|Home | About | Journals | Submit | Contact Us | Français|
Fluid supported lipid bilayers provide an excellent platform for studying multivalent protein-ligand interactions because the two-dimensional fluidity of the membrane allows for lateral rearrangement of ligands in order to optimize binding. Our laboratory has combined supported lipid bilayer-coated microfluidic platforms with total internal reflection fluorescence microscopy (TIRFM) to obtain equilibrium dissociation constant (KD) data for these systems. This high throughput, on-chip approach provides highly accurate thermodynamic information about multivalent binding events while requiring only very small sample volumes. Herein, we review some of the most salient findings from these studies. In particular, increasing ligand density on the membrane surface can provide a modest enhancement or attenuation of ligand-receptor binding depending upon whether the surface ligands interact strongly with each other. Such effects, however, lead to little more than one order of magnitude change in the apparent KD values. On the other hand, the lipophilicity and presentation of lipid bilayer-conjugated ligands can have a much greater impact. Indeed, changing the way a particular ligand is conjugated to the membrane can alter the apparent KD value by at least three orders of magnitude. Such a result speaks strongly to the role of ligand availability for multivalent ligand-receptor binding.
Phospholipid membranes demarcate the interface between the inside of living cells and their external environment. Their two-dimensional fluidity enables a great variety of complex ligand-receptor binding processes to occur. This includes pathogen attack, the trafficking of lymphocytes, the response of the immune system, and the signaling of cells (Choi, 2004; Kiessling et al., 2006). Indeed, multivalent binding creates expanded contact between binding surfaces (Mammen et al., 1998). As such, multivalent strategies have been used clinically to improve immunotherapeutics for cancer treatment (Sulchek et al., 2005).
The binding of proteins and viruses with multiple binding sites to membrane interfaces typically results in the reorganization of membrane-bound components. Many inhibitory drugs function by disrupting these interactions, especially by binding to components within the membrane (Liu et al., 2006; Morgenstern, 1998). Understanding multivalent interactions as a function of membrane chemistry should provide critical insight into biological recognition. The information gleaned from such work could be used to improve drug design including the inhibition of viral entry and the termination of cancer metastasis (Long et al., 2008). A proper understanding of multivalent ligand-receptor interactions is also of great importance for the development of highly selective biosensor devices that exploit these interactions.
Various model systems have been used to mimic the cell surface environment in multivalent binding studies. Three common platforms are self-assembled monolayers (SAM) (Rao et al., 1999; Smith et al., 2003), phospholipid vesicles (Lee et al., 2005), and supported lipid bilayers (Doyle et al., 2003; Pisarchick and Thompson, 1990; Yang et al., 2003). Both vesicles and supported bilayers are composed of the same lipid molecules found in the plasma membranes of living cells. Moreover, they possess the same two-dimensional fluidity as cell membranes (Cremer and Boxer, 1999), and are capable of mimicking the lateral rearrangements that take place in vivo on the cell surface (Hlavacek et al., 1999). For studies of multivalency, the supported bilayer systems are particularly useful because they can be interrogated by a wide variety of surface specific microscopies and spectroscopies. It is known that a thin water layer (approximately 0.5-1.5 nm in thickness) generally resides between the lower leaflet of a supported bilayer and the underlying substrate. This enables individual lipid molecules to facilely translate along the surface (Kim et al., 2001). Therefore, several ligand molecules within the membrane can bind to a single aqueous protein with multiple binding sites simply by undergoing two-dimensional rearrangement (Fig. 1). Since a wide variety of lipid-conjugated ligands can be incorporated into the membrane, it is possible to study the effects of their specific chemistry and presentation on multivalent binding in a highly controlled manner. Results for several different hapten-antibody and ganglioside-toxin interactions are provided below. These model system studies demonstrate that ligand presentation is more important than ligand density in determining the overall protein affinity for the membrane surface.
Exploiting microfluidic devices for the quantitative investigation of multivalent ligand-receptor interactions in lipid membranes was established by our laboratory over the past decade (Yang et al., 2001). Traditional binding measurements had previously been done using a standard flow cell geometry (Kalb et al., 1990). Such experiments usually required long periods of time to make sequential binding measurements as well as large sample volumes of protein solutions. Consequently, limited information about ligand-receptor interactions could be abstracted from a given set of measurements. By contrast, microfluidic platforms provided a high throughput/low sample volume approach to such measurements. Moreover, binding data at multiple protein concentrations can be collected simultaneously. Therefore, these methods often avoid several sources of noise associated with temporal variations in illumination intensities from an arc lamp source as well as detector drift.
It should be noted that multivalent ligand-receptor binding interactions have been studied by a wide variety of techniques from total internal reflection fluorescence microscopy (Pisarchick and Thompson, 1990) and isothermal titration calorimetry (Goins and Freire, 1988) to surface plasmon resonance spectroscopy (Terrettaz et al., 1993) and quartz crystal microbalance analysis (Janshoff et al., 1997). Flow cytometry (Lauer et al., 2002), fluoroimmunoassays (Singh et al., 2000), fluorescence resonance energy transfer (Ma and Cheng, 2006), atomic force microscopy (Rinker et al., 2008; Sulchek et al., 2005), and a colloid particle phase transition method (Baksh et al., 2004) have also been used. We have largely relied on fluorescence-based methods, which are compatible with our microfluidic approach. The typical setup employed in these ligand-receptor binding studies is illustrated in Fig. 2. Generally, the same bilayer chemistry is present in each channel, but the bulk solution contains various protein concentrations. By imaging all the microchannels simultaneously, it is possible to watch an entire binding curve evolve over time (Jung et al., 2008).
The specific binding of antibodies to their target antigens on cell membrane surfaces is a key step in immune response. A common approach for studying antibody-antigen interactions has been to use phospholipid membranes containing hapten-conjugated phosphatidylethanolamine (Thompson et al., 1993). Common model haptens include dinitrophenyl (DNP), trinitrophenyl (TNP), nitroxide, dinitrophenyl nitroxide, fluorescein, and biotin. All of these moieties have been found to be active in antibody binding studies (Hafeman et al., 1981; Pisarchick and Thompson, 1990; Tamm, 1988; Tamm and Bartoldus, 1988; Wright et al., 1988).
Antibodies may bind either monovalently or bivalently to the corresponding antigenic surface. McConnell and co-workers first claimed that IgG molecules interact with the membrane through both of their binding sites (Parce et al., 1979). Thompson and co-workers estimated that 10-50 % of the IgGs were bound monovalently in their studies (Pisarchick and Thompson, 1990). These investigators also found that the apparent equilibrium dissociation constant (KD) of the anti-DNP antibody was ~10 fold stronger than the corresponding Fab fragment under conditions with very high lipid-conjugated DNP concentrations. It should be noted, however, that ligand density should play a role in ligand-receptor bind affinity since both the ligand distribution and inter-ligand distance (Huskens et al., 2004) might affect the ability of IgG molecules to become bivalently bound to the membrane.
The dependence of KD values for antigen-antibody binding on supported membranes was first noted by Tamm and co-workers from studies at three different ligand densities (Kalb et al., 1990; Tamm and Bartoldus, 1988). Our laboratory made binding isotherm measurements for the DNP/anti-DNP system at eleven different ligand densities ranging from 0.1 to 5 mol % by employing microfluidic platforms (Yang et al., 2003). These experiments demonstrated that the KD value tightened in a systematic manner according to a hyperbolic function as the ligand densities was increased from 0.1 to 5.0 mol % (Fig. 3). Specifically, we were able to abstract the individual binding constants for the sequential two-step binding process: KD1 = 25 μM and KD2 = 1.38 × 10-6 mol/dm2. Such results clearly indicate that the binding of an antibody to a membrane surface is enhanced by the second binding event. This corresponds to a positively cooperative binding effect with an entropic difference between the two binding events of ΔS = 24.1 J/(mol·K). Although IgG molecules are largely monovalently bound at low hapten density (74% at 0.1 mol% of DNP), nearly all the antibodies are bivalently bound at higher ligand density (91% at 5 mol% DNP). Moreover, it was found that increasing the density of DNP-PE in phospholipid membranes from 0.1 to 5.0 mol strengthened the apparent KD value by roughly a factor of 13 (Yang et al., 2003). This is a rather modest increase given the fact that the ligand density was increased by a factor of 50.
These results could be explained by ligand-ligand interactions involving the clustering of the glycolipids into islands within the bilayer (Fig 4B). Indeed, corresponding atomic force microscopy studies confirmed the presence of ganglioside islands. The diameter of these islands increased with increasing ligand concentration. In fact, the change in mean island diameter was highly correlated with the weakening of the KD value found from the binding measurements. Such a finding is strong evidence that ligand-ligand interactions within the membrane can actually inhibit binding for strongly interacting ligands. GM1-GM1 interactions are undoubtedly related to hydrogen bonding between adjacent oligosaccharide headgroups (Peters and Grant, 1984; Sharom and Grant, 1978).
The results for the GM1-cholera toxin system are highly significant because they demonstrate that merely increasing ligand density within a membrane is no guarantee that the binding affinity for a multivalent protein will increase. It has been hypothesized that ligand densities within cell membranes can vary. For example, certain ligands may be present at higher concentrations within lipid raft regions. Such higher number densities may increase the number of proteins that are bound, even if the free energy of binding were somewhat attenuated (van der Goot and Harder, 2001; Wolf et al., 1998).
Although our cholera toxin-GM1 studies showed evidence for binding inhibition with increased ligand density, the effects were not very pronounced. Indeed, one order of magnitude weakening or tightening in the dissociation constant is rather modest. Moreover, relatively large changes in ligand density were required to achieve such effects. Therefore, one must search for other factors that can influence the strength of multivalent binding between aqueous proteins and membrane surfaces. To address this question, studies of ligand presentation and lipophilicity were undertaken. It should be noted that the cell membrane interface itself has been thought to affect ligand presentation (Leckband et al., 1995). In fact, ligand molecules may interact with the interior of the fluid lipid bilayers, rendering them less available to incoming proteins (Balakrishnan et al., 1982a; Balakrishnan et al., 1982b). Moreover, since cell surfaces are typically covered by a glycocalix (Schneider et al., 2002), interference with binding by aqueous proteins may occur (Albersdorfer et al., 1997) (Dori et al., 2000; Kim et al., 2000; Moore and Kuhl, 2006). These qualitative notions led us to ask several questions. First, how large of an influence can ligand lipophilicity have upon protein binding? Second, how large could the effect of a glycocalix mimic be on protein binding? Finally, is it possible to substantially alter the free energy of binding through the placement of ligands at various positions within the membrane?
To answer the questions posed above, antibody-antigen binding experiments were performed at phospholipid membrane surfaces with two different haptens (Jung et al., 2008). The first hapten was biotin, which has an octanol-water partition coefficient (log P value) of 0.11. The second hapten was DNP with a partition coefficient of 1.5. Therefore, the DNP is expected to be far more lipophilic than the biotin moiety. In a first set of experiments, KD values for IgG binding to the respective hapten moieties were determined in bulk solution by fluorescence anisotropy measurements (Fig 5B). This pronounced difference can be attributed to the relative propensity of the DNP moiety to become buried within the aliphatic portions of the lipid bilayer and thereby be unavailable to the incoming anti-DNP antibody. By contrast, the biotin contains an ureido moiety (-NH-CO-NH-), which should lead to a preference for residing in the aqueous phase.
In a second set of experiments, we wished to see whether the binding affinities for anti-biotin and anti-DNP could be reversed by changing the way their respective haptens are presented on the surface of the phospholipid bilayer. To weaken anti-biotin binding, a polyethylene glycol (PEG) cushion was added to the membrane through direct conjugation of the polymer chain to phosphatidylethanolamine headgroups (Fig 5C). This strengthened the KD value by roughly two orders of magnitude. Such results are in line with the notion that that a sufficiently long, flexible, and hydrophilic ethylene oxide linker can maximize the binding interaction (Ahlers et al., 1992; Leckband et al., 1995). On the other hand, short hydrophobic spacers should weaken binding (Cooper et al., 1981; Kimura et al., 1990).
The results described in the studies outlined above clearly demonstrate that ligand presentation in simple model systems is generally more effective for altering interfacial binding events, as compared to modulating ligand density or clustering. These findings might have implications for the binding of proteins at the surfaces of cell membranes. However, future work will need to be undertaken in vivo to test the hypothesis that the ligand's lipophilicity or its presentation within the cell membrane's glycocalix are indeed more important factors in determining binding strength than either its number density or its ability to cluster with neighboring moieties.
The thermodynamic binding process is now well understood for a few model systems such as the ones discussed here. Nevertheless, a general understanding of the molecular level mechanisms governing multivalent binding interactions and the associated lateral rearrangements which take place within membranes is still far from complete. One intriguing direction is to try to obtain data at the single molecule level as individual binding events are inherently stochastic. Recently, multivalent biomolecular interactions have been visualized at the single molecule level (Rinker et al., 2008). Moreover, individual monovalent and bivalent events have been separately detected using picoamp current measurements in association with protein pores (Howorka et al., 2004). Finally, it should be noted that cell membranes represent complex heterogeneous mixtures of ligands and receptors with the potential for many simultaneous interactions. Therefore, future studies with model systems will need to ever more closely mimic the actual cellular environment in order to fully appreciate the role that lipid and protein chemistry play in these binding events.
We thank Dr. Tinglu Yang for useful discussions. This work was funded by the National Institutes of Health (GM070622) and the Robert A. Welch Foundation (grant A-1421).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.