance. Effects of protein secondary structure. Biochemistry 31:2670–2677CrossRefPubMedGoogle Scholar
- Subramaniam V, Hanley QS, Clayton AHA, Jovin TM (2003) In: Marriott G, Parker I (eds) Methods in enzymology: biophotonics part A: photophysics of green and red fluorescent proteins: implications for quantitative microscopy Academic Press, San DiegoGoogle Scholar
- Wolfs CLAM, Horváth LI, Marsh D, Watts A, Hemminga MA (1989) Spin label ESR of bacteriophage M13 coat protein in mixed lipid bilayers. Characterization of molecular selectivity of charged phospholipids for the bacteriophage M13 coat protein in lipid bilayers. Biochemistry 28:995–1001CrossRefGoogle Scholar