Kainate receptor pore‐forming and auxiliary subunits regulate channel block by a novel mechanism

Key points Kainate receptor heteromerization and auxiliary subunits, Neto1 and Neto2, attenuate polyamine ion‐channel block by facilitating blocker permeation. Relief of polyamine block in GluK2/GluK5 heteromers results from a key proline residue that produces architectural changes in the channel pore α‐helical region. Auxiliary subunits exert an additive effect to heteromerization, and thus relief of polyamine block is due to a different mechanism. Our findings have broad implications for work on polyamine block of other cation‐selective ion channels. Abstract Channel block and permeation by cytoplasmic polyamines is a common feature of many cation‐selective ion channels. Although the channel block mechanism has been studied extensively, polyamine permeation has been considered less significant as it occurs at extreme positive membrane potentials. Here, we show that kainate receptor (KAR) heteromerization and association with auxiliary proteins, Neto1 and Neto2, attenuate polyamine block by enhancing blocker permeation. Consequently, polyamine permeation and unblock occur at more negative and physiologically relevant membrane potentials. In GluK2/GluK5 heteromers, enhanced permeation is due to a single proline residue in GluK5 that alters the dynamics of the α‐helical region of the selectivity filter. The effect of auxiliary proteins is additive, and therefore the structural basis of polyamine permeation and unblock is through a different mechanism. As native receptors are thought to assemble as heteromers in complex with auxiliary proteins, our data identify an unappreciated impact of polyamine permeation in shaping the signalling properties of neuronal KARs and point to a structural mechanism that may be shared amongst other cation‐selective ion channels.


Introduction
Polyamines are expressed ubiquitously in bacterial, plant and animal cells where they fulfil a variety of roles essential for life (Tabor & Tabor, 1964, 1985. Mammals possess two polyamines, the tetra-and tri-amines, spermine (Spm) and spermidine (Spd), respectively, that are derived from the diamine precursor, putrescine (Put), through tightly regulated biosynthetic pathways. Disruption in the synthesis or catabolism of polyamines leads to a variety of disease states from cancer to neurodevelopmental disorders (Nowotarski et al. 2013;Pegg, 2014), underlining their physiological importance. Given their cationic nature, polyamines interact with negatively charged domains of biomolecules (Tabor & Tabor, 1984) including the electrostatic pore regions of voltage-and ligand-gated ion channels where they bind and consequently block ion flow with micromolar affinity (Lopatin et al. 1994;Bowie & Mayer, 1995;Gomez & Hellstrand, 1995;Haghighi & Cooper, 1998;Lu & Ding, 1999;Kerschbaum et al. 2003;Fu et al. 2012). In this capacity, cytoplasmic polyamines are recognized as important determinants of neuronal signalling by regulating action potential firing rates (Fleidervish et al. 2008) as well as the strength of neurotransmission (Rozov & Burnashev, 1999;Aizenman et al. 2002).
Polyamines are often referred to as permeant channel blockers due to their ability to both block and traverse the ion-permeation pathway of cation-selective ion channels (Bowie et al. 1999). For almost all ion channels, cytoplasmic polyamine block is observed at negative and physiologically relevant membrane potentials, making it an ideal regulator of cellular excitability (Nichols & Lopatin, 1997;Bowie et al. 1999;Lu, 2004;Baronas & Kurata, 2014). In contrast, polyamine permeation occurs at extreme (>+50 mV) positive membrane potentials (Bahring et al. 1997) and therefore has not been considered particularly significant. Given this arrangement, however, the fraction of channels blocked by polyamines at any given membrane potential is the sum of these two opposing mechanisms. Consequently, the overall level of polyamine block could, in principle, be shaped by any mechanism that enhances the relative contribution of polyamine permeation.
At glutamatergic synapses of the developing and mature CNS, polyamines have been well characterized to act as permeant channel blockers of both AMPAand kainate-type (KARs) ionotropic glutamate receptors (iGluRs) (Bowie & Mayer, 1995;Kamboj et al. 1995;Koh et al. 1995). In each case, the degree of polyamine block is voltage-dependent and dynamically regulated, with two of the most prevalent mechanisms occurring through either changes in receptor subunit composition or by receptor association with auxiliary proteins (Cull-Candy et al. 2006;Perrais et al. 2010;Jackson & Nicoll, 2011). At KARs, polyamine block is attenuated by either the formation of heteromers that contain the GluK4 or GluK5 subunit (Barberis et al. 2008) or by co-assembly with auxiliary proteins, Neto1 and Neto2 (Fisher & Mott, 2012). Interestingly, removal of the positively charged Arg-Lys-Lys motif of the Neto1 or Neto2 C-tail restores polyamine block, suggesting that relief of block may involve a charge screening mechanism that affects the pore (Fisher & Mott, 2012). Whether a similar electrostatic mechanism explains the attenuation of block at GluK2/GluK5 heteromers remains to be established.
Here, we have tested the hypothesis that KAR heteromers and auxiliary proteins attenuate channel block by a common mechanism. We show that in both cases relief of polyamine block results from an enhanced rate of polyamine permeation. Relief of block in heteromers is due to a proline residue from GluK5 that is predicted to alter the pore architecture, thereby facilitating polyamine permeation. The effect of auxiliary proteins is additive to that of the pore-forming subunits and we therefore conclude that they enhance polyamine permeation through a different mechanism. As native receptors primarily assemble as heteromers in complex with auxiliary proteins, our data identify an unappreciated impact of polyamine permeation in shaping the signalling properties of KARs in neuronal circuits.

Plasmids and molecular biology
All experiments were performed using cDNA of rat GluK2 and GluK5 kainate-type iGluRs, mouse Neto1-HA and rat Neto2. For electrophysiology, GluK2a(V,C,Q) and GluK5 cDNA each contained a downstream internal ribosome entry site (IRES) sequence encoding mCherry and eGFP, respectively, which were used to identify transfected cells. cDNA of GluK2(G615P) and GluK5(P599G) were made using site-directed mutagenesis with all constructs verified by restriction digest analysis and sequencing.

Cell culture and transfection
HEK293T/17 cells (ATCC, Manassas, VA, USA) were maintained in minimal essential medium (MEM) containing glutaMAX C supplemented with 10% fetal bovine serum (Invitrogen, Carlsbad, CA, USA). Cells were plated at low density (1.6-2.0×10 4 cells ml −1 ) on poly-D-lysine-coated 35 mm plastic dishes and were transiently transfected 24 h later using the calcium phosphate technique. A GluK2:GluK5 or GluK2:Neto1/2 cDNA molar ratio of 1:10 was used for all co-transfections. outside-out patches excised from transfected cells using a piezoelectric stack (Physik Instrumente, Auburn, MA, USA). Solution exchange (10-90% rise time of 250-350 μs) was determined in a separate experiment by measuring the liquid junction current. All recordings were performed using an Axopatch 200B (Molecular Devices, Sunnyvale, CA, USA) using thick-walled borosilicate glass pipettes (3-6 M ) coated with dental wax to reduce electrical noise. Current records were filtered at 5 kHz, digitized at 25 kHz and series resistance (3-12 M ) was compensated for by 95%. Recordings were performed at a range of holding potentials from −100 to +100 mV to study polyamine channel block. Data acquisition was performed using pClamp9 or pClamp10 software (Molecular Devices) and tabulated using Excel. All experiments were performed at room temperature.
All chemicals were purchased from Sigma-Aldrich (St Louis, MO, USA) unless otherwise indicated. External solutions contained (in mM) 150 NaCl, 5 HEPES, 0.1 MgCl 2 and 0.1 CaCl 2 , pH 7.3-7.4. The internal solution contained (in mM) 115 NaCl, 10 NaF, 5 HEPES, 5 Na 4 BAPTA (Life Technologies, Invitrogen, Burlington, ON, Canada), 1 MgCl 2 , 0.5 CaCl 2 and 10 Na 2 ATP, pH 7.3-7.4. The osmotic pressure of all solutions was adjusted to 295-300 mOsm with sucrose. In experiments where polyamine block was examined, Na 2 ATP in the internal solution was replaced with 60 μM Spm and the osmotic pressure adjusted to 295-300 mOsm with sucrose. For Ca 2+ permeability experiments, the external solution contained (in mM) 105 NaCl, 30 CaCl 2 , 5 HEPES and 0.1 MgCl 2 , pH 7.3-7.4 (295 mOsm). For polyamine permeability experiments, the NaCl in the external solution was replaced by 90 mM Spm, pH was adjusted to 7.3-7.4 using the Spm free base and the osmotic pressure was adjusted with sucrose. Concentrated (10× or 100×) agonist stock solutions were prepared by dissolving agonist [L-glutamate or AMPA (Tocris, Ellisville, MO, USA)] in the appropriate external solution, adjusting the pH to 7.3-7.4, and stored frozen at −20°C. Stocks were thawed on the day of the experiment and used to prepare agonist-containing external solutions. For heteromeric GluK2/GluK5 receptors, only patches where the contribution of the slowest exponential component (%A3) was above 25% were considered for analysis, unless otherwise stated.

Data analysis and fitting
Agonist-evoked membrane conductance (G) was calculated using eqn (1): where I is the current at V holding potential and V rev is the reversal potential. Conductance-voltage (G-V) relationships were fit using Origin 7 (OriginLab, Northampton, MA, USA) with the following equation (from Bowie et al. 1998): where G max is the maximal conductance, [Spm] is the internal Spm concentration and K d is the dissociation constant. Binding (k on ), unbinding (k off ) and permeation (k perm ) rates were defined as previously (Bowie et al. 1998): To estimate k off and k perm , K d was first calculated as: which was redefined to eqn (6): Binding rates were assumed to be constant for GluK2/GluK5, GluK2+Neto1 and GluK2+Neto2 (see Results for more details), and the values for a and b were taken from Bowie et al. (1998).
The voltage-dependent conductance of KARs under basal conditions is appreciably different between receptor types within the range ±100 mV. This emphasizes the impact of auxiliary proteins in changing the basic gating properties of KARs and the importance of accounting for these significant differences in gating properties. The intrinsic G/V relationships were fit with eqn (7) using Origin 7: where G 0 is the minimal conductance and V is the membrane potential. Values are summarized in Table 1. The Ca 2+ permeability relative to Na + (P Ca /P Na ) was determined using the constant field equation from Kamboj et al. (1995): where P Ca and P Na are the permeability coefficients for Ca 2+ and Na + , respectively, V rev is the reversal potential shift, F is Faraday's constant, R is the gas constant and T is the temperature in Kelvin. Relative Spm permeability (P Spm /P Na ) was calculated from the reversal potential (V rev ) measurements in Spm-rich external solutions using the following equation from Bahring et al. (1997): where I x is the current carried by ion X at a membrane potential V m , P x is the corresponding permeability (P Na = 1), [X] i and [X] o are the internal and external Spm concentrations, z is the valence, and F, R and T are as above. Experimentally determined reversal potentials were used for V m . We obtained P Spm /P Na using the definition that I Na + I Spm = 0 at the reversal potential. The reversal potentials were corrected for liquid junction potentials that were determined experimentally. All data were illustrated using Origin 7 and Adobe Illustrator.

Statistics
Statistical analysis was performed using the statistical software SPSS Statistics. Data were assumed to be distributed normally. The numbers provided for 'n' refer to the number of individual patches.

Molecular dynamics simulations
Molecular dynamics (MD) simulations were performed to study whether inserting proline residues into the α-helical region of the P-loop had an effect on the pore architecture. No atomic-resolution open-channel structure of an iGluR has been determined, so an open-channel structure of the non-selective NaK cation channel from Bacillus cereus determined to a resolution of 1.6Å was used as a model system (PDB #3E86) (Alam & Jiang, 2009a). Figure 7 below shows a structural alignment of the GluA2 (PDB #3KG2) and the NaK (PDB #3E86) channel pores. Given that they have very similar topologies, the NaK pore structure was used as this structure has the selectivity filter resolved at atomic resolution and is described as being in the open state as opposed to the GluA2 structure which is considered closed and where the selectivity filter region is unresolved.
To generate a membrane slab for further simulation, the tetrameric physiological unit of the NaK channel (residues 23-113, ignoring ions and other small molecules) was converted to Coarse Grain (CG) representation (Martini v2.1) (Monticelli et al. 2008) to allow for self-assembly and equilibration of a bilayer consisting of 200 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) molecules around the protein structure. This system was simulated for 100 ns at 323 K in Gromacs v4.6.3 (Hess et al. 2008) to allow formation of a POPC bilayer (Stansfeld et al. 2015).
Following this, the system was converted back to atomistic representation using a CG2AT-align method (Stansfeld & Sansom, 2011) and this membrane slab was then used for further system setup with the different protein mutants aligned to the position of the protein. The (4S)-2-methyl-2,4-pentanediol molecule, observed in the crystal structure, was deleted. The crystal structure has four ions bound in the filter region: one calcium ion in the external site, one sodium ion in the vestibule site, one caesium ion in site 3 and another sodium ion in site 4. To keep the ion composition simple, the calcium ion and the caesium ion were changed to potassium and sodium, respectively. Both monovalent and divalent cations can bind at the external site, but sodium does not appear to be a good binder at this site (Alam & Jiang, 2009b), and thus the calcium ion was exchanged for potassium. The vestibular site and sites 3 and 4 all appear to prefer sodium, so the caesium ion in site 3 was exchanged for sodium. Crystallographic water molecules were retained. The protein along with ions and protein-bound water molecules was inserted into the pre-generated POPC membrane slab, and water molecules for solvation were further added using the Solvate Plugin of VMD (Humphrey et al. 1996). Ions were added with the Autoionize Plugin in VMD to create a neutral system with an approximate ion concentration of 150 mM. The anion was chloride and because the NaK channel transports both sodium and potassium, approximately half of the cations were sodium and the other half potassium.
The system was then energy minimized for 5000 steps and equilibrated for 10 ns with protein heavy atoms along with ions bound to the filter harmonically restrained with a force constant of 7 kcal mol -1Å-2 . Lipid molecules which, after 10 ns, were still stuck in odd positions between protein helices, resulting from the self-assembly process, were deleted and the system was energy minimized for 5000 steps and 500 ns of unbiased MD simulation was performed. This structure (with the oddly positioned lipid molecules removed) was further used for construction of the mutant systems using the Mutator Plugin of VMD. These systems were likewise energy minimized for 5000 steps followed by 500 ns of unbiased MD simulation. In addition to the NaK crystal structure (WT, mimicking the GluK2 homotetramer), we simulated the NaK structure with S57P in two opposing chains (A and C in 2Pro, mimicking the GluK2/GluK2 heteromer if assuming a 2:2 assembly with GluK2-GluK5-GluK2-GluK5 packing at the membrane level) and the NaK structure with the S57P mutation imposed in all four chains [4Pro, mimicking the GluK2(G615P) homotetramer]. Each 500 ns simulation was repeated, so the final data set consisted of 2×500 ns for each of the protein constructs.
All atomistic simulations were performed with NAMD v2.9 (Phillips et al. 2005) using the CHARMM 36 force field (Klauda et al. 2010;Best et al. 2012) for protein and lipids and the TIP3P water model (Jorgensen et al. 1983). Simulations were performed in the NPT ensemble at a temperature of 310 K and a pressure of 1 atm. Langevin dynamics with a damping coefficient of 0.5 ps −1 was employed to control the temperature. The pressure was maintained using the Langevin piston Nosé-Hoover method with a piston period of 100 fs and a damping timescale of 50 fs. Lennard-Jones interactions were treated with a cutoff of 12Å and the switching function applied from 10Å, while full electrostatics were accounted for using Particle-Mesh Ewald (PME) (Darden et al. 1993) summation with a real-space cutoff of 12Å and a 1Å grid spacing. The pairlist, with a cutoff of 14Å, was updated every 20 fs. Periodic boundary conditions were applied in all simulations. The time step was 2 fs with full electrostatics calculated every second step. Water molecules were kept rigid by the SETTLE (Miyamoto & Kollman, 1992) algorithm and other bonds to hydrogen atoms by the SHAKE (Ryckaert et al. 1977) algorithm. Snapshots were collected every 5 ps.

NaK-polyamine complexes
The most symmetrical and the most asymmetrical snapshot in terms of cross-pore distance (see Fig. 6D below) was taken from each of the simulations of WT, 2Pro and 4Pro, leaving out the first 100 ns of simulation. Spm was inserted into the filter region by aligning the charged nitrogen atoms of Spm with the original position of the four cations in the filter. These four ions were then deleted to remove overlap and ensure charge neutrality, and furthermore water molecules in the filter overlapping with Spm were deleted. Spd and Put were inserted to the filter region of WT snapshots in a similar way, although due to the smaller number of nitrogen atoms in these ligands, central carbon atoms were used along with the nitrogen atoms in the alignment to original ion positions. For Spd, J Physiol 594.7 three of the original cations in the filter were deleted, and for Put, only two. No other ions were binding to the filter region in any of the complexes. The simulation systems were then energy minimized for 2000 steps followed by 30 ns of unbiased MD simulation.
All polyamines remained bound to the filter region throughout the 30 ns of relaxation. To probe the energetics involved in the unbinding of the polyamines from the filter, 15 ns of steered MD (SMD) simulations were performed on all NaK-polyamine complexes, starting after 30 ns of unbiased simulation. The SMD atoms, here the terminal nitrogen atom facing the pulling direction, were harmonically constrained with force constant of 7.2 kcal mol -1Å-2 , corresponding to 500 pNÅ -1 , to move with velocity 2.5Å ns -1 in the negative direction along the z-axis (towards the extracellular side for the NaK channel, corresponding to the intracellular side for the KAR channel). Forces were saved every 200 fs. To avoid the whole system being pulled along z, the C α atoms of residues 23-28 were restrained harmonically with a force constant of 7.2 kcal mol -1Å-2 .
The work was calculated from the force profiles by numerical integration according to: where v is the constant velocity (2.5Å ns -1 ), f(t) is the force at the specific time step given in the SMD output and dt is the time step between the data points (200 fs).

Heteromerization and auxiliary proteins attenuate inward rectification of KARs
To better understand how heteromerization and auxiliary proteins attenuate channel block, we studied KAR responses in outside-out membrane patches at a range of membrane potentials to observe when polyamines enter the pore to block it. To mimic the time course of neurotransmitter release in the synaptic cleft, 1 mM L-glutamate (L-Glu) was applied to excised patches for 1 ms and 60 μM Spm was included in the internal solution to match known cytoplasmic levels of the blocker (Bowie & Mayer, 1995). Figure 1 summarizes these experiments, where the degree of inward rectification induced by polyamine block of GluK2 homomers was compared to that of GluK2/GluK5 heteromers or GluK2 homomers co-assembled with the auxiliary proteins, Neto1 and Neto2.
To better understand the voltage-dependency of the onset and relief of block, we converted the I-V data from each dataset into G-V plots (Fig. 2). KARs exhibit significant outward rectification at positive membrane potentials in the absence of channel blockers (Bowie et al. 1998). To take this into account, we performed additional experiments to obtain control I-V data for all receptor combinations and then subtracted this non-linearity from data obtained in the presence of Spm (Figs 2 and 3). Corrected G-V plots were then fit with a single binding site model of channel block (Bowie et al. 1998) (see Methods) to estimate the equilibrium dissociation constant of Spm at 0 mV (K d(0mV) ) as well as the voltage-dependency of the onset and relief of block (h and k, respectively; Table 2).
In agreement with previous work (Bowie et al. 1998), the K d(0mV) of Spm for GluK2 homomers was estimated to be 5.8 ± 0.3 μM, with voltage-dependencies of the onset and relief of block of −16 ± 0.5 and +20.4 ± 1.0 mV per e-fold change in membrane potential, respectively (Table 2). In comparison, KAR heteromerization and association with auxiliary proteins affected both Spm binding and the voltage-dependency of block. Fits of corrected G-V plots estimated the K d(0mV) of Spm to be 39 ± 2, 71 ± 13 and 119 ± 30 μM for GluK2+Neto2, GluK2+Neto1 and GluK2/GluK5 channels, respectively (Table 2), representing a 6-to 20-fold reduction in blocker affinity. Interestingly, heteromerization and auxiliary protein association mainly affected the voltage-dependency of the relief of block (i.e. k), which was about 2-to 3-fold weaker compared to that of GluK2 homomers (Table 2). For example, fits of corrected G-V plots estimated the voltage-dependency of relief for GluK2/GluK5 heteromers to be +69.7 mV per e-fold change in membrane potential compared to +20.4 mV for GluK2 homomers. In keeping with this, selective activation of GluK2/GluK5 heteromers with either 1 mM AMPA (Herb et al. 1992) or 300 μM L-Glu (Barberis et al. 2008) estimated K d(0 mV) to be 81 ± 40 and 170 ± 33 μM, respectively, with the voltage-dependency of relief corresponding to +56.8 and +107.7 mV per e-fold change in membrane potential, respectively (Table 2). In contrast, the differences between the voltage-dependency of onset of block (i.e. h) were more modest (Table 2), and the binding rates for Spm (k on , Fig. 2C) were assumed to be constant. The unbinding (k off ) and permeation rates (k perm ) were estimated using previous estimates of k on (Fig. 2C) (Bowie et al. 1998). Plots of unbinding (k off ) and permeation (k perm ) for Spm at GluK2 homomers predict that relief of block occurs only at extreme membrane potentials (i.e. > +50 mV), where the rates for k perm become dominant (Fig. 2C). In contrast, similar plots of k off (Fig. 2E) and k perm (Fig. 2F) with GluK2/GluK5 heteromers, GluK2+Neto1 or GluK2+Neto2 ( Fig. 2E and F) reveal that there are substantial shifts in estimates of unbinding and permeation rates that account for the overall attenuation in channel block. Specifically, there is a substantial hyperpolarizing shift in k perm so that Spm permeation is expected to be appreciable even at membrane potentials of −50 mV (Fig. 2F).
Given the specific nature of these effects, we reasoned that the attenuation of polyamine channel block by heteromerization or association with auxiliary proteins occurs because Spm permeation rates are enhanced under these conditions. To test this experimentally, we performed bi-ionic reversal potential experiments to determine the extent to which KAR composition affected polyamine permeation rates.

Heteromerization and auxiliary proteins relieve channel block by facilitating polyamine permeation
To determine the permeability of Spm relative to that of Na + (P Spm /P Na ), we measured the reversal potential of KAR responses in solutions where the main external permeant cation was either 150 mM Na + or 90 mM Spm. In all cases, the main permeant cation of the internal solution was 150 mM Na + (Fig. 4 and see Methods).
As expected, responses mediated by GluK2 (V rev = −0.2 ± 1.5 mV, Fig. 4A and C, inset), GluK2/GluK5 (V rev = −0.5 ± 0.8 mV, Fig. 4D and F, inset) and GluK2+Neto2 (V rev = −0.7 ± 0.8 mV, Fig. 4G and I, inset) in symmetrical 150 mM Na + solutions reversed close to 0 mV, demonstrating that Na + ions are equally permeable from both sides of the plasma membrane. In contrast, reversal potentials determined with external 90 mM Spm as the main permeant cation were dependent upon the KAR composition. With GluK2 homomers, we failed to observe a reliable inward current in 90 mM external Spm and therefore were unable to estimate P Spm /P Na (Fig. 4B and C), an observation similar to a previous study of GluK2+Neto1 GluK2+Neto2 polyamine permeation (Bahring et al. 1997). In keeping with this, preliminary experiments suggested that polyamines with fewer charged groups, namely Put (z = +2) and Spd (z = +3), were more permeant than Spm (z = +4). Reversal potentials with 90 mM external Put or Spd as the main permeant cation were −8.0 ± 2.7 mV (n = 2) and −34.4 ± 1.6 mV (n = 3), respectively, estimating P Put /P Na to be 0.23 ± 0.02 and P Spd /P Na to be 0.049 ± 0.002 for GluK2 homomers. This indicates that permeant channel blockers with smaller valencies have a greater tendency to permeate through the KAR channel, in agreement with prior work (Bahring et al. 1997;Cui et al. 1998). In contrast to GluK2 homomers, Spm-mediated inward currents were observed with both GluK2/GluK5 ( Fig. 4E and F) and GluK2+Neto2 ( Fig. 4H and I), with reversal potentials of −43.2 ± 1.2 mV (n = 3) and −33.6 ± 4.5 mV (n = 4), respectively. As a result, the calculated P Spm /P Na for GluK2/GluK5 and GluK2+Neto2 receptors were 0.024 ± 0.001 and 0.042 ± 0.009, respectively. Taken together, these data provide evidence that GluK2/GluK5 and GluK2+Neto2 receptors were more permeable to Spm compared to GluK2 homomers. A reduction in polyamine block is thought to go hand-in-hand with Ca 2+ permeability in non-NMDA receptors (Burnashev, 1996). Despite these differences, KAR heteromerization or association with the auxiliary proteins do not attenuate Ca 2+ permeability. For example, in Ca 2+ -rich external solutions (i.e. 30 mM CaCl 2 + 105 mM NaCl), the reversal potential of GluK2 I-V plots was +1.0 ± 0.5 mV (n = 5) which was shifted to a slightly more positive potential of +5.0 ± 1.8 mV (n = 3) with GluK2/GluK5 heteromers ( Fig. 5E and F). As a result, the relative Ca 2+ permeabilities (P Ca /P Na ) of GluK2 homomers and GluK2/GluK5 heteromers were estimated to be 2.3 and 3.1, respectively. As a previous study has reported similar findings with the auxiliary proteins, Neto1 and Neto2 (Fisher & Mott, 2012), we concluded that differences in the pore properties of KARs induced by heteromerization or auxiliary proteins could not be revealed by studying divalent permeability; we therefore focused on polyamine block. Thus, KARs essentially . k on was estimated using values from Bowie et al. (1998). D, corrected G-V plots for GluK2+Neto1 (cyan, n = 12) and GluK2+Neto2 (orange, n = 7). All data are represented as the mean ± SEM. G-V plots were fit with eqn (2). E and F, comparison of the estimated unbinding (k off , E) and permeation (k perm , F) rates for GluK2 (1), GluK2/GluK5 (2), GluK2+Neto1 (3) and GluK2+Neto2 (4).
behave as molecular sieves (Burnashev, 1996;Bahring et al. 1997) such that the movement of bulky permeating ions (i.e. Spm) is hindered in the permeation pathway while that of much smaller divalent cations, such as Ca 2+ ions, is not. As explained below, we performed MD simulations to better understand this distinction by looking at how the pore region of KARs might be affected by modest changes in the ion-permeation pathway.

MD simulations predict a structural change in the channel pore
Sequence alignment of all KAR subunits was performed to identify differences between the GluK2 and GluK2/GluK5 receptors (Fig. 6A). The most striking difference was found in the α-helical region of the re-entrant loop (P-loop; Fig. 6A) where a glycine residue (i.e. GluK2 Gly615) is found in all primary subunits (GluK1-3) while a proline residue is found in the equivalent position (i.e. GluK5 Pro599) of secondary subunits (GluK4 and GluK5), which do not form functional homomers. As each residue is located within a predicted α-helical region of the re-entrant P-loop of KARs (Panchenko et al. 1999) (Fig. 6A), we reasoned that the Gly and Pro may exert differential effects on channel block and permeation. Specifically, we hypothesized that the Pro of the GluK5 subunit would result in structural changes in the heteromeric KAR pore regions, as has been proposed for Pro residues in other α-helical regions of ion-channels, such as the S6 helix of Shaker  and the M2 helix of KcsA K + channels (Fowler & Sansom, 2013). To explore this, MD simulations were used to study how the introduction of the Pro residue may affect the pore structure and its dynamics (Fig. 6B-D). As a high-resolution structure of the KAR pore region is not yet available, we performed MD simulations using the structure of the Bacillus cereus NaK channel pore [PDB 3E86 (Alam & Jiang, 2009a), 1.6Å resolution] as it has been used successfully to provide insight into the NMDA-type iGluR (Siegler Retchless et al. 2012). This structure has 4-fold symmetry, as expected for the GluK2 tetramer (Wilding et al. 2014) and furthermore illustrates an open conformation ( Fig. 6B and C). Although the iGluR pore structure has been compared to the KcsA structure (Sobolevsky et al. 2009), it also shares remarkable structural similarities with the NaK channel pore (Fig. 7). Moreover, because the NaK channel is a non-selective cation channel, its pore properties are expected to be closer to that of the iGluR. For these reasons, the NaK structure was used as a model of the pore in our MD simulations. To simulate a heteromer having a 2:2 stoichiometry, as is the case for GluK2/GluK5 KARs (Reiner et al. 2012), prolines were introduced into the equivalent α-helical positions of the NaK structure on opposing subunits (i.e. A and C), in agreement with prior work on KAR heteromers (Kumar et al. 2011). Changes in cross-pore distances between the centre of mass of the α-carbons of residues 50-53 in opposing subunits are summarized in Fig. 6D. As expected, the A/C and B/D cross-pore distances in the wild-type (WT) NaK pore quickly reached equilibrium and remained fairly constant for the rest of the 500 ns simulation (Fig. 6D and supplementary Movies S1 and S2), consistent with a 4-fold symmetrical pore arrangement. In contrast, the pore adopted a 2-fold symmetry with prolines on opposing subunits ( Fig. 6D and Movies S1 and S2). For example, the mean distance between helices A/C increased by approximately 1Å, while the distance between WT helices (B/D) decreased reciprocally (Fig. 6D and Movies S1 and S2). Interestingly, the two simulation repeats with prolines introduced in all four helices illustrated two different outcomes (Fig. 6D, 4Pro). In one case, helices adopted 30% Pk -100mV 5% Pk -100mV 10 ms 30% Pk -100mV 10 ms 30% Pk -100mV 10 ms 5% Pk -100mV  a 4-fold symmetry but with larger separation than the WT (Fig. 6D, Run 1); in the second case, they adopted a 2-fold symmetry (Fig. 6D, Run 2). Thus, despite the trajectories showing two different effects of proline insertions, they probably reflect the same result, namely that the Pro residue induces a more stochastic behaviour of the P-loop compared to the WT pore conformation.
To better understand how the Pro residues affect the dynamics of the NaK α-helical P-loop, we examined the root mean square deviation (RMSD) of each helix in comparison to the reference structure ( Fig. 8; PDB #3E86). Interestingly, the mean RMSD of Pro-containing helices in all simulations was about 0.5Å, a 2-fold increase compared to the RMSD of WT helices (Fig. 8), indicating that they are able to move further away compared to their WT counterparts. Moreover, the RMSD fluctuation in the Pro-containing helices was much greater than in WT helices, spanning almost 0.2Å, while the WT fluctuations spanned less than 0.1Å (Fig. 8). Given this, we hypothesized that increased dynamics in the P-loop α-helices may be responsible for the enhanced polyamine permeation rates observed experimentally.

MD simulations predict increased polyamine permeation
To test whether Pro-induced changes in the position of the helices could facilitate blocker permeation, we compared the work required to pull pore-bound polyamines to the extracellular side of WT and mutant NaK pore regions, corresponding to the intracellular side in KARs (Fig. 9). Polyamines were also pulled to the intracellular side of the NaK pore, but the results from these experiments were inconclusive. From each of the WT trajectories in Fig. 6D, the most symmetrical and the most asymmetrical snapshot (in terms of the cross-pore distance) was found and used as a starting structure, giving rise to a total of four different protein structures. Three polyamines of different structure and charge (Put, Spd and Spm) were modelled into the filter by overlaying with the original cations bound in the filter of the NaK structure ( Fig. 9A) (Alam & Jiang, 2009a). After 30 ns of relaxation, all polyamines were still bound to the filter region. To probe the strength with which the polyamines bind to the filter, SMD simulations were subsequently performed. In the SMD simulations, polyamines were pulled out of the filter along the z-axis toward the intracellular side of the membrane (Movie S3). Each SMD simulation produced a force profile from which the work involved in the process was calculated (Fig. 9B). The eight snapshots shown in Fig. 10 correspond to time steps at which the force peaks, each denoted in Fig. 9B by an asterisk, along with snapshots before and near the end of the SMD simulation. From the SMD simulations with different polyamines, a clear trend in the work values was observed ( Fig. 9C and Movie S3), suggesting that, in the WT pore, it is easier for the smaller polyamines (Put and Spd) to pass through from the filter than Spm, in agreement with our data and that of others (Bahring et al. 1997). Similar SMD simulations were performed with Spm inserted into the most symmetrical or the most asymmetrical snapshots from the 2Pro and 4Pro trajectories from Fig. 6D. As . Prolines in GluK5 pore are predicted to alter pore dimensions A, sequence alignment of the kainate receptor subunits. A proline is present in the secondary subunits where a conserved glycine is present in primary subunits. This residue is located within a region with a predicted α-helical structure of the P-loop. B, crystal structure of the NaK channel (PDB#3E86). Proline residues were mutated in the equivalent α-helical structure of subunits A and C (yellow spheres). C, side view of the A/C (cyan) and B/D (orange) subunits of the inverted NaK pore before (grey) and after (coloured) 257 ns simulations. D, cross-pore distance measurements (see black spheres in C) were measured for 500 ns (two repeats for each condition).
before, all Spm molecules were still bound to the filter region in the 12 different systems (4×WT, 4×2Pro, 4×4Pro) after 30 ns of relaxation. The calculated work values for pulling Spm to the solvent were lower for the 2Pro and 4Pro mutants than for the WT pore (Fig. 9D). Thus, this model system suggests that the Pro residue in the α-helical region of the GluK5 P-loop may be enough to disturb Spm binding and ensure higher permeation and unblock in the GluK2/GluK5 heteromers.

Pro599 in the GluK5 pore is the main determinant of polyamine block
To assess whether GluK5 Pro599 is responsible for enhanced polyamine permeation, we measured the degree of channel block after substituting it with a Gly residue, as is found in GluK2 (Fig. 9E-G). As anticipated, the degree of inward rectification of GluK2/GluK5(P599G) channels was significantly more pronounced than observed with WT heteromers (Fig. 9F). This finding demonstrates that Pro599 is a critical determinant of polyamine block. Fits of the corrected G-V plot estimated the K d(0mV) of GluK2/GluK5(P599G) receptors to be 19 ± 2 μM, which was 4-5 fold more potent than WT heteromers with a K d(0mV) of 119 ± 30 μM ( Fig. 9G and Table 2).

GluA2 PDB#3KG2
NaK PDB#3E86 Figure 7. The NaK structure is similar to the GluA2 structure The inverted structure of the open NaK channel (PDB #3E86, cyan) is overlaid with that of the closed GluA2 (PDB #3KG2, orange), illustrating that the two channels share a similar architecture. The proline position is indicated by a sphere. Only two chains are shown and for GluA2, only transmembrane helices M1-3 are included. The two structures are aligned using M2, C α atoms of residues 573-586 in GluA2 and of residues 50-63 in the NaK channel. The position equivalent to the proline in GluK5 is highlighted by showing the C α atom of this residue with a sphere (Ser580 in GluA2, Ser57 in NaK). A sequence alignment is shown, highlighting the Gly and Pro residues in GluK2 and GluK5, respectively, which correspond to an Ser in NaK.
Interestingly, the K d(0mV) of Spm for GluK2/GluK5 (P599G) was statistically indistinguishable from GluK2 homomers [one-way ANOVA, F 7,28 = 8.506, P = 0.000014; post hoc Tukey honest significant difference (HSD) pairwise comparison, P = 1.000; Table 2], indicating that the removal of Pro599 from GluK2/GluK5 heteromers enhances channel block. As a result, the differences between the G-V plots of GluK2 (Fig. 2B) and mutant heteromers (Fig. 9G) are found in the voltage-dependency of block (Table 2). To test if the reverse substitution would render GluK2 receptors polyamine-insensitive, Gly615 was replaced with a Pro residue, as is found in GluK5 (Fig. 9H-J). In contrast to WT GluK2, the I-V plot for GluK2(G615P) in the presence of 60 μM internal Spm was almost entirely linear, demonstrating that the introduction of Pro615 in the pore region of GluK2 eliminates polyamine block (Fig. 9I). In agreement with this, the corrected GluK2(G615P) G-V plot lacked voltage-dependency (Fig. 9J), further underlining that this Pro residue is the principal determinant of Spm block in GluK2/GluK5 heteromers.

Auxiliary proteins and GluK5 relieve polyamine block by distinct mechanisms
To test whether auxiliary proteins relieve KAR channel block by a similar or different mechanism, we took advantage of the fact that GluK2(G615P) channels are unaffected by Spm block (Fig. 9H-J). We reasoned that if auxiliary subunits attenuate Spm block by a different mechanism, their co-expression with GluK2(G615P) The RMSD values calculated for the C α atoms of the P-loop helix (residues 50-62) of each chain of the NaK channel over the course of the 500 ns simulation. The alignment and RMSD calculation were performed for one chain at a time, aligning by residues 50-62 of the given helix for which the RMSD is calculated. WT helices are shown in black/grey, helices from the 2Pro simulation in red/orange and those of the 4Pro simulation in blue/cyan. would increase the relative Spm permeability of the channel complex. Conversely, if they attenuate Spm block by a similar mechanism, the ability of Spm to permeate GluK2(G615P) channels would remain unchanged by co-expression with an auxiliary subunit. In contrast to WT GluK2 ( Fig. 11A and D), Spmmediated inward currents were readily observed at negative membrane potentials with GluK2(G615P) (Fig. 11B and E). The reversal potential of GluK2(G615P) with 90 mM external Spm was −25.5 ± 2.4 mV (n = 5) ( Fig. 11B and E), estimating the P Spm /P Na to be 0.062 ± 0.008 (Fig. 11G), similar to GluK2/GluK5 and GluK2+Neto2 (Fig. 11G).
Interestingly, the magnitude of Spm-induced inward currents in patches containing GluK2(G615P) channels co-expressed with Neto2 was increased by about 2-fold ( Fig. 11C and F). For example, the Spm-mediated current at −100 mV was 8 ± 5% (n = 5) of the Na + -mediated current with GluK2(G615P) and 19 ± 12% (n = 7) with GluK2(G615P)+Neto2. In addition, the reversal potential observed with 90 mM external Spm was more positive for GluK2(G615P)+Neto2 than GluK2(G615P) alone (V rev = −15.6 ± 2.3 mV; n = 4), estimating P Spm /P Na to be 0.12 ± 0.02 (Fig. 11G). One-way ANOVA performed on the relative Spm permeability revealed that P Spm /P Na for GluK2(G615P)+Neto2 was significantly larger than for GluK2(G615P), GluK2+Neto2 and GluK2/GluK5 (Fig. 11, see figure legend for P values). These results indicate that co-expression of KARs with Neto2 increases blocker permeation in an additive manner, suggesting that it is acting through a mechanism that is distinct from that of the Pro residues in GluK5 subunits.

Discussion
The present study advances our understanding of polyamine block of KARs in several fundamental ways. First, we report that heteromerization and auxiliary proteins unexpectedly attenuate channel block by enhancing polyamine permeation rates. Second, MD simulations suggest that relief of block is due to a structural change in the α-helical region of the KAR pore, which is distinct from the charge-screening mechanism ascribed to GluA2-containing AMPARs. Our data identify that each of two GluK5 subunits in the GluK2/GluK5 heteromer possesses a critical proline residue that is predicted to push the helices outward, leading to a 2-fold symmetrical pore arrangement and increased pore helix fluctuation. Third and finally, heteromerization and auxiliary proteins relieve polyamine block by a distinct pathway. Consequently, Neto1 and Neto2 do not attenuate polyamine block by channelling their effect through the proline residue of GluK5 subunits. Our study reveals that subunit composition affects the functional properties of KARs differently than AMPARs. In AMPARs, the Q/R site of the GluA2 subunit is >99% edited to an Arg but remains unedited in all other subunits (Sommer et al. 1991). This distinction is significant because it generates the two major AMPAR classes that are either GluA2-containing, polyamine-insensitive and divalent cation-impermeable or GluA2-lacking, polyamine-sensitive and divalent cation-permeable (Cull-Candy et al. 2006) (but see Bowie, 2012). Importantly, polyamine-insensitivity of GluA2-containing AMPARs is thought to be due to the electrostatic effect of Arg at the Q/R site of GluA2, which repels polyamines in the narrowest region of the pore (Bowie et al. 1999). However, this reliance on the editing state of the Q/R site is not thought to be as important for KAR heteromers because it is edited only 40% in GluK1 subunits with a range of estimates for GluK2 from <5 to 75% in the developing and adult rodent brain (Sommer et al. 1991;Puchalski et al. 1994;Schmitt et al. 1996).
It has yet to be tested whether the Q/R site of other KAR subunits (i.e. GluK3-5) is edited or not. Our data demonstrate that GluK2/GluK5 heteromers attenuate polyamine block not by electrostatic repulsion but rather through a novel mechanism involving a structural change in the pore helices. Accordingly, the binding affinity for polyamines in the pore is reduced, enhancing their permeation rates. This is achieved through the distinctive cyclical structure of the proline side chain that is conformationally rigid and lacks the N-H hydrogen bond donor, and thus can uniquely distort α-helical structures. As proline residues are found exclusively in GluK4-5 subunits and are absent from GluK1-3 subunits, we predict that heteromers containing either GluK4 or GluK5 will exhibit a diminished level of polyamine block even if the Q/R site of all four subunits is unedited, as is the case in this study. This arrangement is distinct from AMPARs where the coupling of polyamine block and divalent cation permeability is governed by a common structural determinant, namely the Q/R site. As different structural domains govern polyamine block and Ca 2+ permeability in KAR heteromers, native receptors will exhibit appreciable Ca 2+ permeability that is unconstrained by significant voltage-dependent modifications in membrane conductance due to polyamine block.
the NaK tetramer, resulting in a 2-fold symmetrical pore arrangement that alters the pore dynamics. In keeping with this, recent studies have shown that GluK2/GluK5 heteromers have a subunit copy number of 2:2 (Reiner et al. 2012). Interestingly, prolines are found in many transmembrane α-helices of transport proteins and ion channels, as is the case in the present study, but are uncommon in water-soluble helices (Brandl & Deber, 1986;Barlow & Thornton, 1988;Tieleman et al. 2001). They are most frequently located on the hydrophilic side of the helix (Woolfson & Williams, 1990), where they destabilize the helix and tend to produce hinges, kinks or swivels . Given these observations, we hypothesized that the proline found in the re-entrant P-loop of the GluK5 subunit might result in a similar structural rearrangement in the KAR channel pore, as has been proposed for the S6 α-helix in Shaker  and the M2 helix in KcsA potassium channels (Fowler & Sansom, 2013). In the latter case, the introduction of prolines in the M2 helices predicted the formation of kinks and helical rearrangement after only 10 ns of MD simulation, with the pore helices adopting 2-fold symmetry instead of 4-fold symmetry (Fowler & Sansom, 2013). Similarly, the introduction of prolines in the P-loop of the NaK channel pore, which was used in the present study as a model of the KAR channel pore, resulted in small changes in the position of the outer region of the helices. Accordingly, our data suggest that reduced channel block and enhanced polyamine permeation rates occur through an increased structural flexibility of the pore α-helices due to proline residues that bring about an asymmetrical re-arrangement and enlargement of the KAR channel pore.

Auxiliary proteins and polyamine block
Auxiliary proteins of AMPARs and KARs are not pore-forming components of the channel (Chen et al. 2000;Tomita et al. 2003;Zhang et al. 2009) and therefore it has been unclear how they cause attenuation of polyamine channel block. The proximal end of the C-terminal region of the TARP, γ-2 or stargazin, has been shown to attenuate polyamine block of AMPARs through a mechanism that involves a direct interaction with the intracellular domain of GluA1 or GluA4 subunits (Soto et al. 2014). A similar finding has been reported with Neto2, where positive residues in the proximal intracellular C-terminal domain are implicated in reducing polyamine block (Fisher & Mott, 2012). Although it is still unclear how these residues affect inward rectification of GluK2 channels, we found that the auxiliary protein achieves this by increasing the blocker permeation, which seems at odds with the charge screening hypothesis. Our data demonstrate that Neto1 and Neto2 exert a structural effect on the channel pore that is distinct from that seen in GluK2/GluK5 heteromers. Exactly how auxiliary proteins achieve this remains to be investigated. However, given the similarity between the effect of auxiliary proteins on KARs and AMPARs, we predict that TARPs and cornichons are likely to attenuate polyamine block of AMPARs by the same mechanism, i.e. by increasing polyamine permeation. J Physiol 594.7

Supporting information
The following supporting information is available in the online version of this article.
Movie S1. MD simulation of the NaK channel with 2 Pro substitutions. This movie illustrates the MD simulation of the 2Pro mutant of the NaK channel over 500 ns. Ser57 was substituted with a Pro residue in opposing subunits A and C, and the Cα of the inserted Pro is shown as a yellow sphere. The NaK structure is inverted so that the orientation of the helices matches that of the kainate receptor. The four subunits of the tetramer are shown as pairs of opposing subunits. The A and C subunits are on the left (cyan) and the B and D subunits are on the right (orange). The original protein structure is shown in white; the selectivity filter is highlighted in magenta. The Pro-containing α-helices (cyan, including the selectivity filter) tend to move outward, while the WT helices (orange) tend to move inward.
Movie S2. MD simulation of the NaK with 2 Pro substitutions, intracellular view. This movie illustrates the same MD simulation as in Movie S1, viewed from the intracellular side. The Pro-containing helices (cyan) seem to bend back, pulling the selectivity filter (magenta) away from the original crystal structure (white). In contrast, the selectivity filter for the WT helices tends to move slightly closer, causing a more asymmetrical arrangement of the filter, which in turn may affect polyamine permeation. Movie S3. SMD simulation pulling pore-bound polyamines to the intracellular side. This movie illustrates the binding (from left to right) of Spm, Spd and Put to the WT NaK pore region simulated for 30 ns of relaxation before being pulled to the intracellular side. Ligands are illustrated with cyan carbon atoms and for clarity only chains A and C are shown. Protein residues in the filter are shown with licorice representation along with other protein residues and water molecules within 4Å of the ligands.