Selective effect of cell membrane on synaptic neurotransmission

Atomistic molecular dynamics simulations were performed with 13 non-peptidic neurotransmitters (NTs) in three different membrane environments. The results provide compelling evidence that NTs are divided into membrane-binding and membrane-nonbinding molecules. NTs adhere to the postsynaptic membrane surface whenever the ligand-binding sites of their synaptic receptors are buried in the lipid bilayer. In contrast, NTs that have extracellular ligand-binding sites do not have a similar tendency to adhere to the membrane surface. This finding is a seemingly simple yet important addition to the paradigm of neurotransmission, essentially dividing it into membrane-independent and membrane-dependent mechanisms. Moreover, the simulations also indicate that the lipid composition especially in terms of charged lipids can affect the membrane partitioning of NTs. The revised paradigm, highlighting the importance of cell membrane and specific lipids for neurotransmission, should to be of interest to neuroscientists, drug industry and the general public alike.

composition exerts on ligand-receptor entry kinetics in the synapse has been largely ignored in previous studies. For example with ionotropic glutamate receptors (Table S1), valuable pharmacological data can be acquired by concentrating only on their extracellular ligand-binding domains [22][23][24][25][26][27] . However, when studying for instance acetylcholine entry into the M 3 muscarinic acetylcholine receptor, the lipid composition is a factor that should not be overlooked 28 . Although the positively charged NT is not expected to partition to the cell membrane based on its hydrophilic profile, it is yet able to bind effectively into the membrane-buried binding sites of muscarinic receptors as well as into the extracellular binding sites of acetylcholinesterase and nicotinic acetylcholine receptors 28 .
Several experimental and computational studies have indicated that the lipid composition in membranes has a role in NT dynamics. The lipid composition has been shown by molecular dynamics (MD) simulations and calorimetry 29,30 to affect the NT-membrane dynamics with dopamine. The same result has been shown also for glutamate, acetylcholine, γ -aminobutyric acid (GABA) and glycine using MD simulations, calorimetry, and dialysis equilibrium experiments 31,32 . With serotonin, melatonin, and peptidic NT encephalin, attachment to neutral membranes has been also documented by means of neutron scattering 33 , MD simulations [34][35][36] , and nuclear NT was considered bound to the membrane, if it resided within + /− 1 nm from the head group nitrogen atom's average position along the normal to the bilayer surface of either bilayer leaflet. 3 The receptors have extracellular (EX) and/or membrane-buried (MB) binding sites (Table S1). The binding sites are circled (green). The extracellular (red) and intracellular (blue) surfaces are indicated with lines.
Scientific RepoRts | 6:19345 | DOI: 10.1038/srep19345 magnetic resonance measurements 36 . In general, the presence of anionic lipids and the zwitterionic state of NTs enhances membrane attachment. Further, determining the role of membrane environment for NT binding in the synapses is clearly relevant given that lipids occupy roughly half of the cellular membrane surface area, while the other half is taken up by various proteins 37 . In this work we performed atomistic MD simulations for three lipid bilayer models with 13 non-peptidic NTs and the dopamine precursor L-dopa. The pre-equilibrated lipid membrane models (64 lipids/leaflet; see SI) reproduce experimentally determined local bilayer properties such as the order parameter, membrane thickness and average area per lipid (e.g, ref. 38,); accordingly, the models are suitable for studying NT-membrane interactions that take place at the water-membrane interface.
The first "extracellular" membrane model represents the extracellular leaflet of a synaptic membrane, being composed of dioleoylphosphatidylcholine, sphingomyelin, and cholesterol (DOPC/SM/CHOL) 37 . The second "intracellular" model represents a more polar intracellular leaflet and contains dilinoleicphosphatidylcholine, dilinoleicphosphatidylethanolamine, and dilinoleicphosphatidylserine (DLPC/DLPE/DLPS) 37 . Since the exact contents of postsynaptic membranes are not known, the lipid compositions of these two models follow the lipid contents of typical animal cell membranes. Finally, the third "control" model, containing only DLPC lipids, was used as a less polar control membrane. Altogether, 42 NT-membrane systems were simulated for 200-400 ns each; furthermore, the membrane binding energetics were determined for four individual NTs using the umbrella sampling method.
Based on the MD simulation results the current view of neurotransmission, which does not account for the role of lipids 1,2 , was revised and a holistic theory highlighting the selective effect of cell membrane on synaptic neurotransmission was formulated. In short, the simulations provide strong evidence that neurotransmission follows either the membrane-independent or the membrane-dependent mechanism, and the one chosen by nature depends on the position of the receptor's ligand-binding site.

Results
As expected, the NTs did not permeate the lipid bilayers; instead they displayed varying degrees of reversible membrane attachment. Catecholamines such as dopamine and norepinephrine, as well as serotonin, melatonin, and adenosine, all attached strongly to the lipid bilayers in the NT-membrane simulations ( Fig. 1; Table 1). The hydrophobic ring systems were found to be essential for the favorable alignment on the membrane surface. In comparison, small NTs containing charged moieties, such as GABA or glutamate ( Fig. 2; Table 1), did not attach to the membranes. Hydrogen bonding was prominent when NTs attached to the models of the intracellular and extracellular leaflets. With the less polar control membrane, hydrophobic/steric attraction was required for close contact. The specific NT-membrane interactions are discussed thoroughly in the SI (Figures S3-S11 and Tables  S2-S4).
The simulations indicate that NTs can be divided into membrane-binding ( Fig. 1) and membrane-nonbinding groups (Fig. 2). The MD simulations (Table 1), free energy computations ( Fig. 3; Table S5), and log P (the octanol/ water partition coefficient) values of the NTs ( Fig. 4; Table S3) support this view. Although octanol clearly cannot depict all of the physical properties and the diversity of an actual lipid bilayer, the log P values corroborate the simulation results (Table S3). What is more, this categorization of NTs corresponds to the positioning of the ligand-binding sites of their receptors in relation to the postsynaptic membrane (Tables 1 and S1). As a rule of thumb, if a NT attached to a membrane, then its ligand-binding site is membrane-buried. In contrast, whenever the ligand-binding site is extracellular, then the NT did not attach to the membrane (Table 1).
Based on the divergent behavior observed in the simulations, neurotransmission can be split into two different mechanisms (Fig. 5). The classical membrane-independent mechanism (Fig. 5a), shown in neurobiology textbooks, dictates that NTs, whose receptors' ligand-binding sites are extracellular, travel across the synaptic cleft without adhering to the membrane. Thus, these NTs enter their binding site directly from the bulk water. The membrane-dependent mechanism (Fig. 5b) proposed in this work dictates that NTs, whose binding sites are located in the transmembrane part of the receptor, reversibly attach first to the membrane and then diffuse along the membrane surface to the membrane-buried binding site.

Discussion
Although the lipid types residing in synapses are known, the exact lipid compositions of specific synapses or, more importantly, their extracellular and intracellular leaflets are yet to be characterized [39][40][41][42] . Keeping this in mind, let us consider the perspectives emerging from the simulation data for specific lipid types' role in the context of NT binding into proteins embedded at the postsynaptic membrane.
The ligand-binding sites of G protein-coupled receptors are typically membrane-buried (Table 1) and their representative NTs follow the membrane-dependent mechanism (Fig. 5b). Adenosine, epinephrine, norepinephrine, dopamine, serotonin, and melatonin were found to attach reversibly predominantly to the extracellular and intracellular leaflets in the MD simulations (Table 1), stressing the preference for lipid binding. The positively charged histamines attached preferentially to the DLPS-containing intracellular leaflet instead of the extracellular membrane surface (Table 1; Figures S8,S9), which suggests that negatively charged lipids could be more abundant in synapses housing histamine receptors. For serotonin, there exist several 5-HT receptors with membrane-buried binding sites but also 5-HT3 ion channel receptors with extracellular ligand-binding sites (Figs 1,2, S6 and S7; Table 1). As serotonin was found to adhere to all three membrane models, assuring sufficient amount of NT binding into the extracellular binding sites of 5-HT3 receptors might require for example relatively high serotonin levels.
The ligand-gated ion channels such as ionotropic glutamate receptors have extracellular ligand-binding sites. Accordingly, their representative NTs follow the membrane-independent mechanism (Fig. 5a). GABA, glycine, serine, glutamate, and aspartate prefer the bulk water to the extracellular leaflet (Table 1). Interestingly, in the  (Table S1).  (Table 1), but the NT attached to all membranes. Also shown (on the right, bottom) are the extracellular binding sites of the GluA2 glutamate receptor and the 5HT-3 serotonin receptor (Table S1). cases of serine and acetylcholine, the polar intracellular leaflet was preferred to the bulk water ( Figures S8,S9). The fact that acetylcholine attaches to the intracellular surface, and not to the extracellular leaflet, provides an explanation for how muscarinic acetylcholine receptors can have membrane-buried ligand-binding sites while nicotinic acetylcholine receptors have extracellular binding sites (Table 1). Namely, those synapses or membrane regions populated by muscarinic receptors could yet again have more negatively charged lipids attracting the positively charged acetylcholine molecules than those populated by nicotinic receptors. Altogether, the results suggest that specific lipids, and charged types in particular, may modulate the association of NTs to their host membranes, and hence the binding to their receptors.
The above described results suggest that specific lipid types and the lipid composition in general could have an important role in synaptic neurotransmission. While the concentrations of negatively charged glycolipids are usually small in typical animal cells, their levels are known to be higher in neurons 39,40 . Thus, these anionic lipids at the extracellular leaflet of the postsynaptic membrane could facilitate the membrane attachment of positively charged histamine or acetylcholine molecules. While the effect of glycolipids was not addressed in this work, the simulations indicate that charged lipids (here DLPS) do affect the partitioning of NTs (Table 1).
Tightly regulated lipid composition differences between synapses could ensure that acetylcholine can bind into both membrane-buried and extracellular ligand-binding sites (Table 1; Figures S8,S9). In fact, the lipid composition may also be an important factor in neurological diseases and/or their diagnostics. Post mortem studies  have indicated changes in the lipid composition with schizophrenia, Parkinson's disease, and Alzheimer's disease patients [43][44][45][46] . The membrane reversible attachment of a NT could be either enhanced or weakened, if the membrane became less or more charged due to changes in lipid composition. Although the absolute NT levels would not necessarily be affected by these changes, an imbalance in NT-membrane interactions could contribute to disease states such as depression.
Synaptic receptors are not the only membrane-embedded proteins that interact with NTs at the postsynaptic membrane. The timely removal of NTs from the synaptic cleft is vital and, for example, if glutamate transport is obstructed, neurotoxic effects follow 47 . In this regard, the NT-membrane simulations shed new light also on NT transport across the postsynaptic membrane following neurotransmission.
It may seem contradictory, but all NTs eventually bind into the membrane-buried ligand-binding sites of their transporters (Figure S12a,b). The Na + gradient across the membrane powers the Na + /NT co-transport. In addition, the positive ion(s) at the ligand-binding sites form ionic bonds with NTs that have carboxylate groups (shown for leucine in Figure S12b) 48 . With monoamines such as serotonin, which partition readily to the membrane (Table 1), an aspartate side chain acts as an intrinsic ligand which neutralizes the positive charge 48 . In the glutamate transporter the Na + binding takes place prior to the NT entry 49 . Thus, the positively charged ions inside the membrane-buried ligand-binding sites of NT transporters could act as sinks that attract negatively charged NTs such as glutamate and aspartate.
This sink hypothesis is not valid for neutral ligands, but all non-peptidic NTs were somewhat attracted to the polar intracellular membrane in the simulations, serine even preferred it over the bulk water (Figures S10,S11; Table 1). Conceivably, the NT transporters could be surrounded by high concentrations of anionic lipids such as glycolipids. Indeed, lipid composition differences have been suggested to affect, for example, glutamate uptake 50 . Notably, the removal of acetylcholine from the synaptic cleft is different as the positively charged NT binds into the extracellular ligand-binding site of acetylcholinesterase ( Figure S12c) and the reaction products are transported separately into the cytoplasm.
In a broader context, the simulations suggest a strong interplay between lipids, membrane proteins, and NTs, and coevolution at the postsynaptic membrane. The NTs not only have to pass the on/off test of the membrane and match the properties of their receptors' active sites in the synapse, but the membrane effect likely applies to the intracellular NT dynamics as well. However, it is important to recognize that drugs or natural compounds such as toxins do not necessarily follow the same logic as the NTs regarding the membrane partitioning. For example, potent acetylcholinesterase inhibitors can be very lipophilic despite the fact that the enzyme's active site is extracellular (Table S6). This discrepancy highlights the extraordinary importance of fast and coordinated NT kinetics for neurotransmission. Despite this, avoiding ligand-membrane mismatches during drug development should reduce off-target effects and increase efficacy with a multitude of target proteins.
In conclusion, the atomistic MD simulations suggest that non-peptidic NTs are divided into membrane-binding and membrane-nonbinding categories that in turn correspond to the positioning of their receptor's ligand-binding sites in relation to the postsynaptic membrane (Table 1). Hence, neurotransmission follows either the membrane-dependent or the membrane-independent mechanism (Fig. 5). Despite the apparent simplicity of this finding, it represents a fundamentally different outlook on neurotransmission with far-reaching ramifications. The selective effect of the membrane ensures a more coordinated response to NT release than random diffusion could accomplish alone. By controlling the lipid composition, and paying attention to the extracellular leaflet in particular in different synapses in the nervous system, NTs such as acetylcholine could bind efficiently into membrane proteins with both extracellular and membrane-buried ligand-binding sites. Moreover, anionic lipids promote the membrane partitioning of all studied non-peptidic NTs in the simulations (Table 1) and, thus, the presence of negatively charged lipids in the synapse could be a requirement for efficient removal of NTs from the synaptic cleft.

Methods
The atomistic molecular dynamics (MD) simulations were performed with GROMACS 4.5 51 using the OPLS-AA force field 38,52,53 . The MD simulation protocol was presented in a previous study 38 . Free energy curves were calculated using umbrella sampling 54 together with the Weighted Histogram Analysis Method 55 . VMD1.9.1 was used to draw the 3D structures of receptors 56 . Altogether, 42 NT-membrane systems were simulated, each for 200-300 ns (total simulation time being ~12 μ s). The computational methods are explained in detail in SI.