β11-12 linker isomerization governs acid-sensing ion channel desensitization and recovery

Acid-sensing ion channels (ASICs) are neuronal sodium-selective channels activated by reductions in extracellular pH. Structures of the three presumptive functional states, high-pH resting, low-pH desensitized, and toxin-stabilized open, have all been solved for chicken ASIC1. These structures, along with prior functional data, suggest that the isomerization or flipping of the β11–12 linker in the extracellular, ligand-binding domain is an integral component of the desensitization process. To test this, we combined fast perfusion electrophysiology, molecular dynamics simulations and state-dependent non-canonical amino acid cross-linking. We find that both desensitization and recovery can be accelerated by orders of magnitude by mutating resides in this linker or the surrounding region. Furthermore, desensitization can be suppressed by trapping the linker in the resting state, indicating that isomerization of the β11–12 linker is not merely a consequence of, but a necessity for the desensitization process in ASICs.


Introduction
6 synthesis (Genescript, USA) using published sequences (Ye et al., 2008). Transfection was 22 performed using PEI 25k in a mass ratio of 1:3 (cDNA:PEI) for 6 to 8 hours, then the media was 23 replaced with fresh supplemented MEM containing 40 µM MeO-Bpa, a methyl ester derivative 24 of Bpa. Transfected cells were used for experiments 24-30 hours after the beginning of 25 transfection. 26 Electrophysiology and UV trapping 27 Culture dishes were visualized using a 20x objective mounted on a Nikon Ti2 microscope with 28 phase contrast. A 470 nm LED (Thorlabs) and dichroic filter cube were used to excite GFP and 29 detect transfected HEK cells. Outside-out patches were excised using heat-polished, thick-walled 30 borosilicate glass pipettes of 3 to 15 MΩ resistance. Higher resistance pipettes were preferred 31 for non-stationary noise analysis experiments. The pipette internal solution contained (in mM) 32 135 CsF, 33 CsOH, 11 EGTA, 10 HEPES, 2 MgCl2 and 1 CaCl2 (pH 7.4). External solutions with a pH 33 greater than 7 were composed of (in mM) 150 NaCl, 20 HEPES, 1 CaCl2 and 1 MgCl2 with pH values 34 adjusted to their respective values using NaOH. For solutions with a pH lower than 7, HEPES was 35 replaced with MES. All recordings were performed at room temperature with a holding potential 36 of -60 mV using an Axopatch 200B amplifier (Molecular Devices). Data were acquired using 37 AxoGraph software (Axograph) at 20-50 kHz, filtered at 10 kHz and digitized using a USB-6343 38 DAQ (National Instruments). Series resistance was routinely compensated by 90 to 95% where 39 the peak amplitude exceeded 100 pA. Rapid perfusion was performed using home-built, triple-40 barrel application pipettes (Vitrocom), manufactured according to MacLean (MacLean, 2015). 41 Translation of application pipettes was achieved using a piezo translator (P601.40 or P212.80, PI) 42 mounted to a manual manipulator and driven by a voltage power supply (E505.00 or E-471.20, 7 PI). Voltage commands to the piezo were first low-pass filtered Frequency 44 Devices) at 50-100 Hz. Solution exchange was routinely measured at the end of each patch 45 recording using open tip currents with exchange times ranging from 250 to 500 µs. 46 For UV modulation, a high-power UV LED (KSL2-365, Rapp Optoelectronic) was used as the UV light source. The UV LED was set to maximum power and triggered by TTL input. The light 48 emission was reflected off a 425 nm long-pass dichroic mirror held in a beam combiner (which 49 combined the light from the 470nm LED for GFP visualization), on through the epifluorescence 50 port of the Ti2 microscope then reflected off of a 410 nm long-pass dichroic mirror before being 51 focused onto the sample through a 20x objective. For trapping experiments, a single sweep of 52 UV involved 14 LED pulses of 50 ms in duration spaced by 450 ms, leading to a total of 700 ms 53 exposure time spread across 7 seconds. 54 Molecular Dynamics Simulations 55 Molecular dynamics simulations were performed using a structure of chicken ASIC1 suggested to 56 be in the desensitized state (PDB code 4NYK, (Gonzales et al., 2009)), solved to a resolution of 3 57 Å. Residues 42-455 were resolved in the crystal structure. Of these residues, 23 had missing side 58 chain atoms. The missing atoms were added using MODELLER 9v20 (Sali and Blundell, 1993), 59 while the intracellular N-and C-termini were ignored. For each chain, the bound chloride ion and 60 the 50 crystallographically resolved water molecules were retained. The initial membrane 61 position was obtained from the Orientation of Proteins in Membranes database (Lomize et al.,62 2012). The simulated system, consisting of the protein, the chloride ions and crystallographic 63 water molecules, embedded in a POPC lipid bilayer and surrounded by TIP3P water molecules 8 and a NaCl concentration of 150 mM, was generated using the CHARMM GUI (Jo et al., 2008;Lee 65 et al., 2016). Disulphide bonds for each chain was maintained between the following cysteine 66 pairs: C94-C195, C173-C180, C291-C366, C309-C362, C313-C360, C322-C344 and C324-C336. The 67 POPC bilayer was 120 Å x 120 Å, and the box length 146 Å. In the desensitized state, a number of 68 acidic residues are believed to be protonated, however, exactly which residues is unclear. Since 69 covalent bonds cannot be formed or broken during classical molecular dynamics simulations, 70 which residues to protonate must be determined prior to performing simulations. Based on 71 preliminary work, combining molecular dynamics simulations and PROPKA pKa prediction (Olsson 72 et al., 2011) (Musgaard, unpublished data), we chose to protonate a set of acidic residues to 73 stabilize the given structural conformation of the protein. Two histidine residues were 74 protonated as well, giving the following list of protonated residues: H74, E98, H111, E220, D238, 75 E239, E243, E255, E314, E354, D408 and E417. All other side chains were retained in their 76 standard protonation states. For the simulations mimicking a higher pH value, all residues were 77 kept in their standard ionization state (i.e., deprotonated for the acidic residues, neutral for 78 histidine). 79 For simulations of the L414A mutant, the L414 side chain was manually changed to an alanine 80 side chain prior to constructing the simulation systems. 81 The CHARMM36 force field was employed for proteins (Best et al., 2012) and lipids (Klauda et al.,82 2010), and the simulations were performed using GROMACS v 5.0.7 (Abraham et al., 2015). The 83 systems were simulated in the NPT ensemble using periodic boundary conditions, and the 84 equilibration protocol was as follows. The constructed systems were first energy minimized for 85 10,000 steps or until the maximum force acting on any atom was less than 1000 kJ mol -1 nm -1 . 9 This was followed by six shorter simulations, gradually releasing the position restraints as 87 suggested by the default CHARMM-GUI protocol. The first three short simulations were 25 ps 88 long and used a time step of 1 fs; the fourth and the fifth were 100 ns long, while the final part 89 of the equilibration was run for 2 ns. The equilibration simulations 4-6, as well as the production 90 run, used a time step of 2 fs. In all steps, the Verlet cutoff scheme was used with a force-switch 91 modifier starting at 10 Å and a cutoff of 12 Å. The cutoff for short-range electrostatics was 12 Å 92 and the long-range electrostatics were accounted for using the particle mesh Ewald (PME) 93 method (Darden et al., 1993;Essmann et al., 1995). The temperature was maintained at 310 K 94 for all steps of the equilibration using a Berendsen (Berendsen et al., 1984) thermostat, while the 95 Nose-Hoover thermostat (Hoover, 1985;Nose, 1984) was used to keep the temperature at 310 96 K for the production run. For the final four steps of the equilibration as well as for the production 97 run, the pressure was maintained at 1 bar, using semi-isotropic pressure coupling. The Berendsen 98 barostat (Berendsen et al., 1984) was employed for the equilibrations while the  Rahman barostat (Nose and Klein, 1983;Parrinello and Rahman, 1981) was used for the 100 production run. Covalent bonds including hydrogen atoms were constrained using the LINCS 101 algorithm (Hess, 2008). Snapshots were saved every 5 ps, and generally every fourth snapshot 102 was used for analysis. Three repeats for each setup were performed (a, b and c), using different 103 starting velocities for the first step of the equilibration. The simulation times were 3 x 200 ns for 104 the protonated systems (wild type and L414A) and 3 x 400 ns for the deprotonated systems (wild 105 type and L414A). Analysis was performed using standard tools in GROMACS as well as in-house 106 tcl scripts run through VMD v 1.9.3 (Humphrey et al., 1996) Where It is the fraction of the test peak at an interpulse interval of t compared to the conditioning 117 peak, τ is the time constant of recovery and m is the slope of the recovery curve. Each protocol 118 was performed between 1 and 3 times on a single patch, with the resulting test peak/conditioning 119 peak ratios averaged together. Patches were individually fit and averages for the fits were 120 reported in the text. N was taken to be a single patch.

121
For dose-response curves, patches were placed in the middle of a three-barrel application pipette 122 and jumped to either side to activate channels with the indicated pH. Responses to higher pH 123 values were interleaved with pH 5 applications on either side to control for any rundown. Peak 124 currents within a patch were normalized to pH 5 and fit to: where Ix is the current at pH X, pH50 is the pH yielding half maximal response and n is the Hill 127 slope. Patches were individually fit and averages for the fits were reported in the text. N was 128 taken to be a single patch.

129
For non-stationary fluctuation analysis, runs of between 50 and 200 responses from a single 130 patch were recorded. Within each recording, we identified the longest stretch of responses 131 where the peak amplitude did not vary by more than 10%. We further eliminated individual 132 traces with spurious variance such as brief electrical artifacts, resulting in blocks of 40-90 traces. 133 To further correct for rundown or drift in baseline values we calculated the variance between 134 successive traces, as opposed to calculating from the global average, using: Where δi 2 is the variance of trace i, Ti is the current value of the trace i. The ensemble variance 137 and current for each patch were divided into progressively larger time bins. The baseline variance 138 was measured from a 50 ms time window just prior to pH 5 application. The resulting mean 139 current-variance data were then fitted in Originlab using: Where σi(I) 2 is the variance, i is the single channel current, I is the average current, N is the 142 number of channels in the patch and σbaseline 2 is the baseline variance. For all experiments, N was 143 taken to be a single patch. Nonparametric two-tailed, unpaired randomization tests with 100,000 144 iterations were implemented in Python to assess statistical significance.

1
Leu414 strongly influences entry to and recovery from desensitization 2 The large extracellular domain of individual ASIC subunits has been likened to a hand shape with 3 distinct thumb, finger, knuckle and palm domains ( Figure 1A) ASIC recovery from desensitization has been mechanistically examined rarely in general (Kusama we employed a paired pulse protocol where an outside-out patch expressing cASIC1 was 23 incubated at pH 8 to maximally populate the resting state, followed by a jump for 1.5 seconds 24 into pH 5 to fully desensitize the channel population. Following this conditioning pulse, the patch 25 was exposed to pH 8 again for variable intervals, ranging from 3 ms to 30 seconds, to enable 26 some fraction of channels to recover before a 500 ms test pulse of pH 5 was applied ( Figure 1C).

27
A ratio of the second peak to the first enabled us to determine the fraction of the response 28 recovered as a function of the interval between the end of the conditioning pulse and the 29 beginning of the test pulse. We elected to use the chicken ASIC1 subunit for these experiments 30 for two important reasons. First, cASIC1 is the same subunit used from structural studies 31 (Baconguis et al., 2014;Baconguis and Gouaux, 2012;Gonzales et al., 2009;Jasti et al., 2007;32 Yoder et al., 2018). Second, in our hands cASIC1a does not undergo the strong tachyphylaxis 33 mammalian ASIC1a does in outside out patches (Chen and Grunder, 2007). Such strong 34 tachyphylaxis prevents a thorough mapping of the recovery time course and non-stationary noise 35 analysis (see below). 36 We initially examined the recovery time course of cASIC1 wild type and found that cASIC1 37 essentially completely desensitized with a time constant of 181 ± 6 ms ( Figure 1C, E-F) and fully 38 recovered in about 10 seconds (τrec = 840 ± 90 ms, n = 5, Figure 1C, E-F). Consistent with our 39 hypothesis that a smaller residue in the Leu414 position would be more nimble and subject to 40 less steric hindrance, the L414A mutation underwent faster desensitization (41 ± 1 ms, n =5, p < 41 1e -5 vs wild type) but also recovered exceptionally fast. This can be seen in Figure 1C where an 42 L414A patch is overlaid with a wild type patch. At the shortest inter-pulse interval of 10 ms, wild 43 type channels show negligible recovery yet L414A has recovered by more than 50%. To properly 14 45 resolve this highly accelerated time course, a modified pulse protocol was used with very short 46 inter-pulse intervals ( Figure 1D). This revealed L414A was essentially fully recovered in ~ 20 ms 47 (τrec = 4.0 ± 0.5 ms, n = 5, p < 1e -5 vs wild type, Figure  15 than wild type. In past studies (Kusama et al., 2013;Li et al., 2012), recovery from desensitization 49 has been well described as a mono-exponential process. This was the case for cASIC1 wild type, 50 however, the L414A mutation was poorly fit by a single exponential function ( Figure   to obtain four records of cASIC1 wild type of between 50 and 100 sweeps not varying in amplitude 110 by more than 10% (see Methods). Figure 2A illustrates one such patch where the variance for 50 111 consecutive sweeps was calculated and plotted as a function of the current amplitude ( Figure   112 2B). The NSFA indicated a peak Popen of 0.86 ± 0.02 (n = 5, Figure 2C) with an estimated single 113 channel conductance of 10 ± 1 pS, consistent with previously published conductance data 19 (Lynagh et al., 2017;Zhang and Canessa, 2002). Importantly, to our knowledge this represents open probability without substantially altering proton potency (Figures 1 and 2, Sup Fig 1). Taken 123 together, these data argue that the 'flipping' motion of the β11-12 linker is crucial for 124 desensitization but not activation. We next sought to explore the specific molecular interactions  26 versus wild type), 135 ± 7 (n = 5, p < 1e -5 versus wild type), 140 ± 3 (n = 4, p < 1e -5 versus wild 188 type), and 520 ± 50 ms (n = 5, p = 0.01 versus wild type), respectively ( Figure 4). Interestingly, 189 L281A showed the largest acceleration of recovery but also markedly increased the rate of 190 channel desensitization (τdes = 47 ± 3 ms, p < 1e -5 versus wild type). However, I306A and M364A 191 did not substantially alter desensitization (τdes = 240 ± 11; 108 ± 3 ms, respectively). We therefore 192 made the double I306A/M364A mutation with the goal of dramatically altering recovery without 193 effecting entry into desensitization. However, this double mutation did not exhibit an increased 194 effect on recovery as compared to the single mutations (τrecovery 110 ± 9 ms, n = 5, Figure 4). 195 Nonetheless, these data suggest that the desensitized state is partially stabilized by interactions  28 this rank order. Instead, we found that no clear pattern emerged in either the entry to or exit 211 from desensitization. L414R desensitized and recovered very slowly (τdes = 1600 ± 380 ms; τrec = 212 41000 ± 6400 ms, n = 3, p < 1e -5 versus wild type for both, Figure 5) as predicted. However, every 213 other mutation ran counter to the simple hypothesis that size and polarity alone predict 214 desensitization ( Figure 5). In contrast to expectations and previous reports using two-electrode 215 recordings in oocytes (Roy et al., 2013;Wu et al., 2019), the large aromatic side chain 216 substitutions of Phe and Tyr actually resulted in much faster entry and exit in both cases (L414F: 217 τdes = 5.6 ± 0.2 ms; τrec = 21 ± 2 ms, n = 5; L414Y: τdes = 3.2 ± 0.2 ms; τrec = 29 ± 2 ms, n = 4, p < 1e -218 5 for all comparisons to wild type, Figure 5). Similarly, L414N was expected to enter and exit 219 slightly slower than Ala, however, it showed comparable behavior (τdes = 48 ± 3 ms; τrec = 2.7 ± 220 0.4 ms, n = 6). The L414I substitution was an additional surprise. If the only factors at play are 221 size and polarity, then this mutation should have minimal effect. However, we found that the 222 L414I construct entered desensitization 5-fold slower (τdes = 920 ± 50 ms, p < 1e -5 versus wild 223 type) and recovered nearly 6-fold faster (τrec = 145 ± 6 ms, n = 5, p < 1e -5 versus wild type) than The following video is available for figure 6:  flipping for desensitization (Yoder et al., 2018). However, putative disulfide trapping between 265 these two residues resulted in partial suppression of desensitization, possibly indicating other 32 mechanisms are at play. Therefore, to investigate the necessity of β11-12 flipping in 267 desensitization, we turned to non-canonical amino acid (ncAA) incorporation and UV trapping. 268 The ncAA p-benzoyl-L-phenylalanine (Bpa) generates a free radical when exposed to 365 nm light 269 (Klippenstein et al., 2014;Pless and Ahern, 2013;Ye et al., 2008). The resulting free radical 270 spontaneously forms a covalent bond with a nearby atom, preferentially reacting with C-H bonds. Interestingly, the current rise times following UV application did change (10-90% rise time: before 316 UV, 12 ± 1 ms; post UV, 830 ± 400 ms, n = 7, p < 1e -5 between pre and post-UV, Figure 7C Representative whole cell recording of cells transfected with L414TAG plus R3 and YAM and supplemented with MeO-Bpa and responding to pH 5 application. Following five pH 5 applications, high power UV light is pulsed while the channels are in the resting state for additional pH 5 applications (purple traces) followed by applications without UV. (B) Summary of steady state current divided by peak current during pH 5 application before, during and after UV for L414Bpa (triangles) or wild type (circles). (C) Example responses from the same cell as A, before and after UV application. (D) Summary of steady state current divided by peak during pH 5 application for wild type patches with MeO-Bpa (circles), L414TAG with MeO-Bpa (upward triangles), or L414TAG without MeO-Bpa (downward triangles). In the present study we investigated the molecular underpinnings of entry to and exit from 2 desensitization in cASIC1. We corroborate and extend structural and functional studies 3 implicating the β11-12 linker as a regulator of desensitization. Indeed, we report that a simple 4 L414A mutation imparts a 5-fold and 200-fold acceleration in entry into and exit from 5 desensitization, respectively (Figure 1). The acceleration of desensitization was strong enough to and desensitized structures. This was further supported by mutations to either L414 or N415, 24 which alter desensitization kinetics and/or extent (Li et al., 2010a;Roy et al., 2013;Wu et al., 25 2019). It has also been suggested that the β11-12 linker is an important determinant of activation 26 as certain mutations (L414F, Y and A for example) have been reported to profoundly shifted 27 activation curves measured in oocytes (Roy et al., 2013;Wu et al., 2019). We suggest that such 28 apparent shifts in activation curves arise due to the slower solution exchange of oocytes, the 29 extremely rapid desensitization of such mutations (Figures 1 and 5) and the emergence of a 30 sustained current with a right shifted pH-dependence (Supplemental Figure 1). However, we did 31 observe that the activation times of these 'fast' mutants were generally faster than wild type (10-32 90% rise time: wt, 7 ± 3 ms, n = 15; L414A, 4 ± 1 ms, n = 11; L414Y 1.0 ± 0.3, n = 4). Further, the 33 rise times of whole cell L414Bpa responses were much slower following UV trapping than before 34 (10-90% rise time: before UV, 12 ± 1 ms; post UV, 830 ± 400 ms, n = 7, p < 1e -5 between pre and 35 post-UV) and this effect was not observed with cASIC1_GFP (10-90% rise time: before UV, 10 ± 6 36 ms; post UV, 10 ± 6 ms, n = 3). Thus there are likely some local re-arrangements of the β11-12 37 loop during the resting to open transition but these re-arrangements comprise a smaller portion 38 of the energy barrier than was previously proposed. 39 It has recently been suggested that the mechanism of ASIC desensitization can be likened to a 40 valve mechanism, wherein Gln277 acts as a clamp or valve controlling the β11-12 linker flip, and 41 hence desensitization (Wu et al., 2019) is quite striking and will undoubtedly be useful to the field, however, the 'valve' model is