Tuning of the Na,K-ATPase by the beta subunit

The vital gradients of Na+ and K+ across the plasma membrane of animal cells are maintained by the Na,K-ATPase, an αβ enzyme complex, whose α subunit carries out the ion transport and ATP hydrolysis. The specific roles of the β subunit isoforms are less clear, though β2 is essential for motor physiology in mammals. Here, we show that compared to β1 and β3, β2 stabilizes the Na+-occluded E1P state relative to the outward-open E2P state, and that the effect is mediated by its transmembrane domain. Molecular dynamics simulations further demonstrate that the tilt angle of the β transmembrane helix correlates with its functional effect, suggesting that the relative orientation of β modulates ion binding at the α subunit. β2 is primarily expressed in granule neurons and glomeruli in the cerebellum, and we propose that its unique functional characteristics are important to respond appropriately to the cerebellar Na+ and K+ gradients.

We have investigated the molecular and functional role of β 2 and find that it significantly influences the E1P-E2P equilibrium with any of the α subunits studied. To determine the molecular mechanism of β 's functional effect, we constructed chimeras of β 1 and β 2, which pinpointed the transmembrane domain as the main determinant for the observed electrophysiological characteristics. Molecular dynamics (MD) simulations suggest that the transmembrane helices of β 1 and β 2 have different tilt angles, and we propose that the tilt angle of β can influence the relative stability of the Na + occluded E1P state.

Results
The β2 subunit is highly expressed in mouse cerebellum. The α 2β 2 combination has been suggested to be important for K + clearing in hippocampus and fast-twitch muscles 16,18 , but β 2's role in motor coordination 13 implies an important role of its high expression in cerebellum 19 . To determine the relative levels of β 2 expression in the brain, we performed Western blot analysis of isolated regions of mouse brain. We found the highest expression of the β 2 isoform in cerebellum (Fig. 2a), and to further delineate, which cells express β 2, brain slices of adult three months old mice were immunostained. There was intense staining for β 2 in NeuN-positive granule neurons 22 and at glomeruli in the GCL (Fig. 2c), while the Purkinje cell layer had diffuse staining with no signals in the Purkinje cell bodies and only weak, punctuated staining in the molecular layer (ML).
We similarly stained for β 1 and observed it in the NeuN-positive granule neuron plasma membrane and at glomeruli in GCL as well as in the cell bodies and pinceau of Purkinje cells (Fig. 2b). In the ML, fibers and cell bodies showed punctuated β 1 staining. Staining for the α 1 subunit also showed expression in granule neurons and at glomeruli ( Supplementary Fig. S1), suggesting that both α 1β 1 and α 1β 2 pumps may form in the GCL.
Like cerebellum, the dentate gyrus of hippocampus has high neuronal density. To examine if β 2 expression is a general phenomenon in regions with high neuronal density, we examined the dentate gyrus GCL. However, no β 2 staining was observed, and only modest, punctate staining was seen in the ML and Hilus, while both β 1 and α 1 clearly stained the GCL, ML and Hilus in the dentate gyrus ( Supplementary Fig. S1).
Pumps with β2 have high affinity for extracellular Na + . To delineate the functional difference between α 1β 1 and α 1β 2, we expressed the pumps in oocytes from Xenopus laevis and determined their electrophysiological characteristics with two-electrode voltage clamping (TEVC). In the absence of extracellular K + , the NaKA is restricted to binding and releasing Na + from the extracellular medium 9,10 , and the relatively slow charge movement associated with the third Na + can readily be recorded (Fig. 1). The charge translocation in response A simplified Post-Albers scheme with Na + in purple and K + in green is shown in (a). Difference curves in K + -free buffer with and without 10 mM ouabain for (b) α 1β 1,(c) α 1β 2 are shown. The curves were fitted with single exponentials, giving the voltage dependent (d) charge translocation from the off currents and (e) rate constants from the on currents. N = 3-10 with oocytes from at least two Xenopus laevis females. Data are represented as mean ± SD.
Scientific RepoRts | 6:20442 | DOI: 10.1038/srep20442 to a change in the membrane potential (Q/V curve) reflects the voltage-dependent E1P-E2P transition 9 , and right-shifting of the Q/V curve signifies a relative stabilization of the E1P state.
It is well known that mutations and subunit differences in the α subunit can influence the midpoint potential 23,24 . To examine if the β 2 effect is specific for α 1, we therefore expressed the various combinations of α 1, α 2 and α 3 with β 1, β 2 and β 3 and determined their V 0.5 and rate constants. We found that with any given α subunit, pumps with β 2 have right-shifted Q/V curves and higher rate constants compared to pumps with β 1 or β 3 (Supplementary Fig. S2 and Supplementary Table S1).
Voltage dependence of maximal turnover is similar with different β subunits. At hyperpolarized potentials, α 2β 2 has much lower apparent K + affinity than the other combinations of α 1, α 2 or α 3 with β 1 or β 2 15,16 . To estimate the voltage dependence at maximal turnover, we determined the steady-state currents with 15 mM K + in the extracellular buffer. No significant effect of β is evident for the α 1 or α 2 combinations ( Supplementary Fig. S3a-c). With sub-saturating K + -concentrations, the α 2β 2 pumping has previously been reported to have a stronger voltage-dependence than the other combinations 15,16 , suggesting that the main determinant is the voltage-sensitive K + -affinity, which may partly reflect its stronger E1P preference. In the absence of extracellular Na + and K + , NaKA carries an inwardly rectifying proton current 25,26 . The inward proton currents were similar although slightly larger for α 1β 2 ( Supplementary Fig. S3d). Omission of extracellular Na + and K + has also been associated with an uncoupled Na + efflux 27 , but this reaction is orders of magnitude slower than the regular pumping 28 , while the proton current can be even larger than the forward pumping current at hyperpolarized membrane potentials ( Supplementary Fig. S3d).
The effect of the β subunit is determined by its transmembrane helix. Next, we asked if we could map the region of β 2 that markedly influences the E1P-E2P equilibrium. We constructed three chimeras of β 1 and β 2, replacing either the N-terminal cytoplasmic domain, the transmembrane domain or the C-terminal extracellular domain of β 1 with the corresponding β 2 sequence, giving β 1/β 2 NT, β 1/β 2TM and β 1/β 2CT, respectively (Fig. 3a).
The Q/V curve for β 1/β 2NT (V 50 -80 mV) is very close to that of β 1 (V 50 -83 mV), and β 1/β 2CT is also only slightly shifted (V 50 -70 mV). In contrast, β 1/β 2TM (V 50 -51 mV) is closer to β 2 (V 50 -29 mV) than to β 1 ( Fig. 3a and Supplementary Table S1). This suggests that the main determinant of β 's effect on the E1P-E2P equilibrium is the transmembrane domain with a small, additive contribution from the large extracellular C-terminal domain. No significant differences in steady state currents with 15 mM K + were observed between wild type and chimeras ( Supplementary Fig. S3). In the β transmembrane domain, the N-terminal part of the helix (towards the cytoplasmic side) has the highest degree of sequence conservation ( Supplementary Fig. S4), so the more divergent C-terminal part of the helix in β 1 was replaced by the β 2 sequence. Surprisingly, the resulting chimera, β 1/β 2TMC (β 1 with the β 2 transmembrane C-terminus) had a Q/V curve similar to that of β 1 (Fig. 3a). We therefore fine-tuned the mutational approach and introduced the most prominent differences between β 1 and β 2 in the remaining part of the helix into β 1, changing an AGI motif to TAM in the middle (residues 47-49 in β 1 giving β 1TAM) and FK to AF towards the intracellular interface (residues 33-34 in β 1giving β 1AF) ( Supplementary Fig. S4). The charge translocation curve of β 1AF and β 1TAM were only shifted 5 and 7 mV, respectively, towards β 2 (Fig. 3b), changes too modest to explain the effect of the β transmembrane domain on the pump properties. However, combining the two mutations and a motif at the extracellular side, ISE to VSD (residues 61-63 in β 1), in the mutant β 1/3mut, gave a charge translocation curve that was shifted 18 mV towards β 2 (Fig. 3c). This synergistic effect of the three mutated areas suggests an overall structural difference between β 1 and β 2 at the interface with α in the membrane.

Molecular Dynamics simulations suggest different tilt angles of β1 and β2.
To examine the interactions between the α and β subunits further, we analysed crystal structures and compared MD simulations of α 1β 1, α 1β 2 and α 1β 3. The crystal structure of the pig α 1β 1γ 4 in the E1P state was used as the starting model, since it has the most complete structure of β , including the cytoplasmic domain, which is unresolved in other structures. α 1β 2 and α 1β 3 structures were constructed with homology models replacing β 1 ( Fig. 4 and Supplementary Fig. S5).
Comparing β 1 and β 2 shows that the different tilt angles manifest differences in interaction patterns between α and β , especially between the β helix and the α M10 helix (Fig. 4c,d) For example, the β 1-AGI / β 2-TAM motif, Figure 3. Charge translocation curves of chimeras and pocket mutants. Charge translocation was determined for α 1 coexpressed with (a) β chimeras, where the N-terminal, the C-terminal or the transmembrane region of β 1 was replaced with that of β 2, with (b) β mutants where smaller stretches in the transmembrane region of β 1 were replaced with the corresponding β 2 sequences, FK with AF N-terminally, AGI with TAM in the middle or the C-terminal 16 residues, or (c) a combination of the FK toAF , AGI to TAM and VSD to ISE (at the C-terminus, cf.  which had a small but measurable effect in electrophysiology, shows differences in interactions with the α M10: there are hydrophobic interactions between the isoleucine in β 1 and residues in the α M10 approximately 10 Å from ion binding site III (Fig. 4c), but not between the longer methionine in β 2 and the α residues (Fig. 4d).
Interestingly, β 1 and β 2 appear to interact differently with the α C-terminus, which is known to be an important regulator of the Na + binding site III 1,9,29,30 . A cation-π interaction between β 1 K34 and α1 W1009 is absent with β 2, which has an F at this position (Fig. 4C,D). A K in β 1 and an F in β 2 are conserved between species (human, rat, sheep, chicken and dolphin). With β 2, the α1 C-terminus is consequently slightly displaced in the MD simulations, and there is higher fluctuation in the hinge region connecting it to M10 (Fig. 5), suggesting that sequence differences in the β subunits can affect the structure and flexibility of the α C-terminus and hence the E1P-E2P equilibrium.
The electrophysiological data suggest that single regions in the β transmembrane region have little or no effect, but combining mutations gives a pronounced shift (Fig. 3). If the subunit differences depend on the membrane tilt angle as the MD simulations imply, we would expect the transmembrane helices to interact differently with the lipid head groups. From the MD simulations, we calculated radial distribution functions (RDF) between the lipid headgroups and the expected anchoring points of the transmembrane helix ( Supplementary Fig. S6). With β 1, the helix is anchored on the intracellular side at the FK motif and at the extracellular side at the ISE motif ( Supplementary Fig. S4). With β 2, the lipid interactions are weaker at the intracellular side and stronger at the extracellular side, which likely contributes to the change in the tilt angle and thus to altered functional properties, strengthening the electrophysiological finding that a change in both membrane anchor points together with the AGI to TAM mutation in the middle changes β 1 towards β 2.

Discussion
The unique properties of α 2β 2 have led to the suggestion that NaKA with this subunit composition is optimized for clearing of K + from the extracellular fluid in hippocampal glia cells 16 and in fast-twitch glycotic muscles 17,18 . The severe motor phenotype of mice lacking β 2 13 is likely to be due to impairment in the cerebellum, a region central to neuromuscular processing, where high levels of β 2 are detected in rat 20 and mouse (Fig. 2). The cerebellum has the highest ratio of neurons to glia cells in the brain, an estimated 4.3 31 , and about 40-50% of the cerebellar ATP is used to fuel NaKA in the granule cells 21 , where we observe intense β 2 staining (Fig. 2c), suggesting that K + clearance in cerebellar neurons may rely on β 2 containing NaKA. Unlike a previous study 20 , we did not find β 2 expression in Purkinje cells.
Specific subunit localization may be due to developmental or targeting requirements, but our findings indicate that the functional characteristics of β 2 are also likely to be important. We therefore focussed on determining the functional differences between β 1 and β 2. In the P-type ATPase family, the NaKA, H,K-ATPase (HKA) and the lipid flippases are unique in their strict dependence on a β subunit for trafficking of the holoenzyme to the plasmamembrane and for modulation of the catalytic properties. Combining the NaKA α with different β subunit isoforms or the β subunit from the HKA has previously shown that β affects the apparent K + and Na + affinities and changes the rate of formation of the phosphoenzyme 15,16,[32][33][34][35][36][37][38][39][40][41][42][43] , while surface expression, turnover number and ouabain binding were similar for different β isoform combinations 15 .
NaKAs with β 2 have shown the most significant differences with an unusually high K 1/2 for K + activation of α 2β 2 at hyperpolarized potentials 16 . Our data similarly imply that NaKA β 2 combinations form the most divergent pumps. We show that, compared to β 1 and β 3, β 2 markedly shifts the E1P-E2P equilibrium towards the E1P state (right-shift of the Q/V curve) ( Fig. 1 and Supplementary Fig. S2) and increases the rate constants at hyperpolarized potentials (Supplementary Fig. S2).
To understand the mechanism of the functional effects of the β isoforms, we constructed chimeras of β 1 and β 2. Both the N-terminal cytoplasmic region and the C-terminal extracellular region of β 1 could be exchanged for the corresponding β 2 region without major effect on the Q/V curve's midpoint potential. In contrast, replacing the transmembrane domain in β 1 with that of β 2 markedly shifted the Q/V curve towards that of β 2 (Fig. 1d).
The β transmembrane domain was previously suggested to be involved in retention of the holoenzyme in the endoplasmatic reticulum, since chimeras of β 1 with the transmembrane region of the HKA β are retained 34 . Furthermore, glutathionylation of a cysteine in the middle of β 1's transmembrane region was shown to influence the catalytic properties of the pump 44 , but the cysteine is absent from both β 2 and β 3 ( Supplementary Fig. S4), and β 3 behaves largely like β 1 (Supplementary Fig. S2), so a difference in glutathionylation is unlikely to explain why β 2 differs from the other β s. near the membrane interface. Important residues forming ion binding sites I, II and III are shown as sticks, and bound sodium ions are shown as pink spheres. Water and the γ subunit are omitted for clarity, but included in the model. (b) Comparison of the helix tilt between the transmembrane helix of β 1 (blue) and β 2 (light green). The tilt is defined as the angle between the helix axis and the z-axis, which is perpendicular to the membrane surface. Presented values are the averages from the last 40 ns, with error estimations obtained with block averaging. (c,d) Interaction patterns between the β helix (blue β 1 C) and light green β 2 D)) and the M7 helix (yellow) and the C-terminus (orange) of the α subunit in α 1β 1 C) and α 1β 2 D). Interactions between selected residues (shown in licorice) are shown as purple springs, with minimum distances recorded in simulations (last 40 ns) indicated in italics. Hydrophobic carbon atoms are shown in cyan, oxygen atoms bearing partial negative charge are shown in red and nitrogen atoms bearing partial positive charge are shown in deep blue.
Scientific RepoRts | 6:20442 | DOI: 10.1038/srep20442 The transmembrane domain of the β subunit is not highly conserved ( Supplementary Fig. S4), and one or more of the 17 residues that differ between the β 1 and β 2 helices must account for the functional differences, but mutational studies did not identify any single motif as a key determinant (Fig. 3a,b). However, a significant shift of the charge translocation curve towards α 1β 2 was seen when combining mutations in three areas, at the N-and C-terminal membrane anchor points and in a central motif, suggesting that the tilt angle of the transmembrane helix may be important for the functional effects (Fig. 3c).
To examine the structural foundation for the differences, we performed MD simulations of α 1β 1 and of α 1β 2 and α 1β 3 homology models. The most striking difference between the structures was the β tilt angle relative to the membrane plane with β 1 and β 3 being similar, while β 2 was 5° more tilted than β 1, which changes e.g. the interactions between the α M10 and the highly conserved β F38 and between the α C-terminus and the AF/FK motif of β 1 and β 2 (Fig. 4). Thus, we find that the β isoforms can influence structural elements in the α subunit that are known to be important for the kinetic properties of Na + binding 29,45,46 .
From the related calcium pump SERCA, several structures of different functional states have been determined, which indicate that the M7, M8, M9 and M10 move in concert as a C-terminal domain [47][48][49] . In NaKA, the corresponding C-terminal domain of α is important for ion binding and release at the unique site III 5,29 . There are only structures of a few functional NaKA states, but an overlay of the structures of the potassium-bound E2Pi-like state 2 and the sodium-bound E1P-like state 4,5 shows that the M10-β 1 interface, including the β 1 tilt angle, is conserved. The β helix thus appears to move together with the C-terminal helix bundle of α during the catalytic cycle, but the differences described here are likely to affect the overall dynamics of the bundle and thereby the properties of Na + binding at site III.
In summary, the electrophysiological studies and MD simulations presented here indicate that a main determinant of the functional differences between β 1 and β 2 is the tilt angle of the transmembrane helix, which alters the interaction between β and the α C-terminus and thereby the relative stability of the E1P and E2P states in the catalytic cycle. Because of its low apparent K + affinity, α 2β 2 was previously suggested to be specifically geared for high activity in astrocytes when extracellular [K + ] is elevated, making it optimized for K + clearing after neuronal bursts 16 . We suggest that another primary physiological role of the NaKA β 2 is in the cerebellum, where there are only relatively few astrocytes. Future studies will be required to determine if β 2 containing pumps in the cerebellar granule cells and glomeruli can compensate for the high neuronal density.

Methods
Chemicals. Chemicals used were obtained in the highest grade of purity from Sigma Aldrich (St. Louis, MO, USA) and VWR chemicals (Radnor, PA, USA).

Molecular Biology.
Plasmids encoding human α 1, α 2, α 3 and β 1 subunits of NaKA were purchased from Origene (Origene, Rockville, MD, USA) and subcloned into the pXOON vector 50 using EcoRI and NotI. Plasmids encoding β 2 and β 3 subunits of the NaKA were purchased from Source Bioscience Lifesciences (SourceBioscience, Notingham, UK). β 3 was subcloned into the pXOON vector using EcoRI and XhoI. β 2 was amplified from the supplied vector using primers listed in Supplementary Table S2 introducing HindIII and BamHI restriction sites. The PCR product and pXOON were treated with BamHI and HindIII and the β 2 insert was henceforth subcloned into pXOON. Chimeras were constructed by amplifying the N-terminus of β 2, the transmembrane region of β 2 or the N-terminus plus transmembrane region of β 1 with primers listed in Supplementary Table S2 respectively. The amplified products were purified by agarose gel-electrophoresis (1% agarose (w/v)) in combination with the Qiaquick Gel Extraction Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. The purified product was used as a primer on pXOON β 1 (β 1/β 2NT, β 1/β 2TM, β 1/β 2TMC) or pXOON β 2 (β 1/β 2CT). Single mutations in wild type β 1 were constructed with primers stated in supplementary table S2. All mutations were finally constructed using the quick change lightning site directed mutagenesis kit according to the manufacturer's instructions (Agilent Technologies). Constructs were sequenced to verify successful mutagenesis. α isoforms contained mutations Q116R and N127D to reduce ouabain resistance 51 .
In preparation of mRNA transcription, the desired plasmids were linearized using NheI (α 1, α 2, α 3, β 1, β 2 and β 1/β 2 mutants) or XhoI (β 3) for 20-30 minutes. Linearized plasmids were purified using standard phenol/chloroform extraction. The restriction buffer containing the linearized plasmid and restriction enzyme was diluted to 200 μ l in ddH2O and 200 μ l Tris saturated phenol, pH 7.3 was added, vigorously mixed and centrifuged at approximately 17.000 x g for 2 minutes. The top fraction was mixed with chloroform and centrifuged again. After centrifugation, the top fraction was mixed with 15% (v/v) 3 M sodium acetate, pH 5.4, 2.5 volumes of ethanol and kept at -20 °C for a minimum of 30 minutes. Following that was another centrifugation at 17.000 x g and 4 °C for 30 minutes. The DNA pellet was washed with 70% (v/v) ethanol and resuspended in approximately 5-10 μ l ddH2O.
mRNA was transcribed using the mMessage mMachine T7 Ultra Kit (Ambion, Life Technologies, Carlsbad, CA, USA) according to manufacturer's instructions.
Oocytes from Xenopus laevis were isolated and defolliculated. 50 nl of a mixture of α (10 ng) and β (5 ng) mRNA was injected into Stage V and VI oocytes. Oocytes were incubated at 11 °C for 3-8 days prior to electrophysiological analysis. Low ouabain concentrations in the measuring buffers inhibit the endogenous Xenopus laevis NaKA, and NMDG, which is too large a cation to be transported by the NaKA, is used to replace Na + . Measurements were performed in 200 ms voltage jumps in steps of 10 mV and a holding potential of -50 mV. Measurements in 10 mM ouabain buffer were subtracted from measurements without a high ouabain concentration yielding currents solely generated by the NaKA.

Electrophysiology. Electrophysiological measurements
Charge translocation was determined by fitting single exponentials to the ouabain-sensitive pre-steady-state currents in K + occluded buffer (Fig. 2): f(t) = A * exp(-t/τ ) + C where t is time after the voltage jump, A the amplitude (current at t = 0), τ the relaxation rate and C a constant (current at t = ∞).
Data was recorded and analysed with pClamp 10.4 (Molecular Devices) and Graph Pad Prism 6 (Graph Pad Software) 46 .

Molecular Dynamics simulations.
All-atom MD simulations were performed in a manner similar to one described in a recent publication 52 . The recently determined, crystal structure of the [Na 3 ] E1-AlF4-ADP form of pig kidney NaKA (PDB ID: 3WGU 4 ) was used as a starting point of the simulations. This crystal structure represents the pig α 1β 1γ ternary complex and was used as the model of the α 1β 1 combination of the E1P phosphoenzyme. Subsequently, this structure was embedded in a fully hydrated 1-palmitoyl,2oleoyl-sn-glycero-3-posphocholine (POPC), including bound sodium ions and ADP, and an AlF 4 phosphoryl transfer mimic was replaced by a fully phosphorylated Asp369 residue (the phosphorylation site 47 ). To study the α 1β 2 and α 1β 3 combinations, the β 1 structure was replaced by β 2 and β 3 homology models, respectively. For the β 1/3mut and β 1/β 2TM combinations, the point mutations were introduced into the β 1 structure using Pymol (The PyMOL Molecular Graphics System, Version 1.7.4 Schrödinger, LLC).
System construction -The Na + coordinating residues E327 and E779 at binding sites I and II , as well as E954 at the unoccupied site IIIa were kept protonated 52 . The remaining glutamate and aspartate residues were kept in the charged state. The α β γ complex was embedded in the equilibrated POPC membrane (~460 lipid molecules), using the g_membed tool, and surrounded with ~60.000 water molecules. Electroneutrality was achieved by additional sodium ions, placed randomly in the aqueous solution. The resulting model was treated as the input for the MD simulations of the α 1β 1 combination. The atomistic homology models of the β 2 and β 3 subunits were constructed using the atomistic coordinates of the β 1 subunit crystal structure (PDB ID: 3WGU 4 ) and the human sequences of β 2 and β 3, respectively, using MODELLER 53 . Subsequently, the β 1 subunit in the α 1β 1 combination was replaced by either β 2, β 3, β 1/3mut or β 1/β 2TM, resulting in models that were treated as inputs for the MD simulations of the α 1β 2, α 1β 3, α 1β 1/3mut or α 1β 1/β 2TM combination, respectively. Simulation details -GROMACS version 5.0.1 was used to propagate the MD equations of motions using the leap-from algorithm. The CHARMM36 force field in GROMACS format was employed [54][55][56] for proteins, lipids, ADP and ions. Parameters for the phosphorylated aspartate were adapted from Damjanović et al. 57 . Water was modelled with the CHARMM TIP3P model, with Lennard-Jones interactions between water hydrogens, which ensures the correct phase behaviour of the POPC membrane 58 . A timestep of 2 fs was used. Periodic boundary conditions were applied in all three directions. The van der Waals interactions were switched off from 0.8 to 1.2 nm, using the force-switch option in GROMACS. The Particle Mesh Ewald (PME) 59 with a 1.2 nm cut-off was employed for electrostatic interactions. The simulated systems were maintained at the temperature of 310 K and the pressure of 1 bar, realizing the NPT statistical ensemble. Temperature coupling was realized using the Berendsen thermostat 60 for the equilibration (10 ns) and with the Nose-Hoover thermostat 61,62 for the production runs (100 ns each), separately for the solute (proteins, lipids, ADP) and the solvent (water and ions). Pressure coupling was realized with the Berendsen barostat 60 for the equilibration, and the Parinello-Rahman barostat 63 for the production runs. Prior to MD simulations, all systems were energy minimized with the steepest decent algorithm (10000 steps). Trajectories were sampled every 25 ps. The analysis was carried out using GROMACS suite programs: gmx bundle, gmx rms, gmx distnace, gmx rdf. Visualizations were made using VMD 64 . Western Blot. 3 months old C57BL/6J mice (Janvier) were killed by cervical dislocation, brain regions were dissected and lysed in 10 mM Tris, 150 mM NaCl, 2 mM EDTA with 1% IGEPAL and protease inhibitor (Complete, Roche, Basel, Switzerland). Lysates were separated by SDS-PAGE and electro-blotted onto nitrocellulose membranes (Pharmacia-Amersham, Amersham, UK). Membranes were blocked in PBS with 0.5% Tween-20 and 5% skimmed milk powder and subsequently incubated with primary antibodies (β 2 1:1000 (HPA010698, Atlas Antibodies, Stockholm, Sweden) and GAPDH 1:1000 (ab9485, Abcam, Cambridge, UK)) overnight at 4 °C. The following day, membranes were incubated with horseradish peroxidase-conjugated secondary antibody (pig anti-rabbit 1:2000 (Dako, Glostrup, Denmark)) for 1 hour at room temperature. Visualization was done using a LAS 3000 imager (Fujifilm, Tokyo, Japan) with Amersham ECL Western Blotting Detection Kit (GE Healthcare, Buckinghamshire, UK) as detection reagent.  22 were applied in 1% donkey serum PBS/Triton X-100 0.25% for 1 hour at RT and then overnight at 4 °C. The following day secondary labelling was done with Alexa Fluor fluorescent-conjugated secondary antibodies (Alexa Fluor 488 donkey anti rabbit (A21206, Life Technologies, Carlsbad, CA, USA), Alexa Fluor 568 donkey anti mouse (A10037, Life Technologies, Carlsbad, CA, USA), Alexa Fluor 568 Streptavidin (S11226, Life Technologies, Carlsbad, CA, USA)) (1:350) in 1% donkey serum PBS/Triton X-100 0.25% for 1 hour at RT. Hoechst (1:10000) (Life technologies, Carlsbad, CA, USA) in PBS was used to counterstain the nuclei. Sections were mounted using fluorescence mounting medium (Dako, Glostrup, Denmark) and analysed on a LSM510 laser-scanning confocal microscope using a 40x and 63x C-Apochromat water immersion objective NA 1.2 (Carl Zeiss, Göttingen, Germany). Zen 2011 software (Carl Zeiss, Göttingen, Germany) was used for image capturing and subsequent image analysis.