Association of Model Neurotransmitters with Lipid Bilayer Membranes

Abstract

Aimed at reproducing the results of electrophysiological studies of synaptic signal transduction, conventional models of neurotransmission are based on the specific binding of neurotransmitters to ligand-gated receptor ion channels. However, the complex kinetic behavior observed in synaptic transmission cannot be reproduced in a standard kinetic model without the ad hoc postulation of additional conformational channel states. On the other hand, if one invokes unspecific neurotransmitter adsorption to the bilayer—a process not considered in the established models—the electrophysiological data can be rationalized with only the standard set of three conformational receptor states that also depend on this indirect coupling of neurotransmitters via their membrane interaction. Experimental verification has been difficult because binding affinities of neurotransmitters to the lipid bilayer are low. We quantify this interaction with surface plasmon resonance to measure equilibrium dissociation constants in neurotransmitter membrane association. Neutron reflection measurements on artificial membranes, so-called sparsely tethered bilayer lipid membranes, reveal the structural aspects of neurotransmitters’ association with zwitterionic and anionic bilayers. We thus establish that serotonin interacts nonspecifically with the membrane at physiologically relevant concentrations, whereas γ-aminobutyric acid does not. Surface plasmon resonance shows that serotonin adsorbs with millimolar affinity, and neutron reflectometry shows that it penetrates the membrane deeply, whereas γ-aminobutyric is excluded from the bilayer.

Significance

Receptor ion channels in the postsynaptic membrane and their neurotransmitter agonists enable fast communication between neuronal cells. Electrophysiology studies reveal surprisingly complex kinetics that apparently requires a variety of protein conformational states for its quantitative interpretation, but an alternate hypothesis invoking neurotransmitter membrane association reduces the complexity of the underlying reaction schemes significantly. Although their affinity may be low and is hard to quantify experimentally, neurotransmitter membrane association can be relevant because of their large temporary concentration in the synaptic cleft. With thermodynamic and structural measurements, we quantify membrane-bound states of serotonin, establishing this neurotransmitter as membrane binding, whereas the affinity of the more hydrophilic γ-aminobutyric acid is too low to register in our sensitivity-optimized measurement techniques.

Introduction

Synaptic transmission between neurons is mediated by ligand-gated ion channels embedded in the postsynaptic membrane. These proteins selectively allow either cations (e.g., acetylcholine and glutamate receptors) or anions (e.g., γ-aminobutyric acid (GABA) and glycine receptors) to flow into the postsynaptic neuron (1). Conventional kinetic reaction schemes require a surprisingly large number of protein conformational states (2, 3, 4, 5, 6), yet they do not accurately model experimental results (7). For example, desensitization of their ion-conducting conformation decreases with complex kinetics, requiring multiple decay exponentials for data modeling. Although the molecular origin of this complex response is not known, it must be essential to the proper functioning of the central nervous system because organisms with mutations that alter desensitization kinetics exhibit major phenotypic changes detrimental to the health of the organism (1). To address these issues, an alternative hypothesis has been developed that simplifies the kinetic models that describe this complex gating behavior. It posits that neurotransmitters (NTs), which activate postsynaptic receptors through binding to specific sites, also modulate receptor activity indirectly (8, 9, 10). This hypothesis invokes NT interactions with the postsynaptic lipid bilayer surrounding the receptor proteins, which may alter its physical properties and, in turn, modulate receptor activity (11). For a specific case, the interaction of GABA with its receptor GABA_A, a minimal kinetic scheme based on these ideas described experimental agonist-induced and anesthetic-induced activation-deactivation traces of ion flow across the membrane remarkably well (9). Additional support for this hypothesis came from the observation that the response of a receptor to its cognate NT is altered by the presence of a different, unrelated NT (12). Although this provided an impressive validation of the unconventional receptor activation concept, it does not directly demonstrate that NT-membrane interactions do indeed occur.

Various experimental approaches to measure nonspecific NT binding to lipid bilayers have been reported. Such studies showed that the cationic acetylcholine and the zwitterionic GABA and glycine NTs bind to anionic membranes containing phosphatidylserine (PS) or phosphatidylglycerol (PG) (13) and that the aromatic 5-hydroxytryptamine (5-HT; serotonin) has a greater partitioning coefficient into such membranes than polar NTs (14). Atomistic molecular dynamics (MD) simulations suggested that serotonin adsorbs to the anionic phosphate in phosphatidylcholine (PC) through interaction of its cationic primary amine and is then anchored by aligning its aromatic ring with the hydrocarbon tails. This association has been observed to shield against oxidation in erythrocytes (15). The serotonin-derived hormone melatonin associates with planar bilayers in a similar orientation; however, this has only been measured qualitatively, not quantitatively (16,17). Another study indicated that the electrostatic interaction of dopamine with the inner leaflet of presynaptic vesicles must be screened to prevent NT aggregation and thus facilitate synaptic release (18).

In MD simulations, hydrophilic NTs (e.g., GABA and serine) are generally observed to remain in the solution adjacent to a membrane or interact weakly with charged headgroups, whereas hydrophobic species (e.g., serotonin and dopamine) partition into the bilayer and often localize near the lipid phosphates (19, 20, 21). Postila and co-workers studied a broad range of NTs for their membrane interactions and reported a strong correlation between their membrane-binding prevalence and the location of the binding sites on their cognate receptors (19). NTs with extracellular binding sites on the receptor showed low affinity, but NTs whose binding sites are hidden within the membrane—typically in the lipid headgroup region—consistently showed high membrane affinities and were classified as membrane binding. If these observations from simulation can be confirmed with experimental data, this will provide strong support for the hypothesis that the bilayer plays a critical role in the regulation of neurotransmission: whereas hydrophilic NTs bind directly to their binding sites after diffusing across the synaptic cleft, hydrophobic NTs may first adsorb to the bilayer and subsequently encounter their receptors through two-dimensional (2D) diffusion.

Here, we investigate the interactions of two prototypical NTs—serotonin, which was identified as a membrane-binding NT, and GABA, for which simulation results showed low membrane affinity—with model membranes as a function of lipid headgroup charge density. Surface plasmon resonance (SPR) provides quantitative measurements of membrane affinities and shows striking differences between serotonin and GABA. These thermodynamic results are quantitatively consistent with neutron reflection (NR) results by which we study the impact of NTs on membrane structure and localize NTs in the bilayer. In combination, these results show that the aromatic NT serotonin, a cation at physiological pH, binds neutral and charged membranes with millimolar affinity. It intercalates into the bilayer with a distribution that has its highest density at the depth of the phospholipid headgroups, at a position where it can bind its effector protein efficiently. Beyond this specific interaction, it may affect other membrane-embedded proteins, conceivably by modification of the pressure profile across the bilayer. GABA interaction with bilayers was much weaker: the sensitivity of both methods was insufficient for a quantification of surface-adsorbed amounts of this NT. However, our SPR measurements established an upper limit of its dissociation constant from the membrane that is above the typical concentration of NTs in the synaptic cleft.

Materials and Methods

Materials

1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1-rac-glycerol) (POPG, sodium salt) from Avanti Polar Lipids (Alabaster, AL) were used as received. The tether lipid HC18 [Z20-(Z-octadec-9-enyloxy)-3,6,9,12,15,18,22-heptaoxatetracont-31-ene-1-thiolacetate] (22) was provided by David Vanderah (Institute for Bioscience and Biotechnology Research, University of Maryland, Rockville, MD). Other chemicals, including β-mercaptoethanol (βME), salts, and the NTs GABA and serotonin, were from Sigma-Aldrich (St. Louis, MO).

Preparation of sparsely-tethered bilayer lipid membranes

Sparsely-tethered bilayer lipid membranes (stBLMs) (Fig. 1) were prepared as follows. Self-assembled monolayers (SAMs) composed of βME and HC18 (7:3 in solution) were formed on gold-covered solid substrates, followed by bilayer completion through rapid solvent exchange (23) or vesicle fusion (24). For SPR, we used microscopy slides made of glass (Thermo Fisher Scientific, Waltham, MA) or sapphire (Rubicon, Bensonville, IL). Reflectivity measurements were performed on 3-inch-diameter, 5-mm-thick silicon waters (El-Cat, Ridgefield Park, NJ). The bare substrates were cleaned in 5 vol% Hellmanex solution (Hellma Analytics, Müllheim, Germany), incubated in sulphuric acid with Nochromix (Godax Laboratories, Cabin John, MD) for 15 min, rinsed with ample volumes of ultrapure water (EMD Millipore, Billerica, MA) and ethanol, and dried in an N₂ stream. They were then coated in a magnetron (Denton Vacuum Discovery 550; Moorestown, NJ) with an ∼40 Å thick chromium bonding layer, followed by a terminal gold layer that was ∼450 Å thick for SPR and ∼150 Å thick for NR measurements. In all cases, the root mean-square interfacial roughness values of the gold surfaces were σ < 5 Å.

Figure 1.

The solid-supported membrane model system used in this study and its structural characterization. A sparsely-tethered bilayer lipid membrane (stBLM) consists of a substrate-proximal lipid layer that is tethered to a gold surface (25,62) and a distal layer that is fully fluid (63). Such systems have been extensively characterized with NR, and their out-of-plane structure is described in terms of distributions in which the molecular components occupy volume within the plane of the bilayer, as exemplified in the lower section, and is referred to as component volume occupancy (CVO). The proximal lipid layer contains more chain volume than the distal layer because the tether lipids are ether based and pack more densely than the intervening phospholipids. In the quantitative model, water (distribution not shown) fills any space where the combined CVOs of the organic molecules is less than unity (34,38).

For vesicle fusion on SAMs to achieve bilayer formation for NR experiments, stock lipids dissolved in organic solvents were mixed and dried under vacuum at 50°C for at least 12 h. Vesicles were then formed at a lipid concentration of 5 mg/mL in a concentrated salt solution (1 mol/L NaCl, 100 mmol/L NaH₂PO₄ (pH 7.4)). The solution was sonicated until it became translucent, typically after 90 min. The vesicles were allowed to incubate the tether SAM for 60–90 min before flushing with a low-salt buffer (100 mmol/L NaCl, 10 mmol/L NaH₂PO₄ (pH 7.4)) to assist stBLM formation by osmotic shock. Rapid solvent exchange was used to form stBLMs for SPR (25). 50 μL of lipids in organic solvent (10 mg/mL) were added onto the tether SAM and incubated for ∼1 min. The system was then flushed with at least 10 mL of the low-salt buffer. For both procedures, the bilayers were flushed with subsequent rinses of 5 mL water, 10 mL ethanol/water (20 vol%), 5 mL water, and 10 mL of low-salt buffer. The formation and quality of the stBLMs was assessed with electrochemical impedance spectroscopy (EIS) in terms of its bilayer resistance and capacitance (26), and SPR was used to quantify mass adsorption onto the membrane once the stBLMs were exposed to NTs in solution.

EIS

EIS measurements were performed between 1 Hz and 100 kHz, with 60 points logarithmically spaced, with a Solartron (Farnborough, UK) 1287A potentiostat and 1260 frequency analyzer in a three-electrode configuration. A saturated silver-silver chloride (Ag-AgCl-KCl(aq,sat)) microelectrode (M-401F; Microelectrode, Bedford, NH) served as the reference electrode with a 0.25-mm-diameter platinum wire (99.9% purity; Sigma-Aldrich) coiled around the reference electrode as the auxiliary electrode. The working electrode was secured to the gold substrate with copper conducting tape (SPI Supplies/Structure Probe, West Chester, PA). Data were collected and analyzed using Zplot and Zview (Scribner Associates, Southern Pines, NC) and fitted to models as described (27).

SPR

SPR measurements were performed at T = 25.00 ± 0.01°C in single-batch mode using two custom-built instruments that reflect light at wavelengths λ = 660 and 720 nm from the membrane-covered interfaces and work with glass and sapphire, respectively, as support media (SPR Biosystems, Germantown, MD). Gold-coated slides with an SAM layer were assembled by index matching to a prism (Kretschmann configuration), and stBLMs were prepared in situ as described above. The intensity of light reflected from the gold-buffer interface was recorded in a 2D-CCD detector as a function of time, and the characteristic reflection minimum due to surface plasmon generation, located on the detector at a position R, was determined in the raw signals using SPRAria (SPR Biosystems). To determine a baseline, the neat bilayer was measured for 30 min before adding NTs in increasing concentrations, c. R(t) was recorded for each concentration until equilibrium was approached at R_∞ = R(t → ∞), and the changes in R_∞ as a function of c were fitted to a modified Langmuir isotherm:

$$R_{eq}=\frac{c{R}_{\infty }}{c+K_D}+c\frac{dR}{dn}\frac{dn}{dc},$$ (1)

where K_D is the dissociation constant (28), and dR/dn is the instrument response to changes in refractive index of the bulk solution above the interface. The Langmuir binding model is based upon the assumption that all binding sites are equivalent, each ligand interacts with a single binding site, and ligands do not interact with each other. Because binding is expected to be weak, measurements were performed with millimolar NT concentrations, which made corrections, in the form of the second term on the right-hand side of Eq. 1, necessary to account for changes in the refractive index of the solution. The refractive index increment, dn/dc, was determined using a prism refractometer at λ = 660 nm for blank buffer and known NT concentrations in solution.

For the quantitative interpretation of SPR signals, one may use the transfer matrix method (29) to calculate the shift of the optical reflection minimum in response to the adsorption of a molecular species to an interface whose structure is approximately known. The sample consists of a sequence of homogeneous layers, i, with thicknesses d_i, refraction indices n_i, and optical thickness δ_i = d_ik_i, where k_i is the wave vector component normal to the interface. The coefficients r_{n, n + 1} and t_{n, n + 1} then characterize reflection and transmission at each interface between subsequent layers and give rise to a transfer matrix element (30),

$$M_n=t_{n,n+1}^{-1}\left(\begin{array}{cc}{e}^{-i{\delta }_n}& 0\\ 0& e^{-i{\delta }_n}\end{array}\right)\left(\begin{array}{cc}1& r_{n,n+1}\\ r_{n,n+1}& 1\end{array}\right).$$ (2)

The product of the transfer matrices for all layers define the global optical properties of the composite film

$$\hat{M}=t_{\mathrm{0,1}}^{-1}\left(\begin{array}{cc}1& r_{\mathrm{0,1}}\\ r_{\mathrm{0,1}}& 1\end{array}\right)\prod _i\left(M_i\right),$$ (3)

which relate its total reflectance and transmission as

$$\left(\begin{array}{c}1\\ r\end{array}\right)=\hat{M}\left(\begin{array}{c}t\\ 0\end{array}\right).$$ (4)

With an approximate structure of the interfacial film known from NR and knowledge of the localization of adsorbed molecules with respect to bilayer components, which gives rise to estimates of the optical indices of the individual layers in the interfacial film, one can then relate shifts in SPR signal, Δθ_SPR, to mass adsorption, as shown in the Results.

Neutron reflectometry

NR is a sensitive method to characterize the organization of layered organic materials at interfaces and surfaces along their normal direction, z (31). Neutrons transmitted through the Si slab that supports an stBLM are reflected from its interface with the buffer at a shallow specular angle α with the momentum transfer Q_z = (4π/λ) × sinα, where the neutron wavelength λ is ∼5 Å. If the neutron scattering length densities (nSLDs) of substrate and adjacent liquid phase are ρ_substrate < ρ_buffer, as is the case for Si bordering D₂O-based buffer, the reflectivity R is unity between Q_z = 0 and Q_z^c, the critical momentum transfer for total internal reflection. Beyond Q_z^c, R(Q_z) decays sharply because almost all intensity is transmitted into the buffer. The reflection from an ideal interface follows the Fresnel reflectivity R_F, which decays as (Q_z^c/2Q_z)⁴ for sufficiently large Q_z (32). If the idealized interface is blurred, usually modeled with a Gaussian roughness σ, the reflectivity decays even faster in Q_z because R = R_F × exp(−Q_z²σ²). It is therefore essential to achieve ultralow roughnesses of the substrate, such as σ = 3–5 Å on polished Si/SiO₂.

A molecularly layered organic surface structure between the Si substrate and the buffer modifies R_F because of interferences between subsequent strata—the gold film, submembrane aqueous layer, membrane, and surface-adsorbed molecules—and partial transmission and reflection amplitudes depend on the nSLDs, ρ, of individual layers. The neutron reflectivity can be approximated as R/R_F = ${\left|\int \left(d\rho /dz\right)e^{i{Q}_{z}z}dz\right|}^2/{\rho }_{substrate}^2$ (33). Because of the lack of phase information, the reflectivity cannot be directly inverted into a unique structural model. Yet, because the bilayer structure is approximately known, forward modeling of the data works well in practical terms.

Models of stratified interfacial structures of membranes on solid supports were traditionally constructed from molecular slabs of homogeneous nSLDs (box functions in the one-dimensional profiles; hence, the frequently used term “box models”). More realistic descriptions are provided by models in which these strata are blurred into each other, parameterized as continuously variable distributions using analytic functions (34). Refinement of their parameters then approximates the underlying structure. Unique to the probing of interfacial structures with “cold” neutron beams with kinetic energies in the milli-electronvolt range is the absence of beam damage and thus the ability to perform subsequent measurements on the same sample. For example, as-prepared bilayers under neat buffer and bilayers exposed to NT in solution can be measured subsequently in isotopically distinct buffers. Thus, a sequence of individual measurements forms a large data set, and their evaluation in context of each other restrains the parameters of the model strongly (35).

Measurements at T = 22 ± 1°C were performed on the Magik reflectometer (36) at the National Institute of Standards and Technology (NIST) Center for Neutron Research. Reflectivity curves were recorded for momentum transfer values 0.001 ≤ Q_z ≤ 0.25 Å⁻¹, for which counting statistics with adequate quality were typically obtained after 6 h of beam time per condition. A flow cell (volume ≈ 1.3 mL) allowed for in situ buffer exchange (37), accomplished by flushing the sample with ∼10 mL of buffer. Thereby, sequential measurements on membranes without and with NTs in H₂O- or D₂O-based buffer were performed in repeated scans on the same sample footprint.

One-dimensional component volume occupancy (CVO) profiles (38) along the lipid bilayer normal (cf. Fig. 1) were determined using a hybrid model that comprised a slab parameterization for the solid substrate (39), the continuous distribution of submolecular components for the stBLM (34), and a freeform model based on monotonic Hermite splines for adsorbed NTs. The model of the solid substrate was composed of slabs of bulk silicon, silicon oxide, chromium (deposited as a bonding layer), and the terminal gold film. Fit parameters were their thicknesses and nSLDs. For the stBLM, the submolecular groups were the interfacial βME, polyethylene glycol chains, and glycerol groups of the HC18 SAM. HC18 alkyl chains were not distinguished from the acyl chains of the substrate-proximal phospholipid monolayer they intercalate. The phospholipid bilayers were parsed into their substrate-proximal and -distal PC and PG headgroups, substrate-proximal and -distal polymethylene chains, and the chain-terminal methyls in the bilayer center. Many of these parameters are interdependent, and overall, they were subject to the condition that they fill the available space, with water filling any volume not occupied by the surface architecture. Parameters fitted in the model were bilayer completeness, the surface-immobilized ratios of βME/HC18 and HC18/substrate-proximal phospholipid, the submembrane layer thicknesses, and the hydrocarbon layer thicknesses for each bilayer leaflet (Fig. 1). Volumes of the molecular components were constants in the model, and dependent parameters reveal system properties such as the area per phospholipid in the bilayer (38). Hermite splines, defined by control points separated by 15 Å on average, provided a model-independent description of the in-plane averaged NT distributions along z, interfacing seamlessly with the continuous distributions of stBLM components (34). Volume occupancies at the control points and deviations of their positions from equidistant separation were fitted parameters. A global roughness σ was applied to the substrate surface and all submolecular distributions. Parameter optimization was achieved using the Refl1D and ga_refl software packages developed at the NIST Center for Neutron Research (37). Fit parameters that were not expected to change across measurements, such as the structure of the solid substrate, were determined by a simultaneous fit of all reflectivity curves in a single data set. Parameter confidence limits were determined by a Monte Carlo Markov chain (MCMC)-based global optimizer (37). Median values and 68% confidence limits on the CVO profiles were calculated as a statistical average over parameter sets from all MCMC iterations after chain equilibration. Medians (50% percentile) and 68% confidence limits (50 ± 34% percentiles) of the profiles were calculated for every z and displayed as solid lines (blue for error calibration—see below—and red for NT contributions to the overall profiles) and shaded areas of the same colors, respectively.

Results

Our investigations of NT interactions with phospholipid bilayers, reported below, pushed the limits of sensitivity of both the SPR characterization of adsorption thermodynamics and the structural characterization of NT-bilayer arrangements by NR. This required corrections for index changes of the bulk solution because of high NT concentrations in SPR measurements and a critical calibration of the NR model to account for deficiencies in the parametric descriptions of the interfacial structures. Both issues are described at the beginning of their respective sections.

SPR

To interpret SPR readings correctly at high concentrations of dissolved NT molecules, which affect the optical index n of the buffer, their index increment, dn/dc, was measured. The neat buffer at λ = 660 nm had n = 1.3345. The addition of 100 mmol/L serotonin⋅HCl increased the index to n = 1.339 for dn/dc = 0.0021 ± 0.0001 dL/g or (4.45 ± 0.21) × 10⁻⁵ L/mmol. For GABA, dn/dc = 0.0017 ± 0.0002 dL/g or (1.75 ± 0.21) × 10⁻⁵ L/mmol.

Adsorption isotherms of two model NTs, serotonin and GABA, to stBLMs of different surface charges in which POPC bilayers contained 0, 10, 30, or 50 mol% POPG were determined with SPR. Representative data sets are shown in Fig. 2, and overall results are listed in Table 1. All experiments were performed at least in triplicate and are quoted by their error-weighted means, with uncertainties given as the error-weighted standard deviations. Serotonin adsorption occurred to all bilayer compositions and was detected at bulk concentrations above c ≈ 0.05 mmol/L. At c > 1 mmol/L, bulk phase index increases reached detectable levels, and a correction for dn/dc was necessary to fit the data. The corresponding models of the SPR data in Fig. 2, A–C show the contributions of surface accumulation and bulk correction (red and blue lines in the online version of this article); the sum of the two (black lines) fit the experimental data well. For modeling of the entire isotherm, dn/dc was treated as a fit parameter and ranged from (4.21 ± 0.16) × 10⁻⁵ L/mmol for POPC to (4.64 ± 0.21) × 10⁻⁵ L/mmol for 7:3 POPC/POPG (Table 1), consistent with the value of dn/dc determined for serotonin in buffer by bulk index measurements. The affinity of serotonin to the bilayer surface is constant within experimental error near K_D = 1 mmol/L at low bilayer charge (0 and 10 mol% POPG) and high bilayer charge (50 mol% POPG). Remarkably, it drops by a factor of 2 for the bilayer that contains 30 mol% POPG (Fig. 3 A), the concentration of charged phospholipids characteristic for the inner plasma membrane. The surface accumulation of serotonin, evaluated as R_∞(c → ∞), was approximately constant within the experimental error for all bilayer compositions (Fig. 3 B).

Figure 2.

SPR measurements of stBLMs exposed to NTs in aqueous solutions. Equilibrium SPR responses, R(t → ∞), were converted into equivalent refractive index changes, Δn, of the bulk solution adjacent to the membrane and are plotted in this form as a function of NT solution concentration. The overall signals contain contributions from membrane adsorption and/or intercalation of NT molecules and refractive index changes at high concentrations of dissolved NTs. The two contributions are disentangled by fitting the data to Eq. 1. (A)–(C) show serotonin adsorption to stBLMs of various phospholipid compositions as indicated. An inset in (B) exemplifies an original data set. (D) shows an stBLM incubated with GABA. An inset here shows data for three independently measured bilayers on a linear scale. To see this figure in color, go online.

Table 1.

Fitted SPR Results to Quantify the Adsorption of Serotonin, a Membrane-Binding NT, and GABA, a Membrane-Inert NT, to stBLMs of Different Compositions

stBLM Composition	PC	PC/PG 9:1	PC/PG 7:3	PC/PG 1:1
Serotonin
Dissociation constant, KD (mmol/L)	1.1 ± 0.2	1.6 ± 0.5	0.42 ± 0.16	1.1 ± 0.2
Index change by surface accumulation, R∞ (10−5)	35 ± 3	30 ± 5	30 ± 3	35 ± 3
Bulk index change increment, dn/dc (10−5 L/mmol)	4.21 ± 0.16	4.45 ± 0.15	4.64 ± 0.21	4.29 ± 0.12
model fit quality, χ2	2.122	1.549	2.351	0.966
GABA
Dissociation constant, KD (mmol/L)	n/a	n/a	n/a	n/a
Index change by surface accumulation, R∞ (10−5)	n/a	n/a	n/a	n/a
Bulk index change increment, dn/dc (10−5 L/mmol)	1.77 ± 0.12	1.78 ± 0.16	1.66 ± 0.10	1.73 ± 0.04
Model fit quality, χ2	4.227	3.305	5.938	5.107

Whereas the serotonin results required fitting to both arguments in Eq. 1, GABA results were well modeled by fitting only to the right-hand side of Eq. 1. n/a, not applicable.

Figure 3.

Comparison of serotonin binding to stBLMs that contain varying amounts of POPG. (A) The affinity (1/K_D) shows a maximum at ∼30 mol% of the charged lipid, which coincides with the physiological concentration of anionic lipids—mostly PS—in the inner leaflet of the plasma membrane. (B) The extrapolated mass adsorption (c_NT → ∞), displayed in equivalents of bulk index changes, is independent of bilayer surface charge. Error bars show standard deviations from at least three independent measurements.

In contrast to serotonin, the adsorption of the zwitterionic GABA to pure POPC or POPC/POPG bilayers was below detection limits in our SPR experiments. Fits that only used the second term in Eq. 1.—i.e., the bulk contribution to the signal—were entirely adequate to model the experimental data (Fig. 2 D), with similar quality as the serotonin results fitted by the full adsorption model. This clearly implies that interfacial adsorption of GABA is below the sensitivity of the experiment.

Neutron reflectometry

NR was used to localize NTs adsorbed to stBLMs and quantify their adsorbed masses as a function of bilayer surface charge. Because the determination of adsorbed mass in SPR depends on the location of an adsorbed molecular species with respect to the bilayer, localization of adsorbed NT also permits quantitative estimates of the adsorbed mass as seen by the SPR experiments and thus a cross validation between the methods.

As in the SPR experiments, it was imperative to quantitatively assess the sensitivity of NR to NT adsorption to the interface in each measurement. To rationalize our sensitivity calibration procedure, we assume that there are systematic deviations from ideality of the sample structures in each individual measurement, such as a slight waviness of the substrate that will vary for each sample and is unaccounted for by data modeling. This implies that a description of the as-prepared stBLM—for which the model only contains the slab parameterization of the substrate and the continuous distribution of stBLM components, but no Hermite spline for an adsorbent—may be fraught with displacements of the best-fit parameters from their true values. If, nevertheless, a Hermite spline is added, it will also populate to adjust for such systematic errors, even when there are no adsorbent molecules contained in the sample. Because the same as-prepared bilayer is subsequently exposed to NTs, measured with NR under otherwise identical conditions, and evaluated with the same model, we can then with confidence distinguish the NT CVO distribution in the sample that contains NTs from its systematic uncertainty. Because this approach is sample specific—because each physical sample may contain different structural nonidealities—we repeat this check for each sample, assuming that the Hermite splines observed on the as-prepared bilayers provide quantitative measures of the detection limits for the adsorbent for each particular sample.

We characterized serotonin association with stBLMs containing pure POPC and POPC/POPG (7:3) at various NT concentrations in the buffers. The concentration of the charged lipid component was chosen in view of the SPR results and the fact that charged phospholipids (mostly PS) account for ∼30 mol% in the composition of the plasma membrane. Generally, the as-prepared bilayers were first measured in H₂O-based and in D₂O-based buffers to establish a baseline structure and calibrate the sensitivity of the ensuing experiments on bilayers under NT in solution. We then repeated the NR measurements on the same samples with serotonin at c = 1 and 10 mmol/L on POPC and c = 100 μmol/L to 10 mmol/L in three logarithmically spaced measurements for POPC/POPG. Fig. 4 shows an exemplary NR data set for 10 mmol/L serotonin. The differences between the as-prepared and NT-affected bilayer measurements are small but significant, as shown in the error-weighted residuals. Although counting errors are large at momentum transfers Q_z > 0.15 Å⁻¹, the residual plots also demonstrate that there is information in the data up to ∼0.25 Å⁻¹.

Figure 4.

Neutron reflectivities of a 7:3 POPC/POPG stBLM before and after addition of 10 mmol/L serotonin in (A) D₂O-based and (B) H₂O-based buffer. Bottom: error-weighted residuals show the differences between the NR curves measured with and without NT. All data sets were simultaneously evaluated and yielded a structural model, shown in Fig. 5D, in which adsorbed NT molecules were localized within the charged lipid bilayer. Error bars are from neutron counting statistics. To see this figure in color, go online.

Data modeling of an as-prepared stBLM containing only DOPC established a combined thickness of the hydrophobic chains d_h ≈ 29 Å, an average area per lipid A ≈ 70 Å², and a bilayer completeness of >99%. The corresponding CVO profile is shown in Fig. 5 A, and model parameters are provided in the first column in Table 2. Included in this panel is an estimate of the systematic error of the CVO profile, shown in blue near the abscissa of the plot, which was obtained as discussed above. The blue line defines the median of the systematic error on the NT CVO profiles shown in the remaining panels of Fig. 5, visualizing the sensitivity limit of the NR measurements to NT adsorption. Light blue areas provide 68% confidence limits on this estimate, determined from the MCMC approach that underlies the fitting procedure, and the upper bound of this area provides a visual representation of the worst-case error. On this particular sample, this analysis established a detection limit for serotonin of 6 ± 3 ng/cm², above which we expect to be able to discern adsorbed NT.

Figure 5.

One-dimensional stBLM CVO profiles before and after incubation with serotonin. (A) An example of an as-prepared bilayer is given. The shaded blue area near the bottom of the panel visualizes model deficiencies that determine the sensitivity of the measurement to NT adsorption (see text). (B) A PC bilayer exposed to 10 mmol/L serotonin and (C and D) PC/PG (7:3) bilayers exposed to (C) 1 mmol/L and (D) 10 mmol/L serotonin are shown. The color scheme is as in Fig. 1. Substrate-terminal gold layer: yellow; tether chemistry: green; lipid headgroups: turquoise; lipid chains: blue. Membrane-associated serotonin is shown in red, with a line indicating the profile and shaded areas its 68% confidence limits. The solid red area visualizes minimal distributions of adsorbed NT when accounting for statistical and systematic errors.

Table 2.

Model Parameters from the Structural Characterization by NR of an As-Prepared POPC stBLM and the Same stBLM under Buffer that Contained Serotonin

Parameter	As Prepared	Serotonin Concentration
		1 mmol/L	10 mmol/L
Bilayer completeness (%)	99.5 ± 0.5	99.7 ± 0.3	99.6 ± 0.4
Area per lipid in distal leaflet, A (Å2)	69 ± 3	70 ± 4	67 ± 3
Total hydrocarbon thickness, dh (Å)	29.4 ± 0.9	29.4 ± 0.9	29.7 ± 1.0
Hydrocarbon thickness change, Δdh (Å)	n/a	0.0 ± 1.3	0.3 ± 1.0
Hermite spline center of mass, z (Å)	9 ± 6a	10 ± 7	7 ± 5
Adsorbed NT mass per area (ng/cm2)	9.5 ± 4.0a	9.6 ± 4.0b	15.2 ± 5.6
NT/lipid in the substrate-distal leaflet	n/a	0.23 ± 0.10b	0.35 ± 0.13
Model fit quality, χ2	1.17	1.06	1.10

Median parameter values and 1σ (68%) confidence limits obtained from MCMC-based model refinement are shown. n/a, not applicable.

^aThe significance of a Hermite spline for the as-prepared bilayer lies in the fact that it determines deficiencies in the model that can be mistaken for adsorbed molecules on the same bilayer in contact with solutions containing NT. As such, these data provide an objective estimate of the sensitivity of the NR measurements to adsorbed molecular species (see text).

^bNot significant.

Fig. 5, B–D shows representative CVO profiles of neutral and charged stBLMs (PC/PG = 7:3) in the presence of serotonin in the adjacent buffers. NTs localized on the bilayers are shown in red, with a solid line for their most likely distributions and light red areas indicating 68% confidence. Dark red areas visualize the minimum of likely distributions of NT molecules adsorbed to these bilayers in view of statistical and systematic errors, as described above. In particular, a comparison with Fig. 5 A shows conclusively that the amounts of adsorbed NT exceed the sensitivity of the measurements. Fig. 5, C and D show serotonin content in a charged bilayer upon incubation with serotonin at c = 1 and 10 mmol/L, respectively. These panels represent snapshots in a series of incubations with first 100 μmol/L, then 1 mmol/L, and finally 10 mmol/L, which resulted in serotonin mass adsorption between 1.8 ± 1.1 and 26 ± 5 ng/cm² (Table 3). In response to this progression, we observed an ∼10% decrease in the average area per lipid, whereas the hydrophobic thickness d_h remained constant. Bilayer completeness remained constant, suggesting that high concentrations of the NT do not induce defects in the membrane structure. A comprehensive list of the structural data determined for serotonin adsorption to stBLMs composed of POPC and POPC/POPG (7:3) is provided in Tables 2 and 3, respectively.

Table 3.

Model Parameters from the Structural Characterization by NR of an As-Prepared stBLM Composed of 70 mol% POPC and 30 mol% POPS and the same stBLM under Buffer that Contained Serotonin

Parameter	As Prepared	Serotonin Concentration
		0.1 mmol/L	1 mmol/L	10 mmol/L
Bilayer completeness (%)	99.7 ± 0.3	99.7 ± 0.3	99.8 ± 0.2	99.8 ± 0.1
Area per lipid in distal leaflet, A (Å2)	83 ± 5	77 ± 3	77 ± 5	78 ± 6
Total hydrocarbon thickness, dh (Å)	28.9 ± 0.9	27.8 ± 0.7	29.4 ± 1.1	29.8 ± 1.2
Hydrocarbon thickness change, Δdh (Å)	n/a	−1.1 ± 1.3	0.5 ± 0.14	0.9 ± 1.6
Hermite spline center of mass, z (Å)	16 ± 7a	11 ± 8	9 ± 6	8 ± 3
Adsorbed NT mass per area (ng/cm2)	5.7 ± 2.4a	1.8 ± 1.1b	13.0 ± 3.9	26.0 ± 5.1
NT/lipid in the substrate-distal leaflet	n/a	0.05 ± 0.03b	0.34 ± 0.11	0.70 ± 0.14
Model fit quality, χ2	1.47	1.38	1.39	1.59

Median parameter values and 1σ (68%) confidence limits obtained from MCMC-based model refinement are shown. n/a, not applicable.

^aThe significance of a Hermite spline for the as-prepared bilayer lies in the fact that it determines deficiencies in the model that can be mistaken for adsorbed molecules on the same bilayer in contact with solutions containing NT. As such, these data provide an objective estimate of the sensitivity of the NR measurements to adsorbed molecular species (see text).

^bNot significant.

Intriguingly, serotonin was primarily localized within the bilayer upon adsorption to the membranes, essentially coinciding with the distribution of the substrate-distal phospholipids. Already at bulk serotonin concentrations c < 1 mmol/L, NT molecules were detected close to the interface between lipid hydrocarbon tails and headgroups of the distal monolayer. Upon incubation with subsequently higher concentrations, the amount of the trapped NT increased, primarily in the outer leaflet (Fig. 5, C and D). For the exposure of pure POPC stBLMs to serotonin at high concentration (c = 10 mmol/L), NR showed a similar distribution of the serotonin CVO profiles, but the amount of adsorbed material was slightly lower than in the charged bilayers (Fig. 5 B; Table 2).

We also measured the structures of stBLMs composed of pure POPC and POPC/POPG (7:3) at various concentrations of GABA up to 100 mmol/L in the adjacent buffer. Sensitivity calibrations of the corresponding real-space models established a sensitivity limit of 10 ng/cm² for GABA (Fig. 6 A; Table 4). In all models for bilayers incubated with GABA, we observed minute changes of the structural parameters, but the amounts of adsorbed or incorporated NT were at best borderline significant, as shown for 100 mmol/L of GABA on a zwitterionic and a charged bilayer in Fig. 6, B and C, respectively. A comprehensive list of the structural data determined for GABA adsorption to stBLMs composed of POPC and POPC/POPG (7:3) is provided in Tables 4 and 5, respectively.

Figure 6.

One-dimensional stBLM CVO profiles before and after incubation with GABA. (A) An example of an as-prepared bilayer is given. (B and C) A PC bilayer (B) and a PC/PG (7:3) bilayer (C) exposed to 100 mmol/L GABA are shown. Details are the same as in Fig. 5.

Table 4.

Model Parameters from the Structural Characterization by NR of an As-Prepared POPC stBLM and the Same stBLM under Buffer that Contained GABA

Parameter	As Prepared	GABA Concentration
		10 mmol/L	100 mmol/L
Bilayer completeness (%)	99.7 ± 0.3	99.7 ± 0.3	99.7 ± 0.3
Area per lipid in distal leaflet, A (Å2)	68 ± 3	72 ± 4	70 ± 3
Total hydrocarbon thickness, dh (Å)	29.9 ± 0.9	29.9 ± 1.0	29.4 ± 0.9
Hydrocarbon thickness change, Δdh (Å)	n/a	0.0 ± 1.0	−0.5 ± 1.3
Hermite spline center of mass, z (Å)	10 ± 4a	10 ± 5	10 ± 6
Adsorbed NT mass per area (ng/cm2)	9.7 ± 4.1a	11.5 ± 6.2b	9.9 ± 4.8b
NT/lipid in the substrate-distal leaflet	n/a	0.49 ± 0.26b	0.41 ± 0.20b
Model fit quality, χ2	1.15	1.01	0.98

Median parameter values and 1σ (68%) confidence limits obtained from MCMC-based model refinement are shown. n/a, not applicable.

^aThe significance of a Hermite spline for the as-prepared bilayer lies in the fact that it determines deficiencies in the model that can be mistaken for adsorbed molecules on the same bilayer in contact with solutions containing NT. As such, these data provide an objective estimate of the sensitivity of the NR measurements to adsorbed molecular species (see text).

^bNot significant.

Table 5.

Model Parameters from the Structural Characterization by NR of an As-Prepared stBLM Composed of 70 mol% POPC and 30 mol% POPS and the Same stBLM under Buffer that Contained GABA

Parameter	As Prepared	GABA Concentration
		1 mmol/L	10 mmol/L	100 mmol/L
Bilayer completeness (%)	99.0 ± 0.9	99.0 ± 0.8	98.9 ± 1.0	98.2 ± 1.2
Area per lipid in distal leaflet, A (Å2)	69 ± 3	71 ± 3	74 ± 3	85 ± 4
Total hydrocarbon thickness, dh (Å)	33.4 ± 0.7	33.2 ± 0.7	32.6 ± 0.7	31.1 ± 0.8
Hydrocarbon thickness change, Δdh (Å)	n/a	−0.2 ± 1.0	−0.8 ± 1.0	−2.3 ± 1.1
Hermite spline center of mass, z (Å)	13 ± 5a	14 ± 5	14 ± 6	11 ± 7
Adsorbed NT mass per area (ng/cm2)	8.6 ± 3.9a	7.7 ± 3.5b	7.9 ± 3.5b	6.5 ± 3.3b
NT/lipid in the substrate-distal leaflet	n/a	0.32 ± 0.14b	0.34 ± 0.15b	0.32 ± 0.16b
Model fit quality, χ2	1.34	1.37	1.54	1.30

Median parameter values and 1σ (68%) confidence limits obtained from MCMC-based model refinement are shown. n/a, not applicable.

^aThe significance of a Hermite spline for the as-prepared bilayer lies in the fact that it determines deficiencies in the model that can be mistaken for adsorbed molecules on the same bilayer in contact with solutions containing NT. As such, these data provide an objective estimate of the sensitivity of the NR measurements to adsorbed molecular species (see text).

^bNot significant.

Quantitative evaluation of SPR data

SPR and NR both provide quantitative accounts of adsorbed NT on stBLM bilayers, as shown in Figs. 2, 5, and 6. However, whereas the NR quantification of adsorbed NT can be directly derived from integration of the CVO components, a quantitative interpretation of the SPR results depends on assumptions made about how the adsorbent affects membrane structure. If serotonin intercalation into the membrane increases bilayer volume, the resulting SPR increase, at the same NT load, will differ from the increase expected if specific volumes of lipids change upon NT intercalation such that bilayer volume remains (approximately) constant, as indicated by the NR results shown in Tables 2 and 3. Because molecular structural information is not inherent to the SPR data, we were not able to quantify NT mass adsorption (Fig. 2) without reference to an independent structural method. With the structural information obtained from NR (previous section), we can now revisit this issue and determine an absolute measure of the adsorbed molecular mass as seen by SPR.

The shift in the minimum of the SPR curve upon serotonin adsorption, Δθ_SPR, was calculated using the matrix transfer mechanism (Eqs. 2, 3, and 4) for a number of models composed of five or six distinct layers that represented the substrate and stBLM between the Si wafer and bulk water (inset in Fig. 7). The substrate was composed of either a glass or sapphire base for the λ = 660 and 720 nm instruments, respectively, topped with chromium and gold layers. A simplified stBLM model included a water-rich tether; distinct substrate-proximal and substrate-distal lipid monolayers, each assumed to be homogeneous within; and adjacent buffer. The thickness of individual layers matched the sputtering protocol and NR fits. Informed by the NR results, we considered changes in the optical index and geometric properties only of the substrate-distal monolayer and kept the proximal monolayer fixed. This is indeed reasonable: the proximal lipid layer of an stBLM includes a substantial concentration of tether lipids with ether-linked alkyl chains that pack more densely than the acyl chains of genuine phospholipids in the distal layer (as seen in the differences of the CVO profiles of the “inner” and “outer” lipid chains in Figs. 5 and 6). It is therefore conceivable that the density of intercalated NT molecules differs substantially in the distal layer and proximal monolayers of the membrane, as observed in the NR results. The optical indices of the substrate components were adopted from published data (40, 41, 42, 43), and that of the as-prepared bilayer was assumed to be n = 1.45 (44). Although the refractive indices and dispersion coefficients of the substrate differed for the two instruments, they were equal for the stBLMs within the validity of our assumptions. The optical index of serotonin, n = 1.711, was estimated using the Lorentz-Lorenz relation

$$A=\frac{M_w}{\varrho }\frac{n^2-1}{n^2+2},$$ (5)

where A = 53.5 cm³/mol is the molar refractivity determined from its chemical structure, and M_w and ρ are, respectively, its molar mass and density. Although the validity of Eq. 5 is limited to noninteracting media, this condition is satisfied because NT affinities to the bilayer are low.

Figure 7.

Comparison of adsorbed NT mass (left scale) as determined from measurements with SPR (blue diamonds) and NR (black dots) of a PC/PG (7:3) stBLM exposed to 10 mmol/L serotonin in aqueous buffer. The best-fit Langmuir adsorption isotherm, determined from SPR and corrected for bulk index changes due to dissolved NT (Eq. 1), is shown in red for comparison (right scale). The inset shows the expected SPR signals for serotonin adsoption as a function of its adsorbed mass according to the distinct models discussed in the text: (I) NT adsorption on the bilayer surface; (II) NT intercalation into the substrate-distal bilayer leaflet without expansion of the bilayer; (III) NT intercalation into the substrate-distal bilayer leaflet that results in an increase in bilayer thickness; and (II^∗) NT intercalation into the substrate-distal bilayer leaflet under the assumption that lipid and NT volumes are conserved, resulting in expansion of the bilayer in plane. Distinct colors in the model depiction in the inset indicate distinct homogeneous layers in the modeling of the SPR data using the transfer matrix method (Eqs. 2, 3, and 4). Correspondence between the right- and left-hand axes of the main figure is achieved with model (II); although models (1) and (III) result in similar agreement, they are inconsistent with the NR results. Error bars for SPR data correspond to fluctuations in the traces of a single measurement (see inset in Fig. 2B); for NR data, the error bars show differences in integrating median NT CVO profiles +/− 1σ (c.f. Fig. 5D).

We then calculated the refractive indices of serotonin-loaded substrate-distal monolayers in the membrane for a variety of adsorption models, referred to as models (I)–(III) in the inset of Fig. 7, and were thus able to estimate quantitatively the resulting SPR shifts as a function of NT load. In model (I), we assumed an interfacial layer of serotonin outside of the bilayer membrane, making this optical system a six-layer model. Model (II) had its NT load intercalated into the distal lipid monolayer without changing the geometric properties of the bilayer, as informed by the NR results. Specifically, the thickness of the membrane with and without adsorbed NT was conserved in this model, which implies a reduction of the specific volumes of lipids and/or NT molecules in the intercalated state. If, on the other hand, one assumes conservation of lipid- and NT-specific volumes upon serotonin intercalation of the membrane, model (III) allowed for an increase in the distal monolayer thickness (which is, however, not observed in the NR data). Finally, a variant (II^∗) provides another viable model that satisfies the observation that membrane thickness is conserved, as in model (II). Here, the volume increase of the lipid layer considered in model (III) was compensated by lateral expansion. This implies that lipid molecules within this layer are displaced by intercalated serotonin, leading to a reduction of the in-plane lipid density. The increased spacing between lipid cores is accompanied by increased spacing between lipid headgroups; therefore, their hydration increases. Both effects reduce the optical index of the distal lipid monolayer substantially compared to the other models. The results of these model calculations are shown in the inset of Fig. 7. The SPR signal increases expected for models (I)–(III) are within a few percent of each other and therefore indistinguishable without independent structural information. Of these models, only model (II) is consistent with the NR results. Model (II^∗) is also consistent with the structural data. However, because of the low index of the intercalated layer, the observed SPR signal would imply a serotonin load on the bilayer that is about a factor of 3 larger than that observed with NR at bulk NT concentrations in the NR experiments that match those of the SPR experiments. The correspondence of serotonin load in the bilayer determined by SPR and NR shown in the main panel of Fig. 7 is therefore only achieved in model (II). Moreover, although the density of NR data points is necessarily sparse, the plot also indicates that these data follow the same Langmuir isotherm. This suggests that the joint interpretation of SPR and NR results can identify a likely structural model and account for the serotonin load in the bilayer in a rigorous quantitative way.

Discussion

We measured the nonspecific adsorption of two NTs to bilayers composed of POPC and POPG and observed that serotonin associates with the membranes to a significant extent, whereas GABA adsorption is below our sensitivity limits. SPR measurements on planar bilayers to establish K_D and R_∞ are appropriate because of the high sensitivity of the devices (∼3 × 10⁻⁷ by refractive index) (45). This sensitivity is achieved by using atomistically flat substrates with root mean-square roughnesses σ < 5 Å. As demonstrated in Fig. 2, fitting NT adsorption to Eq. 1 models the data accurately by separating an increase in SPR response due to bilayer adsorption from that due to changes in the bulk index (46). In the case of serotonin adsorption to PC and PC/PG bilayers, the bulk effect is small up to c ≈ 1 mmol/L, where R_∞ ≈ 0.5 ng/cm². There is a crossover of bulk and bilayer effects at c < 10 mmol/L, after which the bulk effect dominates surface adsorption. Serotonin surface accumulation was verified with NR, which is inherently surface sensitive (32). Beyond quantification of surface accumulation, however, NR provides further information on the localization of the NTs within the bilayers.

In line with predictions from MD simulation (19), the binding characteristics of the NTs chosen for this study differ significantly. GABA does not show measurable affinities to bilayers composed of PC or PC/PG, whereas serotonin interacts with such bilayers to a significant extent and embeds deeply. Electrostatic interaction contributes to the recruitment of serotonin to charged bilayers because we observe a drop of K_D near 30 mol% PG, although its surface load, R_∞, remains approximately constant between 0 and 50 mol% of PG (Fig. 3). This confirms a tendency observed in MD simulations (19) in which the addition of 20% PS to POPC in a bilayer also reduced K_D by ∼50%, as estimated from the free-energy profiles reported there, but in absolute terms, our measured K_D values are lower by a factor of ∼10 than those inferred from (19). Distinct from serotonin, the SPR response for GABA was linear over a broad concentration range, between 0 and 1 mol/L, with a slope consistent with the concentration-dependent increase of the bulk index. Indeed, fitting of the GABA adsorption isotherms to the full Eq. 1 did not yield meaningful results because the parameter uncertainties were typically larger than the parameter values. In essence, the total amount of GABA adsorbed to the bilayer was below the instrumental resolution, estimated from the uncertainties in R_∞ to be 0.5 ng/cm².

Our NR results show serotonin adsorbed to the hydrophilic-hydrophobic interface between the lipid headgroup and hydrocarbon tails of the exposed leaflet of a PC bilayer. This, again, agrees with MD simulations, which indicated that serotonin may be hydrogen-bonded to lipid phosphates and in which the long NT axis was oriented perpendicular to the bilayer (14). A careful analysis of our NR-based attempts to quantify GABA association with phospholipid bilayers showed that any changes in the membrane structure that might be attributed to NT accumulation are below, or at most equal to, systematic uncertainty, estimated at 10 ± 4 ng/cm² (Tables 4 and 5). This is consistent with the SPR results, as well as with earlier equilibrium binding experiments that evaluated GABA binding to DMPC (dimyristoylphosphatidylcholine) liposomes by dialysis (13). However, our measurements do not confirm other findings in that study because we did not detect any GABA association with anionic bilayers. On the other hand, our GABA results are consistent with MD simulations that observed that this NT remains excluded from both neutral and charged bilayers (19). The data reported here put a lower limit on GABA membrane affinity that corresponds to K_D ≈ 300 mmol/L. This exceeds by far GABA’s peak concentration, ∼5 mmol/L, in the synaptic cleft (47,48).

The differences in bilayer affinities of serotonin and GABA correlate with their distinct chemical characteristics and can also be rationalized by the distinct localizations of the binding sites on their postsynaptic receptor proteins. Serotonin is hydrophobic, so insertion into the bilayer is more favorable than for the more hydrophilic GABA. This hydrophobicity effect has been observed for other small molecules that partition into the bilayer—for example, in the form of the Meyer-Overton rule, which relates the partitioning of volatile anesthetics in mineral oil to their physiological efficacy (49,50). The hydrophobic NT melatonin is structurally related to serotonin and has been reported to adsorb to bilayers and intercalate at a similar depth (16).

Our results on GABA suggest that it does not partition to bilayers at concentrations assumed to be relevant by Lee et al. (9) in their modeling of electrophysiology data. In line with our findings, the ligand binding site on the GABA_A receptor is exposed to the synaptic cleft (51). Therefore, GABA adsorption to the bilayer, even if a weak effect, would compete with its binding to the receptor. In contrast, serotonin receptor proteins have binding sites buried in the membrane (52). Thus, nonspecific membrane adsorption that enables 2D diffusion once an NT hits the surface would decrease the overall diffusion time to its binding site in comparison with a purely three-dimensional random walk in the aqueous environment (53). In addition to this effect of membrane adsorption on binding kinetics to its cognate receptor, the unspecific binding of serotonin to the bilayer can elicit indirect effects on transmembrane proteins because they might alter the lateral pressure profile across the membrane (11,54). An indication that this may indeed be relevant is our observation that neither membrane thicknesses nor areas per lipid increase proportionally to the NT volumes inserted into the bilayer. Serotonin has a molecular volume that is ∼20% of that of phospholipids (∼240 Å³ vs. ∼1150 Å³), and its adsorbed mass saturates at ∼0.7 NT molecules per membrane lipid (Table 3). Yet, we observe no significant volume changes of the membrane, suggesting that the inserted volume can only be accommodated by lateral pressure increases, which, because of the inhomogeneous distribution of intercalated material across the membrane, would in turn suggest changes in the pressure profile. Although our observations make such changes likely, of the two NT molecules studied here, this applies to serotonin but probably not to GABA. Nevertheless, theoretical (8,11,55) and computational (56) approaches and, often indirectly, experimental evidence (57, 58, 59, 60) suggest that changes in the pressure profile within the bilayer can affect the functionality of transmembrane proteins as well as the association, and thus the functionality, of peripheral membrane proteins (54,61). Our results from the work reported here demonstrate conclusively that serotonin binds to zwitterionic and charged membranes with significant affinities. The K_D values we determined are by far lower than serotonin concentrations in the intercellular space after ejection from the presynaptic dendrite. In extrapolation to a class of NTs with moderate hydrophobicity, characterized as membrane binding in comprehensive MD simulations (19), it is thereby likely that their association with the postsynaptic membrane affects the dynamic response of membrane receptors by mechanisms other than the specific binding to their cognate receptors.

Conclusions

We measured the nonspecific adsorption of two NTs to model membranes and found that serotonin adsorbs at physiological concentrations, whereas GABA does not. SPR experiments determined membrane dissociation constants, K_D, in the millimolar range for serotonin, matching its concentrations upon ejection into the synaptic cleft. In contrast, GABA was not observed to adsorb to neutral or charged membranes up to bulk concentrations approaching 1000 mmol/L. NR experiments confirmed these findings and showed that serotonin intercalates into the distal leaflet of the bilayer, where it is centered near the lipid headgroup phosphates. GABA adsorption could not be quantified in these measurements above the intrinsic detection limits. The two experimental methods were cross-validated by simulating the SPR response of a bilayer with the structural results obtained from NR. Indeed, in an optical model that interpreted the SPR results quantitatively, the results of the two techniques were consistent with each other only if we assumed that serotonin intercalates the bilayer. This evidence suggests that aromatic NTs adsorb to the postsynaptic membrane, as predicted by atomistic MD simulations (19). Their presence in the bilayer is likely to alter the energetic landscape of the membrane-embedded receptor proteins and should be taken into account when modeling electrogenic responses of the postsynaptic membrane (11).

Author Contributions

F.H. and M.L. designed the research. B.P.J. and F.H. prepared samples and performed experiments. B.P.J., F.H., V.S., and M.L. analyzed data. B.P.J., F.H., and M.L. wrote the manuscript, and all authors approved the final version.

Editor: Tommy Nylander.